Tsiampousi A 2011 PHD Thesis

NUMERICAL ANALYSIS OF SLOPES
IN UNSATURATED SOILS
A thesis submitted to Imperial College London in partial fulfilment of the requirements

for the degree of Doctor of Philosophy
By
Aikaterini Tsiampousi
Department of Civil and Environmental Engineering
Imperial College London
London SW7 2AZ
APRIL 2011
This thesis is dedicated to my father
2
«Σα βγεις ζηο πηγαιμό για ηην Ιθάκη,
Να εύχεζαι να ‘ναι μακρύς ο δρόμος»
Κ. Π. Καβάθης
‘As you set out for Ithaca

hope that the journey is a long one’
K. P. Kavafis
3
declaration
The work presented in this dissertation was carried out in the Department of
Civil and Environmental Engineering at Imperial College London. This thesis is
the result of my own work and any quotation from, or description of, the work of
others is acknowledged herein by reference to the sources, whether published
or unpublished.
This dissertation is not the same as any that I have submitted for any degree,
diploma or other qualification at any other university. No part of this thesis has
been or is being currently submitted for any such degree, diploma or other
qualification.
Aikaterini Tsiampousi
London, April 2011
4
abstract
Conventional constitutive models developed for saturated soils are inadequate

when analysing problems involving unsaturated conditions. Although
unsaturated constitutive models are available in the Imperial College Finite
Element Program (ICFEP), there are aspects of unsaturated soil response that
are not adequately simulated. The aim of the present thesis is to develop and
implement numerical expressions describing the most relevant of these features
and to apply them in combination with the existing ICFEP capabilities to
boundary value problems involving unsaturated soils. The over-prediction of the
peak shear stress exhibited by overconsolidated soils and the simplicity of the
soil-water retention relationship employed, constitute the focal points of the
improvements suggested.
A new surface is introduced to substitute for the yield and plastic potential
functions on the dry side of critical state, in order to prevent the available
constitutive models from overestimating the peak deviatoric stress. The
development, implementation and calibration of this surface are presented,
followed by analyses of laboratory experiments demonstrating the improved
simulation of soil behaviour.
Novel formulations are proposed for the soil-water retention curve, which
defines the relationship between the degree of saturation or the water content
and the applied suction, modelling its hysteretic nature and incorporating the
effect of specific volume. Ultimately, a three-dimensional hysteretic surface,
defined in terms of degree of saturation, suction and specific volume, is
presented.
The new developments are subsequently applied to the numerical analysis of

boundary value problems involving (a) the stability of slopes in overconsolidated
unsaturated soils and (b) the behaviour of unsaturated soil slopes under
seasonal changes of suction, highlighting the importance of adopting
appropriate constitutive models.
5
acknowledgements
First, I would like to acknowledge the contribution of my supervisors Dr. Lidija

Zdravkovic and Professor David M. Potts and thank them for their interest in my
work and for their guidance. Not only did they help me find my way to Ithaca
but, most importantly, they taught me how to appreciate the journey itself.
Actually, they made this journey worth taking.
I would also like to thank Dr. K. Georgiadis for his suggestions during the initial
stages of my research.
My tuition fees were covered by the EPSRC Doctoral Training Account of the
Department of Civil and Environmental Engineering, Imperial College London.
Their support is kindly acknowledged.
Over the last years I was privileged to work in a very active environment and to
be engaged in numerous academic conversations with both the staff and the
researchers of the Geotechnics Section at Imperial College London. Their
feedback and comments are very well appreciated. In particular I would like to
thank Dr. J. Standing whose engaging MSc lectures on the behaviour of
unsaturated soils inspired my interest in the specific field.
There are no words to describe the help and support I received from David
Taborda. I would have to write a whole new thesis and I am afraid that I have
exhausted my strength in this one.
Special thanks go to my uncle Tasos who encouraged me to embark on this

journey.
Finally, I would like to thank my family for their love and affection and for their
unconditional encouragement: my mum for endorsing my decision to take on
research and for finding the patience to listen about it every time she called or
visited; my sister and her children, Christos and Phaedra, for sending me an
affectionate smile over the Internet whenever I needed it and my father for
supporting my journey. His persistence that I finish my homework before
6
acknowledgements
playing, which annoyed me so much as a child, was the foundation on which I

built my life. This thesis is dedicated to him.
7
table of contents
table of contents
declaration ........................................................................................................ 4
abstract ............................................................................................................. 5
acknowledgements .......................................................................................... 6
table of contents............................................................................................... 8
list of figures ................................................................................................... 14
list of tables .................................................................................................... 29
list of symbols ................................................................................................ 31
chapter 1: Introduction ............................................................................. 40
1.1 General......................................................................................................... 40
1.2 Scope of the research ................................................................................... 41
1.3 Thesis layout ................................................................................................ 42
chapter 2: Recent Advances in the Modelling of Unsaturated Soils ..... 45
2.1 Introduction ................................................................................................... 45
2.2 Stress variables ............................................................................................ 46
2.3 Isotropic compression and yielding ............................................................... 52

2.3.1 Changes in suction at constant confining isotropic stress ...................... 52
2.3.2 Isotropic loading at constant suction ...................................................... 57
2.3.3 Uniqueness of the isotropic compression line (ICL) ............................... 60
2.4 Critical state .................................................................................................. 64
2.5 Soil-water retention curve (SWRC) ............................................................... 73
2.6 Modelling the soil-water retention curve (SWRC) .......................................... 76
2.7 Coupling of the mechanical and the hydraulic behaviour .............................. 88
2.8 Constitutive modelling ................................................................................... 91
8
table of contents
2.8.1 The Gallipoli et al. (2003a) model .......................................................... 92

2.8.2 The Wheeler et al. (2003) model............................................................ 95
2.8.3 The Sheng et al. (2008) model............................................................. 102
2.9 Summary and conclusions .......................................................................... 112
chapter 3: Existing ICFEP Capabilities for Modelling Unsaturated Soil

Behaviour ................................................................................................ 117
3.1 Introduction ................................................................................................. 117
3.2 Extension of the finite element method to unsaturated soil mechanics........ 118
3.2.1 Elasto-plastic stiffness matrix ............................................................... 118
3.2.2 Formulation for coupled problems........................................................ 121
3.2.3 𝜴 variation ........................................................................................... 124
3.3 Constitutive models for the mechanical behaviour of unsaturated soils ....... 124
3.3.1 Stress variables ................................................................................... 124
3.3.2 Yield (YS) and plastic potential (PP) surfaces ...................................... 126
3.3.3 Fully saturated isotropic compression line (ICL) and yield stress ......... 130
3.3.4 Unsaturated isotropic compression line (unsat-ICL) and yield stress ... 131
3.3.4.1 Linear and bi-linear unsat-ICL (Model 1) ....................................... 132
3.3.4.2 Non-linear unsat-ICL (Model 2)..................................................... 134
3.3.5 Suction induced wetting/drying line (WDL) ........................................... 136
3.3.6 Critical state line (CSL) ........................................................................ 138
3.3.7 Hardening/softening rules .................................................................... 141
3.3.8 Elastic behaviour ................................................................................. 142
3.3.9 Initialisation of the hardening parameters ............................................ 143
3.3.10 Summary of the model parameters ...................................................... 144
3.4 Soil-water retention curve (SWRC) model................................................... 147
3.5 Permeability models ................................................................................... 149
3.6 Hydraulic boundary conditions .................................................................... 152

3.6.1 Precipitation boundary condition .......................................................... 153
3.6.1.1 Automatic-incrementation algorithm for the precipitation boundary
condition ..................................................................................................... 155
3.6.1.2 Other applications of the precipitation boundary condition ............ 159
3.6.2 Vegetation boundary condition............................................................. 161
9
table of contents
chapter 4: Modelling of Overconsolidated Unsaturated Soils ............. 169
4.1 Introduction ................................................................................................. 169
4.2 Formulation of the Hvorslev surface for unsaturated soils ........................... 170
4.2.1 Yield surface ........................................................................................ 171
4.2.2 Plastic potential surface ....................................................................... 179
4.3 Implementation of the Hvorslev surface ...................................................... 182

4.3.1 Yield function derivatives ..................................................................... 183
4.3.2 Plastic potential function derivatives .................................................... 189
4.3.3 Plastic hardening parameter 𝑨 ............................................................. 190
4.3.4 Elastic and plastic strains due to changes in equivalent suction ........... 191
4.3.5 Yield function derivatives with respect to the Factor of Safety .............. 191
4.4 Validation of the Hvorslev surface ............................................................... 194

4.4.1 Validation under full saturation ............................................................. 195
4.4.2 Validation under unsaturated conditions .............................................. 203
4.4.3 Effect of the parameter 𝒏 ..................................................................... 207
4.4.4 Effect of the parameter 𝒎 .................................................................... 210
4.5 Calibration of the Hvorslev surface ............................................................. 212

4.5.1 Calibration under full saturation ........................................................... 213
4.5.1.1 Simulation of the Zhu & Yin (2000) laboratory tests on HK marine
clay ..................................................................................................... 213
4.5.1.2 Simulation of the Alonso et al. (2003) laboratory tests on Ancona clay
..................................................................................................... 219
4.5.2 Calibration under unsaturated states ................................................... 225
4.5.2.1 The soil tested by Estabragh & Javadi (2008) and the experimental
procedure followed ......................................................................................... 226
4.5.2.2 Model parameters employed in the analyses ................................ 227
4.5.2.3 Results and discussion ................................................................. 234
chapter 5: Modelling of the Soil Water Retention Curve ...................... 254
5.1 Introduction ................................................................................................. 254
5.2 Specific volume dependent Soil-Water Retention Curve (v-SWRC) ............ 255
5.2.1 Formulation and implementation .......................................................... 255
5.2.2 Validation ............................................................................................. 266
10
table of contents
5.2.3 Calibration ........................................................................................... 267
5.3 Hysteretic Soil-Water Retention Curve (hysteretic-SWRC) ......................... 273

5.3.1 Formulation.......................................................................................... 273
5.3.1.1 Primary drying and wetting paths .................................................. 274
5.3.1.2 Scanning drying and wetting paths ............................................... 274
5.3.1.3 Model parameters ......................................................................... 276
5.3.2 Implementation .................................................................................... 278
5.3.2.1 Initialisation of the reversal parameters......................................... 279
5.3.2.2 Updating of the reversal parameters ............................................. 281
5.3.2.3 Solution of the system .................................................................. 282
5.3.2.4 Calculation of the degree of saturation and of the gradient of the
SWRC ..................................................................................................... 287
5.3.3 Validation ............................................................................................. 288
5.3.4 Calibration ........................................................................................... 291
5.4 Specific volume dependent, hysteretic Soil-Water Retention Curve (v-

hysteretic-SWRC).................................................................................................. 298
5.4.1 Formulation and Implementation .......................................................... 298
5.5 Degree of saturation dependent soil compressibility with suction ................ 301
5.6 Suggestions for future improvement of the v-hysteretic-SWRC model ........ 313
chapter 6: Stability of Highly Overconsolidated Unsaturated Soil Slopes

................................................................................................ 320
6.1 Introduction ................................................................................................. 320
6.2 Problem description .................................................................................... 323
6.3 Factor of safety and determination of slope failure ...................................... 327
6.4 Effect of the yield and plastic potential surfaces .......................................... 332
6.5 Effect of type of analysis performed ............................................................ 337

6.5.1 Comparison between unsaturated and dry analyses ............................ 338
6.5.2 Comparison between unsaturated and effective stress analyses ......... 341
6.6 Effect of initial conditions and of excavation depth ...................................... 348

6.6.1 Effect of G.W.T. and excavation depth ................................................. 350
6.6.2 Effect of OCR ...................................................................................... 352
11
table of contents
6.6.3 Displacements ..................................................................................... 353
chapter 7: Behaviour of Unsaturated Soil Slopes under Seasonal

Changes of Suction ...................................................................................... 366
7.1 Introduction ................................................................................................. 366
7.2 Overview of the numerical analyses performed........................................... 367
7.3 Problem description .................................................................................... 370
7.4 Initial degree of saturation ........................................................................... 375
7.5 Seasonal suction variation .......................................................................... 377
7.6 Monotonic SWRC models ........................................................................... 386
7.7 Hysteretic, specific volume independent SWRC model (hysteretic-SWRC) 392
7.8 Hysteretic, specific volume dependent SWRC model (v-hysteretic-SWRC) 397
7.9 Effect of initial conditions (K0 and OCR) ...................................................... 407
7.10 Effect of root growth .................................................................................... 413
7.11 Variable permeability model ........................................................................ 427
chapter 8: Conclusions and Suggestions for Further Research ......... 437
8.1 Introduction ................................................................................................. 437
8.2 Shearing behaviour of unsaturated soils ..................................................... 438
8.3 The Hvorslev surface .................................................................................. 440
8.4 The soil-water retention curve (SWRC) ....................................................... 442
8.5 Overview of the mechanical and hydraulic modelling of unsaturated soil

behaviour in ICFEP ............................................................................................... 444
8.6 Suggestions for future developments in the modelling of unsaturated soils with
ICFEP ................................................................................................................... 446
8.7 Suggestions for future finite element analyses with ICFEP ......................... 447
references ..................................................................................................... 449
12
table of contents
appendix A .................................................................................................... 467
A.1 Supplementary derivatives for the implementation of the Hvorslev surface . 467
A.2 Development and implementation of the newly introduced Hvorslev surface in

combination with the modified Cam-clay model ..................................................... 470
appendix B .................................................................................................... 472
B.1 Supplementary analyses for the v-hysteretic-SWRC model ........................ 472
B.2 Validation and calibration of the v-hysteretic-SWRC ................................... 476
B.3 Supplementary data for the degree of saturation dependent soil

compressibility with suction ................................................................................... 478
13
list of figures
Figure 2-1: The loading-collapse (LC) yield surface (reproduced from Alonso et al.,
1990) ................................................................................................................. 53
Figure 2-2: (a) Stress paths for wetting tests performed under three different applied
stresses; (b) predicted volumetric strains for wetting tests performed under three
different applied stresses (after Gens, 2010) ..................................................... 53
Figure 2-3: Volume changes due to drying; conceptual model by Toll (1995).............. 54
Figure 2-4: Wetting-drying cycles on Boom clay (after Alonso et al., 1995) ................. 55
Figure 2-5: Wetting-drying cycle on compacted bentonite-kaolin (after Sharma, 1998):
(a) specific volume; (b) degree of saturation ...................................................... 56
Figure 2-6: Wetting-drying cycle on compacted bentonite-kaolin (after Sharma, 1998,
reported in Gallipoli et al., 2003a) ...................................................................... 56
Figure 2-7(a) Stress paths and (b) yield points in the 𝑠-𝑝 plane for compacted
speswhite kaolin (after Wheeler & Sivakumar, 1995) ......................................... 58
Figure 2-8: (a) Idealised scheme of consolidation lines at different suction levels and
(b) definition of the LC curve in the isotropic plane 𝑠-𝑝 (after Gens, 2010) ......... 59
Figure 2-9: Influence of suction on the elastic and elasto-plastic coefficients of soil
compressibility, 𝜅 and 𝜆 (after Rampino et al., 2000).......................................... 59
Figure 2-10: Isotropic compression lines corresponding to full (straight line) and partial
(curved line) saturation and amount of potential collapse predicted (after
Georgiadis, 2003) .............................................................................................. 59
Figure 2-11: Influence of compaction pressure on the isotropic compression behaviour
of speswhite kaolin (after Sivakumar & Wheeler, 2000) ..................................... 61
Figure 2-12: Influence of compaction water content on the isotropic compression
behaviour of speswhite kaolin (after Sivakumar & Wheeler, 2000) ..................... 62
Figure 2-13: Pressure-volume relationships for samples with isotropic stress history: (a)
𝑠 = 0.0 kPa; (b) 𝑠 = 100.0 kPa; (c) 𝑠 = 200.0 kPa; (d) 𝑠 = 300.0 kPa (after
Sivakumar et al., 2010) ...................................................................................... 63
Figure 2-14: Loading-collapse (LC) yield loci for isotropically prepared (IS) and one-
dimensionally (ID) prepared samples (after Sivakumar et al., 2010) .................. 64
Figure 2-15: (a) Variation of the critical state ratios and (b) variation of the critical state
compressibilities with the degree of saturation for Kiunyu gravel (after Toll & Ong,
2003) ................................................................................................................. 66
14
list of figures
compressibilities with the degree of saturation for Jurong soil (after Toll & Ong,
2003) ................................................................................................................. 66
Figure 2-17: Test paths for constant suction shear tests conducted at suction, 𝑠, of
200.0 kPa (after Wheeler & Sivakumar, 1995) ................................................... 68
Figure 2-18: (a) Deviator stress, 𝑞, and (b) specific volume, 𝑣, plotted against mean net
stress, 𝑝, at critical state (after Wheeler & Sivakumar, 1995) ............................. 68
Figure 2-19: Conditions at critical state at suction of 200.0 kPa (after Sivakumar et al.,
2010a) ............................................................................................................... 70
Figure 2-20: Critical state parameters: (a) intercept 𝜇𝑠 and slope 𝑀𝑠 of the critical state
lines in terms of 𝑞-𝑝; (b) slope of critical state lines in terms of 𝑣𝑤-𝑝 (after
Sivakumar et al., 2010a) .................................................................................... 70
Figure 2-21: Effects of stress induced anisotropy at the critical state in the 𝑣-𝑝 plane:
(a) 𝑠 = 0.0 kPa; (b) 𝑠 = 100.0 kPa; (c) 𝑠 = 200.0 kPa; (d) 𝑠 = 300.0 kPa (after
Sivakumar et al., 2010a) .................................................................................... 71
Figure 2-22: Effects of stress induced anisotropy at on the stress-strain curve: (a) fully
drained; (b) constant 𝑝 (after Sivakumar et al., 2010a)....................................... 71
Figure 2-23: Typical shape of the soil-water retention curve (SWRC) in terms of degree
of saturation, 𝑆𝑟 ................................................................................................. 75
Figure 2-24: Hydraulic hysteresis in the soil-water retention curve (SWRC) ................ 76
Figure 2-25: (a) Isotropic loading and unloading test at constant suction on compacted
speswhite kaolin; (b) Triaxial shear test at constant suction and constant mean
stress on compacted speswhite kaolin (reported in Gallipoli et al., 2003b) ......... 76
Figure 2-26: Three-dimensional plot of 𝑆𝑟 against 𝑠 and 𝑣 predicted by the Gallipoli et
al. (2003b) expression (after Gallipoli et al., 2003b) ........................................... 78
Figure 2-27: Predicted soil-water retention curves at different values of specific volume
𝑣 (after Gallipoli et al., 2003b) ............................................................................ 78
Figure 2-28: Experimental and predicted variation of 𝑆𝑟 (a) Isotropic loading and
unloading test at constant suction; (b) Triaxial shear test at constant suction and
constant mean stress (after Gallipoli et al., 2003b) ............................................ 79
Figure 2-29: Retention curve at zero vertical stress plotted in terms of water ratio 𝑒𝑤
(after Tarantino, 2009) ....................................................................................... 80
Figure 2-30: Definition of hydraulic state variables in (a) 𝑆𝑟-𝑠 plane; (b) 𝑆𝑟-𝑠 ∗ plane (after Li,
2005) .................................................................................................................. 83
Figure 2-31: Predicted scanning paths for (a) 𝛽 = 2.0; (b) 𝛽 = 4.0 (after Li, 2005)....... 83
Figure 2-32: Definition of reference curves (after Pedroso & Williams, 2010) .............. 86
15
list of figures
Figure 2-33: Definition of vertical distance 𝐷 (after Pedroso & Williams, 2010) ........... 87
Figure 2-34: Algorithm employed for the Pedroso & Williams (2010) model ................ 87
Figure 2-35: Validation of the Pedroso & Williams (2010) model against experimental
data from tests on Hostun sand (after Pedroso & Williams, 2010)...................... 87
Figure 2-36: Influence of wetting-drying cycle on subsequent behaviour during isotropic
loading, bentonite-kaolin sample (after Sharma, 1998, deducted from Wheeler et
al., 2003): (a) specific volume; (b) degree of saturation ...................................... 90
Figure 2-37: Wetting-drying cycle on compacted kaolin performed under isotropic
stress state, with 𝑝 − 𝑢𝑎 = 50.0 kPa (reported in Wheeler et al., 2003) .............. 91
Figure 2-38: Relationship between ratio 𝑒/𝑒𝑠 and bonding factor 𝜉 during isotropic
virgin loading at constant suction: (a) data by Sharma (1998); (b) data by
Sivakumar (1993) (after Gallipoli et al., 2003a) .................................................. 94
Figure 2-39: (a) isotropic compression lines at constant values of 𝜉; (b) yield loci (after
Gallipoli et al., 2003a) ........................................................................................ 95
Figure 2-40: LC, SI and SD yield curves for isotropic stress states (after Wheeler et al.,
2003) ............................................................................................................... 101
Figure 2-41: Model for water retention behaviour (after Wheeler et al., 2003) ........... 101
Figure 2-42: Coupling of the LC, SI and SD yield surfaces in the Wheeler et al. (2003)
model............................................................................................................... 102
Figure 2-43: Yield surfaces in the 𝑞-𝑝 ∗-𝑠 ∗ stress space (after Wheeler et al., 2003) 102
Figure 2-44: Void ratio versus suction and mean net stress (after Sheng et al., 2008):
(a) path ABD; (b) path (ACD) ........................................................................... 108
Figure 2-45: Isotropic compression lines at different suctions (after Sheng et al., 2008)
........................................................................................................................ 109
Figure 2-46: Void ratio versus suction under constant mean net stress and comparison
with experimental data (after Sheng et al., 2008) ............................................. 109
Figure 2-47: Yield surface for saturated conditions (after Sheng et al., 2008) ........... 110
Figure 2-48: Initial yield surface for a soil consolidated to 300 kPa at zero suction and
its evolution when the soil is then loaded under different suction levels (after
Sheng et al., 2008) .......................................................................................... 110
Figure 2-49: Initial yield surface for a soil consolidated to 300 kPa at zero suction and
its evolution when the soil is then dried under different mean net stresses (after
Sheng et al., 2008) .......................................................................................... 111
Figure 2-50: SWRC employed in the in Sheng et al. (2008) model (after Sheng et al.,
2008) ............................................................................................................... 111
Figure 2-51: Elastic region deafened by the yield surfaces and the suction increase (SI)
and suction decrease (SD) surfaces (after Sheng et al., 2008) ........................ 112
16
list of figures
Figure 3-1: Loading-Collapse (LC) and Suction Increase (SI) surfaces (after
Georgiadis, 2003) ............................................................................................ 129
Figure 3-2: Examples of yield (YS) and plastic potential (PP) functions reproduced by
the Lagioia et al. (1996) expression for unsaturated soils (after Georgiadis, 2003)
........................................................................................................................ 130
Figure 3-3: Apparent cohesion in the 𝑝-𝐽 plane (after Georgiadis, 2003) ................... 130
Figure 3-4: Fully saturated isotropic compression line (ICL) and swelling line ........... 131
Figure 3-5: Unsaturated Isotropic Compression Line (un-ICL); (a) linear unsat-ICL
(option 1) and (b) bi-linear unsat-ICL (option 2) (after Georgiadis, 2003) ... 134
Figure 3-6: Non-linear unsat-ICL for Model 2 developed by Georgiadis (2003) ......... 136
Figure 3-7: Variation of the amount of potential collapse with isotropic yield stress (after
Georgiadis, 2003) ............................................................................................ 136
Figure 3-8: Wetting/drying line – WDL (suction induced compression line) (after
Georgiadis, 2003) ............................................................................................ 137
Figure 3-9: Calculation of (a) 𝑣1𝑠𝑒𝑞 for Model 1 (Options 1 and 2) and (b) 𝑁𝑠𝑒𝑞 for
Model 2 (after Georgiadis, 2003) ..................................................................... 137
Figure 3-10: Critical state line (CSL) for modified Cam-clay (Roscoe & Burland, 1968)
........................................................................................................................ 139
Figure 3-11: Critical state point in the 𝑝 - 𝐽 stress plane ............................................ 140
Figure 3-12: Critical state line (CSL) in the 𝑝 - 𝐽 - 𝑠𝑒𝑞 stress space ........................... 140
Figure 3-13: Calculation of the specific volume 𝑣𝑐𝑠 (after Georgiadis, 2003)............. 140
Figure 3-14: Comparison of the CSL’s predicted by the two models for full and partial
saturation (after Georgiadis, 2003)................................................................... 141
Figure 3-15: Simple non-hysteretic, void-ratio independent, non-linear SWRC
implemented in ICFEP by Melgarejo (2004) ..................................................... 148
Figure 3-16: Assumed variation of soil permeability with suction (after Nyambayo,
2003) ............................................................................................................... 151
Figure 3-17: Increase in soil permeability due to desiccation cracks (after Potts &
Zdravkovic, 1999) ............................................................................................ 152
Figure 3-18: Precipitation boundary condition (after Smith, 2003) ............................. 154
Figure 3-19: Schematic operation of the automatic-incrementation procedure
implemented for the precipitation boundary condition (after Smith, 2003) ........ 157
Figure 3-20: The tolerance zone for the precipitation boundary condition (after Smith,
2003) ............................................................................................................... 158
Figure 3-21: Determination of sub-increment size during application of the precipitation
boundary condition (after Smith, 2003) ............................................................ 158
17
list of figures
Figure 3-22: Precipitation boundary condition in tunnel problem (after Potts &
Zdravkovic, 1999) ............................................................................................ 160
Figure 3-23: Precipitation as a recharge model (after Smith, 2003) ........................... 161
Figure 3-24: Assumed shape of root extraction function in the rooted zone (after
Nyambayo & Potts, 2010) ................................................................................ 164
Figure 3-25: Linear variation of 𝑎 function (after Nyambayo & Potts, 2010)............... 164
Figure 4-1: The yield surface (YS) on the wet and on the dry side of the critical state
for the MCC model (Potts & Zdravkovic(1999)) ................................................ 176
Figure 4-2: The Hvorslev surface in relation to the yield (YS) and plastic potential (PP)
functions; (a)attached to the plastic potential (PP) surface;(b) attached to the
yield surface (YS); (c) attached to the actual critical state point on the yield
surface (YS)..................................................................................................... 177
Figure 4-3: Area of illegal stress states allowed by a straight line Hvorslev surface
(shown in grey) ................................................................................................ 178
Figure 4-4: New shape of the Hvorslev surface, introduced to bound the mean
equivalent stress, p, within acceptable limits .................................................... 178
Figure 4-5: The Hvorslev surface for different values of the parameter n .................. 178
Figure 4-6: Flow vector for the current stress state denoted by point C 𝑝𝑐, 𝐽𝑐 ........... 181
Figure 4-7: Variation of the gradient of the flow vector, 𝛽, with the mean equivalent
stress, p ........................................................................................................... 181
Figure 4-8: Yield surfaces reproduced by the modified Cam-clay (MCC) and
unsaturated soil model (Unsat. M.) on the 𝑝′ − 𝐽 plane, at full saturation (OCR =
50) ................................................................................................................... 199
Figure 4-9: Drained and undrained stress paths followed in triaxial compression for
various OCR values, together with their respective yield surfaces ................... 199
Figure 4-10:Drained and undrained compression results produced by the modified
Cam-clay (MCC) and unsaturated soils model (Unsat. M.) .............................. 200
Figure 4-11: Drained and undrained stress paths followed in triaxial extension for
various OCR values, together with their respective yield surfaces ................... 201
Figure 4-12: Drained and undrained extension results produced by the modified Cam-
clay (MCC) and unsaturated soils model (Unsat. M.) ....................................... 202
Figure 4-13: Drained extension results for OCR = 20 and 50 when 𝑛 = 0.5............... 203
Figure 4-14: Drained stress paths followed in triaxial compression at various suction
levels, together with their respective yield surfaces .......................................... 205
Figure 4-15: Results in terms of deviatoric stress, 𝐽 – axial strain, 𝜀𝑎 for drained triaxial
compression analyses at various suction levels ............................................... 206
18
list of figures
Figure 4-16: Results in terms of volumetric, 𝜀𝑣𝑜𝑙, and axial strain, 𝜀𝑎, for drained
triaxial compression analyses at various suction levels .................................... 206
Figure 4-17: Hvorslev surfaces generated for different values of the parameter 𝑛 and
the corresponding drained stress paths, under saturated and unsaturated
conditions ........................................................................................................ 209
Figure 4-18: Numerical results produced for different values of the parameter n ....... 209
Figure 4-19: The Hvorslev surface generated for n=0.5 and the corresponding stress
path for saturated and unsaturated conditions ................................................. 211
Figure 4-20: Numerical results produced for different values of the parameter 𝑚 ...... 212
Figure 4-21: Normalised laboratory results of undrained triaxial compression and
extension tests for OCR of 8 and 4 and position of the simulated Hvorslev surface
and of the CSL ................................................................................................. 217
Figure 4-22: Undrained triaxial compression and extension laboratory results in terms
of 𝑝′ − 𝐽 and numerical simulations adopting the Hvorslev surface, for two
different values of OCR, 8 and 4 and considering fully saturated behaviour ..... 218
of 𝐽 − 𝜀𝑎 and numerical simulations adopting the Hvorslev surface, for two
different values of OCR, 8 and 4 and considering fully saturated behaviour ..... 218
of 𝛥𝑢 − 𝜀𝑎 and numerical simulations adopting the Hvorslev surface, for two
different values of OCR, 8 and 4and considering fully saturated behaviour...... 219
Figure 4-25: Peak and critical state stress points and the Hvorslev surface and CSL
employed for their reproduction, in terms of deviatoric stress, 𝐽, and mean
effective stress, 𝑝′ ............................................................................................ 223
Figure 4-26: Drained triaxial stress paths under full saturation, followed in the
laboratory for three value OCR values and the yield surface adopted for their
numerical simulation ........................................................................................ 224
Figure 4-27: Drained triaxial compression laboratory results in terms of 𝐽 − 𝜀𝑎 and
numerical simulations adopting the Hvorslev surface, considering fully saturated
behaviour ......................................................................................................... 224
Figure 4-28: Drained triaxial compression laboratory results in terms of 𝜀𝑣𝑜𝑙 − 𝜀𝑎 and
numerical simulations adopting the Hvorslev surface, considering fully saturated
behaviour ......................................................................................................... 225
Figure 4-29: Degree of saturation measured in the laboratory at the commencement of
shearing for each suction level studied and the Van Genuchten (1980) type
SWRC fitted ..................................................................................................... 231
19
list of figures
Figure 4-30: Laboratory data for the critical state lines (CSLs) in the 𝐽 – 𝑝 space for
different suction levels (shown in symbols) and; (a) their respective positions
predicted by the current model; (b )their positions as could be predicted if
cohesion at zero equivalent suction was allowed and (c) their position if the
critical state line was allowed to become curved with increasing equivalent
suction ............................................................................................................. 232
Figure 4-31: Calibration of the parameters 𝛼𝐻𝑉, 𝛽𝐻𝑉, 𝑛 and 𝑚, associated with the
Hvorslev surface based on the laboratory results for OCR = 11.0 and 𝑠𝑒𝑞 = 200.0
kPa .................................................................................................................. 233
Figure 4-32: Variation of the slope of the ICL, 𝜆(𝑠𝑒𝑞) with equivalent suction measured
by Estabragh and Javadi (2008) and comparison with the variation reproduced by
the constitutive model ...................................................................................... 234
Figure 4-33: Loading-Collapse (LC) curve reproduced by the model parameters 𝑟, 𝛽
and 𝑝𝑐 .............................................................................................................. 234
Figure 4-34: Yield surfaces adopted and stress paths observed in the laboratory, for
𝑠𝑒𝑞 = 0.0 kPa ................................................................................................. 242
Figure 4-35: Laboratory results and numerical simulations adopting four different
surfaces; the Hvorslev surface(HV), the Cam-clay surface (CC), the Modified
Cam-clay surface (MCC) and the Sinfonietta Classica surface (SC), for 𝑠𝑒𝑞 =
0.0 kPa ............................................................................................................ 242
𝑠𝑒𝑞 = 100.0 kPa ............................................................................................. 243
100.0 kPa......................................................................................................... 243
𝑠𝑒𝑞 = 200.0 kPa ............................................................................................. 244
200.0 kPa......................................................................................................... 244
𝑠𝑒𝑞 = 300.0 kPa ............................................................................................. 245
20
list of figures
300.0 kPa......................................................................................................... 245
Figure 4-42: Laboratory results in terms of 𝐽 − 𝜀𝑎 and numerical simulations adopting
the Hvorslev surface(HV) attached to the Sinfonietta Classica (SC), for various
values of OCR and at 𝑠𝑒𝑞 = 0.0 kPa ................................................................ 246
Figure 4-43: Laboratory results in terms of 𝜀𝑣𝑜𝑙 − 𝜀𝑎 and numerical simulations
adopting the Hvorslev surface(HV) attached to the Sinfonietta Classica (SC), for
various values of OCR and at 𝑠𝑒𝑞 = 0.0 kPa.................................................... 246
the Hvorslev surface(HV) attached to the Sinfonietta Classica (SC),, for various
values of OCR and at 𝑠𝑒𝑞 = 100.0 kPa ............................................................ 247
adopting the Hvorslev surface (HV) attached to the Sinfonietta Classica (SC), for
various values of OCR and at 𝑠𝑒𝑞 = 100.0 kPa................................................ 247
Figure 5-1: 3-dimensional SWRS in the 𝑠 - 𝑆𝑟 - 𝑣 space and SWRC’s at constant
specific volume (iso-volumetric SWRC’s) ......................................................... 261
Figure 5-2: Projection in the 𝑠 - 𝑆𝑟 plane of iso-volumetric SWRC’s ......................... 261
Figure 5-3: 3-dimensional SWRS in the 𝑠 - 𝑆𝑟 - 𝑣 space generated for 𝜓 = 2.5 ........ 262
Figure 5-4: Projection in the 𝑠 - 𝑆𝑟 plane of the SWRC generated for 𝜓 = 2.5 ........... 262
Figure 5-6: Projection in the 𝑠 - 𝑆𝑟 plane of the SWRC generated for 𝜓 = 1.0 .......... 263
Figure 5-8: Projection in the 𝑠 - 𝑆𝑟 plane of the SWRC generated for 𝜓 = 0.0 ........... 264
21
list of figures
Figure 5-9: Variation of the degree of saturation, 𝑆𝑟, due to changes in the specific
volume occurring during drained analyses, for (a) 𝜓 = 2.5; (b) 𝜓 = 1.0 and (c) 𝜓 =
0.0 ................................................................................................................... 265
Figure 5-10: (a) Deviatoric stress, 𝐽, and (b) volumetric strains, 𝜀𝑣𝑜𝑙, versus axial strain,
𝜀𝑎,calculated for different values of the parameter 𝜓 ....................................... 265
Figure 5-11: The yield surface and critical state line generated for 𝜓 = 2.5 and the
stress path followed for OCR = 11.0 ................................................................ 265
Figure 5-12: Comparison of two SWRC’s produced by ICFEP with their analytical
solution, for two values of specific volume, 𝑣, 1.50 and 1.70 ............................ 266
Figure 5-13: Changes in the (a) specific volume, 𝑣, and (b) in the degree of saturation,
𝑆𝑟, measure during drying and wetting from the as-compacted states (laboratory
data after Jotisankasa, 2005) ........................................................................... 271
Figure 5-14: The LC curve obtained by Jotisankasa (2005) and its numerical
reproduction..................................................................................................... 271
Figure 5-15: Changes in the degree of saturation generated in the laboratory and
numerically, adopting two SWRC models; the v-SWRC and a simple Van
Genuchten type SWRC.................................................................................... 272
Figure 5-16: Primary and scanning paths assumed for the hysteretic-SWRC model . 277
Figure 5-17: Wetting and drying cycles from initial retention point A ......................... 279
Figure 5-18: Possible solutions of the system of equations when drying from an initial
retention point A............................................................................................... 286
Figure 5-19: ICFEP and Excel calculation of the degree of saturation, 𝑆𝑟, generated
during primary drying and wetting, as well as on drying and wetting from random
initial retention points, A and B; Test 1 ............................................................. 290
during cyclic changes of suction; Test 2 ........................................................... 290
Figure 5-22: ICFEP and Excel calculation of the degree of saturation, 𝑆r, generated
Figure 5-23: Hydraulic paths followed by specimens of London clay fill (data after
Melgarejo, 2004); (a) numerical reproduction of the primary paths; (b) numerical
reproduction of a cyclic scanning hydraulic path .............................................. 296
Figure 5-24: Hydraulic paths followed by compacted specimens of artificial soil A (data
after Jotisankasa, 2005)and their numerical reproduction ................................ 297
22
list of figures
Figure 5-25: Hydraulic paths followed by specimens of reconstituted Weald clay (data
after Melgarejo, 2004); (a) numerical reproduction of the primary paths; (b)
numerical reproduction of a cyclic hydraulic path ............................................. 297
Figure 5-26: 3-dimensional hysteretic SWRS in the 𝑠 - 𝑆𝑟 - 𝑣 space ......................... 300
Figure 5-27: Projection in the seq ∗ - Sr plane of the 3-dimensional SWRS ............... 300
Figure 5-28: (a) Specific volume change and (b)volumetric strains 𝜀𝑣𝑜𝑙 evaluated
during the cyclic changes of suction applied for Tests 3 and 4 (Figures 5.21 and
5.22, respectively)............................................................................................ 306
Figure 5-29: Changes in void ratio measured during unconfined drying and wetting of
compacted Soil A (after Jotisankasa, 2005) ..................................................... 307
Figure 5-30: Changes in specific volume measured during unconfined drying and
wetting of reconstituted Soil A (after Cunningham, 2000) ................................. 307
Figure 5-31: SWRC measured in the laboratory for compacted Soil A (after
Jotisankasa, 2005)........................................................................................... 308
Figure 5-32: SWRC measured in the laboratory for reconstituted Soil A (after
Cunningham, 2000) ......................................................................................... 308
Figure 5-33: Numerical reproduction of the experimentally observed behaviour of
compacted Soil A (Jotisankasa, 2005) employing the v-hysteretic-SWRC in
combination with the saturation dependent soil compressibility with suction 𝜅𝑠 ∗
........................................................................................................................ 309
reconstituted Soil A (Cunningham, 2000) employing the v-hysteretic-SWRC in
combination with the saturation dependent soil compressibility with suction 𝜅𝑠 ∗
........................................................................................................................ 310
compacted bentonite-kaolin mixture (Sharma, 1998) employing the v-hysteretic-
SWRC in combination with the saturation dependent soil compressibility with
suction 𝜅𝑠 ∗ ...................................................................................................... 311
Figure 5-36: Monotonic SWRC in the 3-dimensional stress-space s-𝑆𝑟-𝑣 ................. 311
Figure 5-37: Projection of the monotonic SWRC shown in Figure 5.36 on the (a) s-𝑆𝑟
(b) s-𝑣 and (c) v-𝑆𝑟 planes ............................................................................... 312
Figure 5-38: Numerically reproduced SWRC for soil A (data after Cunningham , 2000)
........................................................................................................................ 315
Figure 5-39: Numerically reproduced SWRC for soil B1M9 (data after Cunningham ,
2000) ............................................................................................................... 316
Figure 5-40: Numerically reproduced SWRC for soil M8K1B1(data after Cunningham ,
2000) ............................................................................................................... 316
23
list of figures
Figure 6-1: Geometry of the problem ........................................................................ 326

Figure 6-2: Finite element mesh................................................................................ 326
Figure 6-3: Groundwater table (G.W.T.) position at the end of the excavation .......... 326
Figure 6-4: Shrinkage of the yield (YS) and plastic potential (PP) surfaces due to an
increase in the factor of safety 𝐹𝑠 .................................................................... 330
Figure 6-5: Movement of the stress point due to the shrinking of the yield surface from
the initial position YS (0) to the subsequent position YS (1) and (YS (2) and to the
final position (YS (3) due to the increase in the factor of safety 𝐹𝑠 ................... 331
Figure 6-6: Vectors of incremental displacements at failure ...................................... 331
Figure 6-7: Horizontal and vertical displacements of the toe of the slope signifying
failure............................................................................................................... 331
Figure 6-8: Vectors of incremental displacement for Analyses 1, 2, 3 and 4.............. 334
Figure 6-9: Contours of plastic volumetric strains generated at the end of Analyses 1, 2,
3 and 4 ............................................................................................................ 335
Figure 6-10: Contours of plastic deviatoric strains generated at the end of Analyses 1,
2, 3 and 4......................................................................................................... 336
Figure 6-11: Vectors of incremental displacement for Analysis 3 (OCR = 5.5, BBM-
type) corresponding to 𝐹𝑠=2.9 ......................................................................... 337
Figure 6-12: Variation of Factor of safety 𝐹𝑠 with the assumed groundwater table
(G.W.T.) depth, for the unsaturated and dry analyses ...................................... 340
Figure 6-13: Variation of the factor of safety with groundwater table (after Georgiadis et
al., 2007).......................................................................................................... 340
Figure 6-14: Comparison of factors of safety computed from unsaturated and dry (or
conventional) analyses during the current study and by Georgiadis et al. (2007)
........................................................................................................................ 341
Figure 6-15: Groundwater table (G.W.T.) positions at -5.0 m and -10.0 m for dry
analyses .......................................................................................................... 341
Figure 6-16: Changes in pore pressure and factor of safety during the excavation of a
cut in clay (after Bishop & Bjerrum, 1960) ........................................................ 345
Figure 6-17: Variation of the factor of safety with time for slopes excavated in Brown
London clay (after Lerouil, 2001, deducted from Chandler, 1984) .................... 346
Figure 6-18: Groundwater table (G.W.T.) positions reached at the end of the
excavation for the unsaturated and the effective stress analyses with initial G.W.T
at -5.0 m .......................................................................................................... 346
Figure 6-19: Short-term and steady-state G.W.T. for: (a) the unsaturated analysis; (b)
the effective stress analysis ............................................................................. 347
24
list of figures
Figure 6-20: Failure surfaces in the deviatoric plane (after Potts & Zdravkovic, 1999)
........................................................................................................................ 347
Figure 6-21: Variation of critical state angle of shearing resistance, 𝜑𝑐𝑠 , with Lode’s
angle, θ, predicted by: the Mohr-Coulomb hexagon; the Matsuoka-Nakai (1974)
failure criterion; the van Eekelen (1980) expression ......................................... 348
(G.W.T.) depth, for the unsaturated and effective stress analyses ................... 348
(G.W.T.) depth, for different OCR values and excavation levels ...................... 355
Figure 6-24: Variation of Factor of safety 𝐹𝑠 with the water table/excavation depth ratio
R, for OCR values of 5.5 and 11.0 ................................................................... 356
Figure 6-25: Increase in factor of safety against slope failure for an increase in
cohesion (after Fredlund, 1981) ....................................................................... 356
Figure 6-26: Variation of the factor of safety of slopes in the BT formation for different
groundwater table (GW) position (after Rahardjo et al., 2010) ......................... 357
Figure 6-27: Variation of the factor of safety of slopes in the JF formation for different
groundwater table (GW) position (after Rahardjo et al., 2010) ......................... 357
Figure 6-28: Schematic representation of the yield surfaces corresponding to OCR =
11.0 and OCR = 5.5, for a random isotropic initial stress state ......................... 357
Figure 6-29: Displacement of Node 1 (crest): (a) horizontal and (b) vertical, for
excavation depth of -10.0 m (deep).................................................................. 358
Figure 6-30: Displacement of Node 2 (mid-slope): (a) horizontal and (b) vertical, for
Figure 6-31: Displacement of Node 3 (toe): (a) horizontal and (b) vertical, for
Figure 6-32: Displacement of Node A (crest): (a) horizontal and (b) vertical, for
excavation depth of -4.0 m (shallow)............................................................... 361
Figure 6-33: Displacement of Node B (toe): (a) horizontal and (b) vertical, for
excavation depth of -4.0 m (shallow)................................................................ 362
Figure 7-1: Comparison of the primary drying and wetting paths generated by the
hysteretic-SWRC-model with the SWRC reproduced by the simple, Van
Genuchten (1980) type expression .................................................................. 376
Figure 7-2: Primary curves reproduced by the v-hysteretic-SWRC model, assuming
initial void ratio equal to 0.7.............................................................................. 377
Figure 7-3: SWRC’s reproduced by the simple, Van Genuchten (1980) type model and
the v-SWRC model, assuming initial void ratio equal to 0.7 ............................. 377
25
list of figures
Figure 7-4: Distribution of subsurface water, according to Bear (1972) (after Smith,
2003) ............................................................................................................... 381
Figure 7-5: Seasonal variation of the groundwater table (G.W.T.) predicted for years
1970 to 1974; Analysis D.1 (v-hysteretic-SWRC model, 𝜓 = 1.0)..................... 382
Figure 7-6: Typical summer (August) and winter (February) pore water pressure
distribution with depth for Analysis D.1 (v-hysteretic-SWRC model, 𝜓 = 1.0) ... 383
Figure 7-7: Vectors of incremental displacements showing (a) typical swelling during
the winter and (b) typical shrinkage during the summer for Analysis D.1 (v-
hysteretic-SWRC model, 𝜓 = 1.0) .................................................................... 383
Figure 7-8: Horizontal displacements along vertical lines starting at the (a) crest and at
(b) mid-slope; Analysis D.1 (v-hysteretic-SWRC model, 𝜓 = 1.0)..................... 384
Figure 7-9: Vertical displacements along horizontal lines starting at the (a) crest and at
(b) mid-slope; Analysis D.1 (v-hysteretic-SWRC model, 𝜓 = 1.0)..................... 385
Figure 7-10: Crest displacements for analyses A(simple SWRC); B.1; B.2 and B.3 (v-
SWRC): (a) horizontal and; (b) vertical ............................................................ 388
Figure 7-11: Mid-slope displacements for analyses A(simple SWRC); B.1; B.2 and B.3
(v-SWRC): (a) horizontal and; (b) vertical ........................................................ 389
Figure 7-12: Toe displacements for analyses A(simple SWRC); B.1; B.2 and B.3 (v-
Figure 7-13: Groundwater Table (G.W.T.) positions at the end of August 1974 for
Analyses A, B.1, B.2 and B.3 ........................................................................... 390
Figure 7-14: Contours of degree of saturation, 𝑆𝑟, evaluated at the end of August 1974
for Analyses A, B.1, B.2 and B.3 ...................................................................... 391
Figure 7-15: Crest displacements for analyses A(simple SWRC); C.1; C.2 and C.3
(hysteretic-SWRC): (a) horizontal and; (b) vertical ........................................... 394
Figure 7-16: Mid-slope displacements for analyses A(simple SWRC); C.1; C.2 and C.3
(hysteretic-SWRC): (a) horizontal and; (b) vertical ........................................... 395
Analyses C.1, C.2 and C.3 ............................................................................... 396
for Analyses C.1, C.2 and C.3.......................................................................... 396
Figure 7-19: Crest displacements for analyses C2(hysteretic-SWRC); D.1; D.2 and D.3
( v-hysteretic-SWRC): (a) horizontal and; (b) vertical ....................................... 400
Figure 7-20: Mid-slope displacements for analyses C2(hysteretic-SWRC); D.1; D.2 and
D.3 ( v-hysteretic-SWRC): (a) horizontal and; (b) vertical ................................. 401
Analyses D.1, D.2 and D.3 ............................................................................... 401
26
list of figures
for Analyses D.1, D.2, D.3 and D.4 ................................................................. 402
Figure 7-23: Crest displacements for analyses D.3 and D.4 ( v-hysteretic-SWRC): (a)
horizontal and; (b) vertical ................................................................................ 403
Figure 7-24: Mid-slope displacements for analyses D.3 and D.4 ( v-hysteretic-SWRC):
(a) horizontal and; (b) vertical .......................................................................... 404
Analyses D.3 and D.4 ...................................................................................... 404
Figure 7-26: Crest displacements for Analysis D.1.time ( v-hysteretic-SWRC): (a)
Figure 7-27: Mid-slope displacements for Analysis D.1.time ( v-hysteretic-SWRC): (a)
Figure 7-28: Deformed shape of the finite element mesh at the end of the 15 th year of
analysis (sub-accumulated from the end of the excavation) ............................. 406
Figure 7-29: Crest displacements for analyses D.1, D.1.K 0A and D.1.K0B ( v-hysteretic-
Figure 7-30: Mid-slope displacements for analyses D.1, D.1.K 0A and D.1.K0B ( v-
hysteretic-SWRC): (a) horizontal and; (b) vertical ............................................ 410
Figure 7-31: Crest displacements for analyses D.1 and D.1.OCR ( v-hysteretic-SWRC):
Figure 7-32: Mid-slope displacements for analyses D.1 and D.1.OCR ( v-hysteretic-
Figure 7-33: Vectors of incremental displacements calculated at (a) March 1970 and
(b) May 1970, for Analysis D.1.OCR (OCR = 1.5) ............................................ 413
Figure 7-34: Crest displacements for analyses D.1 and D.1.root (v-hysteretic-SWRC):
Figure 7-35: Mid-slope displacements for analyses D.1 and D.1.root ( v-hysteretic-
Figure 7-36: Schematic variation of the sink term, 𝑆𝑚𝑎𝑥, with depth for deep and
shallow maximum root depth, 𝑟𝑚𝑎𝑥 ................................................................. 419
Figure 7-37: Groundwater Table (G.W.T.) positions for Analyses D.1 and D.1root at the
end of (a) August 1970; (b) February 1971; (c) August 1971; (d) August 1974 420
Figure 7-38: Horizontal displacements along vertical lines starting at the (a) crest and
at (b) mid-slope; Analyses D.1 (𝑟𝑚𝑎𝑥 = 2.0 m) and D.1.root (𝑟𝑚𝑎𝑥 growing by 0.5
m / year), August 1970..................................................................................... 421
27
list of figures
at (b) mid-slope; Analyses D.1 (𝑟𝑚𝑎𝑥 = 2.0 m) and D.1.root (𝑟𝑚𝑎𝑥 growing by
0.5 m / year), August 1972 ............................................................................... 422
at (b) mid-slope; Analyses D.1 (𝑟𝑚𝑎𝑥 = 2.0 m) and D.1.root (𝑟𝑚𝑎𝑥 growing by
0.5 m / year), August 1974 ............................................................................... 423
Figure 7-41: Vertical displacements along horizontal lines starting at the (a) crest and
at (b) mid-slope; Analysis D.1 (𝑟𝑚𝑎𝑥 = 2.0 m) and D.1.root (𝑟𝑚𝑎𝑥 growing by 0.5
m / year), August 1970..................................................................................... 424
Figure 7-42 Vertical displacements along horizontal lines starting at the (a) crest and at
(b) mid-slope; Analysis D.1 (r𝑚𝑎𝑥 = 2.0 m) and D.1.root (𝑟𝑚𝑎𝑥 growing by 0.5 m
/ year), August 1972 ........................................................................................ 425
Figure 7-43: Vertical displacements along horizontal lines starting at the (a) crest and
at (b) mid-slope; Analysis D.1 (𝑟𝑚𝑎𝑥 = 2.0 m) and D.1.root (𝑟𝑚𝑎𝑥 growing by 0.5
m / year), August 1974..................................................................................... 426
Figure 7-44: Crest displacements for maximum root depth 𝑟𝑚𝑎𝑥=2.0 m and maximum
root depth 𝑟𝑚𝑎𝑥 increasing by 0.5 m/year: (a) horizontal and; (b) vertical........ 427
Figure 7-45: Variation of permeability with (a) equivalent suction, 𝑠𝑒𝑞, and (b) degree of
saturation, 𝑆𝑟, for Analyses D.1 (suction dependent permeability) and D.1.perm
(degree of saturation dependent permeability) ................................................. 430
Figure 7-46: Crest displacements for analyses D.1 (suction dependent permeability)
and D.1.perm (𝑆𝑟 dependent permeability): (a) horizontal and; (b) vertical ...... 431
Figure 7-47: Mid-slope displacements for analyses D.1 (suction dependent
permeability) and D.1.perm (𝑆𝑟 dependent permeability): (a) horizontal and; (b)
vertical ............................................................................................................. 432
28
list of tables
Table 2-1: Classification of constitutive models for unsaturated soils according to the
first constitutive variable (adapted from Gens, 2010) ......................................... 51
Table 3-1: Summary of the model input parameters (after Georgiadis, 2003) ........... 146
Table 4-1: Model parameters employed during validation under fully saturated states
........................................................................................................................ 198
Table 4-2: Isotropic yielding and critical state mean effective stress ......................... 198
Table 4-3: Isotropic yield stress at zero equivalent suction, 𝑝0 ∗, and current isotropic
yield stress, 𝑝0 ................................................................................................. 205
Table 4-4: Model parameters calibrated for fully saturated undrained conditions ...... 217
Table 4-5: Model parameters calibrated for fully saturated drained conditions .......... 223
Table 4-6: Initial stresses and degree of saturation prior to shearing ........................ 230
Table 4-7: Model parameters calibrated for unsaturated drained conditions.............. 230
Table 4-8: Specific volumes calculated at the beginning of the analyses based on the
model parameters, in comparison with those measured in the laboratory by
Estrabragh & Javadi (2008) ............................................................................. 231
Table 5-1: v-SWRC model parameters employed for the generation of the 3-
dimensional surface illustrated in Figure 4-49 .................................................. 260
Table 5-2: v-SWRC model parameters employed for the validation of the model ...... 266
Table 5-3: Parameters calibrated for the simulation of isotropic compression tests with
the constitutive model developed by Georgiadis (2003) ................................... 270
Table 5-4: Initial states adopted in the analyses of isotropic compression tests ........ 270
Table 5-5: Initialisation of the reversal parameters .................................................... 280
Table 5-6: Updating of the reversal parameters ........................................................ 280
Table 5-7: System of equations to be solved on drying ............................................. 282
Table 5-8: System of equations to be solved on wetting ........................................... 283
Table 5-9: Derivatives of Equations 5.14 and 5.15 with respect to 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 and 𝑟;
drying............................................................................................................... 284
Table 5-10: Derivatives of Equations 5.16 and 5.17 with respect to 𝑙𝑜𝑔𝑠𝑐𝑜𝑚𝑚𝑜𝑛 and 𝑟;
wetting ............................................................................................................. 285
Table 5-11: Parameters employed in the validation of the hysteretic-SWRC model .. 289
Table 5-12: Changes in suction applied for the validation of the hysteretic-SWRC.... 289
29
list of tables
Table 5-13: Model parameters produced by the calibration of the hysteretic-SWRC

model for three different types of soil; London clay fill, artificial soil A and Weald
clay .................................................................................................................. 295
Table 5-14: Model parameters employed for the prediction of the variation of the
specific volume with suction, measured by Jotisankasa (2005)on compacted soil
A ...................................................................................................................... 305
specific volume with suction, measured by Cunningham (2000) on reconstituted
soil A................................................................................................................ 305
specific volume with suction, measured by Sharma (1998) on compacted
bentonite-kaolin ............................................................................................... 306
Table 5-17: Model parameters employed by the improved version of the v-hysteretic-
SWRC for the reproduction of three retention curves; soil A, B1M9 and M8K1B1
........................................................................................................................ 315
Table 6-1: Analyses undertaken for the investigation of cut slopes in highly
overconsolidated unsaturated soils .................................................................. 322
Table 6-2: Model parameters employed in the parametric study of slope stability ..... 325
Table 6-3: Factors of safety associated with use of different yield surfaces and OCR
values .............................................................................................................. 333
Table 6-4: Factors of safety obtained for the unsaturated and dry analyses .............. 340
Table 6-5: Factors of safety obtained for the unsaturated and effective stress analyses
in the long-term conditions ............................................................................... 345
Table 6-6: Factors of safety associated with the use of different groundwater table
(G.W.T.) depths and OCR values and achieved for two different excavation
depths; 10.0 m (deep) and 4.0 m (shallow) ...................................................... 355
Table 7-1: Analyses undertaken for the investigation of the effect of seasonal changes
of suction on unsaturated soil movement ......................................................... 369
Table 7-2: Constitutive model and transpiration properties employed in the study .... 373
Table 7-3: SWRC and variable permeability model parameters employed in the study
........................................................................................................................ 373
Table 7-4: Potential Evapotranspiration and precipitation data, Building Research
Establishment testing site Chattenden, Kent (U.K.) .......................................... 374
30
list of symbols
Abbreviations and acronyms

𝐵𝐵𝑀 Barcelona Basic Model (Alonso et al., 1990)
𝐶𝐶 Cam-clay
𝐶𝑆𝐿 Critical state line
𝐹𝐶𝑉 First constitutive variable
𝐺𝑊𝑇 Ground-water table
𝐻𝐾𝑀𝐶 Hong Kong marine clay
𝐻𝑉 Hvorslev surface
𝐼𝐶 Isotropically compressed
𝐼𝐶𝐹𝐸𝑃 Imperial College Finite Element Program (Potts & Zdravkovic, 1999)
𝐼𝐶𝐿 Isotropic compression line
𝐼𝐷 One-dimensionally
𝐿𝐶 Loading-collapse
𝑀𝐶𝐶 Modified Cam-clay
𝑂𝐶𝑅 Overconsolidation ratio
𝑃𝑃 Plastic potential
𝑅𝑊𝑈𝑃 Root water uptake model
𝑆𝐶 Sinfonietta Classica
𝑆𝐶𝑉 Second constitutive variable
𝑆𝐷 Suction decrease
𝑆𝐹𝐺 Sheng-Fredlund-Gens
𝑆𝐼 Suction increase
𝑆𝑊𝑅𝐶 Soil-water retention curve
𝑢𝑛𝑠𝑎𝑡 − 𝐼𝐶𝐿 Unsaturated isotropic compression line
𝑊𝐷𝐿 Wetting/drying line
𝑌𝑆 Yield surface
31
list of symbols
Greek alphabet
Suction at projection centre (Li 2005); model parameter in the v-
𝛼
SWRC model
𝛼∗ Log-scaled suction at projection centre (Li 2005)
𝛼𝑖 Constitutive model parameter
𝛼𝐻𝑉 Tangent inclination of the Hvorslev surface
Fitting parameter (Li, 2005); time-stepping factor; constitutive

𝛽
model parameter
Model parameter in the SWRC proposed by Pedroso & Williams

𝛽1
(2010)

𝛽𝑑
(2010)
Gradient of the flow vector associated with the plastic potential of
𝛽𝐻𝑉
the Hvorslev surface (at 𝑝 = 𝑓(𝑠𝑒𝑞 ))

𝛽𝑤
(2010)
𝛾 Bulk unit weight
𝛾𝑑 Dry unit weight
𝛾𝑤 Unit weight of water (9.81 kN/m3 for 1𝑔 conditions)
𝛤𝑎𝑏 Specific volume for 𝑝=𝑠=0.0 (Toll, 1990)
𝛤𝑠 Intercept of the CSL at full saturation (Toll, 1990)
Intercept of the CSL in the 𝑣-ln 𝑝 plane as a function of suction

𝛤 𝑠
(Wheeler & Sivakumar, 1995)
𝛿𝑖𝑗 Kronecker delta
𝛥 Increment; variation
𝜀 Strain
𝜀𝑎 Axial strain
𝜀𝑖𝑗 Strain tensor
𝜀𝑠 Deviatoric strain – triaxial formulation: 2/3∙(𝜀𝑣 - 𝜀𝑕 )
𝜀𝑣𝑜𝑙 Volumetric strain – first invariant of the strain tensor
𝜂 Generalised normalised stress ratio
𝜃 Lode’s angle – third invariant of the stress tensor
𝜅 Elastic coefficient of soil compressibility
32
list of symbols
Slope of the scanning SWRC drying and wetting paths (Wheeler et

𝜅𝑠
al., 2003); elastic coefficient of soil compressibility with suction
𝜅𝑣𝑝 Slope of the swelling line at full saturation (Sheng et al., 2010)
Slope of the swelling line with respect to suction (Sheng et al.,

𝜅𝑣𝑠
2010)
𝜆 Elasto-plastic coefficient of soil compressibility
𝜆 Slope of the SWRC (Pedroso & Williams, 2010)
Critical state compressibility with respect to mean net stress (Toll &
𝜆𝑎
Ong, 2003)
𝜆𝑏 Critical state compressibility with respect to suction (Toll & Ong,
2003)
Slope of the primary SWRC drying and wetting paths (Wheeler et
𝜆𝑠 al., 2003); elasto-plastic coefficient of soil compressibility with
suction
𝜆𝑣𝑝 Slope of the ICL at full saturation (Sheng et al., 2010)
Slope of the compression line with respect to suction (Sheng et al.,

𝜆𝑣𝑠
2010)
𝜆 0 Elasto-plastic coefficient of soil compressibility at full saturation
𝜆 𝑠 Elasto-plastic coefficient of soil compressibility at suction 𝑠

Elasto-plastic coefficient of soil compressibility at equivalent suction
𝜆 𝑠𝑒𝑞
𝑠𝑒𝑞
Λ Plastic strain multiplier
𝜇 Poisson’s ratio
𝜇𝑖 Constitutive model parameter
Function of suction and/or degree of saturation within the first

𝜇1
constitutive variable
Function of suction and/or degree of saturation within the second

𝜇2
constitutive variable
Intercept of the CSL in the 𝑝 - 𝑞 plane as a function of suction

𝜇 𝑠
𝜈 Poisson’s ratio
𝜉 Bonding second stress variable employed by Gallipoli et al. (2003a)
𝜌 Mass density
𝜌𝑎 Air density
𝜎 Total stress; equivalent stress
33
list of symbols
𝜎′ Effective stress
𝜎 Net total stress (Sheng et al., 2010)
𝜎𝑖𝑗 Total stress tensor
𝜎𝑛𝑒𝑡 Net total stress
𝜎𝑛𝑒𝑡 ,𝑖𝑗 Net total stress tensor
𝜎′𝑖𝑗 Effective stress tensor
𝜎𝑖𝑗∗ Bishop’s effective stress (Wheeler et al., 2003)
𝜏 Shear stress
𝜑 Fitting parameter in the SWRC proposed by Gallipoli et al. (2003b)
𝜑𝑐𝑠 Critical state value of angle of shearing resistance
Parameter in Bishop’s effective stress: parameter in the expression

𝜒 introduced for the degree of saturation dependent soil
compressibility with suction
Fitting parameter in the SWRC proposed by Gallipoli et al. (2003b);

𝜓
model parameter in the v-SWRC model
Slope of the CSL in the 𝑣 - ln 𝑝 plane as a function of suction

𝜓 𝑠
Parameter in the expression introduced for the degree of saturation

𝜔
dependent soil compressibility with suction
Parameter governing the volume of water that flows for a given
Ω
change in the volume of voids
Roman alphabet
Fitting parameter in the constitutive model proposed by Gallipoli et
al. (2003a); Fitting parameter in the SWRC proposed by Van
𝑎 Genuchten (1980); Fitting parameter in the SWRC proposed by
Tarantino (2009); function of suction in the vegetation boundary
condition
𝐴 Plastic hardening parameter
Fitting parameter in the constitutive model proposed by Gallipoli et

𝑏 al. (2003a); Fitting parameter in the SWRC proposed by Tarantino
(2009)
𝐵 Strain matrix
Slope of the CSL in the 𝑣𝑤 - ln 𝑝 plane as a function of suction

𝐵 𝑠
34
list of symbols
𝑑 Displacement
𝐷 Constitutive matrix
𝑒 Void ratio
Void ratio at suction 𝑠 (Gallipoli et al., 2003a); water ratio (Tarantino

𝑒𝑤
& Tombolato, 2005); equivalent void ratio (Toll, 1995)
Microstructural water ratio – fitting parameter (Tarantino &

𝑒𝑤𝑚
Tombolato, 2005)
𝐸𝑑 Deviatoric strain – second invariant of the strain tensor
𝐹 Yield surface function
𝐺 Shear modulus; Plastic Potential function
𝐺𝑠 Density of soil particles
𝐽 Deviatoric stress – second invariant of the stress tensor
𝐽2𝜂 Square of the stress ratio
𝐽2𝜂𝑖 Failure value of 𝐽2𝜂
𝐽𝑚𝑎𝑥 Maximum deviatoric stress (𝐹𝑃𝑀)
Fitting parameter (Toll, 1990); permeability; constitutive model

𝑘
parameter
𝑘 State parameters vector
𝑘0 Coefficient of permeability at zero 𝑝 ′
Coupling parameter in the constitutive model proposed by Wheeler

𝑘1
et al. (2003)
Coupling parameter in the constitutive model proposed by Wheeler

𝑘2
et al. (2003)
𝑘𝑚𝑎𝑥 Parameter defining the maximum permeability
𝑘𝑚𝑖𝑛 Parameter defining the minimun permeability
𝑘𝑠𝑎𝑡 Saturated coefficient of permeability
𝐾 Bulk modulus
𝐾0 Coefficient of earth pressure at rest
𝐾𝑚𝑖𝑛 Minimum bulk modulus
𝐾𝐺 Global stiffness matrix

𝑚 model parameter for the plastic potential associated with the
Hvorslev surface; model parameter in the v-SWRC model
35
list of symbols
𝑀𝑎 Total stress ratio (Toll, 1990)
𝑀𝑏 Suction ratio (Toll, 1990)
Gradient of the critical state line in the 𝑝-𝑞 stress space for triaxial
𝑀𝑔
compression
𝑀𝑖 Constitutive model parameter
𝑀𝑁 Mass matrix
𝑀𝑠 Saturated critical state stress ratio (Toll, 1990)
Critical state stress ratio as a function of suction 𝑠 (Wheeler &

𝑀 𝑠
Sivakumar, 1995)
model parameter controlling the position of the Hvorslev surface;
𝑛
model parameter in the v-SWRC model; Porosity; Fitting parameter
in the SWRC proposed by Tarantino (2009)
Intercept of the ICL at suction at full saturation (Gallipoli et al.,
𝑁
2003a)
𝑁 Matrix of shape functions
𝑁𝑝 Matrix of pore pressure interpolation functions
𝑁 𝑠 Intercept of the ICL at suction 𝑠
Mean total stress; mean net stress (Alonso et al., 1990; Toll, 1990;
𝑝
Wheeler & Sivakumar, 1995)
𝑝’ Mean effective stress – first invariant of the stress tensor
𝑝 ′′ Average skeleton stress (Bishop’s stress) (Gallipoli et al., 2003a)
𝑝 Mean net total stress (Sheng et al., 2010)
𝑝𝑎𝑡𝑚 Absolute atmospheric pressure
𝑝𝑐 Characteristic pressure (constitutive model parameter)
𝑝𝑓 Pore fluid pressure
𝑝0 Isotropic yield stress at the current value of suction
Hardening/softening parameter, equivalent fully saturated yield

𝑝0∗
stress corresponding to zero equivalent suction
𝑞 Deviatoric stress – triaxial formulation: 𝜎’1 – 𝜎’3

𝑞𝑛 Nodal flow rate
𝑄 Flow
Constitutive model parameter; root depth; radius of the scanning

𝑟
path (hysteretic and v-hysteretic-SWRC models)
36
list of symbols
𝑟𝑚𝑎𝑥 Maximum root depth
Gradient of the SWRC (in terms of suction and volumetric water

𝑅
content)
𝛥𝑅𝐺 Right-hand side vector
𝑠 Matric suction
Yield value of suction; suction at which the degree of saturation

𝑠0
reaches its residual value
Modified suction (Wheeler et al., 2003); Log-scaled suction variable

𝑠∗
(Li 2005)
𝑠∗ Log-scaled image suction (Li 2005)
𝑠0∗ Log-scaled image suction of the projection centre (Li 2005)
𝑠𝑎𝑒 Air-entry value of suction (Sheng et al., 2010)
𝑠𝑎𝑖𝑟 Air-entry value of suction (Georgiadis, 2003)
𝑠𝑒𝑞 Equivalent suction
𝑠𝑑𝑒𝑠 De-saturation value of suction
𝑠𝑟𝑒 Residual value of suction (Sheng et al., 2010)
𝑠𝑤𝑒 Water-entry value of suction (Sheng et al., 2010)
𝑆𝑎𝑐𝑐 Actual sink term
𝑆𝑚𝑎𝑥 Maximum sink term
𝑆𝑟 Degree of saturation
𝑆𝑟0 Residual degree of saturation
Degree of saturation of the macropores (Tarantino & Tombolato,

𝑆𝑟𝑀
2005)
𝑡 Time
𝑇𝑝 Potential evapo-transpiration rate
𝑢𝑎 Pore air pressure
𝑢𝑤 Pore water pressure
𝑣 Specific volume
𝑣𝑤 Specific water volume
𝑉𝑣 Volume of voids
𝑉𝑤 Volume of water
𝑊 Work input
37
list of symbols
𝑋 Van Eekelen (1980) constant
Degree of saturation on the primary drying path (Pedroso &

𝑦𝑑
Williams, 2010)
Degree of saturation on the primary wetting path (Pedroso &

𝑦𝑤
Williams, 2010)
𝑌 Van Eekelen (1980) constant
𝑧 Geometric coordinate; distance to ground surface
𝑧𝐺𝑊𝑇 Position of the ground water table
𝑍 Van Eekelen (1980) constant
Subscripts
0 Initial values
1 Related to the primary yield surface (𝐵𝑆𝑃𝑀)
0 Related to the secondary yield surface (𝐵𝑆𝑃𝑀)
𝑎 Axial
𝑖 Incremental
𝑠𝑐𝑎𝑛 Scanning
𝑑 Drying
𝑒𝑞 Equivalent
𝐻𝑉 Hvorslev
𝑚𝑎𝑥 Maximum value
𝑚𝑖𝑛 Minimum value
𝑛𝐺 Nodal global
p Related to loading/unloading
𝑝𝑒𝑎𝑘 Peak value
𝑝𝑟 Primary
𝑠 Related to suction
𝑟𝑒𝑣 Reversal
𝑣 Volumetric
𝑤 Wetting
𝑥 Component of a quantity along the x-axis
38
list of symbols
𝑦 Component of a quantity along the y-axis
𝑧 Component of a quantity along the z-axis
Superscripts
′ Effective stresses
𝑑𝑟 Drying
𝑤𝑒𝑡 Wetting
39
chapter 1: INTRODUCTION
1.1 General
Unsaturated soils are met commonly, either in the form of natural soils above
the water table or as compacted materials used for engineering purposes.
However, the soil mechanics of unsaturated soils, and in particular its
application in numerical analysis, is not yet developed effectively enough to
attract industrial interest and application. Theoretical and experimental
difficulties have restricted interest to the academic community.
The Geotechnics Section at Imperial College London has made significant

progress in the past decade in both experimental testing and numerical analysis
of unsaturated soils. Numerical advances include the development and
implementation in the Imperial College Finite Element Program – ICFEP (Potts
& Zdravkovic, 1999) – of: appropriate constitutive models (Georgiadis, 2003);
an algorithm describing the fluid flow in unsaturated soils (Smith, 2003);
relevant hydraulic boundary conditions, such as the precipitation boundary
condition (Smith, 2003) and the vegetation boundary condition (Nyambayo,
2003); a Van Genuchten (1980) type soil-water retention curve (SWRC)
40
Introduction
(Melgarejo, 2004). These existing ICFEP capabilities formed the basis of the
research presented in this thesis.
ICFEP has been used in the numerical analyses of boundary value problems
involving unsaturated conditions, such as shallow and deep foundations
(Georgiadis, 2003), natural slopes (Smith, 2003) and embankments
(Nyambayo, 2003). Nevertheless, as the earlier research projects were
undertaken concurrently, the existing ICFEP capabilities were only recently
used in conjunction, in the boundary value problem of a natural slope in Italy
(Pirone, 2009).
Despite the above mentioned advancements, there are aspects of unsaturated

soil behaviour which are not simulated with the desirable level of accuracy. The
yield surface employed by the constitutive models of Georgiadis (2003) is
shown in the current thesis to be inaccurate in the prediction of the peak
deviatoric stress demonstrated by highly overconsolidated soils. Furthermore,
the SWRC curve implemented by Melgarejo (2004) does not account for the
effect of the specific volume and the hydraulic hysteresis. Finally, coupling
between the mechanical and the hydraulic components of behaviour is limited
to the increase of apparent cohesion with suction. These limitations are
addressed in the present thesis before employing ICFEP in the numerical
analysis of slopes in unsaturated soils.
1.2 Scope of the research
The scope of the research presented in this thesis was two-fold: to overcome
the modelling limitations summarised above and, after implementing the
solutions into ICFEP, to employ the new developments, in combination with the
existing capabilities of the numerical code, in the analysis of slopes in
unsaturated soils.
The first aim can be subdivided into two parts:
 the first part refers to improving the prediction of the peak stress
demonstrated in the laboratory by highly overconsolidated samples and
41
Introduction
constitutes the development, implementation, validation and calibration

of a new surface to substitute for the existing one on the dry side of the
critical state;
 the second part refers to the development, implementation, validation

and calibration of SWRC models accounting for the effect of specific
volume on the retention behaviour and the hydraulic hysteresis
commonly exhibited by unsaturated soils. Modelling of the coupling
between the hydraulic and the volumetric mechanical behaviour is also
included in this second part of the first aim of the research.
The second aim can also be subdivided into two parts:
 application of the solution proposed for the modelling of the shearing

behaviour of highly overconsolidated unsaturated soils in the study of
slope stability;
 application of the SWRC models in the study of the behaviour of

unsaturated soil slopes under seasonal changes of suction.
The primary purpose of the numerical analyses was to investigate the impact of
the improved soil behaviour simulation on the predicted results of a relevant
boundary value problem.
1.3 Thesis layout
Chapter 2 of the thesis is focused on the recent advances in the field of

unsaturated soil mechanics with particular emphasis on the modelling of such
soils. The use of two stress variables is discussed and the isotropic and critical
state lines for unsaturated conditions are presented. Following the description of
the soil-water retention behaviour, models for its replication proposed in the
literature are presented. Finally, the coupling of the mechanical and hydraulic
components of unsaturated soil behaviour is discussed and constitutive models
available in the literature to account for this coupling are presented.
42
Introduction
The existing ICFEP capabilities for the modelling of unsaturated soil behaviour
are explained in Chapter 3. The constitutive models available are described,
followed by the expression for the SWRC. The models for the variation of the
soil permeability due to de-saturation and desiccation are then presented.
Finally, two hydraulic boundary conditions relevant to the unsaturated numerical
analysis, the precipitation and the vegetation boundary conditions, are
described.
In Chapter 4 the development of a new surface, termed the Hvorslev surface, to

substitute for the yield and plastic potential functions on the dry side of the
critical state, is explained. The implementation of the Hvorslev surface in ICFEP
is described, followed by the validation of the implementation. Finally, the
surface is calibrated based on laboratory data available in the literature.
Chapter 5 is focused on the development of three SWRC models and the

modelling of the coupling between the hydraulic and the volumetric components
of unsaturated soil behaviour. The first of the SWRC models (v-SWRC)
accounts for the effect of specific volume on the retention behaviour, the second
model (hysteretic-SWRC) replicates the hydraulic hysteresis and the third (v-
hysteretic-SWRC) incorporates the combined effect of the specific volume and
the hysteretic behaviour. Finally, the soil compressibility due to changes in
suction is coupled to the SWRC, introducing an expression where the soil
compressibility is a function of the degree of saturation.
In Chapter 6 the stability of highly overconsolidated unsaturated soil slopes is

studied. The Hvorslev surface is employed in the analysis and the numerical
results are compared to those produced by the existing surface. Subsequently,
three different types of analyses are performed – unsaturated, dry and effective
stress analyses – with the purpose of investigating the influence of accounting
for unsaturated conditions in the numerical predictions. Finally, the effect of the
initial conditions of the soil and the geometry of the excavation is parametrically
examined.
In Chapter 7 the behaviour of unsaturated soil slopes under seasonal changes

of suction is investigated, with the purpose of studying the effect of the newly
implemented SWRC features on the numerical prediction of the volumetric
43
Introduction
response to a cyclic variation of suction. Seasonal suction variations were

imposed by the precipitation and vegetation boundary conditions, by employing
meteorological data with reference to rainfall and potential evapotranspiration.
The three SWRC models developed were used in the analyses for various
combinations of the relevant model parameters and the generated results were
compared. Other aspects considered were the modelling of the root growth and
of the variation of soil permeability and the impact of the initial conditions of the
soil.
Chapter 8 summarises the main results and conclusions of the research

presented in this thesis, providing suggestions for future research.
44
chapter 2: RECENT ADVANCES IN THE
MODELLING OF UNSATURATED SOILS
2.1 Introduction
Unsaturated soil behaviour has been shown to be significantly different from

saturated soil behaviour. The principles of classical soil mechanics, such as the
principle of effective stress, are not generally applicable to unsaturated soils.
Changes in strains are not restricted to changes in effective stresses, resulting
in a complex mechanical behaviour that cannot be explained by conventional
saturated soil mechanics.
In addition to the mechanical behaviour, the hydraulic behaviour of unsaturated

soils is also very different. Drying and wetting of the soil are irreversible
procedures as different values of suction are required in order to empty and to
flood a void with water. A physical explanation for the occurrence of hydraulic
hysteresis is given by Lu & Likos, 2004. It should be noted at this point that the
term suction is used in the present thesis meaning matric suction, for brevity.
The behaviour of unsaturated soils has been extensively discussed in previous

PhD theses at Imperial College London. Detailed reviews on the subject have
45
Recent Advances in the Modelling of Unsaturated Soils
been given in the last decade by Cunningham (2000), Georgiadis (2003), Smith
(2003), Nyambayo (2003), Melgarejo (2004), Jotisankasa (2005) and Monroy
(2006). Therefore, the present chapter is focused on the recent advances in the
field of unsaturated soils and particular emphasis is given to the theoretical
approaches and modelling of such soils.
First, the use of two stress variables in constitutive modelling is discussed and
the different combinations of stress variables proposed in the literature are
presented. The behaviour of unsaturated soils when isotropically tested is
subsequently discussed: the effect on behaviour of changes in suction and in
confining stress is examined and the uniqueness of the isotropic compression
line is discussed. The shearing behaviour of unsaturated soils is then
considered within the concept of critical state extended to unsaturated
conditions. The soil-water retention behaviour is explained and models
proposed for the representation of the soil-water retention curve are presented.
Finally, the coupling between the mechanical and the hydraulic components of
unsaturated behaviour and constitutive models accounting for it are presented.
2.2 Stress variables
Based on the early assumption that unsaturated states could be incorporated

within the effective stress principle, Bishop (1959) introduced the notion of
generalised effective stress, 𝜎 ′ , as a combination of the total stress, 𝜎, the pore
water pressure, 𝑢𝑤 , and the pore air pressure, 𝑢𝑎 :
𝜎 ′ = 𝜎 − 𝑢𝑎 + 𝜒 𝑢𝑎 − 𝑢𝑤 (2.1)
where 𝜒 is a function of the degree of saturation, 𝑆𝑟 .
However, Jennings & Burland (1962) demonstrated the inability of Bishop’s

generalised effective stress to predict wetting induced collapse of unsaturated
soils and, therefore, to describe their volumetric behaviour under certain
conditions. Bishop & Blight (1963) further challenged the applicability of the
generalised effective stress principle in unsaturated soils, demonstrating that a
46
change in matric suction (𝑢𝑎 − 𝑢𝑤 ) produces a different change in strain than a

change in net normal stress (𝜎 – 𝑢𝑎 ).
Consequently, the generalised effective stress principle was abandoned in

favour of the use of two independent stress variables – the terms ‘state’
variables (e.g. Wheeler et al., 2003) and ‘constitutive’ variables (e.g. Gens,
2010) have been alternatively used. Various combinations of stress variables
that have been proposed in the literature are presented in the current section.
Fredlund & Morgenstern (1977) showed that any two of the following three
independent stress state variables are sufficient for the description of the stress
state of unsaturated soils:
𝑠 = 𝑢𝑎 − 𝑢𝑤 (2.2)
𝜎𝑛𝑒𝑡 = 𝜎 − 𝑢𝑎 (2.3)
𝜎 ′ = 𝜎 − 𝑢𝑤 (2.4)
where 𝑠 is the matric suction, 𝜎𝑛𝑒𝑡 is the net total stress and 𝜎 ′ is the effective
stress. The authors suggested that changing the individual components of each
one of the stress state variables (i.e. 𝜎, 𝑢𝑤 and 𝑢𝑎 ) without modifying the stress
variable itself, results in no distortion or volume change. Carrying out such
experimental ‘null’ tests, they verified that the above stress state variables are
independent. Furthermore, as these were extensively employed in the
formulation of constitutive equations (e.g. Alonso et al., 1990; Toll, 1990;
Wheeler & Sivakumar, 1995; Cui & Delage, 1996; Georgiadis, 2003 and more
recently Sheng et al., 2008) a tensorial notation is more appropriate:
𝑠 = 𝑢𝑎 − 𝑢𝑤 (2.5)
𝜎𝑛𝑒𝑡 ,𝑖𝑗 = 𝜎𝑖𝑗 − 𝑢𝑎 𝛿𝑖𝑗 (2.6)
𝜎 ′ 𝑖𝑗 = 𝜎𝑖𝑗 − 𝑢𝑤 𝛿𝑖𝑗 (2.7)
where 𝛿𝑖𝑗 is the Kronecker delta.
47
Tarantino et al. (2000) noted that the tests by Fredlund & Morgenstern (1977)
were performed employing the axis translation technique (i.e. in the positive
pore water pressure range) and under high degrees of saturation, 𝑆𝑟 , where the
air phase would likely be in the form of occluded bubbles. Therefore, they
argued that the conclusions of the earlier study required verification. In order to
confirm or disprove the work of Fredlund & Morgenstern (1977), Tarantino et al.
(2000) performed experiments in the negative pore water pressure range and at
degrees of saturation corresponding to a continuous air phase. Maintaining the
strain state and the water content constant, they applied changes in the pore air
pressure, 𝑢𝑎 , and measured the response in terms of mean net stress, (𝑝 − 𝑢𝑎 ),
and matric suction, ( 𝑢𝑎 − 𝑢𝑤 ). They found the changes in the two stress
variables negligible thus confirming the earlier work by Fredlund & Morgenstern
(1977). Furthermore, Tarantino et al. (2000) reasoned that the term ‘stress
variables’ should be reserved for the components 𝜎, 𝑢𝑤 and 𝑢𝑎 and showed that
two of them are independent.
A new perspective regarding the stress variables employed resulted from the
work of Houlsby (1997), who calculated the increment of work input, 𝑑𝑊, per
unit volume of an unsaturated granular material to be:
𝑑𝜌𝑎
𝑑𝑊 = 𝑢𝑎 𝑛 1 − 𝑆𝑟 − 𝑢𝑎 − 𝑢𝑤 𝑛 ∙ 𝑑𝑆𝑟 +
𝜌𝑎
(2.8)
+ 𝜎𝑖𝑗 − 𝑆𝑟 𝑢𝑤 + 1 − 𝑆𝑟 𝑢𝑎 𝛿𝑖𝑗 ∙ 𝑑𝜀𝑖𝑗
where 𝜌𝑎 is the air density, 𝑛 is the porosity and 𝜀𝑖𝑗 is the strain tensor.
Neglecting the term for air compressibility, the above equation becomes:
𝑑𝑊 = − 𝑢𝑎 − 𝑢𝑤 𝑛 ∙ 𝑑𝑆𝑟 + 𝜎𝑖𝑗 − 𝑆𝑟 𝑢𝑤 + 1 − 𝑆𝑟 𝑢𝑎 𝛿𝑖𝑗 ∙ 𝑑𝜀𝑖𝑗 (2.9)
Equation 2.9 indicates that the stress tensor 𝜎𝑖𝑗 − 𝑆𝑟 𝑢𝑤 + 1 − 𝑆𝑟 𝑢𝑎 𝛿𝑖𝑗 (or
equally the stress tensor 𝜎𝑖𝑗 − 𝑢𝑎 𝛿𝑖𝑗 + 𝑆𝑟 ∙ 𝑢𝑎 − 𝑢𝑤 𝛿𝑖𝑗 , which is equivalent
to Bishop’s stress written in tensorial notation and assuming that parameter 𝜒 is
equal to 𝑆𝑟 ), is work-conjugate with the strain tensor 𝑑𝜀𝑖𝑗 whereas the matric
suction 𝑢𝑎 − 𝑢𝑤 is work-conjugate with the strain-like variable −𝑛 ∙ 𝑑𝑆𝑟 .
48
Houlsby (1997) suggested incorporating the porosity, 𝑛 , within the stress

variable 𝑢𝑎 − 𝑢𝑤 rather than within the increment 𝑑𝑆𝑟 .
Based on the above, Wheeler et al. (2003) introduced the modified suction 𝑠 ∗ :
𝑠 ∗ = 𝑛 ∙ 𝑢𝑎 − 𝑢𝑤 (2.10)
as an additional variable to Bishop’s stress (for 𝜒 = 𝑆𝑟 ):
𝜎𝑖𝑗∗ = 𝜎𝑖𝑗 − 𝑆𝑟 𝑢𝑤 + 1 − 𝑆𝑟 𝑢𝑎 𝛿𝑖𝑗 (2.11)
and formulated an elasto-plastic constitutive model, which couples the hydraulic

hysteresis and the stress-strain behaviour of unsaturated soils (see also Section
2.8). Although the considered stress variables are more complex than the net
stress, (𝜎𝑖𝑗 − 𝑢𝑎 𝛿𝑖𝑗 ), and the matric suction,(𝑢𝑎 − 𝑢𝑤 ), the authors identified
several advantages associated with their use: Bishop’s stress tensor, 𝜎𝑖𝑗∗ ,
reduces to the effective stress tensor when the soil is saturated, ensuring a
smooth transition from one state to the other; the choice of the stress variables
resulted in the generation of very simple shapes for the yield surfaces in the
isotropic stress space; the behaviour is coupled to the soil-water retention
curve. Nonetheless, the authors also recognised the following complexity:
Bishop’s stress tensor, 𝜎𝑖𝑗∗ , includes the degree of saturation, which is
influenced by the strain-like variable 𝑑𝑆𝑟 and the modified suction, 𝑠 ∗ , depends
on the porosity, which in turn is influenced by the strain variable 𝑑𝜀𝑖𝑗 .
Gallipoli et al. (2003a) employed Bishop’s stress, which they called average
skeleton stress after Jommi (2000), as the first stress variable in the elasto-
plastic model they developed and introduced the additional constitutive variable:
𝜉 = 𝑓 𝑠 ∙ 1 − 𝑆𝑟 (2.12)
𝜉 was defined as the product of the degree of saturation of the air, 1 − 𝑆𝑟 , and
of a function of suction, 𝑓 𝑠 . The former component accounts for the number of
water menisci per unit volume of solid fraction and it is equal to zero when the
soil is saturated. The function 𝑓 𝑠 is associated with the intensity of the
stabilising normal force acting at an inter-particle contact due to a single
meniscus. It may be evaluated using the Fisher (1926) approach and therefore,
49
it may obtain values between 1.0 and 1.5, corresponding to zero and infinite
suction, respectively.
As a consequence of their definition, the stress variables proposed by Gallipoli

et al. (2003a), have the following characteristic: the first stress variable, 𝜎𝑖𝑗∗ ,
relates the effect of fluid pressure within the pores directly to the skeleton
stress; the second stress variable 𝜉 is a measure of the magnitude of the inter-
particle bonding due to water menisci. In this way, the effect of suction on the
mechanical behaviour of unsaturated soils is naturally separated to its two
distinct roles: it influences the skeleton stress through the presence of bulk
water and it increases the normal stabilising force at the inter-particle contacts
through the presence of meniscus water. The elasto-plastic constitutive model
developed based on the above variables is summarised in Section 2.8.
Employment of Bishop’s stress as one of the two stress variables in some of the
most recent approaches has raised questions as to the correct expression for
the parameter 𝜒. Oberg & Sallfors (1997), Jommi (2000), Wheeler et al. (2003),
Gallipoli et al. (2003a), Georgiadis (2003), Sheng et al. (2004), Nuth & Laloui
(2008) and others have substituted 𝜒 by the degree of saturation, 𝑆𝑟 . However,
according to the work of Gray & Schrefler (2001) parameter 𝜒 should not be the
degree of saturation, but the fraction of the solid surface in contact with the
wetting fluid. Other expressions for 𝜒 include some function of the degree of
saturation, 𝑆𝑟 , and the residual degree of saturation, 𝑆𝑟0 (Vanapalli et al., 1996),
or some function of suction (Khalili & Khabbaz, 1998). A detailed overview of
the matter is given by Nuth & Laloui (2008).
Gens et al. (2006) proposed a classification of constitutive models according to

the stress variables adopted in their formulation. The same subject was also
explored by Gens et al. (2008) and more recently by Gens (2010). In the latter
publication, the author noted that the use of two stress variables is common
practice and referred to them as the first and the second constitutive variables
(FCV and SCV). The FCV usually accounts for the overall stress state of the
soil and normally adopts the following form:
FCV = 𝜎𝑖𝑗 − 𝑢𝑎 𝛿𝑖𝑗 + 𝜇1 𝑠, 𝑆𝑟 𝛿𝑖𝑗 (2.13)
50
Table 2-1: Classification of constitutive models for unsaturated soils according to the first constitutive
variable (adapted from Gens, 2010)
Classification of constitutive models after Gens (2010)
Class I: 𝜎𝑖𝑗 − 𝑢𝑎 𝛿𝑖𝑗 ; 𝜇1 = 0.0
Examples Model features

Alonso et al. (1990)
Gens & Alonso (1992)  Stress paths easily represented
Josa et al. (1992)  Constitutive variable is independent of the material
Wheeler & Sivakumar (1993, 1995) state
Cui et al. (1995)  Constitutive variables are independent
Cui & Delage (1996)
 Continuity of variables and behaviour is not ensured
Alonso et al. (1999)
Vaunat et al. (2000a) at the saturated/unsaturated transition
Rampino et al. (2000)  The hydraulic component of the model is
Chiu & Ng (2003) independent of the mechanical component; specific
Georgiadis (2003) effects of the degree of saturation are not included
Sanchez et al. (2005)  An independent function is required to model the
Thu et al. (2007) increase of shear strength with suction
Sheng et al. (2008)
Class II: 𝜎𝑖𝑗 − 𝑢𝑎 𝛿𝑖𝑗 + 𝜇1 𝑠 𝛿𝑖𝑗

 Representation of stress paths is not straightforward
 Constitutive variable is independent of the material
state
Kohgo et al. (1993)  The constitutive variables are linked
Modaressi et al. (1996)  Continuity of variables and behaviour is not ensured
Geiser et al. (2000)
Loret & Khalili (2000, 2002)
at the saturated/unsaturated transition
Laloui et al. (2001)  The hydraulic component of the model is
Sun et al. (2003) independent of the mechanical component; specific
Russell & Khalili (2006) effects of the degree of saturation are not included
Masin & Khalili (2008)  The increase of shear strength with suction is a
result of the constitutive variable definition
 Strength and elastic behaviour can be unified with
an adequate selection of the FCV
Class III: 𝜎𝑖𝑗 − 𝑢𝑎 𝛿𝑖𝑗 + 𝜇1 𝑠, 𝑆𝑟 𝛿𝑖𝑗

