Nuth - 2009 - Constitutive Modelling of Unsaturated Soils With Hydro-Geomechanical Couplings

Constitutive Modelling of Unsaturated Soils
with Hydro-Geomechanical Couplings
THÈSE NO 4358 (2009)

PRÉSENTÉE le 2 avril 2009
À LA FACULTé ENVIRONNEMENT NATUREL, ARCHITECTURAL ET CONSTRUIT
LABORATOIRE DE MÉCANIQUE DES SOLS
PROGRAMME DOCTORAL EN MÉCANIQUE
ÉCOLE POLYTECHNIQUE FÉDÉRALE DE LAUSANNE
POUR L'OBTENTION DU GRADE DE DOCTEUR ÈS SCIENCES
PAR
Mathieu Nuth
acceptée sur proposition du jury:
Prof. I. Botsis, président du jury

Prof. L. Laloui, directeur de thèse
Dr S. Lacasse, rapporteur
Prof. A. Schleiss, rapporteur
Prof. K. Soga, rapporteur
Suisse
2009
Remerciements
Mes premiers remerciements sont adressés à mon directeur de thèse, Monsieur le
Professeur Lyesse Laloui. Son soutien scientifique et personnel ont été essentiels à
l’aboutissement de la thèse et à ma motivation. Je tiens à souligner les excellentes conditions
de travail qu’il a su créer entre travail en équipe et stimulation continue. Je lui suis tout
particulièrement reconnaissant de m’avoir accordé une autonomie toujours mesurée et sa
confiance en toutes circonstances.
Je remercie également Monsieur le Professeur Laurent Vulliet de m’avoir accueilli au sein
du Laboratoire de mécanique des sols, qui m’a offert des moyens et une structure
exemplaires durant ma thèse.
Toute ma reconnaissance va à Madame le Docteur Suzanne Lacasse, Messieurs les
Professeurs Kenichi Soga et Anton Schleiss, pour avoir accepté de me faire l’honneur d’être
les rapporteurs de ma thèse. Merci également à Monsieur le Professeur Ioannis Bostsis
d’avoir accepté de présider le jury de thèse.
J’adresse une pensée reconnaissante à Monsieur le Professeur Jean-François Molinari pour
son aide dans le cadre de l’école doctorale.
Je tiens à remercier tous les intervenants externes à l’EPFL qui se sont faits partenaires de
cette thèse en favorisant les échanges scientifiques. Que Monsieur le Docteur Frédéric Collin,
Monsieur le Professeur Robert Charlier et Pierre Gérard de l’université de Liège se voient
remerciés. Merci également à Monsieur le Docteur François Laigle pour les échanges et les
données concernant Mirgenbach. Je remercie également tous les partenaires du réseau
TRAMM.
Plusieurs personnes ont directement contribué au contenu de cette thèse. Je remercie ainsi
les stagiaires et assistants pour leurs contributions ainsi que leur sympathie: Réda Traki,
Marine Girard, Adel Yousfi, Christophe Berger, Christoph Knellwolf. Un remerciement tout
particulier est adressé à Patrick Dubey pour son aide au laboratoire et à John Eichenberger
pour les simulations.
Ce dernier me permet de faire la transition vers mes chers amis doctorants, ainsi que tous
mes collègues des Laboratoires de mécanique des sols et des roches. Un grand merci aux
« Lyesse’s angels » du début, Bertrand, Hervé, Azad, bientôt rejoints par Suzanne et Marta.
Je n’oublie pas Rafal et Emilie, ainsi que Nina, Véronique, Simon, Claire, Alessio et Gilbert.
Du côté des roches, je pense en particulier au soutien d’Irene, Stefano et Federica, ainsi que
Rafael, Laurent, Tohid, Julian et Jacopo. Un grand merci à tous les autres membres de
l’équipe pour votre convivialité.
Le soutien et la compréhension de mes amis m’a été indispensable. Merci en particulier à
tous les conjoints des pré-cités, Guillaume et Anne-lise, Florence et à tous mes chers amis des
Polyssons.
Mes parents, mes frères et ma sœur ont été les moteurs de mon travail. J’exprime ici toute
ma reconnaissance à mon noyau familial au sens large qui m’a toujours encouragé à aller de
l’avant dans tous mes projets professionnels et personnels.
Solène, un merci tout particulier à toi pour ta patience, ta compréhension et ton aide
pendant ces années. Ce que tu m’offres quotidiennement est inestimable.
Table of contents
Table of Contents ...............................................................................................................I

Abstract............................................................................................................................ VII
Résumé .............................................................................................................................. IX
List of symbols ................................................................................................................. XI
1. Introduction ................................................................................................................ 1
1.1 General context of the study ........................................................................................ 1
1.2 Objectives and contributions of the thesis ................................................................. 2
1.3 Contents .......................................................................................................................... 3
1.4 References ....................................................................................................................... 4
2. Unsaturated soil mechanics, a review ................................................................... 5
2.1 Introduction .................................................................................................................... 5
2.2 Definitions and physics of unsaturated soils ............................................................. 6
2.2.1 Phases and chemical species ....................................................................................................... 6
2.2.2 Surface tension and suctions...................................................................................................... 7
2.3 State of the art: experimental evidence ..................................................................... 14
2.3.1 Imposing and measuring suction(s): technical solutions ........................................................ 14
2.3.2 Loading paths in unsaturated soils ......................................................................................... 17
2.3.3 Stress-strain response of unsaturated soils ............................................................................. 18
2.3.4 Retention behaviour ................................................................................................................ 30
2.3.5 Air and water flow .................................................................................................................. 35
2.3.6 Conclusions: experimental state of the art ............................................................................... 37
2.4 State of the art: constitutive models .......................................................................... 37
2.4.1 Modelling the skeleton volumetric response ............................................................................ 38
2.4.2 Modelling the soil water retention curve ................................................................................ 46
2.5 Conclusions .................................................................................................................. 50
2.6 References ..................................................................................................................... 51
3. Effective stress-strain framework ........................................................................ 57
3.1 Foreword ....................................................................................................................... 57
3.2 Introduction .................................................................................................................. 57
II CONTENTS
3.3 Conceptual principle of effective stress .................................................................... 58

3.4 Overview of effective stress formulations for saturated soils ............................... 59
3.5 Review of stress frameworks for unsaturated soils ................................................ 61
3.5.1 Bishop single effective stress .................................................................................................... 62
3.5.2 Independent stress variables .................................................................................................... 66
3.5.3 Coming back to Bishop-type effective stresses ......................................................................... 69
3.6 Implications of the use of a generalized effective stress ........................................ 77
3.6.1 Critical state analysis .............................................................................................................. 77
3.6.2 Unsaturated mechanical compression ..................................................................................... 79
3.6.3 Constitutive modelling framework .......................................................................................... 81
3.7 Conclusion .................................................................................................................... 84
3.8 Complementary data ................................................................................................... 85
3.8.1 Uniqueness of critical state line ............................................................................................... 85
3.8.2 Mechanical compression under unsaturated states ................................................................. 86
3.8.3 Generalized effective stress paths ............................................................................................ 89
3.9 References ..................................................................................................................... 90
3.9.1 Publications from the author ................................................................................................... 90
3.9.2 Other references ....................................................................................................................... 90
4. Stress-strain model for unsaturated soils ........................................................... 95

4.1 Foreword ....................................................................................................................... 95
4.2 Introduction .................................................................................................................. 95
4.3 Influence of hydro-mechanical field boundary conditions ................................... 96
4.4 On the use of the generalized effective stress in the constitutive modelling of
unsaturated soils ............................................................................................................................ 99
4.4.1 Introduction ............................................................................................................................. 99
4.4.2 Differences between Bishop’s effective stress and generalized effective stress ......................... 99
4.4.3 Is suction a hardening parameter? ........................................................................................ 101
4.4.4 Discussion on advantages due to the use of generalized effective stress ................................ 103
4.4.5 Conclusion ............................................................................................................................. 103
4.5 ACMEG-s: an elasto-plastic constitutive model .................................................... 104
4.5.1 Overview of the model formulation ....................................................................................... 104
4.5.2 Mechanical constitutive framework for unsaturated soils ..................................................... 104
4.5.3 Provisional model for the Soil Water Retention Curve ......................................................... 113
4.5.4 Conclusion on the formulation: review of parameters ........................................................... 114
4.6 Validation of the mechanical part of ACMEG-s .................................................... 114
4.6.1 Isotropic stress paths ............................................................................................................. 115
4.6.2 Deviatoric stress paths .......................................................................................................... 117
4.6.3 Oedometric conditions ........................................................................................................... 118
CONTENTS III
4.6.4 Swelling pressure .................................................................................................................. 120

4.7 Conclusions ................................................................................................................ 122
4.8 References ................................................................................................................... 122
4.8.1 Publications from the author.................................................................................................. 122
4.8.2 Other references ..................................................................................................................... 122
5. A model for the soil water retention curve ....................................................... 125

5.1 Foreword ..................................................................................................................... 125
5.2 Introduction ................................................................................................................ 125
5.3 Capillary hysteresis ................................................................................................... 126
5.3.1 Conceptual remarks ............................................................................................................... 126
5.3.2 Influence of capillary hysteresis ............................................................................................ 128
5.4 Dependency of retention on mechanical behaviour ............................................. 128
5.4.1 Intrinsic shape of the retention curve .................................................................................... 128
5.4.2 Dependency of air entry value on void ratio ......................................................................... 133
5.5 Soil water retention model ....................................................................................... 137
5.5.1 Mathematical formulation ..................................................................................................... 137
5.5.2 Model response ...................................................................................................................... 140
5.6 Conclusion .................................................................................................................. 142
5.7 Complements: experimental tasks .......................................................................... 142
5.7.1 Description of the tested soil .................................................................................................. 142
5.7.2 Investigated behaviour and objective ..................................................................................... 143
5.7.3 Experimental procedure ........................................................................................................ 145
5.7.4 Results ................................................................................................................................... 148
5.8 References ................................................................................................................... 149
5.8.1 Publications from the author ................................................................................................. 149
6. Double-way coupling in ACMEG-s .................................................................. 151

6.1 Introduction ................................................................................................................ 151
6.2 On the integration of the double way coupled model ......................................... 152
6.2.1 Integration of Hujeux’s model ............................................................................................... 153
6.2.2 Integration of ACMEG-s ...................................................................................................... 157
6.2.3 Remarks on initialization ...................................................................................................... 162
6.3 Implementation into LAGAMINE Finite element code ....................................... 163
6.3.1 Multi-phasic flow model ........................................................................................................ 163
6.3.2 Weak formulation .................................................................................................................. 167
6.3.3 Coupled finite element ........................................................................................................... 168
6.3.4 Global algorithm of the finite element ................................................................................... 171
IV CONTENTS
6.4 Validation of the fully coupled model .................................................................... 171

6.4.1 Motivations for introducing double-way coupling: a review ................................................ 171
6.4.2 Double way coupling: case studies for validation .................................................................. 173
6.4.3 Conclusions on the validation of double-way coupling ........................................................ 178
6.5 Calibration of ACMEG-s and influence of parameters ........................................ 178
6.5.1 Remarks on the calibration procedure: a case study .............................................................. 179
6.5.2 Influence of coupling parameters ........................................................................................... 187
6.5.3 Stress paths ............................................................................................................................ 194
6.5.4 Undrained response ............................................................................................................... 196
6.6 Conclusions ................................................................................................................ 197
6.7 References ................................................................................................................... 197
6.7.1 Publication by the author ...................................................................................................... 197
7. Application: numerical modelling of earthdams .................................................. 199

7.1 Introduction ................................................................................................................ 199
7.2 Earth dam practice and verification: a brief review .............................................. 202
7.2.1 Mechanisms of failure ........................................................................................................... 202
7.2.2 Overview of practice in earth dam modelling ........................................................................ 203
7.3 Reproducing the failure of Mirgenbach dam ........................................................ 206
7.3.1 Presentation of the case study ............................................................................................... 206
7.3.2 Qualitative interpretation(s) of the phenomena .................................................................... 208
7.3.3 Previous numerical modelling of Mirgenbach case ............................................................... 209
7.3.4 Parameter determination and reference simulations ............................................................. 212
7.3.5 Fully coupled analysis ........................................................................................................... 223
7.3.6 Conclusions for the case study .............................................................................................. 227
7.4 Conclusion .................................................................................................................. 228
7.5 References ................................................................................................................... 228
8. Application: numerical modelling of a shallow landslide ........................... 229
8.1 Introduction ................................................................................................................ 229
8.2 Presentation of Rüdlingen experiment ................................................................... 230
8.2.1 Site view ................................................................................................................................ 230
8.2.2 Geology and soil .................................................................................................................... 230
8.2.3 Principle of experiment ......................................................................................................... 231
8.2.4 Instrumentation and hydraulic inputs .................................................................................. 233
8.3 List of parameters ...................................................................................................... 233
8.4 Mesh, boundary conditions and loading scenarios .............................................. 236
8.4.1 Mesh and mechanical boundary conditions .......................................................................... 236
CONTENTS V
8.4.2 Hydraulic boundary conditions ............................................................................................. 237

8.5 Discussion on results of the numerical simulation ............................................... 239
8.5.1 Loading scenario 1 ................................................................................................................. 240
8.5.4 Comments on the possible mechanisms of failure .................................................................. 246
8.5.5 Conclusions ............................................................................................................................ 246
8.6 Conclusion .................................................................................................................. 247
8.7 References ................................................................................................................... 247
9. Conclusion .............................................................................................................. 249
9.1 Achievements .............................................................................................................. 249
9.1.1 Clarification of effective stress for unsaturated soils and establishment of ............................ 249
a new theoretical framework ................................................................................................................ 249
9.1.2 Numerical modelling and Boundary Value Problems ........................................................... 251
9.2 Perspectives ................................................................................................................. 251
Appendix A : Conventions, notations ....................................................................... 253
Appendix B : Parameter determination for Sion silt .............................................. 257
Appendix C : Parameter determination for Kaolin ................................................ 267
Appendix D : Parameter determination for Bentonite .......................................... 275
Appendix E : ACMEG-s: influence of parameters and stress paths ................... 281
Appendix F : Mirgenbach earth dam: review of published characteristics ....... 291
VI CONTENTS
Abstract
Even though the saturation in water of most natural and engineered soils is partial,
constitutive models in geotechnical engineering have long made the assumption of complete
saturation. The increasing knowledge of the rheological behaviour of unsaturated soils is
now favourable to new constitutive formulations to understand the stress-strain behaviour
under the effect of suction. Yet, it is not agreed that there is a unique way to account for
capillary effects, and several families of constitutive models appear to be either insufficient to
describe the complete soil behaviour or else far too complex to be handled in practical
engineering. The principal objective of the PhD work is to overcome those limitations by
formulating a new unified and advanced constitutive model to understand and predict the
behaviour of unsaturated soils. It is specified that the model shall be applicable to the
broadest panel of civil engineering cases involving changes in water content of the soil, for
instance landslides and earthdams.
The methods and contributions of the presented work can be summarized in four main
themes that are (i) the effective stress (ii) the constitutive model for stress-strain behaviour,
(iii) the model for water retention behaviour (iv) the integration of the coupled model in a
finite element code.
A review and clarification of the possible effective stresses for unsaturated soils is
proposed. The generalized effective stress is identified and justified as the most suitable and
convenient unique mechanical effective stress for stress-strain behaviour. It combines the
mechanical stress and the capillary variables that are degree of saturation and matric suction.
The complete stress-strain framework is built on the basis of the generalized effective stress
by relying on work input equation.
A new constitutive model named ACMEG-s is formulated from a reference critical state
elasto-plastic model which features two mechanisms of plasticity. The proposed model uses
bounding plasticity. The effects of capillarity are introduced by the means of a modified
yield limit, and the internal couplings due to the effective stress itself. The mechanical
behaviour is thus dependent on the matric suction and degree of saturation. The model
shows particularly good performances in modelling wetting pore collapse and swelling
pressure tests.
On the basis of experimental results obtained with an unsaturated oedometer, new
contributions to the understanding of the soil water retention curve were produced. A new
model linking the degree of saturation and the matric suction has been formulated and
validated on a number of experimental datasets. An innovative interpretation of the coupling
between the void ratio and air entry value is proposed, in parallel with discussions on the
capillary hysteresis. The retention model, based on kinematic hardening, is thus added to the
stress-strain model which requires the management of double-way coupling.
The coupled model ACMEG-s has been implemented into the finite element code
LAGAMINE, in the objective of application to boundary value problems. The
VIII ABSTRACT
implementation required the development of specific integration routines for the constitutive
models with a constant updating of the terms of coupling. The numerical code enables a
three-level coupling between mechanical, retention, and fluid flow behaviours. Two case
studies are presented with innovative insight into the spatial distribution of pore water
pressures: the earth dam of Mirgenbach and the shallow landslide experiment of Rüdlingen.
Recommendations are also formulated with respect to parameter determination.
Keywords: Unsaturated soils, constitutive modelling, effective stress, capillarity,

retention, collapse, finite element method, numerical simulation
Résumé
En dépit de l’état de saturation partielle en eau de la plupart des sols naturels et remaniés,
les modèles constitutifs géotechniques ont longtemps reposé sur l’hypothèse de saturation
complète en eau. L’état des connaissances du comportement rhéologique des sols en
conditions non saturées permet maintenant la formulation de nouveaux modèles constitutifs
pour interpréter le comportement contraintes-déformations sous l’effet de la succion.
Cependant, il n’y a pas consensus quant à la manière de prendre en compte les effets
capillaires, et plusieurs familles de modèles constitutifs s’avèrent soit insuffisants à décrire
complètement le comportement du sol ou bien au contraire trop complexes pour prétendre à
une utilisation au niveau de l’ingénierie pratique. L’objectif principal de la thèse présentée
est de dépasser ces limitations en formulant un nouveau modèle constitutif unifié et avancé
pour comprendre et prédire le comportement des sols non saturés. Le cahier des charges du
modèle souligne son applicabilité à l’éventail le plus large d’applications du génie civil où les
changements de teneur en eau sont prépondérants, tels que les glissements de terrain ou bien
les barrages en terre.
Les méthodes et contributions du travail de recherche peuvent être résumés en quatre
thèmes principaux qui sont (i) la contrainte effective (ii) le modèle constitutif pour le
comportement en contrainte et déformation (iii) la modèle de comportement en rétention et
(iv) l’intégration du modèle couplé dans un code aux éléments finis.
On propose un état de l’art et une clarification des contraintes effectives possibles pour les
sols non saturés. La contrainte effective généralisée a été identifiée comme la plus appropriée
en tant que contrainte mécanique unique pour le comportement contrainte-déformation, et
est justifiée comme telle. Cette contrainte combine la contrainte mécanique et les variables
capillaires qui sont ici le degré de saturation et la succion matricielle. Le cadre complet de
contrainte-déformation est construit sur la base de considérations énergétiques.
Un nouveau modèle constitutif appelé ACMEG-s est formulé sur la base d’un modèle
élasto-plastique de référence, défini dans le cadre de l’état critique et possédant deux
mécanismes de plasticité, permettant de modéliser une plastification progressive. Les effets
capillaires sont introduits par le biais d’une surface de charge modifiée, ainsi que par les
couplages intrinsèques dus à la contrainte effective. Le comportement mécanique dépend
ainsi de la succion et du degré de saturation. Le modèle se montre particulièrement
performant dans la modélisation de l’effondrement hydrique et des tests de pression de
gonflement.
La thèse contribue à la caractérisation de la courbe de rétention en se basant sur un
programme expérimental utilisant un oedomètre non saturé. Un nouveau modèle reliant le
degré de saturation à la succion est formulé et validé sur plusieurs jeux de données
expérimentales. On propose une interprétation inédite du couplage entre l’indice des vides et
la succion d’entrée d’air accompagnée de discussions sur l’hystérèse capillaire. Le modèle de
rétention, utilisant l’écrouissage cinématique, est ainsi ajouté au modèle en contrainte-
déformation, ce qui nécessite la gestion des couplages à double sens.
X ABSTRACT
Le modèle couplé ACMEG-s a été implémenté dans le code aux éléments finis
LAGAMINE, dans le but de l’appliquer à des problèmes aux limites. Des routines
d’intégrations spécifiques pour l’intégration des modèles constitutifs ont été formulées dans
cette optique. Le code permet un triple couplage entre les comportements mécanique, de
rétention et d’écoulement de fluides. Deux études de cas sont présentées, avec des résultats
innovants quant à la distribution spatiale des pressions interstitielles. Il s’agit du barrage en
terre de Mirgenbach et de l’expérience de déclenchement de glissement de terrain de
Rüdlingen. On formule aussi des recommandations pour la détermination des paramètres
du modèle.
Mots-clefs: sols non saturés, modélisation constitutive, contrainte effective, capillarité,

rétention, effondrement, méthode éléments finis, simulation numérique.
List of symbols
Roman alphabet
a material parameter controlling the rate of plasticity inside the
external deviatoric yield limit
b materiel parameter defining the shape of the deviatoric yield
limit
c materiel parameter controlling the rate of plasticity inside the
external isotropic yield limit
c′ cohesion
C mechanical constitutive tensor
D elastic stiffness matrix
CKA1 coefficient for the relation between water permeability and
degree of saturation
CKA2 coefficient for the relation between water permeability and
CKW1 coefficient for the relation between water permeability and
CKW2 coefficient for the relation between water permeability and
d ratio between the preconsolidation pressure, pc′ , and the
′
critical pressure, pCR
e void ratio
e0 initial void ratio
E Young’s modulus
f mechanical yield limit
fiso isotropic yield limit
f dev deviatoric yield limit
g gravity acceleration
g plastic potential
g dev plastic potential of the deviatoric mechanism

XII LIST OF SYMBOLS
giso plastic potential of the isotropic mechanism
G shear elastic modulus

Gref ′
shear elastic modulus at a reference pressure pref
H matrix of hardening moduli

H hardening modulus
I identy matrix
Ip plasticity index
kw intrinsic water permeability
kr relative water permeability
kwsat satured water permeability
K bulk elastic modulus

K ref bulk elastic modulus at a reference pressure (…)
K0 coefficient of earth pressure at rest
M slope of the critical state line

n unit normal vector
n porosity
ne non-linear elasticity exponent

OCR overconsolidation ratio
p′ mean effective stress
pnet mean net stress
p0′ initial mean effective stress
′
pref reference mean effective stress
pc′ preconsolidation pressure
pc′ 0 initial preconsolidation pressure
′
pcR mean effective stress at the critical state
′ 0
pCR initial value of the critical pressure
pa pore air pressure
pg gas pressure
pv pore water vapour pressure
pw pore water pressure

LIST OF SYMBOLS XIII
pl liquid pressure
q deviatoric stress
r radius of the capillary tube
rdev degree of mobilization of the deviatoric mechanism
riso degree of mobilization of the isotropic mechanism
s matric suction
sD drying yield suction
sI wetting yield suction
se reference air-entry suction
se H updated air-entry suction
Sr degree of saturation
Sr ,w degree of saturation in water
S res residual degree of saturation
t time
T temperature
u displacement vector
V volume of the soil
Va volume of air
Vα volume of the constituent α
Vv volume of voids
Vw volume of water
w water content
wL liquid limit
wP plastic limit
W work input per unit volume

x vector of nodal values
Greek alphabet
α flow rule parameter
β plastic compressibility modulus
β0 plastic compressibility modulus at zero suction

XIV LIST OF SYMBOLS
γd dry unit weight
γs parameter for LC curve
δ Kronecker’s symbol
κ swelling index (Cam-clay)
λ compression index (Cam-clay)
λdev
p
plastic multiplier of the deviatoric mechanism
λisop plastic multiplier of the isotropic mechanism
ε total strain tensor
εe elastic strain tensor
εp plastic strain tensor

εv volumetric strain
εd deviatoric strain
ε ve volumetric elastic strain
ε de deviatoric elastic strain
ε vp volumetric plastic strain
ε dp deviatoric plastic strain
Ω parameter for evolution of compressibility

πH term of coupling in retention model
φ′ friction angle at critical state

ρ bulk density
ρa bulk dry density
ρd bulk density of gas
ρg bulk density of solid
ρw bulk density of water
σ′ effective stress
σ total stress
σ′ total stress tensor
σ effective stress tensor
σv vertical stress in oedoemeter
ψ dilatancy angle
χ effective stress parameter
1. Introduction
1.1 General context of the study

Soil mechanics, in the broadest sense of the term, relate the behaviour of the soils to their
properties, actual state and history. Knowing that more than one third of the earth surface is
considered as arid and semi arid (Fredlund and Rahardjo 1993), not to mention the less
extreme cases of drought in soils, it is obvious that the general state of saturation in water of
most soils is partial. In other words, the pore space within the soils is generally filled with
water and air which defines a problem with at least 3 phases.
Practically, it can be observed that any soil surface that is in contact with air will very
likely be assorted with a zone of negative pore water pressure (suction) in the soil mass in
the neighbourhood of that dry surface. This observation does not only have an implication
on natural untouched grounds, but also on most operations of civil engineering practice that
will cause the soil to be in contact with air, by remoulding, excavating or else compacting
processes.
Many examples enter the category of problems of partial saturation, starting with the
most obvious case of earth dams. Such a structure being by definition designed to retain
water within a reservoir, the normal use of the dam will be at the origin of seepage within a
given portion of the dam that is considered as fully saturated in water. In the rest of the
geometry, the state of saturation is partial which raises the issue of a possible different
behaviour and resistance. The problem of partial saturation already arises at the stage of the
construction, where the successive layers are compacted (rolled) and need a control of pore
water pressure.
The stability of natural slopes is also depending on the state of saturation, as it is observed
that a dryer medium will possess a better apparent resistance. The stake here is to predict the
risk of instability due to environmental changes, for instance sudden heavy rains. Again, a
key issue to be addressed is the level of pore water pressures.
Concerning the works in the underground space, the zones of soil that are the most
susceptible to partial saturation are the front of excavation, for instance in the case of tunnels.
The local decompression causes suctions to appear and loss of water in the surfaces in
contact with air. Any vertical excavation used for foundation is also prone to failure if the
water supply is not controlled.
The historical developments of soil mechanics are widely attributed to Karl Terzaghi
(1936), with the definition of the effective stress principle that is still largely used nowadays.
A number of interpretations on the issue of soil deformability and resistance are indeed
limited to this context of soils that are fully saturated with water. However, the concepts of
saturated soils mechanics, while consistent, represent only a part of soil mechanics and
cannot be considered as suitable for the majority of the engineering problems reviewed
above. The engineering practice is however in the majority of cases to neglect the partial
2 CHAPTER 1
saturation, which might either permit to remain on the side of safety (e.g. slope stability) or
on the contrary miss out dramatic volume changes (e.g. expansive soils).
One possible historic explanation for the delay in accounting for partial saturation in soil
mechanics might be the difficulty to relate the soil response to the state of stress in partial
conditions. The knowledge of the behaviour of unsaturated soils has, however, progressed
significantly in the last 50 years, and the themes of testing and modelling partially saturated
soils are at the centre of a large part of journal publications and international conferences. A
number of techniques for testing partially saturated soils in-situ and in the laboratory is now
available at the research and engineering levels, resulting in a good characterization of the
main features of behaviour of soil submitted to suction. In many research laboratories, it
appears that the mechanics of unsaturated soils even tend to replace the mechanics of
saturated soils as the basis for other research themes (e.g. effect of temperature, chemo-
mechanics).
However, some inconsistency can still be found between the experimental results from
one team to the other, and there is not yet a perfect consensus on several particular features
of behaviour (for instance the capacity of retention). On the other hand, no single effective
stress could ever be found to be the equivalent of Terzaghi’s formulation generalized to
partially saturated soils, so that several different schools of thought emerged from the
debates on constitutive modelling.
While some numerical tools on the market already account for partial saturation, the
corresponding formulations remain most of the time heavily simplified and only provide a
partial picture of the hydro-mechanical interactions. Advanced constitutive model and
performing numerical tools are nevertheless useful firstly to model and understand the
physical process observed experimentally, and secondly to help dimensioning new
experiments, particularly for boundary value problems.
1.2 Objectives and contributions of the thesis

The main objective of the PhD thesis is to draw a link between the accepted features of
behaviour of partially saturated soils and a numerical tool designed for general civil
engineering purposes. A key intermediary goal to be fulfilled is the formulation of a new
constitutive model for unsaturated soils.
The PhD. work presented here follows a logical axis of research of the soil mechanics
laboratory of the Ecole Polytechnique Fédérale de Lausanne on unsaturated soil mechanics.
The previous PhD. theses, to be mentioned later, that were published within this context
focused mostly on the experimental characterization of the stress-strain and water retention
behaviour of silts. The present work shall then (i) complement the available data base and (ii)
use the data for validation of the theoretical developments. It will be explained later that the
developed constitutive model also serves as a reference framework for other themes of
research under progress or to be carried out in the laboratory.
Given the fact that there is not a general agreement among geotechnical research teams on
the best practice for the modelling, the discussion on the effective stress(es) for unsaturated
soils shall be at the core of the manuscript. The objective is to not only to clarify the
theoretical developments published in literature but also to propose to the geotechnical
research community a clear conclusion to the debate on stress variables.
A new constitutive model is developed to answer the need for a comprehensive stress-
strain and water retention framework for soils. The constitutive model accounts for complex
INTRODUCTION 3
interactions between the internal modules and shall develop a unique level of performance
with respect to the recent modelling contexts.
The model is designed for a straightforward implementation into a finite element code,
particularly due to the optimized stress variables. The last objective of the thesis is to
complete a working numerical tool for boundary value problems with partial or complete
saturation in water. The model is to be applied to two engineering case studies.
A theme that is omnipresent throughout the manuscript is the balance between the
unique performances of the model and the convenience of use and applicability of the
modelling framework.
1.3 Contents
The thesis is organised in 9 chapters that cover the literature review and developments
carried out during the PhD. research. Following the usual practice, the manuscript starts
with a chapter dedicated to the start-of-the art in testing and modelling partially saturated
soils. Yet, some of the following chapters (e.g chapters 3 and 5) will feature a separate and
denser literature review to provide a more detailed insight into the particular behaviours of
interest.
Chapter 2 draws the state of knowledge on the mechanics of unsaturated soils. A review
of the physics and mechanisms involved in partially saturated soils is firstly proposed, in
order to set the limits of the present research and to contribute already to the clarification of
the terminology used throughout the manuscript. The major behavioural trends in terms of
stress-strain and suction are then presented on the basis of experimental evidence, with a
particular focus on the soil water retention behaviour. While the older, elementary
constitutive models are only briefly mentioned, the chapter essentially reviews the existing
modelling frameworks published in recent years.
The following chapter 3 has the objective to discuss and clarify the issue of stress and
strain framework in partially saturated soils. It is demonstrated that the definition of an
adequate and consistent stress and strain framework involving the whole set of state
variables is a mandatory step for the formulation of the constitutive model. The chapter
contributes also to the clarification of the terminology of effective stresses for saturated and
partially saturated soils, and proposes a new classification of the existing frameworks. The
chosen generalized effective stress is justified and assessed from the viewpoint of
experimental control, constitutive implications and graphical representation.
Chapter 4 introduces the formulation of a new constitutive stress-strain model for
partially saturated soils, called Advanced Constitutive Model for Unsaturated Soils extended
to suction changes (ACMEG-s). The key ingredients of the model are the generalized
effective stress, the water retention model and elasto-plasticity with two mechanisms of
plasticity. The model is validated on several sets of data taken from literature and features a
provisional simplified description of the evolution of the degree of saturation with respect to
suction.
Chapter 5 addresses the relationship between saturation and negative pore water pressure
(soil water retention behaviour). The chapter starts with an extensive review of the recent
advances in experimental characterization of the water retention behaviour. The review
leads to the mathematical formulation of the terms of coupling between the air entry value
and the volume change. A new model for the retention behaviour is then formulated, and
accounts for the identified coupling as well as the capillary hysteresis. The chapter also
presents the experimental works realised during the PhD. thesis that focused on the
4 CHAPTER 1
characterization of the air entry value of Sion silt (Switzerland) under different initial void
ratios.
The purpose of chapter 6 is to gather the two sub-models presented in chapter 4 (stress-
strain model) and chapter 5 (retention model) into a single double-way coupled constitutive
frameworks. The chapter develops the numerical integration of the model and introduces the
models of flow for water and air. In this part is also presented the numerical implementation
of the developed model into a finite element code called LAGAMINE.
The chapters 7 and 8 present two different types of applications for the constitutive model
ACMEG-s, respectively earth dams and landslides. In chapter 7, the case study of
Mirgenbach earth dam (France) is back analysed with the developed numerical tool. The
results show a consistent prediction of the interstitial pore pressures in partially saturated
conditions throughout the geometry of the dam. In the case of Rüdlingen landslide (chapter
8), a parametric analysis is carried out to investigate the effect of rain infiltration within a
natural slope.
Chapter 9 draws the conclusions of the thesis and presents the perspectives of research
within the theme of advanced modelling of unsaturated soils.
In general, the content of each chapter has been the object of an separate publication. For
this reason, each chapter can be read almost independently and thus features its own list of
references.
1.4 References
Fredlund D. G., Rahardjo H. (1993). Soil mechanics for unsaturated soils, John Wiley & Sons.
Terzaghi K. (1936). "The shearing resistance of saturated soils and the angle between the planes of
shear." International Conference on Soil Mechanics and Foundation Engineering, Harvard
University Press, pp. 54-56.
2. Unsaturated soil mechanics, a review
2.1 Introduction
This chapter establishes the state of knowledge on the behaviour of unsaturated soils in
the broad sense that is from the experimental and conceptual viewpoints. The objective is to
assess the current and historic practice in unsaturated soils testing and modelling in order to
define the objectives of the PhD thesis with more precision. The chapter will thus mainly
contribute to identifying the advances and limitations of research works taken from
literature, which will set the foundations for new experimental tests and/or constitutive
models within the present research. The chapter also provides the theoretical basis for all the
assumptions made in the rest of the manuscript.
As evocated in previous chapter 1, the possible range of soils to study is extremely wide.
It should not be a priori possible to cover the full range of soils (from coarse granular
materials to active clays) with a single conceptual framework or experimental programme. It
is one of the goals of this chapter to set the limits of the analysis with respect to soil type, and
also to define already the possible generalisation of anticipated results to materials that are
out of the studied range.
Similarly, the panel of possible applications of the present research work is large enough.
Consequently the nature and direction of in-situ (environmental) loads might be very
different from one geotechnical case to the other. The study of grounds in partially saturated
conditions might also raise new issues with respect to in-situ stresses and pore water
pressures. It is thus intended to review the possible boundary conditions and loads imposed
to the pre-cited materials. This information is of particular value for the validation of models.
The chapter is built in three main parts. Firstly, the definitions and physics of soils in
partially saturated conditions are reviewed in a preliminary section. These definitions will be
considered as approved for the rest of the manuscript, and all the defined variables and
constants are considered as implicit in the coming chapters (unless otherwise stated).
Secondly, a section is dedicated to the review of the rheological knowledge on unsaturated
soils, concerning the features of behaviour and experimental techniques. Even though it will
be concluded that the present PhD work should include only little experimental research, the
state of knowledge on experimental techniques is of interest for the identification of the basic
mechanicsms, in order to meet the experimental limitations for the parameter determination.
The third part reviews the up-to-date constitutive models for unsaturated soils, from the
viewpoint of stress-strain behaviour, retention properties and fluid flow. The historic
evolution of constitutive models is also overviewed in order to understand better the process
of model formulation and the progressive increase in performances and complexity of the
reviewed models.
The chapter ends with a list of specifications for the experimental and modelling tasks
providing also already some tracks to follow for the new developments, explicitly inspired
from the existing concepts.
6 CHAPTER 2
2.2 Definitions and physics of unsaturated soils

The lexicon of unsaturated soils has been historically defined simultaneously in the
disciplines of soil physics (e.g. Brooks and Corey 1934) and soil mechanics (e.g. Bishop 1960)
. As a consequence, some variations between the terms can be noticed in literature. The
purpose of this section is on the one hand to clarify the terminology to be used in the
manuscript and on the other hand to provide all at once the basic useful definitions for
unsaturated soils. Provided that the final developments of the PhD thesis are mainly aimed
at practitioners of soil mechanics, the latter discipline will be strongly favoured as for the
choice of terminology.
2.2.1 Phases and chemical species