 Representation of stress paths is not straightforward
and some times impossible
Bolzon et al. (1996)  Constitutive variable incorporates a state parameter
Jommi (2000)
Wheeler et al. (2003)
𝑆𝑟
Gallipoli et al. (2003a)  The constitutive variables are linked
Sheng et al. (2004)  Continuity automatically ensured at the
Tamagnini (2004) saturated/unsaturated transition
Pereira et al. (2005)  The hydraulic component of the model is closely
Santagiuliana & Schrefler (2006) coupled with the mechanical component
Sun et al. (2007a, 2007c)
 The increase of shear strength with suction is a
Kohler & Hofstetter (2008)
Buscarnera & Nova (2009) result of the constitutive variable definition
 Strength and elastic behaviour can be unified with
an adequate selection of the FCV
51
The SCV is usually associated with the effects of suction changes:
SCV = 𝜇2 𝑠, 𝑆𝑟 (2.14)
Three categories were identified by Gens (2010), depending on the different

assumptions made regarding the function 𝜇1 within the FCV:
 Class I models adopt 𝜇1 = 0.0
 Class II models adopt a function of suction 𝜇1 𝑠
 Class III models adopt a function of both suction and degree of saturation
𝜇1 𝑠, 𝑆𝑟
The general modelling characteristics relevant to each one of the Classes are
summarised in Table 2-1. Also given in the same table are examples of models
falling in each Class, as categorised by Gens (2010). According to the author
the selection of constitutive variables remains a matter of convenience.
2.3 Isotropic compression and yielding
Changes in the volume of unsaturated soils may occur as a consequence of

changes either in the applied suction or in the applied stress. First the
volumetric behaviour due to changes in suction at constant confining isotropic
stress is considered, followed by the volumetric behaviour during isotropic
loading at constant suction. The uniqueness of the isotropic compression line
(ICL), recently investigated by Sivakumar et al. (2010b), is finally discussed.
2.3.1 Changes in suction at constant confining isotropic stress
 wetting at constant confining stress
An unsaturated soil may either collapse or expand upon wetting depending on

the confining stress (Alonso et al., 1987). If the confining stress is sufficiently
low expansion occurs, whereas if it is sufficiently high collapse takes place. A
reversal in the volumetric behaviour during wetting may be exhibited and the
initial expansion may be followed by collapse. This behaviour has been reported
52
by several authors, including Escario & Saez (1973), Josa et al. (1987), Ridley
& Burland (1993). For its modelling Alonso et al. (1990) introduced the concept
of the loading-collapse (LC) curve, which acts as a yield surface in the isotropic
stress plane 𝑝-𝑠, 𝑝 being the mean net stress and 𝑠 being the matric suction.
The LC curve is schematically illustrated in Figure 2-1.
The way the above feature is captured by the LC curve was schematically
illustrated by Gens (2010), as shown in Figure 2-2: on wetting from point A
solely elastic swelling is produced; on wetting from point B initial elastic swelling
is followed by plastic collapse; on wetting from C solely plastic collapse is
generated.
Figure 2-1: The loading-collapse (LC) yield surface (reproduced from Alonso et al., 1990)
Figure 2-2: (a) Stress paths for wetting tests performed under three different applied stresses;
(b) predicted volumetric strains for wetting tests performed under three different applied stresses (after
Gens, 2010)
53
 drying at constant confining stress
On drying from zero suction the soil remains fully saturated and the total volume
change is equal to the pore water volume change, until the air-entry value of
suction is reached and de-saturation occurs. Upon further increase of suction,
the total volume change is smaller than the pore water volume change.
The conceptual model proposed by Toll (1995), shown in Figure 2-3, accounts
for this feature of unsaturated soils. The total volume changes are expressed in
terms of void ratio 𝑒 and the water volume changes in terms of equivalent void
ratio 𝑒𝑤 , which is equal to the volume of water over the volume of solids. For
fully saturated conditions (line AB in the figure), the void ratio line 1 and the
equivalent void ratio line 2 coincide. According to Toll (1995) the AB line is
equivalent to the fully saturated virgin compression line (VCL). It should,
however, be noted that the VCL is defined in the 𝑒-log 𝑝′ plane and therefore,
the x-axis in Figure 2-3 should preferably read log 𝑝 + 𝑠 . After de-saturation
(point B) the two lines deviate from the VCL with the void ratio, 𝑒, exhibiting
slight further changes and the equivalent void ratio, 𝑒𝑤 , decreasing sharply.
Alonso et al. (1990) speculated that increasing the suction beyond a yielding
value, 𝑠0 , which is assumed to be independent of the confining stress and equal
to the maximum previously attained value of suction, elasto-plastic deformations
are produced. Solely elastic strains are generated for lower suction changes. In
the constitutive model they formulated, a yield surface (suction increase surface
– SI) is included to account for this change of behaviour. Nonetheless,
experimental evidence for the existence of such a yield surface is still limited.
Figure 2-3: Volume changes due to drying; conceptual model by Toll (1995)
54
Figure 2-4: Wetting-drying cycles on Boom clay (after Alonso et al., 1995)
 cyclic changes of suction at constant confining stress
Cyclic changes of suction at constant confining stress have been shown by

Alonso et al. (1995) and by Sharma (1998) to produce significant irreversible
strains. Irreversible strains may occur even for suction changes that do not
exceed the previously attained suction level. This behaviour has been linked to
the occurrence of hydraulic hysteresis on the soil-water retention curve (Alonso
et al., 1995).
Alonso et al. (1995) performed controlled-suction oedometer tests on a highly

expansive clay (Boom clay). An example of their results (for vertical net stress
equal to 100.0 kPa) is presented in Figure 2-4 (tension is positive in the figure).
Wetting from a suction of 100.0 MPa (path C1 in the figure), the sample initially
experienced swelling followed by collapse at low suction levels. On subsequent
drying and wetting (paths C2 and C3, respectively) significant irreversibility of the
volumetric strains was demonstrated. During the second drying path C4 and the
third wetting path C5 the irreversible components of the strain were significantly
smaller.
Sharma (1998) tested compacted samples of a bentonite-kaolin mixture under

isotropic stress conditions in a triaxial cell. Keeping the mean net stress
constant and equal to 10.0 kPa, a sample was wetted from 300.0 kPa to 20.0
kPa of suction and was subsequently dried back to 300.0 kPa. The obtained
measurements in terms of specific volume and degree of saturation are
presented in Figure 2-5. Swelling occurred during the wetting path a-b and
compression during the subsequent drying path b-c (Figure 2-5 (a)).
Nonetheless, compression during drying was significantly larger than swelling
55
during wetting. As a consequence distinct paths were followed and a hysteresis

in the specific volume was obtained, similar to the hydraulic hysteresis shown in
Figure 2-5 (b). Similar irreversible reduction in void ratio, induced upon cycles of
suction under constant confining stress, is shown in Figure 2-6, which illustrates
results from another test performed by Sharma (1998).
In order for their constitutive model to reproduce the occurrence of irreversible

changes of volume under wetting-drying cycles, Wheeler et al. (2003) exploited
the suction increase (SI) and the suction decrease (SD) surfaces introduced by
Vaunat et al. (2000a). The model developed by Wheeler et al. (2003) is briefly
presented in Section 2.8.
Figure 2-5: Wetting-drying cycle on compacted bentonite-kaolin (after Sharma, 1998): (a) specific
volume; (b) degree of saturation
Figure 2-6: Wetting-drying cycle on compacted bentonite-kaolin (after Sharma, 1998, reported in
Gallipoli et al., 2003a)
56
2.3.2 Isotropic loading at constant suction
The behaviour of unsaturated soils when subjected to isotropic loading at

constant suction was extensively studied by Wheeler & Sivakumar (1995).
Compacted samples of speswhite kaolin were wetted from their initial suction to
pre-selected suction levels. The samples were subsequently loaded
isotropically, maintaining the pre-selected values of suction constant. An
increase in the isotropic mean net yield stress, 𝑝0 , with increasing suction
levels, as illustrated in Figure 2-7, was reported. The yield points corresponding
to the stress paths in Figure 2-7 (a) are plotted in Figure 2-7 (b). The increase of
yield stress with suction is evident for the unsaturated tests (at zero suction
yield had already occurred during wetting to full saturation, increasing the
isotropic mean net yield stress, 𝑝0 ). Similar behaviour has been reported by
several authors, including Maatouk et al. (1995), Cui & Delage (1996), Dineen
(1997), Rampino et al. (1999, 2000) and more recently Sivakumar et al.
(2010b). In the latter study the LC line obtained was straight, as presented later.
Gens (2010) illustrated the connection between the ICL’s at various suction
levels and the obtained shape of the LC curve, schematically, as shown in
Figure 2-8. The yield points for unsaturated states are located to the right of the
saturated ICL (𝑠 = 0.0 kPa) and their position moves further to the right for
increasing suction, demonstrating that the larger the suction the larger the
stress that can be sustained before yield occurs. Plotting the yield points in
terms of suction, 𝑠, and mean net stress, 𝑝, and joining them together, the LC
curve is obtained.
The slope of the ICL (i.e. the compressibility of unsaturated soils) is also
affected by the suction level. Rampino et al. (2000) measured the variation with
suction of the elastic and the elasto-plastic coefficients of soil compressibility 𝜆
and 𝜅, as illustrated in Figure 2-9. The variation of the coefficient 𝜅 was found to
be small (9%). In constitutive modelling it is commonly assumed that 𝜅 is
independent of suction. On the contrary, the coefficient 𝜆 exhibited a significant
decrease (in the range of 40%) with increasing suction. Nonetheless, other
authors, such as Wheeler & Sivakumar (1995) and more recently Monroy
(2006) and Sivakumar et al. (2010b), reported an increase in 𝜆 with suction.
57
The variation of the coefficient 𝜆 with suction is particularly important and

essentially controls the amount of potential collapse predicted by a constitutive
model. The potential collapse is represented by the vertical distance between
the ICL’s corresponding to unsaturated and saturated conditions, as illustrated
in Figure 2-10 (after Georgiadis (2003)). According to Wheeler & Karube (1996),
examined over a sufficient range of mean net stress, 𝑝, the ICL’s at different
suction levels initially diverge from the fully saturated ICL, at low values of 𝑝,
and subsequently converge to it, at high values of 𝑝 (as illustrated in Figure
2-10). In this way, the amount of potential collapse initially increases and
subsequently decreases with 𝑝, exhibiting a maximum at an intermediate value
of mean net stress (Josa et al., 1992).
The formulation of the Barcelona Basic Model (Alonso et al., 1990) is based on
the assumption that the potential collapse increases continuously with
increasing mean net stress, 𝑝. Nonetheless, Wheeler et al. (2002) demonstrated
that through appropriate adjustment of the relevant parameters (𝑝𝑐 and 𝑟, see
Alonso et al., 1990 and Chapter 3), the constitutive model can also be used for
soils exhibiting decreasing amount of potential collapse with increasing 𝑝.
(a) (b)
Figure 2-7(a) Stress paths and (b) yield points in the 𝑠-𝑝 plane for compacted speswhite kaolin (after
58
Figure 2-8: (a) Idealised scheme of consolidation lines at different suction levels and (b) definition of the
LC curve in the isotropic plane 𝑠-𝑝 (after Gens, 2010)
Figure 2-9: Influence of suction on the elastic and elasto-plastic coefficients of soil compressibility, 𝜅 and
𝜆 (after Rampino et al., 2000)
Figure 2-10: Isotropic compression lines corresponding to full (straight line) and partial (curved line)
saturation and amount of potential collapse predicted (after Georgiadis, 2003)
59
2.3.3 Uniqueness of the isotropic compression line (ICL)
Alonso et al. (1990), when formulating the Barcelona Basic Model (BBM),
suggested the existence of an isotropic compression surface in the 𝑝-𝑠-𝑣 space,
𝑝 being the mean net stress, 𝑠 the suction and 𝑣 the specific volume. The
following equation was employed:
𝑝
𝑣 = 𝑁 𝑠 − 𝜆 𝑠 ln (2.15)
𝑝𝑐
where 𝑝𝑐 is a reference stress state for which 𝑣 = 𝑁 𝑠 and 𝜆 𝑠 is the slope of

the ICL at a specific suction level, 𝑠.
Wheeler & Sivakumar (1995) alternatively suggested the following equation:
𝑝
𝑣 = 𝑁 𝑠 − 𝜆 𝑠 ln (2.16)
𝑝𝑎𝑡𝑚
where 𝑝𝑎𝑡𝑚 is the absolute atmospheric pressure and 𝑁 𝑠 is the intercept at

mean net stress 𝑝 = 𝑝𝑎𝑡𝑚 .
The effect of the initial soil conditions on the position of the isotropic
compression surface in the 𝑣-𝑝-𝑠 stress space was studied by Sivakumar &
Wheeler (2000) and by Sivakumar et al. (2010b).
 tests by Sivakumar & Wheeler (2000)
Sivakumar & Wheeler (2000) tested a series of samples of unsaturated

speswhite kaolin, statically compacted at different pressures and at different
water contents. Two compaction pressures were considered: 400 kPa (Series
1) and 800 kPa (Series 2). The water content for Series 1 and 2 was 25% (4%
dry of optimum). Another series of samples, Series 4, at a water content of
28.5% (close to optimum) was compacted at 500 kPa of pressure so that
samples in Series 2 and 4 were at the same dry density. Samples of Series 3
were dynamically compacted. The method of compaction did not affect the
results and therefore Series 3 is not considered herein. The samples were
wetted from their initial state to pre-selected suction levels and were
subsequently subjected to isotropic loading.
60
The compaction pressure affected the positions of the ICL’s obtained for
different values of suction. Indeed, this behaviour is demonstrated in Figure
2-11. The ICL’s for Series 2 (high compaction effort) are systematically shown
in the figure to lie below the corresponding ICL’s for Series 1 (low compaction
effort). The compaction pressure also affected the mean net yield stress, so that
higher compaction effort (Series 2) was associated with larger yield stresses
upon isotropic loading (different position of the LC curves resulting from
different initial states).
Additionally, the positions of the ICL’s were significantly influenced by the

compaction water content. Figure 2-12 shows the values of specific volume 𝑣
and specific water volume 𝑣𝑤 , measured during isotropic loading for Series 2
and 4. The ICL’s in Series 4 lay substantially below those for Series 2,
indicating that the difference in the compaction water content resulted in the
preparation of effectively different materials.
Sivakumar & Wheeler (2000) demonstrated that the effect of compaction effort
on the soil behaviour could be accounted for by employing an elasto-plastic
model with rotational hardening and changing the initial state according to the
compaction effort. On the contrary, the effect of compaction water content could
not be modelled in a similar way and the authors concluded that samples of the
same soil compacted at different water contents may have to be treated as
fundamentally different materials.
Figure 2-11: Influence of compaction pressure on the isotropic compression behaviour of speswhite
kaolin (after Sivakumar & Wheeler, 2000)
61
Figure 2-12: Influence of compaction water content on the isotropic compression behaviour of speswhite
kaolin (after Sivakumar & Wheeler, 2000)
 tests by Sivakumar et al. (2010b)
In a recent study Sivakumar et al. (2010b) investigated the uniqueness of the

isotropic compression surface. The authors commented that the earlier study of
Sivakumar & Wheeler (2000), as well as the studies of other authors such as
Cui & Delage (1996) and Sharma (1998), were based on testing samples
compacted one-dimensionally into a mould. This process induces anisotropy
and complex soil fabric. The authors, therefore, argued that validation of critical
state type constitutive models for unsaturated soils, such as the BBM, should be
done analysing laboratory data of samples with isotropic stress-strain
properties. For this purpose they tested samples isotropically prepared and
compared the results with those of earlier studies.
Samples of kaolin were prepared at 25% water content (4% dry of optimum),
using isotropic compression. Some were lightly compressed and are identified
as IS(A), while others were heavily compressed and are referred to as IS(B).
The specific volumes obtained by isotropic compression were the same as in
the one-dimensionally prepared samples by Sivakumar & Wheeler (2000),
which are identified as ID(A) and ID(B), respectively. The isotropic samples
were wetted from their initial states under constant mean net stress to pre-
selected values of suction of 300.0, 200.0, 100.0 and 0.0 kPa and were
subsequently loaded isotropically at constant suction.
62
Figure 2-13 illustrates the results of isotropic loading at constant suction for the
isotropically prepared samples. With the exception of a small shift at zero
suction (Figure 2-13 (a)), the isotropic compression curves for IS(B) merged
with the corresponding curves for IS(A) once yielding had occurred. The normal
ICL at each value of suction studied was unique regardless of the initial
condition of the sample. According to the authors, this verified the existence of a
unique isotropic compression surface in the 𝑝-𝑠-𝑣 space for samples with a
previous isotropic history. The uniqueness of the surface, indicated by its
independence on the initial condition of the samples, is probably shown for the
first time, as a unique surface does not seem to exist for samples with induced
anisotropy, such as the ones commonly used in previous studies.
Figure 2-13: Pressure-volume relationships for samples with isotropic stress history: (a) 𝑠 = 0.0 kPa;
(b) 𝑠 = 100.0 kPa; (c) 𝑠 = 200.0 kPa; (d) 𝑠 = 300.0 kPa (after Sivakumar et al., 2010)
63
Figure 2-14: Loading-collapse (LC) yield loci for isotropically prepared (IS) and one-dimensionally (ID)
prepared samples (after Sivakumar et al., 2010)
Figure 2-14 illustrates the LC curves for the isotropically (IS) and one-
dimensionally (ID) prepared samples. The authors noted that, remarkably, the
shape of the LC lines produced by the samples with isotropic stress history was
linear, in contrast to the generally accepted form.
2.4 Critical state
The increase of shear strength with suction has long been acknowledged.
Several attempts have been made to expand the Mohr-Coulomb failure criterion
(e.g. Bishop et al., 1960; Fredlund et al., 1978; Gan & Fredlund, 1996) as well
as the critical state framework to unsaturated conditions, in order to account for
the increase in shear strength. As constitutive modelling of unsaturated soils is
commonly based on the principles of the critical state framework, emphasis is
given in this section to the critical state concept for unsaturated soils.
Toll (1990) proposed that the critical state for unsaturated conditions could be
expressed in terms of the deviatoric stress 𝑞, the mean net stress 𝑝, the suction
𝑠, the specific volume 𝑣 and the degree of saturation, 𝑆𝑟 . Five parameters are
required, 𝑀𝑎 , 𝑀𝑏 , 𝛤𝑎𝑏 , 𝜆𝑎 and 𝜆𝑏 , according to the following equations:
𝑞 = 𝑀𝑎 𝑝 + 𝑀𝑏 𝑠 (2.17)
64
𝑣 = 𝛤𝑎𝑏 − 𝜆𝑎 ln 𝑝 + 𝜆𝑏 ln 𝑠 (2.18)
The author called parameters 𝑀𝑎 and 𝑀𝑏 total stress ratio and suction ratio,
respectively. The intercept 𝛤𝑎𝑏 represents the specific volume when both 𝑝 and
𝑠 equal 1.0. 𝜆𝑎 and 𝜆𝑏 are the slopes of the critical state plane (the term ‘critical
state compressibilities’ was used by Toll & Ong, 2003). Toll (1990) suggested
that the intercept 𝛤𝑎𝑏 may be related to the intercept 𝛤𝑠 which corresponds to
fully saturated conditions, through the degree of saturation, 𝑆𝑟 :
𝛤𝑠 − 1
𝛤𝑎𝑏 = 1 + (2.19)
𝑆𝑟
Additionally, the author showed that for a compacted lateric gravel (Kiunyu
gravel), 𝑀𝑎 , 𝑀𝑏 , 𝜆𝑎 and 𝜆𝑏 did not have unique values but varied as a function
of the degree of saturation, 𝑆𝑟 . Toll & Ong (2003) suggested that the
unsaturated critical state ratios, 𝑀𝑎 and 𝑀𝑏 , can be referenced to the saturated
critical state ratio 𝑀𝑠 , according to the following expressions:
1.0, 𝑆𝑟 ≥ 𝑆𝑟1
𝑀𝑏 𝑘
𝑆𝑟 − 𝑆𝑟2
= , 𝑆𝑟1 > 𝑆𝑟 > 𝑆𝑟2 (2.20)
𝑀𝑠 𝑆𝑟1 − 𝑆𝑟2
0.0, 𝑆𝑟 ≤ 𝑆𝑟2
1.0, 𝑆𝑟 ≥ 𝑆𝑟1
𝑘
𝑀𝑎 𝑀𝑎 𝑆𝑟 − 𝑆𝑟2
𝑀𝑎 − −1 , 𝑆𝑟1 > 𝑆𝑟 > 𝑆𝑟2
= 𝑀𝑠 𝑚𝑎𝑥
𝑀𝑠 𝑚𝑎𝑥
𝑆𝑟1 − 𝑆𝑟2 (2.21)
𝑀𝑠
𝑀𝑎
, 𝑆𝑟 ≤ 𝑆𝑟2
𝑀𝑠 𝑚𝑎𝑥
where 𝑘 is a fitting parameter and may be different for the ratios 𝑀𝑏 𝑀𝑠 and
𝑀𝑎 𝑀𝑠 . 𝑆𝑟1 and 𝑆𝑟2 are the degrees of saturation for which the ratios obtain their
maximum or minimum value according to the above equations.
The same normalised functions for 𝑀𝑏 𝑀𝑠 and 𝑀𝑎 𝑀𝑠 were shown to fit the
experimental data both for a lateric gravel (Kiunyu gravel in Figure 2-15 (a)) and
a residual soil (Jurong soil in Figure 2-16 (a)). Only the saturated critical state
65
ratio 𝑀𝑠 was adjusted to the experimental data for the two soils, while the rest of
the parameters (i.e. 𝑘, 𝑆𝑟1 and 𝑆𝑟2 ) were fixed. A similar observation was made
for the critical state compressibilities 𝜆𝑎 and 𝜆𝑏 , which also vary with the degree
of saturation and may be related to the slope of the critical state line at full
saturation, 𝜆𝑠 . However, the authors provided no equations to substantiate the
above. The fitting of the experimental data for the Kiunyu gravel is shown in
Figure 2-15 (b) and for the Jurong soil in Figure 2-16 (b). The authors
concluded that the form of these functions may be common to a wide range of
soil types.
(a) (b)
compressibilities with the degree of saturation for Kiunyu gravel (after Toll & Ong, 2003)
(a) (b)
compressibilities with the degree of saturation for Jurong soil (after Toll & Ong, 2003)
66
Wheeler & Sivakumar (1995) presented data from a series of controlled-suction

triaxial tests on samples of compacted speswhite kaolin, which they used in the
development of an elasto-plastic critical state framework for unsaturated soils.
The framework was defined in terms of four state variables: the mean net stress
𝑝, the deviatoric stress 𝑞, the suction 𝑠 and the specific volume 𝑣. Included in
the framework was the ICL that was presented in Section 2.3.3 (Equation 2.16),
written in the form: 𝑣 = 𝑓1 𝑝, 𝑠 . The authors postulated the existence of a
critical state hyperline defined by the equations 𝑞 = 𝑓2 𝑝, 𝑠 and 𝑣 = 𝑓3 𝑝, 𝑠 .
On the termination of the shearing tests Wheeler & Sivakumar (1995) reported
that the four state variables 𝑞, 𝑝, 𝑠 and 𝑣 had stabilised (measurements of the
specific water volume 𝑣𝑤 showed that this was still varying at the end of the
experiments). Different paths were followed to critical state, producing a unique
critical state line (CSL) at each value of suction, as shown in Figure 2-17 for 𝑠 of
200.0 kPa.
Figure 2-18 illustrates the critical state values of 𝑞 and 𝑣 plotted against 𝑝 for
constant suction tests performed at different values of suction. Wheeler &
Sivakumar (1995) defined the critical state hyperline as:
𝑞 = 𝑀 𝑠 𝑝+𝜇 𝑠 (2.22)
𝑝
𝑣 = 𝛤 𝑠 − 𝜓 𝑠 ln (2.23)
𝑝𝑎𝑡𝑚
The parameters 𝑀 𝑠 , 𝜇 𝑠 , 𝛤 𝑠 and 𝜓 𝑠 all varied with suction. Fitting these

parameters to the laboratory data the full lines in Figure 2-18 were obtained (for
the values used refer to Wheeler & Sivakumar, 1995).
Combining the critical state and the isotropic compression hyperlines, the
authors proposed an equation of the form of 𝑣 = 𝑓4 𝑝, 𝑞, 𝑠 as a state boundary
hypersurface.
67
Figure 2-17: Test paths for constant suction shear tests conducted at suction, 𝑠, of 200.0 kPa (after
(a) (b)
Figure 2-18: (a) Deviator stress, 𝑞, and (b) specific volume, 𝑣, plotted against mean net stress, 𝑝, at
critical state (after Wheeler & Sivakumar, 1995)
Wheeler & Sivakumar (2000) showed that the CSL is independent of the
compaction pressure. Compaction pressure was shown by Sivakumar &
Wheeler (2000) to affect the positions of the ICL’s at different values of suction,
as already explained in Section 2.3.3. The authors concluded that fabric
differences caused by a change in the compaction pressure are of a type that
can be erased when the samples are sheared to the critical state.
On the contrary, the compaction water content, which was shown by Sivakumar
& Wheeler (2000) to affect the position of the ICL’s, also affected the position of
the CSL’s (Wheeler & Sivakumar, 2000). The differences in the soil fabric
68
caused by a change in the compaction water content were not erased by

shearing to the critical state and continued to influence soil behaviour.
Sivakumar et al. (2010a) performed a number of tests on isotropically prepared

samples which were brought to critical state through different stress paths,
followed at constant suction. The types of paths were fully drained, constant 𝑝
and curved paths with varying slope and are illustrated in Figure 2-19 at suction
of 200.0 kPa. Similar plots for different suction levels confirm the existence of a
unique critical state in the 𝑞-𝑝 and 𝑣-𝑝 planes.
The authors employed Equations 2.22 and 2.23, proposed by Wheeler &
Sivakumar (1995), and evaluated the relevant parameters, 𝑀 𝑠 , 𝜆 𝑠 , 𝜓 𝑠 ,
through linear regression. The variation with suction of the slope 𝑀 𝑠 and of
the intercept 𝜇 𝑠 of the CSL in the 𝑞-𝑝 plane is shown in Figure 2-20 (a). The
effect of suction on 𝑀 𝑠 was relatively small at low suction levels and
increased significantly for 𝑠 of 300.0 kPa. This contradicted the generally
accepted view that 𝑀 𝑠 remains constant with suction (e.g. Alonso et al.,
1990).
The intercept 𝜇 𝑠 initially exhibited a linear increase with suction before

demonstrating a significant decrease at 𝑠 of 300.0 kPa. The authors
commented that a similar increase of 𝜇 𝑠 has been reported in other cases,
such as by Alonso et al. (1990), Maatouk et al. (1995) and Wheeler &
Sivakumar (1995). They added that it is not unusual for 𝜇 𝑠 to exhibit a
reduction as the one illustrated in Figure 2-20 (a), although the reasons for this
are not yet clear. A possible justification is that at high values of suction the
water menisci, and thus shearing resistance, are reduced (Gallipoli et al., 2003a
and Wheeler et al., 2003).
In constitutive modelling the slope of the CSL, 𝜓 𝑠 , in the 𝑣 - 𝑝 plane is

considered equal to the slope of the ICL, 𝜆 𝑠 . The tests by Sivakumar et al.
(2010a) showed that this is so at full saturation and at 𝑠 of 300.0 kPa. For
intermediate values of suction the difference between 𝜓 𝑠 and 𝜆 𝑠 was
significant, as illustrated in Figure 2-20 (b) (𝐵 𝑠 also shown in the latter figure
is the slope of the CSL in the 𝑣𝑤 -𝑝 plane, 𝑣𝑤 being the specific water volume).
69
The authors compared the positions of the CSL’s for isotropically prepared
samples (IS) with the equivalent positions obtained for one-dimensionally
prepared samples (ID). Similar to the ICL’s (see Section 2.3.3) the CSL’s for the
samples with isotropic history lay above the corresponding lines for the one-
dimensionally compressed samples in the 𝑣 - ln 𝑝 plane (Figure 2-21).
Furthermore, the shear strength of the IS samples was notably greater than that
of the ID samples (Figure 2-22).
(a) (b)
Figure 2-19: Conditions at critical state at suction of 200.0 kPa (after Sivakumar et al., 2010a)
(s),
Figure 2-20: Critical state parameters: (a) intercept 𝜇 𝑠 and slope 𝑀 𝑠 of the critical state lines in
terms of 𝑞-𝑝; (b) slope of critical state lines in terms of 𝑣𝑤 -𝑝 (after Sivakumar et al., 2010a)
70
Figure 2-21: Effects of stress induced anisotropy at the critical state in the 𝑣-𝑝 plane: (a) 𝑠 = 0.0 kPa;
(b) 𝑠 = 100.0 kPa; (c) 𝑠 = 200.0 kPa; (d) 𝑠 = 300.0 kPa (after Sivakumar et al., 2010a)
Figure 2-22: Effects of stress induced anisotropy at on the stress-strain curve: (a) fully drained; (b)
constant 𝑝 (after Sivakumar et al., 2010a)
71
Tarantino (2007) suggested that generalisation of the critical state to

unsaturated conditions would require the following three equations:
𝑓1 𝑝, 𝑠, 𝑞, 𝑒, 𝑒𝑤 = 0 (2.24)
𝑓2 𝑝, 𝑠, 𝑞, 𝑒, 𝑒𝑤 = 0 (2.25)
𝑓3 𝑝, 𝑠, 𝑞, 𝑒, 𝑒𝑤 = 0 (2.26)
combing the state variables: mean net stress 𝑝, suction 𝑠, deviator stress 𝑞,
void ratio 𝑒 and water ratio 𝑒𝑤 . The above set of equations would allow the
prediction of the ultimate deviator stress, 𝑞 , both under drained conditions,
where 𝑝 and 𝑠 are known, and under undrained conditions, where 𝑝 and 𝑒𝑤 are
known. Furthermore, the above set of equations implicitly includes the degree of
saturation, 𝑆𝑟 , as only two volumetric variables among 𝑒 , 𝑒𝑤 and 𝑆𝑟 are
independent (i.e. having defined two the third one is also defined).
The author suggested the following equation, introduced by Tarantino &

Tombolato (2005), for the function 𝑓1 :
𝑒𝑤 − 𝑒𝑤𝑚
𝑞 = 𝑀 𝑝 + 𝑠𝑆𝑟𝑀 = 𝑀 𝑝 + 𝑠 (2.27)
𝑒 − 𝑒𝑤𝑚
where 𝑆𝑟𝑀 is the degree of saturation of the macropores and 𝑒𝑤𝑚 is the
microstructural water ratio. In the absence of any direct information about
𝑒𝑤𝑚 and given that its direct measurement may not be easy, this might be
determined as a best-fitting parameter.
The second function 𝑓2 might be given by the following expression for the soil-
water retention curve, introduced by Gallipoli et al. (2003b):
𝑚
𝑒𝑤 1
𝑆𝑟 = = (2.28)
𝑒 1 + 𝜑𝑒 𝜓 𝑠 𝑛
where 𝜑, 𝜓, 𝑚 and 𝑛 are fitting parameters (see also Section 2.6).
72
Finally, the third function 𝑓3 may be obtained from Gallipoli et al. (2003a), who,
using the average skeleton stress (or Bishop’s stress) and the bonding variable
𝜉, as explained in Section 2.2, suggested that:
𝑒 = 𝛤 − 𝜆 ln 𝑝 + 𝑠𝑆𝑟 ∙ 1 − 𝑎 1 − exp 𝑏𝜉 (2.29)
where 𝛤 and 𝜆 are the intercept and the slope of CSL at full saturation and 𝑎
and 𝑏 are two extra parameters as further explained in Section 2.8. Tarantino
(2007) noted that substituting 𝑆𝑟 = 𝑒𝑤 𝑒 and 𝜉 = 𝑓 𝑠 1 − 𝑆𝑟 into the above
equation, this can be rewritten as follows:
𝑒𝑤 𝑒𝑤
𝑒 = 𝛤 − 𝜆 ln 𝑝 + 𝑠 ∙ 1 − 𝑎 1 − exp 𝑏𝑓 𝑠 1− (2.30)
𝑒 𝑒
With the appropriate adjustments the critical state framework has been
extensively employed in the constitutive modelling of unsaturated soils. Some
examples of such models are given in Section 2.8.
2.5 Soil-water retention curve (SWRC)
The degree of saturation, 𝑆𝑟 , generally reduces with increasing suction, 𝑠 ,

following the so called soil – water retention curve (SWRC). The retention curve
has the typical s – shape shown schematically in Figure 2-23. Although it may
be plotted in terms of the volumetric water content, the SWRC in constitutive
modelling is commonly defined in terms of the degree of saturation, 𝑆𝑟 , and this
definition has been employed throughout this thesis.
In saturated conditions, the whole volume of voids is occupied by water and the
degree of saturation, 𝑆𝑟 , is 1. It is common for a soil to withstand a significant
amount of suction and still maintain a degree of saturation, 𝑆𝑟 , of unity.
On drying the soil further, the largest of the pores empty of water and fill with
air. The corresponding value of suction is called the air – entry value of suction,
𝑠𝑎𝑖𝑟 . The switch from saturated to unsaturated conditions is associated with the
air – entry value of suction, 𝑠𝑎𝑖𝑟 .
73
As suction increases with further drying, air exists in a continuous form and
water retreats to smaller voids, loosing its continuity. At very high values of
suction water is present only in the form of menisci at the interparticle contacts.
Any further increase in suction is attributed to the meniscus water and causes
only an insignificant further decrease of the degree of saturation, 𝑆𝑟 , which is
assumed to have reached its residual value.
A subsequent decrease of suction to zero due to wetting brings the soil back
towards saturated conditions. Nevertheless, drying and wetting are not
reversible processes and the wetting path followed to saturation lies beneath
the drying path, demonstrating the hydraulic hysteresis shown in Figure 2-24. It
is possible for air to be present in the form of occluded bubbles even when the
soil is wetted to zero suction (𝑆𝑟 ≠ 1.0 at 𝑠 = 0 kPa).
The retention state of a soil can exist anywhere between the above mentioned
drying and wetting paths (e.g. point A in Figure 2-24). These two paths are
followed only when a reconstituted soil sample is dried from slurry to residual
conditions and is subsequently wetted to full saturation and are, therefore,
referred to as primary paths. The primary paths bound an infinite number of
scanning paths, as illustrated in Figure 2-24. On wetting from an initial retention
point A, the scanning path followed rejoins the primary wetting path at lower
values of suction. The reverse behaviour is observed during drying; the soil
follows a scanning drying path which converges onto the primary one at larger
values of suction.
Even in soils where the retention relationship exhibits no hysteresis, the SWRC
is not unique. As explained by Gallipoli et al. (2003b) the retention relationship
in a deformable soil is affected by the variation of void ratio. Changing the
dimension of the voids and of passageways between them, affects the retention
curve.
Gallipoli et al. (2003b) employed the two examples shown in Figure 2-25, to
illustrate this feature. The degree of saturation is shown to change irreversibly
under an isotropic loading-unloading test involving irreversible changes of void
ratio (Figure 2-25 (a)) under constant suction: during loading plastic reduction in
void ratio occurred and the degree of saturation, 𝑆𝑟 , increased significantly;
74
during subsequent elastic unloading, the amount of swelling was small and the
degree of saturation, 𝑆𝑟 , changed insignificantly. Furthermore, the degree of
saturation, 𝑆𝑟 , increased considerably under the application of deviator stress, 𝑞
(Figure 2-25 (b)): maintaining constant suction and constant mean stress, 𝑝, the
change in 𝑆𝑟 was attributed to the changes in void ratio induced during
shearing.
A direct consequence of the non-uniqueness of the SWRC is that samples of

the same soil can exhibit significantly different degrees of saturation under the
same value of suction, depending on their drying-wetting and loading-unloading
history. The mechanical behaviour of unsaturated soils, however, is intrinsically
related to the degree of saturation (see Section 2.7) and therefore, samples with
different degree of saturation but in all other respects identical subjected to
similar stress states can display different mechanical responses. Furthermore,
the degree of saturation is indicative of the water storage and contributes to the
evolution of water flow within unsaturated soils. The necessity of modelling the
non-uniqueness of the soil-water retention curve becomes obvious.
Figure 2-23: Typical shape of the soil-water retention curve (SWRC) in terms of degree of saturation, 𝑆𝑟
75
Figure 2-24: Hydraulic hysteresis in the soil-water retention curve (SWRC)
(a) (b)
Figure 2-25: (a) Isotropic loading and unloading test at constant suction on compacted speswhite kaolin;
(b) Triaxial shear test at constant suction and constant mean stress on compacted speswhite kaolin
(reported in Gallipoli et al., 2003b)
2.6 Modelling the soil-water retention curve (SWRC)
Various relationships have been proposed for the modelling of the soil-water
retention curve (SWRC), such as those by Van Genuchten (1980) and Fredlund
& Xing (1994). These relationships may be used in combination with existing
constitutive models in order to predict the degree of saturation, 𝑆𝑟 . In the
present section some of the most recent expressions for the SWRC are briefly
76
presented. The expressions presented are either capable of predicting the

variation of 𝑆𝑟 due to changes in specific volume, 𝑣 (additionally to the changes
in suction, 𝑠) or they account for the hydraulic hysteresis.
 The Gallipoli et al. (2003b) model
Making the general hypothesis that, in the absence of hydraulic hysteresis,

there is a unique relationship between the degree of saturation 𝑆𝑟 , the suction 𝑠
and the specific volume 𝑣 , Gallipoli et al. (2003b) proposed the following
expression to model the variation of 𝑆𝑟 in a deformable soil:
𝑚
1
𝑆𝑟 = 𝜓𝑠 𝑛
(2.31)
1+ 𝜑 𝑣−1
where 𝑚, 𝑛, 𝜑 and 𝜓 are soil constants. According to this equation, the degree
of saturation is 1.0 at zero suction and tends to zero at infinite suction. To obtain
Equation 2.31, Gallipoli et al. (2003b) modified the Van Genuchten (1980)
expression, written in terms of degree of saturation, 𝑆𝑟 :
𝑚
1
𝑆𝑟 = 𝑛
(2.32)
1 + 𝑎𝑠
𝜓
by replacing 𝑎 with the function 𝜑 𝑣 − 1 . In this way, the SWRC is modelled
as a 3-dimensional surface in the 𝑆𝑟 -𝑠-𝑣 space, as illustrated in Figure 2-26.
Figure 2-27 shows the same information plotted in the 𝑆𝑟 -𝑠 plane for different
values of specific volume 𝑣 . The authors commented that the variation of
specific volume and the variation of suction may both have a significant effect
on the degree of saturation.
When combined with the elasto-plastic stress-strain model of Wheeler &

Sivakumar (1995), Equation 2.31 was able to predict accurately the
experimental results already presented in Figure 2-25, as shown in Figure 2-28
(also shown in the figure is the prediction produced by the state surface
proposed by Lloret & Alonso, 1985).
77
Figure 2-26: Three-dimensional plot of 𝑆𝑟 against 𝑠 and 𝑣 predicted by the Gallipoli et al. (2003b)
expression (after Gallipoli et al., 2003b)
Figure 2-27: Predicted soil-water retention curves at different values of specific volume 𝑣 (after Gallipoli
et al., 2003b)
78
(a) (b)
Figure 2-28: Experimental and predicted variation of 𝑆𝑟 (a) Isotropic loading and unloading test at
constant suction; (b) Triaxial shear test at constant suction and constant mean stress (after Gallipoli et
al., 2003b)
 The Tarantino (2009) model
Tarantino (2009) proposed a similar expression in order to include the effect of

void ratio, 𝑒, in the expression for the primary drying and primary wetting:
1 𝑛 −𝑏 𝑛
𝑒𝑤 𝑒 𝑏
𝑆𝑟 = = 1+ 𝑠 (2.33)
𝑒 𝑎
where 𝑒𝑤 is the water ratio and 𝑎, 𝑏 and 𝑛 are parameters which may obtain
different values for drying and for wetting. According to the author this
expression is mathematically equivalent to the one proposed by Gallipoli et al.
(2003b) (Equation 2.31) but requires one less parameter. Parameters 𝑎 and 𝑏
have a physical meaning: they are associated with the intercept and slope of
the straight line interpolating experimental data in the ln(𝑠)–ln(𝑒𝑤 ) plane, at
high suctions (Figure 2-29). Therefore, 𝑛 is the only fitting parameter.
Introducing the concept of normalised suction 𝑠 ∗ :
1
𝑠∗ = 𝑒 𝑏 𝑠 (2.34)
Tarantino (2009) reduced the surface given by Equation 2.33 to the 2-

dimensional primary curve:
79
𝑛 −𝑏 𝑛
1
𝑒𝑤 1 𝑏
𝑆𝑟 = = 1+ 𝑠∗ (2.35)
𝑒 𝑎
Despite being called hysteretic model, no expression was defined for the
scanning paths. The model essentially consists of an expression for the primary
paths which may be fitted to laboratory data for primary drying employing one
set of parameters and to laboratory data for primary wetting employing a
different set of parameters.
Figure 2-29: Retention curve at zero vertical stress plotted in terms of water ratio 𝑒𝑤 (after Tarantino,
2009)
 The Li (2005) hysteretic model
Li (2005) proposed an expression for the scanning paths followed by a soil

exhibiting hydraulic hysteresis in the SWRC. The advantage of the proposed
expression is that it may be used in combination with existing expressions for
the primary paths, which serve as bounds of the scanning paths.
In Li’s (2005) model, the hydraulic state is prescribed by the suction 𝑠, the
degree of saturation, 𝑆𝑟 , and the suction at the projection centre 𝛼 which is
illustrated in Figure 2-30 (a). The projection centre is the point at which the last
reversal from drying to wetting or vice versa took place. One fixed projection
centre corresponds to each scanning path. Each state is associated with a
degree of saturation, 𝑆𝑟 , and three suction values: 𝛼, 𝑠 and 𝑠. 𝑠 is the suction of
the image of the current point on the congruent primary path, as illustrated in
Figure 2-30 (a). The image of 𝛼 is denoted as 𝑠0 in the same figure.
80
To facilitate modelling in a semi-logarithmic scale, Li (2005) defined the log-

scaled stress variable 𝑠 ∗ = ln 𝑠 (Figure 2-30 (b)). Therefore, 𝛼 ∗ = ln 𝛼 , 𝑠0 ∗ =
ln 𝑠0 and 𝑠 ∗ = ln 𝑠.
Upon a change in the degree of saturation, 𝑑𝑆𝑟 , a change in suction 𝑑𝑠 ∗ and a

change in image suction 𝑑𝑠 ∗ occur. The changes 𝑑𝑠 ∗ and 𝑑𝑠 ∗ are generally
different, with 𝑑𝑠 ∗ ≥ 𝑑𝑠 ∗ . For each scanning path 𝑠 ∗ varies between 𝛼 ∗ , which is
fixed, and 𝑠 ∗ , which also varies. The difference between 𝑑𝑠 ∗ and 𝑑𝑠 ∗ depends
on the position of 𝑠 ∗ relative to 𝛼 ∗ and 𝑠 ∗ :
∗ ∗
𝑑𝑠 ∗
when 𝑠 → 𝛼 then →∞ (2.36)
𝑑𝑠 ∗
∗ ∗
𝑑𝑠 ∗
when 𝑠 → 𝑠 then → 1.0 (2.37)
𝑑𝑠 ∗
A simple equation which complies with the above observation is:
𝛽
∗
𝑠∗ − 𝛼 ∗
𝑑𝑠 = ∗ 𝑑𝑠 ∗ (2.38)
𝑠 − 𝛼∗
where 𝛽 is a material parameter. The above equation shows that:
∗ ∗
𝑑𝑠 ∗ 𝑑𝑠 ∗ 𝑑𝑠 ∗
when 𝑠 = 𝛼 then = ∙ =∞ (2.39)
𝑑𝑆𝑟 𝑑𝑠 ∗ 𝑑𝑆𝑟
𝑑𝑠 ∗ 𝑑𝑠 ∗
when 𝑠 ∗ ≈ 𝑠 ∗ then ≈ (2.40)
𝑑𝑆𝑟 𝑑𝑆𝑟
So, the scanning path always starts with a horizontal tangential direction and
approaches the congruent primary path asymptotically. Integrating Equation
2.38, the following expression for the scanning paths is obtained:
1
𝑠∗ = 𝛼 ∗ ± 𝑠 ∗ − 𝛼 ∗ 𝛽 +1
− 𝑠0 ∗ − 𝛼 ∗ 𝛽 +1 𝛽 +1 (2.41)
where + is used during drying and – during wetting. The actual suction may
then be calculated as:
81
∗
𝑠 = e𝑠 (2.42)
The author described the implementation of the model, where the sign of 𝛥𝑆𝑟
needs to be checked at each increment. Whenever the sign of 𝛥𝑆𝑟 changes, 𝛼 ∗
and 𝑠0 ∗ need to be updated correspondingly. In this way the change in suction
in response to a change in the degree of saturation may be calculated. If the
change in the degree of saturation in response to change in suction is instead
required, an iterative procedure must be implemented. It should be noted that in
coupled consolidation finite element analysis increments of suction are applied
and the response in terms of degree of saturation is predicted, rather than the
other way around.
Figure 2-31 illustrates multiple scanning paths reproduced by the model. For
their reproduction the author employed the Fredlund & Xing (1994) expression
for the primary paths and adopted two values for parameter 𝛽: 2.0 in Figure
2-31 (a) and 4.0 in Figure 2-31 (b). Multiple drying-wetting loops were induced
under otherwise identical conditions. Parameter 𝛽 affects the size of the
hysteretic loops generated.
Li (2005) proposed several optional refinements for Equation 2.38: the initial
slope of the scanning path may be adjusted to a pre-selected value by
introducing an additional material parameter; the material parameters may be a
function of some internal/external state variable; the primary curves may be
made dependent on the stress state or the soil density. Nevertheless, no
comparison of the proposed scanning paths to experimental data was
presented.
The expression proposed by Li (2005) may be combined with the primary

curves proposed by Gallipoli et al. (2003b) (or similarly by Tarantino, 2009).
Different model parameters could be fitted to reproduce the primary drying and
wetting curves. These would be then used to calculate the suction at the image
points as a function of the degree of saturation, 𝑆𝑟 , and the specific volume, 𝑣
(or the void ratio, 𝑒). Since the image suctions would depend on the specific
volume, the actual suction computed by Li’s (2005) expression would also
depend on specific volume. Given that 𝑣 might change during an increment of
82
the analysis, the application of the expression proposed by Li (2005) might

require the use of a numerical integration technique.
It should be noted as a comment to the above procedure that the expression

employed for the definition of the primary paths should preferably reach a
minimum 𝑆𝑟 (i.e. the residual degree of saturation) at a finite value of 𝑠 .
Otherwise, a continuous smooth transition from the primary drying path to the
primary wetting path cannot be obtained i.e. a jump occurs similar to the one
shown in Figure 2-30 (a) at suction equal to 105 kPa. The expressions by Van
Genuchten (1980), Fredlund & Xing (1994), Gallipoli et al. (2003b) or Tarantino
(2009) do not reach the residual degree of saturation at a finite value of suction
and alterations may be found necessary.
(a) (b)
Figure 2-30: Definition of hydraulic state variables in (a) 𝑆𝑟 -𝑠 plane; (b) 𝑆𝑟 -𝑠 ∗ plane (after Li, 2005)
(a) (b)
Figure 2-31: Predicted scanning paths for (a) 𝛽 = 2.0; (b) 𝛽 = 4.0 (after Li, 2005)
 The Pedroso & Williams (2010) hysteretic model
Pedroso & Williams (2010) proposed a hysteretic model where two reference
curves (similar to the primary curves) bound the hysteretic scanning paths. The
83
overall SWRC is defined based on the vertical distance between the current
state and the corresponding state on the reference curves.
The reference curves are enveloped by three straight lines, 𝜆0 , 𝜆1 and 𝜆2 , as

illustrated in Figure 2-32 (a). For simplicity, it is assumed that 𝜆1 = 𝜆𝑑 = 𝜆𝑤 and
that 𝜆0 = 𝜆2 = 0. To fix these three lines the points 𝑥𝑅𝑑 , 1 , 𝑥𝑅𝑤 , 0 and 0, 𝑦𝑅
shown in Figure 2-32 (b) are employed. Smooth transitions from one line to the
other are generated through the introduction of three coefficients, 𝛽𝑑 , 𝛽𝑤 and
𝛽1 , later explained. The SWRC is formed in the 𝑦 = 𝑆𝑟 and 𝑥 = ln 𝑠 + 1 plane.
The following expression for the drying reference curve is employed:
1 𝑑
𝑦𝑑 𝑥 = −𝜆𝑑 𝑥 + ln 𝑐3𝑑 + 𝑐2𝑑 e𝑐1 𝑥 (2.43)
𝛽𝑑
where 𝑐1𝑑 , 𝑐2𝑑 and 𝑐3𝑑 are constants which depend on the model parameters:
𝑐1𝑑 = 𝛽𝑑 𝜆𝑑 (2.44)
𝑐2𝑑 = e𝛽𝑑 𝑦𝑅 (2.45)
𝑑
𝑐3𝑑 = e𝛽𝑑 𝑦0 +𝜆 𝑑 𝑥 𝑅𝑑
− 𝑐2𝑑 e𝑐1 𝑥 𝑅𝑑 (2.46)
where 𝑦0 = 𝑆𝑟 = 1.0. 𝛽𝑑 is a model parameter which controls the ‘speed’ of the

transition from 𝜆1 to 𝜆2 , shown in Figure 2-32 (a) (𝜆1 = 𝜆𝑑 and 𝜆2 = 0).
Similarly, for the wetting reference curve the following expression is employed:
1 𝑤
𝑦𝑤 𝑥 = −𝜆𝑑 𝑥 + ln 𝑐3𝑤 + 𝑐2𝑤 e𝑐1 𝑥 (2.47)
𝛽𝑤
where 𝑐1𝑑 , 𝑐2𝑑 and 𝑐3𝑑 are constants which depend on the model parameters:
𝑐1𝑤 = 𝛽𝑤 𝜆𝑑 (2.48)
𝑐2𝑤 = e𝛽𝑤 𝑦0 (2.49)
𝑤
𝑐3𝑤 = e𝛽𝑤 𝜆 𝑑 𝑥 𝑅𝑤 − 𝑐2𝑤 e𝑐1 𝑥 𝑅𝑤 (2.50)
84
where 𝑦0 = 𝑆𝑟 = 1.0. 𝛽𝑤 is a model parameter which controls the ‘speed’ of the

transition from 𝜆1 to 𝜆0 , shown in Figure 2-32 (a) (𝜆1 = 𝜆𝑑 and 𝜆0 = 0).
For the determination of the SWRC the vertical distance to the reference curve,
shown in Figure 2-33, is required. This is:
𝐷 = 𝑦𝑑 𝑥 − 𝑦 (2.51)
on drying and :
𝐷 = 𝑦𝑤 𝑥 − 𝑦 (2.52)
on wetting.
The tangent inclination 𝜆 for the SWRC is a priori adopted and is equal to:
𝜆 = 𝜆𝑑 ∙ e−𝛽𝑤 𝐷 (2.53)
where:
𝜆𝑑 = 𝜆𝑑 1 − e−𝛽𝑤 𝐷𝑑 (2.54)
𝐷𝑑 = 𝑦 − 𝑦𝑅 (2.55)
on drying and :
𝜆 = 𝜆𝑤 ∙ e−𝛽𝑤 𝐷 (2.56)
where:
𝜆𝑤 = 𝜆𝑑 1 − e−𝛽𝑤 𝐷𝑤 (2.57)
𝐷𝑤 = 𝑦0 − 𝑦 (2.58)
on wetting. Note that in the above equations the parameter 𝛽𝑤 may be

substituted by the parameters 𝛽2 and 𝛽1 , for drying and wetting respectively,
which control the ‘speed’ of transition from 0 to 𝜆𝑑 and from 0 to 𝜆𝑤 (see also
Figure 2-33).
85
The SWRC 𝑦 𝑥 can then be calculated from:
𝑑𝑦
= −𝜆 (2.59)
𝑑𝑥
Equation 2.59 cannot be integrated analytically and consequently a solution has

to be found numerically. Increments of the degree of saturation or increments of
suction can be alternatively applied. Consideration of whether the soil
undergoes drying or wetting is required: negative increments of 𝑆𝑟 or positive
increments of 𝑠 signify drying whereas positive increments of 𝑆𝑟 or negative
increments of 𝑠 signify wetting. The complete algorithm is presented in Figure
2-34 (note that not all the terms quoted are discussed here, i.e. for the SWCC
transition 𝛽𝑤 – for more details refer to Pedroso & Williams 2010).
The model was successfully employed in the reproduction of laboratory

measurements of the retention behaviour of Hostun sand, as shown in Figure
2-35.
(a)
(b)
Figure 2-32: Definition of reference curves (after Pedroso & Williams, 2010)
86
Figure 2-33: Definition of vertical distance 𝐷 (after Pedroso & Williams, 2010)
(a) (b)
Figure 2-34: Algorithm employed for the Pedroso & Williams (2010) model
Figure 2-35: Validation of the Pedroso & Williams (2010) model against experimental data from tests on
Hostun sand (after Pedroso & Williams, 2010)
87
The SWRC models presented above can be combined with existing elasto-
plastic constitutive models (such as the BBM) to improve simulation of specific
aspects of soil behaviour: the Gallipoli et al. (2003b) and the Tarantino (2009)
models incorporate the effect of void ratio variation on the SWRC whereas the
Li (2005) and the Pedroso & Williams (2010) models account for hydraulic
hysteresis.
Small modifications to the Gallipoli et al. (2003b) model are proposed in

Chapter 5: a finite residual degree of saturation (this is zero in the original
model) and the air-entry value of suction, 𝑠𝑎𝑖𝑟 , are introduced in the expression.
Similar refinements may be applied to the Tarantino (2009) model.
The hysteretic SWRC models proposed by Li (2005) and Pedroso & Williams
(2010) predict the slope of the scanning paths based on the slope of the
congruent primary paths. The scanning paths are then integrated from their
slope. The scanning paths approach the primary paths (Li, 2005), or the
reference paths (Pedroso & Williams, 2010), asymptotically without ever
converging to them. Although the approach by Li (2005) is simpler, a fitting
parameter 𝛽 is required, which controls the size of the loops generated.
Calibration of this parameter was not explained and the model predictions were
not compared to laboratory data. The Pedroso & Williams (2010) model is
slightly more complicated, requiring in total 7 parameters to be calibrated. A
completely different approach is proposed in Chapter 5: the scanning paths
converge to the primary paths, which are then followed upon further suction
changes while no fitting parameters are required for the simulation of the
scanning paths.
2.7 Coupling of the mechanical and the hydraulic behaviour
The mechanical and hydraulic components of the behaviour of unsaturated soils

are coupled and the SWRC affects and is affected by the mechanical
behaviour: changes in the degree of saturation (or the volumetric water content)
induce mechanical effects, while soil deformations induce changes in the
degree of saturation (or in the volumetric water content).
88
Wheeler et al. (2003) summarised the following typical examples, where the
coupling of the hydraulic and mechanical components is demonstrated:
 Volumetric strains affect the position of the SWRC, as discussed in the

previous two sections.
 The transition between saturated and unsaturated conditions may occur

at different values of suction on drying and on wetting. Until the air-entry
value of suction, 𝑠𝑎𝑖𝑟 , is reached upon drying, the effective stress
principle is sufficient for the reproduction of the mechanical behaviour.
On wetting, a soil might remain unsaturated down to a value of suction
which is lower than 𝑠𝑎𝑖𝑟 , requiring two stress variables for its modelling.
 Cyclic changes in suction may affect the subsequent soil behaviour

during isotropic loading. To illustrate this feature the authors used the
laboratory data by Sharma (1998), shown in Figure 2-36. One isotropic
loading-unloading cycle, a-b-c, was performed under constant suction of
200.0 kPa. Subsequently, a wetting-drying cycle, c-d-e, under constant
applied confining stress was carried out. During this stage the degree of
saturation increased (Figure 2-36 (b)) but the specific volume was not
affected (Figure 2-36 (a)). During the second loading-unloading cycle, e-
f-g, the soil yielded on the e-f path at a mean net stress which was lower
than the value of 100.0 kPa previously applied, demonstrating that the
yield point during isotropic loading at a given value of suction is reduced
by a preceding cyclic change of suction.
 The irreversible volumetric changes observed upon cyclic changes in

suction at constant applied confining stress are linked to the occurrence
of the hydraulic hysteresis, as presented in Section 2.3.1. To illustrate
this feature Wheeler et al. (2003) presented the laboratory data of
Sharma (1998), shown in Figure 2-5. Wheeler et al. (2003) noted that
yielding is suggested by the shape of the drying path b-c in Figure 2-5
(a). According to the authors, yielding on drying was also suggested by
the paths illustrated in Figure 2-37.
89
The BBM (Alonso et al., 1990) and other elasto-plastic constitutive models
formulated in terms of net stress and suction are unable to reproduce the
behaviour described above, as the degree of saturation, 𝑆𝑟 , and its influence on
soil behaviour are ignored. For the hydraulic component to be directly included
within a constitutive model, the degree of saturation, 𝑆𝑟 , needs to be
incorporated in the formulation of the mechanical component. This may be done
by including 𝑆𝑟 within the stress (or constitutive) variables, as in the models
summarised in the following section. Alternatively, 𝑆𝑟 may be included in
specific features of the constitutive model, such as the increase of apparent
cohesion with suction and the soil compressibility with suction. This second
approach was followed in the constitutive models by Georgiadis (2003),
presented in Chapter 3, which are available in ICFEP and were used in the
current research project (the increase of apparent cohesion with suction
depends on the degree of saturation, obtained from the SWRC). In the course
of this project, the degree of saturation, 𝑆𝑟 , was additionally introduced into the
expression for the soil compressibility with suction (Chapter 5).
Figure 2-36: Influence of wetting-drying cycle on subsequent behaviour during isotropic loading,
bentonite-kaolin sample (after Sharma, 1998, deducted from Wheeler et al., 2003): (a) specific volume;
(b) degree of saturation
90
Figure 2-37: Wetting-drying cycle on compacted kaolin performed under isotropic stress state, with
𝑝 − 𝑢𝑎 = 50.0 kPa (reported in Wheeler et al., 2003)
2.8 Constitutive modelling
The Barcelona Basic Model (BBM), developed by Alonso et al. (1990), was
probably the first attempt to formulate an elasto-plastic constitutive model
consistent with the principles of critical state (Schofield & Wroth, 1968) but
extended to unsaturated conditions. Based on the BBM numerous constitutive
models were proposed, such as those by Josa et al. (1992), Wheeler &
Sivakumar (1995), Cui et al. (1995), Cui & Delage (1996), Georgiadis (2003).
The modifications made by Georgiadis (2003) led to the development of two

constitutive models (further explained by Georgiadis et al., 2005), which are
available in ICFEP and are presented in the subsequent chapter. Therefore, the
presentation of the BBM and of its numerous modifications – already
extensively discussed in previous PhD theses written at Imperial College
London (e.g. Cunningham, 2000; Georgiadis, 2003; Jotisankasa, 2005; Monroy,
2006) – was not thought to be fundamental and the current section is focused
on constitutive models recently developed to account for the coupled effect of
the hydraulic and mechanical components of unsaturated soils.
Three constitutive models are considered herein: the Gallipoli et al. (2003a)
and the Wheeler et al. (2003) models, which have been included in the PhD
theses of Jotisankasa (2005) and Monroy (2006) and are, therefore, briefly
presented herein, and the more recently developed Sheng et al. (2008) model.
91
2.8.1 The Gallipoli et al. (2003a) model
The stress variables adopted by Gallipoli et al. (2003a) were Bishop’s stress,
given by Equation 2.11, and the bonding variable 𝜉, given by Equation 2.12.
The authors used the term ‘average skeleton stress’, introduced by Jommi
(2000), for the first stress variable, rather than the term ‘Bishop’s stress’, and
the notation 𝜎𝑖𝑗′′ rather than 𝜎𝑖𝑗∗ employed in Equation 2.11. The same term and
notation are used herein when referred to their work.
Gallipoli et al. (2003a) reasoned that since the variable 𝜉 is representative of the
inter-particle forces, it may be related to the additional void ratio that a soil is
capable of sustaining under unsaturated conditions. Therefore, they associated
𝜉 to the ratio 𝑒/𝑒𝑠 , where 𝑒 is the void ratio at a specific applied average
skeleton stress and suction and 𝑒𝑠 is the void ratio in the saturated virgin
compression line at the same applied average skeleton stress. Furthermore,
they showed that the ratio 𝑒/𝑒𝑠 is a unique function of the bonding factor 𝜉.
They employed the following expression:
𝑒
= 1 − 𝑎 1 − eb𝜉 (2.60)
𝑒𝑠
where 𝑎 and b are fitting parameters. The shape of the above expression is
shown in Figure 2-38. At full saturation, 𝜉 becomes 0.0 and 𝑒/𝑒𝑠 becomes 1.0.
The isotropic compression state surface was defined as the product of two
factors: 𝑒 𝑒𝑠 𝜉 and 𝑒𝑠 𝑝′′ . The first factor is given by Equation 2.60 and the
second is the equation of the saturated isotropic compression line relating the
variation of 𝑒𝑠 to the isotropic average skeleton stress:
𝑒𝑠 𝑝′′ . = 𝑁 − 𝜆 ln 𝑝′′ (2.61)
where 𝑁 and 𝜆 are the intercept at 𝑝′′ = 1.0 kPa and the slope of the ICL at full
saturation. Therefore, the isotropic compression state surface is expressed as:
𝑒
𝑒 𝑝′′ , 𝜉 = 𝜉 ∙ 𝑒𝑠 𝑝′′ (2.62)
𝑒𝑠
92
The isotropic compression state surface defined by the above equation acts as
a limiting surface in the 𝑒 - 𝑝′′ - 𝜉 space. This surface bounds attainable soil
states, seperating them from non-attainable soil states. Elastic behaviour is
predicted within the surface and elasto-plastic (irreversible) changes of void
ratio develop when the state lies on the surface. The elastic changes of void
ratio are given by:
𝑒
𝑝𝑓′′
𝛥𝑒 = −𝜅 ln ′′ (2.63)
𝑝𝑖
where 𝜅 is the elastic swelling index and 𝑝𝑖′′ and 𝑝𝑓′′ are the initial and final
values of 𝑝′′ , respectively.
Figure 2-39 (a) illsutrates three examples of ICL’s which lie on the isotropic
compression state surface and which correspond to constant values of 𝜉. A
stress path from point 1 to point 2 in the figure produces elastic changes of void
ratio, given by Equation 2.63. The soil states 1 and 2 belong to the state surface
given by Equation 2.62. Combining the above two equations, Gallipoli et al.
(2003a) obtained the following expression for the yield locus (i.e. LC curve) in
the 𝜉 -ln 𝑝′′ plane (corrected from the original paper as reported in Erratum,
2003):
𝑒
𝜆−𝜅 𝜉 −1 𝑁
𝑒
ln 𝑝0′′ 𝜉 = 𝑒 ln 𝑝0′′ 0 + 𝑒𝑠 (2.64)
𝜉 𝜆−𝜅
𝑒𝑠 𝑒𝑠 𝜉 𝜆 − 𝜅
where 𝑝0′′ 𝜉 is the yield value of the isotropic average skeleton stress (i.e. on
the unsaturated isotropic compression line) corresponding to 𝜉 and 𝑝0′′ 0 the
yield value of the isotropic average skeleton stress during isotropic compression
of a saturated sample. The latter is the hardening parameter of the model. The
yield locus in the 𝜉-ln 𝑝′′ plane is shown in Figure 2-39 (b) together with points 1
and 2.
Also shown in the latter figure is an expanded yield locus which is considered to
be the result of volumetric hardening. The irreversible change of void ratio 𝛥𝑒 𝑝
is associated with the expansion of the yield locus from its initial position,
93
identified by the average skeleton stress 𝑝0′′ 0 𝑖 , to a final position, identified by

the the average skeleton stress 𝑝0′′ 0 𝑓 , and can be calculated from the
following equation:
𝑝0′′ 0 𝑓
𝛥𝑒 𝑝 = −(𝜆 − 𝜅) ln (2.65)
𝑝0′′ 0 𝑖
The model proposed by Gallipoli et al. (2003a) was shown to be able to predict
the following aspects of unsaturated soil behaviour, which are not captured by
BBM type constitutive models:
 the irreversible change of void ratio during drying
 influence of the previous history of suction variation on subsequent

response to isotropic loading at constant suction
The relationship between ratio 𝑒/𝑒𝑠 and the bonding factor 𝜉 is the only
information required additionally to the parameters 𝑁, 𝜆 and 𝜅 corresponding to
full saturation.
(a) (b)
Figure 2-38: Relationship between ratio 𝑒/𝑒𝑠 and bonding factor 𝜉 during isotropic virgin loading at
constant suction: (a) data by Sharma (1998); (b) data by Sivakumar (1993) (after Gallipoli et al., 2003a)
94
Figure 2-39: (a) isotropic compression lines at constant values of 𝜉; (b) yield loci (after Gallipoli et al.,
2003a)
2.8.2 The Wheeler et al. (2003) model
The Wheeler et al. (2003) model was formulated in terms of Bishop’s stress,
given by Equation 2.11, and of the modified suction, given by Equation 2.10.
The appropriate work-conjugate strain increment variables, according to
Equation 2.9 (Houlsby, 1997), are the strain increment tensor 𝑑𝜀𝑖𝑗 and the
increment of degree of saturation −𝑑𝑆𝑟 .
In the isotropic 𝑠 ∗ - 𝑝∗ plane the model employs the three yield surfaces
illustrated in Figure 2-40: the loading-collapse (LC), the suction increase (SI)
and the suction decrease (SD) yield surfaces. For simplification, the authors
assumed that the additional stabilising force 𝛥𝑁, caused by the meniscus water
at the inter-particle contacts, is constant whenever the meniscus is present and
it disappears with the meniscus. They, therefore, argued that the value of
suction within the menisci has very little effect on the stability of the contacts
and adopted a straight line for the LC yield curve in the 𝑠 ∗ -𝑝∗ plane. What is
important is the number of inter-particle contacts affected by the meniscus
water. The authors introduced the notion of plastic changes of degree of
saturation, 𝑆𝑟 , which correspond to flooding or emptying of voids with water and
which influence significantly the stabilising effect provided by the meniscus
95
water. Plastic changes of 𝑆𝑟 are induced by the SI and SD yield curves on

drying (draining of voids) and on wetting (flooding of voids) respectively.
The hydraulic hysteresis in the SWRC (Figure 2-41) is modelled as an elasto-

plastic process, while on a scanning drying or wetting path soil behaviour is
modelled as elastic: elastic changes of 𝑆𝑟 occur and the air-water interface is
moved without draining or flooding of voids with water. The stress point in the
𝑠 ∗ -𝑝∗ plane lies in the elastic region defined in between the SI and SD curves
(Figure 2-40). Moving along the primary drying path plastic changes of 𝑆𝑟 occur,
corresponding to draining of voids. The stress point in the 𝑠 ∗ -𝑝∗ plane lies on the
SI curve. Equivalently, moving along the primary wetting path, plastic changes
of 𝑆𝑟 occur once more but they correspond to flooding of voids. The stress point
in the 𝑠 ∗ -𝑝∗ plane lies on the SD curve.
As shown in Figure 2-41, the primary curves of the SWRC are parallel lines with
a slope of 𝜆𝑠 . The scanning paths are also parallel to each other and are
defined by a slope of 𝜅𝑠 . The authors noted that the modelling of the retention
behaviour is crude and commented that refinement may be found desirable.
Within the rectangular region defined by the three yield curves shown in Figure
2-40 behaviour is elastic. Elastic volumetric strains are caused by changes in 𝑝∗
and elastic changes in 𝑆𝑟 are caused by variation of 𝑠 ∗ . Plastic volumetric
strains are produced upon yielding on the LC curve. No plastic changes in 𝑆𝑟
are involved in the process. Plastic changes in 𝑆𝑟 are induced upon yielding on
the SI and SD curves, without involving plastic volumetric strains. This is a
consequence of applying an associated flow rule on all three yield curves
(Figure 2-40).
The three yield curves are coupled. Yielding on the LC curve (i.e. outward
movement of the LC) causes simultaneous upward movement of the SI and SD
curves (Figure 2-42 (a)). Yielding on the SI curve (i.e. upward movement of the
SI) causes simultaneous upward movement of the SD curve and outward
movement of the LC curve (Figure 2-42 (b)). Finally, yielding on the SD curve
(i.e. downward movement of the SD) causes simultaneous downward
96
movement of the SI curve and inward movement of the LC curve (Figure 2-42
(c)).
Movement of the SI and SD curves due to yielding on the LC curve (i.e.

generation of plastic volumetric strains) produces a shift in the SWRC. As the SI
and SD curves are moved upwards and the yield values of 𝑠 ∗ increase, the
primary curves are shifted to the right (Figure 2-42 (d)).
The yield curves are given by the following equations:
𝑝∗ = 𝑝0∗ (2.66)
𝑠 ∗ = 𝑠𝐼∗ (2.67)
𝑠 ∗ = 𝑠𝐷∗ (2.68)
where 𝑝0∗ , 𝑠𝐼∗ and 𝑠𝐷∗ define the current position of the LC, SI and SD curves,
respectively.
The elastic volumetric strains are calculated as:
𝜅 𝑑𝑝∗
𝑑𝜀𝑣𝑒 = (2.69)
𝑣 𝑝∗
where 𝜅 is the slope of an elastic swelling line in the 𝑣 -ln 𝑝∗ plane at fully
saturated conditions.
Elastic increments of the degree of saturation are calculated as:
𝑑𝑠 ∗
𝑑𝑆𝑟𝑒 = −𝜅𝑠 ∗ (2.70)
𝑠
where 𝜅𝑠 is the slope of the scanning (elastic) hydraulic paths.
The plastic volumetric strain increments are given by:
𝑝 𝜆 − 𝜅 𝑑𝑝∗
𝑑𝜀𝑣 = (2.71)
𝑣 𝑝∗
where 𝜆 is the slope of the ICL in the 𝑣-ln 𝑝∗ plane at fully saturated conditions.
97
The plastic changes of 𝑆𝑟 are given by:
𝑝 𝑑𝑠𝐼∗ 𝑑𝑠𝐷∗
𝑑𝑆𝑟 = − 𝜆𝑠 − 𝜅𝑠 = − 𝜆 𝑠 − 𝜅𝑠 (2.72)
𝑠𝐼∗ 𝑠𝐷∗
where 𝜆𝑠 is the slope of the primary (plastic) hydraulic paths.
The flow rule on the LC yield curve is:
𝑝
𝑑𝑆𝑟
𝑝 = 0 (2.73)
𝑑𝜀𝑣
and on the SI and SD curves is:
𝑝
𝑑𝜀𝑣
𝑝 = 0 (2.74)
𝑑𝑆𝑟
When yielding on the SI or on the SD curves, coupled movement of the LC

curve is predicted according to the following equation:
𝑑𝑝0∗ 𝑑𝑠𝐼∗ 𝑑𝑠𝐷∗

= 𝑘1 ∗ = 𝑘1 ∗ (2.75)
𝑝0∗ 𝑠𝐼 𝑠𝐷
The authors explain that the coupling parameter 𝑘1 ‘controls the path traced by
the corner between the LC and the SD yield curves during a wetting stage
producing yielding on the SD curve’.
When yielding on the LC curve, coupled movements of the SI and SD curves

are predicted according to the following equation:
𝑑𝑠𝐼∗ 𝑑𝑠𝐷∗ 𝑑𝑝0∗

= = 𝑘 2 (2.76)
𝑠𝐼∗ 𝑠𝐷∗ 𝑝0∗
The authors explain that the coupling parameter 𝑘2 ‘controls the magnitudes of
the shifts in the primary drying and primary wetting curves caused by plastic
volumetric strains’.
98
The overall movement of the LC curve is the sum of any direct movement
induced by yielding on this curve and any indirect movement induced by
yielding on the SI or SD curves:
𝑑𝑝0∗ 𝑣 𝑝 𝑘1 𝑝
∗ = 𝑑𝜀𝑣 − 𝑑𝑆𝑟 (2.77)
𝑝0 𝜆−𝜅 𝜆𝑠 − 𝜅𝑠
Similarly, the overall movement of the SI and SD curves is the sum of any direct
movement induced by yielding on these curves and any indirect movement
induced by yielding on the LC curve:
𝑝
𝑑𝑠𝐼∗ 𝑑𝑠𝐷∗ 𝑑𝑆𝑟 𝑘2 𝑣 𝑝
∗ = ∗ = − + 𝑑𝜀𝑣 (2.78)
𝑠𝐼 𝑠𝐷 𝜆𝑠 − 𝜅𝑠 𝜆−𝜅
The general expressions for the plastic strain increments are:
𝑝 𝜆−𝜅 𝑑𝑝0∗ 𝑑𝑠𝐷∗

𝑑𝜀𝑣 = − 𝑘1 ∗ (2.79)
𝑣 1 − 𝑘1 𝑘2 𝑝0∗ 𝑠𝐷
𝑝 𝜆𝑠 − 𝜅𝑠 𝑑𝑠𝐷∗ 𝑑𝑝0∗
𝑑𝑆𝑟 = − − 𝑘 2 (2.80)
1 − 𝑘1 𝑘2 𝑠𝐷∗ 𝑝0∗
and
𝑝 𝜆−𝜅 𝑑𝑝0∗ 𝑑𝑠𝐼∗

𝑑𝜀𝑣 = − 𝑘1 ∗ (2.81)
𝑣 1 − 𝑘1 𝑘2 𝑝0∗ 𝑠𝐼
𝑝 𝜆𝑠 − 𝜅𝑠 𝑑𝑠𝐼∗ 𝑑𝑝0∗
𝑑𝑆𝑟 = − − 𝑘 2 (2.82)
1 − 𝑘1 𝑘2 𝑠𝐼∗ 𝑝0∗
The authors explained that when the stress state is at the corner of the LC and
SI curves one of the following is possible, depending on the direction of the
stress path:
 elastic unloading
 yielding on the SI curve only
 yielding on the LC curve only
99
 yielding on both the LC and SI curves
The latter occurs if:
𝑝∗ 𝑑𝑠 ∗ 1
𝑘2 < ∗ ∗
< (2.83)
𝑠 𝑑𝑝 𝑘1
Similar considerations apply when the stress state is at the corner of the LC and
SD curves. In order to avoid numerical problems when the stress state
approaches these corners the following restriction applies: 𝑘1 𝑘2 < 1.
The model requires six soil constants (parameters): 𝜆, 𝜅, 𝜆𝑠 , 𝜅𝑠 , 𝑘1 and 𝑘2 . The

authors suggested that the formulation may be extended to the 𝑞-𝑝∗ -𝑠 ∗ space
by adopting an ellipse, as shown in Figure 2-43. Basic aspects of unsaturated
soil behaviour, such as swelling or collapse compression on wetting and
maximum potential collapse, are captured by the model. Modelling capabilities
include more complicated features such as:
 smooth transition between saturated and unsaturated conditions
 irreversible compression during the drying stages of wetting-drying

cycles
 influence of wetting-drying cycles on subsequent response to isotropic

loading
 influence of hydraulic hysteresis and of plastic volumetric strains on the

variation of the degree of saturation
The existence of three yield surfaces inevitably complicates the implementation

of the Wheeler et al. (2003) model in a finite element code. Furthermore, the
process of initialising the yield values of suction 𝑠𝐼∗ and 𝑠𝐷∗ , which are coupled to
the position of the SWRC, is not clear. Although this process is simple in single
element analyses, where the same value of suction is applied to all the
integration points, complications may arise in a boundary value problem where
the initial suction distribution may vary significantly between different integration
points. Finally, expansion of the model in the generalised stress space is
necessary for its implementation in a numerical code.
100
Despite the adjustments suggested above and despite the crude and unrealistic
shape adopted for the SWRC, the model successfully couples the mechanical
and hydraulic components of unsaturated soil behaviour, in an innovative and
coherent way. From this point of view, the framework proposed by Wheeler et
al. (2003) classifies as a benchmark in modelling unsaturated soil behaviour.
Figure 2-40: LC, SI and SD yield curves for isotropic stress states (after Wheeler et al., 2003)
Figure 2-41: Model for water retention behaviour (after Wheeler et al., 2003)
101
s*, -dSr s*, -dSr s*, -dSr
s*I,2 SI s*I,2 SI
C SI
LC LC s*I,1
s*I,1 s*I,1 B
A B C s*I,2
s*D,2 s*D,2 A
A
SD SD LC
s*D,1 B
s*D,1 s*D,1
s*D,2 C
p*, -dvp p*, -dvp SD p*, -dvp
p*0,1 p*0,2 p*0,1 p*0,2 p*0,2 p*0,1
(a) (b) (c)

Sr
Initial position
Final position
s* (log scale)
s*D,1 s*D,2 s*I,1 s*I,2
(d)
Figure 2-42: Coupling of the LC, SI and SD yield surfaces in the Wheeler et al. (2003) model
Figure 2-43: Yield surfaces in the 𝑞-𝑝∗ -𝑠∗ stress space (after Wheeler et al., 2003)
2.8.3 The Sheng et al. (2008) model
The model proposed by Sheng et al. (2008) is formulated in terms of net stress
(denoted as 𝜎𝑖𝑗 ) and suction 𝑠. The work-conjugate strains are the increment of
102
the soil skeleton strain vector 𝑑𝜀𝑖𝑗 and the increment of the volumetric water
content 𝑑𝜃.
The volume change due to changes in mean net stress 𝑝 and suction 𝑠 is given
by the following equation:
𝑑𝑝 𝑑𝑠
𝑑𝜀𝑣 = 𝜆𝑣𝑝 + 𝜆𝑣𝑠 (2.84)
𝑝+𝑠 𝑝+𝑠
where 𝜆𝑣𝑝 is the slope of the ICL for normally consolidated soils at full
saturation. The slope 𝜆𝑣𝑠 is identical to 𝜆𝑣𝑝 when the soil is fully saturated and
decreases gradually to zero at high suction levels:
𝜆𝑣𝑝 , 𝑠 < 𝑠𝑠𝑎

𝜆𝑣𝑠 = 𝑠𝑠𝑎 + 1 (2.85)
𝜆𝑣𝑝 , 𝑠 ≥ 𝑠𝑠𝑎
𝑠+1
where 𝑠𝑠𝑎 is the saturation suction.
Although each one of the terms in Equation 2.84 can be integrated, the stress-
strain relationship is usually written in incremental form and therefore only the
rate form (Equation 2.84) is needed. An important aspect of the volumetric
behaviour described by this equation is its stress dependency. Figure 2-44
illustrates 3-dimensional surfaces in the 𝑒-ln 𝑠-ln 𝑝 space (𝑒 being the void ratio)
obtained for the paths ABD (Figure 2-44 (a)) and ACD (Figure 2-44 (b)). The
path ABD involves drying under zero mean net stress from point A to point B
and subsequent loading to point D under constant suction. The path ACD
involves loading under zero suction from point A to point C and subsequent
drying to point D at constant mean net stress. The two surfaces are dissimilar.
The projection in the 𝑒-log 𝑝 plane of the surface illustrated in Figure 2-44 (a), at
different suction levels, is illustrated in Figure 2-45. Evidently, the ICL’s for
unsaturated conditions are not straight lines. As suction increases the initial part
of the isotropic compression lines becomes increasingly flatter. Volumetric
collapse is predicted when wetting the soil to full saturation under constant high
values of mean net stress.
103
The projection in the 𝑒-log 𝑠 plane of the surface shown in Figure 2-44 (b), at
different confining stress levels, is illustrated in Figure 2-46. It is shown that
there is little volume change induced by drying when the soil is first loaded to
high mean net stresses. The same pattern is demonstrated by the experimental
data shown in the same figure.
The elastic volumetric volume change caused by stress and suction changes is:
𝑑𝑝 𝑑𝑠
𝑑𝜀𝑣𝑒 = 𝜅𝑣𝑝 + 𝜅𝑣𝑠 (2.86)
𝑝+𝑠 𝑝+𝑠
where 𝜅𝑣𝑝 is the slope of the swelling line at full saturation. 𝜅𝑣𝑠 is identical to
𝜅𝑣𝑝 when the current suction is less than the saturation suction 𝑠𝑠𝑎 and
decreases gradually to zero with increasing suction:
𝜅𝑣𝑝 , 𝑠 < 𝑠𝑠𝑎

𝜅𝑣𝑠 = 𝑠𝑠𝑎 + 1 (2.87)
𝜅𝑣𝑝 , 𝑠 ≥ 𝑠𝑠𝑎
𝑠+1
Any shape for the yield surface may be adopted in the 𝑞-𝑝-𝑠 stress space. The
authors employed the Modified Cam-clay (MCC) surface which was written in
the following form:
𝑓 = 𝑞 2 − 𝑀2 𝑝 − 𝑝0 𝑠 ∙ 𝑝𝑦 𝑠 − 𝑝 (2.88)
where 𝑀 is the slope of the critical state line at full saturation (the authors
pointed out that it may be chosen to be a function of suction), 𝑝𝑦 𝑠 represents
the pre-consolidation yield stress at different suction levels and 𝑝0 𝑠 is the
function that controls the expansion of the elastic region into the tensile region
of net stresses and the increase of apparent cohesion with suction. 𝑝𝑦 and 𝑝0
are given by the following equations:
𝑝𝑦0 − 𝑠, 𝑠 < 𝑠𝑠𝑎

𝑝𝑦 = 𝑠+1 (2.89)
𝑝𝑦0 − 𝑠𝑠𝑎 − 𝑠𝑠𝑎 + 1 ln , 𝑠 ≥ 𝑠𝑠𝑎
𝑠𝑠𝑎 + 1
and
104
−𝑠, 𝑠 < 𝑠𝑠𝑎

𝑝0 = 𝑠+1 (2.90)
−𝑠𝑠𝑎 − 𝑠𝑠𝑎 + 1 ln , 𝑠 ≥ 𝑠𝑠𝑎
𝑠𝑠𝑎 + 1
where 𝑝𝑦0 is the pre-consolidation value of mean net stress at zero suction.
Equations 2.89 and 2.90 are essentially the two traces of the MCC surface on
the isotropic plane 𝑝-𝑠. For suction levels lower than the saturation suction 𝑠𝑠𝑎 ,
the two curves are 45o straight lines, as shown in Figure 2-47. The 𝑝𝑦 line
passes through the pre-consolidation value of mean net stress at zero suction,
𝑝𝑦0 , and reduces linearly with suction, up to the saturation suction 𝑠𝑠𝑎 . The 𝑝0
line is the 45o line passing through the origin of the axes.
Figure 2-48 illustrates the position of the initial yield surface (black line) for a soil
that has a saturation suction, 𝑠𝑠𝑎 , of 100.0 kPa and has been consolidated to
300 kPa at zero suction. Drying such a soil at zero mean net stress yield is
predicted at suction of about 730 kPa.
For isotropically hardening material 𝑝𝑦 changes according to the hardening law:
𝑝 𝜆𝑣𝑝 − 𝜅𝑣𝑝 𝜆𝑣𝑠 − 𝜅𝑣𝑠

𝑑𝜀𝑣 = 𝑑𝑝 + 𝑑𝑠 (2.91)
𝑝+𝑠 𝑝+𝑠
At constant suction, i.e. 𝑑𝑠 = 0.0, the evolution of 𝑝𝑦 depends on the suction

level:
𝑝 𝜆𝑣𝑝 − 𝜅𝑣𝑝
𝑑𝜀𝑣 = 𝑑𝑝𝑦 (2.92)
𝑝𝑦 + 𝑠
If the soil in Figure 2-48 is isotropically loaded under different suction levels, the
yield surface acquires a new position, given by the following equation:
𝑝𝑦𝑛 0 − 𝑠, 𝑠 < 𝑠𝑠𝑎

𝑝𝑦𝑛 = 𝑝𝑦𝑛 0 𝑠+1 (2.93)
𝑝 + 𝑠 − 𝑠𝑠𝑎 − 𝑠𝑠𝑎 + 1 ln − 𝑠, 𝑠 ≥ 𝑠𝑠𝑎
𝑝𝑦0 𝑦0 𝑠𝑠𝑎 + 1
where 𝑝𝑦𝑛 0 is the new yield stress at zero suction. If the new yield stress at zero
suction, 𝑝𝑦𝑛 0 , is known, Equation 2.93 may be used to find the new position of
105
the yield surface. Alternatively, if the yield stress, 𝑝𝑦𝑛 , at a given suction level is
known, then Equation 2.93 may be used to calculate the yield stress at zero
suction, 𝑝𝑦𝑛 0 .
Equation 2.93 is plotted in Figure 2-48 (grey line) for 𝑝𝑦𝑛 0 = 500 kPa. The
shapes of the initial and the new yield surfaces are dissimilar. The yield stress
along the new yield surface does not decrease monotonically with suction, in
contrast with the initial yield surface, but obtains a minimum value after which it
decreases with increasing suction. The suction 𝑠𝑐 corresponding to the
minimum value of 𝑝𝑦𝑛 may be calculated as:
𝑠𝑠𝑎 + 1 𝑝𝑦𝑛 0
𝑠𝑐 = −1 (2.94)
𝑝𝑦𝑛 0 − 𝑝𝑦0
If a soil lying on the 𝑝𝑦𝑛 yield surface is wetted from suctions larger than 𝑠𝑐
collapse is predicted.
Alternatively, drying from the initial yield surface (Figure 2-49), under constant
mean net stress, i.e. 𝑑𝑝 = 0.0, Equation 2.91 reduces to:
𝑝 𝜆𝑣𝑠 − 𝜅𝑣𝑠
𝑑𝜀𝑣 = + 𝑑𝑠𝑦 (2.95)
𝑝 + 𝑠𝑦
where 𝑠𝑦 is the suction value along the yield surface 𝑝𝑦 . The equation of the
yield surface (Equation 2.89) may be rewritten in terms of 𝑠𝑦 :
𝑝𝑦0 − 𝑝, 𝑝 > 𝑝𝑦0 − 𝑠𝑠𝑎

𝑠𝑦 = 𝑝𝑦0 − 𝑠𝑠𝑎 − 𝑝 (2.96)
𝑠𝑠𝑎 + 1 exp , 𝑝 ≤ 𝑝𝑦0 − 𝑠𝑠𝑎
𝑠𝑠𝑎 + 1
The new position of the yield surface, 𝑠𝑦𝑛 , is given by the following equation:
𝑝𝑦0 − 𝑝, 𝑝 > 𝑝𝑦0 − 𝑠𝑠𝑎

𝐴𝑝 − 1
𝑠𝑦𝑛 = 1−𝐴 , 𝑝 ≤ 𝑝𝑦0 − 𝑠𝑠𝑎 (2.97)
𝐵𝑝 − 1
, 𝑝𝑦𝑛 0 − 𝑠𝑠𝑎 ≥ 𝑝 > 𝑝𝑦0 − 𝑠𝑠𝑎
1−𝐵
where:
106
𝑝 −1
𝑝𝑦𝑛 0 𝑠𝑠𝑎 +1 𝑠𝑦 + 1
𝐴= (2.98)
𝑝𝑦0 𝑠𝑦 + 𝑝
𝑝 −1
𝑝𝑦𝑛 0 𝑠𝑠𝑎 +1 𝑠𝑠𝑎 +1
𝐵= (2.99)
𝑝 + 𝑠𝑠𝑎 𝑠𝑠𝑎 +𝑝
The new yield surface 𝑠𝑦𝑛 is plotted in Figure 2-49 (grey line) and its shape is
similar to the shape of the initial yield surface 𝑠𝑦 , indicating that drying at
constant mean net stress does not alter significantly the yield function.
The soil-water retention curve adopted in the model is shown in Figure 2-50 and
is of the following form:
𝑑𝑠
𝑑𝑆𝑟 = −𝜆𝑤𝑠 (2.100)
𝑠
where, upon primary drying, the slope 𝜆𝑤𝑠 is:
0, 𝑠 < 𝑠𝑠𝑎
𝜅𝑤𝑠 , 𝑠𝑠𝑎 ≤ 𝑠 < 𝑠𝑎𝑒
𝜆𝑤𝑠 = (2.101)
𝜆𝑤𝑠 , 𝑠𝑎𝑒 ≤ 𝑠 < 𝑠𝑟𝑒
𝜅𝑤𝑠 , 𝑠 ≥ 𝑠𝑟𝑒
where 𝑠𝑎𝑒 is the air-entry value, 𝑠𝑟𝑒 is the residual suction and 𝜅𝑤𝑠 is the slope
of the scanning paths (see also Figure 2-50).
Upon primary wetting, the slope 𝜆𝑤𝑠 is given by a similar equation, but the
water-entry value of suction, 𝑠𝑤𝑒 is considered rather than the air-entry 𝑠𝑎𝑒 .
In order to include the effect of the hydraulic hysteresis Sheng et al. (2008)
suggested the employment of a suction increase (SI) and a suction decrease
(SD) yield surfaces, as shown in Figure 2-51. The function of the SI and SD
surfaces is similar to the one adopted by Wheeler et al. (2003) and is therefore
not repeated herein.
Sheng et al. (2008) referred to their model as SFG (Sheng-Fredlund-Gens)

model. It requires the following material parameters: the slope of the isotropic
ICL, 𝜆𝑣𝑝 , and of the swelling line, 𝜅𝑣𝑝 , for saturated conditions; the slope of the
107
CSL 𝑀 in the 𝑞-𝑝 plane; the slope of the primary and of the scanning SWRC
paths, 𝜆𝑤𝑠 and 𝜅𝑤𝑠 ; the saturation, the air-entry, the water-entry and the residual
values of suction, 𝑠𝑠𝑎 , 𝑠𝑎𝑒 , 𝑠𝑤𝑒 and 𝑠𝑟𝑒 , respectively; Poisson’s ratio, 𝜇; the initial
soil state prescribed by the initial pre-consolidation pressure, 𝑝𝑦0 , and the initial
void ratio, 𝑒.
In summary, the SFG model predicts curved ICL’s at constant suctions and
employs a yield surface which is highly stress path dependent. The model also
accommodates hysteresis associated with cycles of wetting and drying.
Figure 2-44: Void ratio versus suction and mean net stress (after Sheng et al., 2008): (a) path ABD;
(b) path (ACD)
108
Figure 2-45: Isotropic compression lines at different suctions (after Sheng et al., 2008)
Figure 2-46: Void ratio versus suction under constant mean net stress and comparison with experimental
data (after Sheng et al., 2008)
109
Figure 2-47: Yield surface for saturated conditions (after Sheng et al., 2008)
Figure 2-48: Initial yield surface for a soil consolidated to 300 kPa at zero suction and its evolution when
the soil is then loaded under different suction levels (after Sheng et al., 2008)
110
Figure 2-49: Initial yield surface for a soil consolidated to 300 kPa at zero suction and its evolution when
the soil is then dried under different mean net stresses (after Sheng et al., 2008)
Figure 2-50: SWRC employed in the in Sheng et al. (2008) model (after Sheng et al., 2008)
111
Figure 2-51: Elastic region deafened by the yield surfaces and the suction increase (SI) and suction
decrease (SD) surfaces (after Sheng et al., 2008)
2.9 Summary and conclusions
The behaviour of unsaturated soils differs significantly from the behaviour of

fully saturated soils. While discussing recent developments in the area of
unsaturated soils the most important features of the behaviour were disclosed.
The effective stress principle is not applicable in unsaturated conditions, as

changes in suction and changes in applied stress may have distinct effects on
the soil structure. Two independent stress variables are required in order to
explain the behaviour of unsaturated soils. Different combinations of
independent stress variables, with increasing complexity, have been employed
in elasto-plastic models in order to improve their ability to represent certain
features of behaviour. Nonetheless, the choice of stress variables is thought to
be a matter of convenience.
112
Changes in the volume of unsaturated soils may occur as a consequence of

changes in suction or changes in applied stress. Drying of soils is equivalent to
increasing the confining stress for as long as the soil remains fully saturated.
Once the air-entry value of suction is reached and de-saturation takes places,
the total volume change of the soil is smaller than the pore water volume
change.
Wetting at constant confining stress may cause swelling or collapse or even a

reversal from swelling to collapse, depending on the level of confinement. If the
applied stress is sufficiently low swelling is exhibited, whereas at high confining
stresses collapse may occur.
The amount of potential collapse due to wetting is also stress dependent and it
initially increases with confining stress, reaches a maximum and reduces with
further increase of confinement to small values.
Suction contributes to an increase in yield stress. This increase is commonly

illustrated in the suction-net mean stress plane employing the loading-collapse
(LC) yield locus. Although the shape of the LC locus is generally accepted to be
curved, recent laboratory tests on isotropically prepared samples yielded along
a straight line in the above mentioned plane.
The suction level also affects the compressibility of the soil when isotropically
loaded at constant suction. Although the slope of the swelling paths is generally
thought to be insignificantly affected by the suction level, the slope of the
isotropic compression line (ICL) is significantly affected. Whereas a decrease in
the slope of the ICL with increasing suction has been reported by several
authors, the opposite behaviour had been reported by others. The general
perception is that the slope of the ICL initially decreases with suction at low
confining stress and increases at high confining stresses.
In most of the theoretical approaches it has been suggested that the apparent
cohesion increases with suction, thus justifying the increase in shear strength
with suction. Although this has been shown to be generally true, it is not
unusual for the apparent cohesion to exhibit a reduction at high values of
suction. The reasons for that are yet unclear but a possible justification is that
113
shearing resistance is reduced at very high suctions because the number of

water menisci is reduced.
In constitutive modelling the slope of the critical state line (CSL) in the deviatoric
stress-mean net stress plane has been commonly assumed to be independent
of the suction level. Nonetheless, this simplification contradicts some recent
experimental evidence which show the slope of the CSL to be unaffected only
at low values of suction.
Sample preparation is thought to affect the behaviour of unsaturated soils. For

one-dimensionally prepared samples the compaction pressure and the
compaction water content affect the position of the ICL. The position of the CSL
is affected by the compaction water content but not by the compaction pressure.
On the contrary, the positions of the ICL and CSL of samples isotropically
prepared were not affected by the initial condition of the samples,
demonstrating the uniqueness of the two lines.
The difference in the behaviour exhibited by one-dimensionally and isotropically

prepared samples is attributed to the inherent anisotropy associated with the
former. Therefore, a constitutive model accounting for stress anisotropy should
ideally be used when modelling compacted soils.
Cyclic changes of suction may produce significant irreversible strains even at

constant confining stress and for suction changes that do not exceed the
previously attained suction level. This irreversibility of strains is correlated to the
occurrence of hydraulic hysteresis in the soil-water retention curve (SWRC).
Drying and wetting are not reversible hydraulic processes and distinct hydraulic
paths are followed on drying and on subsequent wetting, with the latter normally
lying beneath the former. When a reconstituted sample is dried from slurry to
residual conditions and is subsequently wetted back to full saturation it follows
the primary paths of the SWRC. The retention state of a soil may exist
anywhere between the primary paths and on drying or wetting it follows
scanning paths which move towards the primary paths. Therefore, the primary
paths bound an infinite number of scanning paths.
114
The SWRC is additionally affected by the variation of void ratio or specific

volume induced by isotropic loading or by shearing. The SWRC shifts to the
right as the specific volume reduces so that a larger suction can be sustained at
a given degree of saturation.
Recent modelling of the SWRC has successfully accounted for the impact of the
specific volume and the SWRC is illustrated as a surface in the suction-degree
of saturation-specific volume space. Furthermore, modelling of the scanning
hydraulic paths for soils exhibiting hydraulic hysteresis has been advanced.
Two models were presented in which the scanning paths were integrated from
their slope. The slope of the scanning paths was determined based on the slope
of the primary paths. In the first of the models the image of the point on the
congruent primary path was employed in the determination of the slope
whereas in the second one the slope was a function of the vertical distance
between the current point and the congruent primary path.
The mechanical and the hydraulic components of unsaturated soil behaviour

are coupled and the SWRC affects and is affected by the mechanical
behaviour. Typical examples of the coupling include the irreversibility of strains
upon cycles of suction, the shift of the SWRC due to volumetric variations, the
impact of cyclic changes of suction on the subsequent soil behaviour during
isotropic loading. Elasto-plastic models formulated in terms of net stress and
suction cannot capture the coupling of the mechanical and hydraulic behaviour.
For the behaviour to be coupled the degree of saturation is either incorporated
in the formulation of the stress variables or is included in the modelling of a
specific feature of behaviour.
Three constitutive models incorporating the effect of the degree of saturation

were presented. The model by Gallipoli et al. (2003a) is formulated in terms of
the average skeleton stress 𝑝′′ (Bishop’s stress), which is defined by the net
stress, the suction and the degree of saturation, and the bonding variable 𝜉,
which is a measure of the magnitude of the inter-particle bonding due to water
menisci. A unique relationship was assumed between the bonding variable 𝜉
and the ratio of current void ratio 𝑒 at unsaturated conditions over the equivalent
void ratio at fully saturated conditions 𝑒𝑠 . In this way an isotropic compression
115
state surface is defined in the 𝑒-𝑝′′ - 𝜉 space and a single yield curve in the 𝑝′′ - 𝜉
space. The model is capable of predicting complex aspects of behaviour such
as the irreversible change in void ratio during drying and the influence of the
previous history of suction variation on the subsequent response to isotropic
loading at constant suction.
The Wheeler et al. (2003) model was formulated in terms of Bishop’s stress 𝑝∗
and modified suction 𝑠 ∗ . It combines three yield surfaces: the loading-collapse
(LC), the suction increase (SI) and the suction decrease (SD) curves. Their
shape is simple (straight lines) and there is coupling between them. Employing
the SI and SD curves plastic changes of the degree of saturation are modelled
and the hydraulic hysteresis is perceived as an elasto-plastic process. Coupling
between the mechanical and hydraulic components of behaviour is successfully
modelled.
Although formulated in terms of net stress and suction, the Sheng et al. (2008)
model incorporates the effect of hydraulic hysteresis employing the SI and SD
yield curves. The shape of the LC curve is highly stress path dependent and
hardening due to confining stress changes results in a different shape than
hardening due to suction changes. The ICL’s for unsaturated conditions are
curved lines which become increasingly flatter at low stress levels as suction
increases.
Constitutive modelling of unsaturated soils has mainly been focused on the

isotropic stress plane with little consideration for the reproduction of the
shearing behaviour. Critical state frameworks and elasto-plastic constitutive
models have suggested the existence of an elliptic yield function in the triaxial
𝑞 - 𝑝 plane but research has been largely limited to shearing of normally
consolidated soils, overlooking the yielding behaviour of overconsolidated soils
when sheared. Furthermore, generalisation in the deviatoric plane would be
required before employing the constitutive models presented in the current
chapter in numerical analysis of boundary value problems.
116
chapter 3: EXISTING ICFEP CAPABILITIES FOR
MODELLING UNSATURATED SOIL
BEHAVIOUR
3.1 Introduction
The current chapter is focused on the capabilities of the Imperial College Finite
Element Program (ICFEP) (Potts & Zdravkovic, 1999) which are relevant to
numerical analysis of unsaturated soils and were available at the
commencement of the present project. The aspects of ICFEP discussed formed
the bases for this research and were either further developed (constitutive and
soil-water retention curve models, in Chapters 4 and 5) or were employed intact
in the analyses presented in Chapters 6 and 7.
The modifications involved in the formulation of the finite element method, in

order to accommodate the presence of two stress variables, are introduced,
before the existing elasto-plastic constitutive models, required for the simulation
of the mechanical behaviour, are thoroughly explained. The available
expressions for the soil-water retention curve (SWRC) and the variation of
permeability are subsequently discussed and boundary conditions associated
117
Existing ICFEP Capabilities for Modelling Unsaturated Soil Behaviour
with the numerical analysis of unsaturated soils, such as the vegetation and the
precipitation boundary conditions, are finally introduced.
3.2 Extension of the finite element method to unsaturated soil

mechanics
3.2.1 Elasto-plastic stiffness matrix
Description of unsaturated soil behaviour requires the employment of two stress

variables, which can both generate total strains (sum of elastic and plastic
strains). Consequently, the stress-strain relationship in the finite element
method cannot be expressed in terms of a single stiffness matrix 𝐷𝑒𝑝 and,
therefore, the alterations presented in this section are necessary.
The two stress variables adopted in ICFEP are the equivalent stress:
𝜎 = 𝜎𝑛𝑒𝑡 + 𝑠𝑎𝑖𝑟 (3.1)
and the equivalent suction:
𝑠𝑒𝑞 = 𝑠 − 𝑠𝑎𝑖𝑟 (3.2)
where 𝜎𝑛𝑒𝑡 is the net total stress and is equal to 𝜎𝑡𝑜𝑡𝑎𝑙 − 𝑢𝑎 , (𝜎𝑡𝑜𝑡𝑎𝑙 being the
total stress and 𝑢𝑎 being the pore air pressure), 𝑠 is the matric suction, 𝑢𝑎 −
𝑢𝑤 , and 𝑠𝑎𝑖𝑟 is the air-entry value of suction. As already mentioned, the term
suction is widely used in the present thesis meaning matric suction, for reasons
of brevity. For 𝑠 < 𝑠𝑎𝑖𝑟 , the effective stress principle is applicable and the
modifications explained in the present section are valid only for 𝑠 > 𝑠𝑎𝑖𝑟 .
Assuming that changes in equivalent total stress 𝜎 produce the elastic and
plastic incremental strains 𝛥𝜀 𝑒 and 𝛥𝜀 𝑝 , respectively, and that the
𝑝
incremental strains 𝛥𝜀𝑠𝑒 and 𝛥𝜀𝑠 are due to changes in equivalent suction
𝑠𝑒𝑞 , the total incremental strains 𝛥𝜀 can be written as:
𝑝
𝛥𝜀 = 𝛥𝜀 𝑒 + 𝛥𝜀 𝑝 + 𝛥𝜀𝑠𝑒 + 𝛥𝜀𝑠 (3.3)
118
𝑝
The plastic incremental strains 𝛥𝜀𝑠 are induced by a secondary yield surface
(SI) introduced later. The incremental equivalent stress changes 𝛥𝜎 are
related to the elastic strains 𝛥𝜀 𝑒 through the constitutive equation:
𝛥𝜎 = 𝐷 ∙ 𝛥𝜀 𝑒 (3.4)
where 𝐷 is the elastic stiffness matrix. Combining the above two equations,
the incremental equivalent stress changes 𝛥𝜎 can be written as:
𝑝
𝛥𝜎 = 𝐷 ∙ 𝛥𝜀 − 𝛥𝜀 𝑝 − 𝛥𝜀𝑠𝑒 − 𝛥𝜀𝑠 (3.5)
The incremental plastic strains 𝛥𝜀 𝑝 are related to the plastic potential function
𝐺 𝜎 , 𝑘 , 𝑠𝑒𝑞 through the plastic strain multiplier Λ, as follows:
𝜕𝐺 𝜎 , 𝑘 , 𝑠𝑒𝑞
𝛥𝜀 𝑝 = Λ ∙ (3.6)
𝜕𝜎
𝑘 being the state parameters vector. Therefore:
𝜕𝐺 𝜎 , 𝑘 , 𝑠𝑒𝑞 𝑝
𝛥𝜎 = 𝐷 ∙ 𝛥𝜀 − Λ ∙ − 𝛥𝜀𝑠𝑒 − 𝛥𝜀𝑠 (3.7)
𝜕𝜎
or equally:
𝑝 𝜕𝐺 𝜎 , 𝑘 , 𝑠𝑒𝑞
𝛥𝜎 = 𝐷 ∙ 𝛥𝜀 − 𝛥𝜀𝑠𝑒 − 𝛥𝜀𝑠 − 𝐷 ∙Λ∙ (3.8)
𝜕𝜎
The consistency condition for conventional analysis of fully saturated soils

(Potts & Zdravkovic, 1999) includes two terms associated with the stress state,
𝜎 , and the state parameters, 𝑘 . In addition to those, a third term is required
for partly saturated analysis, so that:
𝑇 𝑇
𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞 𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞
𝑑𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞 = 𝛥𝜎 + 𝛥𝑘 +
𝜕𝜎 𝜕𝑘
𝑇 (3.9)
𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞
+ 𝛥𝑠𝑒𝑞 = 0
𝜕𝑠𝑒𝑞
119
where 𝛥𝑠𝑒𝑞 is the change in equivalent suction and 𝐹 is the yield (YS)
function.
The plastic strain multiplier Λ is calculated combining Equations 3.7 and 3.9:
𝑇
𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞 𝑝
𝐷 𝛥𝜀 − 𝛥𝜀𝑠𝑒 − 𝛥𝜀𝑠
𝜕𝜎
Λ= 𝑇 +
𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞 𝜕𝐺 𝜎 , 𝑘 , 𝑠𝑒𝑞
∙ 𝐷 ∙ +𝐴
𝜕𝜎 𝜕𝜎
𝑇
(3.10)
𝛥𝑠𝑒𝑞
𝜕𝑠𝑒𝑞
+ 𝑇
∙ 𝐷 ∙ +𝐴
𝜕𝜎 𝜕𝜎
where:
𝑇
1 𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞
𝐴=− 𝛥𝑘 (3.11)
Λ 𝜕𝑘
Substituting Equation 3.10 into Equation 3.8:
𝑝
𝛥𝜎 = 𝐷𝑒𝑝 ∙ 𝛥𝜀 − 𝛥𝜀𝑠𝑒 − 𝛥𝜀𝑠 − 𝑊 ∙ 𝛥𝑠𝑒𝑞 (3.12)
where:
𝑇
𝜕𝐺 𝜎 , 𝑘 , 𝑠𝑒𝑞 𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞
𝐷 ∙ ∙ ∙ 𝐷
𝜕𝜎 𝜕𝜎
𝑒𝑝
𝐷 = 𝐷 − 𝑇 (3.13)
∙ 𝐷 ∙ +𝐴
𝜕𝜎 𝜕𝜎
is the elasto-plastic matrix in conventional analysis and:
𝑇
𝜕𝐺 𝜎 , 𝑘 , 𝑠𝑒𝑞 𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞
𝐷 ∙ ∙
𝜕𝜎 𝜕𝑠𝑒𝑞
𝑊 = 𝑇 (3.14)
∙ 𝐷 ∙ +𝐴
𝜕𝜎 𝜕𝜎
120
In a non-linear analysis the above matrices vary with the stress and suction
levels and the state parameters. Consequently, the incremental global stiffness
matrix 𝐾𝐺 , which relates the vector of the unknown degrees of freedom
𝑖
(incremental nodal displacements) 𝛥𝑑 𝑛𝐺 to the global right-hand side vector
𝑖
𝛥𝑅𝐺 (containing incremental body forces, surface tractions and suction
terms), in the global equation:
𝑖 𝑖 𝑖
𝐾𝐺 𝛥𝑑 𝑛𝐺 = 𝛥𝑅𝐺 (3.15)
is non-constant and varies not only between increments but also throughout an
increment (Potts & Zdravkovic, 1999). Therefore, a non-linear solution
technique is required to solve Equation 3.15. The modified Newton-Raphson
method with an error controlled sub-stepping stress-point algorithm is employed
in ICFEP and more details are given by Potts & Zdravkovic (1999), while its
application to partly saturated analysis is explained by Georgiadis (2003).
3.2.2 Formulation for coupled problems
Potts & Zdravkovic (1999) explain that for coupled problems the finite element
formulation needs to be expanded to account for the time dependency of the
𝑖
changes in pore fluid pressure, 𝛥𝑝𝑓 , which becomes an additional degree of
𝑛𝐺
freedom in coupled analyses. The governing equation for full saturation,

therefore, is:
𝐾𝐺 𝐿𝐺 𝛥𝑑 𝑛𝐺 𝛥𝑅𝐺
= (3.16)
𝐿𝐺 𝑇 −𝛽𝛥𝑡 𝛷𝐺 𝛥𝑝𝑓 𝑛𝐺 𝑛𝐺 + 𝑄 + 𝛷𝐺 𝑝𝑓 𝑛𝐺
𝛥𝑡
where:
𝑁
𝑇
𝐾𝐺 = 𝐵 𝐷′ 𝐵 𝑑𝑉𝑜𝑙 (3.17)
𝑖=1 𝑣𝑜𝑙 𝑖
is the global stiffness matrix,
121
𝑁
𝑇
𝐿𝐺 = 𝑚 𝐵 𝑁𝑝 𝑑𝑉𝑜𝑙 (3.18)
𝑇
𝑚 being equal to 1 1 1 0 0 0 and:
𝐷′ is the effective stress constitutive matrix
𝐵 is the strain matrix
𝑁𝑝 is the matrix of pore pressure interpolation functions, similar to the matrix of

shape functions 𝑁
𝛽 is a time stepping factor reflecting the variation of pore pressure 𝑝𝑓 𝑛𝐺

with
time
𝛥𝑡 is the time increment
𝑄 represents the flow due to any sources and/or sinks and:
𝑁
𝑇
𝐸 𝑘 𝐸
𝛷𝐺 = 𝑑𝑉𝑜𝑙 (3.19)
𝛾𝑓
𝑁
𝑇
𝑛𝐺 = 𝐸 𝑘 𝑖𝐺 𝑑𝑉𝑜𝑙 (3.20)
𝑇
𝜕𝑁𝑝 𝜕𝑁𝑝 𝜕𝑁𝑝
𝐸 = (3.21)
𝜕𝑥 𝜕𝑦 𝜕𝑧
𝑇
𝑖𝐺 = 𝑖𝐺𝑥 𝑖𝐺𝑦 𝑖𝐺𝑧 is the unit vector parallel, but in the opposite direction, to
gravity, 𝑘 is the permeability matrix, 𝛾𝑓 is the unit weight of water (9.81 kN/m3)
and 𝑖, 𝑁 are increment numbers.
Smith (2003) altered the above formulation in ICFEP, in order to accommodate

unsaturated states. The 𝑚 term within the matrix 𝐿𝐺 was substituted by
122
𝐷 𝑚𝐻 , where 𝑚𝐻 = 1 𝐻 1 𝐻 1 𝐻 0 0 0 and 𝐻 is the elastic

modulus of the soil structure with respect to suction, 𝑠. Consequently:
𝑁
𝑇
𝐿𝐺 = 𝐷 𝑚𝐻 𝐵 𝑁𝑝 𝑑𝑉𝑜𝑙 (3.22)
Furthermore, the following expression applies in unsaturated conditions:
𝑇
Ω 𝐿𝐺 𝛥𝑑 𝑛𝐺 − 𝛽𝛥𝑡 𝛷𝐺 + 𝜔 𝑀𝑁 𝛥𝑝𝑓 =
𝑛𝐺
= 𝑛𝐺 + 𝑄 + 𝛷𝐺 𝑝𝑓 𝛥𝑡 (3.23)
𝑛𝐺
where 𝑀𝑁 is the mass matrix and Ω governs the volume of water that flows for
a given change in the volume of voids (see Section 3.2.3). Furthermore:
1 3Ω
𝜔= − (3.24)
𝑅 𝐻
where 𝑅 is the modulus relating a change in volumetric water content to a
change in matric suction 𝑢𝑎 − 𝑢𝑤 , and must, therefore, be the gradient of the
soil-water retention curve (SWRC expressed in terms of the volumetric water
content).
Smith (2003) explains that the terms in Equation 3.23 have a similar sense as
𝑇
for saturated soils; Ω 𝐿𝐺 𝛥𝑑 𝑛𝐺 accounts for flow generated due to changes in
the volume of voids (soil structure displacements), while the consolidation term
−𝛽𝛥𝑡 𝛷𝐺 𝛥𝑝𝑓 is combined with the term −𝜔 𝑀𝑁 𝛥𝑝𝑓 which reflects the
𝑛𝐺 𝑛𝐺
effect of changing matric suction on the soil structure.
Combining the global set of equations with Equation 3.23, the governing
equation for unsaturated soils, as implemented by Smith (2003), becomes:
𝐾𝐺 𝐿𝐺 𝛥𝑑 𝑛𝐺 𝛥𝑅𝐺
𝑇 = (3.25)
Ω 𝐿𝐺 −𝛽𝛥𝑡 𝛷𝐺 − 𝜔 𝑀𝑁 𝛥𝑝𝑓 𝑛𝐺 𝑛𝐺 + 𝑄 + 𝛷𝐺 𝑝𝑓 𝛥𝑡
𝑛𝐺
123
3.2.3 𝜴 variation
The parameter Ω in Equation 3.23 governs the volume of water that flows for a
given change in the volume of voids and its variation with suction should,
therefore, be defined.
𝑇
For fully saturated states Ω should be 1.0 so that the term Ω 𝐿𝐺 𝛥𝑑 𝑛𝐺
𝑇
reduces to the fully saturated term 𝐿𝐺 𝛥𝑑 𝑛𝐺 (Potts & Zdravkovic, 1999).
Smith (2003) made the assumption that Ω reduces linearly with suction beyond
the air-entry value 𝑠𝑎𝑖𝑟 and becomes 0.0 when the water phase becomes
discontinuous. The corresponding value of suction is an input parameter,
included in the soil-water retention curve (SWRC) models developed and
implemented in ICFEP.
3.3 Constitutive models for the mechanical behaviour of unsaturated

soils
Based on the Barcelona Basic Model (BBM) introduced by Alonso et al. (1990),
two constitutive models were developed and implemented into ICFEP by
Georgiadis (2003) and are further explained by Georgiadis et al. (2005). The
modifications made to the BBM are discussed in the following sections.
3.3.1 Stress variables
The models are formulated employing the two stress variables introduced in the
previous section; the equivalent stress 𝜎 and the equivalent suction 𝑠𝑒𝑞 .
Introduction of the air-entry value of suction, 𝑠𝑎𝑖𝑟 , in the formulation of the stress
variables (according to Equations 3.1 and 3.2) is one of the main differences
with the BBM, as the transition from full to partial saturation is not necessarily
modelled at zero suction. Until 𝑠𝑎𝑖𝑟 is reached, the effective stress principle is
applicable and the soil behaviour can be entirely described by a sole stress
variable, the effective stress. Beyond 𝑠𝑎𝑖𝑟 , de-saturation of the soil occurs and
the two stress variables presented above are required to model its behaviour.
124
For partly saturated conditions ( 𝑠 > 𝑠𝑎𝑖𝑟 ), the equivalent direct stress
components 𝜎𝑥 , 𝜎𝑦 and 𝜎𝑧 and the shear stress components 𝜏𝑥𝑦 , 𝜏𝑥𝑧 and 𝜏𝑦𝑧 are
used and the constitutive models are formulated in the 4-dimensional stress-
space 𝐽 - 𝑝 - 𝜃 - 𝑠𝑒𝑞 where 𝐽 is the deviatoric stress:
1 2
2
2 2 + 2 ∙ 𝜏2 + 2 ∙ 𝜏2
𝐽= ∙ 𝜎𝑥 − 𝑝 + 𝜎𝑦 − 𝑝 + 𝜎𝑧 − 𝑝 + 2 ∙ 𝜏𝑥𝑦 𝑥𝑧 𝑦𝑧 (3.26)
2
(note that 𝐽 = 𝑞 3 , where 𝑞 = 𝜎1 − 𝜎3 , 𝜎1 and 𝜎3 being the major and minor

principal stresses, respectively, and 𝑞 being the deviatoric triaxial stress), 𝑝 is the
mean equivalent stress:
𝜎𝑥 + 𝜎𝑦 + 𝜎𝑧
𝑝= (3.27)
3
𝜃 is the Lode’s angle:
1 3 3 det 𝑠
𝜃 = − ∙ sin−1 ∙ 3 (3.28)
3 2 𝐽
where, det 𝑠 is the determinant of the stress matrix:
𝜎𝑥 − 𝑝 𝜏𝑥𝑦 𝜏𝑥𝑧
det 𝑠 = 𝜏𝑦𝑥 𝜎𝑦 − 𝑝 𝜏𝑦𝑧 (3.29)
𝜏𝑧𝑥 𝜏𝑧𝑦 𝜎𝑧 − 𝑝
and 𝑠𝑒𝑞 is the equivalent suction.
Under full saturation (𝑠 < 𝑠𝑎𝑖𝑟 or if 𝑢𝑤 is compressive), the model performs in

the 3-dimensional stress space 𝐽 - 𝑝′ - 𝜃, where 𝑝′ is the mean effective stress:
𝜎𝑥′ + 𝜎𝑦′ + 𝜎𝑧′

𝑝′ = (3.30)
3
as the equivalent suction, 𝑠𝑒𝑞 , is equal to zero. The expressions for 𝐽 and 𝜃 are
the same as in Equations 3.26 and 3.28, respectively, but substituting the
effective stresses 𝜎𝑥′ , 𝜎𝑦′ and 𝜎𝑧′ for the equivalent stresses 𝜎𝑥 , 𝜎𝑦 and 𝜎𝑧 .
125
3.3.2 Yield (YS) and plastic potential (PP) surfaces
The two constitutive models developed by Georgiadis (2003) are based on the
concept of the loading-collapse (LC) and the suction increase (SI) curves,
originally introduced by Alonso et al. (1990). Figure 3-1 illustrates the LC and
the SI curves schematically. It is evident that the elastic region expands with
increasing suction and consequently the isotropic yield stress, 𝑝0 , is suction
dependent, as discussed in the following section. Expansion of the elastic
region in tensile equivalent stresses is assumed in order to model the
experimentally observed increase of apparent cohesion with suction.
The elastic region is bounded in the 𝑝 - 𝐽 stress plane by the yield surface 𝐹1 .
The expression adopted by Georgiadis (2003) for the latter, as well as the one
assumed for the plastic potential (𝐺1 ), is similar to that proposed by Lagioia et
al. (1996), extended to include the equivalent suction, 𝑠𝑒𝑞 :
𝐾2
𝜂 𝛽𝑓
𝐹1 𝑝 + 𝑓 𝑠𝑒𝑞 1+𝐾
2
= − = 0.0 (3.31)
𝐺1 𝑝0 + 𝑓 𝑠𝑒𝑞 𝜂
𝐾1
𝛽𝑓
1+𝐾
1
where:
 𝑝0 is the isotropic yield stress at the current value of equivalent suction,
 𝑓 𝑠𝑒𝑞 is a measure of the increase in apparent cohesion with suction

and controls the expansion of the yield surface into the tensile region,
 𝐾1 , 𝐾2 and 𝛽𝑓 are constants calculated by the following equations:
𝜇𝑖 ∙ 1 − 𝛼𝑖 4𝛼𝑖 ∙ 1 − 𝜇𝑖
𝐾1,2 = ∙ 1± 1− (3.32)
2 ∙ 1 − 𝜇𝑖 𝜇𝑖 ∙ 1 − 𝛼𝑖 2
𝛽𝑓 = 1 − 𝜇𝑖 ∙ 𝐾1 − 𝐾2 (3.33)
𝛼𝑖 and 𝜇𝑖 being model parameters that control the shape of the surface,
126
 𝜂 is the generalised normalised stress ratio:
𝜂= 𝐽2𝜂 𝐽2𝜂𝑖 (3.34)
𝐽2𝜂 being the square of the stress ratio:
2
𝐽
𝐽2𝜂 = (3.35)
𝑝 + 𝑓 𝑠𝑒𝑞
and 𝐽2𝜂𝑖 being the failure value of 𝐽2𝜂 , obtained by solving the following
cubic equation, which is based on the Matsuoka-Nakai failure criterion
(Potts & Zdravkovic, 1999):
3 2
2
𝐶𝑖 ∙ sin 3𝜃 ∙ 𝐽2𝜂𝑖 + 𝐶𝑖 − 3 ∙ 𝐽2𝜂𝑖 − 𝐶𝑖 − 9 = 0 (3.36)
27
where:
9 − 𝑀𝑖 2
𝐶𝑖 = (3.37)
2𝑀𝑖 3 𝑀𝑖 2
27 − 3 + 1
𝑀𝑖 being a model parameter which can be calculated from the critical

state value of the angle of shearing resistance:
6 ∙ 𝑠𝑖𝑛𝜑𝑐𝑠
𝑀𝑖 = (3.38)
3 − 𝑠𝑖𝑛𝜑𝑐𝑠
The parameters 𝛼𝑖 , 𝜇𝑖 and 𝑀𝑖 may obtain distinct values for the yield and the
plastic potential surfaces and, therefore, carry the index 𝑓 referring to the former
or 𝑔 referring to the latter. Depending on the values adopted for the above
parameters, the flow rule may be associated or non-associated. Furthermore,
by adjusting these parameters suitably, a wide range of shapes can be
reproduced, as illustrated in Figure 3-2, including the Cam-clay (Roscoe &
Schofield, 1963), the Modified Cam-clay (Roscoe & Burland, 1968), the
127
Sinfonietta Classica (Nova, 1988) and the Single Hardening Yield (YS) (Kim &
Lade, 1988b) and Plastic Potential (PP) (Kim & Lade, 1988a) surfaces.
Similarly, 𝐽2𝜂𝑓 is calculated based on the value of 𝑀𝑓 and 𝐽2𝜂𝑔 is calculated from
𝑀𝑔 . For triaxial compression 𝐽2𝜂𝑖 is equal to 𝑀𝑖2 3. It should be noted that 𝑀𝑔 is
the gradient of the critical state line (CSL) in the equivalent stress, 𝑝 – deviatoric
triaxial stress, 𝑞, space for triaxial compression (𝜃 = −30°).
The increase in apparent cohesion due to suction is taken into account through
the introduction of the function 𝑓 𝑠𝑒𝑞 in Equation 3.31. 𝑓 𝑠𝑒𝑞 represents the
expansion of the yield surface into the tensile equivalent stress region (Figure
3-3) and controls the apparent cohesion, 𝐽𝑐𝑖 :
𝐽𝑐𝑖 = 𝐽2𝜂𝑖 ∙ 𝑓 𝑠𝑒𝑞 (3.39)
𝑓 𝑠𝑒𝑞 may either increase linearly with equivalent suction:
𝑓 𝑠𝑒𝑞 = 𝑘 ∙ 𝑠𝑒𝑞 (3.40)
where 𝑘 is a model parameter, similar to the one adopted by the BBM, or it may
be dependent on the degree of saturation, 𝑆𝑟 , using the following expression:
𝑓 𝑠𝑒𝑞 = 𝑆𝑟 ∙ 𝑠𝑒𝑞 (3.41)
In the latter case, the degree of saturation, 𝑆𝑟 , is obtained from the soil-water
retention curve (SWRC), presented in Section 3.4.
Even though both of the above options are available, the first one is realistic
only for low values of suction. The second option predicts that the apparent
cohesion initially increases with suction, reaches a peak and reduces to a small
value, at large values of suction, as the degree of saturation reaches its
minimum value. Apart from being more realistic, the second option provides a
direct coupling between the mechanical and the hydraulic behaviour of
unsaturated soils. Furthermore, the sum 𝑝 + 𝑓 𝑠𝑒𝑞 becomes equal to Bishop’s
effective stress for 𝜒 = 𝑆𝑟 when 𝑠𝑎𝑖𝑟 is zero:
128
𝑝 + 𝑓 𝑠𝑒𝑞 = 𝑝𝑛𝑒𝑡 + 𝑠𝑎𝑖𝑟 + 𝑆𝑟 ∙ 𝑠𝑒𝑞 (3.42)
where
𝑝𝑛𝑒𝑡 = 𝑝𝑡𝑜𝑡𝑎𝑙 − 𝑢𝑎 (3.43)
and 𝑠𝑒𝑞 is given by Equation 3.2, consequently:
𝑝 + 𝑓 𝑠𝑒𝑞 = 𝑝𝑡𝑜𝑡𝑎𝑙 − 𝑢𝑎 + 𝑆𝑟 ∙ 𝑠 (3.44)
or equally:
𝑝 + 𝑓 𝑠𝑒𝑞 = 𝑝𝑡𝑜𝑡𝑎𝑙 − 𝑢𝑎 + 𝑆𝑟 ∙ 𝑢𝑎 − 𝑢𝑤 (3.45)
which is Bishop’s effective stress with 𝜒 = 𝑆𝑟 (see also Chapter 2).
Georgiadis (2003) included the suction increase (SI) yield surface introduced by
Alonso et al. (1990) in the constitutive models developed, so that suctions larger
than a limit value 𝑠0 generate elasto-plastic strains. It should be noted that the
hardening/softening parameter for this surface is suction 𝑠0 . Assuming an
associated flow rule, the expression for this secondary yield surface (𝐹2 ) and the
corresponding plastic potential (𝐺2 ) is:
𝑠𝑒𝑞
𝐹2 = 𝐺2 = − 1 = 0.0 (3.46)
𝑠0
Equivalent suction, seq (kPa)
Suction Increase
(SI) yield surface
Loading-Collapse
Elastic region (LC) curve
f(seq) p0* p0
Mean equivalent stress, p (kPa)

Figure 3-1: Loading-Collapse (LC) and Suction Increase (SI) surfaces (after Georgiadis, 2003)
129
Mod. Cam-clay
Cam-clay
Sinf. Class.
Single Hard. YS
Deviatoric stress, J (kPa)
Single Hard. PP
f(seq) p0
Figure 3-2: Examples of yield (YS) and plastic potential (PP) functions reproduced by the Lagioia et al.
(1996) expression for unsaturated soils (after Georgiadis, 2003)
J2i
f(seq)J2i
f(seq) p0
Figure 3-3: Apparent cohesion in the 𝑝-𝐽 plane (after Georgiadis, 2003)
3.3.3 Fully saturated isotropic compression line (ICL) and yield stress
At fully saturated conditions (i.e. 𝑠 < 𝑠𝑎𝑖𝑟 ), the isotropic compression line (ICL) is
a straight line, illustrated in Figure 3-4, and is given by the following expression:
𝑣 = 𝑣1 − 𝜆 0 ∙ ln 𝑝′ (3.47)
130
where 𝑣1 is the specific volume at mean effective stress 𝑝′ =1.0 kPa (shown as
𝑣1 (0) in Figure 3-4) and 𝜆 0 is the coefficient of soil compressibility at
saturated conditions (i.e. 𝑠𝑒𝑞 = 0.0).
The isotropic yield stress 𝑝0′ , also shown in Figure 3-4, is the
hardening/softening parameter which controls the size of the yield surface and
governs the magnitude of the plastic volumetric strains. Violation of the yielding
criterion on the wet side of the critical state (sub-critical side) is associated with
expansion of the yield surface (hardening) and an increase of 𝑝0′ , generating
contractive volumetric strains, if a drained analysis is assumed, or positive
excess pore water pressure under undrained conditions. On the other hand,
violation on the dry side of the critical state (super-critical side) is related to
contraction of the yield surface (softening) and a decrease of 𝑝0′ , producing
dilative volumetric strains or negative excess pore water pressure, in case of a
drained or an undrained analysis, respectively.
Mean effective stress, ln p' (kPa)

p' = 1.0 kPa p'0
v1(0)
Specific volume,  ( )
(0)
Ful
1 Iso ly sat
trop ura
ic C ted
om
pre
ss ion
 L ine
1 (IC
L)
Swellin
g Line
Figure 3-4: Fully saturated isotropic compression line (ICL) and swelling line
3.3.4 Unsaturated isotropic compression line (unsat-ICL) and yield stress
For unsaturated conditions the hardening/softening parameter is the equivalent

fully saturated isotropic yield stress 𝑝0∗ , corresponding to zero equivalent
suction, 𝑠𝑒𝑞 , and is, therefore, defined at the transition from full to partial
saturation (Figure 3-1). It should be noted that at the transition point the
equivalent stresses are equal to the effective stresses, since 𝑠𝑒𝑞 = 0.0 kPa and
𝑠 = 𝑠𝑎𝑖𝑟 :
131
𝑝 = 𝑝𝑛𝑒𝑡 + 𝑠𝑎𝑖𝑟 = 𝑝𝑡𝑜𝑡𝑎𝑙 − 𝑢𝑎 + 𝑢𝑎 − 𝑢𝑤 = 𝑝′ (3.48)
Therefore:
𝑝0∗ = 𝑝0′ (3.49)
The relationship between the yield stress at the current value of suction, 𝑝0 , and
the hardening/softening parameter, 𝑝0∗ , determines the shape of the LC curve in
the 𝑝 - 𝑠𝑒𝑞 plane and depends on the expression adopted for the unsaturated
ICL (unsat-ICL).
Based on different assumptions concerning the shape of the latter, Georgiadis

(2003) developed two constitutive models; one adopting a linear or a bi-linear
unsat-ICL and one assuming a non-linear unsat-ICL.
3.3.4.1 Linear and bi-linear unsat-ICL (Model 1)
The assumption of a linear unsat-ICL, constantly diverging from the ICL for fully
saturated conditions ( 𝑠𝑒𝑞 = 0.0 kPa), as illustrated in Figure 3-5 (a), was
introduced by Alonso et al. (1990) in the BBM. The unsat-ICL for 𝑠𝑒𝑞 is given by
the following expression:
𝑣 = 𝑣1 𝑠𝑒𝑞 − 𝜆 𝑠𝑒𝑞 ∙ ln 𝑝 (3.50)
𝑣1 𝑠𝑒𝑞 is the specific volume determined on the current unsat-ICL for 𝑝=1.0
kPa and its calculation is explained in Section 3.3.5. 𝜆 𝑠𝑒𝑞 is the coefficient of
soil compressibility at equivalent suction 𝑠𝑒𝑞 and may be calculated by the
following empirical equation (Alonso et al., 1990):
𝜆 𝑠𝑒𝑞 = 𝜆 0 ∙ 1 − 𝑟 e−𝛽 ∙𝑠𝑒𝑞 + 𝑟 (3.51)
𝑟 and 𝛽 being model parameters which control the shape of the LC curve.
Based on the above expression for the unsat-ICL, Alonso et al. (1990) proved
that the hardening/softening parameter 𝑝0∗ is related to the isotropic yield stress
132
𝑝0 , corresponding to the current value of suction, through the following equation

(LC curve in the isotropic stress space 𝑝 - 𝑠𝑒𝑞 ):
𝜆 0 −𝜅
𝑝0∗ 𝜆 𝑠𝑒𝑞 −𝜅
(3.52)
𝑝0 = 𝑝𝑐
𝑝𝑐
where 𝑝𝑐 is a characteristic pressure, 𝜆 0 is the coefficient of soil

compressibility at fully saturated conditions and 𝜅 is the coefficient of
compressibility along an elastic path and is assumed to be independent of the
suction level. If a soil sample with an isotropic net stress equal to 𝑝𝑐 and initially
on an unsat-ICL is wetted keeping the net stress constant it will only generate
elastic swelling as it moves towards the saturated ICL.
The vertical distance between the unsat-ICL and the ICL at full saturation is a
measure of the potential wetting induced collapse and it evidently increases
linearly with the logarithm of the confining stress, 𝑝 (Figure 3-5 (a)).Despite its
applicability at low confining stresses, the above assumption is thought to
generate unrealistically high magnitudes of potential collapse due to wetting
under high stresses. Therefore, a bi-linear unsat-ICL, illustrated in Figure 3-5
(b), was developed by Georgiadis (2003), making the alternative assumption
that the ratio 𝑝0∗ 𝑝𝑐 = 𝑎𝑐 remains constant at confining stresses higher than 𝑝𝑚 :
𝜆 0 −𝜅
𝑐
𝑝𝑚 = 𝑝 ∙ 𝛼 𝑐 𝜆 𝑠𝑒𝑞 −𝜅 (3.53)
and, therefore, the expression for the LC curve in the isotropic stress space 𝑝 -
𝑠𝑒𝑞 becomes:
𝜆 0 −𝜆 𝑠𝑒𝑞
(3.54)
𝑝0 = 𝑝0∗ ∙ 𝛼 𝑐 𝜆 𝑠𝑒𝑞 −𝜅
where 𝛼 𝑐 is a model parameter substituting for 𝑝𝑐 .
For confining stresses lower than 𝑝𝑚 Equation 3.52 is employed, resulting in a

linear increase of the amount of potential collapse with stress, while for higher
confining stresses Equation 3.54 is adopted, resulting in a constant amount of
potential collapse.
133
Mean equivalent stress, ln p (kPa)

p = 1.0 kPa
v1(0)
v1(seq,1) (a)
amount of
v1(seq,2) potential
collapse v
(seq,2) seq,2
1
(seq,1) seq
,1
1
se
q =
0.0
(0) kPa
1

p = 1.0 kPa pm
v1(0)
(b)
(seq)
1
(0) se
1
q 0
.0 k
Pa
(0) se
1 q =0
.0 k
Pa
Figure 3-5: Unsaturated Isotropic Compression Line (un-ICL); (a) linear unsat-ICL (option 1) and
(b) bi-linear unsat-ICL (option 2) (after Georgiadis, 2003)
3.3.4.2 Non-linear unsat-ICL (Model 2)
Georgiadis (2003) proposed a non-linear unsat-ICL (Figure 3-6), derived from

the variation of the amount of wetting induced potential collapse with the
isotropic yield stress, illustrated in Figure 3-7. The full derivation is further
explained by Georgiadis et al. (2005) and is omitted herein. The equation for the
ICL at full saturation becomes:
𝑝
𝑣 = 𝑁 0 − 𝜆 0 ∙ ln (3.55)
𝑝𝑐
where 𝑁 0 is the value of specific volume 𝑣 at 𝑝 = 𝑝𝑐 , 𝑝𝑐 being the

characteristic pressure defining the limiting lower value of the equivalent fully
saturated isotropic yield stress, 𝑝0∗ .
134
For unsaturated conditions, the non-linear unsat-ICL becomes:
𝑝 𝑝 −𝑏 𝑝
𝑣 = 𝑁 𝑠𝑒𝑞 − 𝜆 0 ∙ ln 𝑐
+ 𝛼0 ∙ 𝑐 ∙ ln (3.56)
𝑝 𝑝 𝑝𝑐
𝑁 𝑠𝑒𝑞 is the specific volume determined on the current unsat-ICL for 𝑝 = 𝑝𝑐

and its calculation is explained in the following section. 𝑏 is a model parameter
and 𝛼0 determines the initial slope of the non-linear unsat-ICL:
𝜆𝑖𝑛 𝑠𝑒𝑞 = 𝜆 0 − 𝛼0 (3.57)
and is, therefore, a measure of the soil stiffness at low confining stresses. 𝛼0 is
dependent on the equivalent suction, 𝑠𝑒𝑞 and assuming that 𝜆𝑖𝑛 𝑠𝑒𝑞 is given by
Equation 3.51, may be obtained by the following expression:
𝛼0 = 𝜆 0 ∙ 1 − 𝑟 ∙ 1 − e−𝛽 ∙𝑠𝑒𝑞 (3.58)
where 𝛽 is a model parameter. The parameter 𝑏 controls the value of the

isotropic yield stress 𝑝𝑚 which corresponds to the maximum potential collapse
(Figure 3-7):
𝑝𝑚 −1
𝑏 = ln (3.59)
𝑝𝑐
Finally, Georgiadis (2003) calculated the relationship between the

hardening/softening parameter, 𝑝0∗ , and the isotropic yield stress 𝑝0 ,
corresponding to the current value of suction, to be:
𝛼 ∙𝑥 −𝑏
1− 0 𝑝0
𝜆 0 −𝜅 (3.60)
𝑝0∗ 𝑐
= −𝑝 ∙ 𝑥 , 𝑥=−
𝑝𝑐
The above expression defines the shape of the LC curve in the isotropic stress
space 𝑝 - 𝑠𝑒𝑞 .
135

p= pc
N(0)
N(seq)
in(seq) 1
amount of
potential
arctan(0) collapse v
se
q =0
.0 k seq
Pa
(0)
1
Figure 3-6: Non-linear unsat-ICL for Model 2 developed by Georgiadis (2003)

Potential plastic reduction of
1
specific volume, p ( )
p, max
pc pm
1/b Isotropic yield stress, ln p0 (kPa)
Figure 3-7: Variation of the amount of potential collapse with isotropic yield stress (after Georgiadis,
2003)
3.3.5 Suction induced wetting/drying line (WDL)
The two models employ a suction induced compression (wetting/drying) line

(WDL), assuming constant elastic and elasto-plastic coefficients of
compressibility with suction, 𝜅𝑠 and 𝜆𝑠 , respectively, as illustrated in Figure 3-8.
Taking into account that wetting of the soil under the characteristic pressure 𝑝𝑐
generates solely elastic swelling, the specific volume 𝑣1 𝑠𝑒𝑞 (Figure 3-9 (a))
corresponding to 𝑝 = 1.0 kPa and to the equivalent suction 𝑠𝑒𝑞 , can be
calculated by the following expression:
𝑠𝑒𝑞 + 𝑝𝑎𝑡𝑚
𝑣1 𝑠𝑒𝑞 = 𝑣1 − 𝜅𝑠 ∙ ln − 𝜆 0 − 𝜆 𝑠𝑒𝑞 ln 𝑝𝑐 (3.61)
𝑝𝑎𝑡𝑚
136
Equivalent suction, ln seq (kPa)

Specific volume,  ( ) s0
s
1
s
Figure 3-8: Wetting/drying line – WDL (suction induced compression line) (after Georgiadis, 2003)

p = 1.0 kPa p c
v1(0)
(a)
v1(seq) ve = sln ((seq+patm)/patm)
seq
(seq)
1
se
q =0
.0 k
(0) Pa
1

p= pc
N(0)
N(seq) (b)
ve =
sln ((seq+patm)/patm)
se
q =0
.0 k
Pa
(0)
1
Figure 3-9: Calculation of (a) 𝑣1 𝑠𝑒𝑞 for Model 1 (Options 1 and 2) and (b) 𝑁 𝑠𝑒𝑞 for Model 2 (after
Georgiadis, 2003)
137
Similarly, for 𝑝 = 𝑝𝑐 (Figure 3-9 (b)):
𝑠𝑒𝑞 + 𝑝𝑎𝑡𝑚
𝑁 𝑠𝑒𝑞 = 𝑁(0) − 𝜅𝑠 ∙ ln (3.62)
𝑝𝑎𝑡𝑚
It should be noted that the coefficients 𝜅𝑠 and 𝜆𝑠 are assumed to be

independent of the stress level.
3.3.6 Critical state line (CSL)
In conventional critical state type constitutive models, developed for fully

saturated soils, such as the modified Cam-clay (MCC) model (Roscoe &
Burland, 1968), critical state is reached along a unique line defined in the 𝑣 –
ln 𝑝 – 𝐽 stress space (Figure 3-10). Similarly, in the models developed by
Georgiadis (2003), a unique critical state line (CSL) is reached for a given value
of equivalent suction, 𝑠𝑒𝑞 . The mean equivalent stress 𝑝𝑐𝑠 , which corresponds to
the critical state point (Figure 3-11), can be calculated by substituting 𝜂 =
𝐽2𝜂𝑔 𝐽2𝜂𝑓 in Equation 3.31 and employing the yield surface parameters 𝛼𝑓 ,
𝜇𝑓 and 𝑀𝑓 :
𝐾2𝑓
𝛽𝑓
𝐽2𝜂𝑔 𝐽2𝜂𝑓
1+ 𝐾2𝑓
𝑝𝑐𝑠 = 𝑝0 + 𝑓 𝑠𝑒𝑞 ∙ 𝐾1𝑓 − 𝑓 𝑠𝑒𝑞 (3.63)
𝛽𝑓
𝐽2𝜂𝑔 𝐽2𝜂𝑓
1+ 𝐾1𝑓
The deviatoric stress at critical state is:
𝐽𝑐𝑠 = 𝐽2𝜂𝑔 ∙ 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 (3.64)
The CSL in the 𝑝 - 𝐽 - 𝑠𝑒𝑞 stress space is shown in Figure 3-12. From the
projections of the CSL in the 𝑝 - 𝑠𝑒𝑞 and 𝐽 - 𝑠𝑒𝑞 planes becomes evident that
both 𝑝𝑐𝑠 and 𝐽𝑐𝑠 increase with suction, as the elastic region bounded by the LC
curve and extended into tensile stresses, expands. Consequently, the position
138
of the CSL depends on both the equivalent suction, 𝑠𝑒𝑞 , and the corresponding
yield stress, 𝑝0 .
In order to fully define the CSL the relationship between the specific volume and
the confining stress is required. Critical state is reached in the 𝑣 - ln 𝑝 stress
space at point B shown in Figure 3-13, following an unloading path from point A
and therefore:
𝑝𝑐𝑠
𝑣𝑐𝑠 = 𝑣 𝑝0 − 𝜅 ln (3.65)
𝑝0
where 𝑣 𝑝0 is the specific volume at the yield stress 𝑝0 , corresponding to the

current value of equivalent suction, 𝑠𝑒𝑞 , and is defined on the congruent unsat-
ICL:
𝑣 𝑝0 = 𝑣1 𝑠𝑒𝑞 − 𝜆 𝑠𝑒𝑞 ∙ ln 𝑝0 (3.66)
The calculation of 𝑣1 𝑠𝑒𝑞 has already been explained in Section 3.3.5.
Since the CSL was shown to depend on the yield stress, 𝑝0 , its shape and
position vary with the relationship adopted for the unsat-ICL as shown in Figure
3-14.
Figure 3-10: Critical state line (CSL) for modified Cam-clay (Roscoe & Burland, 1968)
139
J2f
Deviatoric stress, J (kPa) J2g

Jcs,PP
CS point
PP
YS
f(seq) pcs p0 pP
Figure 3-11: Critical state point in the 𝑝 - 𝐽 stress plane
Figure 3-12: Critical state line (CSL) in the 𝑝 - 𝐽 - 𝑠𝑒𝑞 stress space
p = 1.0 kPa pcs p0
v1(seq)
unsa
B t-ICL
vcs
v(p0)  A
1
unsa (seq)
t-CS 1
L
Figure 3-13: Calculation of the specific volume 𝑣𝑐𝑠 (after Georgiadis, 2003)
140

CS Model 1; Optio
L n1
sat
u rate
d Mo
M del
od 1; O
el ptio
2 n2
Figure 3-14: Comparison of the CSL’s predicted by the two models for full and partial saturation (after
Georgiadis, 2003)
3.3.7 Hardening/softening rules
Violation of either of the two yield surfaces, 𝐹1 and/or 𝐹2 , results in the

𝑝
generation of plastic volumetric strains, 𝜀𝑣 , the magnitude of which is
associated with the change in the hardening/softening parameters 𝑝0∗ and 𝑠0
according to the following hardening/softening rules:
𝑑𝑝0∗ 𝑣 𝑝
∗ = 𝑑𝜀𝑣 (3.67)
𝑝0 𝜆 0 −𝜅
for the primary surface 𝐹1 and
𝑑𝑠0 𝑣 𝑝
= 𝑑𝜀𝑣 (3.68)
𝑠0 + 𝑝𝑎𝑡𝑚 𝜆𝑠 − 𝜅𝑠
for the SI surface 𝐹2 .
The above two equations imply that the surfaces are coupled and therefore
violation and movement of either of them induces movement of the other, as:
𝑑𝑝0∗ 𝜆𝑠 − 𝜅𝑠 𝑑𝑠0
∗ = ∙ (3.69)
𝑝0 𝜆 0 − 𝜅 𝑠0 + 𝑝𝑎𝑡𝑚
141
3.3.8 Elastic behaviour
For changes in equivalent suction, 𝑠𝑒𝑞 , and in equivalent stress, 𝜎, within the
elastic region, the following elastic volumetric strains are generated:
𝑒
𝜅𝑠
𝑑𝜀𝑣𝑠 =− 𝑑𝑠𝑒𝑞 (3.70)
𝑣 ∙ 𝑠𝑒𝑞 + 𝑝𝑎𝑡𝑚
for elastic wetting/drying and:
𝑒
𝜅
𝑑𝜀𝑣𝑝 =− 𝑑𝑝 (3.71)
𝑣∙𝑝
for elastic loading/unloading.
In order to avoid calculation of infinite volumetric strains when 𝑝 tends to 0.0, a

minimum bulk modulus ( 𝑣𝑝 𝜅 ) 𝐾𝑚𝑖𝑛 was introduced as an extra model
parameter.
Elastic changes of the deviatoric stress, 𝑑𝐽, are associated with the incremental
deviatoric strains 𝑑𝐸𝑑𝑒 (note that 𝐸 is used to be consistent with terminology of
Potts & Zdravkovic, 1999) through the elastic shear modulus 𝐺:
𝑑𝐽
𝑑𝐸𝑑𝑒 = (3.72)
𝐺
The shear and bulk moduli 𝐺 and 𝐾 are related through Poisson’s ratio 𝜇 ,
according to the following relationship:
3(1 − 2𝜇)
𝐺= 𝐾 (3.73)
2(1 + 𝜇)
For the elastic behaviour to be fully defined, one of the parameters 𝐺 or 𝜇 needs
to be prescribed (along with the parameters 𝜅 and 𝜅𝑠 ), while the other one is
allowed to vary with stress level since the bulk modulus 𝐾 is stress dependent.
Alternatively, a constant ratio of 𝐺/𝑝0 may be assumed, where 𝑝0 is the
isotropic yield stress corresponding to the current value of equivalent suction.
142
3.3.9 Initialisation of the hardening parameters
At the beginning of a numerical analysis, the hardening parameter 𝑝0∗ needs to

be initialised based on the initial stresses and the overconsolidation ratio, OCR,
which is an input parameter. The following options were implemented by
Georgiadis (2003):
 𝑝0∗ = OCR ∙ 𝑝 (3.74)
where 𝑝 is the absolute value of the initial mean equivalent stress,
 𝑝0 = OCR ∙ 𝑝 (3.75)
where 𝑝0 is subsequently used in the calculation of the hardening

parameter 𝑝0∗ as explained in Section 3.3.4,
 ∗
𝜎𝑦𝑚𝑜 = OCR ∙ 𝜎𝑦
(3.76)
where 𝜎𝑦 is the absolute value of the vertical equivalent stress (principal

∗
stress). 𝜎𝑦𝑚𝑜 is the equivalent fully saturated vertical stress,
corresponding to a stress state on the yield surface for zero equivalent
suction, 𝑠𝑒𝑞 , and is directly employed in the calculation of 𝑝0∗
 𝜎𝑦𝑚𝑜 = OCR ∙ 𝜎𝑦 (3.77)
where 𝜎𝑦𝑚𝑜 is the vertical stress (principal stress) corresponding to a

stress state on the current partly saturated yield surface and is used in
the calculation of 𝑝0 .
For the latter two options, the rest of the stress components are calculated as:
∗ ∗ ∗ ∗ ∗ ∗
𝜎𝑥𝑚𝑜 = 𝜎𝑧𝑚𝑜 = (1 − sin 𝜑)𝜎𝑦𝑚𝑜 and 𝜏𝑥𝑦𝑚𝑜 = 𝜏𝑥𝑧𝑚𝑜 = 𝜏𝑦𝑧𝑚𝑜 = 0.0 (3.78)
and
𝜎𝑥𝑚𝑜 = 𝜎𝑧𝑚𝑜 = (1 − sin 𝜑)𝜎𝑦𝑚𝑜 and 𝜏𝑥𝑦𝑚𝑜 = 𝜏𝑥𝑧𝑚𝑜 = 𝜏𝑦𝑧𝑚𝑜 = 0.0 (3.79)
143
The equivalent fully saturated isotropic yield stress, 𝑝0∗ , or the mean equivalent
isotropic yield stress, 𝑝0 , for the current value of equivalent suction can then be
evaluated.
In addition to the above options, the following four options were implemented as
part of this project:
 𝑝0∗ = 𝑝 + 𝑥 (3.80)
 𝑝0 = 𝑝 + 𝑥 (3.81)
 ∗
𝜎𝑦𝑚𝑜 = 𝜎𝑦 + 𝑥 (3.82)
and
 𝜎𝑦𝑚𝑜 = 𝜎𝑦 + 𝑥 (3.83)
where 𝑥 is an input parameter. These new options are particularly useful when
modelling the effect of a roller compaction stress applied during the construction
of each layer of an embankment.
The second hardening parameter, 𝑠0 , which controls the initial position of the SI
yield surface, 𝐹2 , is an input parameter and may be obtained from an
unconfined drying test. For soils for which the SI surface is irrelevant, a value of
𝑠0 larger than the maximum suction level anticipated should be prescribed.
Finally, the initial specific volume also needs to be evaluated:
𝑝0
𝑣 = 𝑣1 𝑠𝑒𝑞 − 𝜆 𝑠𝑒𝑞 ln 𝑝0 + 𝜅 ln (3.84)
𝑝
3.3.10 Summary of the model parameters
A total of twenty-two parameters (Table 3-1) is required by each one of the

models developed by Georgiadis (2003) and presented herein. For the
calibration of the models an adequate testing programme is needed and should
consist of the following laboratory experiments:
144
 one fully saturated undrained triaxial compression test to obtain the

parameters 𝑀𝑔 , 𝑀𝑓 and 𝐺, in the situation where known shapes for the
yield and plastic potential surfaces are adopted. Additional triaxial
compression tests would be required, both on the wet and on the dry
side of the critical state, if the above surfaces are to be calibrated in
order to derive the parameters 𝛼𝑔 , 𝜇𝑔 , 𝛼𝑓 and 𝜇𝑓 ;
 one fully saturated isotropic loading and unloading test to obtain the
parameters 𝜆(0), 𝜅, 𝑣1 and 𝑝0∗ ;
 two isotropic loading tests at different values of suction to establish the

parameters 𝑟, 𝛽 and 𝑝𝑐 (Model 1; Option 1) or 𝛼𝑐 (Model 1; Option 2) or 𝑏
(Model 2);
 one unconfined drying/wetting cycle to determine the soil-water retention

curve and the parameters 𝑠𝑎𝑖𝑟 , 𝑠0 , 𝜅𝑠 and 𝜆𝑠 ;
 one unsaturated drained triaxial compression test to determine the

cohesion increase parameter 𝑘 unless the increase in apparent cohesion
is assumed to vary with the degree of saturation, 𝑆𝑟 .
It should be noted that the laboratory tests listed above represent the minimum
required for the calibration of the models. To increase the reliability of the
calibration or to test the adequacy of the models in realistically reproducing soil
behaviour, a considerably larger number of experiments might be necessary.
145
Table 3-1: Summary of the model input parameters (after Georgiadis, 2003)
Unsaturated soil constitutive models (Georgiadis, 2003)
Parameter Description Parameter Description

Coefficient of
𝛼𝑔 𝜆(0) compressibility for fully
Plastic Potential function saturated conditions
parameters Elastic Coefficient of
𝜇𝑔 𝜅 compressibility (along
elastic paths)
Slope of the CSL in the q-p Specific volume at unit
𝑀𝑔 space for triaxial 𝑣1 pressure for fully saturated
compression conditions
Maximum soil stiffness

𝛼𝑓 𝑟 parameter
Soil stiffness increase

𝜇𝑓 Yield surface parameters 𝛽 parameter
Coefficient of
𝑀𝑓 𝜆𝑠 compressibility for suction
changes
Elastic coefficient of
Characteristic pressure
𝑝𝑐 (Model 1 – Option 1) 𝜅𝑠 compressibility for suction
changes
Characteristic stress ratio

𝛼𝑐 (Model 1 – Option 2)
𝑝𝑎𝑡𝑚 Atmospheric pressure
Maximum collapse
𝑏 parameter 𝐾𝑚𝑖𝑛 Minimum bulk modulus
(Model 2)
𝑠0 Yield suction 𝐺 or 𝜇 or Shear modulus or

𝐺/𝑝0 Poisson’s ratio
Overconsolidation ratio or
𝑠𝑎𝑖𝑟 Air-entry value of suction roller stress
employed in the calculation
OCR or 𝑥 of the equivalent fully
Cohesion increase
𝑘 parameter saturated isotropic yield
Constant or 𝑆𝑟 stress 𝑝0∗
146
3.4 Soil-water retention curve (SWRC) model
The soil-water retention curve (SWRC) existing in ICFEP at the commencement

of the research presented herein was the non-hysteretic, void-ratio
independent, non-linear curve shown in Figure 3-15 (Melgarejo, 2004). The
model is based on the work of Van Genuchten (1980) and is formulated in
terms of degree of saturation, 𝑆𝑟 , and matric suction, 𝑠:
𝑚
1
𝑆𝑟 = 𝑛
1 − 𝑆𝑟,0 + 𝑆𝑟,0 (3.85)
1 + 𝛼 𝑠 − 𝑠𝑑𝑒𝑠
where:
 s is the current value of suction
 𝑠𝑑𝑒𝑠 is the value of suction at de-saturation
 𝑆𝑟,0 is the residual degree of saturation
 𝛼, 𝑛 and 𝑚 are fitting parameters controlling the shape of the curve. Note
that 𝛼 > 0.0, 𝑛 > 0.0 and 0.0 ≤ 𝑚 ≤ 1.0. The dimension of parameter 𝛼 is
1/stress (i.e. kPa-1) so that the product 𝑠 ∙ 𝛼 is dimensionless
The slope of the retention curve at the current value of suction, 𝑠, is:
𝑛−1
𝜕𝑆𝑟 𝛼 𝑠 − 𝑠𝑑𝑒𝑠
𝑅= = −𝑚𝑛𝛼 ∙ 1 − 𝑆𝑟0 ∙ 𝑛 𝑚 +1
(3.86)
𝜕𝑠 1 + 𝛼 𝑠 − 𝑠𝑑𝑒𝑠
From Figure 3-15 it is evident that the minimum possible degree of saturation,
𝑆𝑟 , may be other than 0.0 and it is, therefore, possible to model a residual
degree of saturation, 𝑆𝑟,0 , which is asymptotically reached when the suction
tends to infinity, as implied by Equation 3.85. However, a value of suction 𝑠0 can
be identified for which the degree of saturation, 𝑆𝑟 , practically reduces to its
residual value. This suction, however, should not be misinterpreted as being the
suction at shrinkage limit (no further change in void ratio), which would have
been the case had the retention curve been plotted in terms of void ratio.
According to Smith (2003), the above mentioned suction is rather the suction at
which the water phase seemingly becomes discontinuous. Nonetheless,
147
discontinuity of bulk water is theoretically expected to occur at an earlier stage

upon drying and any further changes in the degree of saturation and the
corresponding suction can be attributed to the meniscus water.
The above inconsistency is practically overcome by the introduction of the Ω

parameter, which, as explained by Smith (2003), governs the volume of water
that flows due to a change in the volume of voids and reduces to 0.0 when the
water phase becomes discontinuous (Section 3.2.3). The suction at Ω transition,
𝑠Ω , is an input parameter and may well be smaller than 𝑠0 . The program,
however, checks that 𝑠Ω > 𝑠𝑎𝑖𝑟 , 𝑠𝑎𝑖𝑟 being the air-entry value of suction, which is
an extra model parameter. Furthermore, the value of 𝑠𝑎𝑖𝑟 prescribed for the
SWRC model needs necessarily to agree with the value input in the elasto-
plastic constitutive model adopted.
Evidently, the formulation of the above model assumes a unique retention curve
for any given soil, ignoring the effect of changes in void ratio and the hydraulic
hysteresis exhibited upon cycles of drying and wetting, which were explained in
Chapter 2. These aspects of the retention curve were modelled as part of the
current project and are extensively discussed in Chapter 5.
1.0
Degree of saturation, Sr ( )
Sr,0
sdes sair ssl s0
Matric suction, log s (kPa)
Figure 3-15: Simple non-hysteretic, void-ratio independent, non-linear SWRC implemented in ICFEP by
Melgarejo (2004)
148
3.5 Permeability models
Permeability, which is essential when performing fully coupled consolidation

analyses, is unlikely to remain constant, as it is believed to depend on the void
ratio or equivalently, on the stress level and, in the case of unsaturated flow, on
the degree of saturation or the suction level. Consequently, various permeability
models are available in ICFEP and are thought to be adequate for the purposes
of the analyses presented in this thesis.
In addition to models assuming isotropic or anisotropic permeability, with or

without spatial variation, some models allow the permeability to vary during a
coupled analysis, as a result of one or more of the following factors:
 de-saturation due to suction
 desiccation due to tensile total stresses
 void ratio or stress level changes
When modelling the effect of de-saturation on soil permeability, two options are
readily available in ICFEP; permeability varying (a) with suction and (b) with the
degree of saturation.
The main assumption involved in the first case is that the logarithm of
permeability varies linearly with suction from its initial value 𝑘𝑠𝑎𝑡 , corresponding
to suction 𝑠1 , to a limiting value 𝑘𝑚𝑖𝑛 , corresponding to suction 𝑠2 , as illustrated
in Figure 3-16. The magnitude of permeability corresponding to the current
suction level 𝑠 can, therefore, be obtained from the following equation:
𝑠 − 𝑠1 𝑘𝑠𝑎𝑡
log 𝑘 = log 𝑘𝑠𝑎𝑡 − log (3.87)
𝑠2 − 𝑠1 𝑘𝑚𝑖𝑛
For values of suction smaller than 𝑠1 the permeability is equal to 𝑘𝑠𝑎𝑡 , whereas
for suction levels higher than 𝑠2 the permeability is equal to 𝑘𝑚𝑖𝑛 .
The above expression is thought to be adequate only when monotonic changes

of suction are applied or for soils exhibiting insignificant hydraulic hysteresis, as
a hysteretic relationship between permeability and suction is believed to exist,
149
similarly to the retention behaviour (Liakopoulos, 1965 reported in Fredlund &

Rahardjo, 1993). However, the permeability-water content or degree of
saturation relationship shows essentially no hysteresis, as discussed by
Fredlund & Rahardjo (1993), and therefore, the following expression should
preferably be employed in combination with the hysteretic retention curve
developed during the course of this project and presented in Chapter 5:
2
𝑚 𝑚
𝑘 = 𝑘𝑠𝑎𝑡 ∙ 𝛩 1 2
∙ 1 − 1 − 𝛩1 (3.88)
where:
𝛩= (3.89)
1 − 𝑆𝑟0
𝑆𝑟0 being the residual degree of saturation and 𝑚 a fitting parameter (Van
Genuchten, 1980).
Under the effect of suction, tensile total stresses, arising when the tensile pore
pressure becomes larger than the compressive effective stresses, may lead to
instigation of desiccation cracks within the soil, increasing its overall mass
permeability. The effect of desiccation is modelled by adopting the following
expression (Nyambayo, 2003):
𝜎𝑇 − 𝜎𝑇1 𝑘𝑚𝑎𝑥
log 𝑘 = log 𝑘𝑠𝑎𝑡 + log (3.90)
𝜎𝑇2 − 𝜎𝑇1 𝑘𝑠𝑎𝑡
where 𝜎𝑇 is the current tensile total principal stress and 𝜎𝑇1 and 𝜎𝑇2 are the
tensile stresses illustrated in Figure 3-17. More specifically, 𝜎𝑇1 is the tensile
stress at which the permeability in saturated conditions starts to increase until
the current tensile stress reaches the limit 𝜎𝑇2 .
Desiccation has preference over de-saturation and before investigating whether

the latter occurs, occurrence of the former is checked within the program and
the permeability is calculated accordingly.
The magnitude of permeability, before de-saturation or desiccation take place,

is equal to 𝑘𝑠𝑎𝑡 and can either be prescribed and maintained constant
150
throughout the analysis, or may vary with void ratio or stress level changes
according to one of the three expressions given below (Potts & Zdravkovic,
1999):
 𝑘 = exp 𝛼 + 𝑏 ∙ 𝑒 (3.91)
where 𝑒 is the current void ratio and 𝛼 and 𝑏 are fitting parameters, or
 𝑘 = 𝑘0 ∙ exp 𝛼 ∙ 𝑝′ (3.92)
where 𝑘0 is the magnitude of permeability for zero effective stress 𝑝′ and

𝛼 is a fitting parameter, or
 𝑘 𝑘0 = 𝑝 ′ −𝛼 (3.93)
The last expression is employed only if 𝑝′ > 1.0, otherwise the program
sets 𝑘 𝑘0 equal to 1.0 and Equation 3.93 is ignored.
It should be noted that rather than prescribing the magnitude of 𝑘𝑚𝑖𝑛 and 𝑘𝑚𝑎𝑥 ,
the program requires the ratios 𝑘𝑠𝑎𝑡 𝑘𝑚𝑖𝑛 and 𝑘𝑚𝑎𝑥 𝑘𝑠𝑎𝑡 . In this way, if one of
the above relationships for 𝑘𝑠𝑎𝑡 is adopted (Equations 3.91, 3.92 or 3.93), the
limits 𝑘𝑚𝑖𝑛 and 𝑘𝑚𝑎𝑥 are proportionally affected.
Soil permeability, log k ( )
ksat
kmin
s1 s2
Matric suction, s (kPa)

Figure 3-16: Assumed variation of soil permeability with suction (after Nyambayo, 2003)
151
Soil permeability, log k ( )
kmax
ksat
T1 T2
Tensile total stress, T (kPa)
Figure 3-17: Increase in soil permeability due to desiccation cracks (after Potts & Zdravkovic, 1999)
3.6 Hydraulic boundary conditions
For coupled problems, it is necessary to specify either a pore pressure or a

prescribed nodal flow for each node on the boundary of the mesh (Potts &
Zdravkovic, 1999). The default condition in ICFEP is that of a zero nodal flow.
As Potts & Zdravkovic (1999) explain, prescribed values of incremental nodal

pore pressure 𝛥𝑝𝑓 affect only the left hand side of the system of equations
𝑛𝐺
3.16 (for saturated conditions) or 3.25 (for unsaturated conditions) and are
treated in a similar way as prescribed displacements. Prescribed nodal flow
rates affect the right hand side vector 𝑄 of the system and are dealt with in a
fashion comparable to prescribed nodal forces.
From the various hydraulic boundary conditions available in ICFEP (such as

sources, sinks, infiltration, tied degrees of freedom, precipitation and
vegetation), the precipitation and the vegetation conditions are detailed herein,
as they are necessary in understanding the analyses presented in Chapters 6
and 7.
152
3.6.1 Precipitation boundary condition
The precipitation boundary condition is essentially a dual condition that is

applied on the nodes along a boundary and enables the simulation of rainfall on
the soil surface; it may operate either as an infiltration condition, by specifying a
constant inflow rate, or as a constant pore water pressure condition.
Consequently, for the determination of the boundary condition, a flow rate, 𝑞𝑛 ,
and a pore pressure, 𝑝𝑓𝑏 , need to be specified. The prescribed flow rate may be
set equal to the required rainfall, while the pore pressure may be set equal to
𝑝𝑓𝑏 .
The pore water pressure at boundary nodes 𝑝𝑓 at the beginning of every

increment is compared to 𝑝𝑓𝑏 . If it is found to be more tensile than the 𝑝𝑓𝑏 , an
infiltration boundary condition is applied, employing the specified flow rate, 𝑞𝑛 .
Infiltration is also applied when at the beginning of the increment the flow rate
across the boundary exceeds the prescribed value.
Alternatively, if the pore pressure at the beginning of the increment is more

compressive than the 𝑝𝑓𝑏 , then a constant pore water pressure equal to the
latter is imposed. In order to maintain the prescribed pore pressure at the
boundary, a portion of the specified infiltration is applied, while the rest is
considered as run-off and, since the resulting flow occurs outside the finite
element mesh, is disregarded.
Typically, the 𝑝𝑓𝑏 is set to 0.0 kPa, allowing for the soil to become fully
saturated as a result of intense rainfall, but preventing compressive pore water
pressures larger than 0.0 to build up at the ground surface. Nonetheless, the
latter situation is desirable when modelling surface ponding and, therefore,
compressive values of 𝑝𝑓𝑏 are accepted by the program. Finally, tensile values
of 𝑝𝑓𝑏 can be employed to prevent total loss of suction.
The operation of the boundary condition described above is schematically

illustrated in Figure 3-18 (after Smith, 2003). The main advantage of adopting
the precipitation rather than the infiltration boundary condition is that ponding
and run-off cannot be modelled with the latter.
153
Low rainfall intensity: PWP distribution at start of increment

Infiltration
(surface PWP
more tensile pfb
than pfb)
tensile,
+ve
compressive,
-ve
High rainfall intensity: PWP distribution at start of increment

Ponding
(pfb compressive) or
pfb
Run-off
(surface PWP=pfb) tensile,
+ve
compressive,
-ve
Figure 3-18: Precipitation boundary condition (after Smith, 2003)
154
3.6.1.1 Automatic-incrementation algorithm for the precipitation boundary

condition
Employment of the precipitation boundary condition is not straightforward;

depending on the flow rate and on the soil permeability, conditions may change
during the increment, for one or more nodes, so that a switch of boundary
condition is required. In that case the program splits the current increment into
sub-increments according to the automatic-incrementation algorithm described
by Smith (2003).
For example, if at the beginning of an increment a prescribed flow boundary

condition had been considered appropriate but at the end of the increment the
calculated pore water pressure is more compressive than the threshold value
𝑝𝑓𝑏 , indicating that the boundary condition changed to a prescribed pore
pressure, the increment is automatically cut down to a portion of the original one
and the process is repeated for the new sub-increment. If at the end of the sub-
increment the boundary condition obtained is still incompatible with the initial
condition, the increment is further subdivided. In the opposite case, the rest of
the increment is applied and the consistency of the boundary condition at the
beginning and the end of the second sub-increment is checked. The procedure
is repeated until the entire original increment has been applied.
The algorithm is described in detail by Smith (2003) and its operation is

schematically illustrated in Figure 3-19. A tolerance (denoted as tol in the figure)
is specified around the 𝑝𝑓𝑏 and is employed when comparison between the
latter and the pore pressure occurs, at the end of every sub-increment. Starting
an increment with a prescribed flow boundary condition, as in the example
above, if the pore pressure at the end of the increment is found to be more
compressive than the 𝑝𝑓𝑏 and lies outside the tolerance zone (point A, in Figure
3-20), the increment is rejected and subdivided.
If at the end of the sub-increment, the pore pressure is found to be more tensile
than the 𝑝𝑓𝑏 and lies outside the tolerance zone (point B, in Figure 3-20), the
infiltration boundary condition remains as is and the rest of the increment is
applied (sub-increment 2). Assuming that at the end of sub-increment 2 the
155
pore pressure is found to be more compressive than the 𝑝𝑓𝑏 and lies outside the
tolerance zone (point C, in Figure 3-20), the sub-increment is once more
rejected and further subdivided. The process is repeated until the pore pressure
at the end of the sub-increment lies within the tolerance zone (point D, in Figure
3-20) and the boundary condition is changed to a constant pore pressure
condition on the next sub-increment, the pressure being equal to the 𝑝𝑓𝑏 .
Once an increment or a sub-increment is rejected, a new sub-increment is

calculated and its size is evaluated as a portion of the failed increment, as
illustrated in Figure 3-21; the size of the rejected increment is multiplied by the
ratio X/Y, where X is the difference between the initial boundary pore pressure
and the 𝑝𝑓𝑏 and Y is the overall change in the pore pressure calculated during
the failed increment. This implies that a linear variation of pore water pressure
throughout the increment is assumed and, therefore, the above method rarely
gives a sufficiently accurate result immediately. However, the algorithm
continues to adjust the size of the sub-increments, progressively reducing the
error, up to the moment the change in boundary condition occurs within the
determined tolerance.
It should be noted that the automatic-incrementation capability existed in ICFEP

for stress-strain behaviour (Potts & Zdravkovic, 1999) and was expanded by
Smith (2003) to address pore water pressures. Adjusting the load step
automatically, the error generated during the non-linear finite element analysis
is reduced, offering an improved method of modelling non-linear behaviour. The
procedure is particularly useful when determining the factor of safety against
slope stability, as further discussed in Chapter 6.
156
Figure 3-19: Schematic operation of the automatic-incrementation procedure implemented for the
precipitation boundary condition (after Smith, 2003)
157
pfb
pfb+tol pfb-tol
Tension +ve Compression -ve
B D C A
PWP distribution
Depth with depth
Figure 3-20: The tolerance zone for the precipitation boundary condition (after Smith, 2003)
Y
Tension +ve pfb Compression -ve
X
PWP distribution at the end of

PWP distribution at the start of the rejected sub-increment
the sub-increment
Old (rejected) sub-increment size:

100%
New sub-increment size:

100%*(X/Y) Depth
Figure 3-21: Determination of sub-increment size during application of the precipitation boundary
condition (after Smith, 2003)
158
3.6.1.2 Other applications of the precipitation boundary condition
Although the precipitation boundary condition enables the simulation of rainfall

on the ground surface, its application is not restricted to that. The boundary
condition is also useful in a fully coupled analysis when excavating elements
beneath the phreatic surface or as a recharge model.
At the end of an excavation, which may well be of a tunnel or of an open cut,

the pore water pressure in the surrounding soil could be tensile. Prescribing
zero pore pressure at the nodes of the newly formed boundary of the active
mesh for the subsequent increments, unrealistically results in the generation of
flow of water from the boundary into the soil. If, on the other hand, a no flow
boundary condition is applied, the above mentioned tensile pore water
pressures reduce with time due to swelling and finally become compressive.
Even though the second approach seems more reasonable, unrealistic
magnitudes of pore pressure may build up in the long-term.
To overcome this problem, the precipitation boundary condition can be

alternatively employed, determining zero flow and zero 𝑝𝑓𝑏 . In this way, a no
flow condition is adopted in the first increments following the excavation and
remains for as long as the pore pressures in the surrounding soil are tensile.
Once the latter become compressive, the condition is switched to a constant
pore pressure equal to zero (i.e. the 𝑝𝑓𝑏 value), while the change of boundary
condition may occur at different increments of the analysis for the individual
nodes. The example of the excavation of a tunnel (Figure 3-22) is discussed in
detail by Potts & Zdravkovic (1999) but similar circumstances apply to the
excavation of an open cut.
Smith (2003) presents the additional theoretical example of a slope where

seasonal changes of the depth of the phreatic surface may occur. The
precipitation boundary condition is applied along the base of the mesh, setting
the inflow rate equal to the permeability of the soil underlying the mesh and the
𝑝𝑓𝑏 to the maximum permitted compressive pore pressure. The latter should be
consistent with the maximum height of the phreatic surface under ‘normal’
conditions, assuming a hydrostatic profile. The phreatic surface is free to lower
159
during dry periods and to rise under intense waterfall (Figure 3-23). In the
former case, water is drained down under gravity and the pore pressures,
therefore, reduce below the 𝑝𝑓𝑏 , resulting in inflow being prescribed at the
bottom boundary. In this way a continuous recharge is modelled which
maintains a deep phreatic surface. Wetter periods tend to raise the water table
and a constant pore pressure condition, equal to 𝑝𝑓𝑏 , applies at the bottom
boundary.
The precipitation boundary condition was employed in the analyses presented

in Chapters 6 and 7 of the thesis, to simulate the effect of rainfall as well as to
model excavations below the phreatic surface.
Figure 3-22: Precipitation boundary condition in tunnel problem (after Potts & Zdravkovic, 1999)
160
Figure 3-23: Precipitation as a recharge model (after Smith, 2003)
3.6.2 Vegetation boundary condition
Water demand for transpiration varies throughout the calendar year, peaking
during the summer and dropping during the winter, modifying the pore
pressures within the ground accordingly. In order to predict the induced volume
change in a boundary value problem, seasonal pore pressure profiles may be
prescribed, matching field observations. Nonetheless, this simplification may
lead to unrealistic water flows which cannot be achieved by vegetation. Root
161
water uptake models (RWUM), simulating transpiration itself, provide a more

accurate approach.
Nyambayo & Potts (2010) implemented a non-linear RWUM into ICFEP,

allowing the pore pressures in a coupled analysis to be predicted from the
available evapotranspiration data rather than being prescribed. The RWUM
existing in ICFEP does not involve plant-specific parameters and can, therefore,
be applied to a wide range of vegetation types.
Extraction of water due to vegetation is assumed to vary linearly with depth,

from a maximum value corresponding to the ground surface, to zero at depth
𝑟𝑚𝑎𝑥 , as illustrated in Figure 3-24. Below the maximum root depth, 𝑟𝑚𝑎𝑥 , the
water uptake is assumed to be zero.
For the water uptake to be accounted for in a numerical analysis, a sink term
𝑆𝑚𝑎𝑥 , which is calculated based on the potential evapotranspiration rate 𝑇𝑝
prescribed, is incorporated in the continuity equation of fluid flow (Potts &
Zdravkovic, 1999). Nonetheless, Nyambayo & Potts (2010) clarify that 𝑇𝑝
relates to how much water can be taken from the ground if the supply of
moisture were constant and note that the actual transpiration is a fraction of the
potential one. Therefore, the above mentioned sink term 𝑆𝑚𝑎𝑥 is reduced in
order to obtain the actual one, 𝑆𝑎𝑐𝑐 , according to the relationship introduced by
Feddes et al. (1978):
2 ∙ 𝛼 ∙ 𝑇𝑝 𝑟
𝑆𝑎𝑐𝑐 = 𝛼 ∙ 𝑆𝑚𝑎𝑥 = ∙ 1− (3.94)
𝑟𝑚𝑎𝑥 𝑟𝑚𝑎𝑥
where
 𝑇𝑝 is the potential transpiration rate prescribed
 𝑟𝑚𝑎𝑥 is the maximum root depth for which the boundary condition applies
and which also needs to be set
 𝑟 refers to the depth below the ground surface (Figure 3-24) and cannot
exceed the value 𝑟𝑚𝑎𝑥 (i.e. 0 ≤ 𝑟 ≤ 𝑟𝑚𝑎𝑥 )
162
 𝛼 is a suction dependent function as shown in Figure 3-25. 𝑎 is zero for

suction levels lower than S1 (anaerobiosis point) and larger than S4
(wilting point) and equal to 1.0 for suction levels between S2 and S3. It
increases linearly between S1 and S2 and decreases linearly from S3 to
S4. Suctions S1, S2, S3 and S4 are model parameters, input as
transpiration material properties. As Nyambayo & Potts (2010) explain,
the generally accepted magnitudes for S1, S2 and S4 are 0 kPa, 5 kPa
and 1500 kPa, respectively, while the value of S3 is not expected to have
significant influence on the overall soil behaviour and is, therefore,
usually considered to be 50 kPa. It is interesting to note that, as suction
does not necessarily vary linearly with depth, neither does 𝑆𝑎𝑐𝑐 .
At the beginning of each increment, the numerical code identifies the integration
points within the root zone and calculates their depth 𝑟 as the shortest distance
to the surface boundary. Based on the pore pressure at the end of the previous
increment, a value of 𝛼 is evaluated and is subsequently substituted, along with
the potential transpiration rate 𝑇𝑝 , into Equation 3.94. Multiplying the sink term
𝑆𝑎𝑐𝑐 by the time step corresponding to the current increment, an estimate of the
flow at the integration point is obtained. In order to estimate the equivalent
nodal flow, a volume integral is performed over the whole element, assuming
that the integration points outside the root zone have zero flow.
The estimated nodal flows are then used, together with the other boundary
conditions, to perform the first iteration of the modified Newton-Raphson
process (Potts & Zdravkovic, 1999). However, at the end of the iteration the
pore pressures have changed affecting the value of 𝛼 and, therefore, the sink
term. A new estimate of the equivalent nodal flows, accounting for the variation
of 𝛼 over the increment (see Nyambayo & Potts, 2010) is then calculated. This
is then compared to the nodal flows initially evaluated (i.e. before the pore
pressures were updated) and the difference gives the out-of-balance nodal
flows, which are used in the next iteration. The process is repeated until
convergence is achieved and the computations for the next increment take
place.
163
The procedure described above is expected to be mesh dependent, since the

water flows at the integration points need to be integrated over the
corresponding elements. Nyambayo & Potts (2010) studied the sensitivity of the
model to mesh density and concluded that the element thickness should not
exceed the maximum root depth 𝑟𝑚𝑎𝑥 (for element thickness to 𝑟𝑚𝑎𝑥 ratios of
1.0, 0.5 and 0.25 similar numerical results were yielded in terms of
displacements and pore pressures).
The vegetation boundary condition was employed by Nyambayo (2003) in the

boundary value problem of a railway embankment founded on London clay, and
was shown to be essential for modelling the progressive failure frequently
encountered in this type of geotechnical structures.
Potential evapotranspiration
Ground level
Depth, r
User defined
active root Assumed shape
depth (rmax) of root water
extraction
function
(Smax)
Figure 3-24: Assumed shape of root extraction function in the rooted zone (after Nyambayo & Potts,
2010)
 function for transpiration
1.0
0.0
S1 S2 S3 S4
Matric suction, s (kPa)
Figure 3-25: Linear variation of 𝑎 function (after Nyambayo & Potts, 2010)
164
The existing capabilities of ICFEP, relevant to numerical analysis of unsaturated

soils, were presented in the current chapter. Whereas some of them were
employed intact in the analyses presented in later chapters of this thesis
(permeability models and hydraulic boundary conditions), the constitutive
models and the soil-water retention curve were modified, as explained in the
following two chapters. Nonetheless, the aspects of ICFEP explained herein
formed the basis of the present research and as such, are essential for the
comprehension of the work undertaken.
Two elasto-plastic constitutive models, developed and implemented by

Georgiadis (2003) for the simulation of the soil behaviour under partial
saturation, were available in ICFEP at the commencement of this project. The
sole difference of the two models consists in the shape of the isotropic
compression lines assumed under unsaturated states (unsat-ICL); the first of
the models allows either a linear or a bi-linear unsat-ICL to be employed,
whereas the second one utilises a non-linear compression line. The latter option
is believed to be more realistic as it is derived from the variation of potential
wetting induced collapse with the isotropic yield stress.
Two stress variables are used in the formulation of the models – the equivalent
stress, 𝜎, and the equivalent suction, 𝑠𝑒𝑞 – and are both dependent on the air-
entry value of suction, 𝑠𝑎𝑖𝑟 . In this way, the transition from full to partial
saturation is modelled to occur at suction 𝑠𝑎𝑖𝑟 .
The models adopt the concept of loading-collapse (LC) and suction increase
(SI) curves, which were initially introduced by Alonso et al. (1990). The
Matsuoka-Nakai failure criterion (Potts & Zdravkovic, 1999) is adopted in the
deviatoric plane and the Lagioia et al. (1996) expression, extended to
unsaturated states, is employed for the yield and plastic potential functions in
the 𝑝 - 𝐽 stress plane. Through appropriate adjustment of the parameters
associated with the Lagioia et al. (1996) expression, a wide variety of shapes
can be reproduced and the flow rule may be non-associated.
165
Each of the models requires 22 parameters to be prescribed and a testing

programme consisting of both fully and partly saturated, isotropic and triaxial
tests is needed for their calibration. In case the increase of apparent cohesion
with suction is coupled with the soil-water retention curve (SWRC), one
unconfined drying/wetting cycle is also necessary for the determination of this
curve.
The SWRC model available in ICFEP was implemented by Melgarejo (2004)

and is based on the Van Genuchten (1980) expression. The latter was altered
to include a residual degree of saturation, 𝑆𝑟,0 . The effect of void ratio and the
hydraulic hysteresis are neglected.
Various expressions exist within ICFEP to model water permeability and its
variation and were employed in the analyses presented in Chapters 6 and 7.
Permeability is allowed to vary during coupled analyses, either because of de-
saturation due to suction or as a result of desiccation cracks formed due to
tensile total stresses.
When modelling the effect of de-saturation, two options are available;

permeability varying with suction or with the degree of saturation. The former
option is adequate when monotonic changes of suction are induced, whereas
the latter is more appropriate when cyclic changes are involved. This is so
because, although the permeability-suction relationship is believed to be
hysteretic – similar to the retention behaviour – the permeability-degree of
saturation relationship exhibits no hysteresis (Fredlund & Rahardjo, 1993).
Permeability is modelled to increase under the effect of desiccation cracks,

formed when the tensile pore pressures become larger than the compressive
effective stresses. Desiccation has preference over de-saturation, when the two
are simultaneously employed, and before investigating whether the latter
occurs, occurrence of the former is checked within the program and the
permeability is accordingly varied.
Furthermore, the magnitude of permeability, before de-saturation or desiccation

take place, can either be prescribed and maintained constant throughout an
analysis, or may vary with void ratio or stress changes. In the latter case, the
166
minimum and maximum limiting values of permeability, associated with de-

saturation and desiccation, respectively, are proportionally affected.
From the various hydraulic boundary conditions available in ICFEP, the

precipitation and vegetation boundary conditions were presented in this chapter
and are subsequently used in the analyses in Chapters 6 and 7.
The precipitation boundary condition is a dual condition, applied on the nodes

along a boundary; it may operate either as an infiltration condition, by specifying
a constant flow rate, or as a constant pore water pressure condition, allowing
the flow to vary. Although the condition is employed to simulate the effect of
rainfall and consequent infiltration or run-off/ponding, its use is not restricted to
that; it may be applied when modelling excavations below the phreatic surface
or as a recharge model. Additionally, the automatic-incrementation algorithm
(Smith, 2003) needs to be simultaneously employed, as a switch of boundary
condition (prescribed flow or pore pressure) may be predicted during an
increment.
The vegetation boundary condition, in combination with a root-water uptake

model (RWUM), was implemented by Nyambayo (2003) to model the effect of
potential evapotranspiration rate. The latter is assumed to vary linearly with
depth within the root zone, defined by the input parameter 𝑟𝑚𝑎𝑥 . Nonetheless,
the actual transpiration is a portion of the potential one, and needs to be
evaluated based on the suction level. Root water uptake is assumed to be zero
at suctions below the anaerobiosis point S1 and above the permanent wilting
point S4, while is at its maximum and constant between points S2 and S3. A
linear variation is assumed between S1 and S2 and between S3 and S4. The
potential evapotranspiration rate and suction at points S1, S2, S3 and S4 also
need to be prescribed.
The herein described code capabilities, along with those newly developed and
presented in the subsequent chapters, enable coupled analysis of unsaturated
soils, subjected to complex boundary conditions, to be performed. In the
present thesis, these developments are applied to the numerical analysis of
boundary value problems involving the long-term stability of excavated slopes
(Chapter 6) and the behaviour of excavated slopes under seasonal changes of
167
suction (Chapter 7), highlighting the importance of adopting appropriate

constitutive models.
168
chapter 4: MODELLING OF OVERCONSOLIDATED
UNSATURATED SOILS
4.1 Introduction
The critical state type constitutive models for unsaturated soils, described in the
previous chapter, have been shown to be inaccurate in the prediction of the soil
behaviour on the dry side of the critical state. The elastic region is too large
resulting in a considerable overestimation of the peak deviatoric stress, as
yielding is predicted at higher stresses and strains than those observed in
laboratory experiments.
In the present chapter, an alternative to the Lagioia et al. (1996) formulation,

termed the Hvorslev surface, is proposed in order to replace both the yield and
the plastic potential surfaces on the dry side of the critical state. Following the
formulation, implementation and validation of the new surface, analyses of
laboratory experiments on artificial silt are presented, demonstrating the
improved simulation of soil behaviour. The inadequacy of the Lagioia et al.
(1996) formulation to reproduce the behaviour on the dry side of the critical
state is explored based on the same laboratory data.
169
Modelling of Overconsolidated Unsaturated Soils
4.2 Formulation of the Hvorslev surface for unsaturated soils
One of the techniques usually employed in order to prevent critical state type
constitutive models from over-predicting the peak deviatoric stress at highly
overconsolidated states, is to adopt a planar surface to replace the yield surface
on the dry side. This surface is commonly termed after Hvorslev, who in 1937
demonstrated that the failure envelope in the 𝑞-𝑝′ plane for overconsolidated
soils can be adequately approximated by a straight line, which is the projection
of the Hvorslev surface onto this plane.
The option of using the Hvorslev surface is already available in ICFEP (Potts &
Zdravkovic, 1999) in combination with the Modified Cam-clay model (MCC
model), which was initially developed by Roscoe & Burland (1968). While the
typical ellipse is used to represent the yield and plastic potential functions on
the wet side, the expressions on the dry side of the critical state have been
replaced by the straight line illustrated in Figure 4-1, in the mean effective
′
stress, 𝑝΄ - deviatoric stress, 𝐽 , plane. The critical state point, (𝑝𝑐𝑠 , 𝐽𝑐𝑠 ) , is
common for the Hvorslev surface and for the ellipse so that no discontinuities
are present. Formulation of the Hvorslev yield surface is then straightforward as
its inclination 𝛼𝐻𝑉 , which is an input parameter, and the critical state point are
both known:
′ ′
𝐽 𝑝𝑐𝑠 𝑝𝑐𝑠 𝛼𝐻𝑉
𝐹= − − 1 − =0 (4.1)
𝑝′ 𝑔 𝜃 𝑝′ 𝑝′ 𝑔 𝜃
where:
𝑠𝑖𝑛𝜑′
𝑔 𝜃 =
1 (4.2)
𝑐𝑜𝑠𝜃 + 𝑠𝑖𝑛𝜃 ∙ 𝑠𝑖𝑛𝜑′
3
is the inclination of the critical state line (CSL) in the 𝑝΄ – 𝐽 plane, 𝜑′ being the
angle of shearing resistance.
A non-associated flow rule is considered and the gradient of the flow vector, 𝛽,
is assumed to vary linearly from the initial input value, 𝛽𝐻𝑉 , corresponding to
𝑝΄ = 0 kPa, to zero (vertical direction) at the critical state, where 𝑝′ = 𝑝𝑐𝑠
′
. In
170
this way, the flow vector predicted by the ellipse on the wet side and by the
Hvorslev surface on the dry side is vertical at the critical state ensuring
continuity. The expression for the plastic potential surface is:
′ ′
𝐽 𝑝𝑐𝑠 𝑝𝑐𝑠 − 𝑝𝑐′ 𝛼𝐻𝑉 𝑝′ − 𝑝𝑐′ 𝛽𝐻𝑉 𝑝𝑐𝑠′
− 𝑝𝑐′
𝐺= ′ − ′ + ∙ − ∙ ∙ (4.3)
𝑝𝑔 𝜃 𝑝 𝑝′ 𝑔 𝜃 𝑝′ 𝑔 𝜃 ′
𝑝𝑐𝑠
where, 𝑝𝑐′ refers to the current stress point C on the Hvorslev surface (shown in
Figure 4-1).
4.2.1 Yield surface
Introducing the Hvorslev surface on the dry side of the constitutive models for
unsaturated soils developed by Georgiadis (2003) has proven to be challenging,
since the flow rule on the wet side of the critical state may well be non-
associated. Attaching the new surface to the Lagioia et al. (1996) expression for
the plastic potential can possibly result in discontinuity of the yield function
between the dry and the wet sides, as illustrated in Figure 4-2 (a). Nevertheless,
associating the Hvorslev surface to the apparent critical state point defined by
the yield function, corresponding to 𝑝 𝐶𝑆,𝑌𝑆 in Figure 4-2 (b), introduces an
inconsistency to the CSL itself, as the inclination of the latter is specified to be
𝐽2𝜂𝑔 according to the plastic potential of the wet side but is forced by the
Hvorslev surface to be altered to 𝐽2𝜂𝑓 on the dry side. Furthermore, the CSL is
commonly defined by the inclination 𝐽2𝜂𝑔 rather than the inclination 𝐽2𝜂𝑓 , so
the critical state point 𝑝 𝐶𝑆,𝑌𝑆 shown in Figure 4-2 (b) is not valid. It is, therefore,
evident that the Hvorslev surface needs to be associated to the Lagioia et al.
(1996) expression for the yield function in such a way that continuity of the
inclination of the CSL is also guaranteed.
To facilitate the above requirements, the Hvorslev surface is attached to the

yield surface at the point of intersection with the CSL, as illustrated in Figure 4-2
(c). The model maintains its flexibility in terms of reproducible shapes for the
surfaces on the wet side while exhibiting consistency and robustness.
The mean equivalent stress corresponding to the critical state point, 𝑝𝑐𝑠 , is
calculated from the Lagioia et al. (1996) expression (Equation 3.31, Chapter 3),
171
using a stress ratio equal to 𝐽2𝑛𝑔 and the model parameters 𝛼𝑓 and 𝜇𝑓 , which
control the shape of the yield surface on the wet side:
𝐾2𝑓
𝐽2𝜂𝑔 𝛽𝑓
𝐽2𝜂𝑓
1+ 𝐾2𝑓
𝑝𝑐𝑠 = 𝑝0 + 𝑓 𝑠𝑒𝑞 𝐾1𝑓 − 𝑓 𝑠𝑒𝑞 (4.4)