The so-called unsaturated soils are widely identified in literature as soils whose pores are
filled with water and air. It would be a major challenge to determine which author is at the
very origin of this basic definition, however, the terminology is reported to be already used
in the first ISSMFE conference (International Society for Soil Mechanics and Foundation
Engineering) in 1936. Fredlund and Rahardjo (1993) define unsaturated soils as soils which
have negative pore-water pressures; that definition is also largely acknowledged in literature.
While the unsaturated soil mechanics were to be compared in the 1930’s only to the
conventional soil mechanics (i.e. saturated soil mechanics), more complex cases of saturation
of pores with a higher number of fluids (e.g. oil reservoirs (Collin 2003)) have emerged since.
The definition of unsaturated soils has thus to be reviewed from the viewpoint of phases and
chemical species.
According to the aforementioned description of the unsaturated soils, the problem
features 3 distinct phases that are solid, liquid and gaseous. The solid phase is constituted of
the soil grains. The liquid phase consists of a mixture of liquid water and dissolved air,
including trapped air. The third phase is a mixture of several gases and water vapour,
knowing that the main component is dry air.
Water and air are thus identified as the only two chemical species for the fluid part of the
problem. The quantity of dissolved air into the liquid water and the quantity of water
vapour mixed with dry air obey specific laws that will be presented later in chapter 6. Those
respective proportions of dissolved air and water vapour depend on the state of saturation
and on the levels of pore air and pore water pressure (Fig. 2.1). In Fig. 2.1, the degree of
saturation S r is defined as the ratio between the volume of water Vw and the volume of
voids Vw :
Vw
Sr = (2.1)
Vv
The dissolved air and water vapour can also have an influence on the compressibility of
the liquid and gaseous mixtures. For instance, the compressibility of an air-water mixture is
always inferior to the theoretical value for pure water that is 4.58 ×`10−7 kPa −1 .
It can be deduced from these introductory paragraphs that the initial definition of
unsaturated soils as porous media saturated with water and air still holds. What is less
appropriate however is to associate the species water only to the liquid phase and the species
air to the gas phase. In an effort to adopt a terminology that is the most consistent with the
physical evidence, it is proposed to decompose the unsaturated soil from the viewpoint of
phases. The described medium thus includes a solid phase s , a gas phase g and a liquid
STATE OF THE ART : UNSATURATED SOILS 7
Figure 2.1 Schematic representation of the saturation phases in a soil profile with a ground water table
beyond the surface, after Wroth and Houlsby (1985).
phase w , each phase α possessing its own volume Vα and mass M α (Fig. 2.2). Each of the
fluid phases is considered as one homogeneous fluid with a corresponding pore pressure
pα , knowing that the two equivalent fluids are considered as immiscible. This terminology
will be used in priority throughout the rest of the manuscript. However, as practically the
pore pressure of the gas is considered as equal to that of the air pa , and the pore pressure of
the liquid is approximated to that of water pw , these two last variables are also used in the
text.
As a conclusion, it appears that the conventional expression “unsaturated soils” is
imprecise. Among the possible correct denominations, the terms “soils partially saturated in
water” or “soils saturated with water and air” would be more representative. However, since
the long-lived “unsaturated soils” implicit terminology is widely accepted and still in use in
up-to-date research, it will be also adopted in this manuscript.
2.2.2 Surface tension and suctions

In the continuity of previous paragraph, it is assumed that the pores of the soil are
completely filled by two fluids only, and that those fluids are immiscible. The most
appropriate way to represent the proportions of solid and fluids is to use the phases rather
than the chemical species. To remain in the most general case where the immiscible fluid
phases might be either liquid and gas or two liquids (the unlikely option of having two
immiscible gases is implicitly discarded here), the fluids are called wetting fluid (w) and non-
wetting fluid (nw). The wetting or non-wetting nature of a fluid is only relative; a fluid taken
alone cannot not indeed be qualified of “wetting”. Looking at the interface between the solid
and the two fluid phases, Fig. 2.3, the fluids can be classified with respect to the relative
angle made with the surface of the solid, called contact angle θ . The largest angle will
8 CHAPTER 2
Figure 2.2 Phase diagram for a representative elementary volume
correspond to the non-wetting fluid (fluid 1 in case of Fig. 2.3) while the smallest angle
defines the wetting fluid (fluid 2 in Fig. 2.3).
Using the subscripts w and nw for the fluids, one can redraw the phase diagrams of Fig.
2.2 for unsaturated soils in Fig. 2.4. In the drawing is introduced a new element at the
interface between the two fluids. This element is called “contractile skin” and is defined by
Davies and Rideal (1963). The contractile skin is a membrane-like interface that is purely
elastic and can sustain tension. The contractile skin being not a part of fluid w nor belonging
to fluid nw , it has different properties from those of the two fluids (e.g. density, heat
conductance). The skin, called “interphase” by Fredlund and Rahardjo (1993) is thus a
pseudo-mixture of the two fluids in which a particular state of stress can be developed. This
means that the pressure of interstitial fluid w and nw can be different, and that a new phase
should theoretically be taken into consideration. The action of the “interphase” will indeed
be at the center of numerous discussions throughout the manuscript, and later on called
“capillary effects”. The impact the contractile skin cannot be neglected for the understanding
the behaviour of unsaturated soils. However, from the point of view of volume and masses,
the fraction represented by the contractile skin is negligible (Fredlund and Rahardjo 1993).
The thickness of the contractile skin is in the order of a few molecular layers. Consequently,
the rigorous phase diagram of Fig. 2.4a can be simplified into that of Fig. 2.4b
Supposing that both the phases of the wetting and non wetting fluids are continuous and
that their respective pressures are different, it is observed that the contractile skin might
interact with soil particles and have an effect on the mechanical behaviour of the soil. While
in the upcoming part 2.3 these effects are quantified experimentally from a macroscopic
point of view, the local notion of surface tension is reviewed hereafter.
The surface tension is a property of the contractile skin, resulting from the balance of
intermolecular forces at the “interphase”. In the case where the wetting fluid is liquid water
and the non wetting fluid is dry air, there is a difference between the forces experienced by
the water molecules belonging to the bulk water in a pore and the forces exerted on the
molecules from the contractile skin, as illustrated in Fig. 2.5. In Fig. 2.5a, a molecule from the
free water (molecule A) is submitted to forces that are equal in all directions, while molecule
B located at the air-water interface, experiences “an unbalanced force towards the interior of
the water” (Fredlund and Rahardjo 1993). The only way to ensure the equilibrium of the
contractile skin is to generate a tensile pull along the skin. So is defined the surface tension
Ts expressed as a tensile force per unit length of contractile skin (Fig. 2.5b).
Figure 2.3 schematic representation of solid-fluid-fluid interface. The contact angles are measured at
the tangent of the interface between the two fluids, taken at the contact point with solid.
Figure 2.4 Rigorous and simplified phase diagrams for a soil saturated with two immiscible fluids
(after Fredlund and Rahardjo 1993).
The force equilibrium in the vertical direction requires:
2Ts sin β = 2Δpw Rs sin β (2.2)
Ts
Δpw = (2.3)
Rs
In the case of Fig. 2.5, the variation Δpw corresponds to the difference between the pore
air pressure pa and the pore water pressure pw . The radius of curvature of the interface is
thus inversely proportional to the difference between the pressure of the non wetting fluid
and the pressure of the wetting fluid. The curved interface is called water meniscus. One
complementary observation here is that the surface tension is only linked to a difference in
fluid pressure; the definition is thus relative and does not require an absolute reference in
fluid pressure.
A measure of this variation of pressures is necessary at this point. Indeed, the studied
property of surface tension is only part of the total potential in unsaturated soils. The
definition accepted in geotechnical engineering (Aitchison 1965) states that the total suction
10 CHAPTER 2
Figure 2.5 Surface tension at water-air interface. Ts is the surface tension, Rs is the radius of
curvature, pw is the water pressure, pa is the air pressure (After Fredlund and Rahardjo 1993).
corresponds to the free energy of water, knowing that the free energy can be measured in
terms of the partial vapour pressure of the soil water. According to Aitchison (1965), the total
suction ψ is the sum of the matric suction s and the osmotic suction π (Eq. 2.4). Each of the
components of the free energy are reviewed hereafter.
ψ = s +π (2.4)
Matric suction
The matric suction is defined as the difference between the pressures of non wetting fluid
and wetting fluid. In the present case where the soil is partially saturated in water, the liquid
and gas phase include both water and air species (see part 2.2.1). The imposed (or measured)
pressures are practically those of the phases, so the matric suction is written:
s = pg − pl (2.5)
where pg is the gas pressure and pl the pore water pressure. However, in literature, the
matric suction is commonly written as the difference between air pressure pa and pore
water pressure pw , which is a reasonable simplification (see 2.2.1):
s = pa − pw (2.6)
Referring back to Fig. 2.5 and Eq. 2.2, it appears that the matric suction is directly
associated with the surface tension property of the contractile skin. Indeed, the matric
suction is linked to the capillary phenomena (or capillarity) taking place in partially
saturated porous media. In complement to Fig. 2.5, the capillarity can be physically modelled
by a small tube inserted into water under atmospheric conditions (Fig. 2.6).
Due to the small radius r of the tube, the condition of equilibrium Eq. 2.2 is now written:
2Ts sin β
hc = (2.7)
rγ w
Figure 2.6 Capillary rise in a small glass tube.
And the relationship between the matric suction and the radius of the tube can be written:
Ts sin β
s = hcγ w = (2.8)
r
Thus, matric suction is inversely proportional to the radius of the capillary tube.
Considering that the glass tube is a model of soil pores, the practical implication of this
observation is that the water rise above a water table due to capillarity will be higher in soils
with a lower porosity. Those same fine-grained materials are also favourable to the
development of larger matric suctions. For these reasons, matric suction is also called
capillary component of the free energy (Aitchison 1965).
Osmotic suction
Coming back to the definition of the total suction in term of the partial vapour pressures
of water, a second component of the free energy can be identified as the osmotic or solute
suction. The osmotic suction practically refers to the presence of dissolved salts in the pore
water. It is the suction derived from the difference between the partial pressure of the water
vapour in equilibrium with a solution (i.e. water and solvent) and the partial pressure of
water vapour in equilibrium with free pure water (Aitchison 1965).
When comparing the vapour pressure over a flat surface of soil water containing salt
pv , soil and the vapour pressure over a flat surface of pure water pv , water , the inequality Eq. 2.9
is verified (Fredlund and Rahardjo 1993):
pv , soil < pv , water (2.9)
Eq. 2.9 implies that the higher is the concentration in salts in the pore water is, the lower
the relative humidity will be, which also corresponds to a higher osmotic suction π .
Considering that in general soils are seen as porous media saturated with soil water that
is not pure, the two components of suction might be of importance. Understanding already
that suction will play a major role in the experimental and modelling tasks to be reviewed
and developed in the following, it is essential to quantify the respective contribution of the
matric and osmotic suctions to the total suction.
12 CHAPTER 2
Starting from an unsaturated material in equilibrium with a given degree of saturation

Sr , one can determine the absolute level of total, matric and osmotic suctions. Upon a
change in the degree of saturation (for instance drying), the quantity of water will be
changed, and variations will occur in all suction measures. Fredlund and Rahardjo (1993)
reviewed such suction levels and variations for two materials under the effect of a reduction
of the degree of saturation, table 2.1. Here the information on the quantity of water in the soil
is provided through the (gravimetric) water content w defined as the ratio between the mass
of water M w over the masse of solid M s :
Mw
w= (2.10)
Ms
It can be noticed from table 2.1 that while in the reference step the contributions of the
absolute matric and osmotic suctions to the absolute total suction are of the same order of
magnitude, the variations in osmotic suction is significantly lower than that of the matric
suction upon drying. The changes in total suction are thus mainly piloted by the changes in
matric suction. In the case of the glacial till, the variations in osmotic suction can even be
considered as negligible. Seker (1983) draws similar conclusions and provide an
interpretation for these different variations. Provided that the drying (or wetting) process
does not affect the thermal and chemical conditions within the soil, the osmotic suction
should logically remain fairly stable. In other words, the osmotic suction is not a function of
saturation. This assertion should however be moderated for the case of fine clays, or active
materials, where the osmotic suction seems to be more affected by the water content (table
2.1). Obviously, in the case where other external factors may affect the chemical composition
of the soil water (for instance in problems of chemical contamination or pollutant transfers),
the osmotic suction may display more important changes in comparison with matric suction.
Such soil contamination problems are out of the scope of this PhD research and will be
disregarded here.
In the particular case of clays the mechanism of osmosis can be regarded from a
microscopic point of view. According to the review from Mitchell and Soga (2005), water is
indeed strongly attracted by the surface of the clay minerals. As a consequence, there is
always a layer of adsorbed water around the clay particles or pellets. The mechanisms for
water adsorption by dry clays are of several natures and will not be developed here:
(i) Hydrogen bonding
(ii) Hydration of exchangeable cations
Table 2.1 Variations of suctions for two soils, from Fredlund and Rahardjo (1993)
Osmotic
Water content Matric suction Total suction
Soil Step suction
w (%) s (kPa) ψ (kPa)
π (kPa)
Initial 30.6 273 187 460
Regina clay Final 28.6 354 202 556
Variation -2 +81 +15 +96
Initial 15.6 310 290 600
Glacial till Final 13.6 556 293 849
Variation -2 +246 +3 +249
Figure 2.7 Mechanism of attraction by osmosis, from Mitchell and Soga (2005)
(iii) Attraction by osmosis (Fig. 2.7)

(iv) Charged surface dipole attraction
(v) Attraction by London dispersion forces
(vi) Capillary condensation
All these forces are contributing to the attraction of water towards the soil particles, but
their intensity is decreasing rapidly with the distance from the mineral surface. The point (iii)
is linked to the osmotic suction and the point (iv) to the matric suction. At the microscopic
scale (Fig. 2.7), the attraction by osmosis is due to the negative charge of the clay surface. The
concentration of cations increases at the vicinity of the mineral surfaces which entails a
diffusion of water molecules in the same zone in order to equalize concentrations. This local
diffusion is at the origin of an overall water pressure deficiency.
The existence of adsorbed water indicates that a clayey soil might contain water even at
very low relative humidity. In other words it is not possible to dry completely a soil sample
unless placing it in a oven with a temperature over 100°C. The practical implication of the
adsorption is the existence of a residual degree of saturation ( S res ) at very high suctions.
The volume adsorbed water is thus accounted for in the total volume of water. However,
the pressure of the adsorbed water should be measured locally at each particle surface. Due
to technical and modelling constraints, only a macroscopic measurement and description of
the pore water pressure is possible. The pressure of adsorbed water is thus considered to be
contributing to the average (negative) liquid water pressure. However, in some theoretical
frameworks reviewed later in this chapter, it happens that the adsorbed water is the object of
a dedicate pressure or volume description.
The theoretical basic concepts of unsaturated soils have been clarified with respect to
phases and chemical species. The definition of suction and the indicators for the quantity of
water in the soils have been formulated. Due to the nature of the total suction (free energy),
the effect of suction is considered from now on as exclusively isotropic. Also, it is considered
14 CHAPTER 2
that the changes in total suction are mainly piloted by the changes in matric suction
exclusively. The purpose of the following section is to review to methods of measurement
and imposition of suction(s) and to summarize the features of behaviour of unsaturated soils
on the basis of experimental evidence.
2.3 State of the art: experimental evidence

2.3.1 Imposing and measuring suction(s): technical solutions
Given the experimental know-how for testing conventional saturated soils from the
geotechnical point of view, the major challenge of adding a third phase in the problem is to
keep track of the volumetric response (in all the three solid, liquid and gas phases) while
managing to control the loading paths in terms of imposed stresses and deformations.
Knowing the definitions of degree of saturation, water content, matric and osmotic
suction, the partial saturation in water of a soil can be theoretically controlled (or monitored)
either by the means of the volume of water, by the pressures of interstitial fluids or by the
chemical composition of these fluids. More accurate results could be expected by using
combinations of several techniques.
In the case studies overviewed in chapter 1, the partial saturation is a consequence of
particular environmental loads. In other words, the pore air and water pressures are such
that there is a capillary rise in the unsaturated zone. The degree of saturation is thus a
consequence of the changes in pore fluid pressures. Similarly, in the experimental techniques
reviewed hereafter, the suction(s) are the imposed variables whereas the degree of saturation
is monitored as a response. It remains however possible to carry out tests where a target
water content is aimed at but this type of tests is less conventional.
As it will be concluded later on, the available experimental evidence on the behaviour of
unsaturated soils is the result of more than 80 years of research on the retention properties,
soil resistance and deformation. The amount of generated information is considerable, and
the research from the last 10 years provides evidence for the utmost levels of complexity
about couplings in unsaturated soils. The existing database is judged as sufficient for the
development of a new coupled constitutive model, only with few experimental tasks to
complement the knowledge of a given material or for parameter determination purposes.
The justification of the present review of techniques for imposing and measuring suction is
not indeed only experimental. The knowledge of the currently-used and up-to-date
laboratory techniques also provides a realistic frame for the upcoming theoretical approach.
The choices of stress variables and material parameters, among others, will particularly refer
to the experimental limitations.
The first technique for imposing suction in a soil sample is called the tensiometric plate,
whose principle is very close to the physics of real problems of partial saturation (Fig. 2.8).
The sample is placed in a cell (a) under atmospheric air pressure. The bottom of the soil is in
contact with a ceramic stone with high air entry value. This element, also called high air
entry disk, is a manufactured rigid material with small pores of fairly uniform sizes. The
ceramic stones serves to ensure a separation between the water (liquid) of the underneath
reservoir and the air (gas) in the top part of the cell. Supposing that the ceramic stone is
initially saturated in water, the size of the disk pores prevents air from breaking into the pore
space due to the surface tension property. This property is verified for an excess of air
pressure over the water pressure lower than the specified air entry value of the disk
(typically ranging from 150 to 1500 kPa). It is assumed that the contact between the bottom of
the soil sample and the top of the ceramic stone is perfect so that the liquid water phase is
Figure 2.8 Simplified layout of tensiometric plate
continuous at the sample-stone interface. A container (b) is connected to the cell via a U-
shaped pipe. If placed below the cell, the container (b) generates a negative pressure
proportional to the negative fluid head. The pressure of water in the reservoir being
transmitted to the sample, the soil pore water becomes lower than the external air pressure
still at the atmospheric level. The tensiometric plate is thus a means to impose matric suction
(Eq. 2.6). The container can also be replaced by a water pressure transducer that controls the
negative pore water pressure.
The well-known limitation of such a negative pore-pressure system is the risk of
cavitation within the liquid phase. Cavitation is the formation of air bubbles within the
liquid phase due to an excessive tensile pressure. The bubbles are formed from a
combination of dissolved air and entrapped micro-bubbles. The limit of cavitation of water is
approximately -80kPa.
Using the same technique, matric suction can be measured from a sample that is in
equilibrium at a given negative pore water pressure with a tensiometer. The principle is
exactly that of Fig. 2.8, except that the difference of fluid height between levels (a) and (b) is
measured and not imposed. The tensiometric technique is the most used for in-situ
measurement of matric suction. Again, the classical limit is imposed by the cavitation
threshold of -80kPa. However, it has been demonstrated that the limit can be shifted down to
-1500 kPa (e.g. Jennings and Burland 1962, Tarantino and Mongiovi 2000a) by reducing the
size and thickness of the water chamber in part (a).
It has been determined in part 2.2 that the matric suction arises from the difference
between the air and water pressure in the interstitial space. The curvature of the menisci
strictly depends only on this pressure difference, and not on the absolute value of each
pressure. Consequently, the definition of matric suction can be re-written:
s = pa − pw ∀pa , ∀pw (2.11)
In order to avoid the cavitation effect due to negative pore water pressure, the principle of
axis translation technique is to use a water pressure that is positive or null and to apply an
excess air pressure to generate a positive suction. In the pressure plate apparatus (also called
Richards’ cell), Fig. 2.9 the air pressure is controlled by a pressure transducer while the water
pressure is either maintained to the atmospheric pressure or imposed positive. The artificial
16 CHAPTER 2
Figure 2.9 Simplified layout of pressure plate apparatus.
increase in pressures increases the risk of occluded air bubbles, as discussed for instance by
Tarantino and Mongiovi (2000a).
The osmotic technique is an alternative way to impose suction into the soil sample (e.g.
Cui and Delage 1996, Romero 1999, Tarantino and Mongiovi 2000b). A sample is in contact
with a semi-permeable membrane. On the other side of the membrane, a solution of Poly-
Ethylene Glycol is circulating. The size of the PEG molecules is such that they cannot pass
through the membrane, while pure water molecules can flow through the membrane. Here
the difference of concentration between the two sides of the membrane is at the origin of a
flow of water from the soil sample towards the circulating solution, creating suction in the
soil pores. Considering that the air pressure is taken equal to the atmospheric pressure, the
pore water pressure is truly negative. Unlike the axis translation technique with air
overpressure, the experimental set up can be very simple with no particular safety
disposition due to excess fluid pressures.
The methods reviewed previously are dealing with pore fluid pressure differences and
thus concern matric suction only. To a different extent can be reviewed the methods of
imposition and measurement of total suction. It has been shown previously that the osmotic
suction is little varying with the degree of saturation of the soil; the change in matric suction
can be taken as a measure of the change in total suction, and vice-versa. The measurements
of total suction are of particular interest in the ranges of high suctions, where the
measurement of matric suction gets more difficult due to the discontinuity of the liquid
water phase. In the vapour phase technique, the sample is placed in a confined atmosphere
with controlled humidity until equilibrium is reached. The free energy will correspond to the
imposed relative humidity of the soil (Fredlund and Rahardjo 1993). But the methods using
total suction are mostly used for measurement purposes rather than imposition. The
principle of the psychrometric approach is for instance to measure relative humidity next to
the interstitial water of the partially saturated soil. In the filter paper technique (e.g. Fawcett
and Collis-George 1967), paper foils are maintained in a soil sample until equilibrium of
suction is reached (i.e. suction in the filter paper equals suction in the soil). The water content
of the central foil is determined by weighting and the suction is deduced via the calibrated
retention properties of the filter paper (relationship between water content and suction).
Several different experimental techniques for measuring and imposing suction have been
reviewed in this section. The corresponding devices show various levels of complexity and
require non conventional geotechnical testing equipment. Each of the techniques covers only
one type of suction (matric and total) so that the only way to determine each component of
suction is to combine several techniques. However, in most of the publications reviewed
hereafter, no clear distinction is made between matric suction ( s ) and suction. It can be
reasonably assumed that, unless otherwise specified, the matric suction is the prevalent part
of total suction and that all particular features of behaviour of unsaturated soils can be
related to matric suction instead of total suction. In absence of counter-indication, the term
“suction” alone will designate “matric suction” in the rest of the text.
2.3.2 Loading paths in unsaturated soils

Asssuming perfectly isothermal conditions, the state of a material saturated in water is
completely described via two stress variables, that are total stress σ and pore water pressure
pw and one volumetric variable that is for instance the void ratio e (ratio of volume of voids
over the volume of solid). In this particular case, the two stress variables can be combined
into a so-called “experimental stress variable” that is the effective stress σ ′ of Terzaghi
(1936):
σ ′ = σ − pw (2.12)
The effective stress can be qualified of “experimental stress variable” provided that “all
measurable effects of a change of stress of the soil, that is compression, distortion, change of
shearing resistance, are exclusively due to changes in effective stress” (Terzaghi 1936). The
state of the material is thus completely described by the experimental stress state variable σ ′
and the volumetric state variable e .
Eq. 2.12 cannot be generalised directly to partial saturation. Indeed, from the definitions
of the previous part, the stress variables necessary to characterise the state of the partially
saturated soil are the total stress σ , pore water pressure pw and pore air pressure pa . The
volumetric state variables are the void ratio e and one parameter describing the relative
volumes of water and air, for instance the degree of saturation S r . Like in the previous
saturated case, it is possible to combine the measures of stress ( σ , pw , pa ) into some
“experimental stress variables” that control the complete state of stress of the material.
However, no single experimental stress variable has ever been found to govern alone all the
measurable effects of a change of stress of the unsaturated soil. The critical issue of the choice
of stress variables and possible combinations between the measures of stress σ , pw and pa
will be extensively addressed in next chapter. Indeed, the present chapter is intended to
focus only on the features of behaviour of unsaturated soils, and not yet to adopt a
theoretical concept or follow a certain school of thought.
At this point of the research, the volumetric state variables e and Sr need to be
expressed as functions of experimental stress state variables. Jennings and Burland (1962),
followed by Fredlund and Morgenstern (1977) among other, demonstrated that the two
following variables could be taken into consideration for this purpose:
σ net = σ − pa (2.13)
s = pa − pw (2.14)
σ net is called the net stress, while Eq. 2.14 corresponds to the definition of matric suction.
It has been demonstrated via null tests (defined in part chapter 3, part 3.5.2) that the state of
18 CHAPTER 2
stress within the partially saturated soils is fully described if σ net and s are known. Those
stress variables provide a convenient measure of the stress, provided that they are
independent, and that they correspond precisely to the stresses imposed experimentally (see
part 2.3.1). The net stress and matric suction will thus be adopted tentatively as the
experimental stress variables for the state of the art review.
The methods of imposition (and measurement) of suction are practically compatible with
any conventional soil mechanics testing apparatus, that is for instance oedometer, triaxial
device, true triaxial cell or direct shear box. Suction can be applied in these modified devices,
and will thus become a controlling variable for the partial saturation. The corresponding
measurement of volume for the partial saturation is obviously the degree of saturation.
These two variables will be referred to as the “capillary variables”. By opposition, the net
stress is a measure of the imposed exterior (mechanical) stress. It can be logically related to
the void ratio. Those two variables will be called the “mechanical variables” in the rest of the
chapter.
Given the technical limitations due to partial saturation (time of equalization, fluids under
high pressures), it is usual to impose practically either a suction change or a net stress
variation not simultaneously. Considering the reference case of saturated soils where the
stress paths for laboratory tests are drawn in the space of triaxial stress variables (mean stress
p and deviatoric stress q ), the occurrence of non zero suction necessitates a third axis of
stress ( s ), Fig. 2.10. Two types of stress paths are then possible: (i) a mechanical path
corresponds to variations of net stress in planes of constant suction (paths 2 and 3 in Fig.
2.10) and (ii) a suction loading path is defined as variations of suction in planes of constant net
stress (path 1 in Fig. 2.10).
2.3.3 Stress-strain response of unsaturated soils

Most of the features of the stress-strain behaviour of unsaturated soils are now well
characterised and accepted within the geotechnical community. For instance, the so-called
“wetting collapse” (e.g. Jennings and Burland 1962) and suction-induced stiffening, to be
reviewed hereafter have been thoroughly investigated for almost a century. Following the
principal objective of the chapter that is to build a list of essential features of behaviour to be
included in a modelling framework for unsaturated soils, those conventional elements of
stress-strain response have to be recalled once synthetically. The commonly accepted general
s= c q
o n st
3
4
2 B
s 1 p n e t
t
c o n s A
=
p n et
Figure 2.10. Possible stress loading paths to be considered for unsaturated behaviour.
aspects are thus deliberately overviewed only briefly. Yet, some more complex behavioural
features that have been highlighted only in recent literature make the object of deepened
discussions in this section.
While the loading paths can be either of a purely mechanical nature, or correspond to
suction changes (Fig. 2.10), the soil response can be quantified in terms of changes in volume
and in quantity of water and air. Each of the volumetric state variables (e.g. void ratio e and
degree of saturation S r ) is a priori likely to be affected by any of the two types of loading
paths. Consequently, it is possible to classify the features of behaviour either by the type of
loading path or by the type of volumetric response.
In anticipation of the discussion on the stress and strain variables for unsaturated soils to
come in chapter 3, it is chosen here to review the behaviour of unsaturated soils from the
viewpoint of volume changes. This first part will indeed investigate the evolution of void
ratio (for all combinations of loading paths) and the next section 2.3.4 will focus only on the
variations in the degree of saturation (against suction changes and mechanical loading).
Isotropic stress-strain behaviour and compressibility

In the space ( pnet , q, s ) of Fig. 2.10, one of the most investigated planes of stress paths is
that of the constant level of suction (plane s = const ). The historic reason for such an interest
is certainly that the reference saturated case used in conventional soil mechanics belongs to
one of those planes of constant suction, with the particular case s = 0 . The experimental
techniques and devices reviewed previously are also favourable to carrying out mechanical
compressions on a sample under a fixed level of suction. It is recalled that a mechanical
loading corresponds by convention to an increase in net stress in this chapter.
Restricting the analysis to isotropic mechanical compressions (path 2 in Fig. 2.10), the
quantitative and qualitative response in void ratio can be assessed in Fig. 2.11. The results of
Fig. 2.11a are provided as an example, but similar trends have been observed in a number of
published works (e.g. Matyas and Radhakrishna 1968, Leclercq and Verbrugge 1985,
Sivakumar 1993, Geiser 1999, Vicol 1990, Al Mukhtar 1995, Mâatouk et al. 1995). The
volumetric response to isotropic compression under various levels of suction can be
summarized qualitatively (Fig. 2.11b) with the following remarks:
(i) The stress-strain response is decomposed into an elastic part and an elastoplastic
part, like in the conventional saturated compression test.
(ii) The response is quasi bi-linear when plotted in a ( e − ln pnet ) plane.
(iii) In the elastic zone (when visible), the stiffness is little to moderately affected by the
level of suction.
(iv) The mechanical compressibility in the plastic range decreases with increasing
suction.
(v) The limit of plasticity is suction-dependent.
The last point (v) will make the object of a dedicated paragraph. The overall effect of
suction is often identified as a global stiffening of the material. Yet it appears that the rate of
change of elastic and plastic slopes with suction is not constant from one material to the
other. Geiser (1999) followed by Rifa’i (2002) worked for instance on the quantification of the
evolution of those compression slopes with suction. Using the terminology of Cam-clay
model (Schofield and Wroth 1968) the elastic and elastoplastic slopes in plane ( e − ln pnet )
20 CHAPTER 2
1.2
(a)
1.15
Void ratio e(-)
1.1
1.05
s=0 kPa
1 s=100 kPa
s=200 kPa
s=300 kPa
0.95
20 40 60 80100 300 500
Mean net stress p (kPa)
net
Figure 2.11 Void ratio changes under the effect of isotropic mechanical load at several constant levels
of suction. (a) Results for kaolin (Sivakumar 1993). (b) qualitative synthetic interpretation.
0.1 0.5
(a) (b)
Elastic slope κ, Elasto-plastic slope λ
Sion silt
0.08 0.4 Jossigny silt
Elasto-plastic slope λ
Trois-rivières silt
λ
Kaolin
0.06 0.3
0.04 0.2
0.02 0.1
κ
0 0
0 100 200 300 0 400 800
Matric suction s (kPa) Matric suction s (kPa)
Figure 2.12 Evolution of elastic and elastoplastic slopes with suction for (a) Sion silt, (b) Several
materials, from Geiser (1999) and Rifa’i (2002)).
are respectively called κ and λ (Fig 2.11b). For the clayey silt studied by Geiser and Rifa’i,
the evolution of the slopes is drawn in Fig. 2.12a.
While in the case of Trois-Rivières silt and compacted Kaolin the compressibility index
seems to increase with suction, other materials like Sion silt evidence a maximum value for
the plastic coefficient. Some other cases, e.g. Josa (1988), can be found to show on the
contrary a decrease of λ with suction. There are several possible explanations for these
differences in compressibility changes. The evolution could follow possibly the same trends
(increase and decrease), but at different suction levels. Since the range of applied suction is
generally restricted to a few hundreds of kPa in the reviewed experimental results, it is
probable that the representation of compressibility coefficient evolution is incomplete. The
results for the evolution of the elastic slope show a more consistent pattern with moderate to
negligible variations.
Preconsolidation stress
In complement to the observations on the compressibility, the isotropic compression tests
at various fixed levels of suction evidence a dependency of the size of the elastic domain. The
materials reviewed in the previous paragraph showed a common clear trend for a larger
elastic domain (that is the domain of reversible stress-strain response) with increasing
suctions. Using once more the Cam-Clay terminology (Schofield and Wroth 1968), this
dependency law can be expressed in terms of preconsolidation pressure pc′ . As the
preconsolidation pressure is the indicator of the over-consolidation ratio (OCR), it represents
an appropriate measure of the degree of apparent hardening undertaken by the soil in drier
conditions. Fig. 2.13a shows a typical plot for the evolution of preconsolidation pressure with
matric suction. Knowing that in a saturated material, the limit of plasticity is dependent on
the initial state of the material, the determination of a curve of the type of Fig. 2.13a should
rely on a bunch of samples with a comparable reference state. The fact that in literature pc is
drawn versus matric suction and not versus total suction is not unintentional. It has been
indeed often noticed that the effects of capillarity were largely predominant with regards to
modifications in the size of the elastic domain. Furthermore, it can be observed that the curve
( s − pc ) is highly non linear (e.g. Alonso et al. 1987). In Fig. 2.13b, it is proposed to highlight
three main zones of interest in the suction range:
(a) The first domain A of suction corresponds to the lowest levels of suction. In this
range, the preconsolidation pressure is little affected by the suction change.
(b) In the second zone B, on the contrary, the change in preconsolidation pressure has a
larger amplitude, indicating a fundamental change in the over-consolidation ratio.
(c) For the highest levels of suction (zone C), the dependency of the plastic limit on
6
1.2 10 300
(a) (b)
Zone C
6
1 10 250
Matric suction s(kPa)
Matric suction s(Pa)
5
8 10 200
5
6 10 150 Zone B
5
4 10 100
5
2 10 50
Zone A
0 5 5 5 5 6
0
0 2 10 4 10 6 10 8 10 1 10 280 300 320 340 360 380 400
Preconsolidation pressure p (Pa) Preconsolidation pressure p (kPa)
c c
Figure 2.13 Evolution of the preconsolidation pressure with suction (data from (a) Kane (1973) and (b)
Geiser (1999)).
22 CHAPTER 2
suction gets weaker.