𝐽2𝜂𝑔 𝛽𝑓
𝐽2𝜂𝑓
1+ 𝐾1𝑓
where 𝐾2𝑓 , 𝐾1𝑓 , and 𝛽𝑓 are given by Equations 3.32 and 3.33, in Chapter 3, and
𝑝0 is the yield stress at the current value of equivalent suction, 𝑠𝑒𝑞 . The function
𝑓(𝑠𝑒𝑞 ) controls the increase of apparent cohesion with suction, as explained in
Chapter 3.
The corresponding deviatoric stress is:
𝐽𝑐𝑠 = 𝐽2𝑛𝑔 ∙ 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 (4.5)
The straight line passing through the point (𝑝𝑐𝑠 , 𝐽𝑐𝑠 ) with an inclination 𝛼𝐻𝑉 , is
given by the following equation:
𝐽 − 𝐽𝑐𝑠 = 𝛼𝐻𝑉 𝑝 + 𝑓 𝑠𝑒𝑞 − 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 (4.6)
where (𝑝, 𝐽) is the current stress point. Rearranging the above equation:
𝐽− 𝐽2𝑛𝑔 ∙ 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 − 𝛼𝐻𝑉 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞 + 𝛼𝐻𝑉 ∙ 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 =0 (4.7)
𝐽 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 𝛼𝐻𝑉 𝛼𝐻𝑉 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞

− − + ∙ =0 (4.8)
𝐽2𝑛𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞 𝑝 + 𝑓 𝑠𝑒𝑞 𝐽2𝑛𝑔 𝐽2𝑛𝑔 𝑝 + 𝑓 𝑠𝑒𝑞
the yield surface expression is calculated to be:
172
𝐽 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 𝛼𝐻𝑉

𝐹= − − 1− =0 (4.9)
𝐽2𝑛𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞 𝑝 + 𝑓 𝑠𝑒𝑞 𝑝 + 𝑓 𝑠𝑒𝑞 𝐽2𝑛𝑔
For saturated conditions 𝑓(𝑠𝑒𝑞 ) = 0 kPa and Equation 4.9 is reduced to an

equation similar to Equation 4.1.
For the Hvorslev surface to lie above the CSL and softening to be predicted, it is
necessary that 𝛼𝐻𝑉 ≤ 𝐽2𝑛𝑔 . If 𝛼𝐻𝑉 = 𝐽2𝑛𝑔 , the Hvorslev surface coincides
with the CSL as the above equation becomes:
𝐹 = 𝐽 − 𝐽2𝑛𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞 (4.10)
which is the equation of the CSL in the mean equivalent stress, 𝑝 – deviatoric
stress, 𝐽, plane.
Whereas a no-tension switch is available for the MCC model, which restricts the
effective stress, 𝑝′ , from obtaining illegal negative values, no such switch is
provided for the current model as the stress states should not be prevented
from becoming tensile. In fact, in order to model the increase of apparent
cohesion with suction, the mean equivalent stress, p, is allowed to become
negative under the effect of suction, up to the limiting value of 𝑝 = − 𝑓(𝑠𝑒𝑞 )
(assuming that 𝑠𝑎𝑖𝑟 = 0.0 kPa and 𝑓(𝑠𝑒𝑞 ) = 𝑆𝑟 𝑠𝑒𝑞 , Bishop’s effective stress with
𝜒 = 𝑆𝑟 becomes equal to zero when 𝑝 = − 𝑓(𝑠𝑒𝑞 )). However, in the absence of
any provision, the stress state is incorrectly allowed to exist within the grey area
of Figure 4-3.
Various ways of overcoming similar problems exist, such as employing a

secondary yield surface to be used as a cut-off at low stresses. A new approach
was adopted as part of the present research programme, ensuring continuity of
the yield function on the dry side, as well as continuity of the plastic potential
function, which will be subsequently introduced. Equation 4.7 can be
alternatively rearranged to:
173
𝐽 − 𝑝 + 𝑓 𝑠𝑒𝑞 ∙ 𝑎𝐻𝑉 − 𝐽2𝑛𝑔 − 𝑎𝐻𝑉 ∙ 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 =0 (4.11)
and therefore:
𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞

𝐽∙ − ∙ 𝑝 + 𝑓 𝑠𝑒𝑞 ∙ 𝑎𝐻𝑉 −
𝑝 + 𝑓 𝑠𝑒𝑞 𝑝 + 𝑓 𝑠𝑒𝑞
(4.12)
𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
− 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 ∙ 𝐽2𝑛𝑔 − 𝑎𝐻𝑉 ∙ =0
𝑝 + 𝑓 𝑠𝑒𝑞
Note that in the above equation 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 ∙ 𝐽2𝑛𝑔 − 𝑎𝐻𝑉 = 𝑥, where 𝑥 is
the distance shown in Figure 4-3. Introducing a fitting input parameter 𝑛 to the
𝑝 𝑐𝑠 +𝑓 𝑠𝑒𝑞
term , which is multiplied by the distance 𝑥 in Equation 4.12, allows the
𝑝+𝑓 𝑠𝑒𝑞
Hvorslev surface to obtain the shape shown in Figure 4-4. Equation 4.9 is thus
altered to:
𝑛
𝐽 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
− 𝑎𝐻𝑉 − 𝐽2𝑛𝑔 − 𝑎𝐻𝑉 ∙ =0 (4.13)
𝑝 + 𝑓 𝑠𝑒𝑞 𝑝 + 𝑓 𝑠𝑒𝑞
Normalising by 𝐽2𝑛𝑔 , the yield function finally becomes:
𝑛
𝐽 𝑎𝐻𝑉 𝑎𝐻𝑉 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
𝐹= − − 1− ∙ =0 (4.14)
𝐽2𝑛𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞 𝐽2𝑛𝑔 𝐽2𝑛𝑔 𝑝 + 𝑓 𝑠𝑒𝑞
where 0 ≤ 𝑛 < 1.
For 𝑛 = 0:
𝐽 𝑎𝐻𝑉 𝑎𝐻𝑉
𝐹= − − 1− =0 (4.15)
𝐽2𝑛𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞 𝐽2𝑛𝑔 𝐽2𝑛𝑔
which can be reduced to the equation of the CSL (Equation 4.10). Furthermore,
for 𝛼𝐻𝑉 = 𝐽2𝑛𝑔 , the Hvorslev surface coincides with the CSL, independently of
the value of 𝑛.
174
For 𝑛 = 1:

𝐹= − − 1− ∙ =0 (4.16)
Equation 4.16 can be written in the same form as Equation 4.9 implying that for
𝑛 = 1, the shape of the Hvorslev surface is a straight line. Clearly, it should be
𝑛 < 1, in order to overcome the above mentioned shortcoming. If a value of
𝑛 = 1 is used, the program gives an error and the analysis is stopped.
In Figure 4-5 the Hvorslev surface obtained for various values of 𝑛 is presented.
For 𝑛 equal to 0.0, the Hvorslev coincides with the CSL, as discussed above.
For 0.0 < 𝑛 < 1.0, the Hvorslev surface is curved and its position is dragged
upwards for increasing values of 𝑛 until it becomes a straight line for 𝑛 = 1.0.
Values of 𝑛 outside the range 0,1 result in abnormal shapes of the yield
surface lying either below the critical state line or above the original straight line
given by Equation 4.9.
The main disadvantage of the new expression for the yield surface is that
Equation 4.14 cannot be defined for 𝑝 = −𝑓 𝑠𝑒𝑞 and therefore requires
numerical tolerances to be set (for 𝐹 in Equation 4.14 to be dimensionless
division by 𝑝 + 𝑓 𝑠𝑒𝑞 is necessary). The new expression, however, holds
significant advantages as it prevents illegal stress states to be reached,
employing a unique surface along the entire dry side of the critical state. The
new surface shares a common point with the yield surface used on the wet side,
for 𝑝 = 𝑝𝑐𝑠 , ensuring continuity of the two surfaces.
Indeed, for 𝑝 = 𝑝𝑐𝑠 , Equation 4.14 gives:
𝐽= 𝐽2𝑛𝑔 ∙ 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 (4.17)
The above calculated deviatoric stress, 𝐽 , and the corresponding mean

equivalent stress, 𝑝𝑐𝑠 , given by Equation 4.4, satisfy the Lagioia et al. (1996)
expression for the yield surface (Equation 3.31, Chapter 3):
175
𝐾2𝑓
𝛽𝑓
𝐽2𝜂𝑔
𝐽2𝜂𝑓
1+ 𝐾2𝑓
𝑝0 + 𝑓 𝑠𝑒𝑞 ∙ 𝐾1𝑓
𝛽𝑓
𝐽2𝜂𝑔 𝐾2𝑓 (4.18)
𝐽2𝜂𝑓 𝛽𝑓
𝜂
1+ 1+𝐾
𝐾1𝑓 2𝑓
− 𝐾1𝑓 =0
𝑝0 + 𝑓 𝑠𝑒𝑞 𝛽𝑓
𝜂
1+𝐾
1𝑓
as the stress ratio, 𝜂, is:
𝐽 𝐽2𝑛𝑔 ∙ 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞

𝐽2𝜂 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 𝐽2𝜂𝑔 (4.19)
𝜂= = = =
𝐽2𝜂𝑓 𝐽2𝜂𝑓 𝐽2𝜂𝑓 𝐽2𝜂𝑓

g()
Jcs c
c e
u rfa
)
s
SL
l ev
(C
or s C
e
HV v
l in
H
te
HV
ta
ls
ca
i
rit
C
pcs' p0'
Effective stress, p' (kPa)

Figure 4-1: The yield surface (YS) on the wet and on the dry side of the critical state for the MCC model
(Potts & Zdravkovic(1999))
176
J2f J2g (a)

ce
urfa
s
ev
o r sl PP
Hv
YS
HV
f(seq) pcs,YS pcs,PP p0

J2f J2g (b)

e
ac
s urf PP
vle
v o rs
H
HV YS
f(seq) pcs,YS pcs,PP p0

J2f J2g (c)

ce
urfa PP
s
l ev
or s HV
Hv
YS
f(seq) pcs pcs,PP p0
Figure 4-2: The Hvorslev surface in relation to the yield (YS) and plastic potential (PP) functions;
(a)attached to the plastic potential (PP) surface;(b) attached to the yield surface (YS); (c) attached to the
actual critical state point on the yield surface (YS)
177
J2g

ce
u rfa
vs
le
v ors
H 
HV
f(seq) pcs p0
Figure 4-3: Area of illegal stress states allowed by a straight line Hvorslev surface (shown in grey)
J2g
f(seq) pcs p0
Figure 4-4: New shape of the Hvorslev surface, introduced to bound the mean equivalent stress, p, within
acceptable limits
HV n=1.0
J2g
n=0.9
n=0.7
n=0.3
n=0.0
f(seq) pcs p0
Figure 4-5: The Hvorslev surface for different values of the parameter n
178
4.2.2 Plastic potential surface
The plastic potential function corresponding to the improved version of the

Hvorslev yield surface, given by Equation 4.14, is such that the gradient of the
flow vector varies from an initial value of 𝛽𝐻𝑉 for 𝑝 = −𝑓 𝑠𝑒𝑞 , to a final value of
0.0 at the critical state, where 𝑝 = 𝑝𝑐𝑠 . This variation is no longer linear but is
controlled by an input parameter 𝑚 as subsequently explained.
The flow vector at the current stress state point C 𝑝𝑐 , 𝐽𝑐 , shown in Figure 4-6,
is assumed to have a gradient of 𝛽𝑐 . The flow vector is necessarily
perpendicular to the plastic potential surface and to its tangent, shown in the
figure. The equation for the tangent is:
𝐽 − 𝐽𝑝𝑝 − 𝑝 + 𝑓 𝑠𝑒𝑞 ∙ 𝛽𝑐 = 0 (4.20)
where 𝐽𝑝𝑝 is the value of 𝐽 for 𝑝 + 𝑓 𝑠𝑒𝑞 = 0. The current deviatoric stress is
therefore:
𝐽𝑐 = 𝐽𝑝𝑝 + 𝑝𝑐 + 𝑓 𝑠𝑒𝑞 ∙ 𝛽𝑐 (4.21)
The current deviatoric stress calculated from the yield function (Equation 4.14)
is:
𝑛
𝐽𝑐 = 𝑎𝐻𝑉 ∙ 𝑝𝑐 + 𝑓 𝑠𝑒𝑞 + 𝐽2𝑛𝑔 − 𝑎𝐻𝑉 ∙ 𝑝𝑐 + 𝑓 𝑠𝑒𝑞 ∙ (4.22)
𝑝𝑐 + 𝑓 𝑠𝑒𝑞
𝐽𝑝𝑝 can be evaluated from the two above equations to be:
𝐽𝑝𝑝 = 𝑎𝐻𝑉 ∙ 𝑝𝑐 + 𝑓 𝑠𝑒𝑞 +
𝑛
𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 (4.23)
+ 𝐽2𝑛𝑔 − 𝑎𝐻𝑉 ∙ 𝑝𝑐 + 𝑓 𝑠𝑒𝑞 ∙ − 𝑝𝑐 + 𝑓 𝑠𝑒𝑞 ∙ 𝛽𝑐
𝑝𝑐 + 𝑓 𝑠𝑒𝑞
Substituting the above expression into Equation 4.20 and normalising by
𝐽2𝑛𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞 , the following equation is obtained:
179
𝐽 𝑎𝐻𝑉 𝑝𝑐 + 𝑓 𝑠𝑒𝑞
− ∙ −
𝐽2𝑛𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞 𝐽2𝑛𝑔 𝑝 + 𝑓 𝑠𝑒𝑞
𝑛
(4.24)
𝑎𝐻𝑉 𝑝𝑐 + 𝑓 𝑠𝑒𝑞 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 𝑝 − 𝑝𝑐 𝛽𝑐
− 1− ∙ ∙ − ∙ =0
𝐽2𝑛𝑔 𝑝 + 𝑓 𝑠𝑒𝑞 𝑝𝑐 + 𝑓 𝑠𝑒𝑞 𝑝 + 𝑓 𝑠𝑒𝑞 𝐽2𝑛𝑔
where 𝛽𝑐 varies according to the following expression:
𝑚
𝑝𝑐𝑠 − 𝑝𝑐
𝛽𝑐 = 𝛽𝐻𝑉 ∙ (4.25)
Finally, the plastic potential function is:
𝐺= − ∙ −
𝑛
𝑎𝐻𝑉 𝑝𝑐 + 𝑓 𝑠𝑒𝑞 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
− 1− ∙ ∙ −
𝐽2𝑛𝑔 𝑝 + 𝑓 𝑠𝑒𝑞 𝑝𝑐 + 𝑓 𝑠𝑒𝑞 (4.26)
𝑚
𝑝 − 𝑝𝑐 𝛽𝐻𝑉 𝑝𝑐𝑠 − 𝑝𝑐
− ∙ ∙ =0
𝑝 + 𝑓 𝑠𝑒𝑞 𝐽2𝑛𝑔 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
For 𝑚 = 1, 𝑛 = 1 and for 𝑓 𝑠𝑒𝑞 = 0 (fully saturated conditions), the expression

for the plastic potential becomes similar to Equation 4.3.
The variation of the gradient of the flow vector, 𝛽, with the mean equivalent
stress, 𝑝, for different values of 𝑚 is shown in Figure 4-7. For 𝑚 = 1 a linear
variation is predicted. It should be noted that 𝑚 cannot obtain negative values.
Due to a lack of experimental evidence indicating otherwise, the parameters

𝛼𝐻𝑉 and 𝛽𝐻𝑉 were assumed to be independent of the suction level. However,
this assumption might be oversimplified and ideally a series of triaxial tests,
performed under various values of constant suction, of samples swelled to
different values of OCR, would produce the data required to fully understand
and model the behaviour of highly overconsolidated unsaturated soils.
180
The revised version of the Hvorslev surface for predicting the behaviour on the
dry side of the critical state requires four extra parameters to be defined, two
more than those demanded by the Hvorslev surface of the MCC model (𝛼𝐻𝑉
and 𝛽𝐻𝑉 ). The parameter 𝑛 was introduced in order to restrict the stress states
to within the acceptable limits, while the parameter 𝑚 adds flexibility to the
model regarding the prediction of the plastic volumetric behaviour, as it is
demonstrated in a subsequent section of the present chapter (4.4.4 Effect of the
parameter 𝒎). Despite these two extra fitting parameters which improve the
robustness of the model, the surface is thought to maintain its simplicity.
A similar surface was developed for the modified Cam-clay model. Its
development and implementation are presented in Appendix A.
c J2g
JPP
f(seq) pc pcs p0
Figure 4-6: Flow vector for the current stress state denoted by point C 𝑝𝑐 , 𝐽𝑐
Gradient of the flow vector, 
HV
m = 0.3
m = 0.5
m = 1.0
m = 2.0
m = 3.0

f(seq) pcs
Figure 4-7: Variation of the gradient of the flow vector, 𝛽, with the mean equivalent stress, p
181
4.3 Implementation of the Hvorslev surface
The sign convention that ICFEP employs agrees with the one used in structural
mechanics, where tension is positive. Even though the development of the
Hvorslev surface for unsaturated states was presented in the previous section
in terms of the conventional signs applied in soil mechanics, its implementation
into ICFEP is explained in the current section in the tension positive convention
(note that in all subsequent figures for the validation and calibration of the
model the soil mechanics sign convention was employed, i.e. compression is
positive). Therefore, the compressive mean equivalent stress, 𝑝 , in the
expressions for the yield and the plastic potential surfaces, need to bear a
minus sign and consequently Equations 4.14 and 4.26 become:
𝑛
𝐹=− − − 1− ∙ =0 (4.27)
𝐺=− − ∙ −
𝑛
𝑎𝐻𝑉 𝑝𝑐 + 𝑓 𝑠𝑒𝑞 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
− 1− ∙ ∙ −
𝐽2𝑛𝑔 𝑝 + 𝑓 𝑠𝑒𝑞 𝑝𝑐 + 𝑓 𝑠𝑒𝑞 (4.28)
𝑚
𝑝 − 𝑝𝑐 𝛽𝐻𝑉 𝑝𝑐𝑠 − 𝑝𝑐
− ∙ ∙ =0
𝑝 + 𝑓 𝑠𝑒𝑞 𝐽2𝑛𝑔 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
As explained in Chapter 3, solution of the governing finite element equation

(Equation 3.15) for nonlinear problems, where the incremental stiffness matrix
depends on the current stress and strain levels, is not straightforward and a
non-linear solution technique is required. ICFEP employs the modified Newton-
Raphson method, with a sub-stepping stress point algorithm (Potts &
Zdravkovic, 1999), and the following terms need to be calculated in order to
evaluate both the global stiffness matrix and the residual load vector 𝜓 ′ :
182
4.3.1 Yield function derivatives
The two following vectors need to be calculated:
𝜕𝐹 𝜕𝐹 𝜕𝑝 𝜕𝐹 𝜕𝐽 𝜕𝐹 𝜕𝜃
= + + ∙ (4.29)
𝜕𝜎 𝜕𝑝 𝜕𝜎 𝜕𝐽 𝜕𝜎 𝜕𝜃 𝜕𝜎
𝜕𝐹 𝑇
𝜕𝐹
= 1 1 1 0 0 0 ∙ (4.30)
𝜕𝑠𝑒𝑞 𝜕𝑠𝑒𝑞
The derivatives of the stress invariants, needed for the first of the above
vectors, are:
𝑇
𝜕𝑝
=− 1 3 1 3 1 3 0 0 0 (4.31)
𝜕𝜎
𝑇
𝜕𝐽 𝜎𝑥 − 𝑝 𝜎𝑦 − 𝑝 𝜎𝑧 − 𝑝 𝜏𝑥𝑦 𝜏𝑥𝑧 𝜏𝑦𝑧
= (4.32)
𝜕𝜎 2𝐽 2𝐽 2𝐽 𝐽 𝐽 𝐽
𝜕𝜃 3 3 𝑑𝑒𝑡 𝑠 𝜕𝐽 𝜕 𝑑𝑒𝑡 𝑠
= − (4.33)
𝜕𝜎 2 ∙ 𝑐𝑜𝑠 −3𝜃 ∙ 𝐽3 𝐽 𝜕𝜎 𝜕𝜎
where:
2
𝑑𝑒𝑡 𝑠 = −𝜎𝑥 − 𝑝 ∙ −𝜎𝑦 − 𝑝 ∙ −𝜎𝑧 − 𝑝 − −𝜎𝑥 − 𝑝 𝜏𝑦𝑧 −
2 2
(4.34)
− −𝜎𝑦 − 𝑝 𝜏𝑥𝑧 − −𝜎𝑧 − 𝑝 𝜏𝑥𝑦 + 2𝜏𝑥𝑧 ∙ 𝜏𝑦𝑧 ∙ 𝜏𝑥𝑦
𝜕 𝑑𝑒𝑡 𝑠 1 2 2 2
=− 𝜎𝑦 + 𝑝 𝜎𝑧 − 𝜎𝑥 + 𝜎𝑧 + 𝑝 𝜎𝑦 − 𝜎𝑥 − 2𝜏𝑦𝑧 + 𝜏𝑥𝑧 + 𝜏𝑥𝑦 (4.35)
𝜕𝜎𝑥 3
𝜕 𝑑𝑒𝑡 𝑠 1 2 2 2
=− 𝜎 + 𝑝 𝜎𝑧 − 𝜎𝑦 + 𝜎𝑧 + 𝑝 𝜎𝑥 − 𝜎𝑦 − 2𝜏𝑥𝑧 + 𝜏𝑦𝑧 + 𝜏𝑥𝑦 (4.36)
𝜕𝜎𝑦 3 𝑥
𝜕 𝑑𝑒𝑡 𝑠 1 2 2 2
=− 𝜎 + 𝑝 𝜎𝑦 − 𝜎𝑧 + 𝜎𝑦 + 𝑝 𝜎𝑥 − 𝜎𝑧 − 2𝜏𝑥𝑦 + 𝜏𝑥𝑧 + 𝜏𝑦𝑧 (4.37)
𝜕𝜎𝑧 3 𝑥
183
𝜕 𝑑𝑒𝑡 𝑠
= 2 𝜎𝑧 + 𝑝 𝜏𝑥𝑦 + 2𝜏𝑥𝑧 ∙ 𝜏𝑦𝑧 (4.38)
𝜕𝜏𝑥𝑦
= 2 𝜎𝑦 + 𝑝 𝜏𝑥𝑧 + 2𝜏𝑥𝑦 ∙ 𝜏𝑦𝑧 (4.39)
𝜕𝜏𝑥𝑧
= 2 𝜎𝑥 + 𝑝 𝜏𝑦𝑧 + 2𝜏𝑥𝑦 ∙ 𝜏𝑥𝑧 (4.40)
𝜕𝜏𝑦𝑧
The vectors of derivatives of the stress invariants, 𝑝, 𝐽 and 𝜃, with respect to 𝜎

are independent of the constitutive model adopted. It is the derivatives of the
yield function, 𝐹, with respect to 𝑝, 𝐽 and 𝜃, which are model dependent and
therefore, needed to be altered on the dry side to be consistent with the
Hvorslev surface:
𝑛
𝜕𝐹 𝐽 𝛼𝐻𝑉 𝑛 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
= 2 − −1 ∙ (4.41)
𝜕𝑝 𝐽2𝜂𝑔 𝑝 + 𝑓 𝑠𝑒𝑞 𝑝 + 𝑓 𝑠𝑒𝑞
𝐽2𝜂𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞
𝜕𝐹 1
=− (4.42)
𝜕𝐽 𝐽2𝜂𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞
𝜕𝐹 𝜕𝐹 𝜕 𝐽2𝜂𝑔 𝜕𝐹 𝜕𝑝𝑐𝑠 𝜕 𝐽2𝜂𝑔 𝜕𝐹 𝜕𝑝𝑐𝑠 𝜕 𝐽2𝜂𝑓

= ∙ + ∙ ∙ + ∙ ∙ (4.43)
𝜕𝜃 𝜕 𝐽2𝜂𝑔 𝜕𝜃 𝜕𝑝𝑐𝑠 𝜕 𝐽2𝜂𝑔 𝜕𝜃 𝜕𝑝𝑐𝑠 𝜕 𝐽2𝜂𝑓 𝜕𝜃
where:
𝑛
𝜕𝐹 𝐽 𝛼𝐻𝑉 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
= 2 + 2 ∙ 1− (4.44)
𝜕 𝐽2𝜂𝑔 𝐽2𝜂𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞 𝐽2𝜂𝑔 𝑝 + 𝑓 𝑠𝑒𝑞
and:
𝑛
𝜕𝐹 𝛼𝐻𝑉 𝑛 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
= −1 ∙ (4.45)
𝜕𝑝𝑐𝑠 𝐽2𝜂𝑔 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 𝑝 + 𝑓 𝑠𝑒𝑞
184
𝜕𝑝𝑐𝑠 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 1 1

= ∙ −
𝜕 𝐽2𝜂𝑔 𝛽𝑓 ∙ 𝐽2𝜂𝑓 𝐽2𝜂𝑔 𝐽2𝜂𝑓 𝐽2𝜂𝑔 𝐽2𝜂𝑓 (4.46)
1+ 1+
𝐾2𝑓 𝐾1𝑓
𝜕𝑝𝑐𝑠 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 𝐽2𝜂𝑔

=− ∙ 2 ∙
𝜕 𝐽2𝜂𝑓 𝛽𝑓 𝐽2𝜂𝑓
1 1 (4.47)
∙ −
𝐽2𝜂𝑔 𝐽2𝜂𝑓 𝐽2𝜂𝑔 𝐽2𝜂𝑓
1+ 1+
𝐾2𝑓 𝐾1𝑓
𝜕 𝐽 2𝜂𝑔 𝜕 𝐽 2𝜂𝑓
The derivatives and are calculated from the cubic Equation 3.36,
𝜕𝜃 𝜕𝜃
Chapter 3 (the complete derivation is presented in Appendix A):
3 2
𝜕 𝐽2𝜂𝑔 𝐶𝑔 ∙ 𝑐𝑜𝑠 −3𝜃 ∙ 𝐽2𝜂𝑔
27
=− (4.48)
𝜕𝜃 3
𝐶𝑔 ∙ 𝑠𝑖𝑛 −3𝜃 ∙ 𝐽2𝜂𝑔 + 𝐶𝑔 − 3
27
3 2
𝜕 𝐽2𝜂𝑓 𝐶𝑓 ∙ 𝑐𝑜𝑠 −3𝜃 ∙ 𝐽2𝜂𝑓
27
=− (4.49)
𝜕𝜃 3
𝐶𝑓 ∙ 𝑠𝑖𝑛 −3𝜃 ∙ 𝐽2𝜂𝑓 + 𝐶𝑓 − 3
27
As indicated by Equation 4.30, the derivative of the yield function, 𝐹 , with

respect to the equivalent suction, 𝑠𝑒𝑞 , also needs to be calculated. The term
𝑓 𝑠𝑒𝑞 which includes the equivalent suction, 𝑠𝑒𝑞 , is not only present in Equation
4.27, but also within the term 𝑝𝑐𝑠 , which is given by Equation 4.4. Furthermore,
𝑝0 , which is the yield stress at the current value of equivalent suction, 𝑠𝑒𝑞 , and is
included in Equation 4.4, is also a function of 𝑠𝑒𝑞 . As it was demonstrated in
Chapter 3, depending on the expression adopted for 𝑝0 , which is a function of
the hardening parameter 𝑝0∗ , two models, offering in total three options for the
isotropic compression lines for unsaturated states, are available. The
expressions are presented below for clarity and further details concerning the
parameters are explained in Chapter 3:
185
𝜆 0 −𝜅
Model 1, Option 1: 𝑝0∗ 𝜆 𝑠𝑒𝑞 −𝜅
(4.50)
𝑝0 = −𝑝𝑐 − 𝑐
𝑝
Model 1, Option 2: (4.51)
𝑝0 = 𝑝0∗ ∙ 𝛼 𝑐 𝜆 𝑠𝑒𝑞 −𝜅
1− 0 𝑝0
𝜆 0 −𝜅
Model 2: 𝑝0∗ = −𝑝 ∙ 𝑥𝑐
, 𝑥=− (4.52)
𝑝𝑐
where:
𝜆 𝑠𝑒𝑞 = 𝜆 0 ∙ 1 − 𝑟 𝑒 𝛽∙𝑠𝑒𝑞 + 𝑟 (4.53)
𝛼0 = 𝜆 0 ∙ 1 − 𝑟 ∙ 1 − 𝑒 𝛽 ∙𝑠𝑒𝑞 (4.54)
Therefore:
𝜕𝐹 𝜕𝐹 𝜕𝑓 𝑠𝑒𝑞
= ∙ +
𝜕𝑠𝑒𝑞 𝜕𝑓 𝑠𝑒𝑞 𝜕𝑠𝑒𝑞
(4.55)
𝜕𝐹 𝜕𝑝𝑐𝑠 𝜕𝑓 𝑠𝑒𝑞 𝜕𝑝𝑐𝑠 𝜕𝑝0 𝜕𝜆 𝑠𝑒𝑞
+ ∙ ∙ + ∙ ∙
𝜕𝑝𝑐𝑠 𝜕𝑓 𝑠𝑒𝑞 𝜕𝑠𝑒𝑞 𝜕𝑝0 𝜕𝜆 𝑠𝑒𝑞 𝜕𝑠𝑒𝑞
for Model 1 and
𝜕𝐹 𝜕𝐹 𝜕𝑓 𝑠𝑒𝑞
= ∙ +
𝜕𝑠𝑒𝑞 𝜕𝑓 𝑠𝑒𝑞 𝜕𝑠𝑒𝑞
(4.56)
𝜕𝐹 𝜕𝑝𝑐𝑠 𝜕𝑓 𝑠𝑒𝑞 𝜕𝑝𝑐𝑠 𝜕𝑝0 𝜕𝑝0∗ 𝜕𝛼0
+ ∙ + ∙ ∙ ∙
𝜕𝑝𝑐𝑠 𝜕𝑓 𝑠𝑒𝑞 𝜕𝑠𝑒𝑞 𝜕𝑝0 𝜕𝑝0∗ 𝜕𝛼0 𝜕𝑠𝑒𝑞
for Model 2.
𝜕𝐹
While 𝜕 𝑝 has already been presented (Equation 4.45), the following derivatives
𝑐𝑠
need to be calculated:
186
𝜕𝐹 1
= ∙
𝜕𝑓 𝑠𝑒𝑞 𝑝 + 𝑓 𝑠𝑒𝑞
𝑛 (4.57)
𝐽 𝛼𝐻𝑉 𝑛 𝑝𝑐𝑠 − 𝑝 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
∙ − −1 ∙ ∙
𝐽2𝜂𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞 𝐽2𝜂𝑔 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 𝑝 + 𝑓 𝑠𝑒𝑞
and:
𝐾2𝑓
𝛽𝑓
𝐽2𝜂𝑔
𝐽2𝜂𝑓
1+
𝜕𝑝𝑐𝑠 𝐾2𝑓
= 𝐾1𝑓 −1 (4.58)
𝜕𝑓 𝑠𝑒𝑞 𝛽𝑓
𝐽2𝜂𝑔
𝐽2𝜂𝑓
1+ 𝐾1𝑓
𝐾2𝑓
𝛽𝑓
𝐽2𝜂𝑔
𝐽2𝜂𝑓
1+
𝜕𝑝𝑐𝑠 𝐾2𝑓
= 𝐾1𝑓 (4.59)
𝜕𝑝0 𝛽𝑓
𝐽2𝜂𝑔
𝐽2𝜂𝑓
1+ 𝐾1𝑓
Depending on the expression adopted for the isotropic compression lines for
unsaturated states, the derivative of 𝑝0 with respect to 𝜆 𝑠𝑒𝑞 for Model 1, is:
𝜕𝑝0 𝜆 0 −𝜅 𝑝0∗
Model 1, Option 1: =− 2 ∙ 𝑝0 ∙ 𝑙𝑛 − (4.60)
𝜕𝜆 𝑠𝑒𝑞 𝜆 𝑠𝑒𝑞 − 𝜅 𝑝𝑐
𝜕𝑝0 𝜆 0 −𝜅
Model 1, Option 2: =− 2 ∙ 𝑝0 ∙ 𝑙𝑛𝛼 𝑐 (4.61)
𝜕𝜆 𝑠𝑒𝑞 𝜆 𝑠𝑒𝑞 − 𝜅
187
𝜕𝜆 𝑠𝑒𝑞
Model 1, both Options: = 𝜆 0 ∙ 1 − 𝑟 𝛽𝑒 𝛽 ∙𝑠𝑒𝑞 (4.62)
𝜕𝑠𝑒𝑞
For Model 2, the derivative of 𝑝0 with respect to 𝑠𝑒𝑞 is:
𝜕𝑝0 𝑝0 𝜆 0 −𝜅
= ∙
𝜕𝑝0∗ 𝑝0∗ 𝑝 −𝑏 𝑝 (4.63)
𝛼0 − 𝑝0𝑐 ∙ 𝑏𝑙𝑛 − 𝑝0𝑐 − 1 + 𝜆 0 − 𝜅
𝑝 −𝑏
𝜕𝑝0∗ − 𝑝0𝑐 𝑝0
= −𝑝0∗ ∙ ∙ 𝑙𝑛 − (4.64)
𝜕𝛼0 𝜆 0 −𝜅 𝑝𝑐
𝜕𝛼0
= −𝜆 0 ∙ 1 − 𝑟 𝑏𝑒 𝑏∙𝑠𝑒𝑞 (4.65)
𝜕𝑠𝑒𝑞
𝜕𝑝
The complete derivation of the term 𝜕 𝑝 0∗ is explained in Appendix A.
0
Finally, there are two options concerning the term 𝑓 𝑠𝑒𝑞 ; it can be equal either
to 𝑘 ∙ 𝑠𝑒𝑞 or to 𝑆𝑟 ∙ 𝑠𝑒𝑞 , where 𝑆𝑟 is the degree of saturation obtained from the
soil-water retention curve (SWRC) model. Therefore:
𝜕𝑓 𝑠𝑒𝑞
If 𝑓 𝑠𝑒𝑞 = 𝑘 ∙ 𝑠𝑒𝑞 : =𝑘 (4.66)
𝜕𝑠𝑒𝑞
𝜕𝑓 𝑠𝑒𝑞
If 𝑓 𝑠𝑒𝑞 = 𝑆𝑟 ∙ 𝑠𝑒𝑞 : = 𝑆𝑟 (4.67)
𝜕𝑠𝑒𝑞
It should be noted that for 𝑝 + 𝑓 𝑠𝑒𝑞 = 0 the above presented derivatives

cannot be determined.
188
4.3.2 Plastic potential function derivatives
The following vector needs to be calculated:
𝜕𝐺 𝜕𝐺 𝜕𝑝 𝜕𝐺 𝜕𝐽 𝜕𝐺 𝜕𝜃
= + + ∙ (4.68)
𝜕𝜎 𝜕𝑝 𝜕𝜎 𝜕𝐽 𝜕𝜎 𝜕𝜃 𝜕𝜎
As for the yield function, the vectors of derivatives of the stress invariants with
respect to 𝜎, are given by Equations 4.31 to 4.40. The derivatives of the plastic
potential function, 𝐺, with respect to the stress invariants, 𝑝, 𝐽 and 𝜃, for 𝑝 = 𝑝𝑐
are:
𝜕𝐺 𝐽 𝛼𝐻𝑉 1
= 2+ ∙ −
𝜕𝑝 𝐽2𝜂𝑔 𝑝 + 𝑓 𝑠𝑒𝑞
𝐽2𝜂𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞
𝑛
𝛼𝐻𝑉 1 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
− −1 ∙ −
𝐽2𝜂𝑔 𝑝 + 𝑓 𝑠𝑒𝑞 𝑝 + 𝑓 𝑠𝑒𝑞 (4.69)
𝑚
1 𝛽𝐻𝑉 𝑝𝑐𝑠 − 𝑝
− ∙ ∙
𝑝 + 𝑓 𝑠𝑒𝑞 𝐽2𝜂𝑔 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
𝜕𝐺 1
=− (4.70)
𝜕𝐽 𝐽2𝜂𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞
𝜕𝐺 𝜕𝐺 𝜕 𝐽2𝜂𝑔 𝜕𝐺 𝜕𝑝𝑐𝑠 𝜕 𝐽2𝜂𝑔 𝜕𝐺 𝜕𝑝𝑐𝑠 𝜕 𝐽2𝜂𝑓

= ∙ + ∙ ∙ + ∙ ∙ (4.71)
𝜕𝜃 𝜕 𝐽2𝜂𝑔 𝜕𝜃 𝜕𝑝𝑐𝑠 𝜕 𝐽2𝜂𝑔 𝜕𝜃 𝜕𝑝𝑐𝑠 𝜕 𝐽2𝜂𝑓 𝜕𝜃
where:
𝑛
𝜕𝐺 𝐽 𝛼𝐻𝑉 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
= 2 + 2 ∙ 1− (4.72)
𝜕 𝐽2𝜂𝑔 𝐽2𝜂𝑔 ∙ 𝑝 + 𝑓 𝑠𝑒𝑞 𝐽2𝜂𝑔 𝑝 + 𝑓 𝑠𝑒𝑞
and:
189
𝑛
𝜕𝐺 𝛼𝐻𝑉 𝑛 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞
= −1 ∙ (4.73)
𝜕𝑝𝑐𝑠 𝐽2𝜂𝑔 𝑝𝑐𝑠 + 𝑓 𝑠𝑒𝑞 𝑝 + 𝑓 𝑠𝑒𝑞
𝜕 𝑝 𝑐𝑠 𝜕 𝑝 𝑐𝑠 𝜕 𝐽 2𝜂𝑔 𝜕 𝐽 2𝜂𝑓
The derivatives , , and are given by Equations 4.46,
𝜕 𝐽 2𝜂𝑔 𝜕 𝐽 2𝜂𝑓 𝜕𝜃 𝜕𝜃
4.47, 4.48 and 4.49, respectively.
4.3.3 Plastic hardening parameter 𝑨
The parameter 𝐴 is given by the following expression:
1 𝜕𝐹
𝐴 = − ∙ ∗ ∙ 𝑑𝑝0∗ (4.74)
Λ 𝜕𝑝0
where Λ is the plastic strain multiplier. From the hardening rule:
𝑣 𝑝
𝑑𝑝0∗ = − 𝑝0∗ 𝑑𝑒𝑣 (4.75)
𝜆 0 −𝜅
𝑝
The increment of volumetric plastic strains 𝑑𝑒𝑣 is:
𝑝 𝜕𝐺
𝑑𝑒𝑣 = Λ (4.76)
𝜕𝑝
Based on the above equations, the parameter 𝐴 can be rewritten as:
𝑣 𝜕𝐹 𝜕𝐺
𝐴= 𝑝0∗ (4.77)
𝜆 0 −𝜅 𝜕𝑝0∗ 𝜕𝑝
𝜕𝐹
The derivative 𝜕 𝑝 ∗ has to be calculated as:
0
𝜕𝐹 𝜕𝐹 𝜕𝑝𝑐𝑠 𝜕𝑝0
∗ = ∙ ∙ (4.78)
𝜕𝑝0 𝜕𝑝𝑐𝑠 𝜕𝑝0 𝜕𝑝0∗
𝜕𝐹 𝜕 𝑝 𝑐𝑠
The derivatives and have already been presented (Equations 4.45
𝜕 𝑝 𝑐𝑠 𝜕 𝑝0
and 4.59, respectively). The derivative of 𝑝0 with respect to the hardening

parameter, 𝑝0∗ , is for Model 1:
190
Model 1, Option 1: 𝜕𝑝0 𝜆 0 −𝜅 𝑝0∗ 𝜆 𝑠𝑒𝑞 −𝜅

(4.79)
∗ = − 𝑐
𝜕𝑝0 𝜆 𝑠𝑒𝑞 − 𝜅 𝑝
𝜕𝑝0
Model 1, Option 2: = 𝛼𝑐 𝜆 𝑠𝑒𝑞 −𝜅 (4.80)
𝜕𝑝0∗
while for Model 2 it is given by Equation 4.63.
4.3.4 Elastic and plastic strains due to changes in equivalent suction
The elastic strain vector due to changes in equivalent suction, 𝑠𝑒𝑞 , is:
𝜅𝑠
𝛥𝜀𝑠𝑒 = 1 1 1 0 0 0 𝑇
∙ ∙ 𝛥𝑠𝑒𝑞 (4.81)
3𝑣 𝑠𝑒𝑞 + 𝑝𝑎𝑡𝑚
When the secondary yield surface is inactive, the plastic strain vector due to
changes in equivalent suction, 𝑠𝑒𝑞 , is:
𝑝 𝑇
𝛥𝜀𝑠 = 0 0 0 0 0 0 (4.82)
When the secondary yield surface is active, the above vector becomes:
𝑝 𝑇
𝜆𝑠 − 𝜅𝑠
𝛥𝜀𝑠 = 1 1 1 0 0 0 ∙ ∙ 𝛥𝑠𝑒𝑞 (4.83)
3𝑣 𝑠𝑒𝑞 + 𝑝𝑎𝑡𝑚
The three above vectors remain unchanged on the wet and on the dry side of
the critical state.
4.3.5 Yield function derivatives with respect to the Factor of Safety
The consistency condition (Potts & Zdravkovic, 1999) requires that:
𝑇 𝑇
𝜕𝐹 𝜎 , 𝑘 𝜕𝐹 𝜎 , 𝑘
𝑑𝐹 𝜎 , 𝑘 = 𝛥𝜎 + 𝛥𝑘 = 0 (4.84)
𝜕𝜎 𝜕𝑘
Furthermore, for unsaturated soils, it is:
191
𝑇 𝑇
𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞 𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞
𝑑𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞 = 𝛥𝜎 + 𝛥𝑘 +
𝜕𝜎 𝜕𝑘
𝑇
(4.85)
+ 𝛥𝑠𝑒𝑞 = 0
𝜕𝑠𝑒𝑞
If a factor of safety, 𝐹𝑠 , is applied to the critical state angle of shearing

resistance, 𝑡𝑎𝑛𝜑𝑐𝑠 , so that:
𝐹 𝑡𝑎𝑛𝜑𝑐𝑠
𝑡𝑎𝑛𝜑𝑐𝑠𝑠 = (4.86)
𝐹𝑠
𝐹
where 𝜑𝑐𝑠𝑠 is the factored angle of shearing resistance, the consistency
condition becomes (Potts & Zdravkovic, 2011):
𝑑𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞 , 𝐹𝑠 =
𝑇 𝑇
𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞 , 𝐹𝑠 𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞 , 𝐹𝑠
= 𝛥𝜎 + 𝛥𝑘 +
𝜕𝜎 𝜕𝑘
(4.87)
𝑇 𝑇
𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞 , 𝐹𝑠 𝜕𝐹 𝜎 , 𝑘 , 𝑠𝑒𝑞 , 𝐹𝑠
+ 𝛥𝑠𝑒𝑞 + 𝛥𝐹𝑠 = 0
𝜕𝑠𝑒𝑞 𝜕𝐹𝑠
It is, therefore, necessary to calculate the derivative of the yield function with
respect to the factor of safety, as follows:
𝜕𝐹 𝜕𝐹 𝜕 𝐽2𝜂𝑔 𝜕𝐶𝑔 𝜕𝑀𝑔

= ∙ ∙ ∙ +
𝜕𝐹𝑠 𝜕 𝐽2𝜂𝑔 𝜕𝐶𝑔 𝜕𝑀𝑔 𝜕𝐹𝑠
(4.88)
𝜕𝐹 𝜕𝑝𝑐𝑠 𝜕 𝐽2𝜂𝑓 𝜕𝐶𝑓 𝜕𝑀𝑓 𝜕𝑀𝑔
+ ∙ ∙ ∙ ∙ ∙
𝜕𝑝𝑐𝑠 𝜕 𝐽2𝜂𝑓 𝜕𝐶𝑓 𝜕𝑀𝑓 𝜕𝑀𝑔 𝜕𝐹𝑠
where:
192
9 − 𝑀𝑖 2
𝐶𝑖 = (4.89)
2𝑀𝑖 3 𝑀𝑖 2
27 − 3 + 1
and
𝐹
6 ∙ 𝑠𝑖𝑛𝜑𝑐𝑠𝑠
𝑀𝑖 = 𝐹 (4.90)
3 − 𝑠𝑖𝑛𝜑𝑐𝑠𝑠
as already presented in Chapter 3. The subscript 𝑖 is replaced by the letter 𝑓 or

𝑔 with regard to the yield or plastic potential functions, respectively. As the
inclination 𝑀𝑖 is obviously dependent on the factor of safety, it is assumed that
the ratio between the inclinations 𝑀𝑓 and 𝑀𝑔 remains constant during an
analysis and the following equation can be introduced:
𝑅 = 𝑀𝑓 𝑀𝑔 = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 (4.91)
𝜕𝐹 𝜕𝐹 𝜕𝐹 𝜕𝑝𝑐𝑠
= + ∙ (4.92)
𝜕 𝐽2𝜂𝑔 𝜕 𝐽2𝜂𝑔 𝜕𝑝𝑐𝑠 𝜕 𝐽2𝜂𝑔
𝜕𝐹 𝜕𝐹 𝜕 𝑝 𝑐𝑠
where the derivatives , 𝜕𝑝 and 𝜕 are given by Equations 4.44, 4.45
𝜕 𝐽 2𝜂𝑔 𝑐𝑠 𝐽 2𝜂𝑔
and 4.46 respectively.
𝜕 𝐽 2𝜂𝑖
The derivative is calculated from the cubic Equation 3.36, Chapter 3 (the
𝜕 𝐶𝑖
complete derivation is presented in Appendix A):
2 3 2
𝜕 𝐽2𝜂𝑖 𝑠𝑖𝑛 −3𝜃 𝐽2𝜂𝑖 + 𝐽2𝜂𝑖 − 1
1 27
=− ∙ (4.93)
𝜕𝐶𝑖 2 𝐽2𝜂𝑖 3
𝐶𝑖 ∙ 𝑠𝑖𝑛 −3𝜃 ∙ 𝐽2𝜂𝑖 + 𝐶𝑖 − 3
27
and:
193
𝑀𝑖 3
2𝑀𝑖 ∙
𝜕𝐶𝑖 27 − 𝑀𝑖 + 2
= 2 (4.94)
𝜕𝑀𝑖 2𝑀𝑖 3 𝑀𝑖 2
27 − 3 + 1
Furthermore:
𝜕𝑀𝑓
= 𝑅 = 𝑀𝑓 𝑀𝑔 (4.95)
𝜕𝑀𝑔
The chain rule is used:
𝑡𝑎𝑛𝜑𝑐𝑠
𝜕𝑀𝑔 𝜕𝑀𝑔 𝜕𝑠𝑖𝑛𝜑𝑐𝑠𝑠
𝐹
𝜕𝜑𝑐𝑠𝑠 𝜕
𝐹
𝐹𝑠
= 𝐹 ∙ ∙ ∙ (4.96)
𝜕𝑠𝑖𝑛𝜑𝑐𝑠𝑠 𝜕𝜑𝑐𝑠𝑠 𝜕 𝑡𝑎𝑛𝜑𝑐𝑠
𝐹
𝜕𝐹𝑠 𝜕𝐹𝑠
𝐹𝑠
and finally:
𝑡𝑎𝑛𝜑𝑐𝑠
𝜕𝑀𝑔 18 ∙ 𝑐𝑜𝑠 𝑎𝑟𝑐𝑡𝑎𝑛 𝑡𝑎𝑛𝜑𝑐𝑠
𝐹𝑠
=− 2∙ (4.97)
𝜕𝐹𝑠 𝑡𝑎𝑛𝜑𝑐𝑠 𝐹𝑠 2 + 𝑡𝑎𝑛2 𝜑𝑐𝑠
3 − 𝑠𝑖𝑛 𝑎𝑟𝑐𝑡𝑎𝑛 𝐹𝑠
4.4 Validation of the Hvorslev surface
Following the implementation of the developed Hvorslev surface into ICFEP, its
reliability was tested through a series of single element analyses. For practical
purposes, the surface was validated separately under full and partial saturation.
Even though it is straightforward to verify under saturated conditions that the
suggested improvement was correctly implemented, through comparison with
an existing critical state type model already including the Hvorslev surface, no
such option is available for unsaturated states, where the effect of suction
needs to be taken into account. For this reason, the results obtained from a
194
series of saturated analyses were compared with those yielded by the modified
Cam-clay model, while the results of the unsaturated analyses were examined
only with respect to their plausibility. Finally, the effect of the two newly
introduced parameters, 𝑛 and 𝑚 , is demonstrated and explained both under
saturated and unsaturated conditions.
4.4.1 Validation under full saturation
In order to validate the newly developed surface at fully saturated states, the
results produced were compared to those obtained by the modified Cam-clay
model (MCC), employing the Hvorslev surface on the dry side of critical state.
Starting from the same initial state of 𝑝′ = 100 kPa, five values of OCR were
used in the study; 2, 5, 10, 20 and 50. Drained and undrained analyses of a
single element, subjected to triaxial compression and extension, were
performed.
An ellipse was employed on the wet side of critical state and associated
plasticity was assumed. For the ellipse to be reproduced by the Lagioia et al.
(1996) expression, parameters 𝛼𝑖 and 𝜇𝑖 obtained the values 0.4 and 0.9,
respectively (Georgiadis, 2003). On the dry side, as the MCC adopts the
Hvorslev surface given by Equation 4.1, which is a straight line, the parameter 𝑛
in Equation 4.14 was set to 0.999999, to reproduce the same surface as close
as possible and to predict yielding at similar stresses. Furthermore, if the
appropriate model parameters are adopted so that 𝐽2𝜂𝑔 = 𝑔 𝜃 , the ultimate
strength calculated by the two models should necessarily be identical under
triaxial loading conditions (𝜃 = ±30°), as the Matsuoka-Nakai failure criterion,
assumed by the unsaturated soil model, and the Mohr-Coulomb envelope,
adopted by the MCC, coincide. Moreover, employing a common isotropic
compression line (ICL), the newly developed surface should yield the same
volumetric strains as the MCC, under fully saturated states.
The parameters employed for the two models are shown in Table 4-1 and are
similar to those obtained during the calibration exercise under unsaturated
states, presented and discussed later in Section 4.5.2. The main differences are
the air-entry value of suction, 𝑠𝑎𝑖𝑟 , assumed and the values for the parameters 𝑛
195
and 𝑚, the effect of which is studied in Sections 4.4.3 and 4.4.4, respectively.
Furthermore, the parameters controlling the expansion of the elastic region with
suction were altered for the undrained analyses, so that the yield surface was
not allowed to shift or expand due to changes in suction during shearing and the
results may, therefore, be directly compared to those produced by the MCC.
Specifically, the parameter 𝑟, which controls the shape of the loading-collapse
yield surface, was set to 1.0 while the parameter 𝑘 , which determines the
increase of apparent cohesion with suction, was assumed equal to 0.0.
Moreover, the air-entry value of suction, 𝑠𝑎𝑖𝑟 , was such that de-saturation due to
the generation of negative excess pore pressure was prevented. The rest of the
parameters were the same for the drained and the undrained analyses.
Figure 4-8 illustrates the yield surfaces, on the wet and dry sides of critical
state, reproduced by the two models, for OCR = 50. It should be noted that it is
not possible to exactly reproduce the MCC ellipse with the Lagioia et al. (1996)
expression. However, as can be seen in Figure 4-8 differences are small and
′
result in minor inequalities in 𝑝𝑐𝑠 and 𝑝𝑐𝑠 calculated from the two models (Table
4-2). Despite these differences, the two models are expected to yield practically
equal results.
The drained and undrained stress paths followed in triaxial compression for the
OCR values tested and their respective initial yield surfaces are illustrated in
Figure 4-9. As the assumed initial stress state was common for all the analyses,
larger OCR values resulted in a larger 𝑝0′ and a larger elastic region. The
reproduced behaviour is presented in Figure 4-10, in terms of deviatoric stress
and axial strain, 𝐽 – 𝜀𝑎 , and volumetric and axial strain, 𝜀𝑣𝑜𝑙 – 𝜀𝑎 . The results
computed by the two constitutive models are in good agreement for all OCR
values, both for drained and undrained compression. For the latter case, for
OCR = 50 the critical state was obtained at an insignificantly higher 𝐽, having
produced slightly larger negative excess pore water pressure, when the
unsaturated soil model was employed. This feature, however, is justified by the
difference in computing the critical state point as discussed above and,
therefore, is not thought to be attributed to a false implementation of the new
surface.
196
The paths followed in triaxial extension are shown in Figure 4-11. As Lode’s
angle 𝜃 changed from -30o to +30o, the inclination of the CSL was reduced and
the position of the Hvorslev surface was moved downwards compared to its
position in Figure 4-9. Therefore, the undrained paths yielded generally at lower
values of deviatoric stress, 𝐽, in comparison with those observed under triaxial
compression. The drained paths were inclined to the left with a slope of
𝑝′ : 𝐽 = 1: 3 towards lower values of mean effective stress, 𝑝′ . For OCR = 2, 5
and 10, once yielding occurred, softening took place and the paths were forced
back down the same path towards the CSL. However, no results were produced
by either model for OCR = 20 and 50, as numerical instabilities occurred due to
the combination of low effective stresses and high stress ratios which along with
the corresponding large gradient of the flow vector affected the performance of
the analyses. Moreover, illegal negative effective stress, 𝑝′ , was detected by the
numerical code for OCR = 50 when the MCC was employed and the analysis
was terminated.
Figure 4-12 summarises the results from the above analyses. Similar to
compression, the predictions of the two models are in good agreement and any
discrepancies are attributed to the differences in their formulation. For the
drained extension analyses, the cases for OCR = 20 and 50 were repeated
using the unsaturated soil model with 𝑛 = 0.5, generating a curved Hvorslev
surface passing through the origin of the stress axes. The mean effective
stress, 𝑝′ , was reduced during shearing, obtaining a minimum value when the
yielding condition was violated. Thereafter, the stress paths were brought back
to the CSL producing the results illustrated in Figure 4-13, demonstrating the
efficiency of the new surface and its main advantage over the prior formulation.
The effect of the parameter 𝑛 is thoroughly investigated in Section 4.4.3.
Having demonstrated the validity of the newly developed surface at full

saturation, the assumption is made that this extends to unsaturated states.
Since no other model is readily available for comparison with the present one
under partial saturation, its performance is subsequently illustrated in a series of
simple parametric studies.
197
Table 4-1: Model parameters employed during validation under fully saturated states
Unsaturated soil model Modified Cam-clay
Parameter Drained Undrained Parameter Value

𝑟 0.06 1.00 𝜆 0.086
𝑘 𝑆𝑟 0.00 𝜅 0.005
𝑠𝑎𝑖𝑟 (kPa) 50.0 100000.0 𝑣1 2.120
𝛼𝑔,𝑓 0.4 𝑠𝑖𝑛𝜑′ 0.5373
𝜇𝑔,𝑓 0.9 𝐺/𝑝 15.0
𝑀𝑔,𝑓 1.3039 𝛼𝐻𝑉 0.45
𝛼𝐻𝑉 0.45 𝛽𝐻𝑉 0.25
𝑛 0.999999
𝛽𝐻𝑉 0.25
𝑚 1.00 Soil-water retention

𝜆(0) 0.086 curve (SWRC)
𝜅 0.005 Parameter Value

𝑣1 2.120 𝑠𝑑𝑒𝑠 (kPa) 50.0
𝛽 0.001 𝑠𝑎𝑖𝑟 (kPa) 50.0
𝑝𝑎𝑡𝑚 (kPa) 100.0 𝑠0 (kPa) 100000.0
𝐺/𝑝 15.0 𝑆𝑟,0 0.1
𝜆𝑠 0.08 𝛼 0.015
𝜅𝑠 0.03 𝑛 0.95
𝑝𝑐 (kPa) 1.0 𝑚 0.7
𝑠0 (kPa) 1000000.0
Table 4-2: Isotropic yielding and critical state mean effective stress
Unsat. M. MCC
′
𝑝𝑐𝑠
OCR 𝑝0′ (kPa) 𝑝𝑐𝑠 (kPa) Difference
= 𝑝0′ 2
Eq. 4.4 as % of 𝑝0′
(kPa)
2 200 101.140 100.0 0.57 %

5 500 252.860 250.0 0.57 %
10 1000 505.710 500.0 0.57 %
20 2000 1011.424 1000.0 0.57 %
50 5000 2528.561 2500.0 0.57 %
198
2000
1600
1200
800
Y.S. Unsat. M.
Y.S. MCC
400 CSL Unsat. M.
CSL MCC
0
0 1000 2000 3000 4000 5000
Mean effective stress, p' (kPa)
Figure 4-8: Yield surfaces reproduced by the modified Cam-clay (MCC) and unsaturated soil model
(Unsat. M.) on the 𝑝′ − 𝐽 plane, at full saturation (OCR = 50)
2000
1800 OCR = 50
1600
1400
ne
Li
e
1200
at
St
al
itic
Cr
1000
800
600 OCR = 20
400
OCR = 10
200
OCR = 5
OCR = 2
0
0 500 1000 1500 2000 2500 3000
Figure 4-9: Drained and undrained stress paths followed in triaxial compression for various OCR values,
together with their respective yield surfaces
199
1200 1800
drained compression undrained compression
(a) 1600 (b)
1000

1400 OCR = 50
OCR = 50 800 1200 OCR = 20

OCR = 20
OCR = 10
OCR = 10 1000
600 OCR = 5
OCR = 5 800 OCR = 2
OCR = 2
400 600
400
200
200
0 0
0 10 20 30 40 50 0 2 4 6 8 10
Axial strain,  (%) Axial strain,  (%)
drained compression 800

Excess pore water pressure, pwp (kPa)
(c) (d) undrained compression

-12
OCR = 50 OCR = 2
Volumetric strain, vol (%)
400
OCR = 20 OCR = 5
OCR = 10 OCR = 10
-8
OCR = 5 0
OCR = 2
-4
-400
OCR = 20
0 -800
OCR = 50
4 -1200
0 10 20 30 40 50 0 2 4 6 8 10
Unsat. M. MCC
Figure 4-10:Drained and undrained compression results produced by the modified Cam-clay (MCC) and
unsaturated soils model (Unsat. M.)
200
2000
1800
1600
1400 OCR = 50
1200
e
1000 Lin
te
S ta
ic al
800 C rit
600
OCR = 20
400
OCR = 10
200
OCR = 5
OCR = 2
0
0 500 1000 1500 2000 2500 3000
Figure 4-11: Drained and undrained stress paths followed in triaxial extension for various OCR values,
together with their respective yield surfaces
201
(a) 80 1200 (b)

drained extension undrained extension
1000 OCR = 50

60
800 OCR = 20
OCR = 10
40 600 OCR = 5
OCR = 2
400
OCR = 10
20
OCR = 5
200
OCR = 2
0 0
-50 -40 -30 -20 -10 0 -10 -8 -6 -4 -2 0
drained extension (c) -16 (d) undrained extension

Excess pore water pressure, pwp (kPa)
0
-12
OCR = 10
OCR = 5 -1000
OCR = 2 OCR = 2
-8
OCR = 5
OCR = 10
-2000 OCR = 20
-4
OCR = 50
0 -3000
-50 -40 -30 -20 -10 0 -10 -8 -6 -4 -2 0
Unsat. M. MCC
Figure 4-12: Drained and undrained extension results produced by the modified Cam-clay (MCC) and
unsaturated soils model (Unsat. M.)
202
60 -20

-16
40
-12
-8
OCR = 50 20
OCR = 20 -4
(a) (b)
0 0
-80 -60 -40 -20 0 -80 -60 -40 -20 0
Figure 4-13: Drained extension results for OCR = 20 and 50 when 𝑛 = 0.5
4.4.2 Validation under unsaturated conditions
The performance of the new surface under unsaturated conditions was studied
through a series of drained triaxial compression tests simulated at various
suction levels. The same model parameters as for the validation under full
saturation were employed in the analyses and are shown in Table 4-1.
The initial mean total stress, 𝑝𝑡𝑜𝑡𝑎𝑙 , was assumed equal to 100.0 kPa in all the
examined cases and three different values of pore water pressure were used in
the parametric study; -50.0, -100.0 and -200.0 kPa, generating different levels
of equivalent suction, 𝑠𝑒𝑞 ; 0.0, 50.0 and 150.0 kPa. An OCR value of 10 was
employed, calculated in terms of the hardening parameter 𝑝0∗ , so that OCR =
𝑝0∗ /𝑝, where 𝑝 is the mean equivalent stress. Assuming that 𝑝𝑛𝑒𝑡 = 𝑝𝑡𝑜𝑡𝑎𝑙 and
given that 𝑠𝑎𝑖𝑟 = 50.0 kPa, 𝑝 = 𝑝𝑡𝑜𝑡𝑎𝑙 + 𝑠𝑎𝑖𝑟 = 150.0 kPa and 𝑝0∗ = 1500.0 kPa.
The isotropic yield stress, 𝑝0 , corresponding to the current value of equivalent
suction, 𝑠𝑒𝑞 , was then computed based on the parameters controlling the shape
of the loading-collapse (LC) curve (Table 4-3).
The yield surfaces corresponding to the levels of 𝑠𝑒𝑞 studied are illustrated in
Figure 4-14 in terms of mean equivalent stress, 𝑝, together with the drained
stress paths followed. The position of the yield surface for the drained
compression test with OCR = 10 at full saturation, presented in the previous
section, is also shown in the figure for comparison and it is plotted in terms of
total stress 𝑝𝑡𝑜𝑡𝑎𝑙 . The initial effective stress, 𝑝′ , for the latter case was 100 kPa
which indicates that the initial total stress, 𝑝𝑡𝑜𝑡𝑎𝑙 , was also 100 kPa as the
203
analysis was performed at zero pore water pressure. The stress path followed
may, therefore, be compared to those produced during the unsaturated
analyses.
To accommodate the increase of apparent cohesion with suction, controlled in

the present analyses by the soil-water retention curve (SWRC), the CSL was
moved upwards. As a result, the ultimate deviatoric stress obtained at critical
state, 𝐽𝑐𝑠 , is expected to increase under higher suction levels. Moreover, the
elastic region expanded with suction under the effect of the LC curve and
therefore, the deviatoric stress is expected to reach larger peak values, 𝐽𝑝𝑒𝑎𝑘 , at
lower saturation levels. The volumetric behaviour should follow a similar pattern
with larger elastic contractant volumetric strains and larger post-yielding dilation
observed at higher suction levels.
The results presented in Figure 4-15 in terms of deviatoric stress 𝐽 and axial
strain 𝜀𝑎 , comply with the expected behaviour. Indeed, both 𝐽𝑝𝑒𝑎𝑘 and 𝐽𝑐𝑠 reach
larger values for higher suction levels, demonstrating that the model is capable
of reproducing the increase in strength with this variable. Furthermore, the
behaviour becomes more brittle, reflecting the effect of increasing 𝑝0 under
constant initial value of 𝑝, which results in the evaluation of higher apparent
OCR (OCRapp) values (Table 4-3).
In accordance with the larger softening observed at higher suction levels, the
corresponding volumetric strains produced indicate more dilative behaviour, as
is evident from Figure 4-16, illustrating the relationship between volumetric, 𝜀𝑣𝑜𝑙 ,
and axial strains, 𝜀𝑎 . It is interesting to note that a similar amount of dilation was
predicted for the saturated case and for the case with 𝑠𝑒𝑞 = 0.0 kPa. This is
due to the fact that the two tests were performed under equal apparent OCR
and therefore, reached the Hvorslev surface at the same relative position,
employing identical values for the parameter β which controls the slope of the
flow vector. The minor difference is attributed to the fact that for the latter case
yielding occurred at insignificantly larger strains.
From the set of analyses presented herein it can be concluded that the Hvorslev
surface provides reasonable results under different values of suction. Before
204
advancing to the calibration of the surface, demonstrating its capability to

realistically reproduce the observed behaviour under full and partial saturation,
the effect of the newly introduced parameters 𝑛 and 𝑚 is subsequently
presented.
Table 4-3: Isotropic yield stress at zero equivalent suction, 𝑝0∗ , and current isotropic yield stress, 𝑝0
𝑝𝑡𝑜𝑡 = 𝑝= 𝑝0∗ = OCRapp =

𝑠𝑒𝑞 ′ 𝑝0 𝑝0 /𝑝
pwp 𝑝 ′
𝑝 + 𝑝𝑤𝑝 𝑝𝑡𝑜𝑡 + 𝑠𝑎𝑖𝑟 OCR ∙ 𝑝
(kPa) (kPa) (kPa)
(kPa) (kPa) (kPa) (kPa)
0.0 − 100.0 100.0 − − 1000.0 10.00
-50.0 0.0 150.0 100.0 150.0 1500.0 1500.0 10.00
-100.0 50.0 200.0 100.0 150.0 1500.0 2180.694 14.54
-200.0 150.0 300.0 100.0 150.0 1500.0 4885.537 32.57
2000
1800 seq = 150 kPa

p = 0; ptotal = 0
1600
1400
1200
1000
800
seq = 50 kPa
600
400 seq = 0 kPa
200
saturated
0
0 500 1000 1500 2000 2500 3000
Mean total stress, ptotal (kPa)
Figure 4-14: Drained stress paths followed in triaxial compression at various suction levels, together
with their respective yield surfaces
205
1200
1000
seq = 150 kPa

800
seq = 50 kPa
600
seq = 0 kPa
400
200
saturated
0
0 10 20 30 40 50
Axial strain,  (%)
Figure 4-15: Results in terms of deviatoric stress, 𝐽 – axial strain, 𝜀𝑎 for drained triaxial compression
analyses at various suction levels
-8
seq = 150 kPa

-7
-6 seq = 50 kPa
-5 seq = 0 kPa
saturated
-4
-3
-2
-1
1
0 10 20 30 40 50
Figure 4-16: Results in terms of volumetric, 𝜀𝑣𝑜𝑙 , and axial strain, 𝜀𝑎 , for drained triaxial compression
analyses at various suction levels
206
4.4.3 Effect of the parameter 𝒏
In the analyses presented so far it was assumed that 𝑛 = 0.999999, practically

producing a straight line in the 𝑝 – 𝐽 plane for the Hvorslev surface. As already
discussed, the main disadvantage of employing a straight line is the fact that the
stress state is allowed to obtain values which are generally thought to be illegal.
Therefore, the parameter 𝑛 was introduced in order to generate a curved yield
surface on the dry side of the critical state, which restricts the stress state within
acceptable limits.
The effect of the newly introduced parameter is highlighted through a series of

drained analyses of a single element under triaxial compression, employing
three values of 𝑛; 0.0, 0.5 and 0.75. The same model parameters were used in
the analyses as for the previous section and are presented in Table 4-1.
Starting from an initial total stress, 𝑝𝑡𝑜𝑡𝑎𝑙 , of 100.0 kPa and for OCR = 10, the
element was sheared drained to critical state under two suction levels; 0.0 kPa,
practically reproducing a fully saturated state and 100.0 kPa, reproducing an
unsaturated state with 𝑠𝑒𝑞 = 50.0 kPa.
The stress paths followed and the yield surfaces examined are illustrated in
Figure 4-17 in terms of total stress, 𝑝, for the saturated case and of equivalent
stress, 𝑝, for the unsaturated one. The elastic region is clearly larger for the
unsaturated case due to the simultaneous effects of the LC curve and of the
increase of apparent cohesion, as discussed in the previous section. Therefore,
a higher deviatoric stress, 𝐽 , both at peak and at critical state, and larger
volumetric deformations are expected for the unsaturated case.
While the Hvorslev surface coincided with the CSL when 𝑛 = 0.0, the distance
between the latter and the yield surface became larger for increasing values of
𝑛, both under full and partial saturation, as illustrated in Figure 4-17 (a) and (b),
respectively. Consequently, the peak in the deviatoric stress, 𝐽, is expected to
be more pronounced for higher values of 𝑛, as is evident from Figure 4-18 (a)
and (b). Furthermore, the value of 𝐽 obtained at critical state did not vary with 𝑛,
as the position of the CSL depends solely on the suction level.
207
It is interesting to note that, assuming 𝑛 = 0, the simulated behaviour became

elastic – perfectly plastic and the critical state was achieved without softening
(Figure 4-18 (a) and (b)). However, the point of yielding did not coincide with the
critical state point and dilation took place forcing the hardening parameter, 𝑝0∗ ,
and consequently the current isotropic yield stress, 𝑝0 , to reduce. As a result,
the yield surface contracted and the critical state point was gradually moved
towards the current stress point, which remained fixed in its position, so that the
two finally coincided. During this process dilative volumetric strains, 𝜀𝑣𝑜𝑙 , were
generated, as illustrated in Figure 4-18 (c) and (d).
Moreover, as yielding occurred earlier in the analyses for which the value of 𝑛
was lower, slightly larger plastic changes in volume are expected to correspond
to similar levels of axial deformation. However, the disparity observed in Figure
4-18 (c) and (d) should not be attributed solely to the above justification, given
that a different gradient of the flow vector was evaluated at the yielding point of
each of the cases examined. Despite these discrepancies, volumetric strains,
𝜀𝑣𝑜𝑙 , of similar magnitude were obtained at critical state, independently of the
value of 𝑛 employed, both under saturated and unsaturated conditions, as the
drained paths terminated at a common critical state point.
From the analyses presented in the current section, it can be concluded that the
parameter 𝑛 affects the position of the Hvorslev surface relatively to the CSL
and, therefore, the magnitude of the peak deviatoric stress, 𝐽, and the amount of
softening occurring post-yielding. The values of 𝐽 and 𝜀𝑣𝑜𝑙 at critical state
remain unaffected as the critical state point is independent of the position of the
yield surface. Finally, the effect of this parameter was shown to be similar for
saturated and unsaturated states.
208
1000 1000

saturated, s = 0.0 kPa unsaturated, seq = 50.0 kPa
800 800
n = 0.75
600 600
n = 0.50
n = 0.00 400 400 n = 0.75
n = 0.50
200 200
n = 0.00 (b)
(a)
0 0
0 500 1000 0 500 1000 1500 2000
Mean total stress, ptotal (kPa) Mean equivalent stress, p (kPa)
Figure 4-17: Hvorslev surfaces generated for different values of the parameter 𝑛 and the corresponding
drained stress paths, under saturated and unsaturated conditions
500 500
400 400
300 300
200 200
100 100
(a) (b)
0 0
0 20 40 60 0 20 40 60
-7 -7
-5 -5
-3 -3
-1 -1
(c) (d)
1 1
0 20 40 60 0 20 40 60
n = 0.00 n = 0.50 n = 0.75
Figure 4-18: Numerical results produced for different values of the parameter n
209
4.4.4 Effect of the parameter 𝒎
The variation of the gradient of the flow vector, 𝛽, from its initial input value 𝛽𝐻𝑉 ,
defined at 𝑝 = −𝑓(𝑠𝑒𝑞 ), to 0.0 at the critical state point, is controlled by the
model parameter 𝑚. For 𝑚 = 1.0 this variation is linear whereas for any other
value it is non-linear. The effect of this parameter on the simulated behaviour is
subsequently examined through a series of drained analyses of a single
element under triaxial compression, employing four values of 𝑚: 0.05, 0.5, 1.0
and 1.5.
The model parameters used in the parametric study presented in the previous
sections were employed and are shown in Table 4-1, with the exception of the
parameter 𝑛 which was assumed equal to 0.5. Starting from an initial total
stress, 𝑝, of 100.0 kPa and for OCR = 10, the element was sheared drained to
critical state under the two suction levels studied before, 0.0 kPa and 100.0
kPa. The reproduced yield surfaces and the stress paths followed under
saturated and the unsaturated states are compared in Figure 4-19. As in the
previous validation exercises, larger deviatoric stresses, 𝐽 , and volumetric
strains, 𝜀𝑣𝑜𝑙 , are expected to be mobilised for the unsaturated conditions. In
either case, however, yielding and critical state are expected to be independent
of the value of 𝑚 employed.
Indeed, the magnitude of the deviatoric stress at peak and at critical state,
shown for saturated and unsaturated conditions in Figure 4-20 (a) and (b),
respectively, was not affected by the value of 𝑚. Furthermore, the volumetric
strains (Figure 4-20 (c) and (d)), both at yielding and at critical state, reached
identical magnitudes, indicating that the parameter 𝑚 had no effect on the
amount of softening predicted. The rate of softening of the simulated behaviour,
however, was evidently influenced by this parameter and was shown to be
larger for smaller values of 𝑚 and to decrease for larger ones.
When a linear variation was assumed for the gradient of the flow vector (i.e.
𝑚 = 1.0) the critical state was achieved at axial strains in excess of 60%, while
just 12% was required for the case with 𝑚 = 0.05. In fact, the current gradient of
the flow vector is larger for the latter case (see also Figure 4-7), leading to
210
larger post-yielding volumetric strains, 𝜀𝑣𝑜𝑙 , being evaluated for similar levels of
axial deformation and to sharper changes in the hardening parameter, 𝑝0∗ . The
ultimate change in 𝑝0∗ , however, is independent of the flow vector and is entirely
a function of both the initial and final positions of the stress point, which are not
affected by the studied parameter. As a result, the total volumetric strain, 𝜀𝑣𝑜𝑙 ,
should necessarily reach similar values at the critical state.
From the analyses presented in the current section, it can be concluded that the
parameter 𝑚 controls the rate of softening post-peak without affecting the
overall magnitude of the reduction in deviatoric stress. Moreover, the effect of
this parameter was shown to be similar for saturated and unsaturated states.
1000
p = 0; ptotal = 0
800
600
400 seq = 50 kPa
200
seq = 0.0 kPa
0
0 500 1000 1500 2000 2500
Mean total stress, ptotal (kPa)
Figure 4-19: The Hvorslev surface generated for n=0.5 and the corresponding stress path for saturated
and unsaturated conditions
211
400 400

300 300
200 200
100 100
(a) (b)
0 0
0 20 40 60 0 20 40 60
-7 -7
-5 -5
-3 -3
-1 -1
(c) (d)
1 1
0 20 40 60 0 20 40 60
m = 0.05 m = 0.50 m = 1.00 m = 1.50

Figure 4-20: Numerical results produced for different values of the parameter 𝑚
4.5 Calibration of the Hvorslev surface
The new surface, developed for highly overconsolidated soil states, was
calibrated separately for full and partial saturation. The lack of experimental
data, regarding overconsolidated unsaturated states under different loading
conditions and at various suction levels, was not the only reason for this
separation. It is commonly accepted that any constitutive model formulated to
simulate the behaviour of unsaturated soils needs to perform equally well under
fully saturated conditions, as during the analysis the pore water pressures may
change from one state to the other. Therefore, particular emphasis was given to
212
fully saturated states, with the additional intention to demonstrate the efficiency
of the surface in a broader sense, in order to support the argument that it is
possibly beneficial to implement the new formulation in other categories of
critical state type models.
The calibration for unsaturated states, apart from its obvious purpose, also
serves the necessity to demonstrate the superiority of the Hvorslev surface
compared to the existing modified Lagioia et al. (1996) expression, when
performing at high overconsolidation ratios. The analyses in this second part of
the calibration, may, therefore, be considered as a parametric study of the
influence of the yield surface on the model predictions.
The calibration was focused on the 4 parameters associated with the Hvorslev
surface, 𝛼𝐻𝑉 , 𝑛, 𝛽𝐻𝑉 , and 𝑚, as the calibration process of other aspects of the
model was presented by Georgiadis (2003).
4.5.1 Calibration under full saturation
The Hvorslev surface was calibrated at full saturation under undrained and
drained conditions. For this purpose the experimental data provided by two
different research groups on two different soils were utilised. The undrained
tests were performed by Zhu & Yin (2000) on Hong Kong marine clay (HKMC)
and the drained tests were carried out by Alonso et al. (2003) on Ancona clay.
The results of the numerical simulations are presented herein, establishing the
capability of the new surface to reproduce the experimentally observed
behaviour.
4.5.1.1 Simulation of the Zhu & Yin (2000) laboratory tests on HK marine
clay
Zhu & Yin (2000) conducted numerous undrained triaxial compression and
extension tests, at different overconsolidation ratios (OCR). Their experiments
for OCR = 8 and 4 were numerically simulated using the Hvorslev surface, in
combination with the Georgiadis (2003) constitutive model 1 (option1).
The soil tested was Hong Kong marine clay (HKMC) consisting of 27.5% clay,
46.5% silt and 26.0% fine sand. The specific gravity, 𝐺𝑠 was 2.66, the liquid
213
limit, 𝑤𝑙 , 60%, the plastic limit, 𝑤𝑝 , 28%, the plastic index, 𝐼𝑝 , 32% and the water
content 51.7%. The gradients of the isotropic compression line, 𝜆(0), and of the
swelling line, 𝜅 , in the 𝑣 – 𝑙𝑛𝑝′ space, were reported to be 0.2 and 0.044,
respectively, and the magnitude of the specific volume at 1 kPa of confining
stress, 𝑣1 , was 2.5. Furthermore, the authors measured in the laboratory the
permeability, 𝑘, and found it equal to 6.15∙10-10 m/s and calculated the effective
′
angle of shearing resistance, 𝜑𝑐𝑠 , to be 31.5o. The above values of 𝜆(0), 𝜅, 𝑣1 ,
′
𝑘 and 𝜑𝑐𝑠 were adopted in the numerical analyses (Table 4-4).
For the model to perform genuinely under fully saturated conditions the model
parameters 𝑟 and 𝑝𝑐 , were set to 1.0. In this way, the loading-collapse (LC)
curve, given by Equation 3.52, became a straight line in the 𝑝-𝑠𝑒𝑞 plane. The
parameter 𝛽, also associated with the LC curve, is irrelevant when 𝑟 = 1.0 and
was assumed equal to 0.001. No increase of apparent cohesion with suction
was allowed and the relevant parameter, 𝑘, was set equal to 0.0. The air-entry
value of suction, 𝑠𝑎𝑖𝑟 , was set to 1000.0 kPa, which is significantly larger than
the values of suction expected during the analysis. The secondary hardening
parameter, 𝑠0 , was also set to a large value (1000000.0 kPa) preventing it from
having any effect on the analyses. The above parameters ensure that the model
performs only under fully saturated conditions and partial saturation is neither
reached nor employed. Therefore, the parameters controlling the
compressibility with suction, 𝜅𝑠 and 𝜆𝑠 , can be safely assumed to be equal to 𝜅
and 𝜆(0). Finally, the atmospheric pressure, 𝑝𝑎𝑡𝑚 , was equal to 100.0 kPa. All of
the above parameters are summarised in Table 4-4.
′
Based on the value of the angle of shearing resistance, 𝜑𝑐𝑠 , reported by Zhu &
Yin (2000), the inclination of the critical state line (CSL), 𝑀𝑔 , was calculated
from Equation 3.38 to be 1.26. The inclination 𝑀𝑓 was set equal to 𝑀𝑔 and
associated plasticity was assumed on the wet side of the critical state, where
the modified Cam-clay surface was employed. For this surface to be
reproduced, the model parameters 𝛼𝑔 and 𝛼𝑓 were set equal to 0.4 and 𝜇𝑔 and
𝜇𝑔 were set to be 0.9 (Georgiadis, 2003). The above values are also shown in
Table 4-4.
214
The elastic shear modulus, 𝐺 , and the parameters 𝛼𝐻𝑉 , 𝑛 , 𝛽𝐻𝑉 , and 𝑚
associated with the Hvorslev surface, were chosen so that the numerical results
fitted the experimental data subsequently presented as close as possible.
Two samples were isotropically consolidated to a mean effective stress, 𝑝′ , of

400.0 kPa and another two to 800.0 kPa. They were subsequently allowed to
swell back to 100.0 kPa, resulting in OCR = 4 and 8, respectively. A back
pressure of 200.0 kPa was applied to each specimen, with the exception of the
sample tested in extension for OCR = 8, for which the back pressure was 400.0
kPa. The samples were then sheared undrained to critical state in compression
and in extension.
The calibration of the parameters 𝛼𝐻𝑉 and 𝑛 which are associated with the
position of the Hvorslev surface in the 𝑝′ - 𝐽 plane is explained first. The
calibration of the parameters 𝛽𝐻𝑉 , and 𝑚, associated with the plastic potential
function on the dry side, is presented subsequently.
Setting the parameters 𝛼𝐻𝑉 and 𝑛 equal to 0.2 and 0.3, respectively, the
Hvorslev surface obtained the positions shown in Figure 4-21 relative to the
laboratory data, which were normalised in terms of 𝑝0′ for illustration purposes.
Also shown in the same figure is the position of the CSL for OCR = 4 and 8,
resulting from the value of 𝑀𝑔 mentioned above. It should be noted that
throughout this section, the deviatoric stress 𝐽 corresponding to extension has
been plotted as negative for illustration purposes.
The stress paths numerically reproduced for these values of 𝛼𝐻𝑉 and 𝑛
adequately simulated those observed in the laboratory in terms of deviatoric
stress, 𝐽, and effective stress, 𝑝′ , as shown in Figure 4-22. The reproduced
stress paths exhibited an overall good agreement with the experimental data,
both in compression and extension and for both OCR values. Nonetheless, the
elastic region assumed by the model was significantly larger than the actual
one, as 𝐽 increased under constant 𝑝′, for the simulated paths, up to the yield
surface, while the laboratory data lie on a curved line in all of the cases
examined. This, however, is a general shortcoming of this type of model which
assumes linear elastic behaviour before yield.
215
Even though the critical state value of the deviatoric stress, 𝐽𝑐𝑠 , was well
predicted for OCR = 4, both in compression and extension, the prediction for
OCR = 8 was less accurate. The stress path in compression joined the CSL at
larger deviatoric and mean effective stresses than the path obtained
experimentally. In extension, on the other hand, the path joined the CSL at
lower deviatoric and higher mean effective stress levels.
The latter observation is further confirmed by the path illustrated in Figure 4-23,
in terms of deviatoric stress, 𝐽 , and axial strain, 𝜀𝑎 . Although the observed
behaviour was successfully reproduced for OCR = 4, for OCR = 8 the critical
state was generally obtained at larger 𝜀𝑎 than those measured in the laboratory,
with the corresponding value of 𝐽 over-predicted in compression. It has to be
noted, however, that in the context of the values involved, the obtained error
was negligible.
The elastic shear modulus, 𝐺, which controls the inclination of the numerically
obtained paths in the 𝐽-𝜀𝑎 plane prior to yielding, was calibrated based on the
laboratory data presented in the above figure. Its value was set equal to
10000.0 kPa for which a generally good agreement with the experimental data
was demonstrated.
The parameters 𝛽𝐻𝑉 , and 𝑚, associated with the plastic potential function on the
dry side, were equal to 0.25 and 0.5, respectively, and with these values the
excess pore water pressure, 𝛥𝑢, generated by the programme during shearing
matched the one measured experimentally as illustrated in Figure 4-24.
Although the data for OCR = 4 were closely matched, the simulations for OCR =
8, were not equally successful but the error is thought to be acceptable. The
switch from contractant to dilative behaviour during compression was well
depicted and negative changes in pore water pressure were calculated post-
yielding, in both cases studied. In extension, the behaviour remained dilative
throughout the experiments and the numerical analyses effectively captured this
feature.
From the numerical simulations of the laboratory tests presented in the current
section, it can be concluded that the newly formulated Hvorslev surface is
capable of accurately and successfully reproducing the behaviour of fully
216
saturated soils, in undrained compression and extension. Furthermore, it was

demonstrated that through employment of appropriate values for the
parameters controlling the aspects relevant to partial saturation, the model can
be used in the numerical analysis of fully saturated soils.
Table 4-4: Model parameters calibrated for fully saturated undrained conditions
Unsaturated soil model
Parameter Value Parameter Value
𝑟 1.0 𝜆(0) 0.2
𝑘 0.0 𝜅 0.044
𝑠𝑎𝑖𝑟 (kPa) 1000.0 𝑣1 2.5
𝛼𝑔,𝑓 0.4 𝛽 0.001
𝜇𝑔,𝑓 0.9 𝑝𝑎𝑡𝑚 (kPa) 100.0
𝑀𝑔,𝑓 1.26 𝐺 (kPa) 10000.0
𝛼𝐻𝑉 0.2 𝜆𝑠 0.2
𝑛 0.3 𝜅𝑠 0.044
𝛽𝐻𝑉 0.25 𝑝𝑐 (kPa) 1.0
𝑚 0.5 𝑠0 (kPa) 1000000.0
300
HK marine clay
200
100 lab. data OCR = 8

lab. data OCR = 4
0 HS OCR = 8
HS OCR = 4
CSL OCR = 8
-100
CSL OCR = 4
-200
-300
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Normalised mean effective stress, p'/p'0 ( )

Figure 4-21: Normalised laboratory results of undrained triaxial compression and extension tests for
OCR of 8 and 4 and position of the simulated Hvorslev surface and of the CSL
217
300
HK marine clay C SL
200
100
lab. data OCR = 8

0 lab. data OCR = 4
num. sim. OCR = 8
num. sim. OCR = 4
-100
-200 CSL
-300
0 100 200 300 400
Figure 4-22: Undrained triaxial compression and extension laboratory results in terms of 𝑝′ − 𝐽 and
numerical simulations adopting the Hvorslev surface, for two different values of OCR, 8 and 4 and
considering fully saturated behaviour
300
HK marine clay
200
100
lab. data OCR = 8

lab. data OCR = 4
0
num. sim. OCR = 8
num. sim. OCR = 4
-100
-200
0 5 10 15 20
Figure 4-23: Undrained triaxial compression and extension laboratory results in terms of 𝐽 − 𝜀𝑎 and
numerical simulations adopting the Hvorslev surface, for two different values of OCR, 8 and 4 and
218
100
HK marine clay
Excess pore water pressure, u (kPa)
-100
-200
lab. data OCR = 8
lab. data OCR = 4
num. sim. OCR = 8
num. sim. OCR = 4
-300
-400
0 5 10 15 20
Figure 4-24: Undrained triaxial compression and extension laboratory results in terms of 𝛥𝑢 − 𝜀𝑎 and
numerical simulations adopting the Hvorslev surface, for two different values of OCR, 8 and 4and
4.5.1.2 Simulation of the Alonso et al. (2003) laboratory tests on Ancona
clay
Alonso et al. (2003) performed a series of laboratory experiments on intact

samples of brown Ancona clay, obtained from the Villa Blasi slope in Northern
Italy, as part of a European project investigating the stability of slopes in
overconsolidated clays subjected to Mediterranean climates. Three drained
triaxial tests were performed under fully saturated conditions and their
behaviour was numerically simulated employing the Hvorslev surface. The
results of the numerical analyses are presented in the current section and are
compared with the laboratory data. It should be noted that the experiments on
Ancona clay were selected for the calibration of the new surface not only
because the samples tested were intact but most importantly because they
exhibited significant rate of softening and failed along a well defined plane.
Even though the formation of shear bands cannot be numerically reproduced in
single element FE analyses, it is shown that the stress levels monitored can be
adequately predicted.
219
The material tested was stiff, brown, silty clay of high plasticity with 𝑤𝑙 = 52 -
64%, 𝑃𝐼 = 25 - 34%, clay fraction 40 - 55%, specific particle weight 𝛾𝑠 = 27.2 -
27.8%, and natural water content close to the plastic limit. The samples were
subjected to large overconsolidation vertical in-situ stress, which was measured
in oedometer tests to vary from 2 to 3 MPa. Three samples were isotropically
consolidated to different initial states (𝑝′ = 50, 120 and 245 kPa), producing
distinct values of OCR; 40.8, 17.0 and 8.327 assuming that 𝑝0′ = 2.25 MPa.
The above three tests were numerically reproduced employing Model 1 (option
1) of the constitutive models presented in Chapter 3. The model parameters are
summarised in Table 4-5 and are discussed in detail below.
Similar to the calibration under undrained conditions, in order to ensure full

saturation, the parameters 𝑟 and 𝑝𝑐 , were set to 1.0. The parameter 𝛽 is
irrelevant when 𝑟 = 1.0 and was assumed equal to 0.001. No increase of
apparent cohesion with suction was allowed and the relevant parameter, 𝑘, was
set equal to 0.0. A value of 20.0 kPa for the air-entry value of suction, 𝑠𝑎𝑖𝑟 , was
large enough for full saturation to be maintained. The secondary hardening
parameter, 𝑠0 , was set to 10000.0 kPa and the associated yield surface is not
expected to be employed in the analyses. As fully saturated conditions were
maintained throughout the numerical simulations, the parameters controlling the
soil compressibility with suction, 𝜅𝑠 and 𝜆𝑠 , were irrelevant and were set equal to
𝜅 and 𝜆(0). Finally, the atmospheric pressure, 𝑝𝑎𝑡𝑚 , was equal to 100.0 kPa. All
of the above parameters are summarised in Table 4-5.
Associated plasticity was adopted on the wet side of critical state and the
modified Cam-clay (MCC) shape was employed for the yield and plastic
potential functions. For this surface to be reproduced, the model parameters 𝛼𝑔
and 𝛼𝑓 were set equal to 0.4 and 𝜇𝑔 and 𝜇𝑔 were set to be 0.9 (Georgiadis,
2003).
The parameters 𝛼𝐻𝑉 , and 𝑛 were set to 0.15 and 0.38, respectively, so that the
Hvorslev surface acquired the position shown in Figure 4-25 relative to the
stress points at peak observed in the laboratory. Also shown in the same figure
are the stress points at the critical state and the CSL employed for their
220
reproduction. For the CSL to match the experimental data, the parameter 𝑀𝑔
was set to 1.26, corresponding to a critical state angle of shearing resistance,
𝜑𝑐𝑠 , of 31.4°. The parameter 𝑀𝑓 was equal to 𝑀𝑔 since associated plasticity
was considered on the wet side. The stress paths followed in the laboratory are
presented in terms of deviatoric stress, 𝐽, and mean effective stress, 𝑝′ , in
Figure 4-26 in comparison with the yield surface and the CSL fitted to them.
The magnitude of peak deviatoric stress, 𝐽𝑝𝑒𝑎𝑘 , predicted by the Hvorslev

surface agreed well with the experimental data, shown in Figure 4-27 in terms
of deviatoric stress, 𝐽, and axial strain, 𝜀𝑎 . The magnitude of deviatoric stress
reached at the critical state, 𝐽𝑐𝑠 , was under-predicted for OCR = 40.8 and over-
predicted for OCR = 8.327, as illustrated in the same figure. Nevertheless, the
experimentally observed behaviour for OCR = 17.0 was accurately simulated in
terms of both peak and ultimate deviatoric stress, 𝐽.
The rate of softening of the simulated behaviour depends on the values of 𝛽𝐻𝑉
and 𝑚 employed, as explained for the latter parameter in Section 4.4.4. Setting
these parameters equal to 0.8 and 0.1, respectively, the intense rate of
softening demonstrated in the laboratory was generally reproduced, although it
was overestimated for the highest OCR value and underestimated for the
lowest one. On the contrary, the volumetric behaviour was not successfully
reproduced and the ultimate volumetric strains, 𝜀𝑣𝑜𝑙 , were significantly
overestimated for all three values of OCR, as illustrated in Figure 4-28. The
disagreement between the observed and the predicted volumetric behaviour
may be a consequence of strain localisation during the experiments (the strains
within the sample are not uniform and the volumetric strains measured may be
smaller than the volumetric strains along the shear band). The critical state was
reached at axial strains, 𝜀𝑎 , which agreed well with the experimental data,
despite the large over-prediction of the respective magnitude of the volumetric
strains, 𝜀𝑣𝑜𝑙 . As already explained in the previous sections, the latter does not
depend on the position of the yield surface or the plastic potential employed but
it is solely a function of the initial stress state in relation to the CSL.
The parameters 𝜆 0 , 𝜅 and 𝑣1 , which determine the position and the inclination
of the ICL, require a fully saturated isotropic loading and unloading laboratory
221
test or an oedometer test for their calibration. Given the lack of data obtained
from such experiments, the above parameters could not be directly calibrated.
Nonetheless, they are essential for the numerical analyses as the ratio
𝑣 𝜆 0 − 𝜅 affects the change in the hardening parameter, 𝑑𝑝0 , through
Equation 3.67 and therefore the rate of softening predicted. For this reason, the
values given to 𝜆 0 , 𝜅 and 𝑣1 (0.152 for 𝜆 0 , 0.02 for 𝜅 and 2.85 for 𝑣1 ) were
chosen so that, in combination with the values adopted for the parameters 𝛽𝐻𝑉
and 𝑚, the softening part of the laboratory stress paths shown in Figure 4-27
was reproduced as closely as possible. Nonetheless, the above three
parameters should be revised if more laboratory data, regarding intact samples
of Ancona clay, become available.
The specific volume and the coefficient 𝜅 , also affect the elastic behaviour
simulated by the constitutive model through the bulk modulus K = 𝑣 ∙ 𝑝 𝜅 .
Additionally to these two parameters, simulation of the elastic behaviour
requires knowledge of the Poisson’s ratio 𝜇. When a value of 0.2 was used for
the latter parameter, the inclination of the elastic part of the reproduced stress
paths, shown in Figure 4-27, adequately approximated the one exhibited by the
laboratory data for OCR = 8.327 and 17.0. However, the decrease in stiffness
exhibited for the lower confining stress (i.e. for OCR = 40.8) was not reproduced
by the numerical simulation and yielding was predicted at axial strain lower than
the one measured experimentally.
As the calibration was focused on the parameters associated with the Hvorslev
surface and due to lack of suitable laboratory tests, limited attention was paid
on some of the parameters required by the Georgiadis (2003) model. From the
comparison of the numerical results with the experimental data, it was evident
that the CSL was defectively located, resulting in partly inaccurate computation
of the ultimate deviatoric stress, 𝐽, and, most importantly, in ineffective estimate
of the ultimate volumetric strain, 𝜀𝑣𝑜𝑙 . On the other hand, the deviatoric stress at
yielding, 𝐽𝑝𝑒𝑎𝑘 , was successfully estimated at various OCR values, confirming
the usefulness of the newly introduced parameter 𝑛 in generating a curved yield
surface rather than a straight line. Furthermore, the axial deformation computed
222
when critical state was achieved, agreed well with the laboratory data,
demonstrating the efficiency of the parameter 𝑚.
Table 4-5: Model parameters calibrated for fully saturated drained conditions
𝑟 1.0 𝜆(0) 0.152
𝑘 0.0 𝜅 0.02
𝑠𝑎𝑖𝑟 (kPa) 20.0 𝑣1 2.85
𝛼𝑔,𝑓 0.4 𝛽 0.001
𝜇𝑔,𝑓 0.9 𝑝𝑎𝑡𝑚 (kPa) 100.0
𝑀𝑔,𝑓 1.26 𝜇 0.2
𝛼𝐻𝑉 0.15 𝜆𝑠 0.152
𝑛 0.38 𝜅𝑠 0.02
𝛽𝐻𝑉 0.8 𝑝𝑐 (kPa) 1.0
𝑚 0.1 𝑠0 (kPa) 10000.0
800 Ancona clay
a ce
700
s urf
le v e
Hv
ors Lin
e
600 at
St
al
ir tic
500 C
400
300
200
lab. data, peak
lab. data, critical state
100
0
0 200 400 600 800 1000 1200