A possible preliminary interpretation of the existence of these three zones could rely again
on the capillary effects. It is widely accepted that the surface tension property, at the scale of
soil pores, is at the origin of internal tractions on the soil grains which tend to stiffen the
material (Wheeler et al. 2003). Regarding the matric suction as a scale for the measurement of
the number of water menisci, the three zones can be revised as follows:
Zone A corresponds to the domain of positive suction below the air entry value. The air
entry value se was defined previously in the description of the ceramic stones. It has the
dimension of a pressure difference between water and air, and represents the threshold
below which the surface tension at the capillary skin is sufficient to prevent air bubbles to
break into the pore space (Fredlund and Rahardjo 1993). For any suction below the air entry
value, the soil pores are fully saturated in water, which means that no capillary effect should
take place. Indeed, the preconsolidation pressure remains close to the reference saturated
preconsolidation pressure in this domain.
At a certain level of suction (zone B in Fig. 2.13b), the difference between air and water
pressure is such that the surface tension at sample boundaries cannot sustain any more the
air inflow which penetrates into the pores. The capillary effects are thus activated at this
point. Since the transition between the perfectly saturated state and the pendular state is
sharp enough, the hardening-like induced effect can be expected to be large, which
corresponds to the important variations in preconsolidation pressure.
In the zone C, the suction level is considerably higher. The air phase is presumably
continuous while the action of menisci might by more sporadic due to the little changes in
amount of water. The capillary effects cannot develop further due to the state of residual
saturation, which explains the stabilisation of the preconsolidation pressure. This residual
state, characterized by the residual degree of saturation S res will be discussed more in detail
in the framework of the soil water retention curve.
Shear strength
The results of shearing unsaturated samples in triaxial apparatus or direct shear box (e.g.
Cui and Delage 1996, Geiser 1999, Boso et al. 2005) converge towards the conclusion that the
apparent resistance to shearing is improved by drying (Fig. 2.14). In the deviatoric stress
versus axial strain plane ( q − ε a ) , an increase in suction is at the origin of higher peak
values. Indeed partial saturation has an effect of both the initial shear modulus and the
maximum deviatoric stress. The observed behaviour in the deviatoric planes is consistent
with previous observations on the stiffening effects of capillarity.
The evolution of the critical state line with partial saturation is also of interest. In a
conventional deviatoric stress versus mean net stress plane ( q − pnet ) it appears that the
angle of friction remains constant with respect to the saturated friction angle, while the
apparent cohesion is clearly suction dependent (Fig. 2.15). The relationship between
apparent cohesion and suction is non linear and tends to a maximum for the majority of soils
(e.g. Alonso et al. 1987, Fredlund et al. 1978, Geiser 1999).
Undrained behaviour
While the drained shearing tests reviewed above require an imposition of suction, the
undrained triaxial tests require virtually lighter experimental equipment as it is needed only
to maintain the water content constant. If undrained tests are less discussed in literature,
they still provide relevant information on the short term behaviour of finer materials (Bishop
and Donald 1960, Rahardjo et al. 2004, Chiu and Ng 2003, Bolton 1986). Two examples of
material stress-strain response to shearing in unsaturated drained conditions are provided in
Figure 2.16.
The relevant aspects of the undrained shear behaviour can be summarized as follows:
2000 0
(a) (b)
1600
Deviator stress q (kPa)
0.01
Volumetric strain εv(-)

1200 0.02
800 0.03
s=0 kPa
400 0.04 s=0 kPa
s=100 Pa s=100 kPa
s=200 Pa s=200 kPa
0 0.05
0 0.125 0.25 0 0.15 0.3
Axial strain εa(-) Axial strain εa(-)
Figure 2.14 Evolution of the shear strength with suction for Sion silt, data from Geiser (1999).
400
200
(a)
(b)
350
Deviatoric stress q (kPa)
300
Vertical intercept q (kPa)
150
250
0
200
100
150
100 s = 300 kPa 50

s = 200 kPa
50 saturated s = 100 kPa
CSL s = 0 kPa
0
50 100 150 200 250 300 0
0 50 100 150 200 250 300 350 400
Mean net stress p (kPa)
net Matric suction s(kPa)
Figure 2.15 Typical evolution of critical state line with suction for compacted kaolin, data from
Wheeler and Sivakumar (1995).
(i) The volumetric change is not negligible, due to the compressibility of the gas (air)
phase.
(ii) The level of suction can increase or decrease, depending on the material and initial
state of stress before shearing
24 CHAPTER 2
400 1
(a) Test CD 50-150 (b)
350 Test CW 50-170
Total volume change (cm )

0
3
Deviatoric stress q(kPa)
300
250 -1
200
150 -2
100
-3
50 Test CD 50-150
Test CW 50-170
0 -4
0 10 20 0 10 20
Axial strain εa(%) Axial strain εa(%)
200
(c)
150
100
50
Test CW 50-170
0
0 10 20
Axial strain εa(%)
Figure 2.16 Comparison of responses to drained and undrained triaxial shearing tests for Juong sandy
clay (from Rahardjo et al. 2004). Test CD 50-150 was carried out in drained conditions, under a
confining pressure of 50 kPa and a constant matric suction of 150 kPa. Test CW 50-170 was carried out
at constant water content, with a confining pressure of 50 kPa and an initial suction of 170 kPa.
(iii) The degree of saturation can also show some increasing or decreasing trends.
Provided that the non-zero volume changes will have an influence on the degree of
saturation, the variations in suction will be also a consequence of the change in capillary state
of the material. Only an analysis using a valid experimental effective stress (see chapter 3)
should enable to model the undrained behaviour within a constitutive framework written
primarily for drained conditions.
Wetting pore collapse

Except for the undrained case, the features of behaviour overviewed up to this point were
defined in planes of constant suction. Investigating in turn the planes of constant net stress
(experimental mechanical stress), the effects of suction changes (e.g. path 1 in Fig. 2.10) can
also be interpreted from the perspective of skeleton volume changes. The most cited feature
of behaviour of unsaturated soils in literature, the so-called “wetting-collapse” belongs to
this category. It designates the property of partially saturated soils to undergo a large
volumetric compression under humidification undertaken at a sufficiently large level of
mechanical stress. Fig. 2.17a plots for instance the volumetric response to a suction reduction
in compacted Pearl Clay (Sun et al. 2007) under a constant mean net stress of 98 kPa. Along a
subsequent drying path, it is usually observed that the material keeps a memory of the
maximum volumetric strain previously reached which indicates that the wetting collapse is a
plastic phenomenon. In some cases, the irreversible compression due to wetting is preceded
or followed by a slight swelling (Fig. 2.17b, from Sivakumar 1993)
Several authors observed that the suction ranges corresponding to the phases of swelling
and collapse are highly dependent on the applied mechanical stress and on the loading
history (Alonso et al. 1995, Sharma 1998, Buisson and Wheeler 2000).
Like previously, the wetting-collapse is one of the expressions of the conventional lexicon of
unsaturated soils that is fairly imprecise and misleading. In tunnelling engineering (e.g. Kim
et al. 2006), collapse refers to catastrophic breaking down of an underground structure. In
conventional slope stability problems (e.g. Reid and Brien 2006) a collapsed area designates a
mass of soil that ran down after slope failure. Collapse is thus conventionally associated with
an event characterised by large deformations, suddenness, and overcoming a threshold of
ultimate resistance.
The experimental evidence of Fig. 2.17 meets that implicit definition of collapse in some
aspects, with a larger amount of deformations and a relatively sudden change in
compressibility (e.g. Sudhakar and Revanasiddappa 2006). However, the phenomenon has
nothing to do with failure of the soil in the perspective of ultimate resistance. It corresponds
rather to an irreversible rearrangement of particles, meaning that the material enters a
hardening plasticity mode.
-1 -2
(a) Wetting (b)
0 Wetting
-1
Wetting
Volumetric strain εv (%)
Volumetric strain εv (%)
1
0
2
1
3
Collapse Swelling
e =1.34
0 2
4 e =1.26
0
e =1.17
5
0 3
e =1.03
0
6 4
0 100 200 0 200 400
Figure 2.17 Wetting collapse phenomenon in soils under mechanical stress (a) Pearl clay,
experimental data from Sun et al. (2007) e0 is the initial void ratio before wetting (b) Kaolin,
experimental data from Sivakumar (1993)
26 CHAPTER 2
Since the wetting-collapse terminology is however widely accepted in the geotechnical

community, it will be also employed in this manuscript. Yet it is suggested to clarify the
expression by the use of wetting pore-collapse. The notion of collapse of pores is often used in
the framework of structured materials (e.g. Koliji et al. 2006). That updated appellation is
more compatible with the evidence of plastic compression that is a reduction of the pore
space. In a sense, the failure or collapse occurs at the scale of the pore (which vanishes) and not
at the scale of the sample.
Complete drying-wetting cycles

The wetting pore-collapse is the material response to a very particular case of suction-
loading path. In the real cases where partial saturation is relevant, the suction undergoes
cycles which corresponds to alternate conditions of drying and wetting. As the experimental
stress variable suction alone is at the origin of skeleton (mechanical) deformations, some
authors cited later proposed to use an analogy between the purely mechanical stress-strain
relationship and the suction-strain behaviour. Using the experimental variables, the variation
of void ratio Δe can thus be decomposed into a part due to the mechanical stress ( Δe m ) and
a part due to suction ( Δe s ). The corresponding constitutive equations, written in terms of
skeleton stress ε ije , could theoretically be also decomposed into a part due to the mechanical
stress (Eq. 2.14) and a part due to suction (Eq. 2.15). The elastic strain increment d ε ije is thus
the sum of the individual strain increments d ε ije ,m and d ε ije , s (Eq. 2.16):
d ε ije ,m = Cijhk
m
(dσ hk − dpaδ hk ) (2.14)
d ε ije , s = C s (dpa − dpw )δ hk (2.15)
d ε ije = Cijhk
m
(dσ hk − dpaδ hk ) + C s (dpa − dpw )δ hk (2.16)
Where Cm and C s are matrices of material characteristics. More explicitly, λ and κ have
been introduced previously as the slopes of the virgin compression line and the unloading
reloading line in plane ( e − ln pnet ) . By analogy in the plane ( e − ln s ) , attempts can be found
in literature to characterise the compressibility coefficients with respect to suction load ( λs
and κ s ). The two questions to be answered by the review are the following:
(i) Is it consistent to introduce such compressibility coefficients ?

(ii) If yes, are these coefficients suction dependent ?
A first family of experimental evidence seems to comfort the analogy between planes
( e − ln pnet ) and ( e − ln s ) . Two examples of void ratio evolution in response to a suction
increase (drying path) are provided in Fig. 2.18, taken from Yahia-Aïssa et al. (2000) and
Fleureau et al. (1993). In Fig. 2.18a, one can read a clear plastic threshold along the suction
axis, for around 7 kPa, whatever the level of applied net stress during the drying tests. The
volumetric response is that of an over-consolidated material, as if the material had in
memory a suction of 7 kPa that would have been previously reached. Similarly, in Fig. 2.18b,
a plastic threshold is reached at a negative pore pressure of around 20kPa. The wetting path
follows a unloading-reloading line whose slope κ s is very consistent with that of the initial
elastic part. The response in plane ( e − ln s ) can thus be a priori modelled with a bi-linear
formulation and defining a “suction of preconsolidation” (limit of plasticity). Interestingly,
0.55 2
(a) Drying (b)
1.8
0.5
1.6
Void ratio e(-)
Void ratio e(-)

Drying
λ
s
0.45 1.4
Wetting
1.2
κ
0.4 σnet =5.2 MPa s
σnet =15.6 MPa 1

σnet =20.8 MPa
0.35 0.8
0.1 10 0.1 1 10 100 1000 10000
Matric suction s (MPa) Matric suction s (kPa)
Figure 2.18 Response of void ratio to suction increase (a) FoCa7 clay, from Yahia-Aïssa et al .(2000) (b)
White clay, from Fleureau et al. (1993).
the materials exhibit this elasto-plastic behaviour along the first drying path already,
meaning that they have never experienced a previous non-null level of suction. Indeed, the
initial preconsolidation stress, in the mechanical sense (the mechanical over-consolidation
ratio is more than 1 in the compacted samples of clay) has been identified to be close to the
observed suction of preconsolidation. This implies that the two measures of stress are
somehow interrelated. For instance, in the works of Josa (1988), a drying process is carried
out on remoulded materials (i.e. normally consolidated): the volumetric strain show no
initial elastic part but only plastic compression.
The determined elastic and elasto-plastic slopes ( λs and κ s ).are here considered as
unique. Table 2.2 show some examples of comparison between this slopes and the
conventional rigidity and compressibility coefficients determined under isotropic mechanical
loading in saturated conditions ( λ and κ ). For the reviewed materials, the values of the
respective slopes show very close comparability. Several authors simply assume that virgin
slopes are identical in both ( e − ln p′ ) and ( e − ln s ) planes (E.g. Zerhouni 1991, Taibi 1994).
This again accounts for a strong coupling between the suction and the mechanical stress-
strain behaviour.
Incidentally, all the stress-strain paths of this type were carried out in almost fully
saturated conditions. As discussed previously, there is a range of non-zero suctions such that
the degree of saturation remains equal to 1 and the air cannot break into the pore space.
Indeed the previous assumption of the existence of two slopes ( λs and κ s ) is valid only for
Table 2.2 Comparison between saturated compressibility coefficients and suction-compressibility

coefficients (determined from data of Fleureau et al. 1993).
Material κ κs λ λs
White clay 0.03 0.03 0.18 0.18
Yellow clay - - 0.09 0.085
Montmorillonite - - 0.04 0.045
28 CHAPTER 2
suctions that are lower than the air entry value se . According to Vicol (1990) dealing with a
clay in an initial slurry state, irreversible deformations occur when the suction increases from
zero to the air-entry value and become then fully reversible. Similar results are reported by
Jennings and Burland (1962), Kogho et al. (1993), Fleureau et al. (1993) Péron (2008) among
others. Two examples of experimental evidence for this phenomenon are plotted in Fig.2.19.
Some authors prefer to define a dedicated threshold in suction to for the slope change by
introducing the suction of “shrinkage limit”. Since it is widely accepted that this suction of
shrinkage is close to the air entry value, it is proposed to assume from now on that these two
limits are equal. The determination of the air entry value will be discussed more in depth in
paragraph 2.3.4.
The slope of the elastic part for suctions beyond the air entry value is reported either as
non constant in plane ( e − ln s ) or otherwise fairly stable but still lower from the initial
elastic slope κ s determined in the zone of over-consolidation. The global trend for void ratio
is to stabilize at highest suctions, indicating that the suction effects reach a limit. Also it is
reported that when a sample has been consolidated or compacted above a particular density,
suction increase is no longer able to introduce any compressive volumetric strain (Vicol
1990). The volume response to humidification from this residual state of saturation (Fig.
2.19b) is fully reversible, with again a slope change at the air-entry value.
If the effects of complete and successive suction cycles were seldom investigated in
traditional mechanics of unsaturated soils, that type of stress path makes the object of an
increasing number of publications (e.g. Alonso et al. 1995, Sharma 1998, Buisson and
Wheeler 2000), noticing that the tested materials are often compacted or may be classified in
highly swelling clays. The volumetric response to such wetting-drying cycles is again
strongly dependent on the mechanical stress applied, as well as the initial over-consolidation
ratio. Unlike the previous cases of Fig. 2.19 where a simple cycle of drying and wetting back
is undertaken under a low level of net stress, successive cycles of suction are at the origin of
irreversible compression along both the drying paths and the wetting paths. These
0.84 0.8
(a) (b)
0.7
0.83
0.6
Drying
Void ratio e(-)
Void ratio e(-)
0.82 0.5
Drying
λ 0.4
s
0.81 0.3
0.2 Wetting
0.8
0.1
s
e
0.79 0 -2 -1 0 1 2 3 4 5 6
1 10 100 1000 10000 100000 10 10 10 10 10 10 10 10 10
Figure 2.19 Effect of suction changes on void ratio, with suction values far above the air entry value.
The air entry value is determined in a plane ( S r − s ) or ( w − s ) . (a) Silt, initial condition: slurry
(Jennings and Burland 1962) (b) Ca-Montmorillonite, initial condition slurry (Fleureau et al. 1993).
irrecoverable densifications of the material correspond indeed to the two plastic behaviours
reviewed previously, namely the wetting pore-collapse and the plastic compression upon
drying. The volumetric response to successive suction cycles is highly dependent on the
initial state of the tested materials, so it is a challenge to choose a data set that could be
considered as “generic” to illustrate this paragraph. However, the data of Alonso et al. (1995)
and that of Buisson and Wheeler (2000) show some volumetric trends that are very
consistent together (Fig. 2.20).
In Fig. 2.20, swelling is exclusively associated with wetting paths while pore-collapse
occurs during the first wetting phase. Subsequent drying-wetting cycles reveal an initially
over-consolidated phase followed by virgin compression. The apparent suction of
6 2.1
(a) (b)
4 2.08 B Wetting A
C
Volumetric strain ε (%)
B Wetting
2 Specific volume v(-) 2.06
v
Drying
Drying
0 D 2.04 E
A
F
-2 C 2.02 D
E
-4 2
F
-6 1.98
0.001 0.01 0.1 1 10 100 1000 10 100 1000
Matric suction s (MPa) Matric suction s (kPa)
Figure 2.20 Successive wetting-drying cycles carried out under constant net stress (a) Boom clay,
oedometer conditions, vertical net stress 100 kPa (Alonso et al. 1995) (b) Compacted kaolin, isotropic
stress state, mean net stress 50 kPa (Buisson and Wheeler 2000).
0.1
D
B
0.05
Wetting
Volumetric strain εv (-)
Wetting A
0
Drying
C
-0.05
Drying
-0.1
E
-0.15 4 5 6
10 10 10
Mean effective stress p' (Pa)
Figure 2.21 Wetting drying cycles on compacted mixture of bentonite and kaolin under constant
isotropic net stress of 10kPa, from Sharma (1998).
30 CHAPTER 2
preconsolidation (limit of slope change) seems to decrease with the number of cycles. This
trend is confirmed by the data of Sharma (1998), where the suction of preconsolidation
passes from a value of 110 kPa in the first wetting-drying cycle to a value of 22 kPa in the
second wetting-drying phase (Fig. 2.21).
Conclusions on volumetric behaviour

The principal features of the volumetric behaviour of unsaturated soils have been
reviewed in this part. The volume changes are measured as a response to combination of
pure mechanical loads and cycles of suction. Using the experimental variables, it appears
that each of the stress variables contributes to the modification of void ratio. The most
challenging issue for upcoming constitutive modelling will be to describe accurately the
switching between reversible behaviour and phases of plastic hardening. Preliminary
indications of coupling between the mechanical stress state and the volumetric response to
suction changes have been drawn.
2.3.4 Retention behaviour

The experimental variables identified as the counterpart of the void ratio can be either the
water content or the degree of saturation. Both variables can be used to define the quantity of
water in the pore space. Yet the degree of saturation provides a natural normalisation of the
volume fractions of water and air by using the volume of voids within the material. The
degree of saturation will thus be favoured for the discussions in the present section and
throughout the manuscript. It will be determined that this variable can respond to a change
in any of the experimental stress variables that are matric suction or net stress.
The published data about the conventional Soil Water Retention Curve (SWRC) will be
first overviewed. The SWRC is defined as the relationship between the degree of saturation
and matric suction, and is usually determined under a null or constant level of net stress
(path 1 in Fig. 2.10). Again, the purpose of this chapter is essentially to set the terminological
frame of the research and to fill in the list of specifications for constitutive modelling. With
this respect, the issue of couplings in the soil water retention behaviour needs to be more
thoroughly investigated besides the introductory part presented hereafter. While this
introduction is the prerequisite to the upcoming review of constitutive models, chapter 5 will
address more in detail the soil water retention features.
Figure 2.22 Idealised shape of soil water retention curve, drying path only.
The typical shape of the water retention curve under a drying path is drawn in Fig. 2.22.
The air entry value se can be determined by the method of tangents intersection, while the
residual degree of saturation S res is identified as the y-ordinate of the horizontal asymptote
of the limit state. The funicular range is defined as the domain where the pore water and
pore air phases are continuous, while the pendular domain corresponds to continuous air
phase and discontinuous water phase.
Note on terminology
Describing the behaviour of a soil partially saturated in water (or more precisely fully
saturated in liquid water and gaseous air) requires several types of analyses. Referring to
conventional soil mechanics definitions:
(i) A purely hydraulic problem is commonly associated to water seepage. The
hydraulic process describes water flowing in or out the individual voids. In
hydraulic problems, the main unknowns to focus on are the fluid velocity and
outflow.
(ii) The mechanical straining refers to deformation of the solid skeleton under the effect
of a load. The variables of interest are thus stress and strain.
(iii) Then, the mechanical consolidation problem involving both straining and water
outflow, it is classified as a coupled hydro-mechanical analysis.
(iv) If more than one fluid phase co-exist in the porous space, the repartition and
pressures of fluids have to be quantified. This information is gathered in the
retention behaviour.
Coupled modelling of soils saturated in water thus requires, for the more general case,
both a mechanical model and a water flow model. Once a gas appears in the pores besides
liquid water, a new level of complexity is reached. On the one hand, the previously cited
mechanical and hydraulic models are likely to be influenced by the repartition of fluids, but on
the other hand, this repartition itself has to be modelled via a retention model. Activating
multiple couplings between “submodels” raises the need for a proper and distinct
denomination of each behavioural model. However, many works from literature tend to
confuse water flow and water repartition, using the adjective ‘hydraulic’ for qualifying both.
It is proposed hereafter to clarify the definition and meaning of words and to justify the
terminology used throughout the thesis.
The etymology of “hydraulic” is Greek, and was used in expression hydraulikos organon,
meaning literally "water organ". Word hydraulikos itself comes from hydr-, stem of hydor
"water", and aulos "musical instrument, hollow tube." The word was later extended (in Latin)
to various kinds of water engines.
“Hydraulic” can thus define either a system (or medium) in which liquid water acts, or a
hollow medium through which water flows. It is therefore improper to associate the
restrictive term ‘hydraulic’ with ‘retention curve’. Indeed, soil water retention curves
accounts for variations in degree of saturation Sr with respect to suction s. Both quantities
are calculated on the basis of both liquid and gaseous phases’ information, i.e. water and air.
Thus, ‘hydraulic’ should be dedicated only to –liquid- water flow problems, in link with
permeability. The ‘action’ of water in this type of problem is mainly due to the flow and
percolation.
The retention behaviour, on the other hand has to be associated with a word reflecting the
two fluid phases together. Among the most adequate terms, “retention” and “capillary” are
of interest. In recent years, many authors (Wheeler et al. 2003, Tamagnini 2004, Sheng et al.
32 CHAPTER 2
2004) are calling the hysteresis appearing in the soil water retention curve, to be explicated in
next paragraph, “hydraulic hystereris”. They also use the terminology “hydraulic
behaviour” for soil response to suction changes. Li (2005) defines a “wetting/drying
hysteresis”, and Pham et al. (2003) call it “hysteresis for soil-water characteristic curve”.
This use of “hydraulic” automatically creates an amalgamation between the retention
curve and the flow problem, which has to be avoided, see previous paragraph. The choice of
terminology probably came from the idea that drying and wetting have a “hydraulic”
connotation the same way “drainage” has. However, processes of drying and wetting cycles
call for presence of both water and air (liquid and gas).
When evidencing irreversibility in the retention curve for one of the first times, (Philip
1964) chose to name it “capillary hysteresis”. It seems nowadays appropriate to come back to
that terminology (e.g. Muraleetharan and Wei 2004). By extension, any coupled model that
includes a soil water retention curve (or an equivalent for it) could be said to own a
“mechanical behavioural model” and a “capillary behavioural model” or more simply a
“water retention model”, which will be adopted from now on. The term “hydraulic” should
from now on be exclusively reserved to water liquid phase flow (see for instance chapter 6).
Capillary hysteresis
The hysteretic property of the soil water retention curve is highlighted by numerous
research works. When expressed in terms of degree of saturation versus suction, the path of
drying is situated above the wetting path, Fig. 2.23.
The conventional elements of explanation for the capillary hysteresis are physical
properties of the soil medium. Firstly, according to Taylor (1948) and Fredlund and Rahardjo
(1993), the capillary hysteresis is typical of porous media constituted of variable size pores
interconnected in parallel or in series. In the model with parallel connections of Fig. 2.24a,
during drying, a given suction that corresponds to drainage of the bigger tube is not
sufficient to dry the smallest tube. The water stored in the smallest tube explains the
difference in volume of water between the wetting and drying paths. In the case of
connection in series Fig. 2.24b, several sizes of pores are encountered along the path of water
drainage; during drying, the drainage of a pore is produced only if suction becomes greater
1.2
1
Degree of saturation S (-)
r
0.8
Drying
0.6
Wetting
0.4
0.2
0
600 1000 3000 5000
Matric suction s (Pa)
Figure 2.23 Capillary hysteresis in a sand, data from Engel et al. 2003
Figure 2.24 Patterns of drainage and humidification in a network of (a) parallel tubes (b) pores
connected in series.
than surface tension generated by the radius r. On wetting, the filling of the great bigger pore
is on the contrary controlled by the other radius R.
The last conventional explanation for the occurrence of hysteresis is difference between
the wetting and drying contact angles of water with respect to solid phase (defined
previously in Fig. 2.3). Knowing that the wetting angle is larger than the drainage angle, the
resulting matric suction is higher during drying than wetting, for the same relative volume
of water.
Influence of void ratio
In the framework of constitutive modelling, new relevant questions are raised by Gallipoli
et al. (2003) considering the influence of void ratio changes on water retention curves. It has
been long accepted that the shape of the retention curve varies significantly depending on
the type of soil, see Fig. 2.25. More precisely, the granulometry has a clear influence on the
shape of the soil water retention curve, the location of the point of air entry and the residual
degree of saturation (Fredlund 1997, Li and Zhang 2007).
Granulometry provides information on the size and type of solid particles. However,
provided that the degree of saturation is the ratio between the volume of pore water and the
volume of the voids, it is more consistent to attribute the shape-dependency of the soil water
retention curve to the volume and shape of the pores, which are in fact, smaller in clays than
in sands. Soil material is a deformable medium and might be subject to important variations
of volume, especially during plastic events. Referring to the conventional soil mechanics
assumption (Terzaghi 1936) of incompressibility of soil grains, it is obvious that the
macroscopic volume change of the soil is in fact the variation of volume of the voids Also the
so-called irrecoverable re-arrangement of the grains could be interpreted as the plastic
rearrangement of the pores. Gallipoli et al. (2003a) deduce that a “variation of the void ratio
produces changes in the dimensions of voids and of connecting passageways between voids,
which cause corresponding changes in the SWRC”.
34 CHAPTER 2
Volcanic sand Fine sand

1.2
Glass beads Touchet silt loam

1
r
0.8
0.6
0.4
0.2
0
1 10
Matric suction s (kPa)
Figure2.25 Comparison of soil water retention curves for various materials (from Brooks and Corey
1964)
(a) (b)
Figure2.26 Effect of void ratio changes on degree of saturation in Speswhite kaolin, reported by
Gallipoli et al. (2003) (a) Isotropic mechanical loading unloading test at constant suction, experimental
data from Zakaria (1995) (b) Triaxial shear test at constant suction and constant mean stress,
experimental data from Sivakumar (1993).
They report evidence of those typical coupled effects observed in Speswhite Kaolin.
During an isotropic loading test carried out at constant suction, large plastic reduction of
void ratio occur which induces a significant increase in degree of saturation S r (Fig. 2.26a).
On the contrary, on unloading, only very small elastic component of skeleton swelling
occurred, with almost no change entailed for S r . Considering besides that shearing might
cause changes in void ratio, the process shall influence the degree of saturation too. Under
the application of a deviator stress, the maintenance of a constant suction and the decrease in
the total void volume cause the degree of saturation to increase in consequence (Fig. 2.26b).
More examples of direct dependency of the degree of saturation on void ratio will be
provided in chapter 5.
The previous examples analyse the coupling between the state of saturation and the
mechanical behaviour from the view point of a single fixed suction. The effect of void ratio,
initial density and level of net stress have been investigated recently with respect to the total
soil water retention curve (e.g. Vanapalli et al. 1999, Romero 1999, Sugii et al. 2002, Salager
2007). Fig. 2.27 shows an example of alteration of the shape of the SWRC with void ratio. The
curve seems to be shifted towards higher levels of suctions for denser states. This pattern is
consistent with the previous observations showing a trend for increase in degree of
saturation when the pores (void space) get smaller.
2.3.5 Air and water flow

In this part is discussed the generalisation of Darcy’s law (1856) to unsaturated soils.
Darcy’s law describes initially the flow of water within soils that are fully saturated in water,
for a laminar flow regime:
Kw
qw = ( grad( pw ) + ρ w .g.grad( y ) ) (2.17)
ρ w .g
where q w is the vector of water velocity with respect to the solid phase, K w the
anisotropic tensor of water permeability, pw the pore water pressures, y the vertical
coordinate taken positive upwards. The components of the tensor of water permeability are
called coefficients of permeability k wsat . It has been shown that Eq. 2.17 can also apply for the
flow of water in unsaturated soils (Richard 1931). Darcy’s law also holds for the flow of air
within the porous medium:
Ka
qa = ( grad( pa ) ) (2.18)
ρ a .g
q a , K a and pa are respectively the air velocity, the air permeability and the pore air
pressure. Given the multiple levels of coupled physical processes in soils, there is no a priori
1.2
e =0.96
0
1 e =0.72
0
e =0.60
r
0.8 e =0.47
0
e =0.36
0
0.6
0.4
0.2
0 5
0.1 10 1000 10
Figure2.27 Effect of initial void ratio on the shape of the soil water retention curve of a sandy clayey
silt, drying paths only (From Salager 2007)
36 CHAPTER 2
reason for the permeability coefficients defined in Eqs. 2.17 and 2.18 to be constants. These
permeability coefficients can for instance be compared with their counterparts in the
conventional case of full saturation in water. According to Lambe and Whitman (1979) the
saturated permeability coefficient k wsat is a function of the void ratio exclusively. In the more
general case of unsaturated soils, the water permeability coefficient k w is not only depending
on suction, but also on one of the experimental variables expressing the quantity of soil
water, that is the gravimetric water content w or the degree of saturation S r . Indeed water
can only flow through the part of the pore space that is filled with water. If the applied
suction is high enough to overcome the air entry value, gaseous air can enter the pore space
and reduce this partial volume of pore water. According to the conceptual sketch of Fig. 2.28,
the air-water interface will recede and get closer to the soil particles. Such a phenomenon
will reduce rapidly the capacity of water to flow through the medium, considering that (i)
the largest pores are drained first, (ii) the tortuosity of the flow path increases (iii) the
particles of water have to flow at the vicinity of the particles from which they might be
attracted (see Fig. 2.7). It is thus expected that the water permeability coefficient will decrease
as the degree of saturation lowers. This was indeed observed by Brooks and Corey (1964)
and Seker (1983) among others (Fig. 2.29).
The coefficients of relative permeability for each fluid are defined as the ratio between the
permeability under conditions of partial saturation and the permeability in fully saturated
state:
kα
krα = (2.19)
kαsat
where the subscript α designates the phase (liquid water or gaseous air). The evolution
of relative permeability coefficients with respect to the degree of saturation in water shows
indeed trends that are consistent with the physical insight above.
An important number of well-accepted formulations of the relationship between the
coefficients of permeability and degree of saturation are available. Some authors also express
the permeability in terms of matric suction (Brooks and Corey 1964, Gardner 1958). These
relationships have proved their efficiency, so it will be chosen not to develop further
modelling of the permeability coefficients (see chapter 6) in the present research.
Figure 2.28 Development of an unsaturated soils by the withdrawal of the air-water interface at
different stages of matric suction or degree of saturation (i.e. stages 1-5) (from Childs 1969, reported by
Fredlund and Rahardjo, 1993).
Figure 2.29 Evolution of relative permeability of water and air versus degree of saturation. Data from
Seker (1983), reported by Rifa’I (2002).
2.3.6 Conclusions: experimental state of the art

This part has provided an overview of the experimental features of behaviour of
unsaturated soils, reviewing the conventional features of stress strain and retention
behaviours. New insights into the coupled effects have been drawn from the state of the art
literature. All the reviewed aspects enter a list of specifications to be addressed by any
constitutive model. The possible simplification assumptions have been introduced. The
experimental techniques have been also overviewed in order to keep an optimal balance
between the upcoming theoretical developments and the technical restrictions of parameter
determination on the basis of laboratory tests.
All the investigated behaviours did not feature any temperature change that would be
driven by intrinsic transformation. It will be thus a fundamental assumption for the rest of
the PhD thesis to consider that all the processes are perfectly isothermal. Authors such as
François and Laloui (2008) address thoroughly the issue of non isothermal conditions in
soils. Furthermore, it will be assumed that no internal chemical transformation is occurring
during the studied processes. Temperature and chemics could be considered as
supplementary environmental loads inducing specific responses that are out of the scope of
the present research work.
2.4 State of the art: constitutive models

Unsaturated constitutive models constitute a necessary basis for the comprehension of
experimental data: a model, may it be right or wrong, is required to draw a guideline to
follow. The models enable to check that the experimental programs will give the answers,
and remain an essential predictive tool for the setting up of new tests.
It is proposed in this part to question in which way the various features of mechanical and
retention behaviour have been addressed from the perspective of existing constitutive
models. Provided that the first models for unsaturated soils have been formulated almost at
the time when first experimental tests were designed, it would be a major challenge to
redraw the historical evolution of constitutive modelling for unsaturated soils. Wheeler and
38 CHAPTER 2
Karube (1995) and Jardine et al. (2004) published extensive reviews for this purpose. The
present review is the achievement of a different approach that consists in addressing
successively each of the points of the list of specifications. For each feature of behaviour
characterised experimentally in part 2.3, only the specific elements of constitutive models
from literature are assessed in their capability to reproduce the physical response in
question.
It is again a deliberate choice to post-pone the analysis of stress variables for unsaturated
soils to next chapter 3. However, many advantages and limitations of the reviewed
constitutive models are linked to the choice of stress variables. Whenever needed in this
section, the stress and strain variables for the mechanical behaviour and for the retention
behaviour will be explicated more in detail. Since the main purpose of this part is yet to
identify only the most suitable “local” formulations for each separate behaviour, the adopted
reviewing procedure is appropriate. In absence of any specific information, the stress tensor
σ * will be defined as the “constitutive stress” entering the constitutive equation. The
constitutive law is a functional relationship between extensive and intensive variables
describing the state of the material. In general, all the variables describing the whole system
may appear in the functional relationship describing any single constituent. The constitutive
law is written under incremental form:
σ* = Dep ( ε, σ*, dσ*, ξ ) dε (2.20)
where Dep is the elasto-plastic matrix (material properties), that is in the most general case
a function of the stress and strain state, the direction of the stress increment dσ * , and
internal variables ξ .
2.4.1 Modelling the skeleton volumetric response

All the models reviewed hereafter belong to the family of elasto-plastic stress-strain
models. Indeed, the incremental skeleton total deformation d ε is basically decomposed into
an elastic (or reversible) part d ε e and a plastic (or irrecoverable) part d ε p (Yu 2006):
dε = dε e + dε p (2.21)
The decomposition Eq. 2.21 is essential to reproduce the physical elasto-plastic behaviour
of the soil and account for the irreversible strains.
The models of the fully reversible or elastic type are thus voluntarily not featured in the
point-by-point comparison of constitutive models of the present section. Elastic models have
a long history in the mechanics of unsaturated soils. They are issued from the idea that for
virgin loading conditions and given initial stress state, the void ratio increment is uniquely
related to increments in experimental stress variables (for instance net stress and suction,
that are independent stresses).
The elastic models for unsaturated soils are non-linear and such that stiffness terms
depend on the current stress or strain rate. While Coleman (1962) followed by Bishop and
Blight (1963) first introduced models for both volumetric and water content changes with
respect to experimental stress variables, Matyas and Radhakrishna (1968) proposed the
concept of “state surfaces” (Fig. 2.30). The principle is to plot the experimental void ratio e
and degree of saturation S r with respect to net stress and matric suction, in order to facilitate
the introduction of the non-linear aspects of the behaviour of unsaturated soils. Similar
modelling approaches are used for instance by Fredlund (1979), Hasan and Fredlund
(1980),Lloret and Alonso (1985), Gatmiri and Delage (1995), Ali et al. (1995). Such models can
Figure 2.30 State surfaces for void ratio and degree of saturation, from Matyas and Radhakrishna
(1968)
obviously not sustain comparison with elasto-plastic models on non-monotonic stress paths
or on the issue of capillary hysteresis. Fig. 2.30 is however relevant to appear here in
anticipation of the review of models for the soil water retention behaviour.
Building a constitutive model for unsaturated soils.