Figure 4-25: Peak and critical state stress points and the Hvorslev surface and CSL employed for their
reproduction, in terms of deviatoric stress, 𝐽, and mean effective stress, 𝑝′
223
800 Ancona clay
700
e
Lin
te
600
ta
lS
ica
it
500
Cr
400
300
200
lab. data OCR = 40.8
100
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200
Figure 4-26: Drained triaxial stress paths under full saturation, followed in the laboratory for three
value OCR values and the yield surface adopted for their numerical simulation
500
Ancona clay
400 lab. data OCR = 8.327
OCR = 8.327
300
200
OCR = 17.0
100
OCR = 40.8
0
0 5 10 15
Figure 4-27: Drained triaxial compression laboratory results in terms of 𝐽 − 𝜀𝑎 and numerical
simulations adopting the Hvorslev surface, considering fully saturated behaviour
224
-16 Ancona clay

lab. data OCR = 17.0 OCR = 40.8
-12 lab. data OCR = 8.327
OCR = 17.0
-8
-4 OCR = 8.327
4
0 5 10 15
Figure 4-28: Drained triaxial compression laboratory results in terms of 𝜀𝑣𝑜𝑙 − 𝜀𝑎 and numerical
simulations adopting the Hvorslev surface, considering fully saturated behaviour
4.5.2 Calibration under unsaturated states
The newly developed version of the Hvorslev surface was used in order to
simulate laboratory tests performed by Estabragh & Javadi (2008) on artificial
silty clay (Model 1, Option 1 of the Georgiadis, 2003 model was employed). The
scope of the current section is twofold, as apart from calibrating the newly
developed surface for unsaturated states, its superiority over other shapes
commonly assumed for the yield and plastic potential functions, is also
demonstrated. For this reason, four different surfaces were employed; the
original Cam-clay surface (CC) introduced by Roscoe & Schofield (1963), the
modified Cam-clay surface (MCC) introduced by Roscoe & Burland (1968), the
sinfonietta classica surface (SC) proposed by Nova (1988) and the Hvorslev
surface attached to the latter on the dry side. Drained tests at four levels of
suction, for two OCR values, corresponding to states dry of critical, were
numerically simulated using all of the above surfaces and their respective
predictions are presented and compared herein. Furthermore, the laboratory
results for another three values of OCR, corresponding to states wet of critical,
were simulated using the new version of the unsaturated soil model for reasons
225
of completeness. The presented results demonstrate that even though it is

demanding for the model to reproduce, with the same level of accuracy, the soil
behaviour at different suction levels and in the entire stress space, both on the
dry and wet sides of the critical state, the new surface can adequately
reproduce the soil behaviour and it is other aspects of the model that require
further improvement.
4.5.2.1 The soil tested by Estabragh & Javadi (2008) and the
experimental procedure followed
Estabragh & Javadi (2008) tested a low plasticity silty soil in order to investigate
the evolution of the critical state line with suction, employing the axis translation
technique in triaxial apparatus. It was concluded that the critical state lines in
the 𝑞 − 𝑝΄ stress space, for different suction levels, are not parallel but merge
with each other at a point with confining stress 𝑝 of about 950 kPa. For the soil
tested, it was shown that suction had no significant effect on the soil strength for
𝑝 values greater than 950 kPa.
The soil tested consisted of a mixture of 5% sand, 90% silt and 5% clay with a
plasticity index (𝑃𝐼) of 19% while the liquid limit (𝐿𝐿) was 29%. The optimum
water content in the standard compaction test was 14.5% and the maximum dry
density (𝛾𝑑 ) was 1.74 Mg/m3. Finally, the specific gravity of solids (𝐺𝑠 ) was 2.72.
The samples were statically compacted in 9 layers to ensure uniformity, at a

water content of 10%. The required suction level was achieved through an
equalisation process during which the air and water pressures were allowed to
equilibrate with their respective applied values. Once the desired suction was
achieved, the samples were loaded isotropically to a mean net stress, 𝑝𝑛𝑒𝑡 , of
550 kPa, under constant suction, and unloaded to the 5 different stress levels
presented in Table 4-6, while the normal consolidation and swelling lines for
every suction level were calculated. Subsequently, the samples were subjected
to drained triaxial shearing until the critical state was reached. In total, five
shear tests were conducted from initial OCR values of 11.0, 5.5, 2.75, 1.833
and 1.375 for each one of the four suction levels examined: 0.0, 100.0, 200.0
and 300.0 kPa (Table 4-6). The initial degree of saturation, before shearing, for
each one of the tests performed is also given in Table 4-6, while the SWRC
226
model parameters employed to reproduce these values are presented in Table

4-7. Further details about the experimental procedure are given by Estabragh &
Javadi (2008).
4.5.2.2 Model parameters employed in the analyses
The model parameters employed in the analyses are presented in Table 4-7
and their values are discussed below. The parameters controlling the shape of
the yield surface, 𝛼𝑖 and 𝜇𝑖 , obtained the values shown in Table 4-7 so that the
original Cam-clay (CC), the modified Cam-clay (MCC) and the sinfonietta
classica (SC) surfaces were reproduced (Georgiadis, 2003). Furthermore,
associated plasticity was assumed for the above surfaces and therefore 𝛼𝑔 = 𝛼𝑓
and 𝜇𝑔 = 𝜇𝑓 . It should be noted that the Hvorslev surface was attached to the
SC surface for reasons that become evident in the next section.
The position of the CSL as well as the position of the yield surface itself, are
dependent on the apparent cohesion. The model offers two options regarding
the increase of apparent cohesion with suction; it can either be linear or it can
depend on the degree of saturation, 𝑆𝑟 . Naturally, the latter option is thought to
be more accurate and is expected to yield more reliable results. However, the
soil-water retention curve (SWRC) is not presented by Estabragh & Javadi
(2008). What is available instead, is the degree of saturation, 𝑆𝑟 , measured for
each sample, before drained shearing took place and, therefore, the parameters
required for the reproduction of the Van Genuchten (1980) type SWRC were
fitted to these data, as shown in Figure 4-29. The relevant parameters are
summarised in Table 4-7. At this point it should be mentioned that the air-entry
value of suction, 𝑠𝑎𝑖𝑟 , was assumed to be 0.0 kPa. The equivalent suction, 𝑠𝑒𝑞 ,
is therefore equal to the experimentally applied matric suction, 𝑠, while the
mean equivalent stress, 𝑝, is equal to the mean net stress, 𝑝𝑛𝑒𝑡 .
The slope of the critical state line (CSL), 𝑀𝑔 , in the 𝑞 – 𝑝 space was estimated
to be 1.3, resulting in 𝐽2𝜂𝑔 = 0.75. In this way, the resulting CSL was matched
to the laboratory data in terms of deviatoric stress, 𝐽, and mean net stress, 𝑝, as
shown in Figure 4-30 (a), for the four suction levels studied. Since associated
plasticity was assumed on the wet side, 𝑀𝑓 was equal to 𝑀𝑔 . As the model
227
assumes that the CSLs remain parallel at various suction levels and move
upwards, due to the increase in apparent cohesion with suction, it was not
feasible to fit closely the experimental data which showed not only an upward
movement but also a shift in the slope with increasing suction.
Clearly, the assumption that the slope of the CSL remains constant with
increasing suction is a drawback of the model. If cohesion was permitted at full
saturation the experimental data would be slightly better reproduced, as shown
in Figure 4-30 (b). However, the experimental data indicate that the CSL
became curved when suction increased, which would explain the apparent
change of slope (Figure 4-30 (c)). Nevertheless, further laboratory evidence on
the actual shape of the CSL needs to be available before advancing this
particular aspect of the model.
For 𝑀𝑔 = 1.3, the deviatoric stress at critical state, 𝐽𝑐𝑠 , is underestimated for
higher OCR values and overestimated for lower OCR values, at 0.0, 100.0 and
200.0 kPa of suction while for 𝑠𝑒𝑞 = 300.0 kPa the overestimation of 𝐽𝑐𝑠 was
more general (Figure 4-30 (a)). The above predictions are common for the four
surfaces assumed and it is solely the peak deviatoric stress, 𝐽𝑝𝑒𝑎𝑘 , which varies
depending on the surface employed.
The only surface which provides flexibility on the dry side of the critical state is
the Hvorslev surface as both the yield and plastic potential functions have a
predefined shape in all other cases. Despite the large amount of experimental
data provided by the authors and in order to simplify the process, the
parameters 𝛼𝐻𝑉 , 𝛽𝐻𝑉 , 𝑛 and 𝑚 were calibrated employing solely the data for
OCR = 11.0, at 𝑠𝑒𝑞 = 200.0 kPa and the same values were adopted in the rest
of the analyses. The parameters 𝛼𝐻𝑉 and 𝑛 were set equal to 0.45 and 0.5,
respectively, resulting in the Hvorslev surface shown in Figure 4-31 (a) in
comparison with the stress path followed in the laboratory. The parameters 𝛽𝐻𝑉
and 𝑚 were equal to 0.25 and 0.5, respectively, simulating closely the rate of
softening of the soil behaviour exhibited in the laboratory, as illustrated in Figure
4-31 (b) and reproducing the volumetric behaviour adequately, as shown in
Figure 4-31 (c). Furthermore, the ratio of shear modulus, 𝐺 , over the mean
equivalent stress, 𝑝, which controls the elastic soil stiffness, was set equal to
228
15.0 (Table 4-7) so that the inclination of the stress path in the 𝜀𝑎 -𝐽 plane prior
to softening was adequately estimated at 𝑠𝑒𝑞 = 200.0 kPa (Figure 4-31 (b)).
The same ratio 𝐺 𝑝 was employed in the remaining analyses.
Another important aspect of constitutive modelling in unsaturated soils, is the

isotropic compression lines (ICLs) and the swelling lines which control the soil
compressibility and affect the shape of the LC curve. The isotropic lines during
isotropic triaxial consolidation and swelling at full saturation were calculated by
Estabragh & Javadi (2008) and the experimental values they indicated for 𝜆(0),
𝜅 and 𝑣1 were adopted in the present analyses and are summarised in Table
4-7. The authors also measured the elasto-plastic coefficient of soil
compressibility, 𝜆(𝑠𝑒𝑞 ), at different suction levels and the variation exhibited is
illustrated in Figure 4-32. The variation of 𝜆(𝑠𝑒𝑞 ) computed when the
parameters 𝑟 and 𝛽, also affecting the shape of the LC curve, were set equal to
0.06 and 0.001, is shown in the same figure to compare well with the laboratory
data.
It should be noted that because of the way the tests were conducted in the
laboratory, the current isotropic yield stress, 𝑝0 , was independent of the suction
level and equal to 550.0 kPa. This, however, does not necessarily imply that the
LC curve is a straight vertical line in the 𝑠𝑒𝑞 – 𝑝 stress space, as the slope of the
ICLs shows a typical variation with suction (Figure 4-32). On the contrary, the
LC curve is assumed to have been violated during isotropic loading and brought
to a new position for each suction level, while its initial position is not of interest,
as it is only the shearing part of the tests which was simulated. The current
position of the LC curve, at the beginning of shearing, is dictated through the
choice of the OCR value, calculated in terms of the isotropic yield stress 𝑝0 . The
value of 𝑝𝑐 was assumed to be 1.0 kPa and, in combination with the above
mentioned values of 𝑟 and 𝛽, produced the shape of the LC curve illustrated in
Figure 4-33, for the sample consolidated to 550.0 kPa at 0.0 kPa of suction.
The program automatically calculates the initial specific volume from Equation
3.84. Employing a value of 𝜅𝑠 = 0.01 the equation yields specific volumes
comparable to those measured in the laboratory before shearing (Table 4-8).
Finally, the initial secondary hardening parameter was set equal to 1000.0 kPa
229
and, therefore, the value of 𝜆𝑠 assumed (0.08) was not employed during the
analyses, as the secondary yield surface was not violated.
Table 4-6: Initial stresses and degree of saturation prior to shearing

Initial stress state Suction and saturation level
𝑆𝑟
𝑝′ 𝑠𝑒𝑞 𝑆𝑟
OCR Calculated
(kPa) (kPa) measured
(SWRC)
50.0 11.000 0.0 1.0 1.0

100.0 5.500 100.0 0.56 0.578
200.0 2.750 200.0 0.43 0.451
300.0 1.833 300.0 0.39 0.3942
400.0 1.375 - - -
Table 4-7: Model parameters calibrated for unsaturated drained conditions

Soil-water retention
curve (SWRC)
Parameter MCC CC SC Parameter Value Parameter Value
𝛼𝑔,𝑓 0.4 0.00001 0.7 𝜆(0) 0.086 𝑠𝑑𝑒𝑠 (kPa) 0.0
𝜇𝑔,𝑓 0.9 1.00001 0.9999 𝜅 0.005 𝑠𝑎𝑖𝑟 (kPa) 0.0
𝛼𝐻𝑉 − − 0.45 𝑣1 2.120 𝑠0 (kPa) 100000
𝑛 − − 0.5 𝛽 0.001 𝑆𝑟,0 0.1
𝛽𝐻𝑉 − − 0.25 𝑝𝑎𝑡𝑚 (kPa) 100.0 𝛼 0.015
𝑚 − − 0.5 𝐺/𝑝 15.0 𝑛 0.95
𝑀𝑔,𝑓 1.3039 𝜆𝑠 0.08 𝑚 0.7
𝑟 0.06 𝜅𝑠 0.01 - -
𝑘 SWRC 𝑝𝑐 (kPa) 1.0 - -
𝑠𝑎𝑖𝑟 (kPa) 0.0 𝑠0 (kPa) 1000000.0 - -
230
Table 4-8: Specific volumes calculated at the beginning of the analyses based on the model parameters,
in comparison with those measured in the laboratory by Estrabragh & Javadi (2008)
Suction 0.0 kPa Suction 100.0 kPa
𝑣 calculated 𝑣 𝑣 calculated 𝑣
𝑝′ 𝑝′
from Eq. measured in from Eq. measured in
(kPa) (kPa)
3.85 lab. 3.85 lab.
50.0 1.59 1.54 50.0 1.63 1.68
100.0 1.59 1.58 100.0 1.63 1.74
200.0 1.58 1.45 200.0 1.62 1.67
300.0 1.58 1.56 300.0 1.62 1.64
400.0 1.58 1.57 400.0 1.62 1.66
Suction 200.0 kPa Suction 300.0 kPa

𝑣 calculated 𝑣 𝑣 calculated 𝑣
𝑝′ 𝑝′
from Eq. measured in from Eq. measured in
(kPa) (kPa)
3.85 lab. 3.85 lab.
50.0 1.67 1.70 50.0 1.71 1.70
100.0 1.67 1.70 100.0 1.70 1.71
200.0 1.66 1.70 200.0 1.70 1.70
300.0 1.66 1.66 300.0 1.70 1.70
400.0 1.66 1.66 400.0 1.70 1.70
1
0.9
Degree of satuartion, Sr ( )
0.8
0.7
0.6
0.5
0.4
0.3 lab. data
0.2 SWRC fitted
0.1
0
0.1 1 10 100 1000 10000
Figure 4-29: Degree of saturation measured in the laboratory at the commencement of shearing for each
suction level studied and the Van Genuchten (1980) type SWRC fitted
231
600
seq = 300.0 kPa

500
seq = 200.0 kPa
seq = 100.0 kPa
seq = 0.0 kPa
400
300
seq = 0.0 kPa

200 seq = 100.0 kPa
seq = 200.0 kPa
seq = 300.0 kPa
100
(a)
0
0 100 200 300 400 500 600 700 800

600
500
a
kP
400 0 .0
30
=
s eq a
300 kP
0
0.
=
200 s eq
100
(b) (c)
0
0 100 200 300 400 500 600 700 800
0 100 200 300 400 500 600 700 800
Mean equivalent stress, p (kPa) Mean equivalent stress, p (kPa)
Figure 4-30: Laboratory data for the critical state lines (CSLs) in the 𝐽 – 𝑝 space for different suction
levels (shown in symbols) and; (a) their respective positions predicted by the current model; (b )their
positions as could be predicted if cohesion at zero equivalent suction was allowed and (c) their position
if the critical state line was allowed to become curved with increasing equivalent suction
232
SL
300
C
200
HV
100
0
-100 0 100 200 300 400 500 600
300 -4

-3
200 -2
-1
100 0
1
(b) (c)
0 2
0 4 8 12 16 0 4 8 12 16 20
lab HV
Figure 4-31: Calibration of the parameters 𝛼𝐻𝑉 , 𝛽𝐻𝑉 , 𝑛 and 𝑚, associated with the Hvorslev surface
based on the laboratory results for OCR = 11.0 and 𝑠𝑒𝑞 = 200.0 kPa
233
0.1
Inclination of the ICL, (seq) ( )

0.08
0.06
0.04
0.02 lab. data

num. fitting
0
0 50 100 150 200 250 300
Figure 4-32: Variation of the slope of the ICL, 𝜆(𝑠𝑒𝑞 ) with equivalent suction measured by Estabragh
and Javadi (2008) and comparison with the variation reproduced by the constitutive model
300
200
Loading - Collapse
curve
100
0
1 10 100 1000 10000
Mean equivalent stress, p(kPa)

Figure 4-33: Loading-Collapse (LC) curve reproduced by the model parameters 𝑟, 𝛽 and 𝑝𝑐
4.5.2.3 Results and discussion
The numerical results produced by the CC, the MCC and the SC surfaces, as
well as the Hvorslev surface attached to the latter, are compared in the present
section to the laboratory results provided by Estabragh & Javadi (2008) for the
tests performed at initial mean net stresses, 𝑝𝑛𝑒𝑡 , of 50.0 and 100.0 kPa (i.e.
OCR = 11.0 and 5.5, respectively) under four suction levels; 0.0, 100.0, 200.0
234
and 300.0 kPa. The purpose of the exercise was not only to study the
performance of the Hvorslev surface but also to demonstrate its superiority in
reproducing the experimentally observed soil behaviour in comparison with
other shapes commonly adopted for the yield and plastic potential functions.
Figure 4-34 illustrates the surfaces adopted and the CSL reproduced in the
deviatoric and mean equivalent stress plane at 0.0 kPa of suction, in
comparison with the stress paths followed in the laboratory. The Hvorslev
surface is undoubtedly the one that intersected both paths closer to the peak
stress reached and is therefore expected to yield the most reasonable results.
The position of the SC and MCC shapes, relative to the experimental stress
paths, clearly instigates overestimation of the peak stress of practically similar
magnitudes, while employment of the CC surface is expected to improve the
results significantly only for OCR = 5.5. Nonetheless, the adequate reproduction
of the stress path in the latter case is thought to be coincidental since for the
higher OCR value the surface is shown to over-predict the size of the elastic
region.
The experimental results related to the above tests are presented in Figure
4-35, in terms of deviatoric stress, 𝐽, and volumetric strain, 𝜀𝑣𝑜𝑙 , versus axial
strain, 𝜀𝑎 , for the two OCR values presently examined. The numerical results
produced by the four surfaces are also shown in the same figure. Even though
there was no experimental evidence of a peak in the deviatoric stress, 𝐽, for
either OCR value examined, all four surfaces predicted one, with the largest
being yielded by the SC and the MCC surfaces (Figure 4-35 (a) and (b)). The
over-prediction produced by these two surfaces reduced for the lower OCR
value but remained significant. As already explained, employment of the CC
shape improved the results considerably but not adequately, as a significant
over-estimation of the yield stress was observed for OCR = 11 while the peak
stress produced for OCR = 5.5 was lower than the ultimate stress measured in
the laboratory.
The results obtained for the Hvorslev surface compare better to the
experimental data, yet only corresponding on average to the level of stresses
observed and to the respective strain levels. For example for the case where
235
OCR = 11 (Figure 4-35 (a)), a peak in the deviatoric stress slightly larger than
the stress measured, was calculated at low axial deformation, while its value at
critical state underestimated that observe experimentally. For OCR = 5.5
(Figure 4-35 (b)), although the peak stress reached a value similar to the
ultimate monitored stress, critical state was achieved at lower stresses due to
post-yield softening being predicted.
For the latter case, it should be noted that even though the estimated response
appears to be stiffer than the observed one, this difference could be attributed
to the irregular shift the experimental stress path exhibited at low axial strains.
Furthermore, the underestimation of the deviatoric stress at critical state, 𝐽𝑐𝑠 ,
was common for all the cases studied and resulted from the ineffective
reproduction of the critical state data presented in Figure 4-30.
The inadequate reproduction of the CSL was also evident from the volumetric
behaviour reproduced by the numerical analyses, since critical state was
generally obtained at higher axial deformations than those observed and
following the generation of larger volumetric strains, as illustrated in Figure 4-35
(c) and (d), for OCR = 11 and 5.5, respectively. As explained by Georgiadis
(2003), the position of the critical state line in the 𝑣 − 𝑙𝑛𝑝 plane does not only
depend on the isotropic yield stress, 𝑝0 , but also on the value of mean stress
corresponding to critical state, 𝑝𝑐𝑠 , which was not identical for the surfaces
employed and dissimilar volumetric strains, 𝜀𝑣𝑜𝑙 , were, therefore, predicted. The
SC and the Hvorslev surfaces, which share a common 𝑝𝑐𝑠 , yielded identical
values of volumetric strain, 𝜀𝑣𝑜𝑙 , at critical state, although, due to distinct plastic
potential functions being utilised, different paths were followed to this ultimate
value. The results obtained by the CC shape, exhibiting the lowest value of 𝑝𝑐𝑠 ,
showed a better agreement with the laboratory data for both cases examined.
Shapes adopting larger values for 𝑝𝑐𝑠 , highly over-predicted the volumetric
deformation observed, indicating that the actual CSL is positioned in the
𝑣 − 𝑙𝑛𝑝 plane closer to the initial state than what was defined by the plastic
potential surfaces adopted.
Finally, it should be noted that for the case where OCR = 5.5, the behaviour
was predicted to be dilative, as it would be expected for such a high value of
236
OCR. Nevertheless, contraction was measured in the laboratory, implying that

yielding must have occurred on the wet side and reinforcing further the notion
that the position of the CSL was inaccurately reproduced in the numerical
analyses.
Figure 4-36 illustrates the yield surfaces adopted and the stress paths observed
in the laboratory for 𝑠𝑒𝑞 = 100.0 kPa. Besides the SC shape, which extensively
assumed an elastic region larger than the one indicated by the two stress paths,
the other two surfaces generated by the Lagioia et al. (1996) expression
performed adequately only for one of the two cases examined while they failed
to conform with the other. Specifically, the MCC surface overestimated the peak
stress for OCR = 11 but predicted it well for OCR = 5.5, while, on the contrary,
the CC shape reproduced effectively the path for the larger OCR value but not
for the smaller one. It was exclusively the Hvorslev surface that was consistent
with both stress paths, demonstrating its superiority.
The respective experimental results in terms of deviatoric stress, 𝐽 , and

volumetric strain, 𝜀𝑣𝑜𝑙 , versus axial strain, 𝜀𝑎 , are shown in Figure 4-37, together
with the numerical results yielded by the adopted surfaces. Indeed, only the
Hvorslev surface was capable of adequately estimating the deviatoric stress for
both OCR values examined (Figure 4-37 (a) and (b)), regardless of the elastic
behaviour which was simulated to be stiffer than the actual one and, as a result,
the peak stress was achieved at lower axial strain, 𝜀𝑎 . Even though the
laboratory results gave some evidence of post-yielding strain-softening for OCR
= 5.5 in Figure 4-37 (b), there was no such evidence for OCR = 11 (Figure 4-37
(a)). Nonetheless, a peak in the deviatoric stress, 𝐽 , was produced by the
numerical analyses for the latter case, while for the former one the CC shape
predicted strain-hardening since yielding occurred defectively on the wet side.
Furthermore, in contrast with the analyses performed at 𝑠𝑒𝑞 = 0.0 kPa, the
current analyses produced an adequate estimate of the critical state value of
the deviatoric stress, 𝐽𝑐𝑠 .
The estimation of the critical state in terms of volumetric behaviour, however,

was not well predicted, as Figure 4-37 (c) and (d) show. While dilation was
observed in the laboratory once yielding occurred for OCR = 11, the behaviour
237
for OCR = 5.5 was contractant, despite the apparent softening in the deviatoric
stress (Figure 4-37 (b)). Furthermore, both experiments were terminated before
the critical state was achieved and it is not, therefore, possible to reach any safe
conclusion concerning the simulation of the volumetric deformation. It could,
however, be postulated that for OCR = 11 the critical state would have been
achieved, had the test been continued, for a value of volumetric strains, 𝜀𝑣𝑜𝑙 , not
much dissimilar to that predicted by the MCC surface. The existing data are
positioned between the results provided by the CC and MCC surfaces with a
clear tendency to increase towards the latter. For OCR = 5.5 contraction was
predicted solely by the CC shape, the accuracy of the experimental data is,
however, questionable. The authors reported that after shearing there was no
evidence of failure in the results of the sample tested at 𝑝 = 100 kPa (i.e. OCR
= 5.5) and attributed this fact to development of shear bands in the sample.
Increasing the equivalent suction, 𝑠𝑒𝑞 , to 200 kPa, the surfaces adopted in the
analyses obtained the position illustrated in Figure 4-38 in comparison with the
stress paths followed in the laboratory. For the higher saturation levels studied
above, the difference between the results provided by the various surfaces was
mainly concentrated on the values of deviatoric stress, 𝐽, and volumetric strain,
𝜀𝑣𝑜𝑙 , predicted. For the current case, the type of the behaviour reproduced –
dilative or contractant – adds to the differentiation between the shapes and for
smaller OCR values is expected to dominate the comparison. Indeed, yielding
was predicted on the wet side for the CC surface for the two tests examined,
and also for OCR = 5.5 when the MCC shape was employed, and, therefore,
the predicted behaviour is expected to be contractant. In all other cases, where
yielding is illustrated to occur on the dry side, dilative behaviour is anticipated.
The laboratory results, presented in Figure 4-39, revealed that the observed
behaviour was undoubtedly dilative post-yielding and therefore, the CC and
MCC shapes, which produced contraction for at least one of the cases studied,
were shown to be inadequate for the simulation of the soil behaviour at larger
suction levels. Even though the results, obtained when the SC and the Hvorslev
surfaces were employed, did not exhibit significant disparity for the smaller OCR
value, the difference was well pronounced in the magnitude of the peak
238
deviatoric stress, 𝐽𝑝𝑒𝑎𝑘 , for OCR = 11 (Figure 4-39 (a)). Although, the critical
state was not achieved by the end of the currently examined laboratory tests, it
is concluded that the SC and the Hvorslev surfaces underestimated the
magnitude of the ultimate volumetric strain (Figure 4-39 (c) and (d)).
Similar conclusions can be drawn from the numerical simulation of the

experiments conducted at 𝑠𝑒𝑞 = 300 kPa, as the stress path for OCR = 5.5 is
shown in Figure 4-40 to intersect the CC and MCC surfaces on the wet side,
while the SC and Hvorslev surfaces appear to predict yielding adjacent to the
critical state point. For OCR = 11 contraction is expected to be generated solely
when the CC shape was adopted whereas dilation is predicted by the rest of the
analyses, as confirmed by the numerical results illustrated in Figure 4-41 (c).
Also shown in Figure 4-41 are the experimental data which, for the larger OCR
value, exhibit a well defined peak in the deviatoric stress, 𝐽 (Figure 4-41 (a)).
Although 𝐽𝑝𝑒𝑎𝑘 was successfully calculated by the MCC, the SC and the
Hvorslev surfaces, none was capable of reproducing the observed rate of
softening, since not only was the critical state value of deviatoric stress, 𝐽𝑐𝑠 ,
overestimated but it was mobilised at larger axial deformation than in the
laboratory. Nonetheless, the reproduction of the deviatoric stresses observed
was generally adequate for both OCR values when the Hvorslev surface was
adopted, demonstrating its consistency in simulating effectively the stress states
exhibited during shearing at various suction levels and from different initial
states.
The volumetric behaviour, however, was not well reproduced, as shown in

Figure 4-41 (c) and (d). Although, the Hvorslev surface rightly resulted in
dilation being generated for OCR = 11, the ultimate volumetric deformation was
under-predicted. Furthermore, for OCR = 5.5 the reproduced behaviour was
almost perfectly plastic post-yielding and the experimentally observed dilation
was not captured by the numerical analysis. It should be noted that the
volumetric behaviour observed in the laboratory does not indicate critical state
conditions.
Based on the above discussion it can be concluded that the Hvorslev surface is
the only one, of those examined, that could reasonably reproduce the shear
239
stresses and give the right indication about the type of volumetric behaviour
observed, under various levels of applied suction. The MCC and SC surfaces
were shown to significantly over-predict the peak deviatoric stress, 𝐽𝑝𝑒𝑎𝑘 , at
lower suction levels and for larger OCR values. This could result in unrealistic
and non-conservative predictions in numerical analyses of boundary value
problems, since failure occurs when suction, and consequently the strength, is
reduced. Although it could be argued that at low suction levels the CC surface
was capable of efficiently simulating the observed behaviour, when suction
increased, contraction was predicted instead of dilation which in a boundary
value problem would lead to miscalculation of volume changes, indicating
settlements where swelling should occur. Therefore, implementation of the
Hvorslev surface is thought to be essential in order to overcome the above
demonstrated inaccuracy of the Lagioia et al. (1996) expression regarding the
dry side of the critical state.
Although the Hvorslev surface could have been attached to any other shape,
the SC was believed to be the appropriate option as it was capable of predicting
the dilative behaviour exhibited at lower saturation levels, in contrast with the
CC and MCC surfaces which generated contraction. In the following analyses,
the effect of attaching the Hvorslev to the SC surface on the simulation of the
behaviour on the wet side is evaluated, through comparison of the numerically
obtained results with the experimental data available for OCR = 2.75, 1.833 and
1.375.
In Figure 4-42 the numerical results in terms of deviatoric stress, 𝐽, versus axial
strains, 𝜀𝑎 , are compared to the experimental data, for 𝑠𝑒𝑞 = 0.0 kPa while the
results for OCR = 11 and 5.5 are also shown for completeness. Overall
reasonable agreement of the predicted and observed deviatoric stress at critical
state, 𝐽𝑐𝑠 , was obtained although the numerical simulations reached a plateau at
lower axial strains than those observed in the laboratory, exhibiting a stiffer
behaviour. It should be noted, however, that the ultimate value of deviatoric
stress, 𝐽 , does not depend on the yield surface adopted but on the CSL
introduced in the analysis. The volumetric strains illustrated in Figure 4-43,
confirm that the critical state was predicted earlier than observed. However, the
240
magnitude of the volumetric strains at the critical state was adequately

predicted for OCR = 1.833 and 1.375 while the experimental data for OCR =
2.75 were clearly out of trend and their reliability is, therefore, questionable.
When the equivalent suction, 𝑠𝑒𝑞 , was increased to 100.0 kPa, the ultimate
deviatoric stress was accurately predicted with the exception of the case where
OCR = 1.375 (Figure 4-44). Additionally, lower contractive ultimate volumetric
strains, 𝜀𝑣𝑜𝑙 , were predicted compared with the experimentally observed ones,
which did not reach the critical state (Figure 4-45). For 𝑠𝑒𝑞 = 200.0 kPa,
although the over-prediction of the deviatoric stress, 𝐽 , for OCR = 1.375
increased further, an adequate simulation was achieved for the remaining tests
(Figure 4-46). The volumetric behaviour was also successfully reproduced
(Figure 4-47) with the exception of the case for OCR = 1.375, where, even
though the experimental data did not reach the critical state, the numerical
simulation produced a clearly defined critical state plateau for a lower value of
volumetric strain, 𝜀𝑣𝑜𝑙 . Finally, the ultimate deviatoric stress, 𝐽, obtained by the
lightly overconsolidated samples at 𝑠𝑒𝑞 = 300.0 kPa, was significantly
overestimated by the numerical analysis (Figure 4-48) while the volumetric
behaviour was effectively reproduced for the various overconsolidation ratios
(Figure 4-49).
The inaccurate representation of the CSL and its effect on the simulation of the
soil behaviour was evident from the above results and it is believed that
improvement of the adopted shape would indirectly improve the performance of
the yield and plastic potential functions. Despite this deficiency, the overall
performance of the Hvorslev surface, in combination with the SC shape on the
wet side, was satisfying and capable of modelling the behaviour of both highly
and lightly overconsolidated soils, under saturated and unsaturated conditions.
241

300
SL
seq = 0.0 kPa
C
200
SC
MCC
100 HV
CC
0
-100 0 100 200 300 400 500 600
Figure 4-34: Yield surfaces adopted and stress paths observed in the laboratory, for 𝑠𝑒𝑞 = 0.0 kPa
300 300
OCR = 11.0, seq = 0.0 kPa Deviatoric stress, J (kPa) OCR = 5.5, seq = 0.0 kPa
200 200
100 100
(a) (b)
0 0
0 4 8 12 16 20 0 4 8 12 16 20
-8 -8
OCR = 11.0, seq = 0.0 kPa -7 -7 OCR = 5.5, seq = 0.0 kPa
-6 -6
-5 -5
-4 -4
-3 -3
-2 -2
-1 -1
(c) 0 0 (d)
1 1
0 4 8 12 16 20 0 4 8 12 16 20
lab HV CC MCC SC
Figure 4-35: Laboratory results and numerical simulations adopting four different surfaces; the Hvorslev
surface(HV), the Cam-clay surface (CC), the Modified Cam-clay surface (MCC) and the Sinfonietta
Classica surface (SC), for 𝑠𝑒𝑞 = 0.0 kPa
242
SL

300 C
seq = 100.0 kPa
200 SC
MCC
HV
100
CC
0
-100 0 100 200 300 400 500 600
300 300
200 200
100 100
(a) (b)
0 0
0 4 8 12 16 20 0 4 8 12 16 20
-5 -5
-3 -3
-2 -2
-1 -1
0 0
1 1
2 2
(c) 3 3 (d)
4 4
0 4 8 12 16 20 0 4 8 12 16 20
lab HV CC MCC SC
243
SL
300
seq = 200.0 kPa
C
200 SC
HV MCC
100
CC
0
-100 0 100 200 300 400 500 600
300 300
200 200
100 100
(a) (b)
0 0
0 4 8 12 16 20 0 4 8 12 16 20
-3 -3
OCR = 11.0, seq = 200.0 kPa OCR = 5.5, seq = 200.0 kPa
-2 -2
-1 -1
0 0
1 1
(c) (d)
2 2
0 4 8 12 16 20 0 4 8 12 16 20
lab HV CC MCC SC
244

300 seq = 300.0 kPa
200 SC
HV MCC
100
SL
C
CC
0
-100 0 100 200 300 400 500 600
OCR = 11.0, seq = 300.0 kPa OCR = 5.5, seq = 300.0 kPa
300 300
200 200
100 100
(a) (b)
0 0
0 4 8 12 16 20 0 4 8 12 16 20
-4 -4
-2 -2
-1 -1
0 0
1 1
2 2
(c) (d)
3 3
0 4 8 12 16 20 0 4 8 12 16 20
lab HV CC MCC SC
245
suction, seq = 0.0 kPa

600
laboratory results numerical simulation
500 OCR = 1.375

400
OCR = 1.833
300
OCR = 2.750
200
OCR = 5.5
100
OCR = 11.0
0
0 5 10 15 20 25 30
Figure 4-42: Laboratory results in terms of 𝐽 − 𝜀𝑎 and numerical simulations adopting the Hvorslev
surface(HV) attached to the Sinfonietta Classica (SC), for various values of OCR and at 𝑠𝑒𝑞 = 0.0 kPa

-8
OCR=11
-6 OCR=5.5
OCR=2.75
OCR=1.833
-4
OCR=1.375
-2
6
0 5 10 15 20 25 30
Figure 4-43: Laboratory results in terms of 𝜀𝑣𝑜𝑙 − 𝜀𝑎 and numerical simulations adopting the Hvorslev
surface(HV) attached to the Sinfonietta Classica (SC), for various values of OCR and at 𝑠𝑒𝑞 = 0.0 kPa
246

700
600
OCR = 1.375
500
OCR = 1.833
400
OCR = 2.750
300
OCR = 5.5
200
100
OCR = 11.0
0
0 5 10 15 20
surface(HV) attached to the Sinfonietta Classica (SC),, for various values of OCR and at 𝑠𝑒𝑞 = 100.0
kPa

-4
-2
4
OCR=11
OCR=5.5
6 OCR=2.75
OCR=1.833
OCR=1.375
8
0 5 10 15 20
surface (HV) attached to the Sinfonietta Classica (SC), for various values of OCR and at 𝑠𝑒𝑞 = 100.0
kPa
247

700
600 OCR = 1.375

500
OCR = 1.833
400
OCR = 2.750
300
OCR = 5.5
200
OCR = 11.0
100
0
0 5 10 15 20
surface(HV) attached to the Sinfonietta Classica (SC), for various values of OCR and at 𝑠𝑒𝑞 = 200.0
kPa

-4
-2
4
OCR=11
OCR=5.5
6 OCR=2.75
OCR=1.833
OCR=1.375
8
0 5 10 15 20
kPa
248

700
600
OCR = 1.375
OCR = 1.833
500
400 OCR = 2.750
300 OCR = 5.5
OCR = 11.0
200
100
0
0 5 10 15
kPa

-4
-2
4
OCR=11
OCR=5.5
6 OCR=2.75
OCR=1.833
OCR=1.375
8
0 5 10 15
kPa
249
A curved surface was developed in order to replace the Lagioia et al. (1996)
expression for the yield and plastic potential functions, on the dry side of critical
state and its formulation into a constitutive model, its implementation into a
numerical code and the validity of the latter were presented in the current
chapter. Finally, the new surface, termed the Hvorslev surface, was calibrated
for saturated and unsaturated conditions and the results of the numerical
analyses demonstrated its capabilities and its superiority in simulating the soil
behaviour under various suction levels.
The surface presented herein constitutes an improvement over the Hvorslev

surface adopted in the version of the Modified Cam-clay model (MCC)
implemented by Potts & Zdravkovic (1999). The newly developed expression
requires four parameters, which are assumed to be independent of suction, to
be defined; the inclination of the yield surface, 𝛼𝐻𝑉 , and the gradient of the flow
vector at 𝑝 = −𝑓(𝑠𝑒𝑞 ), 𝛽𝐻𝑉 , similar to the earlier formulation, and two fitting
parameters 𝑛 and 𝑚, which comprise the innovation of the proposed function.
The parameter 𝑛 was introduced in order to bound the stress states within
acceptable limits and controls the position of the yield surface, which coincides
with the CSL when 𝑛 = 0.0, while it becomes a straight line, identical to the one
adopted by Potts & Zdravkovic (1999) when 𝑛 = 1.0 . For values of 𝑛 in
between, the yield surface becomes curved, passing through the 𝑝 axis at
𝑝 = −𝑓(𝑠𝑒𝑞 ) and sharing a common point with the yield surface employed on
the wet side, at 𝑝 = 𝑝𝑐𝑠 . Larger values of 𝑛 result in a larger elastic region being
generated, a larger magnitude of the deviatoric stress at peak, 𝐽𝑝𝑒𝑎𝑘 , a larger
amount of post-yielding softening. However, the deviatoric stress and the
volumetric strains at critical state remain unaffected. Finally, the effect of the
parameter 𝑛 on the soil behaviour simulated was shown to be similar for
saturated and unsaturated conditions.
The parameter 𝑚 controls the variation of the gradient of the plastic flow vector,
𝛽 , from its initial input value, 𝛽𝐻𝑉 , corresponding to 𝑝 = −𝑓(𝑠𝑒𝑞 ), to zero at
𝑝 = 𝑝𝑐𝑠 . Although the overall amount of post-peak softening is not affected, the
250
new parameter specifies how brittle the reproduced behaviour is. The rate of
softening was shown to increase for lower input values of 𝑚 both under fully
and partly saturated conditions.
Despite the two extra parameters, introduced to improve the robustness and
accuracy of the model, the proposed version of the Hvorslev surface is thought
to maintain its simplicity. Its main advantage is that it is prevents the stress
states from obtaining illegal values while employing a unique surface along the
entire dry side of the critical state. Finally, continuity of the yield and plastic
potential surfaces and of the inclination of the CSL, between the dry and wet
sides, was ensured during the formulation of the new expression.
The surface was implemented into the Imperial College Finite Element Program
(ICFEP) to substitute for the Lagioia et al. (1996) expression on the dry side, for
the constitutive models developed by Georgiadis (2003). The alterations
required were presented in terms of the sign convention that the code assumes
(tension is positive) and relate to the derivatives of the yield and plastic potential
functions, with respect to 𝑝, 𝐽, 𝜃, 𝑠𝑒𝑞 and 𝐹𝑠 and the plastic hardening parameter
𝐴.
The validity of the implementation was examined separately for saturated and
unsaturated conditions. Under full saturation, the new expression yielded results
identical to those produced by the existing version of the surface, already
implemented in combination with the MCC model. The additional parameters, 𝑛
and 𝑚 , adopted by the former, were assumed equal to 0.999999 and 1.0,
respectively, so that the two models could be compared. The two models
produced identical results in triaxial compression and extension, under drained
and undrained conditions, for various values of OCR, demonstrating the validity
of the newly developed surface at full saturation. The assumption was made
that this extends to unsaturated states and the performance of the surface, at
various suction levels, was examined in a series of simple parametric studies.
From the drained analyses presented, it was concluded that the Hvorslev
surface provides reasonable results under different values of suction, modelling
successfully the expected increase of strength and predicting larger volumetric
strains at higher suction levels.
251
The model was shown to be capable of accurately simulating the behaviour of a

saturated Hong-Kong marine clay, tested undrained in triaxial compression and
extension, since it successfully reproduced the deviatoric stress, 𝐽, and the
excess pore water pressure, 𝛥𝑢, generated during shearing from two different
initial OCR values; 4 and 8. However, the drained behaviour of saturated
Ancona clay, tested under three values of OCR (40.8, 17.0 and 8.375) was
adequately reproduced only in terms of the deviatoric stress, 𝐽, measured in the
laboratory, while the measured volumetric strains, 𝜀𝑣𝑜𝑙 , were highly
overestimated by the numerical analyses. Nonetheless, the model was shown
to be able to capture the significant rate of softening the samples exhibited,
even though the formation of shear bands cannot be numerically reproduced by
the single element analyses presented.
Furthermore, when the new surface was compared to three typically employed
shapes – the Cam-clay (CC), the modified Cam-clay (MCC) and the sinfonietta
classica (SC) – in terms of their performance on the dry side over a wide range
of suction levels, the Hvorslev surface was the only one which consistently
predicted the peak deviatoric stress, 𝐽𝑝𝑒𝑎𝑘 , exhibited by the artificial silty clay
examined, and at the same time provided the correct indication of the
volumetric behaviour observed experimentally. On the contrary, the SC and
MCC surfaces highly overestimated 𝐽𝑝𝑒𝑎𝑘 at full saturation and for larger OCR
values, with significant potential implications on boundary value problems, as it
is common for failure to occur when suction, and consequently the strength, is
reduced. Moreover, when the CC shape was employed contraction was
predicted instead of dilation (for unsaturated conditions) which in a boundary
value problem would lead to miscalculation of volume changes, indicating
settlements where swelling should occur.
From the calibration of the Hvorslev surface under unsaturated conditions it was
evident that its performance is satisfying despite its simplicity and that its
implementation is essential in order to accurately predict the stresses at highly
overconsolidated states. Further research on the surface for unsaturated
conditions could possibly concentrate on the assumption that the inclination 𝛼𝐻𝑉
and the gradient of the flow vector, 𝛽𝐻𝑉 , are independent of the suction level.
252
For any improvement to be made in this direction, however, further

experimental evidence on the effect that the applied suction has on the yielding
of highly overconsolidated samples, is necessary. Furthermore, it was
demonstrated that the CSL is in need of a similar improvement.
253
chapter 5: MODELLING OF THE SOIL WATER
RETENTION CURVE
5.1 Introduction
In addition to the constitutive model, another important feature in modelling the

unsaturated soil behaviour is the soil-water retention curve (SWRC). The
SWRC defines the relationship between the degree of saturation or the
volumetric water content and the applied suction. It has long been shown that
the retention curve is not unique for a given soil but depends on both the
mechanical and the hydraulic conditions imposed. Mechanical loading of a
sample results in void ratio changes which have been shown to affect the
position of the retention curve. Furthermore, drying and wetting are not
reversible processes and the retention curve may exhibit significant hysteresis
upon cyclic changes of suction, which are thought to instigate irreversible
volumetric strains, even under constant applied load.
In the present chapter, the development of three models for the SWRC, in terms
of degree of saturation, 𝑆𝑟 , and suction, 𝑠, and their implementation into ICFEP
are presented: a specific volume dependent SWRC (v-SWRC), a hysteretic
SWRC (hysteretic-SWRC) and a specific volume dependent hysteretic SWRC
254
Modelling of the Soil Water Retention Curve
(v-hysteretic-SWRC). Finally, the elastic soil compressibility due to changes in

suction is coupled with the SWRC, in an attempt to model the suction induced
irreversible volumetric strains observed in the laboratory.
5.2 Specific volume dependent Soil-Water Retention Curve (v-SWRC)
In a drained deformable soil changes in stresses are related to changes in void

ratio. The degree of saturation, 𝑆𝑟 , is expected to vary accordingly since it is
equal to the ratio of the volume of water, 𝑉𝑤 , over the volume of voids, 𝑉𝑣 .
Numerical analyses of unsaturated soils involving changes of void ratio should,
therefore, take the corresponding effect on the retention curve into account and
appropriate modelling of the SWRC is consequently needed.
The formulation and implementation of such a model is presented and

constitutes a further development over the one already implemented into ICFEP
by Melgarejo (2004) (Section 3.4). The effect of the newly developed curve on
the simulated soil behaviour is then studied and its plausibility is examined
before the curve is employed in the numerical reproduction of laboratory
experiments carried out by Jotisankasa (2005).
5.2.1 Formulation and implementation
Several attempts have been made in the past to model the influence of the void
ratio on the degree of saturation. Wheeler (1996), Vaunat et al. (2000b) and
Wheeler et al. (2003) included the void ratio in the stress variables adopted,
modelling implicitly its effect, while Gallipoli et al. (2003b) suggested that the
SWRC can obtain the form:
𝑚
1
𝑆𝑟 = 𝜓 𝑛
(5.1)
1+ 𝛼∙ 𝑣−1 ∙𝑠
where:
 𝑆𝑟 is the degree of saturation;
 𝑠 is the current suction level;
255
 𝑣 is the specific volume and is equal to 𝑒 + 1, 𝑒 being the void ratio;
 𝛼, 𝑛, 𝑚 and 𝜓 are fitting parameters.
The above equation is based on the one proposed by Van Genuchten (1980),
and assumes that the degree of saturation, 𝑆𝑟 , is 1.0 at zero suction and tends
to 0.0 at infinite suction (i.e. zero residual degree of saturation). These
assumptions, made in the interest of simplicity, limit the applicability of the curve
in combination with the existing models for the mechanical behaviour,
formulated by Georgiadis (2003) and improved during the present research
project, since the transition from saturated to unsaturated conditions is modelled
to occur at the air-entry value of suction, which can assume a value greater
than zero. Furthermore, the minimum degree of saturation, 𝑆𝑟 , achieved in
practice is usually larger than 0.0 as water is present in the form of menisci at
the inter-particle contacts. This water is difficult to remove and small changes in
the degree of saturation result in large increase in suction.
In order to overcome the above limitations, Equation 2.31 was reformulated in

terms of equivalent suction, 𝑠𝑒𝑞 , and degree of saturation at infinite suction, 𝑆𝑟0 :
𝑚
1
𝑆𝑟 = 𝑛 ∙ 1 − 𝑆𝑟,0 + 𝑆𝑟,0 (5.2)
1+ 𝛼∙ 𝑣−1 𝜓 ∙ 𝑠𝑒𝑞
The degree of saturation, 𝑆𝑟 , is 1.0 when 𝑠𝑒𝑞 = 0 (i.e. 𝑠 ≤ 𝑠𝑎𝑖𝑟 ) and, therefore,
de-saturation of the soil is modelled at the air-entry value of suction, while when
𝑠𝑒𝑞 → ∞ the degree of saturation reaches its minimum value, 𝑆𝑟,0 , which is an
input parameter.
The above expression was implemented into ICFEP and can be used in
association with the unsaturated soil constitutive models presented in the
previous chapters, especially when the option of coupling the SWRC with the
increase of apparent cohesion due to suction is employed. The implementation
requires 5 input parameters:
 𝜓 which controls the effect of the specific volume. If 𝜓 = 0.0 the latter has
no effect on the simulated soil behaviour;
256
 𝑠𝑎𝑖𝑟 which is the air-entry value of suction and needs necessarily to be in

agreement with the value prescribed for the mechanical constitutive
model;
 𝑆𝑟,0 which is the degree of saturation corresponding to infinite suction;
 𝛼 , 𝑛 and 𝑚 which are fitting parameters controlling the shape of the

SWRC. Note that 𝛼 > 0.0, 𝑛 > 0.0 and 0.0 ≤ 𝑚 ≤ 1.0. The dimension of
the parameter 𝛼 is 1/stress (i.e. kPa-1) so that the product 𝑠𝑒𝑞 ∙ 𝛼 is
dimensionless;
Furthermore, the slope of the SWRC is required for its implementation in a finite
element program as it is indicative of the water storage and contributes to the
evolution of water flow within unsaturated soils:
𝜕𝑆𝑟 𝑚∙𝑛
𝑅= =− ∙ 1 − 𝑆𝑟0 ∙
𝜕𝑠𝑒𝑞 𝑠𝑒𝑞
𝜓 𝑛 𝑚 (5.3)
𝛼∙ 𝑣−1 ∙ 𝑠𝑒𝑞 1
∙ 𝑛 ∙ 𝑛
1+ 𝛼∙ 𝑣−1 𝜓 ∙ 𝑠𝑒𝑞 1+ 𝛼∙ 𝑣−1 𝜓 ∙ 𝑠𝑒𝑞
The implementation of the above two equations was straightforward, as ICFEP

keeps track of the void ratio during the analysis. Whenever the degree of
saturation, 𝑆𝑟 , or the slope of the SWRC, 𝑅, at a given suction level is required,
the void ratio is readily available for their calculation.
By including the specific volume, 𝑣, in Equation 5.2, the SWRC becomes a

surface (SWRS) in the 𝑠 - 𝑆𝑟 - 𝑣 space, as illustrated in Figure 5-1, for the model
parameters shown in Table 5-1. Also shown in the same figure are hypothetical
retention curves followed under constant specific volume and are referred to, in
the present thesis, as iso-volumetric SWRC’s. It is interesting to note that their
shape resembles the one obtained by the existing model for the retention
behaviour, presented in Chapter 3. By decreasing the specific volume, larger
degrees of saturation, 𝑆𝑟 , correspond to the same suction level, while for
𝑣 → 1.0 fully saturated conditions are sustained throughout the suction range.
Projecting the iso-volumetric curves in the 𝑠 - 𝑆𝑟 plane, the image illustrated in
257
Figure 5-2 is obtained. Evidently, the position of the retention curve is moved
upwards or downwards depending on whether the specific volume is decreasing
or increasing, respectively.
It is highly improbable that unsaturated soils maintain their volume unaffected

during mechanical loading (i.e. changes in applied stress) or hydraulic loading
(i.e. changes in applied suction). Gradual changes of the specific volume
impose a continuous shift of the retention relationship from one iso-volumetric
curve to the next one and the retention point, defined in the 𝑠 - 𝑆𝑟 - 𝑣 space,
moves on the 3-dimensional SWRS.
Figure 5-3 illustrates the retention curve predicted for the model parameters
presented in Table 5-1, when drying, under constant applied load, from full
saturation to 100000.0 kPa of suction, for a soil exhibiting compressibility due to
changes in suction. The initial specific volume – 1.86 at 10.0 kPa of suction –
reduced during drying, however the degree of saturation, 𝑆𝑟 , remained 100%
until the air-entry value of suction was reached at 50.0 kPa. With further drying,
de-saturation occurred and degrees of saturation lower than 1.0 were
calculated, at the same time as the specific volume was reduced according to
the elastic coefficient of compressibility, 𝜅𝑠 , prescribed equal to 0.06. The
SWRC predicted (black line in Figure 5-3) is shown to travel on the retention
surface, crossing successive iso-volumetric curves, as described above.
Increasing the suction above its yield value, assumed to be 1000.0 kPa, caused
an abrupt change in the slope of the SWRC, as the coefficient of compressibility
attained its elasto-plastic value, 𝜆𝑠 = 0.09 . Thereafter, the curve followed a
distinct path, illustrated as the grey line in Figure 5-3, corresponding to post-
yield changes of suction. The dotted line is the extension of the elastic path and
is shown in the figure for comparison with the elasto-plastic one.
The projection of the predicted SWRC in the 𝑠 - 𝑆𝑟 plane is illustrated in Figure

5-4, together with various iso-volumetric curves. The kink in the slope of the
retention curve when yielding occurred at 1000.0 kPa of suction, is clearly
shown as the degree of saturation, 𝑆𝑟 , reduced more slowly on the elasto-
plastic branch of the curve. This feature, however, should not be solely
attributed to the change in compressibility, as the iso-volumetric curves are not
258
equi-spaced, even though constant intervals between the corresponding values

of specific volume are applied (i.e. 2.0, 1.9, 1.8 etc). The distance between
them increases for decreasing values of 𝑣, intensifying the above mentioned
aspect of modelling. This feature may unrealistically result in the prediction of
increasing 𝑆𝑟 with increasing 𝑠, especially for large values of parameter 𝜓, and
therefore, the model should be used with caution at low saturation levels.
The effect of the specific volume on the SWRC predicted and, consequently,
the distance between the iso-volumetric curves, is controlled by the model
parameter 𝜓 and is expected to be less pronounced for decreasing values of
this parameter, while no effect is taken into account when 𝜓 = 0.
The retention curve shown above (Figure 5-3 and Figure 5-4) was obtained for
𝜓 = 2.5 and the distance between the iso-volumetric curves produced was
considerable. The simulation of the drying process was repeated assuming 𝜓 =
1.0 and a different SWRS was generated in the 𝑠 - 𝑆𝑟 - 𝑣 space, predicting a
new retention curve (Figure 5-5) which was steeper than the former one. The
distance between the respective iso-volumetric curves, projected on the 𝑠 - 𝑆𝑟
plane (Figure 5-6), reduced and thus demonstrated the decreasing effect of the
void ratio changes on the retention relationship. Furthermore, the slope of the
SWRC did not visibly reveal the transition from elastic to elasto-plastic changes
of suction, providing further support to the above conclusion.
Finally, for 𝜓 = 0.0, the slope of the SWRS is shown in Figure 5-7 to be
independent of the specific volume and the projections of the iso-volumetric
curves in the 𝑠 - 𝑆𝑟 plane (Figure 5-8) coincided with each other and with the
projection of the SWRC itself, confirming that the effect of 𝑣 was not taken into
account.
The difference in the SWRC’s, observed in the abovementioned figures, is

expected to influence the mechanical behaviour reproduced with the
unsaturated soil constitutive models (Georgiadis, 2003 and Chapter 4 of the
present thesis), when the degree of saturation, 𝑆𝑟 , is employed in the evaluation
of the apparent cohesion. The computed results are affected by the value of the
parameter 𝜓 employed in the analysis, even if no changes of suction occur. To
259
demonstrate the validity of this argument, the drained triaxial compression

analysis for OCR = 11 and at 𝑠𝑒𝑞 = 300.0 kPa, presented in Section 4.5.2, was
repeated for 𝜓 of 2.5, 1.0 and 0.0, using the v-SWCR parameters of Table 5-1.
Even though the analyses were drained and the suction level remained
constant, the specific volume changed during shearing producing the variation
of 𝑆𝑟 shown in Figure 5-9 (a) and (b), for 𝜓 = 2.5 and 1.0, respectively, while it
had no effect when 𝜓 = 0.0 (Figure 5-9 (c)). Although this variation was small,
different degrees of saturation, 𝑆𝑟 , corresponding to the same value of specific
volume, were calculated at the start of each analysis and as a result the
positions of the critical state line (CSL) and the Hvorslev surface varied with 𝜓.
Indeed, the deviatoric stress, 𝐽, both at peak and at critical state, was lower for
decreasing values of 𝜓, as illustrated in Figure 5-10 (a). Furthermore, larger
dilative volumetric strains corresponded to 𝜓 = 0.0 and their value decreased for
𝜓 = 1.0, while contraction was predicted when 𝜓 = 2.5 (Figure 5-10 (b)). For the
latter case, the elastic region expanded significantly to the left and the stress
path violated the yielding condition on the wet side of critical state, as is evident
from Figure 5-11.
It should be noted at this point that had the specific volume been larger than 2
(i.e. 𝑣 − 1 > 1) the opposite of the above trend would have been observed as
smaller degree of saturation, 𝑆𝑟 , would have been calculated for increasing
values of 𝜓. Furthermore, for the particular case where 𝑣 − 1 = 1, the value of
parameter 𝜓 does not affect the position of the SWRC.
The above results highlight the significance of modelling the effect of void ratio
on the retention relationship. The effectiveness of the proposed expression is
demonstrated in the following sections.
Table 5-1: v-SWRC model parameters employed for the generation of the 3-dimensional surface
illustrated in Figure 4-49
v-SWRC model
𝜓 2.5 𝛼 0.01 kPa-1
𝑠𝑎𝑖𝑟 50.0 kPa 𝑛 1.7
𝑆𝑟,0 0.1 𝑚 0.5
260
Figure 5-1: 3-dimensional SWRS in the 𝑠 - 𝑆𝑟 - 𝑣 space and SWRC’s at constant specific volume (iso-
volumetric SWRC’s)
1.0
 = 1.2
0.8
 = 1.3
 = 1.4
 = 1.5
0.6  = 1.6
iso-volumetric
0.4  = 1.6 SWRC's
 = 1.7
 = 1.8
 = 1.9
0.2
 = 2.0
Sr0
sair s0,r
0.0
100 1000 10000 100000
Suction, log s (kPa)
Figure 5-2: Projection in the 𝑠 - 𝑆𝑟 plane of iso-volumetric SWRC’s
261
Figure 5-3: 3-dimensional SWRS in the 𝑠 - 𝑆𝑟 - 𝑣 space generated for 𝜓 = 2.5

1.0
 s0 ln s (kPa)
s
0.8
s
1
0.6
0.4
0.2
Sr0
sair s0 s0,r
0.0
100 1000 10000 100000
Figure 5-4: Projection in the 𝑠 - 𝑆𝑟 plane of the SWRC generated for 𝜓 = 2.5
262

1.0
 s0 ln s (kPa)
s
0.8
s
1
0.6
0.4
0.2
Sr0
sair s0 s0,r
0.0
100 1000 10000 100000

263

1.0
 s0 ln s (kPa)
s
0.8
s
1
0.6
0.4
0.2
Sr0
sair s0 s0,r
0.0
100 1000 10000 100000

264
 = 2.5  = 1.0  = 0.0

0.668 0.518 0.430
0.666
0.664 0.516
0.662
0.660 0.514
0.658
(a) (b) (c)
0.656 0.512 0.429
0 5 10 0 5 10 0 5 10
Axial strain,  (%) Axial strain,  (%) Axial strain,  (%)
Figure 5-9: Variation of the degree of saturation, 𝑆𝑟 , due to changes in the specific volume occurring
during drained analyses, for (a) 𝜓 = 2.5; (b) 𝜓 = 1.0 and (c) 𝜓 = 0.0
400 -0.6
-0.4  = 0.0
300 -0.2
0  = 1.0
200
 = 2.5 0.2
 = 1.0 0.4
100
 = 0.0  = 2.5
(a) 0.6 (b)
0 0.8
0 2 4 6 8 10 0 2 4 6 8 10
Figure 5-10: (a) Deviatoric stress, 𝐽, and (b) volumetric strains, 𝜀𝑣𝑜𝑙 , versus axial strain, 𝜀𝑎 ,calculated
for different values of the parameter 𝜓
300
SC
200
h
HV
pat
ess
SL
100
Str
C
0
-200 -100 0 100 200 300 400 500 600
Figure 5-11: The yield surface and critical state line generated for 𝜓 = 2.5 and the stress path followed
for OCR = 11.0
265
5.2.2 Validation
For the validation of the v-SWRC model, two retention curves generated by
ICFEP were compared to those computed using Microsoft Excel. The model
parameters employed can be found in Table 5-2 and are the same as those
obtained by the calibration of the model, subsequently presented. Starting from
two different values of specific volume, 1.50 and 1.70, a drying path was
induced and the corresponding degree of saturation, 𝑆𝑟 , was computed,
assuming the elastic coefficient of compressibility with suction, 𝜅𝑠 , to be 0.011.
The numerical code and the Excel spreadsheet produced identical results and
the SWRC’s coincided for both initial values of specific volume, as illustrated in
Figure 5-12, indicating that the model was correctly implemented into ICFEP.
Table 5-2: v-SWRC model parameters employed for the validation of the model
v-SWRC model
𝜓 5.0 𝛼 0.03 kPa-1

𝑠𝑎𝑖𝑟 2.0 kPa 𝑛 0.5
𝑆𝑟,0 0.0 𝑚 0.875
1.0
Degree of saturation, Sr
0.8
0.6
0.4  = 1.70 (ICFEP)

 = 1.50 (ICFEP)
0.2  = 1.70 (analytical)
 = 1.50 (analytical)
0.0
1 10 100 1000 10000
Figure 5-12: Comparison of two SWRC’s produced by ICFEP with their analytical solution, for two
values of specific volume, 𝑣, 1.50 and 1.70
266
5.2.3 Calibration
Experimental tests performed by Jotisankasa (2005) on an artificial material

consisting of 70% HPF4 silt, 20% Speswhite Kaolin and 10% London clay, were
employed in the calibration of the v-SWRC model. The same artificial material,
referred to as Soil A, was employed by Cunningham (2000) in a reconstituted
form, whereas Jotisankasa (2005) tested samples of soil A subjected to static
compaction to states dry-of-optimum.
Two different types of experiments were employed in the calibration: (a)

unconfined drying and wetting and (b) isotropic compression of samples
compacted to various initial conditions.
In the first of the above mentioned series of tests, performed in order to

investigate the influence of void ratio on the SWRC, Jotisankasa (2005)
employed four samples. Two were compacted to initial specific volume, 𝑣, of
1.50 and degree of saturation, 𝑆𝑟 , of 53% and the other two were compacted to
initial specific volume, 𝑣, of 1.70 and degree of saturation, 𝑆𝑟 , of 38%. One
sample at each specific volume was subjected to unconfined drying and the
other sample was subjected to unconfined wetting. The resulting changes in the
degree of saturation and the specific volume were monitored.
The initial states are illustrated in Figure 5-13 (a) in terms of specific volume, 𝑣,
and the logarithm of 𝑠. The constitutive models implemented by Georgiadis
(2003) assume a linear variation in this plane. The respective initial retention
points, defined in the 𝑠 - 𝑆𝑟 plane, are presented in Figure 5-13 (b). Also shown
in the above figure are the laboratory data obtained during the first drying and
the first wetting from their as-compacted states.
From the data presented in Figure 5-13 (a) the elastic coefficient of soil
compressibility with changes in suction, 𝜅𝑠 , was estimated to be 0.011. The
wetting path followed by the sample compacted at an initial specific volume of
1.50, was employed in the calibration of the v-SWRC model and the resulting
parameters have already been presented in Table 5-2. The remaining paths
were generated adopting the same parameters and exhibited reasonable
267
agreement with the experimental data, capturing the effect of void ratio on the
position of the retention curve.
As expected, the SWRC for the larger value of specific volume was predicted to
lie below the curve for 𝑣 = 1.50, for the whole range of suctions studied (Figure
5-13 (b)). The experimental data, however, showed that the two curves merged
for suctions larger than 2000.0 kPa, suggesting that a unique relationship exists
between the degree of saturation, 𝑆𝑟 , and the suction, s. Furthermore, the
degree of saturation, 𝑆𝑟 , was over-predicted for 𝑣 = 1.7, for suctions in the range
of 50.0 to 500.0 kPa, as the Van Genuchten (1980) type of equation employed
cannot capture the bi-modality (as defined by Jotisankasa, 2005) of the curve
observed in the laboratory.
Despite the minor deficiencies in the simulation of the retention curve, the v-
SWRC model was used to predict the changes in the degree of saturation
reported by Jotisankasa (2005), during isotropic compression of the same
material compacted to an initial specific volume of 1.70. Each of the three
samples tested, namely TC16, 18 and 29, was wetted to a different water
content and was subsequently compressed maintaining the latter constant. In
tests TC16 and 18 (shown later in Figure 5-15 (a) and (b), respectively), the
samples were isotropically compressed to 800.0 kPa with a relatively small
reduction in suction, which was assumed constant in the numerical analyses
and equal to 150.0 and 300.0 kPa, respectively. In test TC29 (Figure 5-15 (c)),
the sample was loaded to 50.0 kPa and was then dried to 1000.0 kPa of
suction, before isotropic compression to 800.0 kPa under constant water
content took place. Finally, the sample was unloaded to 200.0 kPa of cell
pressure.
For the numerical simulation of the above three tests, the model parameters
presented in Table 5-2 were adopted, since the same initial conditions were
applied. In the drained analyses performed, the constitutive model for
unsaturated soils by Georgiadis (2003) was used for the reproduction of the
mechanical behaviour (Model 1, Option 1) and the increase of apparent
cohesion with suction was controlled by the v-SWRC model. The parameters
used are presented in Table 5-3 and are subsequently explained. Since only
268
isotropic loading/unloading was involved in the analyses, the yield surface in the
𝑝 − 𝐽 plane was not violated and its shape (controlled by the parameters 𝛼𝑔,𝑓 ,
𝜇𝑔,𝑓 and 𝑀𝑔,𝑓 ) was irrelevant. The elastic shear modulus, 𝐺, was also irrelevant
and assumed equal to 1000.0 kPa. As there was no indication of the secondary
yield surface being reached, the parameter controlling its position, 𝑠0 , was set
equal to 10000.0 kPa and the value of the elasto-plastic compressibility with
suction, 𝜆𝑠 , was not, therefore, engaged. The values of 𝑣1 , 𝜆(0) and 𝜅 employed
in the analyses were evaluated by Jotisankasa (2005) through isotropic
compression tests at full saturation. The elastic coefficient of soil compressibility
with suction, 𝜅𝑠 , was 0.011 as already discussed and the air-entry value of
suction, 𝑠𝑎𝑖𝑟 , was equal to 2.0 kPa, as for the v-SWCR model. Finally, the
parameters 𝑟 , 𝛽 and 𝑝𝑐 , controlling the shape of the loading-collapse (LC)
curve, were fitted to the experimental data, as illustrated in Figure 5-14.
Starting from the initial states presented in Table 5-4, the isotropic compression
tests TC16, 18 and 29 were numerically simulated. The degrees of saturation
computed during loading are compared to those measured in the laboratory, in
Figure 5-15. Also shown in the same figure, are the degrees of saturation
calculated when the simple Van Genuchten (1980) type SWRC, already
implemented into ICFEP by Melgarejo (2004) (Section 3.4 Chapter 3), was
adopted in the numerical analyses.
Despite the general overestimation of the magnitude of the degrees of

saturation, 𝑆𝑟 , measured in the laboratory, the v-SWRC evidently produced a
better prediction of the changes in 𝑆𝑟 , in comparison with the simple Van
Genuchten (1980) curve, which predicted no change in 𝑆𝑟 in any of the tests. It
is interesting to note that the model was capable of capturing the change of
slope observed when yielding on the LC curve occurred, for the tests TC 16 and
18. Moreover, the model correctly predicted the difference in the degree of
saturation, 𝑆𝑟 , during isotropic loading and unloading for test TC29 as the
elasto-plastic coefficient of compressibility, 𝜆(𝑠𝑒𝑞 ), employed upon loading (the
LC curve was already violated) was replaced by the elastic one, 𝜅 , upon
unloading.
269
From the analyses presented above it can be concluded that the v-SWRC
model is capable of capturing the effect of void ratio on the retention behaviour
under various loading conditions, and of providing the correct indication of how
the degree of saturation, 𝑆𝑟 , varies due to changes in suction and in applied
stresses. As the SWRC is modelled to control the increase of apparent
cohesion with suction, the above variation is thought to be significant in the
prediction of the mechanical behaviour of unsaturated soils.
Table 5-3: Parameters calibrated for the simulation of isotropic compression tests with the constitutive
model developed by Georgiadis (2003)
Constitutive model for unsaturated soils
𝑟 0.6 𝜆(0) 0.12
𝑘 v-SWRC 𝜅 0.006
𝑠𝑎𝑖𝑟 (kPa) 2.0 𝑣1 2.1
𝛼𝑔,𝑓 0.4 𝛽 0.0085
𝜇𝑔,𝑓 0.9 𝜆𝑠 0.12
𝑀𝑔,𝑓 1.0 𝜅𝑠 0.011
𝑝𝑎𝑡𝑚 (kPa) 100.0 𝑝𝑐 (kPa) 1.5
𝐺 (kPa) 1000.0 𝑠0 (kPa) 10000.0
Table 5-4: Initial states adopted in the analyses of isotropic compression tests
Initial states
TC16 TC18 TC29
Param. Value Param. Value Param. Value
𝑝𝑛𝑒𝑡 13.0 kPa 𝑝𝑛𝑒𝑡 12.0 kPa 𝑝𝑛𝑒𝑡 29.0 kPa
𝑠 150.0 kPa 𝑠 300.0 kPa 𝑠 600.0 kPa
𝑒 0.714 𝑒 0.733 𝑒 0.712
270
1.8
v = 1.7
Specific volume, v ( )
1.6
v = 1.5
(a)
1.4
1 10 100 1000 10000 100000
initial points
drying,  = 1.50 (lab.)
wetting,  = 1.50 (lab.)
drying,  = 1.70 (lab.)
1.0 wetting,  = 1.70 (lab.)
drying,  = 1.50 (ICFEP)

0.8 wetting,  = 1.50 (ICFEP)
v = 1.5 drying,  = 1.70 (ICFEP)
0.6 wetting,  = 1.70 (ICFEP)
0.4
v = 1.7
0.2
(b)
0.0
1 10 100 1000 10000 100000
Figure 5-13: Changes in the (a) specific volume, 𝑣, and (b) in the degree of saturation, 𝑆𝑟 , measure
during drying and wetting from the as-compacted states (laboratory data after Jotisankasa, 2005)
1000.0
Suction, s (kPa )
750.0
500.0
lab. data
250.0 ICFEP
0.0
1 10 100 1000
Mean net stress, log p (kPa)
Figure 5-14: The LC curve obtained by Jotisankasa (2005) and its numerical reproduction
271
Degree of saturation, Sr ( ) 1.0
TC16 test
0.8
0.6
0.4
0.2
(a)
0.0
10 100 1000
Mean net stress, log pnet (kPa)
1.0
TC18 test
0.8
0.6
0.4
0.2
(b)
0.0
10 100 1000
1.0
0.8 TC29 test
0.6
0.4
0.2
(c)
0.0
10 100 1000
lab. data v_SWRC model Van Genuchten model
Figure 5-15: Changes in the degree of saturation generated in the laboratory and numerically, adopting
two SWRC models; the v-SWRC and a simple Van Genuchten type SWRC
272
5.3 Hysteretic Soil-Water Retention Curve (hysteretic-SWRC)
Several authors have attempted to include the hydraulic hysteresis into

numerical modelling of unsaturated soils (Wheeler, 1996; Vaunat et al., 2000b,
Wheeler et al., 2003; Li, 2005; Sun et al., 2007a; Lloret et al., 2009; Pedroso &
Williams, 2010). The models by Li (2005) and Pedroso & Williams (2010) were
presented in Chapter 2. In these models the scanning paths approach the
primary paths asymptotically. Another typical example is the work of Wheeler et
al. (2003) who proposed the basic shape illustrated in Figure 2-41 of Chapter 2.
An alternative approach is proposed in the present thesis. The primary and the
scanning paths are simple geometric curves which have a common tangent at
the point of intersection. Despite its geometric simplicity, the model has
demonstrated effectiveness in the representation of laboratory data.
5.3.1 Formulation
The hysteretic SWRC model developed and presented herein was aimed at
satisfying two fundamental requirements: the need for a realistic shape both for
the primary and the scanning paths and the necessity for a smooth transition
from scanning to primary paths.
The first requirement arises from the fact that the slope of the retention curve is
essential in a coupled analysis as it affects the flow of water. More specifically,
the flow generated due to changes in water content, and therefore in the degree
of saturation, is dependent on the gradient of the SWRC which controls the
water storage within the soil (Section 3.2.2). Therefore, realistic shapes for the
primary and the scanning paths are of importance.
The second requirement, for a smooth transition from one type of path to the
other, relates to numerical singularities. The robustness of the model is believed
to be improved when abrupt changes on the slope of the retention curve are
avoided.
273
5.3.1.1 Primary drying and wetting paths
The model is formulated in terms of degree of saturation, 𝑆𝑟 , and equivalent

suction, 𝑠𝑒𝑞 = 𝑠 − 𝑠𝑎𝑖𝑟 , 𝑠𝑎𝑖𝑟 being the air-entry value of suction.
The shape of the primary paths is shown schematically in Figure 5-16. De-
saturation during drying and full saturation during wetting are assumed to occur
at the air - entry value, 𝑠𝑎𝑖𝑟 , ensuring the two primary paths have a common
point (𝑠𝑒𝑞 , 𝑆𝑟 ) = (0.0, 1.0). Furthermore, it is assumed that the residual point is
also common for the two primary paths and that occurs at 0.0 residual degree of
saturation. The corresponding suction, 𝑠0 , is a model parameter (it should be
noted at this point that the Van Genuchten (1980) expression was not employed
for the reproduction of the primary curves for the reason that the residual
degree of saturation is not achieved at a finite value of suction; as discussed in
Chapter 2, this may lead to numerical inconsistencies). The above assumptions,
made in the interest of simplification (a refined version of the model is proposed
in Section 5.6), allow the following s-shape curve to be adopted for the primary
𝑑𝑟 𝑤𝑒𝑡
drying (𝑆𝑟,𝑝𝑟 ) and the primary wetting (𝑆𝑟,𝑝𝑟 ) curves:
1
1−𝑠 ∙ 𝑠𝑒𝑞
𝑑𝑟 ,𝑤𝑒𝑡
𝑆𝑟,𝑝𝑟 =
0,𝑒𝑞 (5.4)
1 + 𝛼𝑑 ,𝑤 ∙ 𝑠𝑒𝑞
where 𝑠0,𝑒𝑞 = 𝑠0 − 𝑠𝑎𝑖𝑟 and 𝛼 is a fitting parameter, carrying the index 𝑑 for
drying and 𝑤 for wetting. For the wetting path to lie beneath the drying path, 𝛼𝑤
has to be larger than 𝛼𝑑 , while if they are equal a monotonic curve is generated.
The gradient of the primary paths for the current value of equivalent suction is:
1
𝑠0,𝑒𝑞 + 𝛼𝑑,𝑤
𝜕𝑆𝑟,𝑝𝑟
=− (5.5)
2
𝜕𝑠𝑒𝑞 1 + 𝛼𝑑 ,𝑤 ∙ 𝑠𝑒𝑞
5.3.1.2 Scanning drying and wetting paths
On drying from an initial retention point A, defined in the 𝑠 – 𝑆𝑟 plane by its co-
ordinates (𝑠𝐴 , 𝑆𝑟,𝐴 ) and positioned in between the two primary curves, the soil is
274
assumed to follow the scanning path ABdr shown in Figure 5-16. This scanning
path is assumed to be the arc of a circle, centred on the vertical line passing
through point A so that the suction corresponding to the centre of the circle is
equal to the suction at point A, 𝑠𝐴 . The circle and the primary drying curve have
a common tangent at point Bdr (𝑠𝐵𝑑𝑟 , 𝑆𝑟,𝐵
𝑑𝑟
), also shown in Figure 5-16. In this way,
the slope of the scanning path is always zero at point A and a smooth transition
from the scanning to the primary drying path is provided at point B dr. The radius
of the circle, 𝑟𝑑𝑟 , and the suction at point Bdr, 𝑠𝐵𝑑𝑟 , need to be identified.
The expression for the scanning drying path is:
𝑑𝑟 2 2
𝑆𝑟,𝑠𝑐𝑎𝑛 = 𝑆𝑟,𝐴 − 𝑟𝑑𝑟 + 𝑟𝑑𝑟 − log 𝑠𝑒𝑞 − log 𝑠𝐴 (5.6)
The slope of the scanning drying path for the current value of suction is:
𝑑𝑟 log 𝑠𝑒𝑞 − log 𝑠𝐴 1

𝜕𝑆𝑟,𝑠𝑐𝑎𝑛 2 2 − 2
=− ∙ 𝑟𝑑𝑟 − log 𝑠𝑒𝑞 − log 𝑠𝐴 (5.7)
𝜕𝑠𝑒𝑞 𝑠𝑒𝑞 ∙ ln 10
As noted above, to define the scanning drying path, the radius 𝑟𝑑𝑟 is required.
As Bdr is a common point for the two curves given by Equations 5.4 and 5.6:
1
1−𝑠 ∙ 𝑠𝐵𝑑𝑟
0,𝑒𝑞 2 2 (5.8)
= 𝑆𝑟,𝐴 − 𝑟𝑑𝑟 + 𝑟𝑑𝑟 − log 𝑠𝐵𝑑𝑟 − log 𝑠𝐴
1 + 𝛼𝑑 ∙ 𝑠𝐵𝑑𝑟
Furthermore, the two curves share a common tangent at point B dr:
1
𝑠0,𝑒𝑞 + 𝛼𝑑 log 𝑠𝐵𝑑𝑟 − log 𝑠𝐴 2
1
2 − 2
(5.9)
− 2 =− ∙ 𝑟𝑑𝑟 − log 𝑠𝐵𝑑𝑟 − log 𝑠𝐴
1 + 𝛼𝑑 ∙ 𝑠𝐵𝑑𝑟 𝑠𝐵𝑑𝑟 ∙ ln 10
The above two Equations, 5.8 and 5.9, form a system where the suction at point
Bdr, 𝑠𝐵𝑑𝑟 , and the radius 𝑟𝑑𝑟 are the two unknown variables. However, solution of
the system is not straightforward and requires a numerical approach. Newton’s
method was applied to solve the system of equations as subsequently
explained in Section 5.3.2.2.
275
On wetting from the same initial point A(𝑠𝐴 , 𝑆𝑟,𝐴 ), the soil follows the wetting
scanning path ABwet shown in Figure 5-16, rejoining the primary wetting path at
point Bwet (𝑠𝐵𝑤𝑒𝑡 , 𝑆𝑟,𝐵
𝑤𝑒𝑡
). Similar to the drying scanning path, the wetting scanning
path is assumed to be the arc of a circle, centred on the vertical line passing
through point A. The circle and the primary wetting curve have a common
tangent at point Bwet (𝑠𝐵𝑤𝑒𝑡 , 𝑆𝑟𝑤𝑒𝑡
,𝐵 ). The expressions for the scanning wetting path
and its gradient at the current value of equivalent suction, s eq, are given below:
𝑤𝑒𝑡 2 2
𝑆𝑟,𝑠𝑐𝑎𝑛 = 𝑆𝑟,𝐴 + 𝑟𝑤𝑒𝑡 − 𝑟𝑤𝑒𝑡 − log 𝑠𝐴 − log 𝑠𝑒𝑞 (5.10)
and:
𝑤𝑒𝑡 log 𝑠𝐴 − log 𝑠𝑒𝑞 1

𝜕𝑆𝑟,𝑠𝑐𝑎𝑛 2 2 − 2
=− ∙ 𝑟𝑤𝑒𝑡 − log 𝑠𝐴 − log 𝑠𝑒𝑞 (5.11)
𝜕𝑠𝑒𝑞 𝑠𝑒𝑞 ∙ ln 10
Similar to drying, the following system of equations needs to be solved, in terms

of the radius, 𝑟𝑤𝑒𝑡 , and the equivalent suction at point Bwet, 𝑠𝐵𝑤𝑒 𝑡 :
1
1−𝑠 ∙ 𝑠𝐵𝑤𝑒𝑡
0,𝑒𝑞
= 𝑆𝑟 ,𝐴 + 𝑟𝑤𝑒𝑡 − 2
𝑟𝑤𝑒𝑡 − log 𝑠𝐴 − log 𝑠𝐵𝑤𝑒𝑡 2 (5.12)
1 + 𝛼𝑤 ∙ 𝑠𝐵𝑤𝑒𝑡
and:
1
𝑠0,𝑒𝑞 + 𝛼𝑤 log 𝑠𝐴 − log 𝑠𝐵𝑤𝑒𝑡 2 2 −1 2 (5.13)
− =− ∙ 𝑟𝑤𝑒𝑡 − log 𝑠𝐴 − log 𝑠𝐵𝑤𝑒𝑡
1 + 𝛼𝑤 ∙ 𝑠𝐵𝑤𝑒𝑡 2 𝑠𝐵𝑤𝑒𝑡 ∙ ln 10
For the solution of the system the Newton method was once more employed.
5.3.1.3 Model parameters
Four model parameters are required to define the hysteretic-SWRC model

described above:
 two fitting parameters 𝛼𝑑 and 𝛼𝑤 , for the primary drying and wetting
paths, respectively;
276
 the suctions at the air-entry point, 𝑠𝑎𝑖𝑟 , and at zero residual degree of
saturation, 𝑠0 .
For the primary drying path to lie above the primary wetting one, 𝛼𝑑 needs to be
smaller than 𝛼𝑤 . In addition, the suction at zero degree of saturation, 𝑠0 , needs
to be larger than the air-entry value of suction, 𝑠𝑎𝑖𝑟 .
The model parameters dictate the shape and the position of the primary curves,
which remain unvarying during the analysis. On the contrary, the scanning
paths are not directly controlled by the model parameters; their shape is always
circular and the actual path followed is determined primarily by the initial
retention state (point A in Figure 5-16) and indirectly by the model parameters
through the necessity of joining the primary paths with a common tangent. This
lack of explicit control over the scanning paths could be regarded as a limitation
of the model which, however, guarantees simplicity.
rwet rwet
1
pr
scanning
im
Bwet
ar
wetting A
yd
pr
ry
im
in
ar
g
yw scanning
et Bdr
tin drying
g
0
sair sBwet sA sBdr s0
rdr rdr
Figure 5-16: Primary and scanning paths assumed for the hysteretic-SWRC model
277
5.3.2 Implementation
The above mentioned hysteretic-SWRC model was implemented in ICFEP.