The principle for building a constitutive model for unsaturated soils is, to a certain extent,
common to all the state-of-the-art constitutive models. The principle was first introduced
implicitly by Alonso et al. (1990) for the formulation of the model now called Barcelona Basic
Model (BBM), and was later formalized by Jommi (2000), followed by Pereira (2007), among
others. The various steps of the procedure can be summarized as follows:
(i) Choice of a reference stress-strain model. The development of a constitutive model
for unsaturated soils always relies on an elasto-plastic model formulated for
saturated soils. The model for unsaturated states inherits the specific capability of
the reference model in terms of stress-strain modelling (for instance bounding
plasticity or multiple mechanisms of plasticity).
(ii) Choice of a stress-strain framework. The generalisation of the reference model to
partial saturation requires the adoption of experimental stress variables or
combination of them to define the “constitutive stress”. An exhaustive stress-strain
framework should also provide the stress and strain variables for the retention
behaviour.
(iii) Choice of a model for the retention behaviour. This model complements the
mechanical model by accounting for the quantity of pore water.
(iv) Activation of supplementary “cross-effects”. The cross-effects are the effect of the
mechanical variables on the retention model and the effects of the capillary
variables on the mechanical model. These effects include for instance the suction-
induced stiffening effect.
Point (ii) will be addressed in chapter 3. In the earliest models (e.g. Alonso et al. 1990,
Bolzon et al. 1996), the point (iii) was not yet addressed, which means that the published
constitutive frameworks could only cope with skeleton volume change. The upcoming
comparative analysis focuses mainly on point (iv), where the added value of each specific
formulation is the most obvious.
40 CHAPTER 2
Preconsolidation stress and wetting collapse

It is now widely accepted that the way to account for the wetting pore-collapse in
unsaturated soils is to make use in the constitutive model of a yield curve called “Loading
Collapse curve”. This terminology is due to Alonso et al. (1990). The loading collapse (LC)
yield curve indeed addresses both the need for reproducing the plastic compression upon
wetting under a high mechanical load, and the evidence of increase in preconsolidation
stress with suction. In the BBM, the LC curve is fitted in plane suction versus mean net
stress ( s − pnet ) using yield points determined along mechanical loading paths carried out at
constant suction levels (See Fig. 2.13). As the preconsolidation pressure pc′ represents the
limit of the elastic domain, the LC curve expresses to which extent the suction stiffens the
material. Fig. 2.31 recalls the principle for modelling the wetting pore collapse, which is no
less than a plastic compression due to yielding on the LC curve. A necessary condition to
ensure that the path will encounter the plastic limit is that the initial mechanical stress state
(mean net stress pnet is high enough).
The material function introduced in BBM for the evolution of preconsolidation pressure
pc′ is:
λ0 −κ
⎛ p′ ⎞ λs −κ
pc′ = pr ⎜ 0 ⎟ (2.22)
⎝ pr ⎠
with pr a reference mean stress, p0′ the saturated preconsolidation stress

(preconsolidation stress of Cam-Clay model), λ0 , λs the negative slopes of NCL for saturated
and unsaturated states respectively and κ is the slope of the unloading-reloading line.
Most of the “advanced” constitutive models for unsaturated soils feature a Loading
Collapse curve or an equivalent for it (Geiser 1999, Loret and Khalili 2002, Wheeler et al.
2003, Gallipoli et al. 2003b, Sheng et al. 2004, Tamagnini 2004). However, due to the different
constitutive stress variables chosen the shape of the LC-like yield limit can vary significantly
from one model to another. Fig. 2.32 gives an illustration of four different LC curves taken
from literature. In all plots of Fig 2.32, the axis can be considered to be the mechanical
“constitutive stress” while the ordinate is the stress variable used to express the capillarity.
Figure 2.31 Principle for modelling wetting pore-collapse: yielding on the LC curve, from Alonso et al.
(1990).
(a) (b)
Capillary stress
Suction
Mechanical stress
Mechanical stress
(c) (d)
Capillary stress
Mechanical stress
Mechanical stress
Figure 2.32 Possible shapes of the Loading Collapse curve. (a) Sheng et al. (2004) (b) Gallipoli et al.
2003 (c) Tamagnini (2004) (d) Wheeler et al. (2003)
In the models of Gallipoli et al. (2003) and Wheeler et al. (2003), the LC curves are
simplified to simple lines. The price to pay for such simplification in plane retention stress
versus mechanical stress is to make use of:
(i) complex stress variables (part of the suction effects are introduced in the stress
variables)
(ii) Advanced coupled framework with other limits of plasticity (Fig. 2.32d) .
On the contrary, in Fig. 2.32a (Sheng et al. 2004), the formulation of the LC curve is
apparently highly non linear, with a shape more similar to that of the original BBM.
Interestingly, the use of a particular effective stress (defined later as the generalised effective
stress σ ′ ) enables to model the domain where the air entry suction is not reached. In this
zone, the model predicts no change for the preconsolidation pressure with suction, which is
consistent with the experimental evidence of Fig. 2.13. A similar shape was adopted by Loret
and Khalili (2002). In fig. 2.32c, the LC curve is identical to that of the BBM, except that it is
doubled (Tamagnini 2004). This double formulation comes from the expression of pc′ that is
42 CHAPTER 2
a function of the degree of saturation. In the presented case, the SWRC features an hysteresis
that is reflected on the LC curve, expressed in a plane where the vertical axis is matric
suction. The reviewed models show very satisfactory results in modelling the wetting-pore
collapse.
Drying-wetting cycles
The performances of the various formulations for the loading collapse yield curves are
assessed mainly along drying and wetting paths, as well as on successive drying and wetting
cycles. It can be demonstrated that the successive phases of elasticity and plasticity observed
during suction variations under constant net stress (e.g. Fig.2.21) are only due to the stress
point remaining inside the elastic domain or yielding upon a yield limit. Two different
approaches are adopted for this purpose in constitutive modelling, mainly due to the choice
of stress variables.
In a first category of models (e.g. Alonso et al. 1990), the stress path in plane: capillary
stress versus mechanical stress is simply linear, and even vertical in the case of drying and
wetting. The only way to generate plastic strain upon suction increase is to add a
supplementary yield limit. By reference to the BBM, this supplementary limit is often called
“Suction Increase” (SI) yield limit. In some cases a third yield limit is necessary for the
wetting paths. This limit is often called “Suction Decrease” (SD), see for instance Fig. 2.32a.
In the second category of models (e.g. Loret and Khalili 2002), the stress variables are
combination of the experimental variables, including also possibly material parameters. The
stress path in plane: capillary stress versus mechanical stress is now highly non linear, and
there is a direct possibility to yield the existing yield limit upon drying or wetting.
By reference to the experimental evidence of Fig. 2.18, some volumetric change can arise
from suction changes even below the air entry value. Accounting for those changes in void
ratio at low suctions is essential in a constitutive framework, provided that there is a
generation of plastic strain corresponding to a large variation of volume. The required
elements to be integrated for this purpose in a model are (i) the existence of a suction of
preconsolidation, (ii) elastic and plastic slopes that are close to the mechanical ones and (ii) a
fully elastic response when suction overcomes the air entry value (iii) a modified elastic
slope in plane ( e − ln s ) for this domain.
It has been found out that only the constitutive models using a “generalised effective
stress” or “unsaturated effective stress” σ ′ as the constitutive mechanical variable (Eq. 2.20)
could reproduce this particular behaviour below the air entry value. An extensive review of
the possible formulations for effective stresses in partially saturated soils is proposed in
chapter 3. The generic form of the unsaturated effective stresses is written:
σ ′ = σ net + μ 2 ( s, Sr ) (2.24)
where μ2 is a function of suction and/or degree of saturation. Most of the effective

stresses variables used in literature (e.g. Schrefler 1983, Jommi 2000, Loret and Khalili 2002,
Gallipoli et al. 2003b) are such that for suctions below the air entry value se (that is for a
degree of saturation equal to 1), the generic form Eq. 2.24 becomes:
σ ′ = σ − pw (2.25)
where σ is the total stress and pw the pore water pressure. Eq. 2.21 is easily identified as
the conventional effective stress of Terzaghi (1936) for saturated soils. The advantage of the
models based on effective stresses is thus to feature a natural transition between the states of
complete and partial saturation. Loret and Khalili (2002) followed by Sheng et al. (2004)
proposed to introduce a particular shape for the Loading Collapse curve, expressing the
evolution of the preconsolidation pressure pc′ as follows:
⎧⎪ pc′ 0 if s < se
pc′ = ⎨ (2.26)
⎪⎩ pc′ 0 × γ ( s, Sr ) or pc′ 0 + γ ( s, Sr ) else
where pc′ 0 is the saturated preconsolidation pressure and γ ( s, S r ) is a function of suction

and/or degree of saturation. The typical shape obtained with such a formulation is that of
Fig. 2.32a seen previously. The analysis of Eqs. 2.24 and 2.26 together show that in the case
where suction is lower than the air entry value, the preconsolidation pressure is set equal to
the saturated reference one while the effective stress become the conventional effective stress
of Terzaghi. The mechanical stress-strain analysis becomes equivalent to a purely mechanical
saturated case. The typical response to suction changes in this domain is plotted in Fig. 2.33,
taken from Loret and Khalili (2002). In the saturated zone ( s < se ) behaviour is first elastic
due to initial preconsolidation (path AB) and then follows the normal compression line (path
BC). The elastic and elasto-plastic slopes are that of the saturated reference model.
Desaturation begins at point C: from thereon, the capillary effects make the preconsolidation
pressure pc increase faster than the effective mean stress resulting in elastic behaviour. The
response to wetting is fully reversible.
No model using the experimental variables (net stress and suction) could be found to
account for this domain of full saturation at non-zero suction. This is due to the fact that the
net stress alone fails to describe the saturated state of stress. In consequence, it is not possible
to draw a natural transition in the models between the partial saturation and the
conventional saturated case (see chapter 3).
The capability of constitutive models to reproduce the alternate elastic and elasto-plastic
phases due to successive drying and wetting cycles (Figs. 2.20 and 2.21) has also to be
assessed with respect to the shape and evolution of the yield loci. The up-to-date constitutive
models can be sorted in three main categories:
(a) (b)
Figure 2.33 Simulation of a drying wetting cycle under zero net stress, from Loret and Khalili (2002)
44 CHAPTER 2
1) Models where the constitutive stress variables (mechanical and capillary) include part
of the capillary effects. The stress path is complex, but the yield locus is unique (e.g.
Gallipoli et al. 2003, Tamagnini 2004)
2) Models where the stress variables are simpler (for instance net stress and suction), but
for which a single yield locus of the LC type is insufficient. These models often feature
a Suction Increase (SI) and Suction Decrease (SD) yield limit (e.g. Alonso et al. 1990).
3) Models where the stress variables are not the experimental ones, yet requiring
additional yield limits of the SI and SD type (e.g. Wheeler et al. 2003, Sheng et al.
2004), see Fig. 2.32.
Concerning the second and third categories defined above, the expression of couplings
between SI, SD and LC are too specific to each model to be mentioned here in a generic form.
It can be understood however that a variation in suction is likely to generate plastic
compression due to yielding on SI curve on the drying side and SD curve on the wetting
side. While the models show good performances on successive suction cycles, due to the
multiple yield limits involved for this categories of models, the initialisation and calibration
of processes are generally very complex.
Fig. 2.34 shows an example of model response for the category 1. The model predicts
irreversible changes of void ratio during both the wetting and the drying branch of the test,
using only one LC-type yield locus. Yielding occurs initially during dying and the
development of elasto-plastic strains produces the first expansion of the yield locus. Further
Elasto-plastic deformation is also observed during the second drying phase. The use of a
single yield locus is a consequence of the stress path being highly non linear in the plane
(retention stress versus mechanical stress).
In conclusion, all three categories of model show excellent comparability with
experimental evidence reviewed previously. The stiffening effects of suction have to be
either accounted for in the constitutive stress variables, or by the means of sophisticated
yield limits, or else in both the stress and the plastic limits. The reviewed solutions to
reproduce the successive wetting-drying cycles from the viewpoint of volumetric response
often feature a large number of parameters for the calibration of the multiple yield loci,
(a) (b)
Figure 2.34 Prediction of elastic swelling and plastic compression due to successive wetting cycles
(from Gallipoli et al. 2003). p′′ is the mechanical constitutive stress, ξ is the capillary stress.
which might by a true limitation for the engineering applicability of the models and the use
for boundary value problems. The solution of a simplified unique yield limit should thus
receive more attention in the upcoming research.
Coefficients of compressibility
The dependency of the coefficient of plastic compressibility with the state of saturation
has been shown previously in Figs 2.11 and 2.12. Like for the preconsolidation pressure, the
dependency of the compressibility coefficient on capillarity could be in theory expressed as
either a function of matric suction s , degree of saturation S r or a combination of them. In
practice, Alonso et al. (1990) proposed the following law:
λs = λ0 ((1 − r ) exp(− β s) + r ) (2.27)
with λ0 being the slope of the NCL for saturated states. r and β are material
parameters. The dependency of the elasto-plastic slope is here expressed as a function of
suction only. This is also the case for the models reviewed up to now. If some authors do not
provide an explicit dependency law for λs (for instance Sheng et al. 2004), the majority of
published work report very simple relationships as an alternative to Eq. 2.23, with only one
to two parameters (Geiser 1999, Jommi and Di Prisco 1994, Loret and Khalili 2002, Gallipoli
et al. 2003b).
While the formulation of the BBM (Eq. 2.27) predicts a decreasing slope λs with suction,
most models on the contrary model an increased compressibility for dryer material, see Fig.
2.35. This divergence cannot truly be considered as inconsistent provided that the
experimental evidence showed different evolution trends depending on the studied material.
Furthermore, the constitutive stress for the mechanical stress-strain model being each time
different, it influences also the global stress-strain plot and might justify opposite slope
changes from one model to the other. The calibration of elasto-plastic slopes will be
investigated further in chapter 3.
(a) (b)
Void ratio
Figure 2.35 Examples of slope evolution for normal compression lines with suction (a) From Loret and
Khalili (2002) (b) from Gallipoli et al. (2003).
46 CHAPTER 2
As a complementary remark, none of the reviewed models features a direct dependency

of the parameters of elastic rigidity with suction. It seems thus to be a reasonable assumption
to use the elastic slope of the reference saturated case also for simulations in partial
saturation.
2.4.2 Modelling the soil water retention curve

Looking again to the experimental state variables defined in part 2.3.2, it is obvious that
the complete state of the unsaturated material requires not only to be described from the
mechanical point of view, but also from the quantity and interstitial pressure of pore fluids.
The Soil Water Retention Curve defined in part 2.3.4 has been drawn experimentally and
needs now to be conceptualized within the complete constitutive modelling context.
While in the very first constitutive modelling approaches of the non-linear fully elastic
type, the relationship between degree and saturation, suction and void ratio was already
conceptualized (see Fig.2.31b), the soil water retention curve has been later totally decoupled
from the constitutive models for unsaturated soils (e.g. Alonso et al. 1990, Bolzon 1995). This
was a mistake to neglect the soil water retention behaviour and focus only on the skeleton
volume changes. Firstly, the state of the material would not be described completely.
Secondly, the defined stress-strain frameworks were not consistent from the viewpoint of
work input (Houlsby 1997), which states that each stress variable introduced within the
constitutive framework must be work conjugate with the increment of a corresponding strain
variable. In other words, if a mechanical constitutive stress variable and a capillary stress
variable are introduced, they should be related respectively to a mechanical strain variable
and a capillary strain variable. The possible forms of those variables belonging to a complete
stress-strain framework are not discussed hereafter as they will be the object of chapter 3.
Nonetheless, it is mentioned that a possible capillary stress variable is matric suction s and
the corresponding capillary strain variable is degree of saturation S r . Most up-to-date
constitutive models for unsaturated soils now feature a description of the ( Sr − s )
relationship, might it be coupled with the stress-strain mechanical model or not.
While the term of Soil Water Retention Curve is widely accepted, it has the connotation of
a univocal relationship between S r and s , which in some cases does not meet the
experimental evidence (Fig. 2.23). It is thus preferable to name the models reviewed hereafter
as models for the soil water retention behaviour rather than models of the soil water retention curve.
Reversible models
Many analytical functions have been historically given for the soil water retention
relationship. The initial formulations (e.g. Brooks and Corey 1964, Fredlund 1997) belong to
the family of reversible non linear models the sense that they constitute a simple fitting of
experimental data. Some discontinuous gradient functions are sometimes used although
continuous slopes are preferable. All the functions fix a porous medium and tend to describe
its WRC without taking into account the mechanical and hydraulic history. No
interpretation is given, except for the air entry value and the residual saturation, which are
simply derived by means of geometrical constructions. Sillers and Fredlund (2001) provide a
detailed discussion on the mathematical attributes of the different curves proposed. An
example of such a non-linear revesible curve often used is Van Genuchten’s function (1980):
1
Sr = m
(2.28)
⎡⎣1 + (α .s ) n ⎤⎦
where α , n and m are material parameters. Fig. 2.36 shows an example of fitting using
Van Genuchten formulation. These forms, although useful in practical applications, only
give a basic definition of water retention behaviour. The air entry value and the residual
saturation are seldom known, and still remain difficult to obtain by means of standard
procedures. It is well known that the water retention curve is hysteretic. Energy dissipation
occurs along a drying-wetting cycle, which supposes the introduction of an “elasto-plastic”
law, that is to say a law allowing for partial irreversibility. As it will be shown later, the
capillary hysteresis may play a role on the mechanical behaviour. On the other hand, the
retention curve evolves with the void ratio, which reflects an influence of the mechanical
actions. The subsequent evolution of the curve may be homothetic, but more frequently, only
the higher portion of the curve is modified by the macroscopic strains. In order to model
correctly the dependence on void ratio a coupled hardening law must be provided.
Models for capillary hysteresis

Considering that a given suction might correspond to several values of degree of
saturation, a number of formulations have been proposed for the hysteresis of the soil water
retention curve (e.g. Vaunat et al. 2000, Pham et al. 2003, Wheeler et al. 2003, Tamagnini 2004,
Sheng et al. 2004, Li 2005). The possible states of saturation are bounded by the main drying
curve and the main wetting curve, which are respectively the curve of highest possible
degrees of saturation along drying and curve of lowest possible degrees of saturation upon
wetting (Fig. 2.37). Some different denominations can be found for these two branches of the
hysteresis, and will be reviewed in chapter 6. Inside the domain limited by the main drying
and wetting curves, the behaviour is almost reversible. The reversible parts are called
scanning lines (Fig. 2.37).
The theory of elasto-plasticity is is used for the modelling of the reversible and
irrecoverable parts of the degree of saturation, associated with kinematic hardening. The
total variation in the degree of saturation dS r can be decomposed into an elastic part dS re
and a plastic part dS rp :
1.2
1
r
0.8
0.6
0.4
0.2 Exp.
Van Genuchten (1980)
0 5 7
1000 10 10
Figure 2.36 Calibration of the soil water retention curve of Van Genuchten. Experimental data: Sion
silt (from Geiser 1999).
48 CHAPTER 2
Figure 2.37 Idealised representation of advanced soil water retention models.
dSr = dSre + dSrp (2.29)
The analogy between Eqs. 2.30 and 2.21 is obvious. The use of a kinematic yield limit
enables to predict dissipation along both wetting and drying paths and thus to draw a closed
hysteresis loop in plane ( S r − s ) , Fig. 2.37. Two examples of conceptual frameworks for the
retention behaviour, taken from Sheng et al. (2004) and Wheeler et al. (2003) are reported in
Fig. 2.38. The constitutive equation of the capillary model for the elastic behaviour is written
under the form:
dSr
= −κ s* (2.30)
ds
where κ s* is an elastic coefficient which is distinct from the parameter κ s introduced

previously as the slope of the elastic line in plane ( e − ln s ) . The plastic increments,
corresponding to a point located on the main drying or main wetting curve, are written:
dsDry
(
dS rp = − λs* − κ s* )s (
= − λs* − κ s* ) dss
Wet
(2.31)
Dry Wet
Where λs* is the elasto-plastic slope in plane ( S r − ln s ) , sDry is the maximum suction ever
experienced by the material on drying and sWet the minimum suction ever experienced on
wetting ( sDry and sWet might be interpreted as analogous to the preconsolidation pressure in
a stress-strain model). In the case of Fig. 2.38b (Wheeler et al. 2003), the coefficient λs* is a
constant, while in Fig. 2.38a, λs* ( s ) is a suction dependent material function to specify.
Void ratio dependency

Apart from the out-of-date state surfaces for the degree of saturation (Fig. 2.30), only a
little number of recent models do account for the effect of void ratio on the shape and
parameters of the soil water retention model. The works of Wheeler et al. (2003) and
Gallipoli et al. (2003) are essentially summarized in this paragraph.
(a) (b)
Figure 2.38 Examples of models for the retention behaviour with capillary hysteresis (a) from Sheng et
al. (2004) (b) from Wheeler et al. (2003).
The water retention model proposed by Gallipoli et al. (2003) is somehow at the midway
between the state surface description and the advanced modelling contexts reviewed
previously. Starting from the water retention curve of Van Genuchten (Eq. 2.28), the authors
introduce a direct coupling with void ratio e by modifying the parameter α , which gives:
1
Sr = m
(2.32)
⎡⎣1 + (φ .eψ .s ) n ⎤⎦
where φ and ψ are soil constants. Consequently, depending on the initial void ratio e or
specific volume v ( = 1 + e ) , the overall shape of the soil water retention curve is different, as
shown in Fig. 2.39b. The model can predict the increase in degree of saturation due to a
mechanical compression (i.e. decreasing void ratio) at a constant suction. While the
dependency on void ratio is of particular interest here, the problems proper to the non linear
fully reversible fitting are inherited from the Van Genuchten’s equation. In other words, the
model fails to reproduce the capillary hysteresis.
In the formulation of their water retention model, Wheeler et al. (2003) also introduce a
direct dependency on void ratio. More exactly, the plastic volumetric strains have a
kinematic effect on the water retention model by affecting the position of the main drying
and main wetting curves, Fig. 2.39a. In the case where yielding occurs in the stress-strain
model, that is the LC moves forward, the limits of plasticity in the retention model (main
drying and wetting curves, or sDry and sWet in Eq. 2.31) that are fully coupled with LC are
shifted in consequence. According to this conceptual framework, if changes in void ratio
occur in the purely elastic domain only, no effect on the retention properties is entailed. This
assumption is not justified on the basis of experimental evidence.
The effect of activating a direct coupling between the void ratio and the soil water
retention model is not only to draw various retention curves depending on the initial state
(e.g. Fig. 2.39b) but also to predict changes in degree of saturation at constant suction under
the effect of pure mechanical load (see Fig. 2.26).
50 CHAPTER 2
(a) (b)
Figure 2.39 Examples of models including the effect of void ratio on the soil water retention model (a)
Wheeler et al. 2003 (b) Gallipoli et al. (2003).
It appears from the literature review that the only constitutive model for retention
behaviour including both the capillary hysteresis and the dependency on void ratio is that of
Wheeler et al. (2003). Yet this model is limited to the effect of plastic strains only. There is not
however any visible incompatibility to adapt the laws for void ratio dependency (e.g.
Gallipoli et al. 2003) to retention models featuring already an hysteresis, for instance that of
Pham (2003). In this respect, a more detailed insight into the dependency laws between the
parameters of soil water retention models and void ratio will be presented in chapter 5.
2.5 Conclusions
This chapter has been dedicated to recalling the definitions of interest for the PhD. thesis,
and reviewing the experimental and constitutive modelling state of the art for unsaturated
soils. The objectives were (i) to assess the needs for experimental or modelling tasks within
this research work and (ii) to set up a list of specifications to be addressed with respect to
mechanics of unsaturated soils.
It can be concluded that the experimental knowledge provides now a detailed insight into
the features of behaviour of unsaturated soils. Further investigation is still required for
complex combination and cycles of stresses (mechanical and capillary). In particular, the soil
water retention behaviour under different mechanical stresses and initial void ratio
particularly needs to be addressed experimentally, as the laboratory evidence of this type is
only relatively recent and still discussed. This observation will motivate the experimental
tasks presented in chapter 5.
The presented PhD work is however essentially focusing on constitutive and numerical
modelling of unsaturated soils. It is deduced from the present chapter that even if the
features of behaviour are now relatively well quantified in the laboratory (or in-situ), only a
little number of constitutive models can apprehend the whole set of coupled aspects in the
mechanical and capillary parts. The next chapters will present our contribution to the
advanced constitutive modelling of unsaturated soils, in answer to the various points of the
specifications list, and getting inspiration from the best practice in existing formulations.
In essence, the list of specifications gathers the behaviours to be addressed by an

exhaustive constitutive model. Those features of behaviour can be synthesized in the
following list:
(1) Reproduce the behaviour in fully saturated conditions (elasto-plasticity, critical state,
isotropic and deviatoric behaviours)
(2) Increase in preconsolidation pressure with suction (isotropic mechanical model)
(3) Change of the plastic compressibility coefficients with suction (isotropic mechanical
model)
(4) Increase of apparent shear strength with suction (deviatoric mechanical model)
(5) Reproducing swelling and pore-collapse upon wetting (isotropic mechanical model)
(6) Reproducing plastic hardening (compression) along drying and wetting paths
(isotropic mechanical model)
(7) Recover the mechanical stress-strain behaviour in saturated conditions, including in
the zone of saturation with non-zero suction (isotropic mechanical model)
(8) Define the same zone of saturation at non-zero suction in the water retention model
(9) Reproduce dissipations in the retention curve along successive drying wetting cycles
(water retention model)
(10) Comprehend the dependency of soil water retention on void ratio (water retention
model)
A new constitutive model will be formulated in chapters 4 and 5 to answer those 10
points, but a preliminary thorough investigation on the stress variables for unsaturated soils
is required. This last issue is the theme of next chapter 3.
2.6 References
Aitchison G. D. (1965). Moisture equilibria and moisture changes in soils beneath covered areas, a symposium
in print, Butterworths.
Al-Mukhtar M. (1995). Macroscopic behavior and microstructural properties of a kaolinite clay under
controlled mechanical and hydraulic state. Unsaturated Soils: Proc. 1st Int. Conf. on
Unsaturated Soils / UNSAT 95. E. E. Alonso and P. Delage. Paris. 1: 3-9.
Ali S. R., Pyrah I. C., Anderson W. F. (1995). "Constitutive modelling of unsaturated soils using
hyperelasticity." Unsaturated soils, Sols non saturés, Paris, Balkema, pp. 681-686.
Alonso E. E., Gens A., Hight D. W. (1987). "Special problem soils. General report." 9th European
conference on soil mechanics and foundation engineering, Dublin, pp. 1087-1146.
Alonso E. E., Gens A., Josa A. (1990). "A Constitutive Model for Partially Saturated Soils." Geotechnique
40(3), pp. 405-430.
Alonso E. E., Lloret A., Gens A. (1995). Experimental behavior of highly expansive double-structure
clay. Unsaturated Soils: Proc. 1st Int. Conf. on Unsaturated Soils, UNSAT 95. E. E. Alonso and
P. Delage. Paris. 1: 11-16.
Bishop A., Blight G. (1963). "Some aspects of effective stress in saturated and unsaturated soils."
Geotechnique 13, pp. 177-197.
Bishop A., Donald I. (1961). "The experimental study of partly saturated soil in the triaxial apparatus."
5th Int. Conf. on Soil Mechanics and Fundation Engineering, Paris, pp. 13-21.
52 CHAPTER 2
Bishop A. W., Alpan I., Blight G. E., Donald I. B. (1960). "Factors controlling the strength of partly
saturated cohesive soils." Conference Shear Strength Cohesive Soils, American Society of Civil
Engineers, pp. 503-532.
Bolton M. D. (1986). "The strength and dilantancy of sands." Geotechnique 36(1), pp. 65-78.
Bolzon G., Schrefler B. A., Zienkiewicz O. C. (1996). "Elastoplastic soil constitutive laws generalized to
partially saturated states." Geotechnique 46(2), pp. 279-289.
Boso M., Tarantino A., Mongiovi L. (2005). Shear strength behaviour of a reconstituted clayey silt.In:
Unsaturated soils, Advances in testing, modelling and engineering applications. C. Mancuso and A.
Tarantino: 1-14.
Brooks R. H., Corey A. J. (1964). "Hydraulic properties of porous media." Hydrology Papers(3).
Buisson M. S. R., Wheeler S. J. (2000). "Inclusion of hydraulic hysteresis in a new elasto-plastic
framework for unsaturated soils." In: Experimental evidence and theoretical approaches in
unsaturated soils, Trento, Balkema, pp. 109-119.
Childs E. C. (1969). An introduction to the physical basis of soil water phenomena. London, Wiley.
Chiu C. F. N., C.W.W. (2003). "A state-dependent elasto-plastic model for saturated and unsaturated
soils." Géotechnique 53(9), pp. 809-829.
Coleman J. D. (1962). "Stress/strain relations for partly saturated soils." Géotechnique
12(Correspondence), pp. 348-350.
Collin F. (2003). Couplages thermo-hydro-mécaniques dans les sols et les roches tendres partiellement
saturés. Liège, Université de liège. PhD. Thesis.
Cui Y. J., Delage P. (1996). "Yielding and plastic behaviour of an unsaturated compacted silt."
Geotechnique 46(2), pp. 291-311.
Darcy H. (1856). Histoire des fontaines publiques de Dijon. Paris, Dalmont: pp.590-594.
Davies J. T., Rideal E. K. (1963). Interfacial phenomena (2nd edition). New York.
Engel J., Schanz T., Lauer C. (2003). "State parameters for unsaturated soils, basic empirical concepts."
In: Unsaturated Soils: Numerical and Theoretical Approaches, Weimar, pp. 125-138.
Fawcett R. G., Collis-George N. (1967). "A filter paper method for determining the moisture
characteristics of soil." Australian Journal for Experimental agriculture and animal husbandry 7, pp.
162-167.
Fleureau J. M., Kheirbeksaoud S., Soemitro R., Taibi S. (1993). "Behavior of Clayey Soils on Drying
Wetting Paths." Canadian Geotechnical Journal 30(2), pp. 287-296.
François, B. Laloui, L. (2008). “ACMEG-TS: A constitutive model for unsaturated soils under non-
isothermal conditions”. International Journal for Numerical and Analytical Methods in
Geomechanics, vol 32. , pp. 1955-1988.
Fredlund D. G. (1979). "Second Canadian Geotechnical colloquium: Appropriate concepts and
technology for unsaturated soils." Canadian Geotechnical Journal 16(1), pp. 121-139.
Fredlund D. G., Morgenstern N. R. (1977). "Stress state variables for unsaturated soils." Journal of the
geotechnical engineering division, ASCE 103(GT5), pp. 447-466.
Fredlund D. G., Morgenstern N. R., Widger R. A. (1978). "The Shear Strength of Unsaturated Soils."
Canadian Geotechnical Journal 15, pp. 313-321.
Fredlund M. D., Fredlund D. G., Wilson G. W. (1997). "Prediction of the Soil-Water Characteristic
Curve from Grain-Size Distribution and Volume-Mass Properties." 3rd Brazilian Symposium
on Unsaturated Soils, Rio de Janeiro, Brazil.
Gallipoli D., Gens A., Sharma R., Vaunat J. (2003). "An elasto-plastic model for unsaturated soil
incorporating the effects of suction and degree of saturation on mechanical behaviour."
Gallipoli D., Wheeler S. J., Karstunen M. (2003). "Modelling the variation of degree of saturation in a
deformable unsaturated soil." Geotechnique 53(1), pp. 105-112.
Gardner W. R. (1958). "Some stready state solutions of the unsaturated moisture flow equation with
application to evaporation from a water table." Journal of American soil science society.
Gatmiri B., Delage P. (1995). "A new void ratio state surface formulation for the nonlinear elastic
constitutive modeling of unsaturated soil-Code UDAM." 1st International conference on
unsaturated soils, Paris, pp. 1049-1056.
Geiser F. (1999). Comportement mécanique d'un limon non saturé: étude expérimentale et
modélisation constitutive. Lausanne, EPFL. PhD. thesis.
Geiser F., Laloui L., Vulliet L. (1999). Unsaturated Soil Modelling with Special Emphasis on Undrained
Condition. Proc. 7th Int. Symp. On Numerical Models in Geomechanics, NUMOG. Graz,
Austria: 9-14.
Hasan J. U., Fredlund D. G. (1980). "Pore pressure parameters for unsaturated soils." Canadian
Geotechnical Journal 17(3), pp. 395-404.
Houlsby G. T. (1997). "The work input to an unsaturated granular material." Géotechnique 47(1), pp.
193-196.
Jardine R. J., Gens A., Hight D. W., Coop M. R. (2004). "Developments in understanding soil
behaviour." Advances in Geotechnical engineering. The Skempton Conference, Thomas
Telford, pp. 103-206.
Jennings J. E. B., Burland J. B. (1962). "Limitations to the use of effective stresses in partly saturated
soils." Géotechnique 12, pp. 125-144.
Jommi C. (2000). "Remarks on the constitutive modelling of unsaturated soils." Experimental
Evidence and Theoretical Approaches in Unsaturated Soils; Proc. of an International
Workshop, Trento, pp. 139-153.
Jommi C., Di Prisco C. (1994). "Un semplice approcio teorico per la modellazione del comportamento
meccanico di terreni granulari parcialmente saturi." Conf. Il ruolo dei fluidi nei problemi di
ingegneria geotecnica, Mondovi, pp. 167-188.
Josa A. (1988). Un modelo elastoplastico para suelos no saturados. Barcelona, Universitat Politecnica
de Catalunya. PhD. Thesis.
Kane H. (1973). "Confined compression of loess." 7 th ICSMFE, Moscow, pp. 155-122.
Kim J. K., Hemphill G. B., Stewart W., Tinkler J. (2006). "Methods of re-mining tunnel T08 at THSRC
contract C230 after a collapse." Tunnelling and Underground Space Technology 21(3-4).
Kohgo Y., Nakano M., Miyazaki T. (1993). "Theoretical aspects of constitutive modelling for
unsaturated soils." Soils and Foundations 33(4), pp. 49-63.
Koliji A., Laloui L., Cuisinier O., Vulliet L. (2006). "Suction induced effects on the fabric of a structured
soil." Transport in porous media 64, pp. 261-278.
Lambe W., Whitman R. (1969). Soil Mechanics. New-York.
Leclercq J., Verbrugge J. C. (1985). "Propriété sgéomécaniques des sols non saturés." Colloque sur le
travail des sols, Faculté des sciences agronomiques Gembloux, pp.
Li X., Zhang L. (2007). Prediction of SWCC for coarse soils considering pore size changes. In:
Experimental unsaturated soil mechanics. T. Schanz, Springer.
Li X. S. (2005). "Modelling of hysteresis response for arbitrary wetting/drying paths." Computers and
geotechnics 32, pp. 133-137.
Loret B., Khalili N. (2002). "An effective stress elastic-plastic model for unsaturated porous media."
Mechanics of Materials 34(2), pp. 97-116.
Maâtouk A., Leroueil S., La Rochelle P. (1995). "Yielding and critical state of collapsible unsaturated
silty soil." Géotechnique 45(3), pp. 465-477.
Matyas E. L., Radhakrishna H. S. (1968). "Volume change Characteristics of Partialy Saturated Soils."
Géotechnique 18, pp. 432-448.
54 CHAPTER 2
Mitchell J. K., Soga K. (2005). Fundamentals of soil behavior, third edition, Wiley.
Muraleetharan K. K., Wei C. (2004). A unified framework for elastoplasticity of unsaturated soils:
From capillary hysteresis to soil skeletal deformation. In: Unsaturated soils: Numerical and
Theoretical approaches. T. Schanz. Weimar, Springer. 94: 111-124.
Pereira J.-M., Wong H., Dubujet P., Dangla P. (2005). "Adaptation of existing behaviour models to
unsaturated states: Application to CJS model." International journal for numerical and analytical
methods in Geomechanics 29(11), pp. 1127-1155.
Péron H. (2008). Desiccation cracking of soils. Lausanne, EPFL. PhD. Thesis.
Pham H. Q., Fredlund D. G., Barbour S. L. (2003). "A practical hysteresis model for the soil-water
characteristic curve for the soils with negligible volume change." Geotechnique 53(2), pp. 293-
298.
Philip J. R. (1964). "Similarity hypothesis for capillary hysteresis in porous materials." Journal of
geophysical research 69(8), pp. 1553-1562.
Rahardjo H., Heng O. B., Choon L. E. (2004). "Shear strength of a compacted residual soil from
consolidated drained and constant water content triaxial tests." Canadian Geotechnical Journal
41, pp. 421-436.
Reid M. E., Brien D. L. (2006). Assessing massive flank collapse at stratovolcanoes using 3-D slope
stability analysis. In: Landslides from Massive Rock Slope Failure. Springer. 49: 445-458.
Richards L. A. (1931). "Capillary conduction of liquids through porous medium." Journal of Physics 1,
pp. 318-333.
Rifa'i A. (2002). Mechanical testing and modelling of an unsaturated silt, with engineering
applications. Lausanne, EPFL. PhD. Thesis.
Romero E. (1999). Characterisation and thermo-mechanical behaviour of unsaturated Boon clay: An
experimental study. Barcelona, UPC. PhD. Thesis.
Salager S. (2007). Etude de la rétention d'eau et de la consolidation de sols dans un cadre thermo-
hydro-mécanique. Montpellier, Université Montpellier 2. PhD. Thesis.
Schofield A. N., Wroth C. P. (1968). Critical state soil mechanics, McGraw-Hill.
Seker E. (1983). Etude de la déformation d'un massif de sol non saturé. Lausanne, EPFL. PhD. Thesis.
Sharma R. S. (1998). Mechanical behaviour of highly expansive unsaturated clays, Oxford. PhD.
Thesis.
Sheng D., Sloan S. W., Gens A. (2004). "A constitutive model for unsaturated soils: thermomechanical
and computational aspects." Computational mechanics 33(6), pp. 453-465.
Sillers S. W., Fredlund D. G. (2001). "Statistical assessment of soil-water characteristic curve models for
geotechnical engineering." Canadian Geotechnical Journal 38, pp. 1297-1313.
Sivakumar V. (1993). A critical state framework for unsaturated soils. Sheffield, University of
Sheffield. PhD. Thesis.
Sudhakar M. R., Revanasiddappa K. (2006). "Influence of cyclic wetting drying on collapse behaviour
of compacted residual soil." Geotechnical and Geological Engineering 24, pp. 725-734.
Sugii T., Yamada K., Kondou T. (2002). "Relationship between soil-water characteristic curve and void
ratio." UNSAT 2002, Recife, pp. 209-214.
Sun D. a., Sheng D., Xu Y. (2007). "Collapse behaviour of unsaturated compacted soil with different
initial densities." Canadian Geotechnical Journal 44(6), pp. 673-686.
Taibi S. (1994). Comportement mécanique et hydraulique des sols soumis à une pression interstitielle
négative. E. C. Paris. Paris, Ecole Centrale Paris. PhD. Thesis.
Tamagnini R. (2004). "An extended cam-clay model for unsaturated soils with hydraulic hysteresis."
Tarantino A., Mongiovì L. (2000). "Experimental investigations on the stress variables governing
unsaturated soil behaviour at medium to high degrees of saturation." Experimental Evidence
and Theoretical Approaches in Unsaturated Soils; Proc. of an International Workshop, Trento,

pp. 3-18.
Tarantino A., Mongiovì L. (2000). "A study of the efficiency of semi-permeable membranes in
controlling soil matrix suction using osmotic technique." Unsaturated soils for Asia, pp. 303-
308.
Taylor (1948). Fundamentals of soil mechanics, Wiley.
Van Genuchten M. T. (1980). "A closed form of the equation for predicting the hydraulic conductivity
of unsaturated soils." Soil Sciences American Society (44), pp. 892-898.
Vanapalli S. K., Fredlund D., Pufahl D. E. (1999). "The influence of soil structure and stress history on
the soil-water characteristics of a compacted till." Geotechnique 49(2), pp. 143-159.
Vaunat J., Cante J. C., Ledesma A., Gens A. (2000). "A stress point algorithm for an elastoplastic model
in unsaturated soils." International Journal of Plasticity 16(2), pp. 121-141.
Vicol T. (1990). Comportement hydraulique et mécanique d'un limon non saturé. Application à la
modélisation. Paris, ENPC. PhD. Thesis.
Wheeler S., Karube D. (1995). "Constitutive modelling." 1st International Conference on Unsaturated
soils, UNSAT, Paris. pp. 1323-1356.
Wheeler S. J., Sharma R. S., Buisson M. S. R. (2003). "Coupling of hydraulic hysteresis and stress-strain
behaviour in unsaturated soils." Geotechnique 53(1), pp. 41-54.
Wheeler S. J., Sivakumar V. (1995). "An elasto-plastic critical state framework for unsaturated soil."
Géotechnique 45(1), pp. 35-53.
Wroth C. P., Houlsby G. T. (1985). "Soil mechanics- Property characterization and analysis
procedures." 11th international Conference on soil mechanics and foundation engineering,
San Francisco, pp. 1-55.
Yahia-Aïssa M., Delage P., Cui Y.-J. (2000). "Volume change behaviour of a dense compacted swelling
clay under stress and suction changes." In: Experimental Evidence and Theoretical Approaches in
Unsaturated Soils, Trento, Balkema, pp. 65-74.
Yu H.-S. (2006). Plasticity and Geotechnics, Springer.
Zerhouni M. I. (1991). Rôle de la pression interstitielle négative dans le comportement des sols -
apllication au calcul des routes. Paris, Ecole Centrale Paris. PhD. Thesis.
56 CHAPTER 2
3. Effective stress-strain framework
3.1 Foreword
This chapter is dedicated to the determination of a consistent and convenient stress and
strain framework as a basis for the formulation of the constitutive model. The determination
of such a framework lies on the review of the up-to-date effective stresses from literature, the
critical comparison of their possible forms and the assessment and validation of the final
choice. The selected form of stress variable is named generalized effective stress.
The contents of the present chapter based on our paper published in the International
Journal for Numerical and Analytical methods in Geomechanics (Nuth and Laloui 2008). The
complete contents of this paper entitled “Effective stress concept in unsaturated soils:
clarification and validation of a unified framework” are reported as such in the following
with a page setting that is consistent with the rest of the thesis. Yet, additional analyses are
featured in the last part of the chapter to complement the paper and provide a more detailed
insight into the results obtained on different materials. The parts of the paper to which these
complementary data are related are annotated with footnotes throughout the text.
3.2 Introduction
At early stages of soil mechanics, Terzaghi (1936) introduced the concept of effective
stress for the particular case of saturated soils followed by other major contributions to the
definition of effective stress (Skempton 1960; Nur and Byerlee 1971). Since these early works,
the effective stress principle has been widely used in modelling geotechnical engineering
applications, giving a simple link between elastic deformation and stress acting in the soil,
the latter being proportional to the observed deformation in saturated porous media. The
effective stress is a combination of both the externally applied stresses and the internal
pressure of fluid phase(s) and enables the conversion of a multiphase porous medium into
mechanically equivalent single-phase continuum. Indeed, the stress state is considered as
uniquely represented by the effective stress.
The objective of this contribution is to draw back the historic developments of the
effective stress in the framework of porous materials, remaining as long as possible in a
macroscopic conception of saturated or fully dry soil mechanics. Then, the accent is laid on
the extension of the effective stress principle to materials saturated with more than one fluid.
A particular attention is paid to the various stress frameworks elected up to now for
constitutive modelling of unsaturated soils, by covering the early single effective stress, the
independent stress variables approach and the most recent combined effective stress
conceptions. The stress frameworks considerations developed below are only intended to
introduce constitutive modelling of unsaturated soils. Nevertheless some fundamental
modelling aspects such as general constitutive equations or yield limit definition will be
dealt with in the following.
58 CHAPTER 3
Once the appropriate stress framework is elected regarding the analysis led throughout
the paper, the major implications of the ‘Bishop generalized effective stress framework’ are
eventually investigated. The principle is to re-plot experimental data in conventional stress-
strain planes, stress variables being usually net stress and suction, into new effective stress
planes. Isotropic compression, drying-wetting paths and shear results are examined in order
to evidence possible simplifications brought by the use of the unified effective stress
framework.
3.3 Conceptual principle of effective stress

The need for unified formulations of the stress variables raised from the complexity of
modelling the behaviour of porous materials, that are likely to be filled fully or partially with
fluids of different nature from the skeleton matrix, termed ‘solid phase’ in the following.
Several theories tend to reach a homogenization of the phases in order to obtain an
equivalent macroscopic continuum. Then, the elastic or elastic plastic deformations are
supposed to be directly linked to changes in the macroscopic stress quantity.
Although the lexicon of effective stress is due to Terzaghi (1936), as recalled in next
paragraph, a voluntarily more general view is presented in this section. The effective stress
concept is basically applicable to any porous medium that may undergo effects of both
external loads and internal pore pressures, also called interstitial pressures, due to possible
of fluids within the pore space. Imitating Terzaghi’s definition of the principle of effective
stress, one can formulate two statements:
• Seating solely in the solid skeleton of the medium, the effective stress enters the elastic
and elasto-plastic constitutive equations of the soil matrix, linking a change in stress to
strain-like quantity of the skeleton. The effective stress can thus be simply defined as
that inducing the mechanical elastic strain of the solid skeleton, leading to the
following simple constitutive equation:
εije = Cijkl
e
σ kl′ (3.1)
where εije is the rate of elastic strain of the solid skeleton, Cijkl
e
is the drained elastic
compliance matrix, and σ kl′ the increment of effective stress. A unique stress is necessary and
sufficient to describe the mechanical behaviour.
• Expressed as a function of the externally applied stresses and the internal pore fluid
pressures, the effective stress converts the analysis of a multi–phase porous medium
into a mechanically equivalent, single-phase, single-stress state continuum
comprehension.
In a similar fashion to Khalili et al. (2005), Fig. 3.1 gives a schematic representation of the
effective stress principle. The unified effective stress σ ij′ averaged over the total volume
schemed in Fig. 3.1 is written as follows:
n
σ 'ij = σ ij − ∑ α β pβ δ ij (3.2)
β =1
where σ ij designates the total exterior stress, pβ the pressure of fluid β and α β the
scaling factor for phase β . As developed in depth later, parameters α β should account for
the influence of each fluid pressure on the overall behaviour. The different fluid
EFFECTIVE STRESS 59
I 1 I '1
p 2
I p I I '3 I '3
3 1 p 3
3
p n
I 1 I '1
Figure 3.1 Conversion of multiphase and multi-stress medium (solid phase and pore space filled with
n fluids) into single continuum. pβ designates the interstitial pressure of fluid β , σ ij is the exterior
stress.
contributions are also likely to be linked to their respective volumetric fraction.

Independently of the stress state, all the fluids, may they be wetting or not, are assumed to
react in an isotropic way, generating the same pressure in all directions. Obviously, any
supplementary interaction between the different fluid phases should be integrated in the
generalized stress equation or at least accounted for in parallel to Eq. 3.2.
The previous general definition is intended for the widest range of porous media or
geomaterials, fluid phases being possibly liquid (e.g. water, oil) or gaseous (air). Furthermore
the generic Eq. 3.2. let the effective stress concept being applicable to the categories of
materials classified between fully separated grains and solid rock with interconnected pores.
However, no single effective stress formulation has ever been proved to be applicable to all
materials at once; so, even though the effective stress concept remains applicable to many
materials, the formulation of the stress variable itself (see coefficients α β in Eq. 3.2) is very
likely to vary from one structure to another. Nevertheless, a proper definition of the unified
effective stress enables addressing several kinds of geomechanical problems. The range of
applications could thus extend from the constitutive modelling of unsaturated soils to the
interpretation of pollutants draining in geomaterials.
3.4 Overview of effective stress formulations for saturated soils

Historically, the general definition in the previous paragraph has been first written for the
simple case of a porous medium filled with one single fluid (Terzaghi 1936). Focusing on
soils, two limit states of saturation are remarkable. First, interstitial space can be filled with
air, assumed compressible. The effective stress in the dry soil is:
σ ij′ = σ ij (3.3)
Eq. 3.3 is valid only if air pressure remains equal to the atmospheric (reference) pressure.
Secondly, Karl Terzaghi also writes a proper version of the effective stress principle for
soils saturated with water (Fig. 3.2), using the following two sentences (Terzaghi 1936):
• All measurable effects of a change of stress of the soil, that is compression, distortion,
change of shearing resistance, are exclusively due to changes in effective stress.
60 CHAPTER 3
• σ ij′ is defined by
σ ij′ = σ ij − pw .δ ij (3.4)
with pw being the pore water pressure and δ ij the Kroenecker’s delta: δ ii = 1; δ i ≠ j = 0 .
One assumption implicitly made here is the incompressibility of the fluid phase.
Terzaghi’s expression for the effective stress was shown to be the “experimental true
form”, in the sense that it allows an accurate comprehension of consolidation process among
others. However it is only representative of the very particular case of saturated soils with
incompressible grains and a pore space completely filled with incompressible fluid.
Considering that the latter two assumptions could not be verified for the full range of
geomaterials or with particular conditions, other formulations have been proposed for the
effective stress for saturated materials. In the synthesised review of the main proposals
presented hereafter, the effective stresses are written under the following generic
formulation of Eq. 3.5, which is obviously a particular form of Eq. 3.2 limited to a single fluid
phase.
σ 'ij = σ ij − α w pwδ ij (3.5)
In a voluntary effort to remain at a macroscopic level of stress, or at least to focus mostly

on experimental true forms of skeleton stresses, any considerations implying microscopic or
grain-size analyses (e.g. (Skempton 1964; Lu and Likos 2006)) are disregarded here.
In fact, Terzaghi himself first suggested that α w should equal the material porosity n , but
ended up with the final Eq. 3.4 based on experimental observation. Nevertheless, Biot (1955),
among others, argued that the effect of the interstitial pressure pw should actually be scaled
down in Terzaghi’s expression in order to weight the respective reactions proportionally to
the volumetric fractions. This is in agreement with the preliminary comments stated about
Eq. 3.2. A possible form of the effective stress could include the porosity n , which is the
volume of voids over the total volume, as a scaling factor for the fluid pressure, leading to
the form of Eq. 3.6 below.
σ ij′ = σ ij − n. pw .δ ij (3.6)
where one volumetric state parameter is required to be known.

From a different point of view, the grain compressibility is not always negligible versus
I 1 I '1
I 3
p w I 3 I '3 I '3
I 1 I '1
Figure 3.2 Graphical representation of effective stress concept for saturated soils.
EFFECTIVE STRESS 61
that of the skeleton. Coefficient α w in Eq. 3.5 is thus assumed by Skempton (1964) to depend
on the finite compressibility of the grains Cs and the drained compressibility of the granular
skeleton C .
⎛ Cs ⎞
σ ij′ = σ ij − ⎜1 − ⎟ pwδ ij (3.7)
⎝ C ⎠
⎛ Cs ⎞
Nur and Byerlee (1971), among others, determined identically α w = ⎜1 − in the
⎝ C ⎟⎠
particular framework of rock mechanics.
The last main family of effective stresses for saturated soils makes use of both material
parameters cited previously, coefficient α w being a combination of porosity and
compressibility coefficients. For instance, Suklje (1969) proposed the following formulation:
C
α w = 1 − (1 − n) (3.8)
Cs
As a conclusion, the general form of effective stress for saturated soils (Eq. 3.5) is declined
into several main families, involving either the porosity, the compressibility or both, as
reviewed by Jardine et al. (2004) . Letting extra parameters appear in the effective stress may
enable a more general comprehension of natural materials, considering that separated grains
may behave differently than solid rock. However, some basic assumptions often enable a
neat simplification of the effective stress form. This does not mean that a unique effective
stress applicable to all materials could ever be determined. Indeed, in the case of soils, the
assumption of a negligible compressibility of grains versus that of the whole skeleton is
mostly reasonable, as evaluated by Skempton (1964), and enables recovering Terzaghi’s
effective stress (Eq. 3.4). So, in the following, accepting the assumptions of incompressibility
of water and solid grains mentioned previously, Terzaghi’s effective stress is believed to
represent the only stress governing the mechanical elastic strain of the soil material in
saturated conditions. Assumptions on compressibility are supposed to hold for the
unsaturated conditions.
3.5 Review of stress frameworks for unsaturated soils

Even though the long-time debate on the most appropriate stress framework for
unsaturated soils has eventually come to an end, evidencing the need for complete hydro-
mechanical stress framework (Gens 1995; Jardine et al. 2004), several forms of adequate stress
variables are still possible. Those different unified stresses, now used for advanced
constitutive modelling of unsaturated soils, present several levels of complexity, and can be
considered as inherited either from the Bishop single effective stress or from the independent
stress variable approach. The common feature of the new stress conceptions is nevertheless
the use of suction, or a modified version of it, as a second stress variable to build complete
hydro-mechanical frameworks. It is proposed hereafter to overview the arguments at the
origin of these advanced unified stresses, by retracing the chronological evolution of
effective, independent and combined stresses for unsaturated soil modelling.
62 CHAPTER 3
3.5.1 Bishop single effective stress
A single effective stress for unsaturated soils

Partial saturation, corresponding to more than one fluid filling the pores, is considered as
the most common state for natural and engineered soils. As Terzaghi’s effective stress is
valid only for the limit states of full saturation of pores with one fluid alone, namely water or
air, a need for extending the effective stress principle to unsaturated states raised. In the
present case, the soil is a porous medium saturated with two fluids, one being a wetting fluid
with respect to the other non-wetting fluid, respectively water and air for instance. In natural
conditions, the water is obviously a mixture of water and dissolved air rather than pure
water. Reversely, the gaseous phase can be almost saturated with water. In the following,
unless otherwise specified, each of the two fluid phases is assumed to be homogeneous as
idealised liquid and gaseous fluids, respectively. In comparison with the bi-phasic problem
previously treated – solid grains and water, the third phase appearing in the pore space can
not be considered as uncompressible. This basically gives birth to a specific behaviour of the
interstitial fluids as well as particular fluid-fluid interfaces.
In the early works of Bishop (1959) the effective stress in unsaturated soils has been
straightforward defined as “a function of the total stress and the pore pressure which
controls the mechanical effects of a change in stress”, the goal being once more to convert a
multiphase and multi-stress medium into a mechanically equivalent single phase and stress
state continuum. In other words, extending Terzaghi’s proposals to partially saturated soils
means assuming that:
• All measurable effects of a change of stress of the soil are exclusively due to changes in
effective stress.
• In unsaturated soils, the effective stress is defined as the excess of total stress σ ij over
an equivalent pore pressure p * :
σ ij′ = σ ij − p * δ ij (3.9)
The quantity p * may be considered as that portion of the effective stress in a soil
resulting from the pressure of all fluids in the pores. Defining pa as the interstitial air
pressure in addition to the pore water pressure pw , Bishop’s stress takes the peculiar
following form:
σ ij′ = (σ ij − paδ ij ) + χ ( pa − pw )δ ij (3.10)
where χ is called the effective stress parameter or Bishop’s parameter. Terms

σ ij − paδ ij = σ net − ij and pa − pw = s respectively define the net stress and the matric suction.
It is recalled that the formulation Eq. 3.10 is aimed at averaging the stresses over a
representative elementary volume containing all constituents: air, water and solid grains. As
mentioned previously, the effective stress should be only representative of solid skeleton
stress. A question studied later is the validity of the use of the effective stress as a unique
state parameter for the whole partially saturated material. Still, this is verified in the
particular case of saturated conditions and incompressible grains and wetting fluid, acting
with a neutral stress.
The importance of effective stress parameter χ (Eq. 3.10) lies in the width of range of
saturation state it confers to the effective stress. Indeed, the parameter is imposed to vary
EFFECTIVE STRESS 63
Figure 3.3 Effective stress parameter versus degree of saturation for a number of different soils (in
Zerhouni 1991).
from 0 for dry soils to 1 for saturated soils, enabling a simple transition from partially to fully
saturated states, and recovering Terzaghi’s expression (Eq. 3.4) for the saturated case. As a
part of the macroscopic expression of stress in the unsaturated medium, the Bishop’s
parameter was basically introduced to scale down the influence of suction in function of the
volumetric ratios of the different fluid phases. This obviously defines a primary hydro-
mechanical coupling by making the effective stress depending on the matric suction
multiplied by a peculiar parameter:
σ ij′ = σ net − ij + χ sδ ij (3.11)
The effective stress parameter has been thought to be likely linked with volumetric
fraction of fluids occupying the pore space. Moreover, it is stated that χ can be different for
shear strength or volumetric deformation. Bishop and Donald (1961) first attempted to verify
Bishop’s relation experimentally, and assuming the validity of Eq. 3.11 in elastic conditions,
the values of χ could be evaluated as (Fig. 3.3):
χ = χ ( S r ) (3.12)
One experimental measurement technique of χ was proposed by Jennings (1960) by

comparing the behaviour of a soil specimen under changes in applied suction with the
behaviour of an identical saturated sample under changes in external pressure. This process
lies on the tacit assumption of the validity of the effective stress principle over the whole
range of partial saturation, and moreover to verify that all induced deformations from a
definite initial state remain in the reversible domain only. Bishop et al. (1960) also measured
χ for several soils using volume change and shear strength processes. Zerhouni (1991)
updated the recapitulative Fig. 3.3 initially taken from Jennings and Burland (1962).
Fig. 3.3 evidences a seemingly trend for χ to follow the variations of the degree of
saturation. Consequently, the uniqueness of relationships between χ and S r have been
often questioned. A possible elementary choice is written by Schrefler (1984) under the
following identity:
χ = Sr (3.13)
64 CHAPTER 3
which, along with Eq. 3.10 gives the definition of the Bishop’s generalized effective stress,
or average soil skeleton stress (Jommi 2000). Detailed framework around this generalized
stress will be provided in part 3.5.3 showing that this stress is at the basis of numerous recent
constitutive stress frameworks. However, early works hardly used this formulation and
favoured more complex expressions for parameter χ .
Authors argued that, judging on Fig. 3.3, the Eq. 3.13 failed for given ranges of saturation
out of 20 to 80 %, that is for very dry or very wet soils. Aitchison (1960) gives thus a fitted
expression for the effective parameter, written as follows:
⎧1 if S r = 1
⎪
χ = ⎨α (3.14)
⎪⎩ s se if Sr < 1
with se designating the air entry suction and α a coefficient varying from 0.3 to 0.35.
Jennings and Burland (1962) state in addition that whatever the relationship chosen for
parameter χ , there is no unique relationship between volumetric strain and effective stress
as defined by Eq. 3.10 and that above a certain critical degree of saturation, the effective
stress principle is not valid.
In another recent view point (Khalili et al. 2004), authors recalled that, in reference to
(Coleman 1962), parameter χ might also be related to the current stress and stress history,
which, according to the authors, could explain the possibly bad correlation between χ and
Sr which is a volumetric parameter. By plotting χ against the ratio of matric suction over
the air entry value (suction ratio), authors obtain a unique relationship for most soils. Khalili
and Khabbaz (1998) thus determined a mathematical expression for χ as a function of
matric suction and air entry suction se :
⎧⎛ s ⎞−0.55
⎪ if s > se
χ = ⎨⎜⎝ se ⎟⎠ (3.15)
⎪
⎩ 1 if s ≤ se
The advanced framework developed by Khalili and Khabbaz (1998) remains however
different from the early simple effective stress conceptions previously described. In order to
avoid confusion, no more reference to these works will be made until part 3.5.3.
Apparent limitations of the single effective stress

Prior to the nineties, the effective stress principle extended to partially saturated soils
remained basically associated with a fully elastic (linear or not) conception. This elementary
approach brings obviously a number of difficulties for constitutive modelling, and misleads
numerous research teams on the validity of the effective stress. In the following are
synthesised the statements apparently arguing in disfavour of the effective stress framework,
followed by a new interpretation in the light of the introduction of the elasto-plasticity
concepts.
Jennings and Burland (1962) were among the first ones to question the validity of Bishop’s
relation, considering that it could not seem to provide explanation for collapse phenomenon
in unsaturated soils. The range of soils of interest here are the so-called heaving and
collapsing soils. Collapse phenomenon or plastic compression is observed along a wetting
path under a significant constant mechanical load (Jennings and Burland 1962; Matyas and
EFFECTIVE STRESS 65
0.02
Wetting
0.01

0
Swelling
-0.01
-0.02
Collapse
-0.03
-0.04
0 2 105 4 105
Figure 3.4 Swelling collapse phenomenon in kaolin, under a mean net stress of 40 kPa. Suction
decrease is applied (after Sivakumar 1993).
Radhakrishna 1968). Along the same wetting path, reversible swelling can be observed prior
to plastic collapse (Fig. 3.4).
Within the previously defined limited elastic context, during wetting under constant net
stress, the matric suction s decreases down to zero, which in consequence provokes a
decrease in effective stress (Eq. 3.10). This stress reduction should, by definition for a
saturated material, induce a slight increment of void ratio, which is contrary to the
experimental observation (Fig. 3.4). So the Bishop’s generalized effective stress apparently
failed to capture the collapsible behaviour of soil, which supported afterwards the
determination of other stress state variables as developed in depth in part 3.5.2.
Nevertheless, later interpretation of experimental results showed that the collapse
phenomenon is linked to plastic behavioural mechanism. In the early 1960s, Leonards (1962)
already suggests to interpret the observed compression upon wetting as a sliding of particles
with respect to each other, even in all-round compression tests. That rearrangement actually
characterises a plastic phenomenon. These observations show that the arguments against the
effective stress were misleading, in the sense that they imputed the limitations to the stress
itself instead of judging the inadequacy of the whole reversible context. In the end, it
encouraged further use of the effective stress principle associated with complete elastic-
plastic frameworks. In addition, the evolution of unsaturated soils knowledge then gave
birth to modified effective stresses (e.g. incorporating the effects of phases interfaces) used in
a elasto-plastic frameworks. These considerations are the object of paragraph 3.5.3.
Another apparent shortcoming of the single effective stress approach is the definition and
use of the effective stress parameter, see previous paragraph. Provided that the ratio χ S r
appeared not to be equal to one, the possible difficulties to determine parameter χ were
pointed out. Mainly, the parameter determination is likely to require non-conventional
experimental procedures. Once more, these arguments should be handled with care,
considering that the early effective parameter determination and thus the pseudo-validity of
effective stress principle were examined in a fully elastic context. On the other hand
Fredlund and Morgenstern (1977) also argue that the effective stress incorporates a soil
parameter characteristic, rendering the equation a constitutive expression instead of a mere
66 CHAPTER 3
description of the stress state, which means that supplementary stress variables could be
added to get rid of the material parameter.
Other effective stress determination strategies are based on a micromechanical
equilibrium analysis. Lu and Likos (2006) propose a micromechanical conceptualization of
the effective stress. The force equilibrium of a two soil grain system leads to the identification
of an interparticle force balancing the external force, the forces arising from Van der Waals
attraction, electrical double layer repulsion, chemical cementation and lastly the capillary
forces. In an opposite view point Tarantino and Mongiovi (2000) affirm that being a
macroscopic concept, the effective stress principle could be proved or disproved only on the
basis of experimental evidence and not of theoretical models of a microscopic level.
3.5.2 Independent stress variables
Identification of state variables for unsaturated soils

Considering any constitutive model with Bishop’s generalized stress seemed unable to
describe all features of behaviour of unsaturated soils in a simple form; it has therefore been
proposed to adopt a multiple stress variable approach. In immediate response to Jennings
and Burland (1962) breaking down the effective stress for unsaturated soils, Coleman (1962)
suggests to write the soil volumetric strain in the following modified fashion, under triaxial
conditions:
dV ⎛ ⎛1 ⎞ ⎞
− = −C21 ( dpw − dpa ) + C22 ⎜ d ⎜ (σ 1 + 2σ 3 ) ⎟ − dpa ⎟ + C23 ( dσ 1 − dσ 3 ) (3.16)
V ⎝ ⎝3 ⎠ ⎠
with V being the overall volume, σ 1 the axial total stress, σ 3 the lateral total stress, and
each independent parameter C21 , C22 , C23 related to the material characteristics. Merging the
second and third term on the right hand side of Eq. 3.16 into a single one, we write the
constitutive equation as follows, considering small strain hypothesis:
d ε ije = Cijhk
e
(dσ hk − dpaδ hk ) + C s (dpa − dpw )δ hk (3.17)
where C s is termed the elastic hydric modulus (coefficient of proportionality between

strain and suction).
Compared with constitutive Eq. 3.1, two independent stress state variables have emerged,
resulting in the need for a double constitutive matrix. In that particular case, the stress state
is namely described by the net stress and the matric suction. Indeed, the identification of
state variables can be based on multiphase continuum mechanics (Fredlund and
Morgenstern 1977), leading to the conclusion that “any two of the three possible state
variables (σ ; pw ; pa ) can be used to define the stress state”, possible combinations being:
(1) (σ − pa ) and ( pa − pw ) e.g. used in (Alonso et al. 1990)
(2) (σ − pw ) and ( pa − pw ) e.g. used in (Geiser et al. 2006) (3.18)
(3) (σ − pa ) and (σ − pw )
Thus, two stresses are proposed for both the soil particles and the contractile skin. The
experimental verification of this statement consists in null tests, for which each quantity of
the combined stresses is increased by the same amount. In such null tests, there is no
EFFECTIVE STRESS 67
tendency for volume change of the overall sample and no change in the degree of saturation.
Practically, the verification is possible for the usually controlled stresses that are the net
stress and matric suction. Fredlund and Morgenstern (1977) led a number of concluding null
tests on silts and kaolin. Tarantino and Mongiovi (2000) later corroborated the propositions
from Fredlund and Morgenstern, observing neither volume change nor water volume
change in null tests practised on kaolin samples. Different apparatuses were used, to
compare results of a positive air over pressure (axis translation technique) and of a truly
negative water pressure via the osmotic technique.
The most frequently used couple of variables in this approach is the combination (1) of eq.
3.18: the net stress ( pnet = σ − pa ) and the matric suction ( pa − pw ) , for physical and
practical reasons (e.g. Matyas and Radhakrishna (1968), Alonso et al. (1990)) . The scalar
variable ( pa − pw ) , corresponding to suction, has a definite physical meaning, while most of
the time the air pressure may be considered constant and equal to atmospheric pressure:
pa = patm (= 0) . Under this assumption, net stress is simplified to total stress and suction
becomes a negative water pressure. Moreover, this choice is adapted to the axis translation
technique, consisting in the application of pa > 0 . Using this description, the constitutive law
is recalled to be written in the form of Eq. 3.17.
Geiser et al. (2006) adopted however the second combination of stresses in Eq. 3.18 to
overcome problems raised by the use of net stress in describing the whole range of
saturation. Further details about these limitations are given later. The original combination of
saturated effective stress or Terzaghi’s effective stress (σ − pw ) with matric suction
( pa − pw ) lead to a modified constitutive variational relationship written as follows:
′e σ hk
εije = Cijhk ′ + Cs′ sδ ij (3.19)
′ being the increment of Terzaghi saturated effective stress, C ′e the elastic matrix
with σ hk
and Cs′ an elastic proportionality coefficient for the hydric behaviour, both of which
assumed independent each with respect to the other.
Eventually, no matter the combination elected from Eq. 3.18, the use of two independent
stress state variables will enable the decomposition of strain increments ε into a mechanical
part εm and a hydraulic part εh :
ε = εm + εh (3.20)
Eq. 3.20 is identifiable with both forms Eqs. 3.17 and 3.19.
Limitations of the independent stress variables approach

Although the initial proposals by Leonards (1962) well insisted on the need for
considering elasto-plastic mechanisms for behavioural interpretation of unsaturated soils,
the first early approach using independent stress variables of Matyas and Radhakrishna
(1968) still deals only with non linear elasticity. The state surfaces describe the volumetric
state with respect to net stress state on the one hand and matric suction on the other hand.
But what was believed to be a solution to modelling swelling and collapsing behaviour was
only valuable along monotonic paths and remained unable to describe for instance loading-
unloading cycles. The third axis ‘matric suction’ added to the mechanical stress-strain plane
(ε − σ net ) remained a simple artefact for a better stress-strain curve fitting within a non linear
68 CHAPTER 3
fully reversible context. Moreover, the state surfaces have to be determined empirically for
each material under particular stress state conditions and remain applicable only for strictly
similar stress paths. Obviously, the elastic models present the shortcoming of neglecting any
irreversible process.
Nevertheless, further models continued to use independent stress variables frameworks,
and investigations were dedicated to elasto-plasticity contexts rather than on nature of the
stresses. Ever since the pre-eminent contribution of Alonso et al. (1990), most recent models
based on the independent stress variables are formulated within a complete elasto-plastic
framework, often based on those of the constitutive models for saturated soils. Other
weaknesses linked to the nature of stress variables themselves have then to be overcome. In
(Geiser et al. 2000), it is considered that the bi-tensorial approach net stress and suction could
fail to provide a straightforward transition between saturated and unsaturated states, i.e. for
a null suction, Terzaghi’s effective stress cannot be recovered. In other words, the use of net
stress implies that for the saturated state, the constitutive law should be expressed in terms
of the total stress instead of the effective stress. To overcome this shortcoming, authors
proposed to replace net stress by the ‘saturated effective stress’ (σ − pw ) and thus enable a
natural switching from partially to fully saturated states (Geiser et al. 2006). However, letting
the pore water pressure appear in the mechanical variable may cause the representation of
wetting-drying stress paths to be more complex than in conventional net stress and suction
planes.
As the models following the independent variable approach were the first ones to deal
with elasto-plasticity and coupled yield limits, some fundamental concepts are linked to
them, such as the Loading Collapse (LC), Suction Increase (SI) and Suction Decrease (SD)
curves. Some of these concepts will be described in the part 4. However, it can be seen that
more material characteristics are likely to be determined compared to the effective stress
approach (see Eqs. 3.17 and 3.19 where not only the constitutive matrix is needed, but also
the hydraulic coefficient of proportionality). Other attempts to use the independent stress
variable approach, such as Fredlund and Morgenstern’s (1977) for shear strength of
unsaturated soils, base the failure criterion on Mohr Coulomb criterion. The shear strength is
written:
τ = c′ + (σ − pa ) tan Φ′ + ( pa − pw ) tan Φ b (3.21)
with Φ′ being the effective saturated angle of friction, Φ b the angle of friction accounting
for the matric suction contribution to shear strength and c′ the effective saturated cohesion.
The analysis proved to be complex, considering that the material angle Φ b has a non linear
evolution over wide ranges of suction (Fig. 3.5), meaning that a single determination of
Φ b could not be representative. Even for more advanced constitutive models (e.g. Alonso et
al. (1990); Wheeler and Sivakumar (1995)) at least one separate function must be introduced
to account for the increase of strength with suction.
Lastly, separating completely the mechanical stress from the hydraulic stress prevents a
direct accounting of the hydraulic hysteresis effects on the mechanical stress paths. A
deepened discussion on this feature of the hydro-mechanical coupling will be exposed later.
EFFECTIVE STRESS 69
Φ b2
Φ1b
s
Figure 3.5 Non linearity of apparent friction angle with suction.
3.5.3 Coming back to Bishop-type effective stresses

Numerous experimental data enable an advanced knowledge of elasto-plastic
mechanisms linked to suction-induced phenomena along with the water / air interfaces
receding or moving forth.
The average reaction to mechanical external loads has been developed previously and
won’t be recalled here. Yet, the hydraulic loading, consisting in increasing suction under
constant net stress, needs to be clarified. This second type of load, corresponding to drying,
could be qualified as internal even though the fluid pressures are experimentally controlled
from outside the sample. The mere effect of suction change is double: it modifies directly the
global stress state of the material and the saturation being changed, the interfaces’ action on
the mechanical behaviour does vary. These bonding effects are termed suction-hardening.
From a standard elastic-plastic interpretation and on the basis of experimental
observations (Geiser et al. 2006) it is recalled that the latter effects are at the origin of the
increase in preconsolidation pressure with suction, and under certain stress conditions, they
are linked to the well-known wetting collapse.
Starting form this acquaintance of an inner hydro-mechanical coupling for the
unsaturated behaviour, researchers focused back on unified forms of effective stress of the
Bishop type, combined with other stress variables to build a proper exhaustive stress
framework. In other words, state variables identified by Fredlund and Morgenstern could be
combined in different ways to form the effective stress σ ′ and completed with a second
stress ξ . Most of the stress frameworks for the recent models are classified under this family
of combined stress. So, in a similar fashion to Gens (1995) it is proposed to sort the combined
stress in two categories according to the level of complexity in the Bishop’s like stress
formulation and second stress variables.
The first category (C1) involves the following couple of stress variables:
σ ′ = σ net + μ1 ( s ) (3.22)
ξ = ξ ( s, S r ) (3.23)
70 CHAPTER 3
The second catergory (C2) gathers the following type of stress framework:
σ ′ = σ net + μ 2 ( s, Sr ) (3.24)
ξ = ξ ( s, S r ) (3.25)
where μ1 and μ2 are functions of suction and/or degree of saturation.
Justifications for effective stress type frameworks

The work input characterisation within the energetic approaches is a plausible way to
identify the possible conjugations of stress and strain variables for constitutive modelling.
Observing that in all constitutive frameworks using the net stress and suction as stress state
variables the net stress is implicitly assumed to be work conjugate to the strains, Houlsby
(1997) proposes to identify the strain-like quantity that is work conjugate with suction.
Indeed, Houlsby demonstrates that the choice of stress and strain variables is arbitrary
provided that the variables adopted are work conjugate. The rate of input work (per unit
volume) to the soil, W , is expressed as the sum of the products of the stresses with their
corresponding strain rates:
ρ
W = pa n(1 − S r ) a − ( pa − pw )nSr + [σ hk −( Sr pw + (1 − Sr ) pa )δ hk ]εhk (3.26)
ρa
ρ a being the air density and n the soil skeleton porosity. The three terms of Eq. 3.26
describe the rate of input work to compress the air phase, the rate of input work to change
degree of saturation S r and the rate of input work to modify the average deformation,
respectively. A doubtful assumption in the derivation of Eq. 3.26 is the neglecting of the
work dissipated by the air-water interface, the relative velocity between the soil skeleton and
the interface supposed to be null. However, other theoretical studies developed later seem to
corroborate Houlsby’s conclusions.
The power input to compress the air phase assumed negligible, the following simplified
relationship is obtained:
W = −( pa − pw )nSr + [σ hk −( Sr pw + (1 − S r ) pa )δ hk ]εhk (3.27)
The right hand side terms of Eq. 3.27 could possibly be rearranged in several manners, for
instance:
(
W = [ σ hk − pa δ hk ] εhk + [( pa − pw )δ hk ] − nSr + S r εhk ) (3.28)
That means that several combinations of stress-strain conjugates are consistent. For
instance, if the net stress is associated with soil skeleton strain (Eq. 3.28), the generalized
strain quantity ( −nS r + S r ε kk ) must be associated with suction. Eq. 3.27 presents another
appealing combination, associating the Bishop generalized effective stress
σ′ = σ hk − pa + Sr ( pa − pw )δ hk to the increment of skeleton strain and the matric suction
s = ( pa − pw ) to the product − nSr .
Considering that the full implications of Houlsby’s assumptions for deriving Eq. 3.26
remained to be investigated, another approach to identify the conjugate stress and strain
variables was introduced in (Dangla and Coussy 2002). Full demonstrations are also featured
EFFECTIVE STRESS 71
in (Coussy 2004). The authors assume that the soil skeleton is constituted by solid particles
and phase interfaces. This point of view is interpretable as a need for taking into account the
bonding effects in addition to the identification of the mean volumetric stress acting on the
solid particles. Thus, in the case of pore space filled with two fluids all fluid-fluid interfaces
and solid-fluid interfaces possess their own proper interfacial energy and entropy and are
assumed to act directly on the skeleton free energy Ψ S :
Ψ S (ε ij , S r , T ) = ψ S (ε ij , T ) + nU ( S r , T ) (3.29)
ψ S is the free energy of solid matrix per unit of volume, U is the overall interfacial energy
per unit volume, n is the porosity, and T designates temperature.
The skeleton state equations of unsaturated thermoporoelasticity are written:
⎛ ∂ψ S ⎞
σ ij + π ij = ⎜ ⎟⎟ (3.30)
⎜ ∂ε ij
⎝ ⎠T
⎛ ∂ψ S ⎞
SS = − ⎜ ⎟ (3.31)
⎝ ∂T ⎠ε ij
∂Ψ S
ns = − (3.32)
∂Sr
π ij , called the equivalent pore pressure is a function of fluids pressures, degree of

saturation and interfacial energy. S S is the entropy.
Consequently, a unique stress variable σ ij′ = σ ij + π ij is sufficient to govern the

deformation of the solid skeleton. This effective stress is similar to Bishop’s generalized
effective stress (see Eqs. 3.10 and 3.13) to which is added a term accounting for the energy of
interfaces. Besides, a second stress variable is defined under the form of suction scaled by
porosity; it governs the strain-like quantity called degree of saturation. In the limit state of
full saturation, Terzaghi’s expression is naturally recovered with the effective stress used in
this framework.
In addition, based on a thermodynamic mixture theory, Hutter et al. (1999) addressed the
conditions allowing the derivation of the effective stress. Following Laloui et al. (2003), it is
shown that Bishop’s generalized effective stress could be thermodynamically consistent if: (i)
the soil comprises density-preserving constituents, (ii) the water is a perfect fluid and (iii) no
shear stress is introduced for the fluids. Other thermodynamic analyses of multiphase
deformation are exposed in (Hassanizadeh and Gray 1990, Muraleetharan and Wei 1999).
Effective stresses of category 1.