Depending on the suction change and on the suction level, the appropriate path
needs to be selected. The suction change indicates the direction of hydraulic
loading (i.e. drying or wetting), while based on the suction level itself distinction
is made in the employment of the corresponding primary and scanning paths.
For this procedure to be feasible, a number of variables need to be stored

during the analysis. It is essential to register information concerning the last
retention point (i.e. point defined by the applied suction, 𝑠, and the
corresponding degree of saturation, 𝑆𝑟 ) before a change in the direction of
hydraulic loading is detected. This point is commonly referred to as the reversal
point. If the soil is wetted from an initial point A, shown in Figure 5-17, to point
B, point A is considered to be the reversal point for this wetting path. If the soil
is subsequently dried to point C, point B is the new reversal point for this drying
path.
The variables stored are herein referred to as reversal parameters and consist
of the following quantities which require recalculation every time that a reversal
in the direction of hydraulic loading occurs:
 the suction, 𝑠𝑟𝑒𝑣 , and
 the degree of saturation, 𝑆𝑟,𝑟𝑒𝑣 , of the reversal retention point;
 the radius of the corresponding circle, 𝑟;
 the suction, 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 , at the intersection point with the primary wetting
curve if a reversal from drying to wetting has been detected and with the
primary drying path if a change from wetting to drying has been detected;
 the direction of hydraulic loading: 1.0 for drying and -1.0 for wetting.
The initialisation and updating of the reversal parameters are of central

importance and are subsequently explained.
278
Figure 5-17: Wetting and drying cycles from initial retention point A
5.3.2.1 Initialisation of the reversal parameters
The initial soil state, consisting of the stress state as well as the degree of
saturation, has to be established at the beginning of the finite element analysis.
For the initialisation of the reversal parameters, drying is assumed and the
direction of hydraulic loading is equal to 1.0. If fully saturated or residual
conditions apply the model is not employed in the analysis and the reversal
parameters are set to the values presented for both cases in Table 5-5.
For unsaturated conditions which do not exceed the residual point, the initial
equivalent suction is calculated based on the air-entry value of suction, 𝑠𝑎𝑖𝑟 , and
is used in the calculation of the corresponding degrees of saturation on the
primary drying and wetting curves, 𝑆𝑟𝑑𝑟 and 𝑆𝑟𝑤𝑒𝑡 , respectively, which should
bound the initial degree of saturation, 𝑆𝑟 . In case the latter is found to lie outside
the limiting values 𝑆𝑟𝑑𝑟 and 𝑆𝑟𝑤𝑒𝑡 , the program is terminated. Furthermore, the
program checks if the initial retention point prescribed lies on one of the two
primary curves and adopts the exact degree of saturation computed applying
the relevant model parameter, 𝛼𝑑 or 𝛼𝑤 . The equivalent suction and the
calculated degree of saturation form the coordinates of the initial reversal point,
𝑠𝑟𝑒𝑣 and 𝑆𝑟,𝑟𝑒𝑣 .
If the point is found to be positioned on the primary drying curve, the suction
𝑠𝑐𝑜𝑚𝑚𝑜𝑛 is assumed to be equal to 𝑠𝑟𝑒𝑣 and the radius, 𝑟, is set to 0.0. If on the
other hand, the initial retention point is positioned on the primary wetting curve,
279
the system of Equations 5.8 and 5.9 needs to be solved (see Section 5.3.2.3
below) in order to obtain the suction at the common point, 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 = 𝑠𝐵𝑑𝑟 , and
the radius of the corresponding circle, 𝑟 = 𝑟𝑑𝑟 . The same system of equations
also needs to be solved if the initial retention point lies within the primary paths
for the reversal parameters 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 and 𝑟 to be computed. In the latter case,
𝑆𝑟,𝑟𝑒𝑣 is set equal to the initially prescribed degree of saturation, 𝑆𝑟 . The above
procedure is summarised in Table 5-5.
Table 5-5: Initialisation of the reversal parameters
Initial reversal parameters
On the
On the Within the
Full Residual primary
primary primary
saturation conditions wetting
drying curve curves
curve
Param. Value Value Value Value Value
𝑠𝑟𝑒𝑣 0.0 𝑠0,𝑠𝑒𝑞 𝑠𝑒𝑞 𝑠𝑒𝑞 𝑠𝑒𝑞

𝑑𝑟 𝑤𝑒𝑡
𝑆𝑟,𝑟𝑒𝑣 1.0 0.0 𝑆𝑟,𝑝𝑟 𝑆𝑟,𝑝𝑟 𝑆𝑟
𝑠𝑐𝑜𝑚𝑚𝑜𝑛 0.0 𝑠0,𝑠𝑒𝑞 𝑠𝑒𝑞 𝑠𝐵𝑑𝑟 𝑠𝐵𝑑𝑟
𝑟 0.0 0.0 0.0 𝑟𝑑𝑟 𝑟𝑑𝑟
Direction 1.0 (drying) 1.0 (drying) 1.0 (drying) 1.0 (drying) 1.0 (drying)
Table 5-6: Updating of the reversal parameters

Updating of the reversal parameters
Reversal from drying to wetting Reversal from wetting to drying

Resi- Resi-
Solving Full On Solving Full On
dual dual
the satura- primary the satura- primary
condi- condi-
system tion wetting system tion drying
tions tions
Param. Value Value Value Value Value Value Value Value
𝑠𝑟𝑒𝑣 𝑠𝑒𝑞 0.0 𝑠0,𝑠𝑒𝑞 𝑠𝑒𝑞 𝑠𝑒𝑞 0.0 𝑠0,𝑠𝑒𝑞 𝑠𝑒𝑞

former former former former
𝑆𝑟,𝑟𝑒𝑣 𝑆𝑟 1.0 0.0 𝑤𝑒𝑡

𝑆𝑟,𝑝𝑟 𝑆𝑟 1.0 0.0 𝑑𝑟
𝑆𝑟,𝑝𝑟
former former
𝑠𝑒𝑞 𝑠𝑒𝑞
𝑠𝑐𝑜𝑚𝑚𝑜𝑛 𝑠𝐵𝑤𝑒𝑡 0.0 𝑠0,𝑠𝑒𝑞 𝑠𝐵𝑑𝑟 0.0 𝑠0,𝑠𝑒𝑞
former former
𝑟 𝑟𝑤𝑒𝑡 0.0 0.0 0.0 𝑟𝑑𝑟 0.0 0.0 0.0
Dire-
ction
-1.0 -1.0 -1.0 -1.0 1.0 1.0 1.0 1.0
280
5.3.2.2 Updating of the reversal parameters
One drying and one wetting scanning path correspond to every reversal point
and remain unaffected provided that the direction of hydraulic loading remains
unchanged. Once an increment of suction occurs, the direction of hydraulic
loading is identified; positive change of suction signifies drying, while wetting is
detected in the opposite case.
Detecting wetting while the direction of hydraulic loading is equal to 1.0 (drying),
indicates that a reversal has occurred and the direction is reset to -1.0 (wetting).
The degree of saturation, 𝑆𝑟 , is evaluated, based on the thus far unchanged
reversal parameters, for the suction at the end of the previous increment and
the two form the co-ordinates of the new reversal point, 𝑠𝑟𝑒𝑣 and 𝑆𝑟,𝑟𝑒𝑣 . Only
then is the system of Equations 5.12 and 5.13 solved, employing these co-
ordinates, and the reversal parameters are updated to 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 = 𝑠𝐵𝑤𝑒𝑡 and
𝑟 = 𝑟𝑤𝑒𝑡 (Table 5-6).
The process of solving the above mentioned system of equations is avoided

when full saturation or residual conditions apply and the reversal parameters
are set to the values presented in Table 5-6. Furthermore, if the newly
evaluated reversal point lies close to the primary wetting path (tolerance of 0.02
on the degree of saturation), the latter is followed on subsequent wetting and
the corresponding reversal parameters are also presented in Table 5-6.
A similar process is followed if drying is detected to occur while the direction of

hydraulic loading is -1.0 (wetting). The direction is reset to 1.0 and the degree of
saturation for the previous suction level is evaluated. The co-ordinates of the
new reversal point are then updated and the system of Equations 5.8 and 5.9 is
solved, in order to calculate 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 = 𝑠𝐵𝑑𝑟 and 𝑟 = 𝑟𝑑𝑟 (Table 5-6). If full
saturation or residual conditions apply the solution of the system is not needed
as the reversal parameters are set to the values presented in Table 5-6. Finally,
if the reversal point is found to lie close to the primary drying path (tolerance of
0.02 on the degree of saturation), the latter is adopted.
The occurrence of reversals in the direction of hydraulic loading is checked at

the beginning of every increment for which a change in suction is detected. The
281
reversal parameters are stored and are employed in the calculation of the
degree of saturation, 𝑆𝑟 , corresponding to all the subsequent suction levels, as
long as the direction of hydraulic loading remains unchanged.
5.3.2.3 Solution of the system
As already discussed, the system of Equations 5.8 and 5.9 or 5.12 and 5.13
needs to be solved every time a reversal is detected in the direction of hydraulic
loading in order to compute the reversal parameters 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 and 𝑟. However,
solution of the system is not straightforward and requires a numerical approach.
Newton’s method was chosen and proved to be adequate and efficient. A
limited number of iterations was generally required for convergence to be
achieved (generally less than 10 iterations were sufficient).
To avoid the numerical instabilities associated with the presence of the square
root, an equivalent system of equations was solved where both sides of the
equations were squared. So, instead of the system of Equations 5.8 and 5.9,
the equivalent system presented in Table 5-7 is actually solved when drying is
detected, while instead of the system of Equations 5.12 and 5.13 for wetting,
the equations presented in Table 5-8 are solved. Note that for the case of
wetting the system is solved in terms of log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 and subsequently the
suction 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 is calculated.
Table 5-7: System of equations to be solved on drying
System of Equations; drying
2
1
𝐹1 = 1− 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − 𝑆𝑟 ,𝑟𝑒𝑣 − 𝑟 1 + 𝛼𝑑 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 −
𝑠0,𝑒𝑞
(5.14)
− 𝑟 2 − log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − log 𝑠𝑟𝑒𝑣 2
∙ 1 + 𝛼𝑑 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 2
= 0.0
2
2 2 2
1
𝐹2 = 𝑟 − log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − log 𝑠𝑟𝑒𝑣 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ ln 10 + 𝛼𝑑 −
𝑠0,𝑒𝑞
(5.15)
2 4
− log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − log 𝑠𝑟𝑒𝑣 ∙ 1 + 𝛼𝑑 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 = 0.0
282
Table 5-8: System of equations to be solved on wetting
System of Equations; wetting
2
1 𝑥 𝑥
𝐹3 = 1− 10 − 𝑆𝑟,𝑟𝑒𝑣 + 𝑟 1 + 𝛼𝑤 ∙ 10 −
𝑠0,𝑒𝑞
(5.16)
− 𝑟 2 − log 𝑠𝑟𝑒𝑣 − 𝑥 2
∙ 1 + 𝛼𝑤 ∙ 10𝑥 2
= 0.0
2
2 2 𝑥 2
1
𝐹4 = 𝑟 − log 𝑠𝑟𝑒𝑣 − 𝑥 ∙ 10 ∙ ln 10 ∙ + 𝛼𝑤 −
𝑠0,𝑒𝑞
2
− log 𝑠𝑟𝑒𝑣 − 𝑥 ∙ 1 + 𝛼𝑤 ∙ 10𝑥 4
= 0.0 (5.17)
𝑥 = log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛
According to Newton’s method (Press et al., 2007) the solution of a system of

two non-linear equations may be determined iteratively by:
𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ,𝑖+1 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ,𝑖 𝐹1 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ,𝑖 , 𝑟𝑖

𝑟𝑖+1 = 𝑟𝑖 − 𝐽𝑖−1 ∙ (5.18)
𝐹2 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ,𝑖 , 𝑟𝑖
where 𝐽𝑖 is the Jacobian matrix:
𝜕𝐹1 𝜕𝐹1
𝜕𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ,𝑖 𝜕𝑟𝑖
𝐽𝑖 = (5.19)
𝜕𝐹2 𝜕𝐹2
𝜕𝑠𝑐𝑜𝑚𝑚 𝑜𝑛 ,𝑖 𝜕𝑟𝑖
and its inverse can be calculated as:
𝜕𝐹2 𝜕𝐹1
−
1 𝜕𝑟𝑖 𝜕𝑟𝑖
𝐽𝑖−1 = (5.20)
𝐽𝑖 𝜕𝐹2 𝜕𝐹1
−
𝜕𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ,𝑖 𝜕𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ,𝑖
𝐽𝑖 being the Jacobian determinant:
𝜕𝐹1 𝜕𝐹2 𝜕𝐹1 𝜕𝐹2

𝐽𝑖 = ∙ − ∙ (5.21)
𝜕𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ,𝑖 𝜕𝑟𝑖 𝜕𝑟𝑖 𝜕𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ,𝑖
283
The derivatives of Equations 5.14 and 5.15 with respect to 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 and 𝑟 are
presented in Table 5-9.
Table 5-9: Derivatives of Equations 5.14 and 5.15 with respect to 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 and 𝑟; drying
Derivatives for drying
𝜕𝐹1 1
= −2 1− 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − 𝑆𝑟,𝑟𝑒𝑣 − 𝑟 1 + 𝛼𝑑 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙
𝜕𝑠𝑐𝑜𝑚𝑚𝑜𝑛 𝑠0,𝑒𝑞
1
∙ + 𝑆𝑟,𝑟𝑒𝑣 − 𝑟 ∙ 𝛼𝑑 −
𝑠0,𝑒𝑞
(5.22)
2 2
− ∙ log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − log 𝑠𝑟𝑒𝑣 ∙ 1 + 𝛼𝑑 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 −
𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ln 10
− 2𝛼𝑑 ∙ 1 + 𝛼𝑑 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ 𝑟 2 − log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − log 𝑠𝑟𝑒𝑣 2
𝜕𝐹1 1
= 2∙ 1− 𝑠 ∙ 1 + 𝛼𝑑 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 −
𝜕𝑟 𝑠0,𝑒𝑞 𝑐𝑜𝑚𝑚𝑜𝑛
(5.23)
2 2
− 2 𝑆𝑟,𝑟𝑒𝑣 − 𝑟 ∙ 1 + 𝛼𝑑 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − 2𝑟 ∙ 1 + 𝛼𝑑 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛
2
𝜕𝐹2 1
= −2𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ ln 10 ∙ log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − log 𝑠𝑟𝑒𝑣 ∙ + 𝛼𝑑 +
𝜕𝑠𝑐𝑜𝑚𝑚𝑜𝑛 𝑠0,𝑒𝑞
2
2
1
+ 2𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ ln 10 ∙ + 𝛼𝑑 ∙ 𝑟 2 − log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − log 𝑠𝑟𝑒𝑣 2
−
𝑠0,𝑒𝑞
(5.24)
2 4
− ∙ log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − log 𝑠𝑟𝑒𝑣 ∙ 1 + 𝛼𝑑 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 −
𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ ln 10
2 3
− 4𝛼𝑑 ∙ log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − log 𝑠𝑟𝑒𝑣 ∙ 1 + 𝛼𝑑 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛
2
𝜕𝐹2 2
1
= 2𝑟 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ ln 10 ∙ + 𝛼𝑑 (5.25)
𝜕𝑟 𝑠0,𝑒𝑞
284
The derivatives of Equations 5.16 and 5.17 with respect to log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 and 𝑟 are
presented in Table 5-10.
Table 5-10: Derivatives of Equations 5.16 and 5.17 with respect to 𝑙𝑜𝑔 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 and 𝑟; wetting
Derivatives for wetting
𝜕𝐹3 𝑠𝑐𝑜𝑚𝑚𝑜𝑛
=− 1− − 𝑆𝑟,𝑟𝑒𝑣 + 𝑟 ∙ 1 + 𝛼𝑤 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙
𝜕 log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 𝑠0,𝑒𝑞
1
∙ 2𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ ln 10 ∙ + 𝑆𝑟,𝑟𝑒𝑣 + 𝑟 𝛼𝑤 −
𝑠0,𝑒𝑞
(5.26)
2
− 2 log 𝑠𝑟𝑒𝑣 − log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ 1 + 𝛼𝑤 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 −
− 2𝛼𝑤 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ ln 10 ∙ 𝑟 2 − log 𝑠𝑟𝑒𝑣 − log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 2

∙ 1 + 𝛼𝑤 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 2
𝜕𝐹3 1
= −2 ∙ 1 − 𝑠 ∙ 1 + 𝛼𝑤 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 +
𝜕𝑟 𝑠0,𝑒𝑞 𝑐𝑜𝑚𝑚𝑜𝑛
(5.27)
2 2
+ 2 𝑆𝑟,𝑟𝑒𝑣 + 𝑟 ∙ 1 + 𝛼𝑤 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 − 2𝑟 ∙ 1 + 𝛼𝑤 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛
𝜕𝐹4
=
𝜕 log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛
2
2
1
= 2 ∙ log 𝑠𝑟𝑒𝑣 − log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ ln 10 ∙ + 𝛼𝑤 +
𝑠0,𝑒𝑞
2
2
1
+ 2 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ ln 10 ln 10 + 𝛼𝑤 𝑟 2 − log 𝑠𝑟𝑒𝑣 − log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 2
+ (5.28)
𝑠0,𝑒𝑞
4
+ 2 log 𝑠𝑟𝑒𝑣 − log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ 1 + 𝛼𝑤 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 −
2 3
− 4𝛼𝑤 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ ln 10 ∙ log 𝑠𝑟𝑒𝑣 − log 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ 1 + 𝛼𝑤 𝑠𝑐𝑜𝑚𝑚𝑜𝑛
2
𝜕𝐹4 2
1
= 2𝑟 ∙ 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 ∙ ln 10 ∙ + 𝛼𝑤 (5.29)
𝜕𝑟 𝑠0,𝑒𝑞
285
At the end of every iteration, the degree of saturation on the primary path, 𝑆𝑟,𝑝𝑟 ,
and on the respective scanning path, 𝑆𝑟,𝑠𝑐𝑎𝑛 , are calculated based on the
current solution. When the difference between the two becomes less than a
tolerance, which for the results presented in this thesis has been set equal to
0.01% of the mean value of the two quantities, the iterative process is
terminated and the last solution of the system is adopted.
However, the above systems can have two possible solutions, as pictured in
Figure 5-18 for drying. Therefore, if the degree of saturation calculated is larger
(or smaller for wetting) than the degree of saturation at the reversal point A (i.e.
the code has computed a negative radius, 𝑟) the small circle in Figure 5-18 is
obtained and consequently the solution is rejected and an error message is
produced (such an error did not occur in any of the analyses performed here – if
such an error occurs in the future the algorithm will have to be altered so as to
ignore the wrong root and continue the iterative process until the correct root is
found).
The system needs to be resolved only once for every reversal point and only for
the applied direction of hydraulic loading since the information regarding the
point of intersection are stored (reversal parameters 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 and 𝑟).
Wrong
C solution
A Correct
Bdr solution
0
sair sA sBdr s0
Figure 5-18: Possible solutions of the system of equations when drying from an initial retention point A
286
5.3.2.4 Calculation of the degree of saturation and of the gradient of the

SWRC
Having stored the reversal parameters, the calculation of the degree of

saturation, 𝑆𝑟 , as well as of the gradient of the SWRC, 𝑅, corresponding to the
current value of suction, is straightforward.
If the direction of hydraulic loading is 1.0, drying occurs and distinction in the
employment of the primary or the scanning drying path (Equations 5.4 and 5.6,
respectively) is based on the comparison of the current suction, 𝑠𝑒𝑞 , with the
reversal parameter 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 . For suction levels higher than 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 , the primary
drying path is employed, otherwise the scanning drying path is adopted.
If wetting is detected (the direction of hydraulic loading is -1.0), the scanning

wetting path (Equation 5.10) is employed for suction levels higher than the
reversal parameter 𝑠𝑐𝑜𝑚𝑚𝑜𝑛 . On the contrary, the primary wetting path (Equation
5.4) is adopted for lower suction levels.
If a scanning curve is employed to calculate the degree of saturation, 𝑆𝑟 , once

its value has been computed, it is checked against the limiting values defined by
the primary drying and wetting curves for the same suction and the program is
terminated if a violation of those limits occurs (such violation never occurred
during this research – violating the limits set by the primary curves would
indicate that the system of equations was not correctly solved and the solution
technique would have to be revised).
Despite the simplicity in the formulation of the proposed model for the hysteretic
behaviour exhibited by the SWRC, its implementation into a finite element code
is relatively demanding as the algorithm has to keep track of the reversal points
where the path changes from drying to wetting and vice versa. Furthermore,
solution of the system of the non-linear equations presented above is not
straightforward and a numerical approach is necessary. Therefore, the validity
of the implementation, subsequently discussed, needs to be thoroughly tested.
287
5.3.3 Validation
The hysteresis produced by ICFEP, under various changes in applied suction,

was compared to a solution computed using Microsoft Excel. Four tests were
employed in the process and the numerical results showed exact agreement
with the Excel solution. The model parameters employed in the tests are
presented in Table 5-11 and are representative of those expected to be
produced by the calibration of a typical SWRC.
The first test consisted of the generation of the primary paths and of three
independent scanning drying and wetting paths. For the primary paths to be
obtained, a single element was subjected to incremental changes of suction, s,
from an initial value of 0.0 kPa to 100000.0 kPa and subsequently back to 0.0
kPa. Furthermore, starting from point A ( s = 1037.6 kPa and Sr = 0.382),
scanning drying and scanning wetting paths were simulated, as well as a drying
scanning path from point B (s = 15.40 kPa and Sr = 0.854). The initial and final
suction values applied are shown in Table 5-12. The hysteretic-SWRC model
was used for the computation of the respective degree of saturation, 𝑆𝑟 , and the
results illustrated in Figure 5-19 in terms of suction, 𝑠, rather than equivalent
suction, 𝑠𝑒𝑞 , were identical to those obtained employing Excel.
The second test consisted of the cyclic change of suction, s, explained in Table
5-12. The suction, 𝑠, was increased from its initial value of 150.0 kPa to 15000.0
kPa and reduced back to 170.0 kPa, before being increased to 30000.0 kPa.
The scanning path produced by the hysteretic- SWRC model is shown in Figure
5-20 to be equal to the one calculated by the Excel spreadsheet.
Finally, for Tests 3 and 4 the cycles of suction explained in Table 5-12 were
applied assuming the same initial retention point (also presented in Table 5-12),
but a different primary drying path. The scanning paths reproduced by the
hysteretic-SWRC model are illustrated in Figure 5-21 for Test 3 and in Figure 5-
22 for Test 4, and are in excellent agreement with the Excel solution.
The above four tests demonstrate the accuracy of the model in solving the
system of equations, explained in Section 5.3.2.2, and in applying it in the
calculation of the degree of saturation, 𝑆𝑟 .
288
Table 5-11: Parameters employed in the validation of the hysteretic-SWRC model

Validation of the hysteretic-SWRC model – model parameters
Test 1 Test 2 Test 3 Test 4
Param. Value Value Value Value
𝑠𝑎𝑖𝑟 1.0 kPa 0.0 kPa 0.0 kPa 0.0 kPa
𝑠0 1.0E+5 kPa 1.0E+7 kPa 1.0E+4 kPa 1.0E+4 kPa
𝛼𝑑 5.0E-4 3.8E-5 5.0E-3 5.0E-4
𝛼𝑤 2.8E-2 3.5E-3 2.0E-2 2.0E-2
Table 5-12: Changes in suction applied for the validation of the hysteretic-SWRC
Validation of the hysteretic-SWRC model – hydraulic loading
Parameter Test 1
𝑠1 𝑆𝑟 1037.6 0.382 15.40 0.854
Drying to suction 10000.0 kPa 10000.0 kPa
Wetting to suction 1.0 kPa -
Cyclic loading Test 2 Tests 3 and 4
Parameter Value Value
𝑠1 𝑆𝑟 150.0 0.75 10.0 0.93
𝑠2 (kPa) 15000.0 1000.0
𝑠3 (kPa) 170.0 1.0
𝑠4 (kPa) 30000.0 51.0
𝑠5 (kPa) - 1.0
𝑠6 (kPa) - 251.0
𝑠7 (kPa) - 51.0
𝑠8 (kPa) - 401.0
289
1.0
B
0.8
0.6
0.4
A
0.2
0.0
1 10 100 1000 10000 100000
Wetting from point A (Excel)
ICFEP
Drying from point A (Excel)
Primary paths (Excel)
Drying from point B (Excel)
Figure 5-19: ICFEP and Excel calculation of the degree of saturation, 𝑆𝑟 , generated during primary
drying and wetting, as well as on drying and wetting from random initial retention points, A and B; Test 1
1.0
0.8 A
C
0.6
B
0.4
D
0.2
0.0
100 1000 10000 100000
Drying A-B (Excel)
ICFEP
Wetting B-C (Excel)
Primary paths (Excel)
Drying C-D (Excel)
Figure 5-20: ICFEP and Excel calculation of the degree of saturation, 𝑆𝑟 , generated during cyclic
changes of suction; Test 2
290
Degree of saturation, Sr 1.0 A
0.8
0.6
0.4
Primary paths
ICFEP
0.2
Excel
0.0
1 10 100 1000 10000
1.0 A
0.8
0.6
0.4
Primary paths
ICFEP
0.2
Excel
0.0
1 10 100 1000 10000
5.3.4 Calibration
The hysteretic model for the SWRC presented above, was employed in the
numerical simulation of laboratory tests carried out at Imperial College London
by Melgarejo (2004) and Jotisankasa (2005) on three different soil types:
London clay, artificial soil A and Weald clay.
The first series of tests were performed on intact and reconstituted specimens
of London clay fill (Melgarejo, 2004) and the laboratory data are presented in
291
Figure 5-23. Two reconstituted samples, formed from a slurry, were prepared
from the fill material supplied and were subjected to unconfined drying in order
to obtain the primary drying path. However, the maximum suction that could be
measured was reached before the residual degree of saturation was achieved
(Figure 5-23 (a)) and it was, therefore, impossible to develop the full SWRC.
One of the samples was subsequently wetted forming the scanning wetting path
illustrated in Figure 5-23 (a). Adopting the parameters summarised in Table
5-13, the primary paths shown in the same figure were numerically reproduced.
It should be noted that the parameter 𝛼𝑤 , controlling the primary wetting path,
was fitted in such a way that the scanning wetting path, followed by the
reconstituted sample upon wetting, converged to the primary path at lower
values of suction.
Additionally to the reconstituted material, two intact samples of London clay fill,
with significantly different initial retention states, were tested. Starting from an
initial degree of satuartion of 88% (𝑆𝑟,𝐴 = 0.88) at 440.0 kPa of suction (𝑠𝐴 =
440.0 kPa), the first sample (point A in Figure 5-23 (a)) was dried to 56%
degree of saturation and reached a suction of 18500.0 kPa. The path was
numerically simulated adopting the hysteretic-SWRC model and is shown in
comparison to the one measured in the laboratory in Figure 5-23 (a). The model
predicted the experimental data up to 1100.0 kPa of suction. Thereafter, the
experimental data followed a large plateau which was not reproduced by the
model. Despite the consequent underestimation of the degree of saturation, 𝑆𝑟 ,
the simulated path ended up coinciding with the one observed in the laboratory
for suctions larger than 11000.0 kPa.
The second intact sample was dried from its initial retention state (point A in
Figure 5-23 (b)), with 𝑠𝐴 = 150 kPa and 𝑆𝑟,𝐴 = 0.75, to point B (𝑠𝐵 = 1500 kPa,
𝑆𝑟,𝐵 = 0.46). Subsequently, the sample was wetted to point C (𝑠𝐶 = 170 kPa, 𝑆𝑟,𝐶
= 0.74) and re-dried to point D (𝑠𝐷 = 22500 kPa, 𝑆𝑟,𝐷 = 0.28). The numerically
computed scanning paths are shown in Figure 5-23 (b). The first scanning
drying path (AB) was accurately reproduced. The subsequent wetting scanning
path (B-C) was also adequately predicted for suction levels in the excess of
1000.0 kPa, but the degree of saturation, 𝑆𝑟 , was underestimated for lower
292
values of suction. As a result, the subsequent drying path (C-D) was simulated
to initiate from a point lying to the left of the actual reversal point Cactual, so that
the two have a common degree of saturation. Nevertheless, the reproduced
scanning drying path followed closely the experimentally observed behaviour for
values of suction up to 10000.0 kPa, overestimating the degree of saturation
thereafter.
The simulation of the above laboratory tests is considered to be adequate,

indicating that the performance of the hysteretic-SWRC model is efficient,
despite its simplicity in terms of the shape of the curves employed. The primary
paths were effectively reproduced by the model, however, the air-entry value of
suction, 𝑠𝑎𝑖𝑟 , assumed was 0.0 kPa in order to replicate the wetting paths.
Melgarejo (2004) reported that on drying, the reconstituted samples remained
fully saturated to values of suction of the order of 1000.0 kPa. Adopting one
fitting parameter for each primary curve and assuming that full saturation upon
wetting occurs at the same value of suction as de-saturation upon drying, is
clearly a shortcoming of the model.
The second series of laboratory experiments considered for the calibration of

the hysteretic-SWRC model, were performed by Jotisankasa (2005) on the
artificial soil A, already employed in the calibration of the v-SWRC model.
Starting from the as-compacted state, shown in Figure 5-24 (retention point A),
with initial specific volume, 𝑣, of 1.70, and degree of saturation, 𝑆𝑟 , 38%, a
sample was wetted slowly to 1 kPa of suction and was subsequently re-dried to
an air-dried state, which is assumed in the analysis to reproduce the primary
drying curve (Figure 5-24). Another sample was dried from the same initial state
while a third one was dried from the retention point B, also shown in Figure 5-
24. Both scanning drying paths appear to rejoin the primary one at large values
of suction.
The hysteretic-SWRC model was fitted to the experimental data yielding the
parameters summarised in Table 5-13 and the corresponding primary paths are
illustrated in Figure 5-24. For the simulation of the primary wetting path the
assumption was made that the scanning wetting path initiated from point A,
should rejoin the former one at low values of suction.
293
Even though the predicted scanning wetting path from point A was positioned
slightly below the relevant experimental data, before converging to the primary
wetting path, it produced a reasonable estimation of the behaviour exhibited in
the laboratory. Nevertheless, the simulated scanning drying path from point A
rejoined the primary one at a suction significantly lower than the one indicated
experimentally, resulting in an overall overestimation of the degree of
saturation, 𝑆𝑟 . On the contrary, the scanning drying path from point B was
successfully reproduced, even though convergence to the primary path was
slightly delayed for the computed curve in comparison with the laboratory data.
From the above results it can be concluded that the retention behaviour was
successfully simulated. The ineffective prediction of the drying scanning path
initiated from point A, can be attributed to the inadequate reproduction of the
primary drying curve at high suction levels.
The third series of tests employed were carried out on reconstituted samples of
Weald clay by Melgarejo (2004) and the experimental data obtained are
illustrated in Figure 5-25 (a) and (b). Starting from full saturation (point A in
Figure 5-25 (a)), a sample was dried to 30000.0 kPa of suction (point B) and
back to 10 kPa, in order to obtain the primary curves. A second sample (Figure
5-25 (b)) was dried from full saturation to point B and wetted back to 700.0 kPa
of suction (point C). A scanning drying path was subsequently followed to point
D and from there the sample was wetted up to 2500.0 kPa of suction, where it
rejoined the previously observed in the laboratory wetting path (B-C).
Employing the parameters presented in Table 5-13 the primary paths illustrated
in Figure 5-25 (a) and (b) were computed. Similarly to the previous two cases
examined, the primary wetting path was obtained assuming convergence of the
wetting scanning curves observed in the laboratory to the primary path
reproduced. The degree of saturation, 𝑆𝑟 , measured for the first sample, at the
beginning of the drying and at the end of the wetting tests, was not equal to 1.0
and consequently the primary paths were not successfully reproduced at
suctions lower than 500.0 kPa (Figure 5-25 (a)), highlighting the limitations of
the basic shape adopted in the modelling.
294
The experimentally obtained scanning paths, shown in Figure 5-25 (b),

exhibited a double curvature, both upon wetting (path B-C) and drying (path C-
D). Although this shape was not numerically reproduced the model gave an
overall adequate prediction of the scanning paths. Finally, an accurate
estimation of the wetting path D-E followed in the laboratory was obtained.
The capability of the model to reproduce the scanning paths is generally

satisfying, despite the simple geometric shape assumed. As the scanning path
followed entirely depends on the primary path to which it converges, it is
expected that improvement of the expression used for the latter will also
improve the performance of the former.
Overall, it can be concluded that the model is capable of effectively reproducing

the hydraulic paths obtained in the laboratory for reconstituted and intact
samples of London clay fill and of compacted artificial soil A. It was proven,
however, less successful in the reproduction of scanning paths that exhibit
double curvature, such as those observed for reconstituted Weald clay.
Table 5-13: Model parameters produced by the calibration of the hysteretic-SWRC model for three
different types of soil; London clay fill, artificial soil A and Weald clay
Hysteretic-SWRC model
London clay fill Soil A (Jotisankasa, Weald clay (Melgarejo,

(Melgarejo, 2004) 2005) 2004)
𝑠𝑎𝑖𝑟 0.0 kPa 𝑠𝑎𝑖𝑟 1.0 kPa 𝑠𝑎𝑖𝑟 0.0 kPa
𝑠0 1.0E+7 kPa 𝑠0 1.0E+5 kPa 𝑠0 1.0E+6 kPa
𝛼𝑑 3.8E-5 𝛼𝑑 5.0E-4 𝛼𝑑 1.0E-4
𝛼𝑤 3.5E-3 𝛼𝑤 2.8E-2 𝛼𝑤 5.0E-4
295
1.0
(a)
A
0.8
0.6
0.4
Primary paths
0.2 Scanning paths
Drying (lab.)
Wetting (lab.)
Drying from A (lab.)
0.0
1 10 100 1000 10000 100000
1.0
(b)
Cactual
Pr
0.8 A
im
ary
dry
C
ing
pa
th
0.6
0.4
Pr
im
ary
D
we
0.2
tt
ing
path
0.0
100 1000 10000 100000
Drying A-B (lab.) Drying A-B (model)
Wetting B-C (lab.) Wetting B-C (model)
Drying C-D (lab.) Drying C-D (model)
Figure 5-23: Hydraulic paths followed by specimens of London clay fill (data after Melgarejo, 2004); (a)
numerical reproduction of the primary paths; (b) numerical reproduction of a cyclic scanning hydraulic
path
296
1.0
B
0.8
0.6
Wetting from initial point A (lab.)
0.4 Drying from full saturation (lab.)
Drying from initial point A (lab.)
A
Drying from point B (lab.)
0.2
Primary paths (ICFEP)
Scanning paths (ICFEP)
0.0
0.1 1 10 100 1000 10000 100000
Figure 5-24: Hydraulic paths followed by compacted specimens of artificial soil A (data after
Jotisankasa, 2005)and their numerical reproduction
A
1.0
0.8
0.6
0.4
B
0.2 Primary paths (ICFEP)
Drying and wetting, Sample A (lab.)
(a)
0.0
10 100 1000 10000 100000 1000000
A Suction, log s (kPa)
1.0
0.8
C
0.6 Primary drying A-B (lab.)
E
First wetting B-C (lab.) D
0.4 Second drying C-D (lab.)
Second wetting D-E (lab.) B
0.2 Primary paths (ICFEP)
(b)
0.0
10 100 1000 10000 100000 1000000
Figure 5-25: Hydraulic paths followed by specimens of reconstituted Weald clay (data after Melgarejo,
2004); (a) numerical reproduction of the primary paths; (b) numerical reproduction of a cyclic hydraulic
path
297
5.4 Specific volume dependent, hysteretic Soil-Water Retention Curve

(v-hysteretic-SWRC)
Even though the effect of void ratio on the SWRC and the hysteresis it exhibits
were considered separately in the previous two sections, it is likely that they
occur simultaneously, producing a hysteretic 3-dimensional surface in the 𝑠 - 𝑆𝑟
- 𝑣 space. The extension of the hysteretic model presented above to include the
effect of void ratio is studied in the current section and the corresponding model
is herein referred to as v-hysteretic-SWRC model.
5.4.1 Formulation and Implementation
The hysteretic model introduced in Section 5.3 – Hysteretic Soil-Water

Retention Curve (hysteretic-SWRC) – was employed as the base for the
formulation of the v-hysteretic-SWRC, adopting the same assumptions and a
similar shape for the primary and scanning drying and wetting paths, which
were altered to:
1 ∗
1−𝑠 ∙ 𝑠𝑒𝑞
𝑆𝑟,𝑝𝑟 =
0,𝑒𝑞 (5.30)
1 + 𝛼𝑑 ,𝑤 ∙ ∗
𝑠𝑒𝑞
and:
2 2
𝑑𝑟
𝑆𝑟,𝑠𝑐𝑎𝑛 = 𝑆𝑟,𝐴 − 𝑟𝑑𝑟 + 𝑟𝑑𝑟 ∗ − log 𝑠 ∗
− log 𝑠𝑒𝑞 (5.31)
𝐴
2 2
𝑤𝑒𝑡
𝑆𝑟,𝑠𝑐𝑎𝑛 = 𝑆𝑟,𝐴 + 𝑟𝑤𝑒𝑡 − 𝑟𝑤𝑒𝑡 − log 𝑠𝐴∗ − log 𝑠𝑒𝑞
∗ (5.32)
respectively, where
∗ 𝜓
𝑠𝑒𝑞 = 𝑣−1 ∙ 𝑠 − 𝑠𝑎𝑖𝑟 (5.33)
is a function of the currently applied suction and of the specific volume resulting
from the coupled effect of the mechanical and the hydraulic loading and is
herein termed the combined suction. 𝑠𝐴∗ is, therefore, the combined suction at
298
the initial point A and three co-ordinates are required for its determination; the
suction, 𝑠, the specific volume, 𝑣, and the degree of saturation, 𝑆𝑟 .
The primary drying and wetting surfaces are shown schematically in Figure 5-26
and are assumed to bound an infinite number of scanning surfaces, which, for
reasons of clarity, are not presented in the figure. The primary drying surface
(lightly coloured in the figure) lies above the primary wetting one (dark coloured)
for the whole range of specific volume values. As for the hysteretic-SWRC the
parameter controlling the shape of the drying surface, 𝛼𝑑 , needs necessarily to
be smaller than the one employed for wetting, 𝛼𝑤 .
The main advantage of the formulation presented above is the fact that the
SWRC surface becomes a two-dimensional hysteretic curve when plotted in the
∗
𝑠𝑒𝑞 − 𝑆𝑟 plane (Figure 5-27), similar to the hysteretic-SWRC presented in the
previous section. The scanning paths are assumed to be the arcs of circles in
this particular plane and the combined suction of the point of intersection with
the congruent primary path, 𝑠𝐵𝑑𝑟 ,∗ or 𝑠𝐵𝑤𝑒𝑡 ,∗ , needs to be evaluated together with
the corresponding radius, 𝑟𝑑𝑟 or 𝑟𝑤𝑒𝑡 . Furthermore, the suction 𝑠0,𝑒𝑞 ,
corresponding to the residual degree of saturation (assumed to be zero), also
varies with the specific volume and, therefore, the input parameter 𝑠0 refers to
the case where 𝑣 equals 2.0.
∗
Substituting 𝑠𝑒𝑞 for 𝑠𝑒𝑞 , the implementation of the new version of the model is
identical to that of the hysteretic-SWRC and is, therefore, omitted for brevity. It
is, however, interesting to note that a reversal could occur not solely due to
changes in suction but also due to changes in the specific volume. For example,
a reversal is detected when cyclic changes of confining stress are applied under
constant suction and the resulting effect on the SWRC may be mistaken for a
change in the direction of hydraulic loading. Nonetheless, the changes in the
specific volume are not expected to be large enough for the impact on the
simulated behaviour to be significant. This is further explored in Appendix B.
Due to the similarities with the hysteretic-SWRC, the validation and calibration
of the model are also presented in Appendix B.
299
Figure 5-26: 3-dimensional hysteretic SWRS in the 𝑠 - 𝑆𝑟 - 𝑣 space
rwet rwet
1
pr
im
scanning
Bwet
ar
wetting
yd
A
ry
in
pr
im
g
ar
y scanning
we Bdr
tti drying
n g
0
sBwet,* sA* sBdr,* s0*
rdr rdr
Combined suction, log s* (kPa)

∗
Figure 5-27: Projection in the 𝑠𝑒𝑞 - 𝑆𝑟 plane of the 3-dimensional SWRS
300
5.5 Degree of saturation dependent soil compressibility with suction
The constitutive models for unsaturated soils developed by Georgiadis (2003)

assume that changes in suction induce linear changes of the specific volume in
the 𝑣-ln 𝑠 plane. The corresponding coefficient of compressibility is 𝜅𝑠 for elastic
changes of suction and 𝜆𝑠 for elasto-plastic ones (Figure 3-8, Chapter 3).
Therefore, unless the secondary yield surface is violated, cyclic changes of
suction produce reversible volumetric strains.
Indeed, considering the tests 3 and 4 employed for the validation of the
hysteretic-SWRC (Section 5.3.3), the volumetric strains computed were
perfectly reversible, as illustrated in Figure 5-28, despite the cyclic changes in
suction applied (see Table 5-12). Moreover, the magnitude of the volumetric
strains was equal for the two tests even though they correspond to significantly
different retentions curves (Figure 5-21 for test 3 and Figure 5-22 for test 4).
This prediction is believed to be unrealistic, as it is well established by

laboratory experiments (see Chapter 2) that cyclic changes in suction are likely
to produce irreversible volumetric strains, even under constant confining stress.
Furthermore, it is reasonable to expect that soils following different SWRC’s
would also exhibit distinct volumetric behaviour. Therefore, assuming constant
coefficients of compressibility with suction, 𝜅𝑠 and 𝜆𝑠 , appears to be inconsistent
with the adoption of a hysteretic SWRC, when modelling the suction related
volumetric behaviour. Clearly, there is adequate experimental and theoretical
evidence indicating that 𝜅𝑠 varies with changes in suction.
Jotisankasa (2005) and Cunningham (2000) measured the specific volume of

compacted and reconstituted soil A, respectively, during drying and wetting.
Plotted in a semi-logarithmic scale, their results exhibit non-linearity, as
illustrated in Figure 5-29 (after Jotisankasa, 2005) and Figure 5-30 (after
Cunningham, 2000), while the changes in specific volume with suction were
reduced to 0.0 at large suction levels, at the excess of 1000.0 kPa, indicating
that the shrinkage limit was reached. Furthermore, the changes were
irreversible, demonstrating a hysteretic behaviour similar to the one observed in
the SWRC, illustrated in Figure 5-31 for the compacted material and in Figure 5-
32 for the reconstituted sample.
301
The above experimental results suggest that the coefficient of compressibility 𝜅𝑠

varies with the suction level as a function of the corresponding degree of
saturation. The following expression is thought to be indicative of this behaviour:
𝜅𝑠∗ = 𝜅𝑠 𝜒 ∙ 𝑆𝑟 𝜔
(5.34)
where
 𝜅𝑠∗ is the degree of saturation dependent coefficient of compressibility

with suction and applies solely to unsaturated states;
 𝑆𝑟 is the degree of saturation calculated for the current suction from the
SWRC model, which may be either of the models presented in this
chapter;
 𝜒 and 𝜔 are fitting parameters ( 𝜔 ≥ 0.0). For 𝜔 = 0.0 and 𝜒 = 1.0 , a

constant coefficient of compressibility equal to 𝜅𝑠 is obtained.
Equation 5.34 was adopted in the numerical prediction of the specific volume
measured in the laboratory for the tests presented above, in combination with
the v-hysteretic-SWRC model, which is thought to be the most advanced from
the models present in the current chapter. The parameters employed for the
simulation of the tests by Jotisankasa (2005) are shown in Table 5-14 and the
numerical results are illustrated in Figure 5-33. The computed specific volume is
shown to be in good agreement with the laboratory data, increasing upon
wetting from point A and subsequently decreasing upon re-drying from full
saturation but following a distinct path. A third path was predicted for drying
from point B representing the experimentally observed behaviour accurately. All
paths reproduced correctly exhibited limited change of specific volume at
suctions higher than 1000.0 kPa, whereas linear, constantly increasing 𝑣 would
have been predicted, had a constant elastic coefficient of compressibility been
adopted.
For the prediction of the specific volume, 𝑣, observed by Cunningham (2000) on

reconstituted soil A, the primary paths were reproduced, employing the model
parameters presented in Table 5-15. The retention curve computed was a crude
302
estimate of the one measured in the laboratory (Figure 5-34), as de-saturation

upon primary drying and full saturation upon subsequent wetting were falsely
modelled to occur at the same value of suction, according to one of the main
assumptions for the formulation of the model. As a result, the primary wetting
path highly over-predicted the degree of saturation at low suction levels, lying
close to the primary drying curve and exhibiting insignificant hysteresis.
Consequently, the hysteresis in the variation of the specific volume, 𝑣, with
suction was not accurately predicted (Figure 5-34). Nevertheless, the non-
linearity was depicted and more importantly limited changes in volume were
computed for suctions larger than 1000.0 kPa.
Additionally to the laboratory experiments by Jotisankasa (2005) and

Cunningham (2000), a wetting-drying cycle performed by Sharma (1998) and
employed by Wheeler et al. (2003) in the theoretical justification of the model
they proposed, was also simulated. A compacted bentonite-kaolin sample,
tested in a triaxial cell under isotropic mean net stress of 10.0 kPa, was
subjected to a cyclic change of suction from 300.0 kPa to 20.0 kPa and back.
Consistent with the hydraulic hysteresis shown in Figure 5-35 (a), swelling was
observed during wetting (path A-B) followed by greater compression upon
drying (path B-C), as illustrated in Figure 5-35 (b). Employing the parameters in
Table 5-16, the above behaviour was closely reproduced by the new
capabilities of ICFEP.
Improving the expression for the primary paths is expected to improve the
prediction of the specific volume, 𝑣. The idea is further explored in Appendix B,
adopting the improvements suggested for the v-hysteretic-SWRC model in the
subsequent section.
According to the approach proposed herein, the irreversibility of volumetric

strains due to cycles of suction under constant loading, reported in the
laboratory, could be due to the hysteresis of compressibility with suction. In this
sense, they are considered to be elastic irreversible volumetric strains rather
than plastic strains, as assumed by Wheeler et al. (2003).
Based on the idea that the SWRC is a three-dimensional hysteretic curve in the
𝑠 − 𝑆𝑟 − 𝑣 space, it would be appropriate to assume that compressibility with
303
suction, 𝜅𝑠 , is a function of the degree of saturation, 𝑆𝑟 . Plotting a monotonic

SWRC in the 𝑠 − 𝑆𝑟 − 𝑣 space, as illustrated in Figure 5-36, and projecting it in
the three planes 𝑠 − 𝑆𝑟 , 𝑠 − 𝑣, and 𝑣 − 𝑆𝑟 , as shown in Figure 5-37 (a), (b) and
(c), respectively, the relationship between the void ratio and the degree of
saturation becomes evident.
The projection in the 𝑠 − 𝑣 plane, shown in Figure 5-37 (b), displays a shape
similar to that obtained by Equation 5.34 for soil A, presented above, providing
a theoretical justification for its formulation and implementation into a finite
element code. For fully saturated states, where the soil behaviour is described
by the effective stress principle, the variation of the specific volume, 𝑣, with
suction is constant and controlled by the coefficient 𝜅, whereas past the air-
entry value of suction, 𝑠𝑎𝑖𝑟 , this variation becomes non-linear and the specific
volume finally reaches a limiting value, 𝑣𝑙𝑖𝑚 , which is believed to represent the
shrinkage limit.
A similar trend can be seen in Figure 5-37 (c), which illustrates the projection of
the three-dimensional SWRC in the 𝑣 − 𝑆𝑟 plane. The degree of saturation, 𝑆𝑟 ,
remains constant and equal to 1.0 for fully saturated states despite the
volumetric changes induced by the suction changes. Upon de-saturation, a non-
linear relationship is predicted between the degree of saturation, 𝑆𝑟 , and the
specific volume, 𝑣, which reaches a limiting value, 𝑣𝑙𝑖𝑚 , as 𝑆𝑟 tends to 0.0.
The non-linear, degree of saturation dependent expression suggested for the

variation of the elastic coefficient of compressibility with suction, 𝜅𝑠∗ , is the same
for drying and for wetting and the observed hysteresis is a result of the
hysteresis in the SWRC. If the latter is monotonic the 𝑣 − 𝑠 relationship
predicted is also monotonic but remains non-linear. The main advantage of the
proposed approach is that the irreversibility of volumetric strains due to cyclic
changes of suction is explicitly related to the hydraulic hysteresis and does not
require the implementation of a third yield surface, termed the suction decrease
(Wheeler et al., 2003), in the isotropic stress plane 𝑠 − 𝑝.
304
Table 5-14: Model parameters employed for the prediction of the variation of the specific volume with
suction, measured by Jotisankasa (2005)on compacted soil A
Model parameters for compacted soil A
Hysteretic- SWRC 𝑆𝑟 dependent 𝜅𝑠∗
Parameters Value Parameters Value
𝑠𝑎𝑖𝑟 1.0 kPa 𝜒 0.411
𝑠0 1.0E+5 kPa 𝜔 4.000
𝛼𝑑 0.0011 𝜅𝑠 0.020
𝛼𝑤 0.045 - -
𝜓 0.75 - -
suction, measured by Cunningham (2000) on reconstituted soil A
Model parameters for reconstituted soil A
𝑠𝑎𝑖𝑟 300.0 kPa 𝜒 0.958
𝑠0 1.0E+5 kPa 𝜔 4.000
𝛼𝑑 0.004 𝜅𝑠 0.035
𝛼𝑤 0.0075 - -
𝜓 2.0 - -
305
suction, measured by Sharma (1998) on compacted bentonite-kaolin
Model parameters for compacted bentonite-kaolin
𝑠𝑎𝑖𝑟 0.0 kPa 𝜒 0.777
𝑠0 1.0E+5 kPa 𝜔 4.750
𝛼𝑑 5.0E-5 𝜅𝑠 0.150
𝛼𝑤 5.0E-3 - -
𝜓 0.50 - -
3.0
2.5
2.0
Cycle 1 (Test 3)
1.5 Cycle 2 (Test4)
1.0
1 10 100 1000
-10.0

Volumetric strains, vol (%)
-5.0
1 10 100 1000
0.0
5.0
10.0
Cycle 1 (Test 3)
15.0
Cycle 2 (Test4)
20.0
25.0
Figure 5-28: (a) Specific volume change and (b)volumetric strains 𝜀𝑣𝑜𝑙 evaluated during the cyclic
changes of suction applied for Tests 3 and 4 (Figures 5.21 and 5.22, respectively)
306
Figure 5-29: Changes in void ratio measured during unconfined drying and wetting of compacted Soil A
(after Jotisankasa, 2005)
Figure 5-30: Changes in specific volume measured during unconfined drying and wetting of reconstituted
Soil A (after Cunningham, 2000)
307
Figure 5-31: SWRC measured in the laboratory for compacted Soil A (after Jotisankasa, 2005)
Figure 5-32: SWRC measured in the laboratory for reconstituted Soil A (after Cunningham, 2000)
308
1.0
B
(a)
0.8
0.6
0.4
0.2
0.0
0.1 1 10 100 1000 10000 100000
1.9
(b)
1.8
A
1.7
B
1.6
1.5
1.4
0.1 1 10 100 1000 10000 100000
Wetting from initial point A (lab.) Drying from full saturation (lab.)
Drying from initial point A (lab.) Drying from point B (lab.)
Figure 5-33: Numerical reproduction of the experimentally observed behaviour of compacted Soil A
(Jotisankasa, 2005) employing the v-hysteretic-SWRC in combination with the saturation dependent soil
compressibility with suction 𝜅𝑠∗
309
1
(a)
0.8
0.6
0.4
0.2
0
10 100 1000 10000 100000
1.52
(b)
1.50
1.48
1.46
1.44
1.42
10 100 1000 10000 100000
Primary drying (lab.) Primary drying (ICFEP)
Primary wetting (lab.) Primary wetting (ICFEP)
Figure 5-34: Numerical reproduction of the experimentally observed behaviour of reconstituted Soil A
(Cunningham, 2000) employing the v-hysteretic-SWRC in combination with the saturation dependent soil
compressibility with suction 𝜅𝑠∗
310

B
0.8 C (a)
0.6
A
0.4
Laboratory data
0.2
ICFEP
0.0
0.1 1 10 100 1000 10000 100000
2.4
Specific volume, 
B
2.3 (b)
A
2.2
2.1 Laboratory data C

ICFEP
2.0
0.1 1 10 100 1000 10000 100000
Figure 5-35: Numerical reproduction of the experimentally observed behaviour of compacted bentonite-
kaolin mixture (Sharma, 1998) employing the v-hysteretic-SWRC in combination with the saturation
dependent soil compressibility with suction 𝜅𝑠∗
Figure 5-36: Monotonic SWRC in the 3-dimensional stress-space s-𝑆𝑟 -𝑣
311
0.8
0.6
0.4
0.2
(a)
sair
0.0
1 10 100 1000 10000 100000
2.2
2.0
sat
1.8
1.6 lim
sair
(b)
1.4
1 10 100 1000 10000 100000
1.0
0.8
0.6
0.4
0.2
lim sat (c)

0.0
1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2
Figure 5-37: Projection of the monotonic SWRC shown in Figure 5.36 on the (a) s-𝑆𝑟 (b) s-𝑣 and (c) v-𝑆𝑟
planes
312
5.6 Suggestions for future improvement of the v-hysteretic-SWRC

model
As discussed above, the scanning paths depend on their circular shape and the
primary path to which they converge, and it is, therefore, expected that
improvement of the expression used for the latter will also improve the
performance of the former.
Consequently, it is suggested that future improvement of the hysteretic model is

centred on the expression employed for the primary paths. Allowing for de-
saturation upon drying and full saturation upon wetting, to occur at distinct
values of suction is thought to be an essential development. Furthermore, the
assumption of the model, that the degree of saturation reached at residual state
is 0.0, is believed to be limiting and unrealistic. Finally, more fitting parameters
could be introduced in order to obtain a better control of the shape of the
primary paths which is currently controlled solely by one parameter for drying,
𝛼𝑑 , and one for wetting 𝛼𝑤 .
The reproduction of the SWRC is expected to be improved, by employing the

following expressions for the primary drying and for the primary wetting paths,
respectively:
∗ ∗ 𝑚 𝑑𝑟
𝑠𝑒𝑞 − 𝑠𝑑𝑒𝑠
1− ∗
𝑠0 − 𝑠𝑑𝑒𝑠
𝑑𝑟 (5.35)
𝑆𝑟,𝑝𝑟 = 𝑛 𝑑𝑟 ∙ 1 − 𝑆𝑟,0 + 𝑆𝑟,0
∗ − 𝑠∗
1 + 𝛼𝑑 ∙ 𝑠𝑒𝑞 𝑑𝑒𝑠
∗ 𝑚 𝑤𝑒𝑡
𝑠𝑒𝑞
1− 𝑠
𝑤𝑒𝑡 0
𝑆𝑟,𝑝𝑟 = 𝑛 𝑤𝑒𝑡 ∙ 1 − 𝑆𝑟,0 + 𝑆𝑟,0 (5.36)
∗
1 + 𝛼𝑤 ∙ 𝑠𝑒𝑞
where:
 ∗
𝑠𝑑𝑒𝑠 is the de-saturation value of combined suction upon drying. For
∗
suction levels lower than 𝑠𝑑𝑒𝑠 , the degree of saturation is assumed to be
1.0;
313
 𝑆𝑟,0 is the degree of saturation at a residual state, where the suction is 𝑠0

(corresponding to 𝑣 = 2.0, as for the developed v-hysteretic-SWRC
model);
 𝑚𝑑𝑟 and 𝑛𝑑𝑟 are fitting parameters for the primary drying path and;
 𝑚𝑤𝑒𝑡 and 𝑛𝑤𝑒𝑡 are fitting parameters for the primary wetting path;
are the extra parameters required in comparison to the current v-hysteretic-

SWRC model.
The parameters presented in Table 5-17 were employed in the reproduction of

the retention behaviour of three reconstituted artificial soils tested by
Cunningham (2000). All three soils, namely soil A (Figure 5-38), soil B1M9
(Figure 5-39) and M8K1B1 (Figure 5-40), exhibited significantly different values
of suction at de-saturation upon drying and full saturation upon subsequent
wetting, which cannot be modelled by the current v-hysteretic-SWRC model. It
should be noted that the model is not implemented into ICFEP and a Microsoft
Excel spreadsheet was used to compute the SWRC, while 𝜓 was assumed
equal to 0.0 in order to simplify the calculation of 𝑆𝑟 .
The improved expressions, proposed for the primary paths, provided an

accurate estimation of the degree of saturation measured in the laboratory,
confirming their efficiency and highlighting the potential benefit which is
expected to arise from their numerical implementation into ICFEP.
314
Table 5-17: Model parameters employed by the improved version of the v-hysteretic-SWRC for the
reproduction of three retention curves; soil A, B1M9 and M8K1B1
Improved version of the v-hysteretic-SWRC model
Soil A Soil B1M9 Soil M8K1B1

(Cunningham, 2000) (Cunningham, 2000) (Cunningham, 2000)
𝑠𝑎𝑖𝑟 0.0 kPa 𝑠𝑎𝑖𝑟 1.0 kPa 𝑠𝑎𝑖𝑟 0.0 kPa
𝑠0 1.0E+7 kPa 𝑠0 1.0E+5 kPa 𝑠0 1.0E+6 kPa
𝛼𝑑 3.8E-5 𝛼𝑑 5.0E-4 𝛼𝑑 1.0E-4
𝛼𝑤 3.5E-3 𝛼𝑤 2.8E-2 𝛼𝑤 5.0E-4
𝑠𝑑𝑒𝑠 300.0 kPa 𝑠𝑑𝑒𝑠 300.0 kPa 𝑠𝑑𝑒𝑠 450.0 kPa
𝑚 𝑑𝑟 0.7 𝑚 𝑑𝑟 0.9 𝑚 𝑑𝑟 0.9
𝑛𝑑𝑟 1.1 𝑛𝑑𝑟 0.9 𝑛𝑑𝑟 0.9
𝑚 𝑤𝑒𝑡 0.9 𝑚 𝑤𝑒𝑡 1.5 𝑚 𝑤𝑒𝑡 0.5
𝑛𝑤𝑒𝑡 1.1 𝑛𝑤𝑒𝑡 0.8 𝑛𝑤𝑒𝑡 0.6
𝑆𝑟,0 0.05 𝑆𝑟,0 0.02 𝑆𝑟,0 0.015
𝜓 0.0 𝜓 0.0 𝜓 0.0
1.0
0.8
0.6
Soil A
0.4
Primary drying (lab.)
Primary drying (model)
0.2
Primary wetting (lab)
Primary wetting (model)
0.0
10 100 1000 10000 100000
Figure 5-38: Numerically reproduced SWRC for soil A (data after Cunningham , 2000)
315
1.0
Soil B1M9
0.8 Primary drying (lab.)

0.6 Primary wetting (lab)
0.4
0.2
0.0
10 100 1000 10000 100000
Figure 5-39: Numerically reproduced SWRC for soil B1M9 (data after Cunningham , 2000)
1.0
Soil M8K1B1
0.8 Primary drying (lab.)

0.4
0.2
0.0
10 100 1000 10000 100000
Figure 5-40: Numerically reproduced SWRC for soil M8K1B1(data after Cunningham , 2000)
The soil-water retention curve (SWRC), which defines the relationship between
the degree of saturation or the volumetric water content and the applied suction,
is one of the most important features in unsaturated soil mechanics. The
present chapter was devoted to the development and implementation into a
numerical code (ICFEP) of three SWRC models, which account for (a) the effect
of void ratio, (b) the hydraulic hysteresis and (c) the combined effect of the two.
Finally, the elastic soil compressibility with suction was coupled with the SWRC,
to model the suction induced irreversible volumetric strains observed in the
laboratory, completing the 3-dimensional perception of the retention behaviour.
316
The first of the SWRC models presented accounts for the effect of specific
volume, 𝑣 , on the position of the retention curve (v-SWRC), through the
introduction of the parameter 𝜓, in a similar manner to that originally proposed
by Gallipoli et al. (2003b). The position of the SWRC in the 𝑠-𝑆𝑟 plane is moved
upwards or downwards depending on whether the specific volume is decreasing
or increasing, respectively (larger values of 𝑆𝑟 correspond to the same suction
level for decreasing values of 𝑣). Increasing the parameter 𝜓 intensifies this
shift whereas for 𝜓 = 0.0, 𝑣 has no effect on the simulated soil behaviour.
By including the specific volume in the retention behaviour, the SWRC becomes
a 3-dimensional surface (SWRS) in the 𝑠 - 𝑆𝑟 - 𝑣 space. Assuming suction
changes under constant volume, the soil is modelled to follow a hypothetical
retention curve – referred to, herein, as iso-volumetric SWRC – the shape of
which resembles the one proposed by Van Genuchten (1980). Gradual changes
of 𝑣 impose a continuous shift of the retention relationship, from one iso-
volumetric curve to the next one, and the retention point, defined in the 𝑠-𝑆𝑟 -𝑣
space, moves on the SWRS.
The second model was developed to include the hydraulic hysteresis

(hysteretic-SWRC model), commonly exhibited by unsaturated soils, adopting a
realistic, non-linear shape for the curves followed and ensuring a smooth
transition from scanning to primary paths. S-shape curves are employed for the
primary drying and wetting paths, which have two commons points; the point of
de-saturation and the residual point. The scanning paths are assumed to be
arcs of circles, centred on the vertical line passing through the last reversal
point, and to rejoin the corresponding primary path with a common tangent.
Four model parameters are required to fully define the hysteretic-SWRC model;
two fitting parameters – one for each primary curve – and the suctions at the
air-entry point and at the residual point. The model parameters dictate the
shape and the position of the primary curves, which remain unvarying during
the analysis. On the contrary, the scanning paths are not directly controlled by
the model parameters; their shape is always circular and the actual path
followed is determined primarily by the initial retention state and indirectly by the
model parameters through the necessity of joining the primary paths with a
317
common tangent. This lack of explicit control over the scanning paths could be
regarded as a limitation of the model which, however, guarantees simplicity.
Furthermore, improving the expression adopted for the primary curves is
expected to improve the performance of the scanning paths.
Despite its simplified formulation, implementation of the hysteretic-SWRC into a

numerical code, like ICFEP, is relatively demanding, as the calculation of the
scanning paths requires the solution of a system of non-linear equations.
Additionally, depending on the suction change and the suction level, the
appropriate path needs to be selected. For this procedure to be feasible, a
number of variables have to be stored during the analysis. These are termed
reversal parameters and consist of the suction and degree of saturation at the
last reversal point, the radius of the scanning path and the suction at the point
of intersection with the primary path and the direction of hydraulic loading
(drying or wetting). The initialisation and update of those parameters are of
central importance and care should be taken for their correct and accurate
implementation.
The capability of the model to reproduce the scanning paths was shown to be
satisfactory, despite the simple geometric shape assumed. From the analyses
presented in this chapter, it was concluded that the model is capable of
effectively reproducing the hydraulic paths obtained in the laboratory for
reconstituted and intact samples of London clay fill and of compacted artificial
soil A. It was shown, however, to be less successful in the reproduction of
scanning paths that exhibit double curvature, such as those observed for
reconstituted Weald clay.
The third model incorporates the effect of specific volume and the hydraulic
hysteresis (v-hysteretic-SWRC model) in a 3-dimensional formulation. Adopting
the same assumptions and a similar shape for the primary and scanning paths
as for the hysteretic-SWRC, this model is formulated employing the notion of
combined suction, which is a function of the currently applied suction and of the
specific volume resulting from the coupled effect of the mechanical and
hydraulic loading. The main advantage of this approach is that the 3-
dimensional SWRS, defined in the 𝑠 - 𝑆𝑟 -𝑣 space, becomes a 2-dimensional
318
hysteretic curve when plotted in terms of combined suction. Therefore, although

the primary drying and wetting surfaces bound an infinite number of scanning
surfaces in the 𝑠-𝑆𝑟 -𝑣 space, the relevant computations are dealt with in a
similar way to the one followed by the hysteretic-SWRC model.
In addition to the development of the SWRC models, the soil compressibility

with suction, 𝜅𝑠 , assumed to be constant in the constitutive models developed
by Georgiadis (2003), was modified and the alterations undertaken were
presented in this chapter. Experimental evidence provided by various authors,
including Sharma (1998), Cunningham (2000) and Jotisankasa (2005), indicate
a hysteretic relationship between the specific volume and the applied suction,
similar to the hysteresis of the retention curve. Therefore, compressibility due to
suction was modelled to be dependent on the degree of saturation, introducing
two fitting parameters. Through appropriate adjustment of these parameters a
constant compressibility, independent of the SWRC, can be obtained. The
laboratory results reported by the above mentioned authors were closely
reproduced by the proposed relationship, which is believed to realistically
account for the 3-dimensional essence of the retention relationship; the degree
of saturation, the specific volume and the suction are coupled and as such they
should be treated inseparably.
319
chapter 6: STABILITY OF HIGHLY
OVERCONSOLIDATED UNSATURATED
SOIL SLOPES
6.1 Introduction
The numerical analysis of a cut slope in a highly overconsolidated unsaturated

soil is studied under various conditions in the current chapter. The scope of the
study is to apply the Hvorslev surface, developed in Chapter 4, on an
appropriate boundary value problem, in order to demonstrate the impact of
adequately predicting the shear strength of such a soil on slope stability.
First, the results produced by the Hvorslev surface (HV) are compared with
those produced by adjusting the Lagioia et al. (1996) expression to the modified
Cam-clay surface (as in the Barcelona Basic Model – BBM – introduced by
Alonso et al., 1990), aiming to highlight the usefulness of the new surface.
Second, the HV surface is employed in three different types of analysis, namely
unsaturated, dry and effective stress analyses, in order to assess approaches
alternative to unsaturated, under specific conditions. Finally, a large number of
analyses are presented where the initial conditions of the soil and the geometry
of the excavation are parametrically changed and their effect on slope stability
320
Stability of Highly Overconsolidated Unsaturated Soil Slopes
is thus investigated. For this purpose, two OCR values were considered and the
initial groundwater table (G.W.T.) position was varied in the numerical analyses
of both a deep and a shallow excavation.
Table 6-1 summarises the analyses undertaken, giving information about the
shape of the yield and plastic potential (YS and PP, respectively) surfaces
employed (HV or BBM-type), about the type of analysis carried out (unsat., dry
or eff.st), the OCR value (5.0 or 11.0), the G.W.T. depth (at -5.0, -7.5, -10.0 or -
20.0 m) and the depth of the excavation performed (10.0 or 4.0m). The effect of
the YS and PP surfaces is presented in Section 6.4 and Analyses 1, 2, 3 and 4,
shown in Table 6-1 (light grey code on the left of the table), were considered in
the study. In order to investigate the impact of the type of analyses on slope
stability, discussed in Section 6.5, Analyses 1 and 5 to 15 (Table 6-1; grey
code) were taken into account. Finally, the effect of the initial conditions and of
the depth of the excavation were investigated in Section 6.6 using the results of
Analyses 1, 2, 5, 6, 7 and 16 to 24 (Table 6-1; dark grey code).
321
Table 6-1: Analyses undertaken for the investigation of cut slopes in highly overconsolidated unsaturated
soils
Analyses performed
Number YS and
Type of G.W.T. Excav.
Effect studied of PP OCR
analysis depth Depth
Analysis surface
1 HV Unsat. 5.5 -5.0 m 10.0 m
6.4 Effect of YS
and PP surface
2 HV Unsat. 11.0 -5.0 m 10.0 m
3 BBM-type Unsat. 5.5 -5.0 m 10.0 m
4 BBM-type Unsat. 11.0 -5.0 m 10.0 m
5 HV Unsat. 5.5 -7.5 m 10.0 m

Section 6.5 Effect of type of analysis performed
6 HV Unsat. 5.5 -10.0 m 10.0 m
7 HV Unsat. 5.5 -20.0 m 10.0 m
8 HV Dry 5.5 -5.0 m 10.0 m
9 HV Dry 5.5 -7.5 m 10.0 m
10 HV Dry 5.5 -10.0 m 10.0 m
11 HV Dry 5.5 -20.0 m 10.0 m
12 HV Eff. st. 5.5 -5.0 m 10.0 m
13 HV Eff. st. 5.5 -7.5 m 10.0 m
14 HV Eff. st. 5.5 -10.0 m 10.0 m
15 HV Eff. st. 5.5 -20.0 m 10.0 m
16 HV Unsat. 11.0 -7.5 m 10.0 m

Section 6.6 Effect of initial conditions and
17 HV Unsat. 11.0 -10.0 m 10.0 m
18 HV Unsat. 11.0 -20.0 m 10.0 m

of excavation depth
19 HV Unsat. 5.5 -5.0 m 4.0 m
20 HV Unsat. 5.5 -7.5 m 4.0 m
21 HV Unsat. 5.5 -10.0 m 4.0 m
22 HV Unsat. 11.0 -5.0 m 4.0 m
23 HV Unsat. 11.0 -7.5 m 4.0 m
24 HV Unsat. 11.0 -10.0 m 4.0 m
322
6.2 Problem description
A 10.0 m deep excavation, assumed to be performed in the silty soil tested by

Estabragh & Javadi (2008) (Chapter 4), was considered in plane strain and the
stability of the 2:1 slope generated was investigated. The excavation was 50.0
m wide at the original ground surface and 30.0 m wide at its bottom, as
illustrated in Figure 6-1. For the finite element analysis, a domain of 30.0 m
depth and 100.0 m width was discretised as shown in Figure 6-2, employing
isoparametric, quadrilateral 8-noded solid elements. The mesh is adequately
refined behind the face of the excavation, where failure occurs, and the results
are not, therefore, mesh-dependent. The origin of the x and y axes is shown in
the same figure to be at the top left corner of the discretised domain. The out-of-
plane direction is of no interest, as plane strain conditions were applied.
The material properties of the silty soil assumed in the study were presented in
Section 4.5.2. The Georgiadis (2003) unsaturated soil constitutive model (Model
1; option 1), with and without the Hvorslev surface (attached to the SC on the
wet side) was employed in the unsaturated and dry analyses (the increase of
apparent cohesion with suction was coupled to the SWRC) and a form of the
modified Cam-clay model (MCC model; for its implementation into ICFEP refer
to Potts & Zdravkovic, 1999) was used in the effective stress analyses. The
model parameters for the three different types of analyses performed are
summarised in Table 6-2 and for the unsaturated analyses are identical to those
produced by the calibration of the Hvorslev surface, presented in Section 4.5.2.
Similar parameters were adopted in the dry analyses, with the exception of
parameters 𝑘 and 𝑟, which were set equal to 0.0 and 1.0, respectively, in order
to disregard the effect of suction (see also Section 6.5). The MCC model
parameters, employed in the effective stress analyses, were chosen so that the
soil strength, stiffness and compressibility matched those predicted by the
Georgiadis (2003) model at full saturation, as explained in Section 6.5.
The simple van Genuchten (1980) type soil-water retention curve (SWRC)
presented in Section 3.4, was employed. The relative parameters are
summarised in Table 6-2 and were matched to the experimental data given by
Estabragh & Javadi (2008), as already explained in Section 4.5.2 (see also
323
Figure 4-29). The coefficient of saturated soil permeability, 𝑘, was assumed to

be 𝑘 = 10−8 m/s. The logarithm of 𝑘 was assumed to vary linearly with suction
and the relevant parameters are summarised in Table 6-2.
The soil stresses were initialised at the commencement of the analysis

(Increment 0), employing a unit weight 𝛾𝑑𝑟𝑦 = 𝛾𝑠𝑎𝑡 of 19.1 kN/m3. The coefficient
of earth pressure at rest, 𝐾0 , was assumed equal to 1.0 and the initial pore
water pressure distribution with depth was hydrostatic, allowing suctions to
develop above the initially horizontal G.W.T. Although the initial total stresses
were identical in all of the analyses considered, the effective stresses varied
according to the position of the G.W.T. The hardening/softening parameters 𝑝0∗
and 𝑝0 , for unsaturated and fully saturated conditions, respectively, were then
initialised for each Gauss point in the finite element mesh (Figure 6-2), based
on the corresponding total or effective stress and on the OCR value.
Starting from a horizontal ground surface, the 10.0 m deep excavation (Figure
6-1) was performed in 20 stages, in each of which a 0.5 m thick layer of
elements was removed. Each excavation stage was executed in 10 increments
of the coupled consolidation analysis, allowing the initial hydrostatic pore water
pressure regime to be altered.
No change of pore pressure (∆𝑝𝑓𝑏 = 0.0) was prescribed at the right vertical
boundary of the finite element mesh and the no flow condition (𝑞𝑛 = 0.0) was
applied on the remaining boundaries. When excavating below the water table,
the precipitation boundary condition, detailed in Section 3.6.1, was applied to
the newly formed boundary of the slope, prescribing either a 0.0 flow rate, 𝑞𝑛 , or
a 0.0 pore pressure, 𝑝𝑓𝑏 . On the bottom of the excavation a 0.0 pore pressure
condition, 𝑝𝑓𝑏 , was applied. The above conditions on the face and on the
bottom of the cut were applied at each excavation stage below the G.W.T., until
the final excavation level was reached. Each stage was performed in a one
month time-step. At the end of the excavation the G.W.T. reached the position
illustrated in Figure 6-3 for the case where the initial depth of the G.W.T. was
5.0 m. It should be noted that this is not the steady-state condition, as further
explained in Section 6.5.2.
324
Once the excavation was completed, coupled consolidation was deactivated

and drained analysis was thereafter performed. The hydraulic conditions
mentioned above were also switched off and changes in the factor of safety, 𝐹𝑠 ,
were applied, as described in the subsequent section.
Table 6-2: Model parameters employed in the parametric study of slope stability
Modified Cam-clay
model
Dry Effective stress
Unsaturated analyses
analyses analyses
Parameter HV BBM HV Parameter Value
𝛼𝑔,𝑓 0.7 0.4 0.7 𝜆 0.086
𝜇𝑔,𝑓 0.9999 0.9 0.9999 𝜅 0.005
𝛼𝐻𝑉 0.45 - 0.45 𝑣1 2.12
𝑛 0.5 - 0.5 𝐺/𝑝 15.0
𝛽𝐻𝑉 0.25 - 0.25 𝛼𝐻𝑉 0.45
𝑚 0.5 - 0.5 𝑛 0.5
𝑀𝑔,𝑓 1.3039 1.3039 𝛽𝐻𝑉 0.25
𝑟 0.06 1.0 𝑚 0.5
𝑘 SWRC 0.0 𝑋 0.5893
𝑠𝑎𝑖𝑟 (kPa) 0.0 0.0 𝑌𝐹 = 𝑌𝑃 0.6552
𝜆(0) 0.086 0.086 𝑌𝐹 = 𝑌𝑃 0.2300
𝜅 0.005 0.005 Soil-water retention

𝑣1 2.120 2.120 curve (SWRC)
𝛽 0.001 0.001 Parameter Value

𝑝𝑎𝑡𝑚 (kPa) 100.0 100.0 𝑠𝑑𝑒𝑠 (kPa) 0.0
𝐺 (kPa) 15.0 15.0 𝑠𝑎𝑖𝑟 (kPa) 0.0
𝜆𝑠 0.08 0.08 𝑠0 (kPa) 100000.0
𝜅𝑠 0.03 0.03 𝑆𝑟,0 0.1
𝑝𝑐 (kPa) 1.0 1.0 𝛼 0.015
𝑠0 (kPa) 1000000.0 1000000.0 𝑛 0.95
Variable permeability model 𝑚 0.7

-8
𝑘𝑠𝑎𝑡 (m/s) 10 𝑠1 (kPa) 0.0
𝑘𝑠𝑎𝑡 /𝑘𝑚𝑖𝑛 100.0 𝑠2 (kPa) 1000.0
325
100.0 m
50.0 m
10.0 m
1 G.W.T.-5.0 m
2 -7.5 m
-10.0 m
30.0 m
30.0 m
-20.0 m
Figure 6-1: Geometry of the problem
Figure 6-2: Finite element mesh
Figure 6-3: Groundwater table (G.W.T.) position at the end of the excavation
326
6.3 Factor of safety and determination of slope failure
The factor of safety against slope failure, 𝐹𝑠 , is conventionally defined as the

ratio of the actual shear strength over the minimum shear strength required to
prevent failure (Bishop, 1955) and as such is the factor by which the soil
strength needs to be divided in order to reach failure (Duncan, 1996).
The strength reduction can be achieved in two ways in a numerical analysis.

The first way is to input factored strength parameters at the beginning of the
analysis. For the factor of safety, 𝐹𝑠 , to be evaluated the factored strength
parameters for which failure is achieved are compared with the actual strength
parameters. The second approach is to gradually increase the factor of safety,
𝐹𝑠 (i.e. reduce the strength), at a relevant stage of the analysis. The value of 𝐹𝑠
for which failure is reached is the factor of safety.
The first approach has been used in the literature since the 70’s. Zienkiewicz et
al. (1975) employed this technique in finite element analyses of a homogeneous
embankment slope and of an excavated slope and compared the factors of
safety computed with those calculated from a standard slip circle analysis. They
found a generally good agreement. More recently, Dawson et al. (1999)
adopted the strength reduction technique in the finite difference analysis of a
soil slope. The Mohr-Coulomb failure criterion, with an associated flow rule and
without a tension cut-off was employed. A wide range of slope angles and of
soil friction angles and a variety of pore pressure coefficients was considered in
the study. The results were compared to upper-bound limit analysis solutions,
showing little difference in terms of computed factors of safety, when a
sufficiently refined mesh was used. The factors calculated from the strength
reduction technique were generally slightly higher than those predicted by limit
analysis.
Despite the earlier work, inclusion of the method in Eurocode 7 (EC7), in

combination with the increasing sophistication of the constitutive models
currently available, has raised questions concerning its exact application in FE
analysis (Schweiger, 2005). The implications involved in the process of
reducing the strength parameters in both ways are discussed further by Potts &
Zdravkovic (2011).
327
For the particular case of the constitutive models which are relevant to this
thesis, applying a factor of safety 𝐹𝑠 to the tangent of the critical state angle of
shearing resistance, 𝑡𝑎𝑛𝜑𝑐𝑠 , the consistency equation obtains the form
presented in Chapter 4 (Equation 4.87). This alteration enables the user to
prescribe incremental changes of 𝐹𝑠 at any given stage of the analysis. The
𝐹
factored value of the critical state angle of shearing resistance, 𝜑𝑐𝑠𝑠 , is used to
calculate the corresponding inclination of the critical state line (CSL), 𝑀𝑔 , in
triaxial compression (at Lode’s angle 𝜃 = −30°), according to Equation 4.90.
The inclination 𝑀𝑓 , associated with the yield function, as explained in Chapter 3,
is proportionally affected, since the ratio 𝑅 = 𝑀𝑓 𝑀𝑔 is assumed to remain
constant (Equation 4.91, Chapter 4). The corrected values of 𝑀𝑓 𝐹𝑠 and 𝑀𝑔 𝐹𝑠 are
then employed in the cubic Equation 3.36 (Matsuoka-Nakai failure criterion), to
evaluate the inclinations 𝐽2𝜂𝑓 𝐹𝑠 and 𝐽2𝜂𝑔 𝐹𝑠 , respectively, which affect the
position of the yield and plastic potential surfaces (Equation 3.31, Chapter 3).
Consequently, by increasing 𝐹𝑠 the two surfaces reduce in size, as shown for
the Lagioia et al. (1996) expression under unsaturated conditions in Figure 6-4.
As explained in Chapter 4, for the Hvorslev surface to lie above the CSL and
softening to be predicted, it is necessary that 𝛼𝐻𝑉 ≤ 𝐽2𝑛𝑔 . For this condition to
be ensured when applying incremental changes of 𝐹𝑠 and because 𝐽2𝑛𝑔
reduces, the ratio 𝛼𝐻𝑉 𝐽2𝑛𝑔 remains constant (in ICFEP the input parameter
may be the inclination 𝛼𝐻𝑉 or the ratio 𝛼𝐻𝑉 𝐽2𝑛𝑔 ).
A stress point, initially lying within the elastic region, at first remains unaffected
by the changes in the factor of safety, as long as the shrinking yield surface is
not encountered. Once the yielding criterion is violated, the consistency
condition ensures that the stress point remains on the yield surface and is
dragged with it. Therefore, as 𝐹𝑠 increases, the stress point approaches critical
state (Figure 6-5), ultimately leading to failure.
For the cases studied herein, once excavation was completed, 𝐹𝑠 was
incrementally increased from 1.0, until a failure mechanism was formed and the
corresponding factor of safety was thus determined. Nonetheless, establishing
328
slope failure in a numerical analysis is not straightforward. Griffiths & Lane

(1999) listed three possible ways of defining failure: bulging of the slope profile
(Snitbhan & Chen, 1976); limiting of the shear stresses on the potential failure
surface (Duncan & Dunlop, 1969); nonconvergence of the solution (Zienkiewicz
& Taylor, 1989) and noted that these were discussed by Abramson et al. (1995)
but without resolution. In the present study, failure was assumed when a fully
developed mechanism was evident from the vectors of incremental soil
displacements, as in the example shown in Figure 6-6. Furthermore, failure is
indicated by the evolution of the horizontal and vertical displacements at the
crest and at the toe of the slope (dramatic change in the gradient of the
displacement curves), as illustrated for the latter in Figure 6-7. Finally, non-
convergence of a subsequent analysis increment is the third criterion to verify
failure.
Comparable criteria for the determination of slope failure were used by

Georgiadis et al. (2007), even though the conventional strength reduction
approach was employed for the evaluation of the corresponding factor of safety.
For an excavation similar to the one studied herein, Georgiadis et al. (2007)
performed finite element analyses for different initial groundwater table depths.
The analyses were repeated for reducing values of angle 𝜑𝑐𝑠 , until failure
occurred at the last stage of the excavation and the factor of safety was then
𝑓𝑎𝑖𝑙𝑢𝑟𝑒
calculated as 𝐹𝑠 = tan 𝜑𝑐𝑠 tan 𝜑𝑐𝑠 .
The two above mentioned methods of applying the factor of safety in FE

analysis – stepped reduction of the strength parameters at a given stage of the
analysis, or input of factored strength parameters at the beginning of the
analysis – may yield dissimilar results, as reported by Blackwell (2010). For the
case of a cantilever wall, the latter study revealed a minor influence of the
method adopted on the factor of safety calculated. However, the two methods
were shown to produce significantly different structural forces. Furthermore, for
the case of a single propped wall, the excavation level achieved before failure
occurred, did not agree for the two methods employed and neither did the
structural forces, raising questions about the appropriate application of the
factor of safety in finite element analysis.
329
In order to assess the difference of the two methods in the particular case
examined herein, the unsaturated analysis for OCR = 5.5, denoted as Analysis
1 in Table 6-1, was performed adopting both procedures. In the first case,
incremental changes of the factor of safety were applied after the excavation
was completed and the generated slope failed when 𝐹𝑠 = 2.76 (as presented in
Section 6.4), corresponding to 𝑀𝑔 = 0.484 . Repeating the excavation with
𝑀𝑔 = 0.484, failure should be predicted at the last excavation phase, implying
that for the specific inclination of the CSL, the factor of safety is 1.0. Indeed,
failure occurred during the penultimate excavation phase when 𝑀𝑔 = 0.484 and
during the ultimate for 𝑀𝑔 = 0.49 , which is thought to be an adequate
approximation. It can, therefore, be concluded that the two approaches
practically produced similar results for the particular boundary value problem
investigated. This conclusion is in agreement with the study currently
undertaken by Potts & Zdravkovic (2011), which supports the idea that for
drained analysis, the method of factoring the strength parameters is not
expected to have a significant influence on the computed results.
Finally, it should be noted that the automatic incrementation algorithm described

in Chapter 3 was employed in the analyses presented in the current chapter.
YS, initial
PP, initial
YS, factored
PP, factored
f (seq) p0
Figure 6-4: Shrinkage of the yield (YS) and plastic potential (PP) surfaces due to an increase in the
factor of safety 𝐹𝑠
330

stress point
YS (0)
YS (1)
YS (2)
YS (3)
f (seq) p0 (OCR = 5.5) p0 (OCR = 11.0)

Figure 6-5: Movement of the stress point due to the shrinking of the yield surface from the initial position
YS (0) to the subsequent position YS (1) and (YS (2) and to the final position (YS (3) due to the increase in
the factor of safety 𝐹𝑠
Figure 6-6: Vectors of incremental displacements at failure
0.30
horizontal
0.20
vertical
Displacement (m)
0.10
0.00
1 2 3 4 5
Factor of safety, Fs ( )
-0.10
-0.20
Figure 6-7: Horizontal and vertical displacements of the toe of the slope signifying failure
331
6.4 Effect of the yield and plastic potential surfaces
As explained in Chapter 3, the Lagioia et al. (1996) expression can reproduce a

large number of shapes for the yield and the plastic potential surfaces in the 𝑝-𝐽
stress plane. In Section 4.5.2 the original Cam-clay (CC), the modified Cam-
clay (MCC) and the Sinfonietta Classica (SC) surfaces were shown to
inadequately predict the shearing behaviour – under highly overconsolidated
states – of the soil considered in the current slope stability analysis. On the
contrary, when the Hvorslev surface (HV) was attached to the SC surface, the
behaviour was more accurately reproduced. The impact on slope stability of the
shape assumed for the surfaces is studied in the current section.
Four analyses were performed: two employing the HV on the dry side, attached
to the SC on the wet side (Analyses 1 and 2 in Table 6-1) and two adopting the
MCC surface on both sides, in which case a BBM-type approach was followed
(Analyses 3 and 4 in Table 6-1). The plastic potential function was controlled by
the parameters 𝛽𝐻𝑉 and 𝑚 when the HV was used and associated plasticity was
assumed on the remaining surfaces.
The initial G.W.T. was 5.0 m deep and two OCR values, 5.5 and 11.0, were
considered. Once the 10.0 m deep excavation was completed, incremental
changes of the factor of safety were applied, as explained in the previous
section.
When the HV was employed failure was reached for factors of safety equal to
2.76, for OCR = 5.5, and 3.28, for OCR = 11.0 (Table 6-3). The predicted failure
mechanisms, in terms of vectors of incremental displacements at the last stable
increment of the analysis, are shown in Figure 6-8 (a) for OCR = 5.5 and in
Figure 6-8 (b) for OCR = 11.0. These plots indicate the failure of the whole
slope, from its tow to the crest.
The contours of sub-accumulated plastic volumetric and deviatoric strains

(measured from the end of the excavation), shown in Figure 6-9 and Figure
6-10, respectively, also confirm the extent of the failure mechanisms for the
cases of OCR = 5.5 (plots (a)) and OCR = 11.0 (plots (b)) and indicate that the
332
failure is initiated from the toe of the slope (in the above figures tension is
positive).
On the contrary, when the BBM-type surface was employed in the same
analyses, large vectors of incremental displacements were computed at the toe
of the slope, where a very localised failure was generated, as shown in Figure
6-11 for Analysis 3 (for OCR = 5.5). At this stage the factor of safety was 2.9.
The MCC surface, which was employed in Analysis 3, was shown in Chapter 4
to highly over-predict the peak deviatoric stress, 𝐽𝑝𝑒𝑎𝑘 . Furthermore, a greater
strain softening rate was predicted by the MCC surface (see also Figures 4-35
and 4-37). The analysis would not converge further after this point. Therefore, to
compare the results from the two surfaces, it was chosen to present the results
from the BBM-type surface at the same 𝐹𝑠 obtained with the HV surface (i.e.
2.76 and 3.28 for OCR = 5.5 and OCR = 11.0, respectively).
The vectors of incremental displacements are shown in Figure 6-8 (c) and (d),
while contours of sub-accumulated plastic volumetric and deviatoric strains
(measured from the end of the excavation) are shown in Figure 6-9 (c) and (d)
and Figure 6-10 (c) and (d). As in the case of the HV surface, the latter two
plots indicate the initiation of failure from the toe of the slope, but no
progression up the slope.
As discussed above, a larger peak deviatoric stress was predicted by the BBM-
type surface in comparison with the HV surface. The overestimation of the
stresses reported in Section 4.5.2 was shown in the current chapter to have
resulted in an over-prediction of the slope stability in the boundary value
problem examined. Therefore, it is recommended not to employ models
assuming an elliptical yield surface in the 𝑝-𝐽 plane, centred on the 𝑝 axis, in the
analysis of highly overconsolidated soils, without previously addressing this
particular shortcoming. Alternatively, the Hvorslev surface may be employed.
Table 6-3: Factors of safety associated with use of different yield surfaces and OCR values
Yield surface effect
Surface OCR = 5.5 OCR = 11.0
Hvorslev Surface/SC 2.76 (Anal. 1) 3.28 (Anal. 2)

BBM-type - (Anal. 3) - (Anal. 4)
333
Figure 6-8: Vectors of incremental displacement for Analyses 1, 2, 3 and 4
334
Figure 6-9: Contours of plastic volumetric strains generated at the end of Analyses 1, 2, 3 and 4
335
Figure 6-10: Contours of plastic deviatoric strains generated at the end of Analyses 1, 2, 3 and 4
336
Figure 6-11: Vectors of incremental displacement for Analysis 3 (OCR = 5.5, BBM-type) corresponding
to 𝐹𝑠 =2.9
6.5 Effect of type of analysis performed
Despite the advances in both the understanding and modelling of the behaviour
of unsaturated soils, numerical analyses considering partial saturation are rare
in engineering practice. The purpose of the current section is to compare the
factors of safety, 𝐹𝑠 , obtained from different types of analyses and to investigate
the degree of inaccuracy introduced when partial saturation is neglected. All the
analyses presented in the section assumed an OCR value of 5.5. The following
types of analyses were performed:
 Analyses in which the Georgiadis (2003) constitutive model (Model 1;

Option 1), with the newly developed Hvorslev surface on the dry side of
the critical state, was employed. The Hvorslev surface was attached to
the SC surface on the wet side. These analyses are termed ‘unsaturated’
and are shown in Table 6-1 as 1 (G.W.T. at -5.0 m), 5 (G.W.T. at -7.5 m),
6 (G.W.T. at -10.0 m) and 7 (G.W.T. at -20.0 m).
 Analyses using the same model as above but neglecting the effect of
suction on the LC curve and on the increase of apparent cohesion,
through appropriate adjustment of the relevant parameters; 𝑟 = 1.0 and
𝑘 = 0.0 (Table 6-2). These analyses are shown in Table 6-1 as 8, 9, 10
and 11 for G.W.T. at -5.0, -7.5, -10.0 and -20.0 m, respectively. They are
referred to, herein, as ‘dry’, since the suction level above the G.W.T. was
irrelevant and the soil was essentially treated as dry (albeit with the same
bulk unit weight as the saturated soil below the G.W.T.). It should be
337
noted at this point that the constitutive model performs in terms of total
stresses above the G.W.T., where unsaturated conditions apply, and in
terms of effective stresses below, where the soil elements are fully
saturated.
 Analyses adopting the modified Cam-clay model (MCC model),

implemented into ICFEP as explained by Potts & Zdravkovic (1999). A
Hvorslev surface similar to the one developed for the Georgiadis (2003)
constitutive model (Chapter 4), was also implemented for the MCC
model (the relevant equations can be found in Appendix A) and was
employed in the current study. Partial saturation was entirely disregarded
and the effective stress was the sole stress variable used. Negative pore
water pressures were allowed above the G.W.T. thus increasing the
effective stresses. Therefore, these analyses are, herein, termed
‘effective stress’ analyses and are shown in Table 6-1 as 12, 13, 14 and
15 for G.W.T. at -5.0, -7.5, -10.0 and -20.0 m, respectively.
6.5.1 Comparison between unsaturated and dry analyses
The factors of safety predicted for the unsaturated and the dry analyses are
summarised in Table 6-4 and are plotted versus the G.W.T. depth in Figure
6-12. Comparing the two curves in this figure the importance of accounting for
partial saturation, especially as the G.W.T. deepens, is revealed. By modelling
the soil as dry above the G.W.T., the increase in strength with suction was not
captured and slope stability was underestimated.
A similar trend was reported by Georgiadis et al. (2007) who studied the stability
of an unsaturated cut slope in slightly overconsolidated Thanet sand (OCR =
1.5), employing the original version of the Georgiadis (2003) model. The
cohesion increase parameter 𝑘 was set equal to the degree of saturation, 𝑆𝑟 ,
obtained from the van Genuchten (1980) expression. Two SWRC’s were
considered – for Thanet sand and for Lambeth Group sand – in order to assess
the influence of the curve adopted on slope stability. In additional to the
unsaturated analyses, dry (conventional) analyses, similar to those presented in
the current section, were performed.
338
The stability of the 10.0 m deep excavation, which was similar the one
considered in the current chapter, was investigated for G.W.T. depths of 2.5,
5.0, 7.5 and 10.0 m. As already explained, Georgiadis et al. (2007) employed
factored strength parameters at the beginning of each analysis and assumed
failure based on similar criteria as the ones adopted in the present study. The
factors of safety computed for the unsaturated and dry (conventional) analyses
are presented in Figure 6-13. Neglecting the effect of partial saturation led to
more conservative results and smaller factors 𝐹𝑠 were calculated. Furthermore,
the difference between the curves in Figure 6-13 increased with the G.W.T.
depth.
Figure 6-14 combines the results of the earlier study with those of the current
one, supporting the conclusion that disregard of the soil suction in the numerical
analysis systematically caused under-prediction of the slope stability. The effect
of OCR on the factor of safety is discussed later in the same chapter (Section
6.6.2) and the difference observed in the magnitude of 𝐹𝑠 during the present
research project and the earlier study is explained.
Lower G.W.T. positions improved the slope stability in the unsaturated

analyses. Nevertheless, for the dry analyses the increase in 𝐹𝑠 was limited to
G.W.T. depths shallower than the excavation depth (10.0 m), as shown in
Figure 6-12. For G.W.T. = -20.0 m and for G.W.T. = -10.0 m, the soil behind the
slope was modelled exclusively in terms of total stresses, which are
independent of the G.W.T. position and similar factors 𝐹𝑠 were, therefore,
calculated. For G.W.T. = -7.5 m and for G.W.T. = -5.0 m, the soil at the lower
part of the slope was modelled in terms of effective stresses, which due to the
compressive pore water pressures are smaller than the total stress. To explain
things further, Figure 6-15 illustrates the final G.W.T. positions for two of the dry
analyses: 8 (G.W.T. = -5.0 m) and 10 (G.W.T. = -10.0 m). The grey area in the
figure was modelled in terms of effective stresses in Analysis 8, which were
smaller than the total stresses employed in Analysis 10. Therefore, the factor 𝐹𝑠
decreased when the initial position of the G.W.T. was raised from -10.0 m to -
7.5 m and to -5.0 m.
339
Table 6-4: Factors of safety obtained for the unsaturated and dry analyses
Type of analysis
Water Table Unsaturated Dry
-5.0 m 2.76 (Anal. 1) 2.46 (Anal. 8)

-7.5 m 3.01 (Anal. 5) 2.87 (Anal. 9)
-10.0 m 3.64 (Anal. 6) 3.21 (Anal. 10)
-20.0 m 6.52 (Anal. 7) 3.19 (Anal. 11)
unsaturated analysis
dry analysis
6
2
-20.0 -17.5 -15.0 -12.5 -10.0 -7.5 -5.0
Groundwater table position, G.W.T. (m)
Figure 6-12: Variation of Factor of safety 𝐹𝑠 with the assumed groundwater table (G.W.T.) depth, for the
unsaturated and dry analyses
Figure 6-13: Variation of the factor of safety with groundwater table (after Georgiadis et al., 2007)
340
Figure 6-14: Comparison of factors of safety computed from unsaturated and dry (or conventional)
analyses during the current study and by Georgiadis et al. (2007)
Figure 6-15: Groundwater table (G.W.T.) positions at -5.0 m and -10.0 m for dry analyses
6.5.2 Comparison between unsaturated and effective stress analyses
As detailed by Bishop & Bjerrum (1960) for fully saturated conditions, slope
stability reduces with time since the mean effective stresses decrease with
swelling and equilibration of the initially depressed pore water pressures (see
also Figure 6-16). Leroueil (2001) presented the degradation of factor of safety
with time, calculated by Chandler (1984) for slopes excavated in fully saturated
341
Brown London clay, to be as shown in Figure 6-17 and noted that the time
necessary to reach pore pressure equilibration depends on the swelling
properties of the soil, its hydraulic conductivity, the stratigraphy of the deposit
and the geometry of the excavation.
For the effective stress analyses undertaken herein, the time-step employed
was sufficient for the G.W.T. to reach its final (steady-state) position by the end
of the excavation (i.e. full dissipation of the excess pore water pressure had
occurred) and therefore, the factors of safety evaluated correspond to their
long-term, and, therefore, critical values. On the contrary, pore water pressures
in the unsaturated analyses presented in the previous section had not reached
equilibrium by the end of the excavation. The positions of the G.W.T. at the end
of the excavation for the two types of analyses are shown in Figure 6-18 to
differ significantly.
For the results of the unsaturated and the effective stress analyses to be
comparable the former analyses were repeated for a significantly larger time
step (10 years per excavation stage) in order to achieve steady-state
conditions. In fact only Analyses 1 and 5, for initial G.W.T. at -5.0 m and -7.5 m,
were repeated, as in the other cases (i.e. G.W.T. at -10.0 m and -20.0 m)
comparable pore pressures had been generated by the end of the excavation,
independently of the type of analysis performed.
The positions of the G.W.T. (zero pore water pressure contour) in the short term
(time-steps of 1 day per excavation stage were employed to obtain short-term
conditions and the only hydraulic condition applied was a no change pore
pressure on the right hand vertical boundary of the mesh) and at steady-state
are illustrated in Figure 6-19 (a) and (b) for the unsaturated and the effective
stress analyses, respectively (the initial position of the G.W.T. was at -5.0 m).
Although swelling is predicted in the long-term for the fully saturated analysis
(i.e. effective stress analysis), the process in the unsaturated analysis is
complicated: swelling is predicted for the soil below the bottom of the
excavation, whereas consolidation is predicted for the soil behind the face of the
excavation. This is thought to be the consequence of limited negative load
being transferred to the water-phase of the unsaturated soil above the G.W.T.
342
during the excavation. It should be noted that the consolidation process in

unsaturated soils is affected by the shape of the SWRC (Wong et al., 1998).
Nonetheless, this is outside the scope of the present thesis and was not further
investigated. From the above discussion it may be concluded that, for the
particular unsaturated cases examined, the G.W.T. was predicted to depress in
the long-term. Therefore, the factors of safety corresponding to steady-state
(Analyses 1a and 5a in Table 6-5) were larger than the factors of safety for
Analyses 1 and 5 (Table 6-4).
In the deviatoric plane the Georgiadis (2003) constitutive model employs the
Matsuoka & Nakai (1974) failure criterion (Figure 6-20), which allows the
strength to vary with Lode’s angle 𝜃. The stress ratio 𝑀𝑔 𝑇𝐶 in terms of 𝑞 and 𝑝,
corresponding to failure under triaxial compression, is an input parameter and
may be calculated from the critical state angle of shearing resistance 𝜑𝑐𝑠 𝑇𝐶 , at
triaxial compression, from Equation 3.38. The stress ratio at failure in terms of 𝐽
and 𝑝 is 𝐽2𝜂𝑔 and varies with the angle 𝜃, as indicated by Equation 3.36. In the
latter equation the quantity 𝐶𝑔 𝑇𝐶 is calculated from the input parameter 𝑀𝑔 𝑇𝐶