It is recalled that the effective stress framework built in the fashion of category 1 set up
two stress variables, the effective stress (Eq. 3.22) which is a particular form of Bishop’ s
expression, and a second stress variable (see Eq. 3.23)
In this paragraph, it is proposed to review the main stress frameworks in this category
used for constitutive modelling as well as their major advantages and shortcomings. Most of
the category 1 effective stress frameworks were defined prior to energetic or thermodynamic
72 CHAPTER 3
approaches mentioned previously. Consequently, justifications for stress frameworks are

often of other particular nature.
Kohgo et al. (1993) first defined an effective stress classified in category 1, under the form:
σ ′ = σ − peq (3.33)
where peq is called the equivalent pore pressure. This pressure is aimed at averaging the
effects of all fluids pressures within the pores. It also designed to recover Terzaghi’s effective
stress on saturated states. Consequently, authors had to express the equivalent pore pressure
in terms of the air entry suction value se , a critical suction sc and a material parameter ae :
peq = pa − s if s ≤ se
⎛ sc − se ⎞ (3.34)
peq = pa − ⎜⎜ se + ( s − se ) ⎟⎟ if s > se
⎝ ( s − se ) + ae ⎠
This formulation is the Bishop’s effective stress (Eq. 3.10) with χ = ae ( sc − se ) ( s − se + ae ) 2

for unsaturated states and χ = 1 for saturated range. The empiric relationship has been
formulated on the basis of the relationship between shear strength of unsaturated soil and
amount of suction. Out of the promising elastoplastic considerations, authors define a second
stress variable ξ = s* = s − se named effective suction to complete the stress framework. This
emergence of a second stress variable in the context of effective stress is a pioneering
formulation that could be later confirmed by the energetic approach.
Modaressi and Abou-Bekr (1994) basically focused on one of the features presented above
that is the shear strength increasing and reaching a limit upon suction increase. Then, the
proposed effective stress for unsaturated states is simply written as follows:
σ ′ = σ + π c ( s ) (3.35)
π c is named the capillary pressure originally determined by Taibi (1994), defined as a

hyperbolic function of suction varying from the air entry suction to a certain maximum value
linked to the void ratio, surface tension and granulometry of the modelled material. The
capillary pressure is quantified by the means of an averaging process of capillary stresses.
Other particularity lies in the range of applicability of the capillary pressure; actually, the
true form of the effective stress is:
σ′ =σ + p (3.36)
with a condition expressed on air entry suction se to recover Terzaghi stress for saturated
states:
p = − pw if p w ≥ se or p = −π c if p w < se
While it is clearly stated by authors that the generalized effective stress concept is applied,
and thus that the given effective stress alone governs the skeleton deformation, the energetic
consistency can be verified only if a second set of work conjugate stress and strain variables
is defined in parallel. This second stress ξ and strain are very likely to be expressible in terms
of matric suction, porosity and degree of saturation.
Alread cited previously is the constitutive framework from Khalili and Khabbaz (1998), in
which the effective stress is very close to those previously mentioned. The following Eq. 3.37
EFFECTIVE STRESS 73
is exactly Bishop’s expression, with a particular function of suction for the effective stress
parameter χ = χ ( s )
σ ij′ = σ net − ij + χ ( s) × s (3.37)
Once more, the function of suction is defined differently whether the air entry suction is
exceeded or not, see Eq. 3.15, and a material parameter is introduced in χ ( s ) . This form of
effective stress coefficient was determined on the basis of examination of shear strength and
volume change data. Again, similarly to other models of category 1, no description of a
second stress-strain couple is provided although it seems to be necessary at least to describe
the hydraulic behaviour in parallel.
To conclude, several forms for function μ1 ( s ) in Eq. 3.22 have been defined and present
similar characteristics, specially giving importance to the air entry suction and involving
material parameters. However, since no information on the degree of saturation is required
to fully describe μ1 ( s) , most authors seem to have left apart the second couple of work-
conjugate stress and strains, which make the hydro-mechanical behaviour incomplete.
Moroever, category 1 effective stresses, due to their inner complexity, might lead to a
difficult representation in the conventional (experimental) stress-strain planes. Lastly,
similarly to the independent stresses approach, category 1 effective stresses do not offer a
direct unified way to incorporate hydraulic hysteresis effects, as explained later.
Effective stresses of category 2

Category 2 gathers most of the stress frameworks for advanced modelling of unsaturated
soils during the last five years. Again, the skeleton deformations are fully governed by a
single effective stress of the Bishop type, to which a second stress variable is added, often to
complete the description of the hydraulic behaviour. The set of stresses take form of Eqs. 3.24
and 3.25
It has been shown previously that several forms are consistent for functions μ 2 ( s, Sr ) and
ξ ( s, Sr ) . Consequently, the choices made out by authors are a matter of convenience,
provided that in any case the bonding effects and the irreversible processes ought to be
accounted for in addition. Most research works tend however towards simple stress
formulations inside more complex elastic-plastic contexts.
Even though the original definition of the Bishop’s generalized effective stress is due to
Schrefler (1984), one of the first constitutive frameworks making use of this unified effective
stress is that of Jommi and Di Prisco (1994) later reported by Jommi (2000). Basing the stress
identification on Houlsby’s remarks on work input, the effective stress called average
skeleton stress by the authors is defined as the difference between the total stress and the
mean value of the fluid pressure weighted with the saturation degree S r :
σ ′ = σ net + S r s (3.38)
As pointed out by the authors, the formulation includes an implicit direct dependency of
the overall behaviour on the hydraulic state, reflected through variables suction and degree
of saturation. Besides, the knowledge of the soil water retention curve is necessary, both to
complete the stress framework by monitoring a second set of stress and strain variables and
to determine the generalized effective stress itself anytime (Eq. 3.38).
Several similar stress frameworks are reported in literature (e.g. Bolzon and Schrefler
1996) but noticeable differences then appear in choice of complementary stress variable ξ ,
74 CHAPTER 3
Eq. 3.25. Due to the simplicity of its formulation, the Bishop generalized effective stress alone
is not able to include advanced effects of partial saturation. Recalling the statements from
introduction of part 3.5.3, experimental observation showed that an increase in suction is
likely to produce two separate effects.
Firstly, a global deformation is induced. From a constitutive point of view, and without
entering elastic-plastic considerations, the straining of the material is governed by changes in
the effective stress. The formulation of the generalized effective stress Eq. 3.38 unveils a
direct modification of σ ′ upon suction change.
Secondly, the suction induced hardening must be accounted for. These hardenings effects
are believed to be attributable to bonding/debonding effects due do the evolution of
interfaces. It is proposed to gather these effects under the generic term of bonding. Since
bonding is not included in Bishop’s generalized effective stress, some authors suggested
including it into the second stress variables. Others preferred keeping the second stress
variable simple and introduce advanced coupling in the whole elasto-plastic model
formulation. This modification is often referred to as ‘second suction mechanism’ (Jommi
2000).
For instance, Sheng et al. (2004) motivated their choice of the following work-conjugate
stresses and strains:
⎛σ ′ ⎞ ⎛ε ⎞
⎜ ⎟ and ⎜ S ⎟ (3.39)
⎝s ⎠ ⎝ r⎠
on judging the possible implementation of their constitutive model in finite element

codes. Indeed, the proposed stress framework seems to be the best choice provided that most
existing finite element codes are programmed for saturated soils. Consequently, a natural
extension to partially saturated states is provided, by simply exchanging effective stresses by
Bishop’s generalized stresses. Moreover, the second stress strain relationship represents
directly the soil water retention curve, which constitutes a relevant output.
Tamagnini’s (2004) proposal for stress framework is strictly the same as the one above. To
introduce the second suction mechanism, the evolution of the preconsolidation pressure
p 'c is defined in terms of a double-hardening mechanism, under the following form:
p 'c = p 'c ( SAT ) + p 'c (UNSAT ) (3.40)
with the second term on the right hand side depending directly on the degree of
saturation and material parameters. The relative simplicity of the stress variables thus makes
it necessary to introduce refinements in the formulation of the hydro-mechanical couplings.
These are also required for modelling the collapse phenomenon upon wetting.
Wheeler et al. (2003) also make use of the work input considerations to justify their choice
of the Bishop generalized effective stress and a second stress variable named the modified
suction. This stress variable, work conjugate with decrement of the degree of saturation, is
the product between matric suction and porosity:
s* = ns = n( pa − pw ) (3.41)
The proposed constitutive model includes remarkable hydro-mechanically coupled

mechanisms, enabling to reproduce bonding effects in a unified approach. Such a
EFFECTIVE STRESS 75
1.5
1.4
1.3
f(s) 1.2
1.1
1
0 1 104 2 104 3 104 4 104
Figure 3.6 Simplified plot of f ( s ) , from Gallipoli et al. (2003).
constitutive context however implies the use of multiple coupled yield loci and a high level
of complexity in the formulation.
A quite opposite contribution concerning the second stress variable is the work from
Gallipoli et al (2003). The authors consider that many features of the elasto-plastic behaviour
are linked to bonding phenomenon between particles, which are linked to menisci. As
bonding cannot be accounted for by exclusively the average skeleton stress as a constitutive
variable, the variable ξ is introduced as a measure of the magnitude of the interparticular
bonding:
ξ = f ( s )(1 − Sr ) (3.42)
Factor (1 − S r ) is aimed at accounting for the number of water menisci per unit volume of
unit fraction and obviously drops down to zero when the soil is saturated. f ( s ) is a function
of matric suction expressing the ratio between interparticle forces at suction s and at null
suction (Fig. 3.6). f ( s ) is determined via a spherical model of two soil grains.
Other proposals tend to include directly the interfaces effects within the mechanical
effective stress, which leads to a different content for function μ 2 ( s, Sr ) in Eq. 3.24 in
comparison with Bishop’s generalized effective stress.
Studies of interest in this field are Dangla and Coussy (2002) followed by Pereira et al.
(2005). As demonstrated previously, authors write the effective stress as follows:
σ ij′ = σ ij + π ij (3.43)
with π ij being the equivalent pore pressure defined as
⎛ ∂ ( nU ) ⎞
π ij = ( ua − Sr ( pa − pw ) ) δ ij − ⎜
⎜ ∂ε ij ⎟⎟
(3.44)
⎝ ⎠ Sr ,T
76 CHAPTER 3
The second set of conjugate stress and strain variables can be taken from Eq. 3.32. The
advantage of the proposed stress framework is that the energy of interfaces is included into
the effective stress, simplifying the modelling of bonding. However, the quantification of the
interfaces energy U is based on the knowledge of the water retention curve:
1
U ( n, S r ) = ∫ s ( n, S ) dS (3.45)
Sr
A major advantage of this determination is that quantity U , by reference to Eq. 3.45 is

directly dependent on the soil water retention curve. Consequently, if the drying path and
the wetting paths are not confounded in plane ( S r − s ) , evidencing a hydraulic hysteresis,
the interfaces energy will reflect this irreversibility. In the end, the hysteresis effects are
naturally incorporated within the effective stress.
More generally, this natural accounting for the hydraulic hysteresis in the stress state is
valuable for any effective stress belong to category 2, provided that both the degree of
saturation and the matric suction enter its formulation. Basically, if the main wetting curve
(
and main drying curve can be separated, the stress paths in plane ξ ( s, S ) − σ ′ will be
r )
invariably different upon drying and wetting. Graphical representation of this feature is
provided in part 3.6. This direct incorporation of hydraulic hysteresis within the skeleton
stress variable is neither possible for the independent stress variables approach nor in the
category 1 of effective stresses.
As a conclusion, the effective stress frameworks classified in category 2 present first the
advantage of providing a natural transition between saturated and unsaturated states. The
existence of μ 2 ( s, Sr ) facilitates the inclusion of hydraulic effects, and particularly hydraulic
hysteresis. Using an effective stress imposes a higher level of complexity for the stress
variable, but potential advantage of expressing a constitutive model in terms of relatively
complex stress variables is that the relationships between strain, strength and these stress
parameters should be simpler than for a model expressed in terms of net stress and suction,
if the stress variables have a strong physical significance. Some authors still argue that the
representation of stress paths in conventional stress-strain planes is not straightforward, or
almost impossible in the case of extremely sophisticated effective stresses, indicating that if a
category 2 effective stress is to be elected, the simple formulations such as Bishop’s
generalized effective stress should be preferred.
Concluding remarks
While it is now clear that any adequate set of work-conjugate stresses and strains can be
used for constitutive modelling of unsaturated soils (Gens 1995), the difference likely comes
from the convenience in manipulating stress-strain planes and secondary functions.
Examining the pros and cons of stresses formulations (Gens 1995), it has appeared that the
effective stresses from category 2 own the highest number of advantages, including the
straightforward transition between saturated and unsaturated states and the direct
accounting for the hydraulic hysteresis effect on the mechanical response. This framework
makes the effective stress principle applicable for describing the mechanical part of the
behaviour, resulting in very likely simplified stress-strain relations. Besides, since the
determination of the effective stress shall not be fastidious due to high number of specific
parameters, the simplification of the function μ 2 ( s, Sr ) down to the product ( s × S r ) is a
promising option. In the following, the Bishop’s generalized effective stress is thus adopted
EFFECTIVE STRESS 77
as a standard and broadly recognized effective stress for describing the mechanical
behaviour of unsaturated soils. Obviously, the hydraulic part of the behaviour must be
described in addition; as the object of the paper is mostly to validate a mechanical effective
stress, the hydraulic stress and strain are taken simple, as written explicitly in the complete
stress-strain framework below:
⎛ σ ′ = σ net + S r s ⎞ ⎛ε ⎞
⎜ ⎟ and ⎜ ⎟ (1)
⎝ s = pa − pw ⎠ ⎝ Sr ⎠
The adoption of the previous framework lies on the conceptual comparative analysis of
stresses for unsaturated soils. It is proposed to validate this stress selection on the basis of
experimental data, by answering two main issues, namely (i) checking the actual
simplifications due to the effective stress formulation and (ii) investigate the apparent
shortcomings usually attributed to Bishop’s generalized effective stress.
3.6 Implications of the use of a generalized effective stress

The implications of the use of the Bishop’s generalized effective stress for the
interpretation of the unsaturated mechanical behaviour are studied in the following. The
hereby assertions are exclusively based on experimental data and form the pretext of an
advanced constitutive model (Pereira et al. 2005). The principle is to re-plot experimental
data sets expressed in terms of classical unsaturated stress variables, which are net stress and
suction into new effective stress planes. The uniqueness of critical state line for different
suctions, and modifications along unsaturated mechanical loading paths are investigated.
Other implicit effects of the hydro-mechanical coupling are also presented, leading to
introductory discussion on modelling.
For the sake of simplicity, in the following, the conventional triaxial stresses are used for
representations. The deviatoric stress q and mean stress p are defined as follows:
q = σˆ 11 − σˆ 33 (3.47)
σˆ 11 + σˆ 22 + σˆ 33
p= (3.48)
3
σˆ ij being the stress tensor chosen: depending on stress framework, it can be the Bishop’s
generalized effective stress σ ′ or the net stress σ net .
3.6.1 Critical state analysis

Khalili et al. (2004) investigated the uniqueness of the critical state line (CSL) in a deviator
stress versus mean effective stress plane for different levels of suction. The stress variable
used for reinterpretation has been given under the form of Eq. 3.11 with the effective stress
parameter χ being a function of suction and air entry value, see Eq. 3.15 Results evidence
clearly a convergence of all critical state points towards a single line whatever the suction, as
plotted for instance in Fig. 3.7.
However, authors do not provide any conventional net stress interpretation plot for
comparison with the new effective stress plot, preventing to estimate whether it is really
worth or not to use the effective stress and if the conversion is significant. We propose to
reinterpret similarly several shearing datasets with the Bishop’s generalized effective stress,
termed ‘effective stress’ in the following for the sake of simplicity.
78 CHAPTER 3
Figs. 3.8 and 3.9 mirror two stress interpretations for CSL, namely the ( q − pnet ) plane
(Figs. 3.8a and 3.9a) and ( q − p′ ) plane (Figs. 3.8b and 3.9b). Most of unsaturated
experimental behavioural studies report data in conventional net stress planes, the latter
being an experimentally controlled stress variable. Under this form, even though the
variations of the friction angle and cohesion are evidently different from one material to
another, Figs. 3.8a and 3.9a highlight a plurality of critical state lines in the net stress planes
for both materials.
Whereas transforming the plot from net stress conception to effective stress representation
does not affect deviatoric stress level, the isotropic effect of suction scaled by degree of
saturation is added to the net mean stress. The critical state lines at different suctions are
uniformly translated horizontally by the amount of stress mentioned above, the immediate
consequence of which is a reduction of apparent cohesion. The alignment property of points
is conserved in the new plane as well as friction angles.
The processed results of Sivakumar (1993) in Fig. 3.8b are particularly accounting for an
obvious unification of the effective critical state whatever the level of suction between 0 and
300 kPa, which corresponds to 1 ≤ S r ≤ 0.59 . Concerning Sion silt (Geiser et al. 2006), while
the net stress interpretation, Fig. 3.9a, already evidences a narrow arrangement of lines, the
effective stress version, Fig. 3.9b, tends to align even more the experimental points, and sets
the whole scattering much closer to the saturated critical state line, accounting for its
uniqueness.
This encourages simplification of parameter determination, assuming that saturated
shear parameters, namely the saturated angle of friction and cohesion, are sufficient to
describe both saturated and partially saturated critical state behaviours, overcoming the
difficulties linked to the suction-dependent cohesion observed in the ( q − pnet ) plane1.
1600 1000
(a) (b)
1400
800
1200 Saturated
Saturated
CSL CSL
1000
600
800
600 400
s = 0 kPa
400 s = 200 kPa
s = 400 kPa 200
200 s = 800 kPa saturated tests
s = 1500 kPa unsaturated tests
0 0
0 500 1000 1500 0 100 200 300 400 500 600
Mean effective stress pnet (kPa) Mean effective stress pnet (kPa)
Figure 3.7 Critical state evolution with suction in ( q − p′) plane for (a) Jossigny silt (Cui and Delage
1996) and (b) trois rivières silt (Maatouk and Leroueil 1995) , reported by Khalili et al. (2004).
1 The effective representation of critical state is analysed on two more materials in part 3.8.1
EFFECTIVE STRESS 79
400
400
(a) (b)
350
Deviatoric stress q (kPa) 350

300
300
250
250
200
Saturated
200 CSL
150
150
100 s = 300 kPa 100 s = 300 kPa
s = 200 kPa
Saturated s = 200 kPa
50 s = 100 kPa
CSL 50 s = 100kPa
s = 0 kPa
s = 0 kPa
0
50 100 150 200 250 300 0
0 100 200 300 400 500 600
Mean net stress pnet (kPa)
Mean effective stress p' (kPa)
Figure 3.8 Critical state lines for Kaolin at different suctions – experimental data from Sivakumar
(1993).
2000
2000
(a) (b)
1500
1500
Suction level (kPa) 40
200 200
50 Saturated 50 Saturated
280
CSL 280 CSL
1000
1000
75
75
55
500 60 s > 0 kPa
114 500 55 s > 0 kPa
60
s = 100 kPa 114
s = 100 kPa
s = 0 kPa s = 0 kPa
0
0 400 800 1200 1600 0
0 400 800 1200 1600
Figure 3.9 Critical state line for Sion silt at different suctions. Experimental data from Geiser et al.
(2006).
3.6.2 Unsaturated mechanical compression

Detractive arguments regarding an effective stress framework concern the apparent
difficulty to evaluate the stress level and then to interpret results in subsequent effective
planes. However, even if the stress variables are not directly the experimentally controlled
ones, that are net stress and suction, the elected generalized effective stress is a rather simple
combination of them, with the parallel retention information. Nevertheless, data sets lacking
from continuous hydraulic monitoring (that is knowledge of water content, volumetric strain
and matric suction) are obviously excluded from the proposed effective stress interpretation.
Fig. 3.10 indicates that the increase in compressibility and preconsolidation pressure with
suction, observed in the experimental net stress mechanical planes, Fig. 3.10a are recovered
in the effective stress interpretation, Fig. 3.10b. On the whole, those two suction effects
appear to be amplified with the proposed generalized effective stress approach.
80 CHAPTER 3
0.02
0.02
(a) (b)
0
0
-0.02
-0.02

-0.04
-0.04
-0.06
-0.06
-0.08 s = 0 kPa -0.08 s = 0 kPa

s = 100 kPa
s = 100 kPa
-0.1 s = 200 kPa
-0.1 s = 200 kPa
s = 300 kPa
s = 300 kPa
-0.12
20 40 60 80100 300 500 -0.12
20 40 60 80100 300 500
Figure 3.10 Isotropic compressions of kaolin at different suctions, experimental data from Sivakumar
(1993).
1.4
1.3
1.2
Void ratio e(-)
1.1
1
s = 0 kPa
s = 100 kPa
0.9
s = 200 kPa
s = 300 kPa
0.8
40 60 80 100 300 500
Figure 3.11 Normal compression lines at constant suction in the plane void ratio versus mean effective
stress (experimental data from Sivakumar, 1993), after Gallipoli et al. (2003).
Gallipoli et al. (2003) plotted a similar interpretation with the Bishop’s generalized
effective stress, interpreting slightly curved shapes for compressions rather than lines, Fig.
3.11. However, Fig. 3.11 shows that a linear model could fit well the compression lines in the
plastic part; and that inflexions of the experimental curves, if some, are likely to appear only
at high mean effective stresses or might not be significant.
Fig. 3.10 also enables the determination of the Loading Collapse (LC) yield curve, which
original definition is due to Alonso et al. (1990) in the matric suction versus mean net stress
plane. This yield limit accounting for the increase in preconsolidation pressure pc with
suction is a reversible function that reflects the suction hardening, see Fig. 3.12. As expected
from previous observations, the shape of the LC curve in the effective stress interpretation is
similar to the reference one, but neatly amplified. A significant consequence is that
analogous mathematical formulations for the function pc = pˆ c ( s ) could be applied to both
net and effective stress analyses.
EFFECTIVE STRESS 81
4 105 Net stress Effective stress

interpretation interpretation

3 105
5
2 10
5
1 10
s
e
0
0 P
c0
1 105 2 105 3 105
Preconsolidation stress p or p' (Pa)
c-net c
Figure 3.12 Shape of the LC curve in two interpretations, experimental data from Sivakumar (1993).
Concerning the saturated domain, defined by s < se with se being the air entry suction,
the shape of the yield limit is a simple vertical line in the ( s − pc′ ) plane. The effective stress
interpretation induces this independency of the apparent preconsolidation pressure on
suction, unlike in the ( s − pc − net ) plane. This specificity owns a fundamental use for the
modelling of volume changes upon hydraulic loading paths (Laloui and Nuth 2005). Notice
that the transition between saturated and unsaturated states being not straightforward with
the net stress and suction interpretation, it is certainly incorrect to let appear the saturated
zone ( s ≤ se ) in the ( s − pc − net ) plane2.
3.6.3 Constitutive modelling framework

As elastoplastic aspects will be introduced in the present paragraph, the reader must be
aware that the terminology ‘generalized’ is solely applied to stress variables, and has no
relation with generalized plasticity concept.
Unicity of yield limit

Statements in part 3.3 imply that the effective stress is simply defined as that inducing the
mechanical (elastic) strain of the solid skeleton, Eq. 3.1. Judging on Eq. 3.11, the generalized
effective stress is incremented by the application of any combination of the external
mechanical stress σ ij and the matric suction s . So, any of the two separate or combined loads
induce variations in total volumetric strain εv that is equivalent to a pure mechanical
straining of the material, as expected from constitutive equation Eq. 3.1.
On this basis, the mechanical stress state being fully described by the means of the unified
stress, plasticity mobilisation can be evaluated via the generalized stress state alone. A
relevant consequence of the unified stress approach lies thus in the sufficiency of a single
mechanical yield surface to comprehend any elasto-plastic behaviour resulting in skeleton
deformations. So, for the mechanical stress-strain behaviour, no second yield surface in
suction is required, setting the model formulation free from any kind of Suction Increase (SI)
2 The unsaturated mechanical compression of 3 other materials is assessed with the generalized effective stress
in part 3.8.2
82 CHAPTER 3
6
1 10
possible
5 net stress yield limits
8 10 interpretation

6 105
5
4 10
5
2 10
0 effective stress
interpretation
-2 105
0 4 104 8 104 1.2 10 5
Mean stress p or p' (Pa)
net
Figure 3.13 Drying path in two interpretations on Sion Silt, experimental data from Geiser et al. (2006).
or Suction Decrease (SD) yield curves, that are proper to constitutive models written in terms
of independent stress variables such as Barcelona Basic Model family (Alonso et al. 1990) or
others (Laloui et al. 2001).
In the end, the Loading Collapse curve should include a hydro-mechanical coupling,
which leads to the following formulation:
pc′ = pˆ c′ (ζ , s ) ; ∀s (3.49)
where the term ζ gathers the saturated hardening variables. It can be remarked that the
expression for LC yield curve is continuous from the saturated state to the unsaturated state.
Visualizing the stress paths3 in the matric suction versus mean stresses planes help
understanding better the unified stress framework implications, Fig. 3.13. Whereas a drying
path under constant net stress is a simple vertical line in the ( s − pnet ) plane, the effective
stress formulation confers a curved shape in the ( s − p′ ) plane. By reference to Fig. 3.12, the
effective drying path evidences a possible direct yielding on the LC curve, whereas the net
stress interpretation shows the need for a supplementary pseudo-horizontal yield limit.
A questionable issue remains the fact that the curvature of the drying path in the effective
stress conception is narrowly linked to the soil water retention curve shape. Indeed, it has
been seen that the product of s times S r in the formulation of the effective stress confers to it
a non linear evolution upon suction changes. But, upon drying, once a certain limit in suction
reached, called sCR as critical suction, the S r s product tends to decrease, resulting in an
inflexion in the stress path representation (plane ( s − p′ ) ) and a decrease in the effective
stress. However, this inconsistency may appear only at very high suctions, where the degree
of saturation drops down to its residual value close to zero, see Fig. 3.14. Gray and Schrefler
(2001) for instance related this aspect to the residual state of saturation, where water phase is
described as a film coating the particles rather than under the form of a connected reservoir.
3 The representation and interpretation of effective stress path for 2 more materials are presented in part 3.8.3.
EFFECTIVE STRESS 83
5
1.2 8 10
(a) (b)
7 105
1
5
6 10

r
0.8
5 105
s
CR
0.6 4 105
5
3 10
0.4
2 105
0.2
1 105
0 0
1 100 104 s 10
6
0 1 104 2 104 3 104
CR
Matric suction s (Pa) Mean effective stress p' (Pa)
Figure 3.14 Characterisation of the critical suction (a) Soil water retention curve – drying path (b)
corresponding stress path in ( s − p′ ) plane. Experimental data from Fleureau et al. (1993).
Then, a modified water pressure can be introduced to correct the effective stress tensor.
Consequently, a particular attention must thus be paid for interpreting this limit state with
the generalized effective stress proposed here.
Hydraulic hysteresis
A unique feature conveyed by Bishop’s generalized effective stress is the taking in
account of no less than two effects of the hydro-mechanical coupling. First, as developed
previously, any increase in suction is likely to provoke a direct change in the effective stress,
Eq. 3.46. Secondly, the defined stress framework enables a natural repercussion of the
hydraulic hysteresis observed in ( Sr − s ) plane into ( s − p′ ) representation. Fig.3.15a plots a
typical shape of a soil water retention curve in the conventional ( Sr − s ) plane. A particular
attention must be paid to the drying curve being not recovered upon wetting, letting appear
a dissipation characterising the hydraulic hysteresis. The two limiting curves are usually
termed main drying curve and main wetting curve. Notice however that different
intermediary paths are possible within the space bounded by the main wetting and drying
curve. Obviously, the consequence of the hydraulic hysteresis is that for a same level of
suction, two different states of saturation can be obtained according to the suction variation
in progress. Although this has nothing to do directly with suction induced hardening
mentioned previously, or bonding, the stress state and skeleton deformations are all the
same likely to be influenced by the hydraulic irreversibility. Fig. 3.15b plots the same
complete drying wetting path in the ( s − p′ ) plane. Due to the inclusion of product of suction
by degree of saturation inside the effective stress formulation, the hydraulic hysteresis is
naturally featured. In consequence, even with the same yield limit, plastic states could be
reached sooner or later according to the process applied, which might result in relevant
simplifications in the constitutive formulation.
Indeed, this natural repercussion of the soil water retention curve shape can be obtained
only if the mechanical stress variable accounts for both suction and degree of saturation,
which is verified only in the category 2 of effective stresses. Other stress frameworks, for
instance with effective stress based on suction ratio, have to be combined with advanced
coupled elastoplastic contexts in order to approach hydraulic hysteresis effects within the
mechanical part of the model.
84 CHAPTER 3
6
1 10
(a) (b)
105
0.8
drying

r
104
0.6 wetting
wetting
1000
drying
0.4
100
0.2
10
0 1
1 100 104 106 1 100 104
Matric suction s (Pa) Mean effective stress p' (Pa)
Figure 3.15 (a) Soil water retention curve of Jossigny loam (after Fleureau et al. (1993)).
(b) Corresponding stress paths in ( s − p′ ) plane.
3.7 Conclusion
The historical developments of the effective stresses have been reviewed with the
objective of determining a proper stress framework for constitutive modelling of unsaturated
soils. Starting from the saturated state, it has been shown, that under the assumption of
incompressibility of grains and fluid, Terzaghi’s effective stress (Eq. 3.4) is the unique stress
state variable.
The extension of the effective stress principle to unsaturated soils is possible; however, the
terminology of effective stress is quite misleading provided that a second stress variable is
always necessary to fully describe the soil behaviour. Work input characterization, among
other methods enable the determination of possible combinations of stresses to use. It is not
compulsory for the two stresses to be independent. No matter the stress framework adopted,
it should be used within a complete elastoplastic framework, particularly for modelling
suction hardening effects.
Similarly to Gens (1995) it has been proposed to classify the possible stresses into three
types: the independent stress approach, the category 1 effective stresses and the category 2
effective stresses. Table 3.1 sums up the major capabilities and consequences for each stress
framework, studying the following points:
• The ease of representation of the stress paths, linked to the complexity of stress
variables
• The ability to shift from saturated to unsaturated states
• The inclusion of hydraulic couplings in the stresses and effects of hydraulic hysteresis
• The ability to model directly increase in strength
The Bishop’s generalized effective stress has been adopted as the single mechanical stress.
Porosity is voluntarily not included in the stresses and strains variables in order to obtain
directly the soil water retention curve S r = Sr ( s ) for the hydraulic part of the behaviour. The
major implications of the proposed stress framework have been investigated on the basis of
experimental data. Notably, the critical state is assumed unique for any suction level, and a
single mechanical yield surface is sufficient. Other aspects like curvature of the stress path in
EFFECTIVE STRESS 85
Table 3.1 Comparative study of stress frameworks

Saturated- Hysteresis & Direct accounting
Category Representation unsaturated hydraulic of increase in
transition effects strength
Independent
stress variables
+ - - -
Category 1
effective stresses
- + - +
Category 2
effective stresses
- + + +
( s − p′) upon hydraulic loading as well as accounting for the hydraulic hysteresis are also of
interest.
3.8 Complementary data

As mentioned in the foreword to the chapter, the following analyses are not featured in
the published journal paper. Three main representations are investigated below more in
depth with respect to generalized effective stress analysis; namely the critical state, the
compressibility and the stress paths. The results are of interest as they confirm the trends
previously observed.
3.8.1 Uniqueness of critical state line

In complement to data from Cui and Delage (1996), Maatouk (1995), Sivakumar (1993),
Khalili et al. (2004) and Geiser (1999) previously cited, two other sets of data on shear
resistance from literature are re-interpreted with the generalized effective stress. Rampino et
al. (2000) determined the critical state lines of Metramo silty sand under three constant levels
of suction between 0 and 300 kPa. The critical state lines are originally determined in
deviatoric stress versus mean net stress plane (Fig. 3.16a). Again, a common slope M could
be identified for each suction level, indicating that the friction angle remains constant. The
slight deviation of unsaturated critical state lines with respect to the saturated line is reduced
by the use of mean generalized effective stress (Fig. 3.16b). It is deduced from this second
( q − p′ ) plane that the assumption of a unique critical state line whatever the level of suction
is valid for this material.
Toll and Ong (2003) published the points of critical state for a sandy clay under 14
different suctions. These points are reported in Fig. 3.17a in the conventional deviatoric stress
versus mean net stress plane. Only the saturated critical state line is drawn, since only one
point was determined for each non zero suction. The conversion to generalized effective
stress, Fig 3.17b, does not entail here the expected alignment of points on the saturated
critical state line. The points of critical state in unsaturated conditions are indeed shifted
towards a higher mean stress. It seems from figure 3.17b that the saturated friction angle
would overestimate most of the experimental points. An explanation for the scattering could
be found in a statement from the authors mentioning that “a true critical state was rarely
achieved by the end of the constant water content tests”. Most of the tests at non zero suction
were apparently still featuring changes in deviator stress beyond the plotted arbitrary critical
state points. As the points plotted in Fig. 3.17 are taken directly from Toll and Ong (2003) and
could not be reinterpreted on the present study, the original uncertainty might have a
significant influence on the final interpretation with generalized effective stress.
86 CHAPTER 3
2500
2500
(a) Saturated (b) Saturated
CSL CSL
2000
2000

1500
1500
1000
1000
s =0 kPa
s = 0 kPa
500 s = 100 kPa
500 s = 100 kPa
s = 200 kPa
s = 200 kPa
s = 300 kPa
s = 300 kPa
0
0 200 400 600 800 1000 1200 1400 1600 0
0 200 400 600 800 1000 1200 1400 1600
Figure 3.16 Critical state line for Metramo silty sand at different suctions. Data from Rampino et al.
(2000).
700
241 700
600 311 600 311

500 45
171
282 500 45
282
168 171
400 178
44 400 44 178
182
73 182
300 292 73
85 300 292
158 85
Saturated 158
200 186
158 CSL 200 186
158
100
100 Saturated
(a) CSL (b)
0
0 100 200 300 400 500 600 700 0
0 100 200 300 400 500 600 700
Mean effective stress p'(kPa)
Figure 3.17 Critical state points for a sandy clay. Data from Toll and Ong (2003).
3.8.2 Mechanical compression under unsaturated states

The purpose of this paragraph is to provide more support to part 3.6.2 dealing with the
plotting of volume change as a function of generalized effective stress. In anticipation to the
formulation of the constitutive model in next chapter, four data sets are assessed hereafter
with the generalized effective stress interpretation. In agreement with the previous
conclusions, this paragraph enables (i) to validate the assumption of linear normal
compression lines in plane void ratio versus mean effective stress and (ii) to draw the trends
for the evolution of effective compressibility with suction.
Kane (1973) obtained early results for the mechanical compression of a loess under
different suction levels, with information on void ratio and water content which enables to
deduce the degree of saturation. The volumetric response with respect to vertical net stress
and vertical effective stress is plotted in Fig. 3.18. The results expressed in terms of logarithm
of generalized effective stress (Fig. 3.18b) are consistent with the pattern observed in part
3.6.2. In the elastoplastic domain, the compression curves can be reasonably fitted with a
EFFECTIVE STRESS 87
0.9 0.9
(a) (b)
0.8 0.8
0.7 0.7
Void ratio e(-)

Void ratio e(-)
0.6 S =0.93 0.6 Sr0=0.93

r0
Sr0=0.70 Sr0=0.70
0.5 S =0.54 0.5 Sr0=0.54
r0
Sr0=0.43 S =0.43
r0
0.4 Sr0=0.31 0.4 Sr0=0.31
Sr0=0.15 S =0.15
r0
0.3 0.3
10 100 1000 10 100 1000
Vertical net stress σvnet (PSI) Vertical effective stress σ' (PSI)
v
Figure 3.18 Isotropic compression of loess, starting from different initial degrees of saturation Sr 0 .
Experimental data from Kane (1973).
2.3
2.3
(a)
s = 100 kPa (b)
s = 100 kPa
T3
2.2 T3
2.2
Specific volume v(-)
T1 T1
T4 T4
2.1 T5
2.1 T5
2
2
s = 0 kPa
s = 0 kPa
1.9
1.9
s = 200 kPa
s = 200 kPa
1.8
100 1000 1.8
100 1000
Figure 3.19 Consolidation lines of a speswhite kaolin under different levels of suction (with several
tests for suction 100 kPa) . Experimental data from Zakaria (1994).
linear model and trend for the slope change is similar to that in the net stress interpretation
(Fig. 3.18a). The reinterpretation with generalized effective stress of unsaturated
consolidation of speswhite kaolin (Zakaria 1994) are in agreement with those assumptions
(Fig. 3.19).
Using Cam-clay terminology (Wroth and Schofield, 1968), the slope λ of the normal
compression line in plane ( e − ln pnet ) and ( e − ln p′ ) is deduced from the previous figures.
The results for three materials, namely loess (Kane 1973), speswhite kaolin (Zakaria 1994)
and compacted kaolin (Josa et al. 1988) are reinterpreted in Figs. 3.20 to 3.22.
88 CHAPTER 3
0.18
0.16
0.14
Plastic slope
0.12
0.1
Net
0.08 Effective
0 20 40 60 80 100 120 140 160

Matric suction s(PSI)
Figure 3.20 Evolution of plastic slope (determined in plane ( e − ln σ v ) planes) with suction.
Experimental data from Kane (1973).
0.2 Net
Effective
0.18
Plastic modulus λ
0.16
0.14
0.12
0.1
0.08
0 50 100 150 200 250

Figure 3.21 Evolution of plastic modulus with suction for speswhite kaolin. Experimental data from
Zakaria (1994).
Two conclusions can be drawn from Figs 3.20 to 3.22. Firstly, there is not a unique trend
for the evolution of the plastic modulus with suction, as it appears either to increase (Fig.
3.20), decrease (Fig. 3.22) or to reach a maximum before decreasing at high suctions (Fig.
3.21). From the constitutive modelling point of view, the plastic modulus is significantly
dependent on suction and its variation might be different for each material (either positive or
negative). Secondly, the interpretation of stress-strain response in terms of generalized
effective stress shows modified plastic moduli, often higher than in the net stress
representation, but still with an evolution with suction that remains consistent with the
reference net one.
EFFECTIVE STRESS 89
0.11
0.1
Plastic modulus λ
0.09
0.08
0.07
0.06
Net
Effective
0.05
0 0.02 0.04 0.06 0.08 0.1
Matric suction s(MPa)
Figure 3.22 Evolution of plastic modulus with suction for kaolin. Experimental data from Josa (1988).
3.8.3 Generalized effective stress paths

Following the discussion from paragraph 3.6.3, it is proposed to investigate further the
issue of the effective stress paths. In the particular case where suction is modified while the
net stress is kept constant, variations in effective stress are calculated from changes in suction
and degree of saturation only (Eq. 3.38). The plane suction versus mean generalized effective
stress ( s − p′ ) is used hereafter to plot the stress path. While providing a good insight on the
variations of effective stress, this plane of representation is misleading and will used with
care in the other chapters. Indeed, a stress path should be represented in a plane of stresses
(or invariants of stresses). As the matric suction is not a mechanical stress variable, it is
practically not correct to plot the mechanical stress path in ( s − p′ ) plane. This subject is
discussed more in detail in chapter 4.
The stress paths for a complete cycle of drying and wetting on white clay have been
interpreted from the works of Zerhouni (1991). The soil water retention curve (Fig. 3.23a) has
an influence on the effective stress path plotted in Fig. 3.23b. In particular, the capillary
hysteresis in plane ( Sr − ln s ) entails a loop of the stress path in plane ( s − p′ ) .
In the case where successive wetting and drying cycles are applied to a mix of bentonite
and kaolin (Sharma 1998), the water retention curve evidences not only a main drying and a
main wetting curve but also scanning lines. The irreversibility observed in the degree of
saturation versus suction plane (Fig. 3.24a) is automatically featured in the stress path (Fig.
3.24b). In other words, a given value of suction may correspond to several degrees of
saturation and also to several mean effective stresses.
90 CHAPTER 3
4
1 1.4 10
(a) (b)
1.2 10 4
0.8
r

1 104
0.6
8 103
6 103
0.4
4 10
3 Drying
Wetting
0.2
Drying 2 103
Wetting
0 0
0.1 1 10 102 103 104 105 106 0 5 102 1 103 1.5 103 2 103 2.5 10 3
Matric suction s(kPa) Mean effective stress p'(kPa)
Figure 3.23 Drying-wetting cycle on white clay. Data from Zerhouni (1991).
1.2
(a) (b) A
Drying 4 105 C
1 E E
r
D
B Drying
0.8 Wetting
Wetting
C
0.6 Drying
Wetting A 2 105
0.4
0.2 B
D
0 3 0
10 104 105 106 0 2 105 4 105
Matric suction s(Pa) Mean effective stress p'(kPa)
Figure 3.24 Double wetting-drying cycle on a mix of bentonite and kaolin. Data from Sharma (1998).
3.9 References
3.9.1 Publications from the author
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, pp. 771-801.
Nuth M., Laloui L. (2007) "Unified stress framework for modelling unsaturated subsoil behaviour",
International journal of Road materials and Pavement Design, vol. 8(4), p. 767-781.
Nuth M., Laloui L. (2007) "Implications of a generalized effective stress on the constitutive modelling
of unsaturated soils", In: Theoretical and numerical unsaturated soils mechanics - Weimar, p. 75-82,
2007
3.9.2 Other references

Aitchison G. D. (1960). "Relationships of moisture sterss and effective stress functions in unsaturated
soils." In: Pore pressure and suction in soils, London, Butterworths.
EFFECTIVE STRESS 91
40(3), pp. 405-430.
Biot M. A. (1955). "Theory of elasticity and consolidation for a porous anisotropic soil." Journal of
Applied Physics 26(2), pp. 182-185.
Bishop A., Donald I. (1961). "The experimental study of partly saturated soil in the triaxial apparatus."
5th Int. Conf. on Soil Mechanics and Fundation Engineering, Paris, pp. 13-21.
Bishop A. W. (1959). "The principle of effective stress." Tecnisk Ukeblad 39, pp. 859-863.
Bishop A. W., Alpan I., Blight G. E., Donald I. B. (1960). "Factors controlling the strength of partly
saturated cohesive soils." Conference Shear Strength Cohesive Soils, American Society of Civil
Engineers, pp. 503-532.
Coleman J. D. (1962). "Stress/strain relations for partly saturated soils." Géotechnique
12(Correspondence), pp. 348-350.
Coussy O. (2004). Poromechanics, Wiley.
Cui Y., Delage P. (1996). "Yielding and plastic behaviour of anunsaturated compacted silt."
Dangla P., Coussy O. (2002). Approche énergétique du comportement des sols non saturés. In:
Mécanique des sols non saturés. O. Coussy and J.-M. Fleureau, Hermes science publications.
Fredlund D. G., Morgenstern N. R. (1977). "Stress state variables for unsaturated soils." Journal of the
geotechnical engineering division, ASCE 103(GT5), pp. 447-466.
Fredlund D. G., Morgenstern N. R., Widger R. A. (1978). "Shear-Strength of Unsaturated Soils."
Canadian Geotechnical Journal 15(3), pp. 313-321.
Geiser F., Laloui L., Vulliet L. (2000). "Modelling the behaviour of unsaturated silt." Experimental
Geiser F., Laloui L., Vulliet L. (2006). "Elasto-plasticity of unsaturated soils: laboratory test results on a
remoulded silt." Soils and Foundations Journal 46(5).
Gens A. (1995). "Constitutive modelling: Application to compacted soils." Unsaturated Soils, Paris,
Balkema, pp. 1179-1200.
Gray W., Schrefler B. (2001). "Thermodynamic approach to effective stress in partially saturated
porous media." European Journal of Mechanics A-Solids 20(4), pp. 521-538.
Hassanizadeh S. M., Gray W. G. (1990). "Mechanics And Thermodynamics Of Multiphase Flow In
Porous-Media Including Interphase Boundaries." Advances in Water Resources 13(4), pp. 169-
186.
193-196.
Hutter K., Laloui L., Vulliet L. (1999). "Thermodynamically based mixture models of saturated and
unsaturated soils." Mechanics of cohesive-frictional materials 4, pp. 295-338.
Jennings J. E. (1960). "A revised effective stress law for use in the prediction of the behaviour of
unsaturated soils." Pore pressure and suction in soils, London, Butterworths, pp. 26-30.
92 CHAPTER 3
Jennings J. E. B., Burland J. B. (1962). "Limitations to the use of effective stresses in partly saturated
soils." Géotechnique 12, pp. 125-144.
Jommi C., Di Prisco C. (1994). "Un semplice approcio teorico per la modellazione del comportamento
meccanico di terreni granulari parcialmente saturi." Conf. Il ruolo dei fluidi nei problemi di
ingegneria geotecnica, Mondovi, pp. 167-188.
Josa A. (1988). Un modelo elastoplastico para suelos no saturados. Barcelona, Universitat Politecnica
de Catalunya. PhD. Thesis.
Kane H. (1973). "Confined compression of loess." 7 th ICSMFE, Moscow, pp. 155-122.
Khalili N., Geiser F., Blight G. E. (2004). "Effective stress in unsaturated soils: review with new
evidence." International journal of geomechanics 4(2), pp. 115-126.
Khalili N., Khabbaz M. H. (1998). "A Unique Relationship for x for The Determination of The Shear
Strength of Unsaturated Soils." Géotechnique 48(5), pp. 681-687.
Khalili N., Witt R., Laloui L., Vulliet L., Koliji A. (2005). "Effective stress in double porous media with
two immiscible fluids." Geophysical Research Letters 32(15), L 15309.
Kohgo Y., Nakano M., Miyazaki T. (1993). "Theoretical aspects of constitutive modelling for
unsaturated soils." Soils and Foundations 33(4), pp. 49-63.
Laloui L., Geiser F., Vulliet L. (2001). "Constitutive modelling of unsaturated soils." Revue française de
génie civil 5(6), pp. 797-807.
Laloui L., Klubertanz G., Vulliet L. (2003). "Solid-Liquid-Air Coupling in Multiphase Porous Media."
International Journal For Numerical and Analytical Methods in Geomechanics 27, pp. 183-206.
Laloui L., Nuth M. (2005). "An introduction to the constitutive modelling of unsaturated soils." Revue
Européenne de Génie Civil 9(5-6), pp. 651-669.
Leonards G. A. (1962). "Correspondence." Géotechnique 12, pp. 354-355.
Lu N., Likos W. J. (2006). "Suction stress characteristic curve for unsaturated soil." Journal of
Geotechnical and Geoenvironmental Engineering 132(2), pp. 131-142.
Maâtouk A., Leroueil S., La Rochelle P. (1995). "Yielding and critical state of collapsible unsaturated
silty soil." Géotechnique 45(3), pp. 465-477.
Modaressi A., Abou-Bekr N. (1994). "A unified approach to model the behavior of saturated and
unsaturated soils." Computer methods and advances in geomechanics, Morgantown,
Balkema, pp. 1507-1513.
Muraleetharan K. K., Wei C. F. (1999). "Dynamic behaviour of unsaturated porous media: Governing
equations using the theory of mixtures with interfaces (TMI)." International Journal for
Numerical and Analytical Methods in Geomechanics 23(13), pp. 1579-1608.
Nur A., Byerlee J. (1971). "Exact effective stress law for elastic deformation of rocks with fluids."
Journal of Geophysical Research 76(26), pp. 6414.
Pereira J.-M., Wong H., Dubujet P., Dangla P. (2005). "Adaptation of existing behaviour models to
unsaturated states: Application to CJS model." International journal for numerical and analytical
methods in Geomechanics 29(11), pp. 1127-1155.
Rampino C., Mancuso C., Vinale F. (2000). "Experimental behaviour and modelling of an unsaturated
compacted soil." Canadian Geotechnical Journal 37, pp. 748-763.
Schofield A. N., Wroth C. P. (1968). Critical state soil mechanics, McGraw-Hill.
Schrefler B. A. (1984). The finite element method in soil consolidation (with applications to surface
subsidence), University College of Swansea. PhD. Thesis.
EFFECTIVE STRESS 93
Sharma R. S. (1998). Mechanical behaviour of highly expansive unsaturated clays, Oxford.

Skempton A. W. (1960). "Effective stress in soils, Concrete and Rocks." Pore Pressure and Suction in
Soils, London, Butterworths, pp. 4-16.
Suklje L. (1969). Rheological aspects of Soil Mechanics. New York, Wiley.
Taibi S. (1994). Comportement mécanique et hydraulique des sols soumis à une pression interstitielle
négative. E. C. Paris. Paris, Ecole Centrale Paris.
Tarantino A., Mongiovì L. (2000). "Experimental investigations on the stress variables governing
unsaturated soil behaviour at medium to high degrees of saturation." Experimental Evidence
and Theoretical Approaches in Unsaturated Soils; Proc. of an International Workshop, Trento,
pp. 3-18.
Toll D. G., Ong B. H. (2003). "Critical State Parameters for an Unsaturated Residual Sandy Clay."
Zakaria I. (1994). Yielding of unsaturated soil. University of Sheffield. PhD. Thesis.
Zerhouni M. I. (1991). Rôle de la pression interstitielle négative dans le comportement des sols -
apllication au calcul des routes. Paris, Ecole Centrale Paris. PhD. Thesis.
94 CHAPTER 3
4. Stress-strain model for unsaturated soils
4.1 Foreword
This chapter presents the formulation of a new constitutive model for the stress-strain
behaviour of fine-grained soils in partially saturated conditions. The elasto-plastic model
addresses all the specifications listed in chapter 2 in terms of capillary effects. The contents of
this chapter have been published partly in the journal Computers and Geotechnics under the
title “On the use of the generalized effective stress in the constitutive modelling of
unsaturated soils” (Laloui and Nuth 2008) and in the conference proceedings of Unsat Asia
2007, invited lecture, under the title “New insight into the unified hydro-mechanical
constitutive modelling of unsaturated soils” (Nuth and Laloui 2007). The overall structure of
this chapter is that of the paper for Unsat Asia with updates and modifications, while part
4.4 corresponds exactly to the article from Computers and Geotechnics. Even though the
main topic of this second article is the effective stress, it has been chosen to merge the paper
with this chapter dedicated to the model due to the nature of the conclusions drawn in the
article. Indeed, the purpose of the chapter 3 was to select and validate the stress variables
while here all implications of the stress framework are seen exclusively from the constitutive
point of view.
4.2 Introduction
Suction is seen in this chapter as an average internal fluid pressure, provided that the
unsaturated material exhibits either an overall negative pore water pressure (compression
being positive) or else a distributed gaseous air overpressure. The combination of external
mechanical loads and cyclic variations of water content can modify permanently the volume
fine-grained soils, particularly under high mechanical stresses. The generation of swelling
pressure and wetting pore collapse are the main critical issues addressed in this chapter.
According to the literature review made in chapter 2, most constitutive models for
unsaturated soils published up to now feature the effects of partial saturation over the
mechanical behaviour. Conventionally, the model features that are updated to cope with
suction are preconsolidation stress, stiffness and shear strength. Indeed, the occurrence of
partial saturation in problems of geomechanics raise the need for (i) an elasto-plastic
framework, (ii) the identification of state variables and (iii) accounting for intrinsic hydro-
mechanical couplings. While recent modelling frameworks reviewed from the literature
already show good results their applicability to engineering practice is arguable. Indeed, the
performances of a constitutive model should be balanced with respect to its convenience, in
terms of parameter determination, stress-strain representation and numerical
implementation. Jardine et al. (2004), Nuth and Laloui (2008) assessed in particular the
implications of the modelling stress framework; Sheng et al. (2004) discussed the topic of
implementation of the models in Finite Element codes. All discussions lead to the conclusion
that the stress framework should be a natural extension of the saturated effective stress of
96 CHAPTER 4
Terzaghi (1936) (see previous chapter 3). Meanwhile the determination of material
parameters should lie only on laboratory testing at macroscopic scale. Both conditions would
warranty a general applicability of the model as well as an easier implementation in
numerical codes.
The presented unified constitutive model for unsaturated soils is formulated on the basis
of the generalized effective stress framework defined in the previous chapter. Due to the use
of the generalized effective stress, the nature of hardening variables and the consistency
equation are the object of a preliminary discussion (Laloui and Nuth 2008). The subsequent
mathematical formulation of the model is derived from Hujeux’s elasto-plastic model
(Hujeux 1979, Hujeux 1985). The features of the reference saturated model are thoroughly
reviewed besides the new developments for the partial saturation in water. Among the
principal advantages of the framework, the conventional parameter determination is
appreciated.
This chapter describes only the extension of the reference saturated stress-strain model to
partially saturated states. The capability of the mechanical model to reproduce the most
complex stress and strain trends in unsaturated soils depends not only on the modifications
brought to Hujeux’s model but also on the accuracy of the soil water retention curve. The
choice is made here to use at first the non-linear reversible formulation of Van Genuchten
(1983) for computing the degree of saturation versus suction relationship. The principle is to
have the capillary variables ( S r , s ) determined in a simple way; then these variables are
used as static inputs for the mechanical stress-strain variables and model. The purpose is to
de-couple partially and provisionally the mechanical and retention models and hence to
assess the performances of the mere unsaturated stress-strain model. The added value of the
subsequent double-way coupling is discussed in next chapters. The global structure of the
analysis will be:
(i) Extension of a saturated stress-strain model to partial saturation, effect of suction
and simplified model for the soil water retention curve (this chapter)
(ii) Formulation of a new model for the soil water retention curve, featuring hysteresis
and effect of void ratio (chapter 5)
(iii) Activation of double-way coupling between the stress-strain model (i) and the
advanced retention model (ii) (chapter 6).
Data sets have been selected from literature (chapter 2) to validate the formulated
mechanical model. Attention is paid to the back prediction of the volumetric response of
fine-grained soils to complex combinations of suction variations and mechanical loads. In the
validation process at the scale of the sample, the lateral displacements can be blocked, or a
condition of no volume change can be applied. The type of simulated tests ranges from the
free swelling tests to the isochoric swelling pressure tests.
4.3 Influence of hydro-mechanical field boundary conditions

The major effects of capillarity are known to be the increase in preconsolidation pressure
with suction (Geiser et al. 2006), the suction-induced modifications in rigidity,
compressibility and shear strength (Wheeler and Sivakumar 1995). The volumetric response
of partially saturated soils to mechanical loading is expected to be altered by these effects.
Focusing on dehydration and soaking of soils (Fleureau et al. 1993, Sivakumar 1993, Lloret et
al. 2004), fundamental volumetric and retention behaviours are called into question with
respect to free or totally constrained mechanical boundary conditions. The three main
STRESS-STRAIN MODEL 97
0.1
v
Volumetric strain ε
-0.1
-0.2 Drying
-0.3
Wetting
-0.4
-0.5 2 4 6 8
10 10 10 10
Figure 4.1 Complete drying wetting cycle on white clay, after Fleureau et al. (1993)
categories of conditions on stress and displacements for natural ground in the field are
discussed hereafter.
First of all, the case of a free-swelling soil submitted to a complete drying and wetting
back cycle is of interest. Fleureau et al. (1993) for instance conducted hydric loading tests on
loams and clays, starting from slurries or samples being preconsolidated to 10-50 kPa. Fig.
4.1 is representative of the typical volumetric response to suction variations, evidencing
alternating elastic and plastic transformations. The irreversible process takes place only
along drying and as long as the matric suction remains inferior to air entry suction se (to be
defined later). The sample volume also tends towards an asymptotic value at high suctions.
Secondly, to reproduce a more constrained field case, other laboratory tests make use of a
constant significant level of mean net stress besides wetting towards zero suction, Fig. 4.2.
Unexpectedly, in opposition with observations from previous Fig. 4.1, reversible swelling
and plastic compression are detected along hydration path (Fig. 4.2). The highlighted
behaviour is conventionally termed “swelling collapse” upon wetting or “wetting-induced
pore collapse”. Swelling can also occur subsequently to collapse, for lower ranges of suction.
The third category of mechanical boundary conditions gathers the fully confined case
studies. A complete restriction on displacements imposes that total deformations remain
null. As wetting the soil is alleged to provoke swelling or collapse under free displacement
conditions, the confined soaking generates stresses inner to the cell to prevent straining. A
set of results from this standard procedure called “swelling pressure” tests is given as an
illustration in Fig. 4.3. The generated vertical stress obviously does not hold a linear
dependency on matric suction, and its evolution trend even reverts twice.
Even though these advanced behavioural attributes seem to mobilize complex
mechanisms upon suction variations, they are all believed by authors to be inherited from
the capillary effects. In other words, basic hydro-mechanical coupling features are likely to
be at the source of several phenomena raised during hydric loading-unloading. In fact, the
reviewed behaviours are incontestably consequences of the plastic or elastic nature of the
transformation, involving stress history and stress state. Also, compressibility coefficients
shall play a major role in the volumetric response trends.
98 CHAPTER 4
0.02
Wetting
0.01
Wetting
Volumetric strain ε (-)

0
v
-0.01
Collapse Swelling
-0.02
-0.03
-0.04 5 5
0 2 10 4 10
Figure 4.2 Swelling collapse phenomenon on Kaolin, experimental data from Sivakumar (1993)
1000
SP1
SP2
SP3
Matric suction s (MPa)
SP4
100
10
0.1
0 2 4 6 8 10
Vertical net stress σv (MPa)
Figure 4.3 Swelling pressure tests on FEBEX Bentonite, experimental data from Lloret et al. (2004)
Therefore, on the basis of existing concepts such as the loading collapse yield curve
(Alonso et al. 1990) and an advanced understanding of stress variables (e.g. Sheng et al. 2004,
Nuth and Laloui 2008), is possible to formulate here an adequate constitutive model capable
to account for complex stress paths alternating mechanical and suction loads besides keeping
confined conditions.
4.4 On the use of the generalized effective stress in the constitutive

modelling of unsaturated soils
4.4.1 Introduction
The long-lived debate on the use of stress variables for unsaturated soil constitutive
modelling could be coming to an end thanks to the work of Gens (1995) and Jardine et al.
(2004). These publications review various stress frameworks from conventional ones using
independent stress variables based on experimental results to more complex ones dealing
with averaged pseudo-effective stresses.
The critical assessment of stress contexts also reveals that a large number of them are
consistent from a thermo-dynamical viewpoint (Laloui et al. 2003, Borja 2006). Consequently,
the choice of the stress framework appears to be mostly a matter of convenience; causing
several schools of thought to emerge. For instance the independent stress variables
(Fredlund and Rahardjo 1993) can be preferred for their experimental meaning; or else
complex stress variables accounting for micromechanical structure of the phases
(Pietruszczak and Pande 1996) or interface energy (Coussy 2004) can present strong
advantages. Nevertheless, the aim of the present paper is to focus more on Bishop-type
effective stress, in order to review the historical definition, update it and justify the structure
of a comprehensive stress framework. The comparison of the various possible stress
frameworks for unsaturated soils is then out of the scope and will not be developed further
here.
Judging from preliminary remarks on the limitations of Bishop’s effective stress recently
stated (e.g. Estabragh and Javadi 2007), misunderstandings of the ‘effective stress’
terminology still appear to exist. Several arguments can be found to demonstrate that the so-
called Bishop’s effective stress (Bishop 1959) is unsuitable in the modelling of unsaturated
soils. First of all, the original definition puts the effective stress within a fully elastic context,
making it incompatible with modelling of the swelling-collapse upon wetting provided that
the wetting path invariably reduces the Bishop’s effective stress and hence results in swelling
only. Secondly, the initial definition of Bishop’s effective stress disregards the retention
behaviour and the associated hydro-mechanical couplings, even if the effective stress
parameter χ can be related to the degree of saturation. Finally, the parameter χ has been
called into question, mostly concerning the ambiguity of its definition. Instead, the authors
propose to focus on a particular form of Bishop’s effective stress, namely a single mechanical
variable within an adaptated elasto-plastic framework, later referred to as the generalized
effective stress. The use of this variable will be clarified in this paper. In addition, the
bonding aspects induced by partial saturation are briefly discussed.
Throughout the following discussion, matric suction is defined as a negative intergranular
(pore-) pressure, denoting a tensile isotropic stress when using soil mechanics sign
convention of positive compression.
4.4.2 Differences between Bishop’s effective stress and generalized effective

stress
The terminology of “Bishop’s effective stress” is confusing while dealing with constitutive
modelling of unsaturated soils. The common misuse in literature is to associate the term
‘Bishop’s effective stress’ to both (i) the stress variable σ ij′ Bishop defined by Eq. 4.1 and (ii) the
elementary fully elastic concept implicitly set up by Bishop (1959).
σ ij′ Bishop = σ ij − paδ ij + χ ( pa − pw ) δ ij (4.1)

100 CHAPTER 4
with σ ij being the total external stress tensor, pa , pw the pore air and pore water pressure
respectively, and δ ij the Kronecker’s delta (δ ij = 1 if i = j; δ ij = 0 if i ≠ j ) .
As a result, the shortcomings inherent to the genuine stress variable alone and besides
those linked to the elastic constitutive context are commonly amalgamated into the generic
‘Bishop’s stress’. In order to obtain a correct and consistent use of the term effective stress
also for unsaturated soils, the authors encourage the geotechnical community to abandon
this ‘Bishop’s stress’ terminology. It is proposed to re-define a unified stress variable
anchored into an adapted elasto-plastic context. Such a definition must be of a three level
structure and any valuable comparison between stress formulations should consider all the
levels, these being:
(i) The stress variable is the first element of the structure. It consists in defining a
dependency law for the effective stress on the state stress variables and/or
saturation and/or water content ; for instance one stress variable defined by
Khalili and Khabbaz (1998) is
−0.55
⎛ s⎞
σ ij′ = σ net ij + s. ⎜ ⎟ δ ij (4.2)
⎝ se ⎠
for s > se , σ net ij ( = σ ij − paδ ij ) being the net stress tensor, s ( = pa − pw ) the matric
suction and se the air entry suction.
(ii) The stress-strain framework gathers all the stress and strain variables in a
conjugation that is consistent from the thermo-dynamical and work input point of
view. The stress-strain framework of Sheng et al. (2004) can be used as an
illustration:
⎛ σ ij′ = σ net ij + Sr sδ ij ⎞ ⎛ ε ij ⎞
⎜ ⎟ and ⎜ ⎟ (4.3)
⎝ s = pa − pw ⎠ ⎝ Sr ⎠
S r being the degree of saturation and ε ij the strain tensor.
(iii) The constitutive context introduces the constitutive relations between stresses and
strains with the variables defined in (ii) and in the framework of a theory of
plasticity (e.g. Wheeler et al. 2003).
Denomination: generalized effective stress

It seems appropriate to the authors to designate the following stress (Eq. 4.4), used among
others by Schrefler (1984), as the generalized effective stress.
σ ij′ = σ ij − paδ ij + S r ( pa − pw ) δ ij (4.4)
This stress variable, widely used in up-to-date constitutive modelling of unsaturated soils,
has been given several names, e.g. Bishop’s stress (Bolzon et al. 1996, Tamagnini 2004),
Average soil skeleton stress (Jommi 2000, Wheeler et al. 2003, Gallipoli et al. 2003, Kimoto et al.
2007), Average stress (Sheng et al. 2004).
Naming here formulation Eq. 4.4 Bishop’s stress would be misleading for the reasons
mentioned previously. The principle for saturated and unsaturated soils modelling is to
determine an equivalent mechanical stress for the macroscopic continuum. Terminology
‘generalized’ meaning literally ‘spread throughout a body or system’ is hence suitable. Also,
the term ‘effective’ remains essential, as the macroscopic stress actually governs the
mechanical straining.
Following the introductory discussion, the definition of the stress variable alone is
insufficient. In the same fashion Bishop’s stress is implicitly linked to a fully elastic
framework, the generalized effective stress must be defined within (i) a consistent stress-
strain framework and (ii) a superior constitutive context. The stress-strain framework of Eq.
4.3 is for instance used by Nuth and Laloui (2007) on the basis of the generalized effective
stress. Secondly, the elasto-plastic constitutive context makes use of a particular concept for
plasticity, for instance hardening plasticity (Nuth and Laloui 2007) or generalized plasticity
(Bolzon et al. 1996).
Generalized effective stress versus Bishop’s effective stress.

In the following, a point by point comparison of the complete two contexts previously
defined is proposed. Starting from the elementary sublevel of the stress framework structure,
the genuine stress variable formulations open the comparative analysis. The unique
difference at the stress tensor level is the effective stress parameter. Using the degree of
saturation in the generalized effective stress unifies the effective stress parameter
determination for any material. It also brings a superior approach by introducing a built-in
coupling between the retention properties and the mechanical generalized effective stress.
Additionally, the complete constitutive contexts respectively defined around Bishop’s
stress and generalized effective stress can be confronted. The generalized context should
principally be an extensive improvement of the previous basic one. Indeed, the main
additions are the theory of plasticity, the accounting for the retention behaviour (by the use
of the degree of saturation), and the consistent inclusion of two-way hydro-mechanical
coupling. The capillary Sr − s constitutive equation is described in parallel to the mechanical
model, which involves an advanced and realistic model for the retention curve making use
for instance of kinematic hardening and capillary hysteresis. Unlike Bishop’s stress within an
elastic framework, the generalized context is entirely suitable for modelling swelling collapse
(Gallipoli et al. 2003, Laloui and Nuth 2005).
4.4.3 Is suction a hardening parameter?
Capillary forces and bonding effects

It is a common knowledge that the emergence of a second fluid phase, namely air, in a
porous medium initially saturated with water creates menisci. It is also often reported that
the surface tension effects combined with the negative pore water pressure tend to
strengthen the material in a similar mode to that of cementing. The comparison between
hydraulic bonding and cementation is however incorrect given that these phenomena are not
analogous from an effective stress viewpoint. Indeed, cementation is characterised by the
formation of interparticle bonds due to the deposition or precipitation of cementing agent
around and at particle contact (Fernandez and Santamarina 2001). Consequently, the
strength of cemented materials depends on both the strength of the grains and the strength
of the cemented agent. The overall resistance is thus no more a characteristic intrinsic to the
material alone so that cohesion is non null in effective stress planes. By opposition, using a
proper generalized effective stress representation, capillary forces have been shown not to
affect the critical state line (Nuth and Laloui 2008). It is suggested to abandon ‘bonding-
debonding’ terms because of their residual state connotation, and attribute the effects of
partial saturation to the ‘capillary forces’ only.
102 CHAPTER 4
‘Suction hardening’ effects ?