(Equation 3.37).
The stress ratio 𝐽2𝜂𝑔 depends on the loading conditions and so does the stress
ratio 𝑀𝑔 and the angle of shearing resistance 𝜑𝑐𝑠 . The variation of 𝜑𝑐𝑠 with
Lode’s angle 𝜃 is illustrated in Figure 6-21 for the particular case examined in
this chapter; starting from the value 32.38° in triaxial compression (𝜃 = −30°)
(𝜑𝑐𝑠 𝑇𝐶 = 32.38°), the angle 𝜑𝑐𝑠 obtained a maximum of 36.73° at 𝜃 = −11° and
reduced to 32.38° at triaxial extension (𝜃 = +30°).
The MCC model offers two options regarding the failure criterion in the
deviatoric plane: the Mohr-Coulomb hexagon (Figure 6-20), which assumes a
′
constant 𝜑𝑐𝑠 , and the Van Eekelen (1980) expression, for which a wide range of
′
variations of 𝜑𝑐𝑠 with 𝜃 may be reproduced, through appropriate adjustment of
three fitting parameters.
The Mohr-Coulomb hexagon is shown in Figure 6-21 to under-predict the

Matsuoka-Nakai value of 𝜑𝑐𝑠 by a maximum of 4.35°. The two criteria agree
solely under triaxial conditions, as it is also evident from Figure 6-20, where the
343
two shapes coincide for 𝜃 = ±30°. However, triaxial conditions are unlikely to be
maintained in the boundary value problem. For the MCC model to predict
′
values of 𝜑𝑐𝑠 comparable to the Georgiadis (2003) model (i.e. 𝜑𝑐𝑠 ), the Van
Eekelen (1980) expression was used in the effective stress analyses. The
stress ratio 𝑔 𝜃 in terms of 𝐽 and 𝑝′ was calculated from the following equation,
for values of Lode’s angle 𝜃 varying from −30° to +30°:
−𝛧
𝑔 𝜃 = 𝑋 1 + 𝑌 sin 𝜃 (6.1)
where 𝑋, 𝑌 and 𝑍 are constants. Within ICFEP, 𝑌 and 𝑍 are allowed to assume
distinct values for the yield ( 𝑌𝐹 and 𝑍𝐹 ) and plastic potential ( 𝑌𝑃 and 𝑍𝑃 )
surfaces in order to obtain a non-associated flow rule in the deviatoric plane.
′
The value of 𝜑𝑐𝑠 which corresponds to the stress ratio 𝑔 𝜃 may be calculated
from the following expression:
′
sin 𝜑𝑐𝑠
𝑔 𝜃 = ′
sin 𝜃 sin 𝜑𝑐𝑠 (6.2)
cos 𝜃 +
3
The parameters 𝑌𝐹,𝑍𝐹, 𝑌𝑃, 𝑍𝑃 and 𝑋, shown in Table 6-2, were adopted so
′
that the variation of the critical state angle of shearing resistance, 𝜑𝑐𝑠 , with
Lode’s angle, 𝜃, matched the variation of 𝜑𝑐𝑠 predicted by the Matsuoka-Nakai
criterion, as illustrated in Figure 6-21.
The rest of the MCC model parameters, summarised in Table 6-2, matched the
ones employed for the Georgiadis (2003) model at full saturation. The
parameters 𝜆, 𝜅 and 𝑣1 of the MCC model were set equal to 𝜆 0 , 𝜅 and 𝑣1 of
the Georgiadis (2003) model. Furthermore, the ratio 𝐺 𝑝′ was set equal to 𝐺 𝑝
and the same values of 𝛼𝐻𝑉 , 𝛽𝐻𝑉 , 𝑛 and 𝑚, controlling the HV surface, were
used for the two models.
The factors of safety achieved are shown in Table 6-5 and are plotted against
the G.W.T. depth in Figure 6-22. The curve for the unsaturated analyses lies
above the one for the effective stress analyses but the two do not exhibit a
significant difference. The effect of suction on the soil strength was adequately
approximated by the effective stress approach. It could, therefore, be concluded
344
that, for the particular drained analyses examined, if Bishop’s stress for 𝜒 = 𝑆𝑟 ,
is substituted by Terzaghi’s effective stress, only a relatively small error is
introduced. It should, however, be emphasised that effective stress analysis is
not appropriate when suction changes are involved, as their effect on the
volumetric behaviour of unsaturated soils cannot be simulated, according to the
work of Jennings & Burland (1962).
Unsaturated analysis should always be favoured when dealing with unsaturated

soils. Nonetheless, in cases where such an analysis is not feasible, it is
preferable, according to the herein presented results, to follow a conventional,
effective stress approach, rather than performing a dry analysis, as long as
drained conditions are ensured throughout. The results of dry analyses are
shown here to be excessively conservative.
Table 6-5: Factors of safety obtained for the unsaturated and effective stress analyses in the long-term
conditions
Type of analysis (long-term conditions)
Water Table Unsaturated Effective stress

-5.0 m 2.98 (Anal. 1a) 2.68 (Anal. 12)
-7.5 m 3.28 (Anal. 5a) 2.98 (Anal. 13)
-10.0 m 3.64 (Anal. 6) 3.32 (Anal. 14)
-20.0 m 6.52 (Anal. 7) 6.28 (Anal. 15)
Figure 6-16: Changes in pore pressure and factor of safety during the excavation of a cut in clay (after
Bishop & Bjerrum, 1960)
345
Figure 6-17: Variation of the factor of safety with time for slopes excavated in Brown London clay (after
Lerouil, 2001, deducted from Chandler, 1984)
Figure 6-18: Groundwater table (G.W.T.) positions reached at the end of the excavation for the
unsaturated and the effective stress analyses with initial G.W.T at -5.0 m
346
Figure 6-19: Short-term and steady-state G.W.T. for: (a) the unsaturated analysis; (b) the effective stress
analysis
Figure 6-20: Failure surfaces in the deviatoric plane (after Potts & Zdravkovic, 1999)
347
38.0
triaxial plane triaxial
Critical state anlge of shearing
compression strain extension
37.0
resistence, cs (o)
36.0
35.0
Mohr-Coulomb hexagon
34.0
Matsuoka-Nakai (1974)
33.0 van Eekelen (1980)
32.0
31.0
-35 -30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30 35
Lode's angle,  (o)

Figure 6-21: Variation of critical state angle of shearing resistance, 𝜑𝑐𝑠 , with Lode’s angle, θ, predicted
by: the Mohr-Coulomb hexagon; the Matsuoka-Nakai (1974) failure criterion; the van Eekelen (1980)
expression
7
unsaturated analysis
effective stress analysis
6
2
-20.0 -17.5 -15.0 -12.5 -10.0 -7.5 -5.0
Figure 6-22: Variation of Factor of safety 𝐹𝑠 with the assumed groundwater table (G.W.T.) depth, for the
unsaturated and effective stress analyses
6.6 Effect of initial conditions and of excavation depth
In the previous two sections it was demonstrated that numerical analysis of

slope stability in highly overconsolidated unsaturated soils requires employment
348
of a constitutive model which accounts for the beneficial effect of suction without
over-predicting the soil strength. Employing the Hvorslev surface on the dry side
of critical state and performing an unsaturated analysis was shown to be the
most realistic of the approaches examined. The same approach was used in the
parametric study presented in the current section.
The study concerns the following:
 The effect of suction: for the suctions within the ground to vary in each
case, the initial position of the G.W.T. was altered. The depths
considered were 5.0, 7.5, 10.0 and 20.0 m. The excavation was
consequently performed under different initial suction and saturation
conditions. The variation of 𝐹𝑠 with the G.W.T. depth is attributed to the
different pore water pressure regime generated at the end of the
excavation.
 The effect of OCR: two values were considered, 5.5 and 11.0. The initial
effective and total stresses, input at the commencement of the analysis,
were unchanged. On the contrary, the initial hardening parameters 𝑝0′
and 𝑝0∗ for fully and for partly saturated conditions, respectively, were two
times larger for the latter case in comparison with the former. Therefore,
although the stress states remained the same, their position relative to
the yielding criterion changed, affecting the stability of the cut slope.
 The effect of the excavation depth: in addition to the 10.0 m deep

excavation, a shallower one, of 4.0 m, was considered. Once the 8 top
layers (0.5 m deep each), shown in Figure 6-1, were removed the factor
of safety was evaluated. In this way, the geometry of the problem was
altered and its effect on slope stability was investigated.
8 analyses (Analyses 1, 2, 5, 6, 7 and 16 to 18 in Table 6-1) were performed for

the deep excavation: 4 for OCR = 5.5 and 4 for OCR = 11.0. In these analyses
the position of the G.W.T. was varied. Another 6 analyses (Analyses 19 to 24)
were carried out for the shallow excavation; 3 for OCR = 5.5 and 3 for OCR =
11.0. Once more, the depth of the G.W.T. was the parameter to vary. At this
point it should be explained that the analyses for G.W.T. at -20.0 m were not
349
repeated for the shallow excavation. The reason was that the evolution of 𝐹𝑠
with the G.W.T. position, subsequently presented, indicated that failure would
have been achieved for unrealistically low critical state angles of shearing
resistance (note that for Analysis 24, for which OCR was 11.0 and the G.W.T.
𝐹
was at -10.0 m, the value of 𝐹𝑠 calculated was 9.30 corresponding to 𝜑𝑐𝑠𝑠 =
3.63o).
The factors of safety achieved for each one of the above analyses is
summarised in Table 6-6. Also shown in the same table is the ratio R of the
G.W.T. depth over the excavation depth.
Figure 6-23 shows the factors of safety plotted versus the G.W.T. position, for
the two OCR values and for the two excavation depths considered. Slope
stability improved with increasing suction, as 𝐹𝑠 increased with the G.W.T.
depth. It also improved with OCR as the curves for OCR = 11.0 are shown to lie
above those for OCR = 5.5. Finally, larger factors of safety were calculated for
the shallow excavation in comparison with the deep one.
Furthermore, there seems to be a correlation between the ratio R of the G.W.T.

depth over the excavation depth and the factor of safety 𝐹𝑠 . In Figure 6-24, 𝐹𝑠 is
plotted versus the ratio R and two curves may be identified: one for OCR = 11.0
and one for OCR = 5.5 (the numbers in the figure refer to the numbers of the
Analyses in Table 6-1). Also shown in the same figure is the equivalent curve
for OCR = 1.5, which was derived from the Thanet sand data shown in Figure
6-13 (Georgiadis et al., 2007). Figure 6-24 shows that lowering the G.W.T. and
limiting the excavation depth had a similar stabilising effect on the slope. This
conclusion was drawn for both OCR values considered in this study and was
further supported by the earlier investigation of Georgiadis et al. (2007).
6.6.1 Effect of G.W.T. and excavation depth
The observed improvement of slope stability with the G.W.T. depth (Figure
6-23) resulted from the increase of apparent cohesion and therefore of soil
strength with suction (as explained in Chapter 3). A similar improvement was
observed when the ratio R increased (Figure 6-24), indicating that, for slope
stability to be improved, suction needs not necessarily to increase in absolute
350
terms but relative to the excavation depth and, therefore, relative to the active
earth pressures within the deforming ground and especially those at the toe of
the slope.
Indeed, 𝐹𝑠 was larger for Analysis 19 of the shallow excavation (G.W.T. at -5.0
m, OCR = 5.5) than for Analysis 6 of the deep excavation (G.W.T. at -10.0 m,
OCR = 5.5), although the G.W.T. was deeper in the latter case. The suction at
the lower part of the deep excavation was small relative to the large active
pressures, whereas the suction at the lower part of the shallow excavation was
larger relative to the small active pressures.
Moreover, despite the large active pressures at the toe of the slope in Analysis
7 (deep excavation, G.W.T. at -20.0 m, OCR = 5.5), the suction was relatively
large and the slope was more stable than the shallow slope in Analyses 20
(shallow excavation, G.W.T. at -7.5 m, OCR = 5.5).
From the above observations it may be concluded that the depth of the
excavation and of the G.W.T. should be considered concurrently and not
independently.
Various other authors have studied the influence of suction on slope stability.
Fredlund (1981) studied a typical unsaturated slope in Hong Kong and reported
a significant increase in the factor of safety with suction, justified by the increase
of apparent cohesion. The factor of safety was evaluated employing commonly
used methods of slices: the ordinary method of Fellenius (Fellenius, 1936); the
simplified Bishop’s method (Bishop, 1955); Spencer’s method (Spencer, 1967);
Janbu’s simplified method (Janbu, 1973) and Morgenstern-Price method
(Morgenstern & Price, 1965). The results are illustrated in Figure 6-25 in terms
of factors of safety and suction induced cohesion. 𝐹𝑠 increased by almost 150%
for an increase of cohesion of 80.0 kPa, demonstrating the significant influence
of soil suction on slope stability.
In a slightly different study, Rahardjo et al. (2010) investigated the effect of

groundwater table position on slope stability during rainfall. The authors
considered a slope of 15.0 m height, with a 27o angle, formed in two different
geological formations, typically met in Singapore: the Bukit Timah granite (BT)
351
and the Jurong Formation (JF). Three initial positions were assumed for the
groundwater table for each one of the two slopes: one corresponding to the
driest period, one corresponding to the wettest period and one in between
(average). Variations in the factor of safety at the BT and the JF soil slopes,
during rainfall, are given in Figure 6-26 and Figure 6-27, respectively. The
authors noted that the groundwater table position affected the initial factor of
safety, prior to rainfall, so that the shallower the position of the G.W.T., the
lower the initial factor of safety would be. The results of the current study agree
in principle with the above observation of Rahardjo et al. (2010).
6.6.2 Effect of OCR
Slope stability improved for increasing values of OCR. As already mentioned,

this conclusion was supported by the Georgiadis et al. (2007) data for OCR =
1.5, also shown in Figure 6-24. However, it should be noted that the different
factors of safety calculated during the current project and from the earlier study
are due not only to the different OCR values assumed but also to the distinct
material and SWRC parameters employed, despite adopting similar values for
the critical state angle of shearing resistance in triaxial compression (32.38° in
the current study and 33° in the former one).
Employing common initial stresses, the difference observed in the computed

factors of safety for OCR 5.5 and 11.0 was exclusively due to the initialisation of
the hardening parameter 𝑝0∗ , for unsaturated, and 𝑝0′ , for fully saturated
conditions. At the commencement of each analysis, the hardening parameter 𝑝0∗
was computed, for each Gauss point under partial saturation, from the total
stress state and the OCR value input. Subsequently, the isotropic yield stress,
𝑝0 , at the current suction level was derived from the hardening parameter, 𝑝0∗ ,
as explained in Chapter 3. Therefore, larger values of OCR correspond to larger
values of 𝑝0∗ and consequently of 𝑝0 . Similarly, under full saturation, a larger
hardening parameter, 𝑝0′ , was evaluated.
The isotropic yield stress, 𝑝0 , at the current suction level (or 𝑝0′ for fully
saturated conditions) controls the size of the elastic region and the position of
the yield surface. As schematically shown in Figure 6-28 for a random initial
352
isotropic stress state, the elastic region is larger for OCR 11.0 and the Hvorslev
surface predicts higher peak deviatoric stresses, 𝐽𝑝𝑒𝑎𝑘 , justifying the increased
factors of safety, 𝐹𝑠 , observed in comparison with the lower OCR value.
6.6.3 Displacements
The horizontal and vertical sub-accumulated displacements of selected nodes

(measured from the end of the excavation) are presented in Figure 6-29 to
Figure 6-33. Nodes 1, 2 and 3 refer to the crest, the middle and the toe of the
deep excavation and Nodes A and B refer to the crest and the toe of the
shallow excavation. The full lines correspond to OCR = 5.5 and the broken lines
to OCR = 11.0. Finally, the signs are in agreement with the axes shown in
Figure 6-2 and, therefore, negative horizontal displacement signifies movement
towards the open face of the excavation, while negative vertical displacement is
associated with downward soil movement.
Horizontal displacements in the range of 1 to 1.5 cm developed at Node 1 (crest

of the deep slope), before the initiation of failure, as shown in Figure 6-29 (a).
The subsequent, abrupt increase of the displacements coincided with the
occurrence of a failure mechanism and with considerable soil movement
towards the face of the excavation. The vertical displacements, illustrated in
Figure 6-29 (b), revealed downward soil movement at failure, while only small
displacements were computed beforehand.
The horizontal displacements at middle of the deep slope (Node 2), shown in
Figure 6-30 (a), exhibited a pattern comparable to that at Node 1. The
magnitude of the horizontal displacements prior to failure was slightly larger
than at the crest. The vertical displacements (Figure 6-30 (b)) signified
downward movement in some of the cases (Analyses 1, 5 and 6 for which the
G.W.T. was initially at -5.0, -7.5 and -10.0 m, respectively, and OCR was 5.5)
and upward movement in others (Analyses 7 and 18 for which OCR was 5.5
and 11.0, respectively, and the G.W.T. was initially at -20.0 m). For three of the
analyses (Analyses 2, 16 and 17 for which the G.W.T. was initially at -5.0, -7.5
and -10.0 m and OCR was 11.0), the direction of displacement changed at the
initiation of failure from upward to downward.
353
Finally, the toe of the deep slope (Figure 6-31) was displaced upwards and
towards the open face of the excavation, as expected. The pre-failure
magnitudes of the horizontal and vertical displacements were larger in
comparison to those calculated at the crest and the middle of the slope. This is
in accordance with the fact that the failure mechanism developed from the toe,
as already discussed. Combined with the displacements at the crest of the deep
excavation, the displacements in the above figure indicate that the failure
mechanism was sliding, as shown in Figure 6-8 (a) and (b) for Analyses 1 and
2, respectively.
Horizontal displacements in the range of 1 cm developed before the initiation of

failure in the shallow excavation at Node A (Figure 6-32 (a)). The pre-failure
vertical displacement at the same node was small (Figure 6-32 (b)). The
displacement at the toe of the shallow slope (Node B) was larger both in the
horizontal (Figure 6-33 (a)) and the vertical direction (Figure 6-33 (b)), indicating
that the mechanism was instigated from the toe, as expected. Furthermore,
upward vertical movement was predicted at Node B (toe) and downward at
Node A (crest), consistent with a sliding failure mechanism.
Larger pre-failure displacements were computed for the deep excavation in

comparison with those yielded by the shallow one. Comparing the
displacements at the toe prior to failure, the horizontal ones were in the range of
4 cm for the deep slope (Figure 6-31 (a)) and in the range of 2 cm for the
shallow slope (Figure 6-33 (a)). The vertical displacements at the toe exhibited
a similar difference, with those in the deep excavation ranging from 6 to 8 cm
(Figure 6-31 (b)) and those in the shallow one being limited to 3 cm (Figure 6-33
(b)).
For all of the nodes considered, with the exception of Node 2, the initial
conditions imposed, both for the deep and the shallow excavations, had only a
small influence of the displacement prior to failure. The values of suction and
OCR affected primarily the factor of safety, 𝐹𝑠 , and, therefore, the occurrence of
failure, and to a lesser extent the displacements associated with the instigation
of failure.
354
Table 6-6: Factors of safety associated with the use of different groundwater table (G.W.T.) depths and
OCR values and achieved for two different excavation depths; 10.0 m (deep) and 4.0 m (shallow)
Excavation depth
Deep (10.0 m) Shallow (4.0 m)
G.W.T R OCR = 5.5 OCR = 11.0 R OCR = 5.5 OCR = 11.0
-5.0 m 0.50 2.76 (An. 1) 3.28 (An. 2) 1.250 4.69 (An. 19) 5.70 (An. 22)
-7.5 m 0.75 3.01 (An. 5) 3.92 (An. 16) 1.875 6.09 (An. 20) 7.84 (An. 23)
-10.0 m 1.00 3.64 (An. 6) 4.66 (An. 17) 2.500 7.42 (An. 21) 9.30 (An. 24)
-20.0 m 2.00 6.52 (An. 7) 8.64 (An. 18) - - -
10
OCR = 11.0
Shallow
8
OCR = 5.5
OCR = 11.0
Deep
6 OCR = 5.5
4 OCR = 11.0, shallow exc.

OCR = 5.5, shallow exc.
OCR = 11.0, deep exc.
2
-20.0 -17.5 -15.0 -12.5 -10.0 -7.5 -5.0
Figure 6-23: Variation of Factor of safety 𝐹𝑠 with the assumed groundwater table (G.W.T.) depth, for
different OCR values and excavation levels
355
10
OCR = 11.0, shallow exc. OCR = 11.0 24
9 OCR = 5.5, shallow exc. 18
8 23
OCR = 1.5, Thanet Sand

7 21
after Georgiadis et al. (2007)
6 22 7 OCR = 5.5
20
5 17
16 19
4
2
6
3
5
1 OCR = 1.5
2
1
0.0 0.5 1.0 1.5 2.0 2.5
Groundwater table/Excavation depth ratio, R ( )
Figure 6-24: Variation of Factor of safety 𝐹𝑠 with the water table/excavation depth ratio R, for OCR
values of 5.5 and 11.0
Figure 6-25: Increase in factor of safety against slope failure for an increase in cohesion (after Fredlund,
1981)
356
Figure 6-26: Variation of the factor of safety of slopes in the BT formation for different groundwater
table (GW) position (after Rahardjo et al., 2010)
Figure 6-27: Variation of the factor of safety of slopes in the JF formation for different groundwater table
(GW) position (after Rahardjo et al., 2010)
900
800 OCR = 11.0

OCR = 5.5
700
stress path
600
500
400
300
200
100
0
0.0 500.0 1000.0 1500.0 2000.0
Figure 6-28: Schematic representation of the yield surfaces corresponding to OCR = 11.0 and OCR =
5.5, for a random isotropic initial stress state
357
Horizontal displacement, Node 1

(crest of the deep excavation)
0.00 (a)
Horizontal displacement, u (m)
-0.01
-0.02
-0.03
-0.04
-0.05
1 2 3 4 5 6 7 8 9
Vertical displacement, Node 1

(crest of the deep excavation)
0.00
(b)
Vertical displacement, v (m)
-0.01
-0.02
-0.03
-0.04
-0.05
1 2 3 4 5 6 7 8 9
OCR = 5.5, G.W.T. 5.00 m (1) OCR = 11.0, G.W.T. 5.00 m (2)
G.W.T. 7.50 m (5) G.W.T. 7.50 m (16)
G.W.T. 10.0 m (6) G.W.T. 10.0 m (17)
G.W.T. 20.0 m (7) G.W.T. 20.0 m (18)
Figure 6-29: Displacement of Node 1 (crest): (a) horizontal and (b) vertical, for excavation depth of -
10.0 m (deep)
358
Horizontal displacement, Node 2

(middle of the deep excavation)
0.00
-0.01
-0.02
-0.03
-0.04
(a)
-0.05
1 2 3 4 5 6 7 8 9
0.03 Vertical displacement, Node 2

(middle of the deep excavation)
0.02
0.01
0.00
-0.01
(b)
-0.02
1 2 3 4 5 6 7 8 9
OCR = 5.5, G.W.T. 5.00 m (1) OCR = 11.0, G.W.T. 5.00 m (2)
G.W.T. 7.50 m (5) G.W.T. 7.50 m (16)
G.W.T. 10.0 m (6) G.W.T. 10.0 m (17)
G.W.T. 20.0 m (7) G.W.T. 20.0 m (18)
Figure 6-30: Displacement of Node 2 (mid-slope): (a) horizontal and (b) vertical, for excavation depth of
-10.0 m (deep)
359
0.02 Horizontal displacement, Node 3

(toe of the deep excavation)
0.00
-0.02
-0.04
-0.06
-0.08
-0.10
-0.12
-0.14
(a)
-0.16
1 2 3 4 5 6 7 8 9
Vertical displacement, Node 3

0.20
(toe of the deep excavation)
0.18
0.16
0.14
0.12
0.10
0.08
0.06
0.04
0.02
(b)
0.00
1 2 3 4 5 6 7 8 9
OCR = 5.5, G.W.T. 5.00 m (1) OCR = 11.0, G.W.T. 5.00 m (2)
G.W.T. 7.50 m (5) G.W.T. 7.50 m (16)
G.W.T. 10.0 m (6) G.W.T. 10.0 m (17)
G.W.T. 20.0 m (7) G.W.T. 20.0 m (18)
Figure 6-31: Displacement of Node 3 (toe): (a) horizontal and (b) vertical, for excavation depth of -10.0
m (deep)
360
0.01 Horizontal displacement, Node A

(crest of the shallow excavation)
(a)
0.00
-0.01
-0.02
-0.03
-0.04
-0.05
1 2 3 4 5 6 7 8 9 10
0.01 Vertical displacement, Node A

(crest of the shallow excavation)
(b)
0.00
-0.01
-0.02
-0.03
-0.04
-0.05
1 2 3 4 5 6 7 8 9 10
OCR = 5.5, G.W.T. 5.00 m (19) OCR = 11.0, G.W.T. 5.00 m (22)
G.W.T. 7.50 m (20) G.W.T. 7.50 m (23)
G.W.T. 10.0 m (21) G.W.T. 10.0 m (24)
Figure 6-32: Displacement of Node A (crest): (a) horizontal and (b) vertical, for excavation depth of -4.0
m (shallow)
361
0.02 Horizontal displacement, Node B

(toe of the shallow excavation)
0.00 (a)
-0.02
-0.04
-0.06
-0.08
-0.10
-0.12
-0.14
-0.16
1 2 3 4 5 6 7 8 9 10
0.20 Vertical displacement, Node B

(toe of the shallow excavation)
0.18
(b)
0.16
0.14
0.12
0.10
0.08
0.06
0.04
0.02
0.00
1 2 3 4 5 6 7 8 9 10
OCR = 5.5, G.W.T. 5.00 m (19) OCR = 11.0, G.W.T. 5.00 m (22)
G.W.T. 7.50 m (20) G.W.T. 7.50 m (23)
G.W.T. 10.0 m (21) G.W.T. 10.0 m (24)
Figure 6-33: Displacement of Node B (toe): (a) horizontal and (b) vertical, for excavation depth of -4.0 m
(shallow)
362
The numerical analysis of a cut slope in a highly overconsolidated unsaturated

soil was studied under various conditions in the current chapter. The Hvorslev
surface was applied and the impact of adequately predicting the shear strength
of such a soil on slope stability was demonstrated.
A 10.0 m deep excavation, with a 2:1 slope was considered. The model
parameters employed in the study were those produced by the calibration of the
Hvorslev surface under unsaturated conditions, in Chapter 4. The excavation
was performed in stages of the coupled consolidation analysis, allowing for
equilibration of the excess pore water pressures (swelling) to concurrently take
place. Once the excavation was completed coupled consolidation was switched
off and drained analysis was thereafter performed. Incremental changes of the
factor of safety, 𝐹𝑠 , were applied until failure occurred.
First, four analyses were performed in order to assess the effect of the yield and
plastic potential surfaces employed on the predicted slope stability; two
employing the Hvorslev (HV) surface on the dry side, attached to the Sinfonietta
Classica (SC) on the wet side, and two adopting the modified Cam-clay (MCC)
surface on both sides. The plastic potential function was controlled by the
parameters 𝛽𝐻𝑉 and 𝑚 when the HV was used and associated plasticity was
assumed on the remaining surfaces. The initial groundwater table (G.W.T.) was
5.0 m deep and two OCR values, 5.5 and 11.0, were considered.
Failure was initiated from the toe for all of the cases examined but for those
employing the MCC surface the available peak stress was enough for
generalised failure to be prevented. Overestimating the peak deviatoric stress,
the MCC surface resulted in a large over-prediction of the slope stability in the
boundary value problem examined. Therefore, it is recommended not to employ
models assuming an elliptical yield surface in the 𝑝-𝐽 plane, centred on the 𝑝
axis, in the analysis of highly overconsolidated soils, without previously
addressing this particular shortcoming. Alternatively, the Hvorslev surface may
be employed.
363
Subsequently, three different types of analyses, namely unsaturated, dry and

effective stress analyses, were compared and the degree of inaccuracy arising
as a result of neglecting partial saturation was investigated. Two OCR values
were considered (5.5 and 11.0) and four initial G.W.T. depths were assumed
(5.0, 7.5, 10.0 and 20.0 m).
Comparing the results of the unsaturated and dry analyses, it was found that
disregard of the soil suction systematically caused under-prediction of the slope
stability. This conclusion was supported by an earlier study by Georgiadis et al.
(2007). The under-prediction was larger for G.W.T. positions deeper than the
excavation depth.
The factors of safety, 𝐹𝑠 , evaluated by the unsaturated and the effective stress
analyses were not significantly different, demonstrating that the effect of suction
on the soil strength was adequately approximated by the effective stress
approach. It was therefore concluded, for the particular drained analyses
examined, where suction changes were not involved, that substituting
Terzaghi’s effective stress for Bishop’s stress with 𝜒 = 𝑆𝑟 , resulted in only a
relatively small error.
If possible, unsaturated analysis should be favoured when dealing with

unsaturated soils, as it is shown to be more appropriate. Nonetheless, in cases
where such an analysis is not feasible, it is preferable, according to the above
study, to follow a conventional, effective stress approach, allowing for negative
pore water pressures above the G.W.T., rather than performing a dry analysis,
as long as drained conditions are ensured throughout.
Finally, a parametric study of the effect of initial conditions and of the effect of
the excavation depth on the stability of the cut slope was presented. A large
number of analyses was involved in the study. The initial conditions varied due
to different assumptions made for the initial position of the G.W.T., which
affected the distribution of pore water pressure with depth. Additionally,
considering two OCR values, the relative position of the initial stress states and
the yield surface was altered. The analyses were performed both for a shallow
excavation of 4.0 m and a deep excavation of 10.0 m.
364
Slope stability improved with the increase of G.W.T. depth and of the ratio R of
the G.W.T. depth over the excavation depth. It was concluded that the depth of
the excavation and the depth of the G.W.T. should be considered concurrently
and not independently, as for slope stability to be improved, suction needs not
necessarily increase in absolute terms but only relative to the excavation depth.
Slope stability improved for increasing values of OCR. This conclusion was
further supported by comparing the results of the current study, for OCR = 5.5
and 11.0, with the data by Georgiadis et al. (2007), for OCR = 1.5.
Finally, it was shown that the values of suction and OCR affected primarily the
factor of safety, 𝐹𝑠 , and, therefore, the occurrence of failure, and to a lesser
extent the displacements associated with the instigation of failure.
Overall, it was demonstrated that numerical analysis of slope stability in highly

overconsolidated unsaturated soils requires employment of a constitutive model
which accounts for the beneficial effect of suction without over-predicting the
soil strength. Employing the Hvorslev surface on the dry side of critical state
and performing an unsaturated analysis was shown to be the most realistic of
the approaches examined. Furthermore, particular emphasis should be given to
the initial conditions within the ground as they were shown to have a significant
effect on the computed factors of safety, 𝐹𝑠 .
365
chapter 7: BEHAVIOUR OF UNSATURATED SOIL
SLOPES UNDER SEASONAL CHANGES OF
SUCTION
7.1 Introduction
The behaviour of unsaturated soil slopes under seasonal changes of suction is

studied in the current chapter, employing the soil-water retention curve (SWRC)
models developed as part of the current research project (Chapter 5). Model 1,
Option 1 of the Georgiadis (2003) constitutive model was used for the
simulation of the mechanical behaviour in conjunction with the Hvorslev surface
presented in Chapter 4. Typically, suctions increase within the ground during
the summer, due to excessive transpiration rates and limited water infiltration,
whereas, during the winter, when precipitation is dominant, the soil is driven
towards saturation. The succession of drying and wetting produces cyclic
volumetric changes, which manifest themselves as shrinkage during the
summer followed by swelling during the winter. The purpose of the study is to
investigate the effect of the newly implemented features on the numerical
prediction of this cyclic volumetric response.
366
Behaviour of Unsaturated Soil Slopes under Seasonal Changes of Suction
Rather than directly prescribing the seasonal suction variation it was indirectly
controlled by the hydraulic boundary conditions (precipitation and vegetation
boundary conditions) which were imposed on the boundary value problem.
Meteorological data with reference to rainfall and potential evapotranspiration
were employed in the coupled consolidation analyses and the performance of
the retention models was assessed primarily in terms of the resulting
displacements.
The specific volume dependent (v-SWRC), the simple hysteretic (hysteretic-

SWRC) and the hysteretic, specific volume dependent (v-hysteretic-SWRC)
soil-water retention curve models were employed, in conjunction with the
degree of saturation dependent soil compressibility with suction, 𝜅𝑠∗ . Various
combinations of the relevant model parameters were considered. The
differences in the formulation of the models influenced the results,
demonstrating the importance of appropriately selecting the retention
relationship in numerical analyses of unsaturated soils.
7.2 Overview of the numerical analyses performed
A total of 17 analyses were performed in order to study the behaviour of

unsaturated soil slopes under seasonal variations of suction. A summary of
these analyses is given in the current section. The various cases considered
are presented in Table 7-1 where the varying parameter is shown in bold.
The soil movements yielded by the simple, Van Genuchten (1980) type, SWRC
expression, presented in Section 3.4, were compared to those yielded by the v-
SWRC model, in order to evaluate the effect of the parameter 𝜓, introduced by
Gallipoli et al (2003b). The parameter 𝜓 was shown in Chapter 5 to control the
shift of the SWRC due to specific volume changes. Therefore, in addition to the
simple expression (Analysis A in Table 7-1), for which essentially 𝜓 = 0.0, three
more cases were examined; 𝜓 = 1.0 (Analysis B.1 in Table 7-1); 𝜓 = 2.0
(Analysis B.2 in Table 7-1) and 𝜓 = 5.0 (Analysis B.3 in Table 7-1). For all of the
analyses, the soil compressibility with suction, 𝜅𝑠∗ , was assumed to be
367
independent of the degree of saturation, 𝑆𝑟 , and, therefore, 𝜒 = 1.0 and 𝜔 = 0.0

(see Section 5.5).
The effect of the soil compressibility with suction on the soil movements was
studied in association with the hysteretic-SWRC model. The parameter 𝜒, to
which the coefficient of compressibility 𝜅𝑠 is raised (Equation 5.34), was set
equal to 1.0. The parameter 𝜔, which controls the impact of the current degree
of saturation, 𝑆𝑟 , on the soil compressibility 𝜅𝑠∗ , was set equal to 0.0, 2.0 and
4.0 in the analyses termed C.1, C.2 and C.3, respectively, in Table 7-1.
The 3-dimensional v-hysteretic-SWRC model was subsequently adopted and

the effect of both parameters 𝜓 and 𝜔 was investigated. Analyses D.1, D.2 and
D.3 (Table 7-1) were performed for 𝜒 = 1.0 and 𝜔 = 2.0, while 𝜓 obtained the
values 1.0, 2.0 and 3.0, respectively. The case for which 𝜓 = 3.0 (Analysis D.3)
was repeated assuming 𝜔 = 4.0 , in the analysis named D.4 in Table 7-1.
Finally, Analysis D.1 was extended to 15 years (D.1.15yr in the same table), by
replicating the available meteorological data.
In the above analyses the same initial conditions, derived from the coefficient of
earth pressure at rest, 𝐾0 , (equal to 1.0) and the overconsolidation ratio, OCR,
(equal to 5.5), were assumed. In order to account for the effect of 𝐾0 and OCR,
Analysis D.1 was repeated but with the values of 𝐾0 and OCR presented in
Table 7-1 (Analyses D.1.K0A, D.1.K0B and D.1.OCR).
Furthermore, a constant maximum root depth, 𝑟𝑚𝑎𝑥 , equal to 2.0 m, was

prescribed for the application of the vegetation boundary condition.
Nonetheless, vegetation may take years to establish itself. Therefore, Analysis
D.1 was repeated for a root depth 𝑟𝑚𝑎𝑥 increasing by 0.5 m/year (D.1.root in
Table 7-1).
Finally, permeability is known to vary with suction and consequently with the
degree of saturation. Although the former relationship is believed to be
hysteretic, the latter was shown to be monotonic, as discussed by Fredlund &
Rahardjo (1993). Two models are available in ICFEP, regarding the decrease of
permeability due to de-saturation and are both monotonic: one adopting a
linear, suction dependent variation of the logarithm of the coefficient 𝑘 and a
368
second one employing a non-linear, degree of saturation dependent variation of

the coefficient 𝑘. Preferably the second model should be used in combination
with the hysteretic models for the SWRC. Nevertheless, the nonlinearity
introduced by the relevant expression (Equation 3.88, in Chapter 3), increased
the computational time required for the herein presented finite element analyses
dramatically. For this reason, the linear variation of log 𝑘 with suction was
instead employed in all of the analyses summarised so far. In order to assess
the inaccuracy instigated by this simplification, Analysis D.1 was once more
repeated, adopting the 𝑆𝑟 dependent variable permeability model and the
corresponding analysis is referred to in Table 7-1 as D.1.perm.
Table 7-1: Analyses undertaken for the investigation of the effect of seasonal changes of suction on
unsaturated soil movement
Analyses performed
Analysis SWRC model 𝜔 𝜓 Other Variables
A simple 0.0 NA 𝐾0 = 1.0; OCR = 5.5
B.1 v-SWRC 0.0 1.0 𝐾0 = 1.0; OCR = 5.5
B.2 v-SWRC 0.0 2.0 𝐾0 = 1.0; OCR = 5.5
B.3 v-SWRC 0.0 5.0 𝐾0 = 1.0; OCR = 5.5
C.1 hysteretic 0.0 NA 𝐾0 = 1.0; OCR = 5.5
D.1 v-hysteretic 2.0 1.0 𝐾0 = 1.0; OCR = 5.5
D.1.15yr v-hysteretic 2.0 1.0 15 year-long analysis
D.1.K0A v-hysteretic 2.0 1.0 𝑲𝟎 = 0.75; OCR = 5.5
D.1.K0B v-hysteretic 2.0 1.0 𝑲𝟎 = 1.25; OCR = 5.5
D.1.OCR v-hysteretic 2.0 1.0 𝐾0 = 1.0; OCR = 1.5
D.1.root v-hysteretic 2.0 1.0 Roots increasing 0.5m/yr
D.1.perm v-hysteretic 2.0 1.0 𝑆𝑟 dependent permeability
369
7.3 Problem description
The 10.0 m deep excavation employed in the previous chapter and illustrated in
Figure 6-1, was also considered in the current chapter. The excavation was
50.0 m wide at the original ground surface and 30.0 m wide at its bottom. For
the finite element analysis, a domain of 30.0 m depth and 100.0 m width was
adopted and was discretised as shown in Figure 6-2, employing isoparametric,
quadrilateral 8-noded solid elements. The origin of the x and y axes is shown in
the same figure to be at the top left corner of the discretised domain. The out-of-
plane direction is of no interest, as plane strain conditions were applied.
The excavation was performed in the silty soil tested by Estabragh & Javadi
(2008) (Chapter 4) and the material properties were presented in Section 4.5.2.
The Georgiadis (2003) unsaturated soil constitutive model (Model 1; option 1)
was employed, adopting the newly developed Hvorslev (HV) surface on the dry
side, attached to the Sinfonietta Classica (SC) surface on the wet side. The
apparent cohesion due to suction was coupled to the SWRC. The model
parameters are those produced by the calibration of the Hvorslev surface under
unsaturated conditions (Section 4.5.2). They are repeated in Table 7-2 for
clarity, together with the potential transpiration parameters S1, S2, S3 and S4,
which control the 𝛼 variation for the calculation of the actual transpiration
(Section 3.6.2).
The SWRC model parameters adopted in the analyses are summarised in

Table 7-3. For the simple, Van Genuchten (1980) type model they are identical
to those employed in the slope stability study, presented in Chapter 6. The
same parameters were used for the v-SWRC model, with the exception of the
parameter 𝜓 which was varied as already described. The parameters assumed
for the two hysteretic models were chosen so that the primary curves generated
bounded the Van Genuchten (1980) curve, as explained in the subsequent
section. The air-entry value of suction, 𝑠𝑎𝑖𝑟 , was 0.0 kPa and therefore the
equivalent suction 𝑠𝑒𝑞 was equal to the matric suction 𝑠.
Also shown in Table 7-3, are the parameters for the variable permeability model
employed, which was a combination of the de-saturation and the desiccation
370
models presented in Section 3.5. The model parameters for the degree of
saturation dependent permeability are presented in Section 7.11. The
coefficient of saturated soil permeability, 𝑘, was assumed to be 𝑘𝑠𝑎𝑡 = 10−8 m/s.
The soil stresses were initialised at the commencement of the coupled

consolidation unsaturated analysis (Increment 0), employing a unit weight 𝛾𝑑𝑟𝑦
= 𝛾𝑠𝑎𝑡 of 19.1 kN/m3. The coefficient of earth pressure at rest, 𝐾0 , was equal to
1.0. The initial pore water pressure distribution with depth was hydrostatic,
resulting from a 5.0 m deep groundwater table (G.W.T.), above which suctions
were allowed to develop. Considering a value of 5.5 for the overconsolidation
ratio (OCR), the hardening/softening parameters 𝑝0∗ and 𝑝0′ , for partly and fully
saturated conditions, respectively, were then initialised for each Gauss point in
the finite element mesh. For the elements under partial saturation, the
hardening parameter, 𝑝0∗ , was directly calculated from the corresponding total
stress (Equation 3.74), and the isotropic yield stress, 𝑝0 , at the current value of
suction was, subsequently, computed from Equation 3.52.
Starting from a horizontal ground surface, the 10.0 m deep excavation was
performed in 20 stages, in each of which a 0.5 m thick layer of elements was
removed. Each excavation phase was executed in 10 increments of the coupled
consolidation analysis, allowing the initial hydrostatic pore water pressure
regime to be altered.
No change of pore pressure (∆𝑝𝑓𝑏 = 0.0) was prescribed at the right vertical
boundary of the finite element mesh and the no flow condition (𝑞𝑛 = 0.0) was
applied on the remaining boundaries. When excavating below the water table,
the precipitation boundary condition was applied to the newly formed boundary
of the slope, prescribing a 0.0 flow rate, 𝑞𝑛 , or 0.0 a pore pressure, 𝑝𝑓𝑏 . At the
bottom of the excavation a 0.0 pore pressure condition, 𝑝𝑓𝑏 , was applied, in
order to accelerate the dissipation (swelling) of the negative excess pore
pressures generated due to the excavation. The above conditions on the face
and on the bottom of the cut are applied at each excavation phase below the
G.W.T., until the final excavation level is reached. Each stage was performed in
a one month time-step. At the end of the excavation the G.W.T. reached the
position illustrated in Figure 6-3.
371
Assuming the initial position of the G.W.T. to be at -5.0 m and employing an

OCR = 5.5, Analysis 1 of Chapter 6, which had yielded a factor of safety of 2.76
at the end of the excavation, essentially formed the basis for the current
parametric study.
Once the excavation was completed, the vegetation and the precipitation
boundary conditions, which were detailed in Section 3.6, were employed for the
numerical reproduction of the combined effect of evapotranspiration and rainfall.
The automatic incrementation algorithm described in Chapter 3 was used in the
coupled consolidation analyses, in combination with the precipitation boundary
condition. The potential evapotranspiration and precipitation data, summarised
in Table 7-4, were applied on the original ground surface and on the newly
formed inclined boundary of the slope, excluding the bottom of the excavation,
where zero change of pore pressure was prescribed ( ∆𝑝𝑓𝑏 = 0.0 ). The
meteorological data, obtained from the Building Research Establishment testing
site at Chattenden, Kent, UK, refer to years 1970 to 1974 and were formerly
employed in finite element analyses of railway embankments by Nyambayo
(2003). They were applied on monthly basis and each month was simulated
over 12 increments of the coupled consolidation analysis.
372
Table 7-2: Constitutive model and transpiration properties employed in the study
Constitutive model and transpiration properties
Transpiration
Unsaturated soil model properties
properties
Parameter Value Parameter Value Parameter Value

𝛼𝑔,𝑓 0.7 𝜆(0) 0.086 𝑆1 (kPa) 0.086
𝜇𝑔,𝑓 0.9999 𝜅 0.005 𝑆2 (kPa) 0.005
𝛼𝐻𝑉 0.45 𝑣1 2.120 𝑆3 (kPa) 2.12
𝑛 0.5 𝛽 0.001 𝑆4 (kPa) 15.0
𝛽𝐻𝑉 0.25 𝑝𝑎𝑡𝑚 (kPa) 100.0
𝑚 0.5 𝐺 (kPa) 15.0
𝑀𝑔,𝑓 1.3039 𝜆𝑠 0.08
𝑟 0.06 𝜅𝑠 0.03
𝑘 SWRC 𝑝𝑐 (kPa) 1.0
𝑠𝑎𝑖𝑟 (kPa) 0.0 𝑠0 (kPa) 1000000.0
Table 7-3: SWRC and variable permeability model parameters employed in the study
SWRC and variable permeability model parameters
Hysteretic and v- Variable permeability

Simple & v-SWRC
hysteretic-SWRC model

-8
𝑠𝑑𝑒𝑠 (kPa) 0.0 𝑠𝑎𝑖𝑟 (kPa) 0.0 𝑘𝑠𝑎𝑡 (m/s) 10
𝑠𝑎𝑖𝑟 (kPa) 0.0 𝑠0 (kPa) 100000.0 𝑠1 (kPa) 0.0
𝑠0 (kPa) 100000.0 𝛼𝑑 0.008 𝑠2 (kPa) 1000.0
𝑆𝑟,0 0.1 𝛼𝑤 0.05 𝑘𝑠𝑎𝑡 /𝑘𝑚𝑖𝑛 100.0
𝛼 0.015 - - 𝜎𝛵1 (kPa) 0.0
𝑛 0.95 𝜓 varying 𝜎𝛵2 (kPa) 100.0
𝑚 0.7 𝜒 1.0
𝑘𝑠𝑎𝑡 /
100.0
- - 𝜔 varying /𝑘𝑚𝑎𝑥
373
Table 7-4: Potential Evapotranspiration and precipitation data, Building Research Establishment testing
site Chattenden, Kent (U.K.)
Analyses performed
Pot. Evapo- Pot. Evapo-
Month & Precipitation Month & Precipitation
transpiration transpiration
Year (mm/month) Year (mm/month)
(mm/month) (mm/month)
Jan-70 4.21 57.75 Jul-72 88.12 25.91
Feb-70 18.25 35.21 Aug-72 67.10 2.17
Mar-70 22.46 50.70 Sep-72 48.67 35.91
Apr-70 44.91 39.44 Oct-72 31.56 0.00
May-70 80.00 7.04 Nov-72 17.07 31.54
Jun-70 105.26 0.00 Dec-72 1.26 52.20
Jul-70 103.86 29.58 Jan-73 2.46 11.45
Aug-70 85.61 22.54 Feb-73 8.88 12.51
Sep-70 60.35 45.07 Mar-73 16.60 16.20
Oct-70 44.91 4.23 Apr-73 38.71 42.12
Nov-70 26.67 126.76 May-73 79.11 24.89
Dec-70 7.02 47.89 Jun-73 88.14 61.26
Jan-71 4.21 42.25 Jul-73 86.72 19.19
Feb-71 8.42 11.27 Aug-73 78.75 0.00

Mar-71 23.86 39.44 Sep-73 55.12 97.15
Apr-71 36.49 28.17 Oct-73 28.84 28.95
May-71 64.56 36.62 Nov-73 14.36 11.71
Jun-71 80.00 104.23 Dec-73 3.79 28.48
Jul-71 80.00 5.63 Jan-74 12.37 49.59
Aug-71 82.81 33.80 Feb-74 11.02 75.46
Sep-71 43.51 4.23 Mar-74 19.61 36.73

Oct-71 35.09 35.21 Apr-74 49.31 14.15
Nov-71 18.25 39.44 May-74 75.30 5.25
Dec-71 8.42 25.35 Jun-74 82.64 28.64
Jan-72 5.16 65.25 Jul-74 101.18 22.23
Feb-72 5.06 33.63 Aug-74 77.46 48.11
Mar-72 18.00 37.32 Sep-74 62.45 85.16
Apr-72 46.65 33.17 Oct-74 38.74 43.96
May-72 88.36 10.70 Nov-74 16.28 79.78
Jun-72 77.78 9.18 Dec-74 13.69 27.38
374
7.4 Initial degree of saturation
At the commencement of an unsaturated finite element analysis, the initial

degree of saturation, 𝑆𝑟 , at each Gauss point under partial saturation is
required, so that the increase of apparent cohesion with suction, controlled by
the product 𝑠𝑒𝑞 ∙ 𝑆𝑟 , can be computed. For the case of a monotonic SWRC, this
can be calculated by the expression adopted, since a unique degree of
saturation corresponds to each suction level. On the contrary, for a hysteretic
SWRC, the initial value of 𝑆𝑟 needs to be prescribed for each integration point.
In the cases where the hysteretic-SWRC model was employed (Analyses C.1,
C.2 and C.3), the initial hydraulic points were assumed to lie on the primary
drying curve. This was fitted as closely as possible to the simple Van
Genuchten (1980) curve adopted in Analysis A, by setting the model parameter
𝛼𝑑 equal to 0.008. Nonetheless, due to the difference in the expressions on
which the two models are based, the fitting was limited to suction levels from
0.0 to 100.0 kPa (Figure 7-1). Given the lack of information regarding the
retention behaviour of the silty soil considered, the wetting curve was assumed
to be defined by the value 0.05 specified for the parameter 𝛼𝑤 . Analyses C.1,
C.2 and C.3 shared a common SWRC (specific volume independent) and
common initial retention points.
Complications in the initiation of the degree of saturation arose when the v-

hysteretic-SWRC model was employed (Analyses D.1, D.2, D.3 and D.4),
caused by the introduction of the specific volume, 𝑣 , in the retention
relationship. Prescribing the degree of saturation to lie on the primary drying
surface is impractical in a boundary value problem, as prior knowledge of the
specific volume is required for every Gauss point in unsaturated conditions.
Nevertheless, the specific volume is automatically adjusted by the program to
be consistent with the input parameters, as explained in Section 3.3.9.
Alternatively, the same magnitudes of 𝑆𝑟 as for the hysteretic-SWRC model
were assumed and the initial retention points did not lie on the primary surfaces
(Figure 7-2). The parameters 𝛼𝑑 and 𝛼𝑤 were set equal to those employed by
the hysteretic-SWRC model.
375
Figure 7-2 illustrates the position of the input initial retention points relative to
the projection of the primary curves in the 𝑠-𝑆𝑟 plane for Analyses D.1, D.2, D.3
and D.4. For the above illustration to be feasible an initial void ratio of 0.7 was
assumed. This value is believed to adequately approximate the average void
ratio which is expected to be initialised for the unsaturated elements of the
mesh. As discussed in Chapter 5, the projection of the primary curves shifted to
the right for higher values of the parameter 𝜓. For 𝜓 = 3.0 (Analyses D.3 and
D.4) the primary wetting path is shown to lie just below the initial retention
points. The relative displacement of the primary surfaces in the 𝑠-𝑆𝑟 plane is
expected to affect the soil movements within the vegetated slope, despite the
initial conditions being identical.
As discussed above, for the case of a monotonic SWRC the initial degree of
saturation needs not be prescribed but is computed by the retention relationship
itself. As the position of the SWRC – illustrated in Figure 7-3 assuming that the
initial void ratio was equal to 0.7 – depends on the parameter 𝜓, Analyses A
and B.1, B.2 and B.3 did not share common initial points.
1.0
0.8
0.6
0.4
Primary Drying
Primary wetting
0.2 Van Genuchten (1980)
Initial hydraulic points
0.0
0.1 1 10 100 1000 10000 100000
Figure 7-1: Comparison of the primary drying and wetting paths generated by the hysteretic-SWRC-
model with the SWRC reproduced by the simple, Van Genuchten (1980) type expression
376
1.0
0.8
0.6
0.4 Prim. Curves D.1

Prim. Curves D.2
Prim. Curves D.3
0.2
Prim. Curves D.4
Initial retention points
0.0
1 10 100 1000 10000 100000
Figure 7-2: Primary curves reproduced by the v-hysteretic-SWRC model, assuming initial void ratio
equal to 0.7
1.0
0.8
0.6
0.4
Analysis A
Analysis B.1
0.2
Analysis B.2
Analysis B.3
0.0
1 10 100 1000 10000 100000
Figure 7-3: SWRC’s reproduced by the simple, Van Genuchten (1980) type model and the v-SWRC
model, assuming initial void ratio equal to 0.7
7.5 Seasonal suction variation
According to the literature, a seasonal variation of the pore pressure distribution

with depth and a seasonal variation of the depth of the groundwater table
(G.W.T.) are associated with the cyclic succession of high precipitation during
the winter and high demand for transpiration during the summer.
377
Infiltration of rainfall water into a soil slope (Pak Kong, Hong Kong) using ICFEP
was originally studied by Smith (2003), who, for this purpose, implemented the
precipitation boundary condition, presented in Chapter 3. When rainfall
infiltrates an unsaturated soil slope, the suctions above the phreatic surface are
reduced, decreasing the available shear strength. Furthermore, as water flows
downwards, the groundwater table may rise, causing instability. Nonetheless,
as noted by Smith (2003), failure may be directly induced by infiltration, without
requiring a rise of the G.W.T. To explain the processes involved, Smith (2003)
utilised the vertical moisture distribution profile, presented by Bear (1972) and
shown in Figure 7-4. The ground may be divided into the zone of saturation,
which lies below the groundwater table and the zone of aeration, which lies
above. The latter can be further sub-divided into three zones: capillary fringe,
intermediate (vadoze) zone and soil water zone.
The capillary fringe is characterised by tensile pore pressures (suction) and high
degrees of saturation, 𝑆𝑟 , in excess of 75%, according to Bear (1972) and at a
minimum in the range of 85-90%, estimated by Fredlund & Rahardjo (1993).
Significant groundwater flow may, therefore, occur within this zone. Transient
gravitational flow may take place within the intermediate zone but it is within the
soil water zone where the water content varies considerably in response to
rainfall and plant uptake and can range between full saturation and air-dried
conditions. Under infiltration the soil becomes fully saturated over a very small
depth from the surface, which might be as shallow as 1.5 cm (Bear, 1972).
Below this zone, a degree of saturation, 𝑆𝑟 , of 80-90% is maintained until the
‘wetting front’ is reached, where 𝑆𝑟 drops sharply to its original value (Lumb,
1962, reported in Smith, 2003).
Further to infiltration of rainfall water, the evapotranspiration effects of

vegetation may cause significant changes of suction within the ground. For the
numerical simulation of these effects, Nyambayo (2003) implemented the water-
uptake model and the vegetation boundary condition presented in Chapter 3.
He explains that during the summer, the amount of water absorbed by the roots
exceeds that supplied by rainfall and the soil is gradually dried out from the top
downwards. During the winter, due to lower transpiration rates, the soil recovers
and in most situations the influence of the trees is entirely seasonal (Shaw,
378
1994 reported in Nyambayo, 2003). Nonetheless, full recovery does not always
occur and, in certain cases, the soil strata are left permanently desiccated.
The predicted variation of suction distribution with depth and the resulting
generalised ground movements are presented and explained in the current
section for one of the analysis performed, namely D.1, which is thought to be
representative of the majority of analyses undertaken. Similar behaviour was
exhibited in the remaining analyses, with the suctions reducing during the
winter, inducing soil swelling, and increasing during the summer, generating soil
shrinkage, and, therefore, the respective outcomes are omitted for reasons of
brevity. On the contrary, the results of these analyses are studied in terms of
displacements evaluated at single points on the crest and the middle of the
slope.
The estimated positions of the G.W.T. are shown in Figure 7-5 at the end of the
two seasons – summer and winter – for the years 1970 to 1974. Starting from
the end of August 1970, the first winter period to be taken into consideration
lasted till the end of February 1971. The position of the G.W.T. at the end of
February 1970 is illustrated in the figure only for comparison, since it is the
result of the accumulated effect of two months, since January 1970, when the
available data start, rather than six. Nonetheless, the G.W.T. is shown to have
acquired a position similar to those corresponding to February 1972, 1973 and
1974 and is, therefore, thought to be realistic.
The G.W.T. at the end of August 1970 is illustrated in Figure 7-5 (a) to have
dropped around the face of the slope and, subsequently, to rise and to drop
during the following winter and summer, respectively. A similar variation was
predicted for the remaining years, shown in Figure 7-5 (b) and (c). The changes
were concentrated behind the open cut and diminished further away, as the
maximum root depth, assumed to be 2.0 m, was not sufficient in order to
extensively affect the depth of the G.W.T., modelled at -5.0 m beneath the
horizontal ground surface.
Furthermore, the pore pressures above the G.W.T. also varied seasonally
under the combined effect of vegetation and precipitation. Figure 7-6 shows the
typical winter (February) and summer (August) pore water pressure evolution
379
with depth, 25.0 m behind the crest of the slope, calculated for the Analysis D.1.
Suctions lower than the hydrostatic developed during the winter and higher than
the hydrostatic during the summer, while the discrepancy was concentrated in
the upper 5.0 m of the ground.
The seasonal suction variation described above resulted in generalised swelling

during the winter and generalised shrinkage during the summer, as indicated by
the vectors of incremental displacements shown in Figure 7-7 for February and
August. The largest ground movements in both seasons occurred closer to the
surface and to the open cut, where confinement was limited. Horizontal
displacements were observed behind the slope but were suppressed further
away. Moreover, along the face of the slope, displacements increased gradually
from the toe to the crest.
Figure 7-8 (a) illustrates the horizontal displacements along a vertical line
starting from the crest of the slope. The horizontal displacements along a similar
vertical line initiating from the middle of the slope are shown in Figure 7-8 (b).
The displacements were sub-accumulated from the end of the excavation. The
maximum absolute horizontal displacement at the crest did not exceed 0.7 cm
(end of August 1973) and the one evaluated at mid-slope was about 1.25 cm for
the same month.
Displacements in Figure 7-8 (a) were concentrated above the G.W.T. and were
limited below but without being zero, in agreement with the generalised
horizontal movement of the face of the excavation. During the summer (end of
August) positive relative displacements, in relation to the past February, were
predicted whereas during the winter (end of February) negative relative
displacements were calculated, both at the crest and at mid-slope. It should be
noted that negative signifies horizontal displacement towards the face of the
slope (swelling), according to the axis convention shown in Figure 6-2.
Comparable results were yielded in terms of vertical displacements measured

along a horizontal line starting from the crest (ground surface), shown in Figure
7-9 (a), and along a horizontal line initiating from the middle of the slope (at the
original G.W.T. depth), illustrated in Figure 7-9 (b). The displacements were
sub-accumulated from the end of the excavation. The maximum absolute
380
displacement at the crest of the slope was in the range of 2.5 cm while it was
less than 1.5 cm at mid-slope, demonstrating a noticeable decrease with depth.
Furthermore, the majority of vertical ground movements extended up to 20.0 m

behind the open cut. Further than this, displacements were still significant at the
ground surface but were almost negligible at 5.0 m depth. During the summer
(end of August) negative relative displacements, in relation to the past
February, were computed whereas during the winter (end of February) positive
relative displacements were calculated, both at the crest and mid-slope. It
should be noted that negative signifies downwards vertical displacement
(settlement resulting from shrinkage), according to the axis convention shown in
Figure 6-2.
Combining the observations made so far, it can be concluded that shrinkage

was predicted during the summer period, followed by swelling during the winter.
Moreover, the ground movements were more pronounced around the open face
of the excavation and at shallow depths at the ground surface, where conditions
were less confined and directly exposed to the applied boundary conditions.
Figure 7-4: Distribution of subsurface water, according to Bear (1972) (after Smith, 2003)
381
Figure 7-5: Seasonal variation of the groundwater table (G.W.T.) predicted for years 1970 to 1974;
Analysis D.1 (v-hysteretic-SWRC model, 𝜓 = 1.0)
382
Pore water pressure (kPa)

-100 -50 0 50
0.000
5.000
10.000
August
Depth (m)
February
Hydrostatic
15.000
Figure 7-6: Typical summer (August) and winter (February) pore water pressure distribution with depth
for Analysis D.1 (v-hysteretic-SWRC model, 𝜓 = 1.0)
Figure 7-7: Vectors of incremental displacements showing (a) typical swelling during the winter and (b)
typical shrinkage during the summer for Analysis D.1 (v-hysteretic-SWRC model, 𝜓 = 1.0)
383
3. Aug. '71
1. Aug. '70
5. Aug. '72
9. Aug. '74
7. Aug. '73
2. Feb. '71
4. Feb. '72
6. Feb. '73
8. Feb. '74
0.000
2 4 3 1 5 9 7
Vertical distance from the crest (m)

6 8
2.000
4.000
6.000
August
February
8.000
(a)
10.000
-0.01 0 0.01
Horizontal displacement (m)
3. Aug. '71
1. Aug. '70
5. Aug. '72
9. Aug. '74
7. Aug. '73
2. Feb. '71
4. Feb. '72
8. Feb. '74
6. Feb. '73
0.000
9 7
Vertical distance from mid-slope (m)
2 4 3 1
2.000
658
4.000
6.000
August
February
8.000
(b)
10.000
-0.01 0 0.01 0.02
Horizontal displacments (m)
Figure 7-8: Horizontal displacements along vertical lines starting at the (a) crest and at (b) mid-slope;
384
0.010
2. Feb. '71 2
4. Feb. '72 4
0.000
Vertical displacement (m)

3. Aug. '71 3
8. Feb. '74 8
6
6. Feb. '73
1
1. Aug. '70 -0.010
5
5. Aug. '72
-0.020
9
9. Aug. '74 7 August
(a)
7. Aug. '73 February
-0.030
0 10 20 30 40 50
Horizontal distance from the crest (m)
0.005
4
2
4. Feb. '72 3
2. Feb. '71
3. Aug. '71 0.000

6. Feb. '73
8. Feb. '74 6
8
5 -0.005
5. Aug. '72
1
1. Aug. '70
7 -0.010
7. Aug. '73
August
(b)
9. Aug. '74 9 February
-0.015
0 10 20 30 40 50
Horizontal distance from mid-slope (m)
Figure 7-9: Vertical displacements along horizontal lines starting at the (a) crest and at (b) mid-slope;
385
7.6 Monotonic SWRC models
The seasonal changes of suction presented in the previous section were shown
to induce generalised soil movements, resulting from successive swelling and
shrinkage. The horizontal and vertical displacements at the crest and the middle
of the slope, calculated when employing different monotonic SWRC’s, are
herein compared, with the intention of studying the effect of the parameter 𝜓,
initially introduced in the retention relationship by Gallipoli et al. (2003b) (see
also Chapter 5). For this reason, along with the simple, Van Genuchten (1980)
type SWRC (Analysis A), for which essentially 𝜓 = 0.0 , the analysis was
performed adopting the v-SWRC model, explained in Section 5.2, for 𝜓 values
of 1.0, 2.0 and 5.0 (Analyses B.1, B.2 and B.3, respectively).
The horizontal displacement of the crest is shown in Figure 7-10 (a) to increase
almost monotonically with time for Analyses A and B.1 and B.2, while the
positive sign signifies soil shrinkage. For analysis B.3, swelling was indicated
during the winter (negative change of displacement) and shrinkage during the
summer (positive change of displacement), with the latter being larger in
absolute magnitude than the former, leading to accumulation of positive
displacement, which reached 5 cm at the end of December 1974.
Successive swelling and shrinkage was evident from the vertical displacement
of the crest, not only for B.3 but also for the remaining analyses, as illustrated in
Figure 7-10 (b). Once more, shrinkage dominated the behaviour, as negative
vertical displacement (settlement) accumulated with time and for the case of
𝜓 = 5.0, exceeded the value of 12 cm in August 1974.
A similar pattern is observed in Figure 7-11, illustrating the horizontal and

vertical displacements calculated at the middle of the slope. Although the
horizontal displacement was comparable to the one evaluated at the crest, the
vertical displacement is smaller, for all of the cases examined. The
displacements at the toe (Figure 7-12) were insignificant with the exception of
those computed for Analysis B.3, which, however, only just exceeded 2 cm in
the horizontal direction during the summer months.
386
Evidently, larger displacements were systematically predicted for higher 𝜓

values. For increasing values of this parameter the G.W.T. was further
depressed, under the influence of the applied boundary conditions, as shown in
Figure 7-13 for August 1974. The trend exhibited in the latter figure justifies the
tendency of the accumulated displacements to increase with 𝜓, as an amplified
seasonal variation of the G.W.T. is naturally associated with larger soil
movements.
The position of the G.W.T. is thought to be linked to the parameter 𝜓 through

the resulting shapes of the SWRC (see also Figure 7-3). For increasing values
of this parameter, nearly saturated conditions can be sustained up to larger
suction levels. Indeed, in Figure 7-14, which illustrates contours of degree of
saturation, 𝑆𝑟 , corresponding to 99%, 90% and 75%, the area characterised by
partial saturation of less than 75% reduced with increasing 𝜓 , since the
corresponding contour progressed upwards. Degrees of saturation in the range
between 99% and 90% were retained for the largest part of the unsaturated soil
mass in Figure 7-14 (d), where 𝜓 = 5.0. Finally, the downward advancement of
the 99% saturation front observed, which is in accordance with the depression
of the G.W.T., is coupled with the intensified seepage resulting from the
increased values of 𝑆𝑟 .
It should, however, be noted that the above explanation is somewhat simplified

and neglects important factors, such as the soil permeability and the actual
transpiration, which were both functions of suction. Furthermore, distinction
between the two possible modes composing the precipitation boundary
condition is pore pressure dependent. Nevertheless, such simplification is
inevitable, as it is unfeasible to directly quantify the effect of each one of these
factors on the predicted results.
387
Crest displacement
0.06
0.04
0.02
0.00
(a)
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.04
(b)
0.00
-0.04
-0.08
-0.12
-0.16
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Time (months) Dec 1974
Anal. A Anal. B.1 Anal. B.2 Anal. B.3

 = 0.0  = 1.0  = 2.0  = 5.0
Figure 7-10: Crest displacements for analyses A(simple SWRC); B.1; B.2 and B.3 (v-SWRC): (a)
horizontal and; (b) vertical
388
Mid-slope displacement
0.06
0.04
0.02
0.00
(a)
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.02
(b)
0.00
-0.02
-0.04
-0.06
-0.08
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973

 = 0.0  = 1.0  = 2.0  = 5.0
Figure 7-11: Mid-slope displacements for analyses A(simple SWRC); B.1; B.2 and B.3 (v-SWRC): (a)
389
Toe displacement
0.03
0.02
0.01
0.00
(a)
-0.01
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.01
(b)
0.00
-0.01
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973

 = 0.0  = 1.0  = 2.0  = 5.0
Figure 7-12: Toe displacements for analyses A(simple SWRC); B.1; B.2 and B.3 (v-SWRC): (a)
Figure 7-13: Groundwater Table (G.W.T.) positions at the end of August 1974 for Analyses A, B.1, B.2
and B.3
390
Figure 7-14: Contours of degree of saturation, 𝑆𝑟 , evaluated at the end of August 1974 for Analyses A,
B.1, B.2 and B.3
391
7.7 Hysteretic, specific volume independent SWRC model (hysteretic-

SWRC)
The results produced by the simple, Van Genuchten (1980) type SWRC
(Analysis A) were compared to those obtained by the hysteretic-SWRC model
(Analysis C.1). Two extra analyses were performed employing the latter model,
named C.2 and C.3, where the soil compressibility with suction, 𝜅𝑠∗ , was
assumed to depend on the degree of saturation, 𝑆𝑟 , according to Equation 5.34
proposed in Section 5.5. The parameter 𝜒 remained unchanged and equal to
1.0 while the parameter 𝜔 assumed the values 2.0 and 4.0, as explained in
Table 7-1.
When the hysteretic model was utilised, the volumetric behaviour was shown to
be less accumulative in comparison to Analysis A, as indicated by the
displacements of the crest and the middle of the slope, summarised in Figure
7-15 and in Figure 7-16, respectively. There was a better balance between
swelling and shrinkage and at the end of the 5th year there was hardly any
displacement accumulated at the crest for Analysis C.1, as shown in Figure
7-15 (a) and (b).
Nonetheless, the difference in displacements between the summer and the

winter for the latter analysis was larger and better pronounced; the crest was
horizontally displaced for Analysis C.1 by approximately 2 cm towards the open
face of the excavation (swelling), during the winter period from end of August
1970 to end of February 1971, and by 2 cm in the opposite direction
(shrinkage), during the summer period from February 1972 to August 1972,
whereas 2 cm of displacement were accumulated throughout the 5 years of
Analysis A (Figure 7-15 (a)). Even larger relative vertical displacements were
computed at the crest for Analysis C.1 (Figure 7-15 (b)).
Although a small horizontal displacement was predicted at the crest from

January to February 1970 for Analysis A, the hysteretic model generated 1.5 cm
of negative horizontal displacement (swelling), for Analysis C.1, as shown in
Figure 7-15 (a). The vertical displacement over the same period was double for
the latter case in comparison to the former (Figure 7-15 (b)), leading to the
392
conclusion that the hysteretic-SWRC model is capable of reproducing the

anticipated volumetric behaviour right from the start of the analysis.
A comparable trend was exhibited by the displacements at the middle of the

slope (Figure 7-16), although the difference in the computed results between
Analysis A and C.1 was generally smaller and for the case of the vertical
displacements, the corresponding curves are shown to be almost parallel, with
larger settlement being predicted at the end of December 1974 for Analysis A.
In order to investigate the effect of the parameter 𝜔 on the volumetric behaviour

predicted, the results of Analyses C.1, C.2 and C.3 were compared. Introduction
of a non-linear, degree of saturation dependent, soil compressibility with suction
affected the relative displacements mostly at the crest (Figure 7-15) and to a
lesser extend at the middle of the slope (Figure 7-16). The distance between
the highs and lows in the curves presented in these two figures reduced and
smaller swelling and shrinkage was generated for increasing values of 𝜔.
Since identical primary paths were prescribed for the retention relationship, the
difference in the volumetric behaviour predicted is believed to be primarily due
to the difference in the elastic compressibility with suction rather than the
distinct scanning paths followed. Indeed, the position of the G.W.T. at the end of
August 1974, illustrated in Figure 7-17, agreed well for the various values of the
parameter 𝜔 employed and no obvious difference in the resulting contours of
degree of saturation was exhibited, as shown in Figure 7-18. Starting from a
common saturated value of 𝜅𝑠 , the 𝑆𝑟 dependent compressibility, 𝜅𝑠∗ , degraded
faster with suction for higher values of the parameter 𝜔, thus justifying the
observed reduction in the magnitude of the computed displacements.
Although 𝜅𝑠∗ is expected to be smaller upon wetting and larger upon drying and,
therefore, to result in reduced swelling in comparison to shrinkage under ideal
conditions, the volumetric behaviour predicted was dominated by the boundary
conditions and the coupling between the mechanical and the hydraulic
components of the analysis. As a result, no such difference was obvious and
the general evolution of displacements followed by Analyses C.2 and C.3
agreed with the one exhibited by Analysis C.1, for which 𝜔 = 0.0.
393
Comparing the results of Analysis A and C.1, it may be concluded that the
hysteretic-SWRC is more realistic in reproducing the swelling induced
displacements of the two nodes examined, especially at the very start of the
analysis. The volumetric behaviour was less accumulative but without exhibiting
a sinusoidal evolution with time around the time axis; swelling seemed to be
driving the soil movements during the first 2.5 years, and shrinkage became
more dominant following the summer period of 1972. Furthermore, larger
relative swelling and shrinkage was generated by the hysteretic-SWRC model.
Nonetheless, the difference in volumetric behaviour, introduced by the
hysteretic-SWRC model, was moderated by the inclusion of a non-constant
compressibility, without, however, allowing the clear distinction between
swelling in the winter and shrinkage in the summer to be diluted.
Crest displacement
0.02
0.01
0.00
-0.01
(a)
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.03
0.02
(b)
0.01
0.00
-0.01
-0.02
-0.03
-0.04
-0.05
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
Anal. A Anal. C.1 Anal. C.2 Anal. C.3

monotonic  = 0.0  = 2.0  = 4.0
Figure 7-15: Crest displacements for analyses A(simple SWRC); C.1; C.2 and C.3 (hysteretic-SWRC): (a)
394
0.03
0.02
0.01
0.00
-0.01
(a)
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.01
(b)
0.00
-0.01
-0.02
-0.03
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Anal. A Anal. C.1 Anal. C.2 Anal. C.3

monotonic  = 0.0  = 2.0  = 4.0
Figure 7-16: Mid-slope displacements for analyses A(simple SWRC); C.1; C.2 and C.3 (hysteretic-
SWRC): (a) horizontal and; (b) vertical
395
Figure 7-17: Groundwater Table (G.W.T.) positions at the end of August 1974 for Analyses C.1, C.2 and
C.3
Figure 7-18: Contours of degree of saturation, 𝑆𝑟 , evaluated at the end of August 1974 for Analyses C.1,
C.2 and C.3
396
7.8 Hysteretic, specific volume dependent SWRC model (v-hysteretic-

SWRC)
The v-hysteretic-SWRC model was used to repeat Analysis C.2 ( 𝜒 = 1.0 ,

𝜔 = 2.0), employing three different values for the parameter 𝜓; 1.0 (Analysis
D.1); 2.0 (Analysis D.2); and 3.0 (Analysis D.3). Furthermore, Analysis D.3 was
repeated, assuming 𝜔 = 4.0, so that the combined effect of the parameters 𝜓
and 𝜔 could be assessed.
Figure 7-19 illustrates the horizontal and vertical displacements of the crest,
whereas the ones computed at mid-slope are shown in Figure 7-20. Similar to
the hysteretic-SWRC, the current model produced well defined changes of
volumetric behaviour between the winter and the summer.
The results of Analysis C.2, for which essentially 𝜓 = 0.0, and of Analysis D.1
( 𝜓 = 1.0 ) practically coincided, as confirmed by both figures. When 𝜓 was
increased to 2.0 (Analysis D.2), a variation, mainly in the horizontal
displacements, was exhibited which was even more pronounced when 𝜓 = 3.0
(Analysis D.3). A larger accumulation of displacements denoting shrinkage, was
generated for higher values of this parameter, which is in agreement with the
observation made in Section 7.6, for the monotonic models.
Nonetheless, the differences in the position of the G.W.T. at the end of August
1974, illustrated in Figure 7-21, were not as significant as for the monotonic v-
SWRC (Figure 7-13). Furthermore, the resulting distribution of degree of
saturation 𝑆𝑟 , shown in Figure 7-22, showed only a very slight dependency on
the parameter 𝜓, in contrast with the contours of 𝑆𝑟 shown in Figure 7-14, for
the v-SWR model. It should, however, be noted that the maximum value of the
parameter 𝜓 was 3.0 for the hysteretic model and 5.0 for the monotonic one.
Moreover, in the analyses of the current section, the model was not exclusively
performing at high degrees of saturation – as was the case for the monotonic
model – and 𝑆𝑟 was reduced upon wetting, due to the hysteretic nature of the
retention behaviour.
Contrary to the effect of the parameter 𝜓, which was depicted in the results of
Analyses D.1, D.2 and D.3 , the parameter 𝜔 affected the displacements
397
produced by Analyses D.3 and D.4 at the crest of the slope (Figure 7-23), to a
much lesser extent. Furthermore, the two analyses yielded practically equal
horizontal and vertical displacements at the middle of the slope (Figure 7-24).
The position of the phreatic surface for the two analyses, is shown in Figure
7-25 to have remained unaffected by the value of 𝜔 adopted. Similarly, there
was no obvious difference in the contours of 𝑆𝑟 produced and presented in
Figure 7-22 (c) and (d).
According to the conclusions of the previous section, the displacements were

expected to decrease for increasing values of 𝜔, due to the degradation of the
coefficient of soil compressibility 𝜅𝑠∗ with suction. Nevertheless, the parameter
𝜔 was shown to be relatively unimportant when combined with high values of
the parameter 𝜓 (its significance was already demonstrated for 𝜓 = 0.0 in the
set of analyses C).
For high saturation levels, the coefficient 𝜅𝑠∗ obtains magnitudes close to the
saturated value, 𝜅𝑠 , which is an input parameter, raised to the power of 𝜒,
assumed in the analyses to be 1.0. At the middle of the slope, 𝑆𝑟 was computed
to be approximately 90% at the end of the summer period, for the year 1974
(Figure 7-22), and is only expected to increase during the winter periods.
Therefore, 𝜅𝑠∗ is believed to be comparable for the two analyses which yielded
similar results at mid-slope. At the crest, where a lower degree of saturation
was calculated, the displacements showed relatively larger variation.
It is worth noting that for Analysis D.4, the horizontal displacement of the crest
(Figure 7-23 (a)) did not exhibit significant negative changes, associated with
swelling. This might be attributed to the reduced value of 𝜅𝑠∗ , resulting upon
wetting, whose effect was not pronounced in the C set of analyses, as
discussed in the previous section. For such a conclusion to be made, however,
further investigation is required, in order to study the effect of the parameter 𝜔
under various conditions and in combination with various values adopted for the
parameter 𝜓.
Applying the same evapotranspiration data summarised in Table 7-4 in two

successive cycles of 5 years each, Analysis D.1 was continued until the end of
398
December 1984. The vertical and horizontal displacements, calculated at the

crest and the middle of the slope during this extended period, are presented in
Figure 7-26 and Figure 7-27, respectively.
Figure 7-26 (a) shows that positive horizontal displacement (shrinkage) has
accumulated at the crest by the end of the 15th year of analysis with the
tendency to increase further with time. Nonetheless, the maximum magnitude of
displacement did not exceed 2 cm and was obtained at the end of summer
1983. At the same time, the vertical displacement at the crest was at the range
of 7 cm, as illustrated in Figure 7-26 (b). Overall, settlement accumulated
throughout the analysis, exhibiting a well defined increase with time. Similar
trends were demonstrated by the horizontal and vertical mid-slope
displacements, shown Figure 7-27 (a) and (b), respectively.
The way the finite element mesh deformed as a consequence of the vegetation
and the precipitation boundary conditions by December 1984, is shown in
Figure 7-28. The displacements in the figure are exaggerated for illustration
purposes and the maximum magnitude is at the range of 5 cm. Furthermore,
they are sub-accumulated from the end of the excavation so that the effect of
the boundary conditions may be isolated from the effect of the excavation.
Although the plot refers to a winter month, overall settlement and shrinkage was
exhibited. The displacement was minimal at the toe and increased towards the
crest where it acquired its maximum value. The settlement of the original
ground surface was slightly larger towards the open face of the cut but generally
uniform a few metres behind. Furthermore, there was no clear sign of
progressive failure having been initiated as a result of the cyclic changes of
suction and the slope was stable throughout the analysis. The stability
demonstrated is believed to be a consequence of the high value calibrated for
the critical state angle of shearing resistance, 𝜑𝑐𝑠 = 32.4°, and the large factor
of safety, 𝐹𝑠 , achieved by the end of the excavation (2.76 according to Table
6-3, for the particular case examined).
399
Crest displacement
0.01
0.00
-0.01
(a)
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.02
(b)
0.01
0.00
-0.01
-0.02
-0.03
-0.04
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Anal. C.2 Anal. D.1 Anal. D.2 Anal. D.3

 = 0.0  = 1.0  = 2.0  = 3.0
Figure 7-19: Crest displacements for analyses C2(hysteretic-SWRC); D.1; D.2 and D.3 ( v-hysteretic-
SWRC): (a) horizontal and; (b) vertical
400
0.02
0.01
0.00
(a)
-0.01
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.01
(b)
0.00
-0.01
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Anal. C.2 Anal. D.1 Anal. D.2 Anal. D.3

 = 0.0  = 1.0  = 2.0  = 3.0
Figure 7-20: Mid-slope displacements for analyses C2(hysteretic-SWRC); D.1; D.2 and D.3 ( v-
hysteretic-SWRC): (a) horizontal and; (b) vertical
Figure 7-21: Groundwater Table (G.W.T.) positions at the end of August 1974 for Analyses D.1, D.2 and
D.3
401
Figure 7-22: Contours of degree of saturation, 𝑆𝑟 , evaluated at the end of August 1974 for Analyses D.1,
D.2, D.3 and D.4
402
Vertical displacement (m) Horizontal displacement (m)
-0.04
-0.03
-0.02
-0.01
-0.01
0.00
0.01
0.00
0.01
Dec 1969 Dec 1969
Apr 1970 Apr 1970
Aug 1970 Aug 1970
Dec 1970 Dec 1970
Apr 1971 Apr 1971
 = 2.0
Aug 1971 Aug 1971
Anal. D.3
Dec 1971 Dec 1971
403
Apr 1972 Apr 1972
vertical
Aug 1972 Aug 1972
Time (months)
Time (months)
Dec 1972 Dec 1972

Crest displacement
Apr 1973 Apr 1973
Aug 1973 Aug 1973
Dec 1973 Dec 1973
Apr 1974 Apr 1974
 = 4.0
Anal. D.4
Aug 1974 Aug 1974
(b)
(a)
Dec 1974 Dec 1974
Figure 7-23: Crest displacements for analyses D.3 and D.4 ( v-hysteretic-SWRC): (a) horizontal and; (b)
0.02
0.01
0.00
(a)
-0.01
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.01
(b)
0.00
-0.01
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Anal. D.3 Anal. D.4

 = 2.0  = 4.0
Figure 7-24: Mid-slope displacements for analyses D.3 and D.4 ( v-hysteretic-SWRC): (a) horizontal
and; (b) vertical
Figure 7-25: Groundwater Table (G.W.T.) positions at the end of August 1974 for Analyses D.3 and D.4
404
Horizontal displacement (m) Crest displacement

0.02
0.01
0.00
(a)
-0.01
1969
1974
1979
1984
Time (months)
0.02
(b)
0.00
-0.02
-0.04
-0.06
-0.08
1969
1974
1979
1984
Time (months)
Figure 7-26: Crest displacements for Analysis D.1.time ( v-hysteretic-SWRC): (a) horizontal and; (b)
vertical
405
Horizontal displacement (m) Mid-slope displacement

0.03
0.02
0.01
0.00
(a)
-0.01
1969
1974
1979
1984
Time (months)
0.02
(b)
0.00
-0.02
-0.04
-0.06
1969
1974
1979
1984
Time (months)
Figure 7-27: Mid-slope displacements for Analysis D.1.time ( v-hysteretic-SWRC): (a) horizontal and; (b)
vertical
Figure 7-28: Deformed shape of the finite element mesh at the end of the 15 th year of analysis (sub-
accumulated from the end of the excavation)
406
7.9 Effect of initial conditions (K0 and OCR)
In the analyses presented so far, limited plasticity was generated, during the
excavation and the subsequent seasonal variation of suction within the ground.
This is justified by the initial values adopted for the coefficient of earth pressure
at rest, 𝐾0 , which was equal to 1.0, and for the overconsolidation ratio, OCR,
which was 5.5, in combination with the large critical state angle of shearing
resistance, 𝜑𝑐𝑠 , calibrated in Chapter 4 to be 32.4 o. The resulting factor of
safety, 𝐹𝑠 , at the end of the excavation for Analysis 1 of Chapter 6, which
formed the basis for the parametric study presented in the current chapter, was
2.76, reflecting the effect of the above parameters.
Analysis D.1 was repeated for two alternative values of 𝐾0 : 0.75 (Analysis
D.1.K0A in Table 7-1) and 1.25 (Analysis D.1.K0B in Table 7-1). The same
analysis was subsequently performed assuming OCR = 1.5 (Analysis D.1.OCR
in Table 7-1), for the original value of 𝐾0 . In this way, the initial stress conditions
and their relative position to the yield surface were altered and their effect was
assessed.
The displacements calculated at the crest (Figure 7-29) and the middle of the
slope (Figure 7-30), in both the horizontal and vertical directions, showed no
dependency on the coefficient of earth pressure at rest. This is believed to be
due to the large value of 𝜑𝑐𝑠 . Although the elements in compression were
brought closer to the yield surface when 𝐾0 = 0.75 and the elements in
extension were brought closer to the yield surface when 𝐾0 = 1.75, the available
shear strength prevented the shift of initial position to influence the evaluated
displacements.
On the contrary, the OCR value adopted did affect the calculated
displacements, mostly at the crest (Figure 7-31) and to a lesser extent at mid-
slope (Figure 7-32). This was so, not only because the factor of safety
decreased (according to the discussion in Section 6.6.2) but also because the
initial stress states approached the loading-collapse (LC) curve. The finite
elements in the upper, unsaturated ground zone violated the LC curve during
the first winter (January and February 1970), producing plastic collapse and
407
decreasing the computed amount of elastic swelling. Consequently, the

horizontal and vertical displacements of the crest (Figure 7-31) were reduced by
50%. Subsequent shrinkage and swelling, during the following years, was
comparable to that calculated for OCR = 5.5 and the two curves remained
parallel. By the end of the 5th year, larger displacements had accumulated for
the analysis with the smallest OCR.
Plasticity generated at the toe of the slope during the excavation, in combination
with wetting induced plasticity during the first months of 1970, caused the
displacement pattern, shown in Figure 7-33 (a). This illustrates relatively large
but uniform displacement vectors down the slope surface perhaps indicating the
early steps of the development of a failure mechanism. In contrast with Figure
7-7 (a), the vectors of incremental displacements illustrated in Figure 7-33 (a)
were uniformly distributed along the face of the slope. Nonetheless, the
mechanism formed was permanently stabilised during summer 1970, as suction
increased and shrinkage was predicted (Figure 7-33 (b)). The failure
mechanism did not recur the subsequent winter since the LC curve had already
been displaced, increasing in effect the elastic region. Had the angle 𝜑𝑐𝑠 been
smaller, failure might have occurred during the first months of the analysis.
The results of the analyses presented above demonstrate that the effect of the
overconsolidation may potentially be more significant than that of the coefficient
of earth pressure at rest. Since little change was observed in the total stresses
during the application of the vegetation and the precipitation conditions, 𝐾0 was
expected to influence the analysis at the stage of the excavation. On the
contrary, the value of OCR employed had an impact on both the excavation of
the slope and its behaviour under cyclic changes of suction, as wetting induced
plasticity may be generated on the LC curve.
408
Crest displacement
0.01
0.00
-0.01
(a)
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.02
(b)
0.01
0.00
-0.01
-0.02
-0.03
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Anal. D.1 Anal. D.1.K0A Anal. D.K0B

K0 = 1.00 K0 = 0.75 K0 = 1.25
Figure 7-29: Crest displacements for analyses D.1, D.1.K0A and D.1.K0B ( v-hysteretic-SWRC): (a)
409
0.015
0.010
0.005
0.000
-0.005
(a)
-0.010
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.010
(b)
0.000
-0.010
-0.020
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
Anal. D.1 Anal. D.1.K0A Anal. D.1.K0B

K0 = 1.00 K0 = 0.75 K0 = 1.25
Figure 7-30: Mid-slope displacements for analyses D.1, D.1.K0A and D.1.K0B ( v-hysteretic-SWRC): (a)
410
Crest displacement
0.01
0.00
(a)
-0.01
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.02
(b)
0.01
0.00
-0.01
-0.02
-0.03
-0.04
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Anal. D.1 Anal. D.1.OCR

OCR = 5.5 OCR = 1.5
Figure 7-31: Crest displacements for analyses D.1 and D.1.OCR ( v-hysteretic-SWRC): (a) horizontal
and; (b) vertical
411
0.02
0.01
0.00
(a)
-0.01
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.01
(b)
0.00
-0.01
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
Anal. D.1 Anal. D.1.OCR

OCR = 5.5 OCR = 1.5
Figure 7-32: Mid-slope displacements for analyses D.1 and D.1.OCR ( v-hysteretic-SWRC): (a)
412
Figure 7-33: Vectors of incremental displacements calculated at (a) March 1970 and (b) May 1970, for
Analysis D.1.OCR (OCR = 1.5)
7.10 Effect of root growth
In the analyses presented so far, the maximum root depth, 𝑟𝑚𝑎𝑥 , required for the
application of the vegetation boundary condition, was constant and equal to 2.0
m. Nonetheless, vegetation may take years to establish itself. Therefore,
Analysis D.1 was repeated for a root depth 𝑟𝑚𝑎𝑥 starting from 0.5 m and
increasing by 0.5 m/year (D.1.root in Table 7-1). Similar analyses were
performed by Nyambayo (2003) on a railway embankment founded on London
clay, showing that the displacements of the soil structure increased with root
depth and that modelling of the root growth played an important effect.
The horizontal and vertical crest displacements (Figure 7-34) and the horizontal
mid-slope displacements (Figure 7-35 (a)) showed very little dependency on the
maximum root depth. The effect of 𝑟𝑚𝑎𝑥 was slightly more exagerated on the
vertical displacements at the middle of the slope (Figure 7-35 (b)).
413
Although the two analyses produced similar swelling during the first few months
of 1970 (Figure 7-35 (b)), Analysis D.1.root, for which 𝑟𝑚𝑎𝑥 = 0.5 m, yielded the
largest amount of subsequent shrinkage and swelling. The original analysis
(𝑟𝑚𝑎𝑥 = 2.0 m) produced the smallest displacements. During the second and the
third year, the largest displacements were once more evaluated for the analysis
with the shortest roots (D.1.root; 𝑟𝑚𝑎𝑥 = 1.0 m in 1971 and 𝑟𝑚𝑎𝑥 = 1.5 m in
1972). By the fourth year the roots in Analysis D.1.root had grown to 2.0 m, i.e.
equal to the roots in Analysis D.1. The curves in Figure 7-35 (b) remained
practically parallel during 1973, indicating that similar shrinkage and swelling
was produced by the two analyses. The distance between the curves
decreased in the fifth year and the displacements produced by Analysis D.1
(𝑟𝑚𝑎𝑥 = 2.0 m) approached those calculated for Analysis D.1.root ( 𝑟𝑚𝑎𝑥 = 2.5
m).
The largest relative swelling and shrinkage were systematically produced by the
shortest roots. This trend might seem unrealistic at first. However, it results from
the fact that the sink term 𝑆𝑎𝑐𝑐 is inversely proportional to the maximum root
depth, according to Equation 3.94, which is repeated below for clarity (refer to
Section 3.6.2 for further information):
2 ∙ 𝛼 ∙ 𝑇𝑝 𝑟
𝑆𝑎𝑐𝑐 = 𝛼 ∙ 𝑆𝑚𝑎𝑥 = ∙ 1− (7.1)
𝑟𝑚𝑎𝑥 𝑟𝑚𝑎𝑥
The maximum potential sink term, 𝑆𝑚𝑎𝑥 , (expressed in flow units, for example
𝑚3 𝑠𝑒𝑐 ) varies linearly with depth as shown schematically in Figure 7-36.
Furthermore, 𝑆𝑚𝑎𝑥 is larger for shallower 𝑟𝑚𝑎𝑥 at the ground surface, despite
degrading more quickly with depth. The triangular areas, defined for the two
root depths assumed, are both equal to the potential transpiration rate 𝑇𝑝 .
The pore pressure distribution at the end of the excavation and prior to the
application of the vegetation boundary condition was identical for Analyses D.1
and D.1.root. Therefore, the 𝛼 function was common. Moreover, the same
potential evapotranspiration 𝑇𝑝 was prescribed in both analyses.
Consequently, 𝑟𝑚𝑎𝑥 was the only parameter in Equation 3.94 to affect the
results. For 𝑟 = 0.0 m, i.e. at the ground surface, the smaller the maximum root
414
depth, 𝑟𝑚𝑎𝑥 , the larger the sink term, 𝑆𝑚𝑎𝑥 , and the larger the surface
displacements.
By the end of August 1970, the G.W.T. was further depressed around the toe of
the slope for Analysis D.1.root, for which 𝑟𝑚𝑎𝑥 was 0.5 m, as confirmed in Figure
7-37 (a). The G.W.T. was less affected for Analysis D.1, which employed a
larger 𝑟𝑚𝑎𝑥 (equal to 2.0 m), thus justifying the smaller settlement exhibited in
Figure 7-35 (b).
During the following winter, the G.W.T. was raised to a comparable level for the
two cases (Figure 7-37 (b)). The largest variation relative to the previous
summer was recorder for Analysis D.1.root, which, accordingly, produced the
largest amount of swelling. At the end of the second summer (Figure 7-37 (c)),
the deepest G.W.T. was obtained once more for Analysis D.1.root which
produced the largest shrinkage (Figure 7-35 (b)). Finally, the analyses resulted
in practically common G.W.T. positions, shown in Figure 7-37 (d), for the end of
the summer period of the year 1974.
The horizontal displacements showed little dependency on the assumed

maximum root depth, 𝑟𝑚𝑎𝑥 , not only at the surface but also with depth.
Calculated along a vertical line initiated at the crest (Figure 7-38 (a)) and at the
middle of the slope (Figure 7-38 (b)), the horizontal displacements produced by
Analyses D.1 and D.1.root, exhibited a maximum difference of few mm by the
end of August 1970. This difference subsided slowly with depth. At the end of
August 1972 (Figure 7-39), the difference in the displacements calculated with
depth was even smaller and no difference was practically exhibited at the end of
August 1974 (Figure 7-40).
The vertical displacements along the original horizontal ground surface were not
significantly affected by the assumed maximum root depth, 𝑟𝑚 𝑎𝑥 . On the
contrary, the vertical displacements along a horizontal line from mid-slope (at
the original G.W.T. depth) showed a slightly larger dependency on 𝑟𝑚𝑎𝑥 . At the
end of August 1970 (Figure 7-41), similar surface displacements were
calculated from Analyses D.1 and D.1.root while the vertical displacements at
5.0 m of depth were larger for the latter case in comparison with the former. The
difference in the results (approximately 5 mm at the face of the cut) extended
415
for not more than 10.0 m behind the face of the slope and thereafter the two
curves coincided. At the end of August 1972, the surface settlements (Figure
7-42 (a)) exhibited little variation for the two cases examined but they were
larger in comparison to August 1970. The difference in the displacements
calculated at 5.0 m of depth ((Figure 7-42 (b)) was similar to that exhibited in
August 1970 but extended slightly further behind the face of the cut. Finally, at
the end of August 1974, the surface settlements (Figure 7-43 (a)) had increased
further in magnitude and showed a slight difference for Analyses D.1 and
D.1.root. The difference in the displacements at 5.0 m depth (Figure 7-43 (b))
extended to almost 30.0 m behind the face of the slope.
From the above observations is clear that the effect of the increase in root depth
with time was mainly exhibited in the vertical displacements produced at some
depth beneath the ground surface. The vertical displacements at the middle of
the slope are primarily due to the seasonal variation of the G.W.T., which is
focused at the lower half of the open face of the cut and at a distance of
approximately 10.0 m behind. The limited variation in the position of the G.W.T.
further away from the face – the maximum root depth was not deep enough to
have a large affect the G.W.T. depth – restricted the effect of root growth on the
surface settlements.
Modelling the root growth with time allows the effect of transpiration, initially
limited to the surface, to gradually progress towards larger depths. Modelling of
this effect is expected to be fundamental in cases where the G.W.T. is shallow,
as subsequently demonstrated.
Figure 7-44 illustrates the displacements at the crest of the same slope for an
initial position of the G.W.T. assumed at -1.0 m, for the case where the
maximum root depth, 𝑟𝑚𝑎𝑥 , was 2.0 m throughout the analysis and for the case
where 𝑟𝑚𝑎𝑥 increased by 0.5 m/year.
In the former analysis, seasonal variation of the crest displacement was limited
to the first 16 months (up to April 1971). Infiltration of rain water during the
winter period from August 1970 to February 1971 fully saturated the soil to the
ground surface (full saturation did not occur earlier because the analysis was
started in January 1970 and only few months of precipitation were taken into
416
account, rather than the whole winter period). The compressive pore water
pressures, which consequently built up, restrained transpiration during the
subsequent summer period; 𝛼 in Equation 3.94 is zero for zero suction and for
compressive pore water pressures (anaerobiosis point). Despite the applied
hydraulic conditions, no further change in pore water pressures was predicted
and the displacements remained unaffected. It should be noted that although
the maximum root depth, 𝑟𝑚𝑎𝑥 , was 2.0 m, only the part of the roots above the
G.W.T. was active and contributed to transpiration during the first summer.
On the contrary, modelling of the root growth allowed for seasonal variation of
the crest displacement to be predicted past April 1971. Despite being shallower,
the roots were active along their whole length at the commencement of the
analysis. Furthermore, the sink term was larger at the ground surface,
generating larger suctions during the first summer. Precipitation during the
subsequent winter was not large enough to fully saturate the entire area and
several finite elements close to the ground surface remained under unsaturated
conditions. The 𝛼 function was, therefore, non zero and transpiration was
predicted during the following summer and so on.
In conclusion, modelling of the root growth is fundamental when the G.W.T. is

shallow but is also considered to be of importance for deeper G.W.T., as the
natural process of the vegetation establishing itself is more accurately
accounted for.
417
Crest displacement
0.01
0.00
-0.01
(a)
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.02
(b)
0.01
0.00
-0.01
-0.02
-0.03
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Anal. D.1 Anal. D.1.root

rmax = 2.0 m rmax 0.5 m /year
Figure 7-34: Crest displacements for analyses D.1 and D.1.root (v-hysteretic-SWRC): (a) horizontal and;
(b) vertical
418

0.02
0.01
0.00
(a)
-0.01
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.01
(b)
0.00
-0.01
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973

rmax = 2.0 m rmax 0.5 m / year
Figure 7-35: Mid-slope displacements for analyses D.1 and D.1.root ( v-hysteretic-SWRC): (a) horizontal
and; (b) vertical
Sink term, Smax (m3/s)
Depth, r (m)
deep roots
shallow roots
Figure 7-36: Schematic variation of the sink term, 𝑆𝑚𝑎𝑥 , with depth for deep and shallow maximum root
depth, 𝑟𝑚𝑎𝑥
419
Figure 7-37: Groundwater Table (G.W.T.) positions for Analyses D.1 and D.1root at the end of (a)
August 1970; (b) February 1971; (c) August 1971; (d) August 1974
420
Crest, 1970
0.000

2.000
4.000
6.000
8.000
(a)
10.000
-0.005 0 0.005 0.01 0.015
Mid-slope, 1970
0.000
2.000 Vertical distance from mid-slope (m)
4.000
6.000
8.000
(b)
10.000
-0.005 0 0.005 0.01 0.015
Anal.D.1 Anal. D.1.root

Analyses D.1 (𝑟𝑚𝑎𝑥 = 2.0 m) and D.1.root (𝑟𝑚𝑎𝑥 growing by 0.5 m / year), August 1970
421
Crest, 1972
0.000

2.000
4.000
6.000
8.000
(a)
10.000
-0.005 0 0.005 0.01 0.015
Mid-slope, 1972
0.000
2.000 Vertical distance from mid-slope (m)
4.000
6.000
8.000
(b)
10.000
-0.005 0 0.005 0.01 0.015

422
Crest, 1974
0.000

2.000
4.000
6.000
8.000
(a)
10.000
-0.005 0 0.005 0.01 0.015
Mid-slope, 1974
0.000
Vertical distance from mid-slope (m)

2.000
4.000
6.000
8.000
(b)
10.000
-0.005 0 0.005 0.01 0.015

423
Crest, 1970
0.000

-0.010
-0.020
(a)
-0.030
0 10 20 30 40 50
Mid-slope, 1970
0.005
0.000 Vertical displacement (m)
-0.005
-0.010
-0.015
(b)
-0.020
0 10 20 30 40 50

Analysis D.1 (𝑟𝑚𝑎𝑥 = 2.0 m) and D.1.root (𝑟𝑚𝑎𝑥 growing by 0.5 m / year), August 1970
424
Crest, 1972
0.000

-0.010
-0.020
(a)
-0.030
0 10 20 30 40 50
Mid-slope, 1972
0.005
-0.005
-0.010
-0.015
(b)
-0.020
0 10 20 30 40 50

Figure 7-42 Vertical displacements along horizontal lines starting at the (a) crest and at (b) mid-slope;
425
Crest, 1974
0.000

-0.010
-0.020
(a)
-0.030
0 10 20 30 40 50
Mid-slope, 1974
0.005
-0.005
-0.010
-0.015
(b)
-0.020
0 10 20 30 40 50

426
Crest displacement, G.W.T. = -1.0 m

0.01
0.00
(a)
-0.01
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
Dec 1974
Time (months)
0.02
(b)
0.01
0.00
-0.01
-0.02
Apr 1970
Aug 1970
Apr 1971
Aug 1971
Apr 1972
Aug 1972
Apr 1973
Aug 1973
Apr 1974
Aug 1974
Dec 1969
Dec 1970
Dec 1971
Dec 1972
Dec 1973
rmax = 2.0 m rmax increasing 0.5 m /year
Figure 7-44: Crest displacements for maximum root depth 𝑟𝑚𝑎𝑥 =2.0 m and maximum root depth 𝑟𝑚𝑎𝑥
increasing by 0.5 m/year: (a) horizontal and; (b) vertical
7.11 Variable permeability model
As explained in Chapter 3, permeability is known to vary with suction and

consequently with the degree of saturation. Although the former relationship is
believed to be hysteretic, the latter was shown to be monotonic, as discussed
by Fredlund & Rahardjo (1993). Two models are available in ICFEP, regarding
the decrease of permeability due to de-saturation and both are monotonic: one
adopting a linear variation of the logarithm of the coefficient 𝑘 with suction and
the second one employing a non-linear variation of the coefficient 𝑘 with the
427
degree of saturation, 𝑆𝑟 . Preferably the second model should be used in

combination with the hysteretic models for the SWRC.
The 𝑆𝑟 dependent variable permeability model available in ICFEP is based on

the Van Genuchten (1980) expression (derived from the Mualem, 1976
equation for predicting the relative hydraulic conductivity from the soil-water
retention relationship) and is repeated herein for clarity (for more information
refer to Section 3.5):
2
𝑚 𝑚
𝑘 = 𝑘𝑠𝑎𝑡 ∙ 𝛩 1 2
∙ 1 − 1 − 𝛩1 (7.2)
where:
𝛩= (7.3)
1 − 𝑆𝑟0
𝑆𝑟0 being the residual degree of saturation and 𝑚 a fitting parameter.
The nonlinearity introduced by the above expression increased the

computational time required dramatically and, therefore, a linear variation of
log 𝑘 with suction was instead employed in the finite element analyses
presented so far. In order to assess the inaccuracy instigated by this
simplification, Analysis D.1 was once more repeated, adopting the 𝑆𝑟 dependent
permeability model and the relevant analysis is referred to in Table 7-1 as
D.1.perm.
The fitting parameter 𝑚 was assumed equal to the one adopted in the simple,
Van Genuchten (1980) type SWRC model, used in Analysis A (Section 7.6), i.e.
0.7, as shown in Table 7-3. The residual degree of saturation, 𝑆𝑟0 , and the
current degree of saturation, 𝑆𝑟 , were calculated from the v-hysteretic-SWRC
model.
In Figure 7-45 (a) the linear variation of the coefficient log 𝑘 with the equivalent
suction, employed in Analysis D.1, is illustrated in black. Also shown in the
same figure (grey curve) is the variation of coefficient 𝑘 with suction, adopted in
Analysis D.1.perm. The latter curve was derived from Equation 7.2; the degree
of saturation, 𝑆𝑟 , computed for the primary drying and wetting retention curves
428
(assuming an initial void ration of 0.7 as in Section 7.4), was used in the
calculation of 𝑘, which was then plotted versus the corresponding equivalent
suction, 𝑠𝑒𝑞 . Since the SWRC was hysteretic (v-hysteretic-SWRC model) the
obtained curve was also hysteretic. For illustration purposes the minimum value
of 𝑘 plotted was 10-11 m/s, although significantly lower values were evaluated.
Consequently, the loop is shown to be open, even though this was not the case.
Similarly, in Figure 7-45 (b) the non-linear variation of 𝑘 with the degree of
saturation, resulting directly from Equation 7.2, is plotted in grey. Also shown in
the figure is the variation of 𝑘 with the degree of saturation adopted in Analysis
D.1; the coefficient 𝑘 was calculated from the equivalent suction but was
subsequently plotted versus the corresponding degree of saturation, 𝑆𝑟 . A
different 𝑆𝑟 corresponds to each suction level, depending on the hydraulic
loading. The curve illustrated in Figure 7-45 (b) was obtained from the primary
retention paths (assuming an initial void ration of 0.7 as above) and is
hysteretic.
The suction dependent variable permeability, adopted in Analysis D.1, was

monotonic when plotted in terms of log 𝑘 and 𝑠𝑒𝑞 and hysteretic when plotted in
the log 𝑘 - 𝑆𝑟 plane. This is in disagreement with the evidence discussed by
Fredlund & Rahardjo (1993). On the contrary, the degree of saturation
dependent permeability, adopted in Analysis D.1.perm, was hysteretic in the
log 𝑘-𝑠𝑒𝑞 and monotonic in the log 𝑘-𝑆𝑟 plane, agreeing with the literature.
The degradation of permeability with suction and the degree of saturation was
generally faster for Analysis D.1.perm. Nevertheless, the results of the two
analyses in terms of horizontal and vertical displacements at the crest (Figure
7-46) and the middle (Figure 7-47) of the slope were not significantly different in
order to justify employment of the computationally costly 𝑆𝑟 dependent model.
In the context of the induced suction (maximum of approximately 100.0 kPa)
and degree of saturation levels (mostly in excess of 75% for Analysis D.1), the
difference in coefficients 𝑘 employed by the two models was within one order of
magnitude.
429
It should, however, be noted, that the displacements were compared only for
the first two years (1970 and 1971), as Analysis D.1.perm was not continued
further. The computational time required for these two years was approximately
four times longer than that required for the five years of Analysis D.1 and for
convergence to be achieved, the stiffness matrix was updated for the first 50
iterations of every increment (Potts & Zdravkovic, 1999). Given the increased
computational effort required and the minor impact of substituting the 𝑆𝑟
dependent model for the suction dependent one, it was thought that the
simplification adopted was necessary and justified in the particular boundary
value problem studied.
1E-008
permeability, k (m / s)
Coefficient of water
1E-009
Primary
1E-010 wetting
Primary
drying (a)
1E-011
0 400 800 1200 1600 2000
1E-008 Primary
permeability, k (m / s )
wetting
Coefficient of water
Primary
1E-009 drying
1E-010
(b)
1E-011
0 0.2 0.4 0.6 0.8 1
Anal.D.1 Anal. D.1.perm

suction dependent permeability Sr dependent permeability
Figure 7-45: Variation of permeability with (a) equivalent suction, 𝑠𝑒𝑞 , and (b) degree of saturation, 𝑆𝑟 ,
for Analyses D.1 (suction dependent permeability) and D.1.perm (degree of saturation dependent
permeability)
430
Horizontal displacement (m) Crest displacement

0.01
0.00
-0.01
Dec 1969
Dec 1970
Dec 1971
Feb 1970
Apr 1970
Jun 1970
Aug 1970
Oct 1970
Feb 1971
Apr 1971
Jun 1971
Aug 1971
Oct 1971
Time (months)
0.02
0.01
0.00
-0.01
Dec 1969
Dec 1970
Dec 1971
Feb 1970
Apr 1970
Jun 1970
Aug 1970
Oct 1970
Feb 1971
Apr 1971
Jun 1971
Aug 1971
Oct 1971
Time (months)
Anal. D.1 Anal. D.1.perm

Figure 7-46: Crest displacements for analyses D.1 (suction dependent permeability) and D.1.perm (𝑆𝑟
dependent permeability): (a) horizontal and; (b) vertical
431

0.01
0.00
-0.01
Dec 1969
Dec 1970
Dec 1971
Feb 1970
Apr 1970
Jun 1970
Aug 1970
Oct 1970
Feb 1971
Apr 1971
Jun 1971
Aug 1971
Oct 1971
Time (months)
0.01
0.00
-0.01
Dec 1969
Dec 1970
Dec 1971
Feb 1970
Apr 1970
Jun 1970
Aug 1970
Oct 1970
Feb 1971
Apr 1971
Jun 1971
Aug 1971
Oct 1971
Time (months)
Anal. D.1 Anal. D.1.perm

Figure 7-47: Mid-slope displacements for analyses D.1 (suction dependent permeability) and D.1.perm
(𝑆𝑟 dependent permeability): (a) horizontal and; (b) vertical
The behaviour of an unsaturated soil slope under seasonal changes of suction

was studied employing the v-SWRC, the hysteretic-SWRC and the v-hysteretic-
SWRC models, which were developed as part of the current research project.
Rather than directly prescribing the seasonal suction variation it was indirectly
controlled by the hydraulic boundary conditions (precipitation and vegetation
boundary conditions). Meteorological data with reference to rainfall and
432
potential evapotranspiration were employed in the coupled consolidation

analyses. The available data, obtained from the Building Research
Establishment testing site at Chattenden, Kent, UK (Nyambayo, 2003), covered
5 years. At the end of this period the displacements computed employing the
different SWRC models were compared and the effect of the newly
implemented features on the soil response was investigated.
Typically, suctions increase within the ground during the summer, due to
excessive transpiration rates, whereas, during the winter, when precipitation is
dominant, the soil is driven towards saturation. Due to this suction variation,
generalised shrinkage was predicted during the summer period, followed by
generalised swelling during the winter. Moreover, the ground movements were
more pronounced around the open face of the excavation and at smaller depth,
where conditions were less confined and directly exposed to the applied
boundary conditions.
Comparing the results produced when monotonic and hysteretic SWRC’s were
employed (Analysis A and C.1), it was concluded that the latter was more
realistic in reproducing the cyclic volumetric soil response to seasonal variations
of suction, whereas a more accumulative and monotonic response was
obtained from the former.
The amount of shrinkage and swelling generated by the hysteretic-SWRC

model was moderated by the inclusion of non-constant soil compressibility with
suction, 𝜅𝑠∗ , without, however, allowing the clear distinction between swelling in
the winter and shrinkage in the summer to be diluted. Larger values of the
parameter 𝜔, which controls the impact of the current degree of saturation, 𝑆𝑟 ,
on the soil compressibility 𝜅𝑠∗ , were associated with smaller cyclic displacements
at the crest and the middle of the slope. Starting from a common saturated
value of 𝜅𝑠 , soil compressibility 𝜅𝑠∗ degraded faster with suction for higher values
of the parameter 𝜔, thus justifying the observed reduction in the magnitude of
the computed displacements.
Larger displacements were predicted at the crest and the middle of the slope for
higher values of the parameter 𝜓, both when the monotonic (v-SWRC) and the
hysteretic (v-hysteretic-SWRC) models were employed. The parameter 𝜓
433
(initially introduced by Gallipoli et al., 2003b) controls the shift of the SWRC due
to specific volume changes. For increased values of this parameter larger
degrees of saturation 𝑆𝑟 were sustained within the ground, facilitating seepage.
Consequently, the seasonal variation of the G.W.T., which is naturally
associated with soil movements, was larger.
The parameter 𝜔 was shown to be relatively unimportant when combined with

high values of the parameter 𝜓. The effect of the former parameter is more
pronounced on the volumetric soil behaviour when large changes in the degree
of saturation are involved in the analysis. For increased values of the parameter
𝜓, however, the model performs under higher degrees 𝑆𝑟 for a given range of
suction levels.
Continuing one analysis to 15 years (Analysis D.1.15yr), shrinkage and

settlement accumulated, showing a clear tendency to increase further with time.
No indication of progressive failure was given and the slope was stable
throughout the analysis. The stability demonstrated was due to the high value of
shearing resistance, 𝜑𝑐𝑠 = 32.4°, of the material used in the analysis and the
large overconsolidation ratio, OCR = 5.5.
When a smaller OCR was employed (OCR = 1.5), the LC curve was violated
during the first winter. The plastic collapse produced decreased the elastic
swelling and caused the initiation of a failure mechanism. This mechanism was
stabilised during the subsequent summer, as suction increased and shrinkage
was predicted. It did not recur the following winter since the LC curve had been
displaced, increasing in effect the elastic region. Therefore, it was the first
winter which was critical for the stability of the slope.
On the contrary, the coefficient of earth pressure at rest, 𝐾0 , had no effect on

the computed displacements at the crest and the middle of the slope.
Modelling the root growth with time primarily affected the vertical displacements
at the middle of the slope and at a maximum distance of 30.0 m behind it. This
was the area where the seasonal variation of the G.W.T. mainly occurred. The
horizontal displacements both at the surface and with depth were much less
affected. The vertical displacements at the surface, where conditions were less
434
confined, were larger in magnitude than those calculated at 5.0 m depth but
showed smaller variation between the two cases examined, i.e. between
Analyses D.1 and D.1root. Furthermore, modelling of the root growth was
shown to be fundamental for shallow G.W.T. Therefore, it is concluded that the
effect of root growth is more significant in the regions of the finite element mesh
where conditions may vary between full and partial saturation.
Smaller mid-slope vertical surface displacements corresponded to deeper roots.

This was a consequence of the fact that the sink term, 𝑆𝑚𝑎𝑥 , which is associated
with the effect of transpiration, is inversely proportional to the maximum root
depth, 𝑟𝑚𝑎𝑥 .
Despite being smaller at the surface, the sink term, 𝑆𝑚𝑎𝑥 , degrades more slowly
with depth, for the case of larger roots, and the effect of transpiration directly
extends to bigger depths. Nonetheless, in the cases studied, the vertical
displacements under the horizontal ground surface, at 5.0 m depth and at a
distance of more than 30.0 m from the open face of the excavation, were not
affected by the different root depths employed in the analyses. Since the
variation in the results was concentrated in regions where conditions varied
between full and partial saturation, as concluded above, the effect of 𝑟𝑚𝑎𝑥 was
not displayed at depth, probably due to the large difference between the G.W.T.
depth and the maximum root depth. The influence of the vegetation boundary
condition on soil displacements should be further investigated under various
ratios of G.W.T. depth over 𝑟𝑚𝑎𝑥 .
Finally, two models were considered for the variation of coefficient of soil
permeability, 𝑘 , due to de-saturation; one adopting a linear variation of the
coefficient log 𝑘 with suction and a second one employing a non-linear variation
of the coefficient 𝑘 with the degree of saturation, 𝑆𝑟 . The results of the two
analyses in terms of horizontal and vertical displacements at the crest and the
middle of the slope were not significantly different. Therefore, employment of
the linear expression, instead of the non-linear one, is advisable for the
boundary value problem studied, as the non-linearity introduced by the latter
increased the computational time and effort dramatically.
435
The SWRC models, developed in the course of this research project, were
successfully employed in the boundary value problem of a slope subjected to
seasonal changes of suction. The differences in the formulation of the models
reflected on the obtained results, demonstrating the importance of appropriately
selecting the retention relationship, in numerical analyses of unsaturated soils.
436
chapter 8: CONCLUSIONS AND SUGGESTIONS FOR
FURTHER RESEARCH
8.1 Introduction
The aim of the research presented in this thesis was (a) to overcome modelling
limitations regarding the shearing behaviour of overconsolidated unsaturated
soils and the retention behaviour of unsaturated soils and (b) to employ the new
developments, in combination with the existing capabilities of ICFEP, in the
analysis of slopes in unsaturated soils.
Nonetheless, only limited laboratory data regarding the shearing behaviour of

overconsolidated unsaturated soils are available and several aspects require
further experimental investigation. These are discussed first in Section 8.2.
The conclusions drawn from this research are subsequently presented in two
parts. The first part refers to the development of the Hvorslev surface and its
application in a boundary value problem. The surface was developed in order to
substitute for the yield and plastic potential functions on the dry side of the
critical state in the 𝑝 - 𝐽 plane. Its implementation in a finite element code
(ICFEP), its validation and its calibration were presented. The Hvorslev surface
437
Conclusions and Suggestions for Further Research
was employed in the study of slope stability in highly overconsolidated

unsaturated soils where its usefulness was demonstrated. Conclusions relevant
to the development and use of the Hvorslev surface are summarised in Section
8.3.
The second part refers to the development of three soil-water retention curve
(SWRC) models and their application in the study of the behaviour of a slope
under seasonal changes of suction. It was demonstrated that the choice of the
SWRC model and of the model parameters employed have a significant
influence on the behaviour predicted. The relevant conclusions are presented in
Section 8.4.
The existing ICFEP modelling capabilities and the alterations made during this
research project were combined to produce a novel approach in the modelling
of the mechanical and the hydraulic behaviour of unsaturated soils, which is
summarised in Section 8.5.
Finally, suggestions for further development of the constitutive and the SWRC
models and suggestions for further numerical analyses are presented in the last
two sections of the chapter.
8.2 Shearing behaviour of unsaturated soils
Constitutive modelling of unsaturated soils has mainly been focused on

simulating the soil behaviour under isotropic stress conditions. Some research
has been carried out in extending these constitutive models to account for
shearing, however, this has been limited to triaxial stress space. Modelling of
the soil behaviour in the generalised stress space has attracted limited interest
despite being necessary for finite element analyses of boundary value
problems. Indeed, it has been suggested that extension of the constitutive
models, such as those presented in Chapter 2, to the generalised stress space
is straightforward, following standard procedures (Gens, 1995). Nonetheless,
the generalisation of these models has not been presented in the literature.
438
Georgiadis (2003) and Georgiadis et al. (2005) have discribed the expansion of
a Barcelona Basic Model (BBM; Alonso et al., 1990) type model into the
generalised stress space. The Matsuoka-Nakai failure criterion was adopted in
the deviatoric plane and the Lagioia et al. (1996) expression, extended to
unsaturated conditions, was employed for the yield and plastic potential
functions in the 𝑝-𝐽 plane. Nevertheless, this formulation was shown in Chapter
4 to be inadequate when simulating the behaviour of highly overconsolidated
unsaturated soils on the dry side of the critical state and the Hvorslev surface
was therefore developed.
To date no generally accepted constitutive model has been proposed that can
reproduce in detail both the isotropic and the shearing behaviour of unsaturated
soils, whilst accounting for the coupling between the mechanical and the
hydraulic components of the unsaturated soil behaviour. The present thesis is a
step towards this direction. Nevertheless, the approach proposed herein is
based on a number of simplifications, made due to the lack of comprehensive
experimental data which would facilitate the understanding and the modelling of
different components of the unsaturated soil behaviour under various shearing
conditions. For example, it is not yet clear:
 what is the shape of the yield surface in the 𝑝-𝐽 stress plane, on the wet
and on the dry side of the critical state;
 if and how the above shape is affected by the suction level and/or by the
degree of saturation;
 if the drying/wetting history influences this shape;
 what is the shape of the critical state line (CSL) in the 𝑝-𝐽 stress plane at
constant suction – the assumption that the CSL is a straight line in the
above plane is usually based on experimental data of samples sheared
from normally consolidated or lightly overconsolidated states (the data by
Estabragh & Javadi (2008), presented in Chapter 4 indicate that the CSL
becomes curved in unsaturated states);
439
 if and how the shape of the CSL in the 𝑝-𝐽 stress plane is affected by the
suction level and/or by the degree of saturation and whether the
drying/wetting history influences this shape;
 how the apparent cohesion increases with suction and/or with the degree
of saturation – assuming a false shape for the CSL in the 𝑝-𝐽 stress plane
may lead to a false understanding of the evolution of the intercept with
suction.
Experimental investigation of the above points would require shearing of

samples at different OCR values and suction levels. The data obtained from
these tests could then be combined with the existing data for both the isotropic
and the retention behaviour to facilitate the calibration of constitutive models
formulated in the generalised stress space.
8.3 The Hvorslev surface
A new curved surface, termed the Hvorslev surface, was developed as part of
the current research project, in order to replace the Lagioia et al. (1996)
expression for the yield and plastic potential functions on the dry side of the
critical state. The newly developed expression requires 4 parameters, which are
assumed to be independent of suction, to be defined; the inclination of the yield
surface, 𝛼𝐻𝑉 , the gradient of the flow vector at 𝑝 = −𝑓(𝑠𝑒𝑞 ), 𝛽𝐻𝑉 , and two fitting
parameters 𝑛 and 𝑚.
The formulation of the Hvorslev surface into a constitutive model, its

implementation into a numerical code and the validity of the latter were
presented in Chapter 4. The constitutive model was calibrated employing
experimental data for an artificial silty clay which were available in the literature.
The new surface was compared to three typically employed shapes – the Cam-
clay (CC), the modified Cam-clay (MCC) and the sinfonietta classica (SC) – in
terms of their performance on the dry side over a wide range of suction levels.
The results of the analyses performed demonstrated the capabilities and the
440
superiority of the new surface in simulating the soil behaviour under various
suction levels.
The Hvorslev surface was employed in the numerical analysis of the stability of
an excavated slope in Chapter 6. The excavation was assumed to be performed
in the above mentioned artificial soil and the model parameters produced by the
calibration of the surface were employed in the analyses. The stability of the
slope was studied under various conditions, for which the factor of safety was
estimated.
The results of the analyses demonstrated that:
 numerical analysis of slope stability in highly overconsolidated

unsaturated soils requires employment of a constitutive model, such as
the one presented herein, which accounts for the beneficial effect of
suction without over-predicting the soil strength;
 disregard of the soil suction causes under-prediction of the slope

stability;
 in cases where an unsaturated analysis is not feasible, it is adequate to

follow an effective stress approach, allowing for suctions to develop
above the groundwater table, as long as drained conditions are ensured
throughout;
 for slope stability to be improved, suction does not necessarily need to

increase in absolute terms but only relative to the excavation depth;
 slope stability improved for increasing values of OCR.
The performance of the Hvorslev surface in simulating the soil behaviour

observed in the laboratory as well as its performance in boundary value
problems is satisfying despite its simplicity. Its implementation is essential in
order to accurately predict the stresses at highly overconsolidated stress states.
441
8.4 The soil-water retention curve (SWRC)
Three SWRC models, which account for (a) the effect of void ratio, (b) hydraulic
hysteresis and (c) the combined effect of the two were developed and
implemented in a finite element code (ICFEP).
The first of the SWRC models (v-SWRC) is a modification of the Gallipoli et al.
(2003b) model and accounts for the effect of specific volume, 𝑣, on the position
of the retention curve, through the introduction of the parameter 𝜓 . The
retention relationship becomes a 3-dimensional surface (SWRS) but remains
monotonic. Gradual changes in suction, 𝑠, and in the specific volume, 𝑣, impose
a continuous change in the degree of saturation causing the retention point to
move from one iso-volumetric curve to the next one, i.e. it moves on the SWRS.
The position of the projection of the SWRC in the 𝑠-𝑆𝑟 plane is moved upwards
or downwards, depending on whether the specific volume is decreasing or
increasing, respectively. Increasing the parameter 𝜓 intensifies this shift
whereas for 𝜓 = 0.0, 𝑣 has no effect on the simulated soil behaviour.
The second model was developed to include the hydraulic hysteresis

(hysteretic-SWRC model), commonly exhibited by unsaturated soils. A smooth
transition from scanning to primary paths was modelled. S-shape curves are
employed for the primary drying and wetting paths, which have two common
points: the point of de-saturation and the residual point. The scanning paths are
assumed to be arcs of circles, centred on the vertical line passing through the
last reversal point, and to rejoin the corresponding primary path with a common
tangent. The capability of the model to reproduce the scanning paths was
shown to be satisfactory, despite the simple geometric shape assumed.
The third model incorporates the effect of specific volume and hydraulic
hysteresis (v-hysteretic-SWRC model) in a 3-dimensional formulation; the
primary drying and wetting surfaces bound an infinite number of scanning
surfaces in the 𝑠 - 𝑆𝑟 - 𝑣 space. The 3-dimensional SWRS becomes a 2-
dimensional hysteretic curve when plotted in terms of combined suction, which
is a function of the currently applied suction and of the specific volume resulting
from the coupled effect of the mechanical and hydraulic loading. The effect of
442
specific volume is controlled by the parameter 𝜓 similar to the v-SWRC model

and the hysteresis is modelled in a way similar to that adopted by the hysteretic-
SWRC model.
Finally, the elastic soil compressibility with suction, 𝜅𝑠∗ , was coupled with the
SWRC, to model the suction induced irreversible volumetric strains observed in
the laboratory, completing the 3-dimensional perception of the retention
behaviour. The 𝑆𝑟 dependent compressibility 𝜅𝑠∗ is equal to 𝜅𝑠 𝜒 ∙ 𝑆𝑟 𝜔
(𝜅𝑠 is a
model parameter in the original version of the model by Georgiadis, 2003),
where 𝜒 and 𝜔 are fitting parameters. The latter parameter controls the
influence of the degree of saturation on the soil compressibility. The proposed
relationship realistically accounts for the 3-dimensional essence of the retention
relationship; the degree of saturation, the specific volume and the suction are
coupled and as such they should be treated inseparably.
The new developments were employed in the numerical analyses of an

unsaturated soil slope subjected to seasonal changes of suction. Meteorological
data with reference to rainfall and potential evapotranspiration were employed
in the coupled consolidation analyses and the effect of the SWRC model
adopted was investigated by comparing the resulting soil movements. The
results of the analyses demonstrated that:
 monotonic SWRC models, such as the Van Genuchten (1980) and the v-
SWRC model, cannot adequately reproduce the expected cyclic
volumetric soil response to seasonal variations of suction;
 for the above behaviour to be simulated a hysteretic model is required –

the hysteretic-SWRC and the v-hysteretic-SWRC models predict
shrinkage during summer and swelling during winter;
 larger displacements at the crest and the middle of the slope were
produced for higher values of the parameter 𝜓;
 larger values of the parameter 𝜔 were associated with smaller cyclic

displacements at the crest and the middle of the slope;
443
 the effect of the parameter 𝜔 was reduced when employed in

combination with high values of the parameter 𝜓;
 shrinkage and settlement showed a clear tendency to increase with time

when the hysteretic models were used;
 the coefficient of earth pressure at rest, 𝐾0 , had no effect on the

computed displacements at the crest and the middle of the slope
whereas the value of OCR employed affected the results during the first
wetting cycle;
 modelling of the root growth has a greater effect on the results for
shallow initial groundwater table depths;
 the effect of root growth is more significant in the regions of the finite
element mesh where conditions may vary between full and partial
saturation;
 the permeability model adopted did not have a significant effect on the
computed displacements at the crest and the middle of the slope.
The results of the numerical analyses undertaken during the duration of this
research project demonstrated that the choice of the SWRC model has a
significant impact on the numerical analysis performed. Therefore, realistic
modelling of the retention relationship is an aspect worth investigating further.
8.5 Overview of the mechanical and hydraulic modelling of

unsaturated soil behaviour in ICFEP
Following the modifications made to the models developed by Georgiadis

(2003) and the development of the SWRC models during this research project,
the behaviour of unsaturated soils is now modelled based on the following
concepts:
 a loading-collapse (LC) curve is used in the isotropic stress plane and

induces isotropic yield and collapse due to wetting (it should be noted
444
that a suction increase surface – SI – is also available but was not

employed in the analyses presented in this thesis);
 three options are available concerning the shape of the isotropic

compression line (ICL): a linear, a bi-linear and a curved ICL - the shape
of the projection of the critical state line (CSL) on the isotropic stress
plane is similar to the shape adopted for the ICL;
 the Lagioia et al. (1996) surface is used for the yield and plastic potential
functions in the 𝑝-𝐽 plane on the wet side of the critical state – the flow
rule may either be associated, or non-associated;
 the Hvorslev surface is employed on the dry side of the critical state – the
flow rule is non-associated;
 the Matsuoka-Nakai failure criterion is adopted in the deviatoric plane;
 the increase of apparent cohesion with suction is controlled by the

SWRC and thus the position of the CSL in the 𝑝-𝐽 plane is coupled with
the hydraulic component (a linear increase of apparent cohesion is also
available but is considered to be unrealistic);
 four SWRC models are available and may be used in combination with
the constitutive model: the simple Van Genuchten (1980), the v-SWRC,
the hysteretic-SWRC and the v-hysteretic-SWRC models;
 the soil compressibility due to changes in suction, 𝜅𝑠∗ , is a function of the

degree of saturation – in this way coupling between the volumetric soil
behaviour and the hydraulic behaviour is achieved.
The modifications proposed in this thesis can be implemented into existing BBM
type constitutive models, in order to couple the mechanical and hydraulic
components of behaviour (increase of apparent cohesion and soil
compressibility with suction are 𝑆𝑟 dependent), to improve the prediction of the
peak deviatoric stress in highly overconsolidated soils and to model the SWRC.
445
8.6 Suggestions for future developments in the modelling of

unsaturated soils with ICFEP
As discussed above, the parameters associated with the Hvorslev surface were
assumed to be independent of the suction level. This assumption was made
due to lack of experimental evidence indicating otherwise and it may be found in
the future that this simplification is not realistic. Refinement of the expression
will then be required and may be referred to the parameters controlling the
shape and the position of the Hvorslev surface (𝛼𝐻𝑉 and 𝑛), the parameters
controlling the plastic potential function (𝛽𝐻𝑉 and 𝑚) or all of the above.
Further improvement of the SWRC models developed herein is suggested to be

focused on the advancement of the expression employed for the primary
surfaces (an improved relationship was proposed in Chapter 5). Increasing the
sophistication of the expression adopted for the scanning paths is less urgent;
the simple geometric shape assumed can adequately simulate the behaviour
observed, requiring no fitting parameter. Alternatively, an expression similar to
the one proposed by Li (2005) may be found more appropriate in the future.
Although the expression for the soil compressibility 𝜅𝑠∗ was shown to fit the
experimental data for various different types of soils, further calibration is
required. This is because the data employed in the calibration either did not
include the primary paths (data by Sharma, 1998) or did not include scanning
paths (data by Cunningham, 2000). Further research may show that the
proposed expression is not always adequate; nevertheless, it is the author’s
opinion that the soil compressibility with suction is some function of the degree
of saturation and when modelled as such realistically accounts for the
irreversible changes in volume observed in the laboratory upon cyclic changes
of suction.
As pointed out in Section 8.2, there is a need for experimental data relative to
the shearing behaviour of unsaturated soils. The following aspects of modelling
may require further development, once such data become available:
 the shape of the CSL and its evolution with suction – as already
discussed, the current shape of the CSL (i.e. straight line) and the
446
assumption that its slope remains constant with suction are based on
limited laboratory data;
 the increase of apparent cohesion with suction – although the

assumption that the increase of apparent cohesion is a function of the
degree of saturation seems realistic, the actual relationship employed
has never been investigated in great depth.
Finally, the models by Georgiadis (2003), which formed the basis of this
research, are based on the behaviour of isotropically consolidated soils. As
discussed in Chapter 2 the behaviour of one-dimensionally prepared samples is
anisotropic and therefore, significantly different compared to the behaviour of
isotropically prepared samples. Further development of the models in order to
account for anisotropy involves rotation of the yield and plastic potential
surfaces in the 𝑝-𝐽 plane.
8.7 Suggestions for future finite element analyses with ICFEP
The stability of slopes in overconsolidated unsaturated soils was studied for a

homogeneous artificial silty clay. It was shown that the factor of safety increases
with the OCR value and with the ratio R of the groundwater table depth over the
excavation depth. It would be interesting to repeat the unsaturated analyses for
OCR values ranging from 5.0 to 1.0 and for various ratios R (including 0.0) and
identify safe and unsafe combinations of OCR and R. The parametric study can
be extended to include the inclination of the slope and the value of the angle of
shearing resistance in order to provide general guidelines relevant to slope
stability.
The behaviour of unsaturated soil slopes under seasonal changes of suction

was also investigated, considering the same artificial silty clay, which was
characterised by a large critical state angle of shearing resistance, 𝜑𝑐𝑠 (in
excess of 30o). Due to the available strength the slope was stable throughout
the analyses. It would be interesting to repeat the analyses for lower values of
𝜑𝑐𝑠 and of OCR and study the effect of the SWRC model employed and the
447
effect of the parameters 𝜓 and 𝜔 on the progressive failure of unsaturated soil

slopes.
The progressive failure of UK railway embankments due to seasonal changes of

suction was studied by Nyambayo (2003). Nonetheless, the embankment fill
was assumed to be fully saturated. The earlier study could be repeated
adopting the constitutive and SWRC models for unsaturated soils employed in
this thesis.
Desiccation of unsaturated soils caused by vegetation is speculated to have a

similar effect on the coefficient of earth pressure at rest, 𝐾0 , as the OCR.
Nevertheless, this effect has not been investigated. The constitutive and SWRC
models presented in this thesis in combination with the hydraulic boundary
conditions available in ICFEP offer a tool for such an investigation to be
performed.
448
references
Abramson, L. W., Lee, T. S., Sharma, S. & Boyce, G. M. (1995) Slope stability
and stabilisation methods. New York, John Wiley and Sons.
Alonso, E. E., Gens, A. & Hight, D. W. (1987) Special problems soils. General
report. Proceedings of the 9th Europ. Conf. Soil Mech., Dublin. pp. 1987-
1146.
Alonso, E. E., Gens, A. & Josa, A. (1990) A constitutive model for partially
saturated soils. Géotechnique, 40 (3), 405-430.
Alonso, E. E., Lloret, A., Gens, A. & Yang, D. Q. (1995) Experimental behaviour
of highly expansive double-structure clay. Proceedings of the 1st Int.
Conf. Unsaturated Soils, Paris. 1, pp. 11-16.
Alonso, E. E., Vaunat, J. & Gens, A. (1999) Modelling the mechanical behaviour
of expansive clays. Engineering Geology, 54 (1-2), 173-183.
Alonso, E. E., Gens, A. & Delahaye, C. H. (2003) Influence of rainfall on the
deformation and stability of a slope in overconsolidated clays: A case
study. Hydrogeology Journal, 11 (1), 174-192.
Bear, J. (1972) Dynamics of fluids in porous media. Dover Publications. Inc.
Bishop, A. W. (1955) The use of slip circle in the stability analysis of slopes.
Geotechnique, 5 (1), 7-17.
Bishop, A. W. (1959) The principle of effective stress. Teknish Ukeblad, 106
(39), 859-863.
449
references
Bishop, A. W., Alpan, I., Blight, G. E. & Donald., I. B. (1960) Factors controlling
the strength of partly saturated cohesive soils. ASCE Res. Conf. Shear
Strength of Cohesive Soils (Univ. of Colorado, Boulder), 503-532.
Bishop, A. W. & Bjerrum, L. (1960) The relevance of the triaxial test to the
solution of stability problems. Proceedings of the ASCE research
conference on shear strength of cohesive soils. Boulder, pp. 437-501.
Bishop, A. W. & Blight, G. E. (1963) Some aspects of effective stress in
saturated and partly saturated soils. Géotechnique, 13 (3), 177.
Blackwell, E. (2010) Accounting for factors of safety to geotechnical parameters
in retaining wall numerical analysis. MSc Thesis. Imperial College
London, UK.
Bolzon, G., Schrefler, B. A. & Zienkiewicz, O. C. (1996) Elastoplastic soil
constitutive laws generalized to partially saturated states. Géotechnique,
46 (2), 279-289.
Buscarnera, G. & Nova, R. (2009) An elastoplastic strainhardening model for
soil allowing for hydraulic bonding-debonding effects. International
Journal for Numerical and Analytical Methods in Geomechanics, 33 (8),
1055-1086.
Chandler, R. J. (1984) Recent european experience of landslides in over-
consolidated clay and soft rock. In: 4th Int. Sym. Landslides, Toronto. 1,
pp. 61-81.
Chiu, C. F. & Ng, C. W. W. (2003) A state-dependent elasto-plastic model for
saturated and unsaturated soils. Géotechnique, 53 (9), 809-829.
450
references
Cui, Y. J., Delage, P. & Sultan, N. (1995) An elasto-plastic model for compacted
soils. Unsaturated Soils, Vols 1 and 2, 703-709.
Cui, Y. J. & Delage, P. (1996) Yielding and plastic behaviour of an unsaturated
compacted silt. Géotechnique, 46 (2), 291-311.
Cunningham, M. R. (2000) The mechanical behaviour of compacted silty clay.
PhD Thesis. Imperial College, University of London, UK.
Dawson, E. M., Roth, W. H. & Drescher, A. (1999) Slope stability analysis by
strength reduction. Géotechnique, 49 (6), 835-840.
Dineen, K. (1997) The influence of suction on compressibility and swelling. PhD
Thesis. Imperial College, University of London, UK.
Duncan, J. M. & Dunlop, P. (1969) Slopes in stiff fissured clays and soils. J Soil
Mech Found Div, ASCE, 95, 467-492.
Duncan, J. M. (1996) State of the art: Limit equilibrium and finite-element
analysis of slopes. Journal of Geotechnical Engineering-Asce, 122 (7),
577-596.
Erratum (2003) Géotechnique, 53 (9), 844.
Escario, V. & Saez, J. (1973) Measurement of the properties of swelling and
collapsing soils under controlled suction. Proceedings of the 3rd Int. Conf.
Expansive Soils, Haifa. pp. 195-200.
Estabragh, A. R. & Javadi, A. A. (2008) Critical state for overconsolidated
unsaturated silty soil. Canadian Geotechnical Journal, 45 (3), 408-420.
451
references
Feddes, R. A., Kowalk, P. J. & Zaradny, H. (1978) Simulation of field water use
and crop yield. New York, John Wiley & Sons.
Fellenius, W. (1936) Calculation of the stability of earth dams. Proceedings of
the 2nd Congress on Large Dams, Washington, D.C. U.S. Government
Printing Office, 4.
Fisher, R. A. (1926) On the capillary forces in an ideal soil: Correction of the
formulae given by W. B. Haines. J. Agric. Sc., 16 (3), 492-505.
Fredlund, D. G. & Morgenstern, N. R. (1977) Stress state variables for
unsaturated soils. J. Geotech. Eng. Div., ASCE, 103 (GT5), 447-466.
Fredlund, D. G., Morgenstern, N. R. & Widger, R. A. (1978) Shear-strength of
unsaturated soils. Canadian Geotechnical Journal, 15 (3), 313-321.
Fredlund, D. G. (1981) The shear strength of unsaturated soils and its
relationship to slope stability problems in Hong Kong. Hong Kong
engineer, April, 37-45.
Fredlund, D. G. & Rahardjo, H. (1993) Soil mechanics for unsaturated soils.
Wiley Interscience Publication.
Fredlund, D. G. & Xing, A. Q. (1994) Equations for the soil-water characteristic
curve. Canadian Geotechnical Journal, 31 (4), 521-532.
Gallipoli, D., Gens, A., Sharma, R. & Vaunat, J. (2003a) An elasto-plastic model
for unsaturated soil incorporating the effects of suction and degree of
saturation on mechanical behaviour. Géotechnique, 53 (1), 123-135.
452
references
Gallipoli, D., Wheeler, S. J. & Karstunen, M. (2003b) Modelling the variation of
degree of saturation in a deformable unsaturated soil. Géotechnique, 53
(1), 105-112.
Gan, J. K. M. & Fredlund, D. G. (1996) Shear strength characteristics of two
saprolitic soils. Canadian Geotechnical Journal, 33 (4), 595-609.
Geiser, F., Laloui, L. & Vulliet, L. (2000) Modelling the behaviour of unsaturated
silt. In: Tarantino, A. and Mancuso, C. (eds.) Experimental Evidence and
Theoretical Approaches in Unsaturated Soils, Balkema, Rotterdam. pp.
155-175.
Gens, A. & Alonso, E. E. (1992) A framework for the behavior of unsaturated
expansive clays. Canadian Geotechnical Journal, 29 (6), 1013-1032.
Gens, A. (1995) "Constitutive laws. Modern issues in non-saturated soils",
Springer-Vergal Wien, pp. 129-158 (reported in Gallipoli et al., 2003a).
Gens, A., Sanchez, M. & Sheng, D. (2006) On constitutive modelling of
unsaturated soils. Acta geotechnica, 1 (3), 137-147.
Gens, A., Guimaraes, L. D., Sanchez, M. & Sheng, D. (2008) Developments in
modelling the generalised behaviour of unsaturated soils. Unsaturated
Soils: Advances in Geo-Engineering, 53-61.
Gens, A. (2010) Soil-environment interactions in geotechnical engineering.
Géotechnique, 60 (1), 3-74.
453
references
Georgiadis, K. (2003) Development, implementation and application of partially
saturated soil models in finite element analysis. PhD Thesis. Imperial
College, University of London, UK.
Georgiadis, K., Potts, D. M. & Zdravkovic, L. (2005) Three-dimensional
constitutive model for partially and fully saturated soils. International
Journal of Geomechanics, 5 (3), 244-255.
Georgiadis, K., Potts, D. M. & Zdravkovic, L. (2007) Numerical study of the
stability of a partially saturated soil slope. In: Pande, G. N. and
Pietruszczak, S. (eds.) X International Symposium on Numerical Models
in Geomechanics (NUMOG X), Rhodes, Greece. Balkema. pp. 545-550.
Gray, W. G. & Schrefler, B. A. (2001) Thermodynamic approach to effective
stress in partially saturated porous media. European Journal of
Mechanics A/Solids, 20 (4), 521-538.
Griffiths, D. V. & Lane, P. A. (1999) Slope stability analysis by finite elements.
Houlsby, G. T. (1997) The work input to an unsaturated granular material.
Janbu, N. (1973) Slope stability computations. In: Embankment Dam
Engineering, Wiley, New York. pp. 47-88.
Jennings, J. E. B. & Burland, J. B. (1962) Limitations to the use of effective
stresses in partly saturated soils. Géotechnique, 12 (2), 125-144.
454
references
Jommi, C. (2000) Remarks on the constitutive modelling of unsaturated soils.
Experimental Evidence and Theoretical Approaches in Unsaturated
Soils, 139-153.
Josa, A., Alonso, E. E., Gens, A. & Lloret, A. (1987) Stress-strain behaviour of
partially satuarted soils. Proceedings of the 9th Europ. Conf. Soil Mech.
Found. Eng. 2, pp. 561-564.
Josa, A., Balmaceda, A., Gens, A. & Alonso, E. E. (1992) An elastoplastic
model for partially saturated soil exhibiting a maximum collapse. In:
Owen, D. R. J., Onate, E. and Hinton, E. (eds.) Computation plasticity III,
Pineridge Press, Swansea. 1, pp. 815-826.
Jotisankasa, A. (2005) Collapse behaviour of compacted silty clay. PhD Thesis.
Imperial College, University of London, UK.
Khalili, N. & Khabbaz, M. H. (1998) A unique relationship for chi for the
determination of the shear strength of unsaturated soils. Géotechnique,
48 (5), 681-687.
Kim, M. K. & Lade, P. V. (1988a) Single hardening constitutive model for
frictional materials I: Plastic potential function. Computers and
Geotechnics, 5, 307-324.
Kim, M. K. & Lade, P. V. (1988b) Single hardening constitutive model for
frictional materials II: Yield criterion and plastic work contours.
Computers and Geotechnics, 6, 13-29.
Kohgo, Y., Nakano, M. & Miyazaki, T. (1993) Theoretical aspects of constitutive
modelling for unsaturated soils. Soils and Foundations, 33 (4), 49-63
455
references
Kohler, R. & Hofstetter, G. (2008) A cap model for partially saturated soils.
International Journal for Numerical and Analytical Methods in
Geomechanics, 32 (8), 981-1004.
Lagioia, R., Puzrin, A. M. & Potts, D. M. (1996) A new versatile expression for
yield and plastic potential surfaces. Computers and Geotechnics, 19 (3),
171-191.
Laloui, L., Geiser, F. & Vulliet, L. (2001) Constitutive modelling of unsaturated
soils. Rev. fr. genie civ., 5 (6), 797-807.
Leroueil, S. (2001) Natural slopes and cuts: Movement and failure mechanisms.
Li, X. S. (2005) Modelling of hysteresis response for arbitrary wetting/drying
paths. Computers and Geotechnics, 32, 133-137.
Liakopoulos, A. C. (1965) Theoretical solution of the unsteady unsaturated flow
problems in soils. Bull. Int. Assoc. Sci. Hydrol., 10, 5-39.
Lloret, A. & Alonso, E. E. (1985) State surfaces for partially saturated soils. In:
11th Int. Conf. Soil MEch. Found. Engng., San Francisco. 2, pp. 557-562.
Lloret, M., Sanchez, M. & Wheeler, S. J. (2009) Generalised elasto-plastic
stress-strain and modified suction-degree of saturation relations of a fully
coupled model. Proceedings of the 4th Asia Pacific conf. Unsaturated
Soils, Newcastle. Taylor & Francis Group, pp. 667-672.
456
references
Loret, B. & Khalili, N. (2000) A three-phase model for unsaturated soils.
Geomechanics, 24 (11), 893-927.
Loret, B. & Khalili, N. (2002) An effective stress elastic-plastic model for
unsaturated porous media. Mechanics of Materials, 34 (2), 97-116.
Lu, N. & Likos, W. J. (2004) Unsaturated soil mechanics. New Jersey, John
Wiley & Sons, Inc.
Lumb, P. (1962) Effect of rain storms on slope stability. In: Symposium on Hong
Kong soils, May 1962, Hong Kong Joint Group of the Institution of Civil,
Mechanical and Electrical Engineers.
Maatouk, A., Leroueil, S. & LaRochelle, P. (1995) Yielding and critical state of a
collapsible unsaturated silty soil. Géotechnique, 45 (3), 465-477.
Masin, D. & Khalili, N. (2008) A hypoplastic model for mechanical response of
unsaturated soils. International Journal for Numerical and Analytical
Methods in Geomechanics, 32 (15), 1903-1926.
Matsuoka, H. & Nakai, T. (1974) Stress-deformation and strength
characteristics of soil under three different principal stresses. Proc. Jap.
Soc. Civ. Eng., 232, 59-70.
Melgarejo, M. L. C. (2004) Laboratory and numerical investigations of soil-water
retention curves. PhD Thesis. Imperial College, University of London,
UK.
457
references
Modaressi, A., Abou-Bekr, N. & Fry, J. J. (1996) Unified approach to model
partially saturated and saturated soil. Unsaturated Soils, Vol 3, 1495-
1502.
Monroy, R. (2006) The influence of load and suction changes on the volumetric
behaviour of compacted London clay. PhD Thesis. Imperial College,
University of London, UK.
Morgenstern, N. R. & Price, V. E. (1965) The analysis of the stability of general
slip surfaces. Geotechnique, 15 (1), 79-93.
Mualem, Y. (1976) A new model for predicting the hydraulic conductivity of
unsaturated porous media. Water Resources Research, 12, 513-522.
Nova, R. (1988) Sinfonietta Classica: An exercise on classical soil modelling. In:
Saada, A. F. and Bianchini, G. F. (eds.) Int. Symp. Constitutive Eq. For
Granular Non-Ceohesive Soils, Cleveland. Rotterdam, Balkema. pp. 501-
520.
Nuth, M. & Laloui, L. (2008) Effective stress concept in unsaturated soils:
Clarification and validation of a unified framework. International Journal
for Numerical and Analytical Methods in Geomechanics, 32 (7), 771-801.
Nyambayo, V. P. (2003) Numerical analysis of evapotranspiration and its
influence on embankments. PhD Thesis. Imperial College, University of
London, UK.
Nyambayo, V. P. & Potts, D. M. (2010) Numerical simulation of
evapotranspiration using a root water uptake model. Computers and
Geotechnics, 37, 175-186.
458
references
Oberg, A. L. & Sallfors, G. (1997) Determination of shear strength parameters
of unsaturated silts and sands based on the water retention curve.
Geotechnical Testing Journal, 20 (1), 40-48.
Pedroso, D. M. & Williams, D. J. (2010) A novel approach for modelling soil
water characteristic curves with hysteresis. Computers and Geotechnics,
37 (3), 374-380.
Pereira, J. M., Wong, H., Dubujet, P. & Dangla, P. (2005) Adaptation of existing
behaviour models to unsaturated states: Application to cjs model.
Geomechanics, 29 (11), 1127-1155.
Pirone, M. (2009) Analysis of slope failure mechanism in unsaturated
pyroclastic soils, based on testing site monitoring. PhD. University of
Naples, Federico II, Italy.
Potts, D. M. & Zdravkovic, L. (1999) Finite element analysis in geotechnical
engineering: Theory. London, Thomas Telford.
Potts, D. M. & Zdravkovic, L. (2011) Application of partial factors of safety in
numerical analysis of bearing capacity. Proceedings of the 2nd
International Symposium in Computational Geomechanics (accepted for
publication), Cavtat, Croatia.
Press, W. H., Teukolsky, S. A., Vetterling, W. T. & Flannery, B. P. (2007)
Numerical recipies: The art of scientific computing. Cambridge University
Press.
459
references
Rahardjo, H., Nio, A. S., Leong, E. C. & Song, N. Y. (2010) Effects of
groundwater table position and soil properties on stability of slope during
rainfall. Journal of Geotechnical and Geoenvironmental Engineering, 136
(11), 1555-1564.
Rampino, C., Mancuso, C. & Vinale, F. (1999) Laboratory testing on an
unsaturated soil: Equipment, procedures, and first experimental results.
Canadian Geotechnical Journal, 36 (1), 1-12.
Rampino, C., Mancuso, C. & Vinale, F. (2000) Experimental behaviour and
modelling of an unsaturated compacted soil. Canadian Geotechnical
Journal, 37 (4), 748-763.
Ridley, A. M. & Burland, J. B. (1993) A new instrument for the measurement of
soil-moisture suction. Géotechnique, 43 (2), 321-324.
Roscoe, K. H. & Schofield, A. N. (1963) Mechanical behaviour of an idealised
'wet' clay. In: 2nd ECSMFE, Wiesbaden. 1, pp. 47-54.
Roscoe, K. H. & Burland, J. B. (1968) On the generalised stress-strain
behaviour of 'wet' clay. In: Heyman, J. and Leckie, F. A. (eds.)
Engineering plasticity, Cambridge, UK, Cambridge University Press. pp.
535-609.
Russell, A. R. & Khalili, N. (2006) A unified bounding surface plasticity model for
unsaturated soils. International Journal for Numerical and Analytical
Sanchez, M., Gens, A., Guimaraes, L. D. & Olivella, S. (2005) A double
structure generalized plasticity model for expansive materials.
460
references
Geomechanics, 29 (8), 751-787.
Santagiuliana, R. & Schrefler, B. A. (2006) Enhancing the bolzon-schrefler-
zienkiewicz constitutive model for partially saturated soil. Transport in
Porous Media, 65 (1), 1-30.
Schofield, A. N. & Wroth, C. P. (1968) Critical state soil mechanics. London,
McGraw Hill.
Schweiger, H. F. (2005) Application of FEM to ULS design (Eurocodes) in
surface and near surface geotechnical stuctures. Proceedings of the 11th
Int. Conf. Computer Methods and Advances in Geomechanics, Bologna.
Patron Editore.
Sharma, R. S. (1998) Mechanical behaviour of unsaturated highly expansive
clays. PhD Thesis. University of Oxford, Oxford.
Shaw, E. M. (1994) Hydrology in practice. 3rd edn. London, UK, Chapman &
Hall.
Sheng, D., Sloan, S. W. & Gens, A. (2004) A constitutive model for unsaturated
soils: Thermomechanical and computational aspects. Computational
Mechanics, 33 (6), 453-465.
Sheng, D., Fredlund, D. G. & Gens, A. (2008) A new modelling approach for
unsaturated soils using independent stress variables. Canadian
Geotechnical Journal, 45 (4), 511-534.
461
references
Sivakumar, V. & Wheeler, S. J. (2000) Influence of compaction procedure on
the mechanical behaviour of an unsaturated compacted clay - part 1:
Wetting and isotropic compression. Géotechnique, 50 (4), 359-368.
Sivakumar, V., Sivakumar, R., Boyd, J. & Mackinnon, P. (2010a) Mechanical
behaviour of unsaturated kaolin (with isotropic and anisotropic stress
history). Part 2: Performance under shear loading. Géotechnique, 60 (8),
595-609.
Sivakumar, V., Sivakumar, R., Murray, E. J., Mackinnon, P. & Boyd, J. (2010b)
Mechanical behaviour of unsaturated kaolin (with isotropic and
anisotropic stress history). Part 1: Wetting and compression behaviour.
Smith, P. G. C. (2003) Numerical analysis of infiltration into partially saturated
soil slopes. PhD Thesis. Imperial College, University of London, UK.
Snitbhan, N. & Chen, W. F. (1976) Elastic-plastic large deformation analysis of
soil slopes. Comput. struc., 9, 576-577.
Spencer, E. (1967) A method of analysis of the stability of embankments
assuming parallel interslice forces. Geotechnique, 17 (1), 11-26.
Sun, D., Sheng, D. & Sloan, S. W. (2007a) Elasto-plastic modelling of hydraulic
and stress-strain behaviour of unsaturated soils. Mechanics of
Geomaterials, 39, 212-221.
Sun, D. A., Matsuoka, H., Cui, H. B. & Xu, Y. F. (2003) Three-dimensional
elasto-plastic model for unsaturated compacted soils with different initial
462
references
densities. International Journal for Numerical and Analytical Methods in
Geomechanics, 27 (12), 1079-1098.
Sun, D. A., Sheng, D. C., Cui, H. B. & Sloan, S. W. (2007b) A density-
dependent elastoplastic hydro-mechanical model for unsaturated
compacted soils. International Journal for Numerical and Analytical
Sun, D. A., Sheng, D. C. & Sloan, S. W. (2007c) Elastoplastic modelling of
hydraulic and stress-strain behaviour of unsaturated soils. Mechanics of
Materials, 39 (3), 212-221.
Tamagnini, R. (2004) An extended cam-clay model for unsaturated soils with
hydraulic hysteresis. Géotechnique, 54 (3), 223-228.
Tarantino, A., Mongiovi, L. & Bosco, G. (2000) An experimental investigation on
the independent isotropic stress variables for unsaturated soils.
Tarantino, A. & Tombolato, S. (2005) Coupling of hydraulic and mechanical
behaviour in unsaturated compacted clay. Géotechnique, 55 (4), 307-
317.
Tarantino, A. (2007) A possible critical state framework for unsaturated
compacted soils. Géotechnique, 57 (4), 385-389.
Tarantino, A. (2009) A water retention model for deformable soils.
463
references
Thu, T. M., Rahardjo, H. & Leong, E. C. (2007) Elastoplastic model for
unsaturated soil with incorporation of the soil-water characteristic curve.
Canadian Geotechnical Journal, 44 (1), 67-77.
Toll, D. G. (1990) A framework for unsaturated soil behavior. Géotechnique, 40
(1), 31-44.
Toll, D. G. (1995) A conceptual model for the drying and wetting of soil. In:
Alonso, E. E. and Delage, P. (eds.) Unsaturated Soils, Balkema,
Rotterdam. pp. 805-810.
Toll, D. G. & Ong, B. H. (2003) Critical-state parameters for an unsaturated
residual sandy clay. Géotechnique, 53 (1), 93-103.
Van Eekelen, H. A. M. (1980) Iostropic yield surfaces in three dimensions for
use in soil mechanics. Int. Jnl. Num. Anal. Meth. Geomech., 4, 89-101.
Van Genuchten, M. T. (1980) A closed-form equation for predicting the
hydarulic conductivity of unsaturated soils. Soil Sci. Soc. Am. J., 44, 892-
898.
Vanapalli, S. K., Fredlund, D. G., Pufahl, D. E. & Clifton, A. W. (1996) Model for
the prediction of shear strength with respect to soil suction. Canadian
Geotechnical Journal, 33 (3), 379-392.
Vaunat, J., Romero, E. & Jommi, C. (2000a) An elastoplastic hydro-mechanical
model for unsaturated soils. In: Tarantino, A. and Mancuso, C. (eds.)
Experimental Evidence and Theoretical Approaches in Unsaturated
Soils, Balkema, Rotterdam. pp. 121-138.
464
references
Vaunat, J., Romero, E. & Jommi, C. (2000b) An elasto-plastic hydromechanical
model for unsaturated soils. Proceedings of the Experimental evidence
and theoretical approaches in unsaturated soils. Rotterdam: Balkema,
pp. 121-138.
Wheeler, S. J. & Sivakumar, V. (1993) Development and application of a
critical-state model for unsaturated soil. Predictive Soil Mechanics, 709-
728.
Wheeler, S. J. & Sivakumar, V. (1995) An elasto-plastic critical state framework
for unsaturated soil. Géotechnique, 45 (1), 35-53.
Wheeler, S. J. (1996) Inclusion of specific water volume within an elasto-plastic
model for unsaturated soil. Canadian Geotechnical Journal, 33 (1), 42-
57.
Wheeler, S. J. & Karube, D. (1996) State of the art report: Constitutive
modelling. Proceedings of the 1st Int. Conf. on Unsaturated soils, Paris.
Balkema, Rotterdam, 3, pp. 1323-1356.
Wheeler, S. J. & Sivakumar, V. (2000) Influence of compaction procedure on
the mechanical behaviour of an unsaturated compacted clay. Part 2.
Shearing and constitutive modelling. Géotechnique, 50 (4), 369-376.
Wheeler, S. J., Gallipoli, D. & Karstunen, M. (2002) Comments on use of the
Barcelona Basic Model for unsaturated soils. International Journal for
Numerical and Analytical Methods in Geomechanics, 26 (15), 1561-
1571.
465
references
Wheeler, S. J., Sharma, R. S. & Buisson, M. S. R. (2003) Coupling of hydraulic
hysteresis and stress-strain behaviour in unsaturated soils.
Wong, T. T., Fredlund, D. G. & Krahn, J. (1998) A numerical study of coupled
consolidation in unsaturated soils. Canadian Geotechnical Journal, 35
(6), 926-937.
Zhu, J. G. & Yin, J. H. (2000) Strain-rate-dependent stress-strain behaviour of
overconsolidated Hong Kong marine clay. Canadian Geotechnical
Journal, 37, 1271-1282.
Zienkiewicz, O. C., Humpheson, C. & Lewis, R. W. (1975) Associated and non-
associated visco-plasticity and plasticity in soil mechanics.
Zienkiewicz, O. C. & Taylor, R. L. (1989) The finite element method. London,
McGraw Hill.
466
appendix A
A.1 Supplementary derivatives for the implementation of the

Hvorslev surface
𝜕 𝐽 2𝜂 𝑔
The derivative presented in Equation 4.48, was calculated from the cubic
𝜕𝜃
Equation 3.36, as follows:
3 2
2
𝐾= ∙ 𝐶𝑔 ∙ 𝑠𝑖𝑛 −3𝜃 ∙ 𝐽2𝜂𝑔 + 𝐶𝑔 − 3 ∙ 𝐽2𝜂𝑔 − 𝐶𝑔 − 9 = 0 (A.1)
27
differentiating in terms of 𝜃:
𝜕𝐾 𝜕𝜃 𝜕𝐾 𝜕 𝐽2𝜂𝑔 𝜕𝐾 𝜕𝐶
+ ∙ + =0 (A.2)
𝜕𝜃 𝜕𝜃 𝜕 𝐽2𝜂𝑔 𝜕𝜃 𝜕𝐶 𝜕𝜃
𝜕𝜃 𝜕𝐶
= 1.0 and 𝜕𝜃 = 0.0. Therefore:
𝜕𝜃
𝜕𝐾
𝜕 𝐽2𝜂𝑔 𝜕𝜃
=− (A.3)
𝜕𝜃 𝜕𝐾
𝜕 𝐽2𝜂𝑔
which gives:
467
appendix A
3
𝜕𝐾 6
= 𝐶𝑔 ∙ 𝑐𝑜𝑠 −3𝜃 ∙ 𝐽2𝜂𝑔 (A.4)
𝜕𝜃 27
2
𝜕𝐾 6
= 𝐶𝑔 ∙ 𝑠𝑖𝑛 −3𝜃 ∙ 𝐽2𝜂𝑔 + 2 ∙ 𝐶 − 3 ∙ 𝐽2𝜂𝑔 (A.5)
𝜕 𝐽2𝜂𝑔 27
The derivative is then calculated to be:
3 2
𝜕 𝐽2𝜂𝑔 𝐶𝑔 ∙ 𝑐𝑜𝑠 −3𝜃 ∙ 𝐽2𝜂𝑔
27
=− (A.6)
𝜕𝜃 3
𝐶𝑔 ∙ 𝑠𝑖𝑛 −3𝜃 ∙ 𝐽2𝜂𝑔 + 𝐶 − 3
27
𝜕 𝐽 2𝜂 𝑓
The derivative presented in Equation 4.49 was calculated in a similar
𝜕𝜃
way.
𝜕 𝐽 2𝜂 𝑖
The derivative required for the calculation of the derivative of the yield
𝜕 𝐶𝑖
function with respect to the Factor of Safety (Equation 4.93), was evaluated
from Equation A.1 in a similar way:
𝜕𝐾 𝜕𝐶𝑖 𝜕𝐾 𝜕 𝐽2𝜂𝑖 𝜕𝐾 𝜕𝜃
+ ∙ + =0 (A.7)
𝜕𝐶𝑖 𝜕𝐶𝑖 𝜕 𝐽2𝜂𝑖 𝜕𝐶𝑖 𝜕𝜃 𝜕𝐶𝑖
𝜕 𝐶𝑖 𝜕𝜃
= 1.0 and 𝜕𝐶 = 0.0. Therefore:
𝜕 𝐶𝑖 𝑖
𝜕𝐾
𝜕 𝐽2𝜂𝑖 𝜕𝐶𝑖
=− (A.8)
𝜕𝐶𝑖 𝜕𝐾
𝜕 𝐽2𝜂𝑖
where:
3 2
𝜕𝐾 2
= 𝑠𝑖𝑛 −3𝜃 𝐽2𝜂𝑖 + 𝐽2𝜂𝑖 −1 (A.9)
𝜕𝐶𝑖 27
𝜕𝐾
and is given by Equation A.5.The derivative is then calculated to be:
𝜕 𝐽 2𝜂 𝑖
468
appendix A
2 3 2
𝜕 𝐽2𝜂𝑖 𝑠𝑖𝑛 −3𝜃 𝐽2𝜂𝑖 + 𝐽2𝜂𝑖 − 1
1 27
=− ∙ (A.10)
𝜕𝐶𝑖 2 𝐽2𝜂𝑖 3
𝐶𝑖 ∙ 𝑠𝑖𝑛 −3𝜃 ∙ 𝐽2𝜂𝑖 + 𝐶𝑖 − 3
27
𝜕𝑝
The derivative 𝜕 𝑝0∗ for Model 2 (Equation 4.63) was calculated from the plastic
0
hardening parameter Equation 4.52, as follows:
1− 0 𝑝0
𝜆 0 −𝜅
𝑝0∗ = −𝑝𝑐 ∙ 𝑥 , 𝑥=− (A.11)
𝑝𝑐
1− 0 (A.12)
𝜆 0 −𝜅
𝐶= 𝑝0∗ 𝑐
+𝑝 ∙𝑥 =0
and:
𝜕𝐶 𝜕𝑝0∗ 𝜕𝐶 𝜕𝑝0
− ∙ =0 (A.13)
𝜕𝑝0∗ 𝜕𝑝0∗ 𝜕𝑝0 𝜕𝑝0∗
𝜕𝐶
𝜕𝑝0 𝜕𝑝0∗
=− (A.14)
𝜕𝑝0∗ 𝜕𝐶
𝜕𝑝0
where:
𝜕𝐶
=1 (A.15)
𝜕𝑝0∗
and:
𝜕𝐶 𝜕𝐶 𝜕𝑥
= ∙ (A.16)
𝜕𝑝0 𝜕𝑥 𝜕𝑝0
𝛼0 ∙ 𝑥 −𝑏
𝛼 ∙𝑥 −𝑏 1−
𝜕𝐶 1− 0 −𝛼0 𝜆 0 −𝜅
𝜆 0 −𝜅
= 𝑝𝑐 ∙ 𝑥 ∙ ∙ −𝑏 𝑥 −𝑏−1 ln 𝑥 + ∙1 (A.17)
𝜕𝑥 𝜆 0 −𝜅 𝑥
𝜕𝐶 𝑝0∗ 1
=− ∙ ∙ 𝛼0 ∙ 𝑥 −𝑏 b ∙ ln 𝑥 − 1 + 𝜆 0 − 𝜅 (A.18)
𝜕𝑥 𝑥 𝜆 0 −𝜅
469
appendix A
𝜕𝑥 1
=− 𝑐 (A.19)
𝜕𝑝0 𝑝
Finally:
𝜕𝑝0 𝑝0 𝜆 0 −𝜅
= ∙
𝜕𝑝0∗ 𝑝0∗ 𝑝 −𝑏 𝑝 (A.20)
𝛼0 − 𝑝0𝑐 ∙ 𝑏𝑙𝑛 − 𝑝0𝑐 − 1 + 𝜆 0 − 𝜅
A.2 Development and implementation of the newly introduced

Hvorslev surface in combination with the modified Cam-clay
model
A surface similar to the one developed in Chapter 4 to substitute for the yield
and plastic potential expressions adopted by the unsaturated soil models, on
the dry side of the critical state, was derived for the modified Cam-clay model
(implemented in ICFEP as detailed in Potts & Zdravkovic, 1999).
The equation for the yield function on the dry side becomes:
′ 𝑛
𝐽 𝑎𝐻𝑉 𝑎𝐻𝑉 𝑝𝑐𝑠
𝐹= − − 1 − ∙ =0 (A.21)
𝑔 𝜃 ∙ 𝑝′ 𝑔 𝜃 𝑔 𝜃 𝑝′
where 0 ≤ 𝑛 < 1 and 𝑔 𝜃 is Lode’s angle, given by Equation 4.2.
The associated plastic potential function is:
𝑛
𝐽 𝑎𝐻𝑉 𝑝𝑐′ 𝑎𝐻𝑉 𝑝𝑐′ 𝑝𝑐𝑠
′
𝐺= − ∙ − 1 − ∙ ′∙ −
𝑔 𝜃 ∙ 𝑝′ 𝑔 𝜃 𝑝′ 𝑔 𝜃 𝑝 𝑝𝑐′
𝑚 (A.22)
𝑝′ − 𝑝𝑐′ 𝛽𝐻𝑉 ′
𝑝𝑐𝑠 − 𝑝𝑐′
− ∙ ∙ =0
𝑝′ 𝑔 𝜃 ′
𝑝𝑐𝑠
where 𝑝𝑐′ is the current value of mean effective stress.
The derivatives of the above two expressions, required for their implementation
into ICFEP, are given below (note that the sign convention is tension positive):
470
appendix A
′ 𝑛−1
𝜕𝐹 𝐽 𝛼𝐻𝑉 𝑛 𝑝𝑐𝑠
= − − 1 ∙ (A.23)
𝜕𝑝 𝑔 𝜃 ∙ 𝑝′ 2 𝑔 𝜃 𝑝′ 𝑝′
𝜕𝐹 1
=− (A.24)
𝜕𝐽 𝑔 𝜃 ∙ 𝑝′
′ 𝑛
𝜕𝐹 𝜕𝐹 𝜕𝑔 𝜃 𝐽 𝛼𝐻𝑉 𝛼𝐻𝑉
𝑝𝑐𝑠 𝜕𝑔 𝜃
= ∙ = 2 + 2 − 2 ∙ ∙ (A.25)
𝜕𝜃 𝜕𝑔 𝜃 𝜕𝜃 𝑔 𝜃 ∙ 𝑝′ 𝑔 𝜃 𝑔 𝜃 𝑝′ 𝜕𝜃
where:
1 ′
cos 𝜃 sin 𝜑𝑐𝑠 − sin 𝜃
𝜕𝑔 𝜃 3 (A.26)
= ′
𝜕𝜃 sin 𝜑𝑐𝑠
and:
′ 𝑛
𝜕𝐺 𝐽 𝛼𝐻𝑉 1 𝛼𝐻𝑉 1 𝑝𝑐𝑠
= + ∙ − −1 ′ ∙ −
𝜕𝑝 𝑔 𝜃 ∙ 𝑝′ 2 𝑔 𝜃 𝑝′ 𝑔 𝜃 𝑝 𝑝′
′ 𝑚 (A.27)
1 𝛽𝐻𝑉 𝑝𝑐𝑠 − 𝑝′
− ′∙ ∙ ′
𝑝 𝑔 𝜃 𝑝𝑐𝑠
𝜕𝐺 1
=− (A.28)
𝜕𝐽 𝑔 𝜃 ∙ 𝑝′
′ 𝑛
𝜕𝐺 𝜕𝐺 𝜕𝑔 𝜃 𝐽 𝛼𝐻𝑉 𝛼𝐻𝑉
𝑝𝑐𝑠 𝜕𝑔 𝜃
= ∙ = 2 + 2 − 2 ∙ ∙ (A.29)
𝜕𝜃 𝜕𝑔 𝜃 𝜕𝜃 𝑔 𝜃 ∙ 𝑝′ 𝑔 𝜃 𝑔 𝜃 𝑝′ 𝜕𝜃
Finally, for the calculation of the plastic hardening parameter A, the following
derivative is also necessary:
′ ′ 𝑛−1
𝜕𝐹 𝜕𝐹 𝜕𝑝𝑐𝑠 𝛼𝐻𝑉 𝑛 𝑝𝑐𝑠 1
= ∙ = −1 ′ ∙ ∙ (A.30)
𝜕𝑝0′ ′
𝜕𝑝𝑐𝑠 𝜕𝑝0′ 𝑔 𝜃 𝑝 𝑝′ 2
′
as 𝑝𝑐𝑠 = 𝑝0′ 2.
471
appendix B
B.1 Supplementary analyses for the v-hysteretic-SWRC model
The performance of the v-hysteretic-SWRC under cyclic changes of suction and

of confining stress was studied simulating the four hypothetical stress paths
illustrated in Figure B-1 and summarised in Table B-1 and employing the model
parameters presented in Table B-2.
The cyclic hydraulic paths were simulated adopting both the v-hysteretic-SWRC
and the hysteretic-SWRC, for the same common model parameters, and
produced significantly different results as illustrated in Figure B-2 for Test 1 and
in Figure B-3 for Test 2. When the suction increase surface SI2 was violated
during drying in Test 2, the elasto-plastic coefficient 𝜆𝑠 was invoked, generating
larger and irrecoverable changes in specific volume (Figure B-3 (b)). For the
analysis with the v-hysteretic-SWRC model, the retention curve was shifted at
the yielding point for Test 2, following a distinct path in comparison with the one
produced during Test 1, as shown in Figure B-4.
The changes in specific volume produced during Test 3 were so small as not to
have an obvious effect on the degree of saturation, 𝑆𝑟 , and the combined
∗
suction, 𝑠𝑒𝑞 , in contrast with Test 4, where unloading coincided with a reversal
∗
in the 𝑠𝑒𝑞 - 𝑆𝑟 relationship (Figure B-5). Nonetheless, the respective changes in
472
appendix B
the degree of saturation, 𝑆𝑟 , are small and the scanning path generated is
practically flat.
Table B-1: Cyclic stress paths

Cyclic stress paths
Equival. Relevant
Suction, 𝑠
Test stress, 𝑝 Type of loading yield
(kPa)
(kPa) surface
1 20–650–20 50 – const. cyclic hydraulic – elastic SI1
2 20–650–20 50 – const. cyclic hydraulic – elasto-plastic SI2
3 20 – const. 50–100–50 cyclic mechanical – elastic LC
4 20 – const. 50–550–50 cyclic mechanical – elasto-plastic LC
Table B-2: Model parameters employed in the simulation of the cyclic stress paths
Soil-water retention
curve (SWRC)
𝛼𝑔,𝑓 0.7 𝜆(0) 0.086 𝜓 1.5
𝜇𝑔,𝑓 0.9999 𝜅 0.005 𝑠𝑎𝑖𝑟 (kPa) 0.0
𝛼𝐻𝑉 0.5 𝑣1 2.120 𝑠0 (kPa) 10000.0
𝑛 0.25 𝛽 0.001 𝛼𝑑 0.005
𝛽𝐻𝑉 0.25 𝑝𝑎𝑡𝑚 (kPa) 100.0 𝛼𝑤 0.02
𝑚 0.5 𝐺/𝑝 15.0 - -
𝑀𝑔,𝑓 1.3039 𝜆𝑠 0.09 - -
𝑟 0.06 𝜅𝑠 0.08 - -
𝑘 𝑆𝑟 - SWRC 𝑝𝑐 (kPa) 1.0 - -
𝑠𝑎𝑖𝑟 (kPa) 0.0 𝑠0 (kPa) 1000000.0 - -
473
appendix B
1000
SI1
600
Suction, s (kPa)
SI2
400
LC Test 1
curve Test 2
200
Test 3
Test 4
0
10 100 1000 10000
Mean equivalent stress, log p (kPa)
Figure B-1: Elastic and elasto-plastic cyclic suction and confining stress paths
1.0
0.8
0.6
0.4
0.2 v-hysteretic-SWRC
hysteretic-SWRC
(a)
0.0
10 100 1000
Equivalent suction, log seq (kPa)
1.0
0.8
0.6
0.4
(b)
hysteretic-SWRC
0.0
1.71 1.7 1.69
Figure B-2: Comparison of hysteretic-SWRC and v-hysteretic-SWRC models in terms of (a) suction, 𝑠,
and (b) specific volume, 𝑣, with degree of saturation, 𝑆𝑟 for cyclic test 1
474
appendix B
1.0
0.8
0.6
0.4
hysteretic-SWRC
(a)
0.0
10 100 1000
1.0
0.8
0.6
0.4
(b)
hysteretic-SWRC
0.0
1.71 1.7 1.69 1.68 1.67 1.66
Figure B-3: Comparison of hysteretic-SWRC and v-hysteretic-SWRC models in terms of (a) suction, 𝑠,
and (b) specific volume, 𝑣, with degree of saturation, 𝑆𝑟 for cyclic test 2
0.46
Test 1
0.44
Test 2
0.42
0.40
0.38
0.36
0.34
0.32
400 500 600 700
Figure B-4: Comparison of the hydraulic paths reproduced by the v-hysteretic-SWRC model for cyclic
tests 1 and 2
475
appendix B
Degree of saturation, Sr 0.876
0.874
0.872
0.870
Test 3
Test 4
(a)
0.868
5 6 7 8 9 10 11 12 13 14 15
Combined equivalent suction, s*eq (kPa)
0.876
Test 3
Test 4
0.874
0.872
0.870
(b)
0.868
1.72 1.7 1.68 1.66 1.64 1.62 1.6 1.58 1.56
Figure B-5: Numerical results in terms of (a) suction, 𝑠, and (b) specific volume, 𝑣, with degree of
saturation, 𝑆𝑟 for cyclic tests 3 and 4
B.2 Validation and calibration of the v-hysteretic-SWRC
The laboratory tests performed by Jotisankasa (2005) on compacted artificial

soil A, already presented in Figure 5-24, were simulated employing the v-
hysteretic-SWRC for the model parameters shown in Table B-3.
The reproduction of the laboratory data, shown in Figure B-6 (a), is thought to
be adequate for values of suction larger than 1000.0 kPa. It should be noted
that the SWRC illustrated in the figure, is the projection of the actual one,
followed in the three-dimensional 𝑠 - 𝑆𝑟 - 𝑣 space, and although the scanning
drying paths are shown in the 𝑠 - 𝑆𝑟 plane to lie outside the primary drying, they
∗
do not exceed the boundaries when plotted in the 𝑠𝑒𝑞 - 𝑆𝑟 plane.
476
appendix B
For validation purposes, the Excel solution for the above scanning paths was
also obtained and is shown to be identical to the numerical ones in Figure B-6
(b).
It is interesting that the hysteretic-SWRC (Figure 5-24) and the v-hysteretic-

SWRC (Figure B-6 (a)) models produced comparable results indicating that the
effect of the hydraulic hysteresis is dominant in comparison with the effect of the
void ratio. Nonetheless, the latter is essential when modelling certain aspects of
the behaviour of unsaturated soils, such as the changes observed in the degree
of saturation, 𝑆𝑟 , during isotropic compression under constant water content –
as explained in Section 5.2.3. Therefore, its inclusion in the hysteretic
relationship for the SWRC is thought to be necessary.
Table B-3: Model parameters employed for the calibration and validation of the v-hysteretic-SWRC
model
v-hysteretic-SWRC model
Param. Value Param. Value
𝑠𝑎𝑖𝑟 1.0 kPa 𝜓 1.7
𝑠0 2.5E+4 kPa 𝛼𝑑 1.1E-3
𝛼𝑤 2.5E-2 𝜅𝑠 0.008
477
appendix B
1.0
B (a)
0.8
0.6
Wetting from initial point A (lab.)
0.4 Drying from full saturation (lab.)
Drying from initial point A (lab.)
A
Drying from point B (lab.)
0.2
Primary paths (ICFEP)
0.0
0.1 1 10 100 1000 10000 100000
1.0
B
(b)
0.8
0.6
0.4
A
0.2 ICFEP
Excel
0.0
0.1 1 10 100 1000 10000 100000
Figure B-6: Hydraulic paths (a) followed by compacted specimens of artificial soil A (data after
Jotisankasa, 2005)and their numerical reproduction employing the v-hysteretic SWRC model and
(b)comparison of ICFEP and Excel reproductions
B.3 Supplementary data for the degree of saturation dependent soil

compressibility with suction
The laboratory data presented by Cunningham (2000) for the artificial soils A,
B1M9 and M8K1B1 were numerically reproduced, adopting the improvements
over the v-hysteretic-SWRC model, suggested in section 5.6, in combination
with the degree of saturation dependent soil compressibility with suction κ∗s
(model parameters in Table B-4). The numerical results are compared with the
experimental data in Figure B-7 for soil A, Figure B-8 for soil B1M9 and Figure
478
appendix B
B-9 for soil M8K1B1 and support the argument that improving the expression
adopted for the primary paths will also improve the prediction of the specific
volume changes induced by suction changes.
Table B-4: Model parameters employed for the prediction of the variation of the specific volume with
suction, measured by Cunningham (2000) on reconstituted soils A, B1M9 and M8K1B1
𝑆𝑟 dependent elastic compressibility with suction, 𝜅𝑠∗
Soil A Soil B1M9 Soil M8K1B1

(Cunningham, 2000) (Cunningham, 2000) (Cunningham, 2000)
param. value param. value param. value
𝜅𝑠 0.022 𝜅𝑠 0.05 𝜅𝑠 0.06
𝜒 1.200 𝜒 1.0 𝜒 0.71
𝜔 2.000 𝜔 0.5 𝜔 1.45

1.0
0.8
Soil A
0.6
0.4 Primary drying (model)
(a)
0.0
10 100 1000 10000 100000
1.56
1.52
1.48
1.44
(b)
1.40
10 100 1000 10000 100000
Figure B-7: (a) SWRC and (b) changes in 𝑣 with suction reported by (Cunningham, 2000) for Soil A
479
appendix B

Soil B1M9
0.8
0.4
0.2
(a)
0.0
10 100 1000 10000 100000
2.10
Specifiv volume,  ( )
2.00
1.90
(b)
1.80
10 100 1000 10000 100000
Figure B-8: (a) SWRC and (b) changes in 𝑣 with suction reported by (Cunningham, 2000) for Soil B1M9
1.0
Soil M8K1B1
0.8
0.4
0.2
(a)
0.0
10 100 1000 10000 100000
2.1
2.0
1.9
1.8
1.7
(b)
1.6
10 100 1000 10000 100000
Figure B-9: (a) SWRC and (b) changes in 𝑣 with suction reported by (Cunningham, 2000) for Soil
M8K1B1
480

Tsiampousi A 2011 PHD Thesis

Uploaded by

Copyright:

Available Formats

Tsiampousi A 2011 PHD Thesis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tsiampousi A 2011 PHD Thesis

Uploaded by

Copyright:

Available Formats

NUMERICAL ANALYSIS OF SLOPES

A thesis submitted to Imperial College London in partial fulfilment of the requirements

Department of Civil and Environmental Engineering

Imperial College London

London SW7 2AZ

‘As you set out for Ithaca

London, April 2011

Conventional constitutive models developed for saturated soils are inadequate

The new developments are subsequently applied to the numerical analysis of

First, I would like to acknowledge the contribution of my supervisors Dr. Lidija

Special thanks go to my uncle Tasos who encouraged me to embark on this

playing, which annoyed me so much as a child, was the foundation on which I

list of figures ................................................................................................... 14

list of tables .................................................................................................... 29

list of symbols ................................................................................................ 31

chapter 1: Introduction ............................................................................. 40

1.2 Scope of the research ................................................................................... 41

1.3 Thesis layout ................................................................................................ 42

chapter 2: Recent Advances in the Modelling of Unsaturated Soils ..... 45

2.1 Introduction ................................................................................................... 45

2.2 Stress variables ............................................................................................ 46

2.3 Isotropic compression and yielding ............................................................... 52

2.4 Critical state .................................................................................................. 64

2.5 Soil-water retention curve (SWRC) ............................................................... 73

2.6 Modelling the soil-water retention curve (SWRC) .......................................... 76

2.7 Coupling of the mechanical and the hydraulic behaviour .............................. 88

2.8 Constitutive modelling ................................................................................... 91

2.8.1 The Gallipoli et al. (2003a) model .......................................................... 92

2.9 Summary and conclusions .......................................................................... 112

chapter 3: Existing ICFEP Capabilities for Modelling Unsaturated Soil

3.1 Introduction ................................................................................................. 117

3.4 Soil-water retention curve (SWRC) model................................................... 147

3.5 Permeability models ................................................................................... 149

3.6 Hydraulic boundary conditions .................................................................... 152

3.7 Summary and conclusions .......................................................................... 165

chapter 4: Modelling of Overconsolidated Unsaturated Soils ............. 169

4.1 Introduction ................................................................................................. 169

4.3 Implementation of the Hvorslev surface ...................................................... 182

4.4 Validation of the Hvorslev surface ............................................................... 194

4.5 Calibration of the Hvorslev surface ............................................................. 212

4.6 Summary and conclusions .......................................................................... 250

chapter 5: Modelling of the Soil Water Retention Curve ...................... 254

5.1 Introduction ................................................................................................. 254

5.2.3 Calibration ........................................................................................... 267

5.3 Hysteretic Soil-Water Retention Curve (hysteretic-SWRC) ......................... 273

5.4 Specific volume dependent, hysteretic Soil-Water Retention Curve (v-

5.7 Summary and conclusions .......................................................................... 316

chapter 6: Stability of Highly Overconsolidated Unsaturated Soil Slopes

6.1 Introduction ................................................................................................. 320

6.2 Problem description .................................................................................... 323

6.3 Factor of safety and determination of slope failure ...................................... 327

6.5 Effect of type of analysis performed ............................................................ 337

6.6 Effect of initial conditions and of excavation depth ...................................... 348

6.6.3 Displacements ..................................................................................... 353

6.7 Summary and conclusions .......................................................................... 363

chapter 7: Behaviour of Unsaturated Soil Slopes under Seasonal

7.1 Introduction ................................................................................................. 366

7.2 Overview of the numerical analyses performed........................................... 367

7.3 Problem description .................................................................................... 370