The actual effects of partial saturation should not either be termed ‘suction-induced
positive hardening’ (Casini et al. 2007) provided that the phenomenon is not compatible with
the mere definition of hardening. Hardening means that the yield surface changes in size,
location and/or shape with the loading history (Yu 2006), so that this particular change in
the yield surface is only due to an irreversible (dissipative) phenomenon, related to a given
plastic work, for instance the plastic rearrangement of particles. In reality, capillary effects
are not depending on the hydraulic loading history but on the saturation state and are
reversible upon saturation changes (without plastic work). In other words, if the
preconsolidation pressure increases with suction, it also decreases back upon suction
reduction.
Consequently, the shape of the yield surface evolves with the level of suction in a
reversible manner, very comparably to the way it changes with deviator stress q . Fig. 4.4
shows how the yield loci are represented in two stress planes, suction versus mean effective
stress p′ and deviator stress versus mean effective stress, respectively.
Obviously, for different values of deviator stress q1 and q2 , the respective yield pressures
p1′ and p2′ are different, but this does not correspond to hardening process between the two
values of deviator stress. Analogous shape evolution for yield limit is observable upon
suction change without mobilisation of hardening between s1 and s2 , so that suction is not a
hardening parameter. In other words, the size of the yield surface can enlarge with suction
even when the stress state is located in the elastic region.
Consistency condition
To illustrate the consistency condition, let us consider the simple case of a generic form of
mechanical yield surface with isotropic hardening. It could be written as follows:
f = f ( state variables, hardening parameters ) (4.5)
s D ry in g p a th s (a ) q S h e a rin g p a th s (b )
Y ie ld lim it f
Y ie ld lim it f
L o a d in g p a th s
s 3
q 1
s 2 q 2
q
s 1
3
p 'c 1 p 'c 2 p 'c 3 p ' p '1 p '2 p '3 p '

Figure 4.4 (a) stress path and yield surface in (s-p’) plane (b) stress path and yield surface in (q-p’)
plane, pc′ is the preconsolidation pressure.
The mechanical stress state is represented by the generalized effective stress alone σ ij′ .
The hardening parameters are primarily the plastic strain ε ijp and secondly other possible
hardening parameters gathered under the generic term ξ . In other words, Eq. 4.5 is
equivalent to f (σ ′ , ε p , ξ ) . Then, Prager’s consistency condition requires:
ij ij
df = ∂σ ij′ fdσ ij′ + ∂ ε p fd ε ijp + ∂ξ fd ξ = 0 (4.6)

ij
It has been demonstrated in the previous section that suction is not considerable as a
hardening parameter. Suction shall consequently not belong to variable ξ . It is therefore
unnecessary either to introduce an independent term ∂ s fds into the increment of yield limit
(Eq. 4.6) for the following three reasons:
• Any variation in generalized effective stress due to suction changes is inherently
accounted for by the third term on the right hand side of Eq. 4.4. This is a consequence
of suction being not a state parameter on its own.
• There is no possibility to vary suction alone keeping a constant effective stress and
producing a plastic strain.
• The relation governing the growth in preconsolidation pressure with suction level is
only a ‘shape’ equation, the latter being valuable upon both suction increase and
suction decrease.
4.4.4 Discussion on advantages due to the use of generalized effective stress.

Gens (1995), followed by Khalili et al. (2004) and Nuth and Laloui (2008) critically
compared the various possible stress frameworks for constitutive modelling of unsaturated
soils, providing the reader a significant assistance in decision making. The generalized
effective stress was indeed noticed to present an interesting number of advantages.
Principally, the inclusion of product S r × s in the effective stress formulation makes the
hydro-mechanical coupling straightforward, featuring a direct modification in effective
stress upon suction change and the accounting of capillary hysteresis on the mechanical
behaviour. Furthermore, simplifications appear in model formulations due to the genuine
application of effective stress principle. As a result, the effective critical state line is unified
whatever the suction and the unique yield limit in ( s − p′ ) plane is suitable for all loading
paths. Also, the transition between saturated description (Terzaghi’s effective stress (1943)
and partially saturated states is conventionally recognized as straightforward.
Furthermore, provided that the generalized effective stress is a relatively basic
combination of experimentally controlled stress variables and degree of saturation, the
determination of the variable remains simple anytime. However, focusing on the complete
constitutive context, the efficiency of the intrinsic hydro-mechanical coupling aspects
previously discussed is highly dependent on the capillary part of the model that is the
retention behaviour. A proper description of S r and s thus raises the need for an advanced
water retention model besides the generalized mechanical framework.
4.4.5 Conclusion
The definition and use of the generalized effective stress variable for unsaturated soils
have been clarified. Provided that the generalized stress is used within a proper constitutive
context owning couplings, plasticity and realistic retention model, the generalization of
effective stress principle brings a number of advantages linked to the uniqueness of the
mechanical yield surface in ( s − p′ ) plane.
104 CHAPTER 4
In the continuation of many conceptual analyses from the past fifteen years, authors
encourage the geotechnical community to take advantage of advances brought by the
generalisation of the effective stress to describe the mechanical behaviour of unsaturated
soils.
4.5 ACMEG-s: an elasto-plastic constitutive model

4.5.1 Overview of the model formulation
The usual laboratory experimental loading of soils partially saturated in water may be of
two natures (see chapter 2): (i) a mechanical path is followed when varying the exterior stress
at a constant level of suction (paths 2 and 3 in Fig. 4.5) and (ii) a suction loading path
corresponds to a suction variation in a constant net stress plane, such as path 1 in Fig. 4.5.
Both are likely to generate either skeleton straining or variations in saturation or both. In the
present chapter, the hydro-mechanical coupling is implemented only partially into the
model. In other words, the retention curve can influence the stress-strain relationship while
the contrary is not studied in this chapter. Consequently, the only volumetric response
studied hereafter is the skeleton strain ε ; the variations of degree of saturation S r due to
hydro-mechanical couplings are discussed more in detail in next chapter.
Hereafter is developed the mathematical formulation of the Advanced Constitutive
Model for Environmental Geomechanics, unsaturated extension, abbreviated to ACMEG-s.
In essence, the model’s construction is based on the following characteristics (Fig. 4.6):
(i) The advanced stresses and strains framework is that of Eq. 4.3 on the basis of
previous discussion
(ii) The skeleton strain ε is basically related to the generalized effective stress
σ ′ using conventional elasto-plastic constitutive theories. This relation is called
from now on the mechanical part of the model. The reference saturated stress-strain
model is that of Hujeux (1985).
(iii) Besides, the distribution of fluids (saturation) and its evolution are expressed
anytime as a function of interstitial fluids’ pressures. The retention part is therefore
describing the evolution in degree of saturation S r with suction s through the soil
water retention curve.
(iv) The mechanical and retention sections are coupled together via the generalized
effective stress and key parameters such as preconsolidation pressure in order to
reproduce alterations in hardening process, volumetric response with respect to
fluids proportions.
4.5.2 Mechanical constitutive framework for unsaturated soils

Within the so-called mechanical part that is the focus of this chapter, the only mechanical
stress state variable is the single generalized effective stress. The terminology ‘mechanical’
refers to skeleton straining in response to changes in generalized effective stress.
Nevertheless, those changes result from any combination of increments in suction and net
stress. The standard elasto-plastic decomposition of strain increment d ε ij gives:
d ε ij = d ε ije + d ε ijp (4.7)

s= c q
o n st
3
4
2 B
s 1 p n e t
t
c o n s A
=
p n et
Figure 4.5 Possible stress loading paths to be considered for unsaturated behaviour.
Figure 4.6 Global structure of ACMEG-s model.
where d ε ije is the elastic strain increment and d ε ijp the plastic strain increment.
Superscripts ‘e’ and ‘p’ will respectively keep on describing elastic and plastic deformations
in the text. The elastic strain increment is governed by Eq. 3.1 seen in previous chapter in
direct relationship with the generalized effective stress. In the following, subscripts ‘v’ and
‘d’ are respectively referring to the volumetric and deviatoric parts of strain:
εV = ∑ ε ii = tr (ε ) (4.8)
i
106 CHAPTER 4
2 1/ 2
εd = ⎡⎣(ε11 − ε 22 ) 2 + (ε 22 − ε 33 ) 2 +(ε 33 − ε11 ) 2 ⎤⎦ (4.9)
3
Introducing the elastic parameters K (bulk elastic modulus) and G (shear elastic
modulus), the stress-strain incremental relations within the elastic domain are non linear:
dp′
d εVe = (4.10)
K
ne
⎛ p′ ⎞
with K = K ref ⎜⎜ ⎟ (4.11)
′ ⎟⎠
⎝ pref
dq
d ε de = (4.12)
3G
ne
⎛ p′ ⎞
with G = Gref ⎜ ⎟⎟ (4.13)
⎜ pref
⎝ ′ ⎠
K ref , Gref and ne are material parameters, defined respectively as the bulk elastic
′ , the reference shear elastic modulus at pref
modulus at a reference mean stress pref ′ , and the
non linearity fitting exponent, n e remaining within the range between 0 and 1. Even though
Eqs. 4.11 and 4.13 do not let appear an explicit dependency on suction, the elastic parameters
K and G are coupled to the soil water retention curve via the generalized effective stress (Eq.
4.4). It was estimated from part 2.4 in chapter 2 that no other modification of the reference
elastic moduli K ref and Gref with respect to suction would be justified. Hence, apart from
the use of unsaturated effective stress, Eqns. 4.10 to 4.13 are strictly identical to the reversible
stress-strain relationships defined in the saturated Hujeux model.
The elastic domain is bounded by the yield surface, and like in the reference saturated
model the concept of multiple mechanisms of plasticity is applied. By definition, each
mechanism ( mec ), corresponding to a type or direction of loading, is defined with an
individual yield surface f mec , a plastic potential g mec and hardening variables α mech . A
mechanism is activated if the point of stress σ is located on the yield surface f mec and if the
variation of stress dσ goes outside of the surface. Several mechanisms of hardening can be
activated simultaneously under the condition that the stress point belongs to all the
corresponding yield surfaces (Fig. 4.7). The total plastic deformation is calculated by
summing the plastic deformations at all active mechanisms. In the particular case of the
model of Hujeux, two mechanisms of plasticity are defined, one isotropic ( iso ) and one
deviatoric ( dev ). The deviatoric yield surface f dev and the isotropic yield limit fiso are drawn
in Fig. 4.8, using the conventional triaxial plane completed with matric suction along the
third axis. The formulation is such that the mechanisms are coupled together, which means
that if hardening occurs on one of the mechanisms, the yield limit of the other mechanism
will also be modified.
Figure 4.7 Multiple mechanisms of plasticity: yield limits and activation (after Michalski and Rahma
1989).
q
L
CS
M
L
CS
se (A)
(B) p’c0 p’
e
ur v
s
LC c
p’CR
p’c
Figure 4.8 Shape of the yield surface and critical state line (CSL) in ( p′, q, s ) space.
Irreversible behaviour of the soil gives birth to a volumetric plastic strain ε vp following
the normal compression line in plane (ε v
p
)
− ln p′ which slope is defined by that of the
critical state line:
108 CHAPTER 4
Figure 4.9 Volumetric stress-strain plane: representation of normal consolidation and critical state line.
Figure 4.10 Graphical representation of the degree of mobilization rmec of a given mechanism mec .
′
pCR
log = βε vp (4.14)
′ 0
pCR
′ 0 and β are respectively the actual and initial critical state pressures, and
′ , pCR
where pCR
the coefficient of compressibility, to be discussed again later. In the same plane, critical state
line and saturated normal compression line are parallel and distant by ln d ( d is a soil
parameter), Fig. 4.9. So, the preconsolidation pressure p 'c verifies:
′
p 'c = d . pCR (4.15)
The first mechanism of plasticity is isotropic; the associated yield locus is defined as:
( )
fiso p′, ε vp , riso = p′ − riso . pc′ (4.16)
The variable riso is linked to the isotropic mechanism and ranges between 0 and 1. riso is a
measure of the degree of plasticization with respect to the state of perfect plasticity. In other
words, it indicates how close the actual position of the actual yield surface is from its limit
position (Fig. 4.10). The main purpose of riso , called “degree of mobilisation of the isotropic
mechanism” is to provide a gradual change from elastic to elastoplastic deformations. Fig.
4.11 compares the model response in the volumetric stress-strain plane with and without
using riso ; it can be deduced that the progressive slope change provides a more accurate
fitting of the real deformations due to progressive plastic rearrangement of particles.
Provided, as detailed later, that the preconsolidation pressure is already suction
dependent it would have been repetitive to set riso to depend on the capillary actions.
Moreover, no experimental evidence clearly accounts for any direct influence of suction on
the suddenness of changes from elastic to plastic domains.
e
Introducing c and riso as material constants, and ε vp ,iso as the part of the volumetric plastic
strain induced by the isotropic mechanism, the degree of mobilisation is written:
ε vp ,iso
riso = risoe + (4.17)
c + ε vp ,iso
The preconsolidation pressure is then seen as a key variable to introduce capillary effects
into the mechanical model. In other words, the preconsolidation stress has to grow with
suction (see experimental points in plane ( s − p′ ) reviewed from chapter 2 in Fig. 4.12) but
decreases back upon suction reduction in a reversible fashion. By analogy with temperature
effects on apparent preconsolidation pressure, a fitting equation inspired by Laloui and
Cekerevac (2003) is entirely suitable to quantification capillary effects on the size of the
elastic domain:
0.01
ACMEG-s
e
r =1
0 iso
v
-0.01
-0.02
ACMEG-s
e
r =0.001
iso
-0.03
Exp. data
-0.04 5
10
Mean effective stress p'(Pa)
Figure 4.11 Effect of the degree of mobilization of the isotropic mechanism riso . Experimental points
from Geiser (1999).
110 CHAPTER 4
p c′ ( s ) = pc′0 for 0 < s < se

⎡ ⎛ s ⎞⎤ (4.18)
p c′ ( s ) = pc′0 ⎢1 + γ s log ⎜ ⎟ ⎥ for s > se
⎢⎣ ⎝ se ⎠ ⎥⎦
pc′0 is the initial preconsolidation pressure at zero suction, γ s is a material parameter,

and se is the air entry suction, beyond which the degree of saturation becomes inferior to 1.
Satisfactory fitting of experimental points for p c′ ( s ) is obtained for several fine grained soils
(see Fig. 4.12 and Appendices B,C). Although using a different stress framework, Alonso et
al. (1990) already defined this yield locus as the Loading Collapse (LC) curve. The same
terminology is adopted for ACMEG-s. Terzaghi’s effective stress being recovered for
saturated states, the definition of the Loading Collapse curve adequately covers the full
range of saturation states. In the saturated domain ( s < se , S r = 1) the pore space remains
filled with a single liquid fluid, so that the capillary mechanisms are not activated; a positive
suction may take place without affecting the preconsolidation pressure that remains equal to
its saturated reference value (Eq. 4.18) in ACMEG-s.
Following the discussion from part 4.4, suction is not considered as a hardening
parameter. Indeed, Eq. 4.18 does not define a true yielding or hardening upon suction load,
but rather a shape relationship. The hardening variables in the present case are ε vp and riso
only. Suction is not either an independent mechanical state parameter due to the use of the
generalized effective stress as a single mechanical stress variable. Consequently, suction does
not appear explicitly in the consistency condition:
∂f iso ∂f ∂f
df iso = dσ ij′ + iso driso + isop d ε vp = 0 (4.19)
∂σ ij′ ∂riso ∂ε v
The flow rule for the isotropic mechanism is associated (with respect to the mean effective
pressure reference system only), the plastic flow being written:
5
4 10
Exp.
ACMEG-s
5
3 10
5
2 10
5
1 10
s
e
p'
c0
0 5 5 5 5 6
2 10 4 10 6 10 8 10 1 10
Preconsolidation pressure p'c (Pa)
Figure 4.12 Model for Loading Collapse yield curve – Experimental points from Geiser (1999).
⎧ λisop
∂f ⎪ if i = j
d ε ijp ,iso = λisop iso = ⎨ 3 (4.20)
∂σ ij′ ⎪
⎩0 if i ≠ j
The plastic multiplier of the isotropic mechanism λiso

p
is:
1
λisop = dp′ (4.21)
H iso
⎛ (1 − riso )2 ⎞
H iso ′ ⎜
= d . pCR + β riso ⎟ is called the hardening modulus of the mechanical
⎜ c ⎟
⎝ ⎠
isotropic mechanism.
Starting from a saturated reference value β 0 at zero suction, the plastic compressibility
coefficient β (Eq. 4.14) is evolving as a function of suction to reproduce changes in slopes of
consolidation lines at different suction levels (Fig. 4.13).
β = β ( s) = β 0 + Ω.s (4.22)
Ω is a material parameter which sign is proper to the modelled material. It has been
chosen to use a linear relationship for the evolution of β on the basis of the database
reviewed in chapter 2. Indeed, the simple Eq. 4.22 proved to be accurate for the range of
suctions and materials investigated here.
The second plastic mechanism is coupled to the isotropic one through the hardening
variable ε vp = ε vp ,iso + ε vp , dev ; ε vp ,dev is the part of the volumetric plastic strain induced by the
deviatoric mechanism. The formal expression for deviatoric yield surface is:
0.01
Exp.
s=100kPa
Exp.
0 s=200kPa
-0.01
-0.02 ACMEG-s
Exp.
s=0
-0.03
-0.04 5 5 6
4 10 7 10 10
Figure 4.13 Modelling the normal compression line slope changes with respect to suction. The
evolution of preconsolidation pressure with suction is activated, but the degree of mobilization of the
isotropic mechanism is not used to improve legibility. Experimental data from Geiser (1999).
112 CHAPTER 4
⎛ p′ ⎞
fdev ( p′, q, rdev , ε vp , ε dp ) = q − Mp′ ⎜1 − b log ⎟r (4.23)
⎝ ′ ⎠ dev
pCR
Taking advantage of the unification of the critical state for any suction due to the use of
the generalized effective stress (see chapter 3), the critical state line is fixed and defined by
Cam-Clay model (Roscoe and Burland, 1968). M is then the slope of the unique critical state
line in plane ( q − p′ ) .
q = Mp′ (4.24)
6sin φ ′
And, using the effective friction angle φ ′ , M = .
3 − sin φ ′
Again, the mobilisation of the deviatoric mechanism is more or less marked thanks to the
use of the degree of mobilisation of the deviatoric mechanism rdev :
ε dp ,dev
rdev = rdev
e
+ (4.25)
a + ε dp
e
with rdev and a being material parameters. ε dp is the deviatoric plastic strain, solely due
to yielding on deviatoric plastic mechanism. Here the flow rule is non-associated and the
dilatancy law writes:
d ε vp ,dev ⎛ q ⎞
=α ⎜M − ⎟ (4.26)
dε d p
⎝ p'⎠
where α is a material parameter. The plastic flow for the deviatoric mechanism is written:
1 ⎡ ∂q 1 ⎛ q ⎞⎤
d ε ijp ,dev = λdev
p
⎢ + α ⎜ M − ⎟⎥ (4.27)
Mp′ ⎣⎢ ∂σ ij′ 3 ⎝ p′ ⎠ ⎦⎥
where λdev
p
is the plastic multiplier of the deviatoric mechanism and is determined to
satisfy the consistency condition:
⎡ ∂f dev ∂f dev ∂f ⎤
λdev
p
⎢ dσ ij′ + drdev + devp d ε vp ⎥ = 0 (4.28)
⎣⎢ ∂σ ij ∂rdev ∂ε v ⎦⎥
⎧ ∂f p′ ⎞ (1 − rdev )
2
⎛
⎪ dev
′
drdev = − Mp ⎜1 − b log ⎟ d ε dp
⎪ ∂rdev ⎝ ′ ⎠
pCR a
With ⎨ (4.29)
p′ ⎞ (1 − rdev ) p
2
⎪∂f dev ⎛
⎪ ∂r drdev = − ⎜1 − b log p′ ⎟ λdev = − H d λdev
p
⎩ dev ⎝ CR ⎠ a
⎧ ∂f dev
⎪ ∂ε p ,dev d ε v
p , dev
= − Mp′bβ rdev d ε vp ,dev
⎪ v
And ⎨ (4.30)
⎪ ∂f dev ε p ,dev = −bβ ⎛ M − q ⎞ r λ p = − H λ p
⎜ ⎟ dev dev
⎩⎪ ∂ε v p′ ⎠
p , dev v V dev
⎝
1 ∂f dev 1 ∂f dev
So λdev
p
= dσ ij′ = dσ ij′ (4.31)
H d + HV ∂σ ij′ H D ∂σ ij′
where H D is the hardening modulus of the deviatoric mechanism

Fig. 4.8 shows the intercept of the two plastic mechanisms. Basically, the size of the
deviatoric Cam-Clay like ellipse is governed by the coupling with the isotropic mechanism,
itself ruled by the preconsolidation pressure pc′ . Consequently, Eq. 4.18 does not only affect
the size of the elastic domain in the isotropic conditions, but it also directly pilots the shape
of the deviatoric yield limit. In Fig. 4.8, a first domain (A) is identified in the ( s − p′ ) plane;
with the suction remaining inferior to the air entry value, the behaviour is of fully saturated
type. In the second zone (B), the capillary effects modify the preconsolidation pressure
resulting in a larger elastic domain.
4.5.3 Provisional model for the Soil Water Retention Curve

Besides the fact that the formulated mechanical part needs information on retention
variables ( s, S r ) anytime, the stress framework Eq. 4.3 assigns the retention part of ACMEG-s
to relate the degree of saturation to suction state. The purposes of the water retention model
are consequently first to predict the state of saturation of the material and secondly to be able
to calculate the generalized mechanical effective stress itself. The scope of the chapter being
principally to assess the capabilities of unified mechanical constitutive framework, the
retention part of the model is voluntarily simplified to Van Genuchten (1980) water retention
model. This fully reversible non linear formulation (each suction corresponds to a unique
degree of saturation) enable to provide the necessary inputs ( s, S r ) to the stress strain model.
mr
⎛ 1 ⎞
Sr = ⎜ ⎟ (4.32)
⎜ 1 + (α .s )ns ⎟
⎝ s ⎠
with α s , ns and mr being material parameters. The shape of the computed Soil Water
Retention Curve (SWRC) is drawn for instance in Figs. 4.14a and 4.18c in next section. The
compatibility of the stress-strain part of ACMEG-s with such simplified retention models is
of interest, in particular for the implementation in numerical codes where the stress-strain
model might be distinct from the retention curve. The shape function for the
preconsolidation pressure with suction (Eq. 4.18) requires the value of the air entry suction
se . In the present chapter the air entry suction is only defined as a material parameter. This
parameter is theoretically obtained via the experimental ( S r − s ) points. Consequently, any
model for the retention curve featuring already the air entry value as a parameter would
optimize the parameter determination as se would be defined only once for both the
retention and the stress-strain models. On the other hand, it will be demonstrated in chapter
5 that the air entry suction is indeed a variable.
114 CHAPTER 4
4.5.4 Conclusion on the formulation: review of parameters

The formulated model ACMEG-s takes thus advantage of the advanced features of the
reference multi-mechanism elasto-plastic model as well as the natural extension of effective
stress to unsaturated soils. The suction and degree of saturation have been seen only as
inputs to calculate the stress state and draw the shape of the yield limits of the isotropic and
deviatoric mechanisms.
Even though the extension of Hujeux model to partial saturation require a low number of
extra parameters (3 for the stress-strain model), ACMEG-s inherits the relatively high
number of material constants from the reference model (i.e. 12) This number of parameter is
necessary to reproduce the complex physical processes involved in soils. Table 4.1 recalls the
parameters of the reference model (Hujeux 1985), the added parameters for the effects of
suction on the stress-strain model in ACMEG-s and the parameters for the model of the
retention curve.
e e
In practice, parameters rdev and riso are taken very small which lower the number of
parameters to 16. The parameter determination process is provided in Appendices B, C, D
for several test materials.
4.6 Validation of the mechanical part of ACMEG-s

The object of the following assessment of ACMEG-s is to mobilize hydro-mechanically
coupled processes to produce innovative simulations of the behaviour of unsaturated soils.
The shape of the loading collapse curve, combined with the use of the generalized effective
stress, is shown to play a pre-eminent role in the unified modelling of complex stress paths.
Table 4.1. List of parameters of ACMEG-s

Symbol Description Typical value
K ref Bulk modulus 100 MPa
Elastic
parameters
Gref Shear modulus 50 MPa
e
n Elastic exponent 0 to 1
Reference φ′ Friction angle 30 °
stress- β0 Compressibility coefficient 30
strain α Dilatancy coefficient 1
model Plastic
parameters a 0.001
(Hujeux
b 0.5
1979)
c 0.01
d 2
Limits of e
rdev 0.01
elastic e
domain riso 0.01
γs Coefficient of LC Curve 1.5
Capillary effects Ω Coefficient of compressibility change 1e-5
se Air entry value 100 kPa
αs 1e-6
Retention model
ns 1
(Van Genuchten 1980)
ms 4
Even though all the predicted results presented below are obtained as a consequence of a
complete parameter determination, only the most pioneering results of back predictions are
communicated. Sets of determined parameters for each simulated material are provided in
Appendices B, C, D.
4.6.1 Isotropic stress paths

In the particular case of the suction variations under a low level of net stress, the
simulated qualitative response of the model is worth being examined in three different stress
and strain planes (Fig. 4.14). For this example, the material parameters of the soil water
retention curve have been first determined to fit experimental points, then the mechanical
stress strain slopes were also calibrated (in plane (ε v − ln p′) ). The initial preconsolidation
pressure is that published in original works. As the mean effective stress is piloted by
changes in both suction s and degree of saturation S r , the modelled soil water retention
1.2
(a)
A B
1 C
E
r
0.8
0.6
0.4
0.2
Exp.
Van Genuchten D
0 5 7 9
1000 10 10 10
Matric suction (Pa)
9
10
(c)
Stress path D A
8
10 Initial LC 0
B Exp.
Final LC Wetting
7 ACMEG-s
10
se C -0.12
6 Drying
10
5 Yielding -0.24
10
zone
B E
4
10 C
Drying -0.36 D
Wetting
1000
A (b) s
E -0.48 e
100 4 6 8 4 6 8
100 10 10 10 100 10 10 10
Mean effective pressure p' (Pa) Matric suction (Pa)
Figure 4.14 Prediction of volumetric response to hydric cycle with a low initial net stress.
Experimental points : White clay (Fleureau et al. 1993).
116 CHAPTER 4
curve plot is of importance (Fig. 4.14a). The mobilization of isotropic plastic mechanism is
assessed only in the ( s − p′ ) stress plane on the basis of the followed stress path (Fig. 4.14b)
and advanced Loading Collapse curve. This is governing the repartition of the elastic and
plastic parts along the stress path. Eventually, the volumetric response ε v is drawn as a
function of matric suction (Fig. 4.14c). This last plot is simply deduced and reinterpreted
from the main ( ε v − p′ ) representation.
In the saturated domain s < se ( se = 1.7 106 taken from the reference), any increase in
suction changes the effective stress by the same amount, such as the loading is equivalent to
an isotropic purely mechanical load. With equality S r = 1 being verified, the volumetric
response to suction change is thus identical in both (ε v − ln p′ ) and (ε v − ln s )
representations. The elastic (path AB and CE) and the elasto-plastic (path BC) behaviour in
plane ( ε v − ln p′ ) is the one of a standard overconsolidated material.
Once the entry value se is reached the preconsolidation pressure is imposed to increase
with suction (Eq. 4.18), faster than the mean effective stress does augment (Fig. 4.14b). The
elastic domain grows, so that the response becomes reversible (path CD) and there is no
more plastic deformation increment. The reversible volumetric stress-strain relation (Eq.
4.10) is always valid and governs a linear path in plane ( ε v − ln p′ ) , whose slope is indicated
in dotted line in Fig. 4.14c for a matter of comparison. The solid line in ( ε v − ln s plane) is
deduced from this dotted line and Eq. 4.4, and reveals non linearity. The correlation with
experimental points along the complete cycle is remarkable.
The swelling collapse behaviour upon soaking is a second inbuilt feature of the model.
The phenomenon corresponds to yielding on the LC curve upon wetting, the stress path
being not straightforward in ( s − p′ ) plane (Fig. 4.15a). The condition for obtaining swelling
collapse behaviour is to set up an initial high level of net stress. A set of experimental results
from Sivakumar (1993) is used here for validation. Fig. 4.15c plots wetting collapse obtained
in laboratory, the same test being carried out three times under identical conditions (see
“Exp 1” to 3). Prior to leading predictions of Fig. 4.15, the water retention curve was fitted
and the normal compression and swelling lines were also calibrated on a standard saturated
compression test. Then the parameters for LC curve are determined via isotropic mechanical
compression tests at three levels of suction (appendix C).
Again, the elastic or elasto-plastic nature of volumetric response keeps on being piloted
by the stress state being or not inside the elastic domain. Eq. 4.18 imposes the apparent
preconsolidation pressure to decrease faster than the mean effective stress upon wetting.
Provided that the initial stress state (point A in Fig. 4.15a) is close enough to the yield locus,
two volumetric responses are predicted (Figs. 4.15b and c); (i) a fully reversible swelling
upon effective stress relief along paths AB and CD and (ii) a plastic compression assorted
with yielding on LC curve along path BC. In Fig. 4.15b, the dotted lines represent the trace of
the volumetric response to virtual mechanical loading paths under the constant levels of
suction of points A and C, for a matter of comparison. The superimposition of numerical and
array of experimental results for kaolin shows discrepancies between the suction levels for
activation of plasticity, mostly attributable to possible inaccuracy in the LC curve
determination. However, the overall volumetric variations are well predicted and the
qualitative trends for alternative swelling and collapsing are reliable.
4.6.2 Deviatoric stress paths

The extension of the saturated ACMEG to moisture variations provides a natural
accounting for the deviatoric states. Due to the suction dependency of the preconsolidation
pressure (Eq. 4.18) the entire deviatoric yield surface grows along a desaturation path. Yet,
the critical state line remains unique whatever the saturation state. Fig. 4.16 shows the model
predictions for a number of deviatoric tests led on slightly overconsolidated Sion silt
samples. The stress path representation in Fig. 4.16a evidences that upon shearing, the yield
limit is reached later for higher suctions resulting in a different repartition of elastic and
plastic strains.
In the presented case study, a brittle behaviour was observed experimentally close to the
critical state. This feature is out of the scope of the model and cannot be predicted with the
present version. Nevertheless, the predicted peak deviatoric stress is satisfactory, and the
6
10 0.03
(a) (b)
A
0.015 B
Wetting
Wetting 0 A
5
10 -0.015
B D
Yielding zone -0.03 C
s
e
C
Wetting path -0.045
Initial LC Wetting path
D Final LC Mech. path
4
10 5
-0.06 5
10 10
Mean effective pressure p' (Pa) Mean effective pressure p' (Pa)
0.02
(c)
B
0.01
A
0
-0.01 Wetting
D
-0.02
C Exp. 1
Exp. 2
-0.03
Exp. 3
ACMEG-s
-0.04 5 5
0 2 10 4 10
Figure 4.15 Simulation of wetting collapse. Experimental points: Kaolin (Sivakumar 1993).
118 CHAPTER 4
6
2 10
(b)
6
1.6 10
Deviator stress q (Pa)

6
1.2 10
5
8 10
Exp. s=0 kPa
ACMEG-s s=0 kPa
5 Exp. s=100 Pa
4 10 ACMEG-s s=100 Pa
Exp. s=200 Pa
ACMEG-s s=200 Pa
0
0 0.15 0.3
Axial strain ε1(-)
0
Exp. s=0 kPa
ACMEG-s s=0 kPa
0.006 Exp. s=100 kPa
ACMEG-s s=100 kPa

Exp. s=200 kPa
ACMEG-s s=200 kPa
0.012
0.018
0.024
(c)
0.03
0 0.15 0.3
Axial strain ε1(-)
Figure 4.16 Simulation of shear tests with ACMEG-s. Experimental points Sion Silt (Geiser et al. 2006).
In figure (a) the deviatoric yield surface is presented in its initial state only.
volumetric behaviour is properly foreseen.
4.6.3 Oedometric conditions

Even though oedometric paths are hardly ever used in conventional validation processes,
they definitely own a major importance in experimental characterisation of unsaturated soil
behaviour, regarding the reduced cost of unsaturated oedometric cell tests versus
unsaturated triaxial or isotropic compression apparatuses.
The numerical integration of the model is set up to cater for zero lateral total strain
condition, so that only the vertical total stress σ v and matric suction are controlled as inputs.
A complex set of stress paths is simulated (Lloret et al. 2004), comprising successively an
equalization to a given level of suction and an oedometric compression at a constant level of
suction (Fig. 4.17a). The main reasons for choosing to model Lloret et al. (2004) experimental
9
10
Test 1 (a) 0.32 (b)
Initial Test 2 Exp. test 1

8
10 point Num. test 1

0.24 Exp. test 3
Num. test 3
Test 3 Exp. test 5
7
10 0.16 Num. test 5
Test 4
0.08 Initial
6
10 point
Final 0 Wetting
5
point Test 5
10
5 7 9
4
10 10
6
10
8 10 10 10
Vertical stress σv (Pa) Matric suction s (Pa)
0.4
(c) Exp. 1
Num. 1
Exp. 2
0.3 Num. 2
Exp. 3
Num. 3
0.2 Exp. 4
Num. 4
Exp. 5
Num. 5
0.1
-0.1 4 5 6 7 8
10 10 10 10 10
Vertical net stress σv(Pa)
Figure 4.17 Back prediction of hydro-mechanical tests under oedometric conditions. Experimental
point: bentonite, (Lloret et al. 2004).
data set is that stress paths in ( s − σ v ) plane actually start from the same initial state, which
eases comparison.
Simulation of the wetting or drying processes from an initial suction of 138 MPa shows to
predict satisfactory volumetric trend and amplitude (Fig. 4.17b). For a matter of legibility,
only 3 representative tests are plotted in Fig. 4.17b. The magnitude of volumetric swelling is
observed to vary depending on the net stress applied during equalization, as evidenced by
the comparison of wetting tests 5 and 1, under a respective vertical net stress of 0.1MPa and
5.1MPa. Even though the global swelling trend is observed upon wetting for all tests,
punctual decrease in ε v is attributed to (i) the occurrence of mechanical compression prior to
or during equalization (vertical paths for test 1 and 3 in Fig. 4.17b) and (ii) the occurrence of
seamless plastic episodes due to the initiation of minor wetting collapse.
120 CHAPTER 4
Subsequent oedometric compression tests (Fig. 4.17c) at constant suctions from 0 (test5) to
500MPa (test1) are also remarkably predicted with the proposed framework, the law for
evolution of compressibility with suction (Eq. 4.22) being particularly reliable. Fig. 4.17c
shows unexpected variations in the vertical net stress during equalization. These are only the
result of numerical effects and are magnified by the scale of the plot.
4.6.4 Swelling pressure

Wetting a material under fully confined conditions causes swelling to be prevented, so
that external stresses have to be generated at the boundaries to compensate volumetric
variations and ensure zero total stress. A very little number of models for interpreting
swelling pressure have been developed up to now, the most reliable being that used by
Lloret et al. (2004). In these works, the conceptual elasto-plastic framework of Alonso et al.
(1999) was employed to model the wetting test of bentonite at a constant volume. Although
promising, the predictions do require a complex parameter determination based on the
quantification of the micro and macrostructural behaviour. It is proposed to use the unified
framework of ACMEG-s to simulate the same confined tests and demonstrate that the
generated stresses are easily predictable with a proper use of hydro-mechanical couplings
and generalized stress framework.
Fig. 4.18a plots the results of a swelling pressure test simulation with ACMEG-s using the
same set of material parameters than for the previous oedometric simulations on a bentonite
(section 4.6.3). The zero volume change prescription upon wetting gives birth to a distinctive
shape for the followed path in the stress planes ( s − p′ ) and ( s − pnet ) . The deduced stress
plane ( s − σ v ) , σ v being the vertical exterior stress, also reflects similarly three zones of
interest (Fig. 4.18b), the repartition of which is linked to the shape of the Loading Collapse
yield curve.
At the initiation of wetting, i.e. in domain A, the process is fully reversible as the stress
state remains within the elastic domain. Since no plasticity threshold is reached (i.e. there is
no increment of plastic strain), only the constitutive Eq. 4.10, which is verified no matter the
type of transformation, is used for calculation. The elastic deformations equal the total
deformations that are imposed null, so that the variations in effective stress are zero.
Nevertheless, the suction variation is experimentally observed to generate “exterior” stress at
boundaries to compensate the swelling that would occur in unconfined conditions (Fig.
4.18b). This phenomenon is in agreement with the unified framework according to which the
net stress is deduced from the generalized effective stress definition:
( ) ( )
Δ σ net ij = Δ σ ij′ − S r sδ ij = −Δ ( S r sδ ij ) (4.33)
Soil water retention curve model thus pilots the non linearity of stress response in
( s − σ v ) plane anytime (Fig. 4.18c). If the wetting path reaches zone B, yielding on the LC
curve is implied. The total deformations remain null but a plastic deformation is generated,
balancing completely the elastic part of the deformation (Eq. 4.7). The occurrence of elasto-
plastic strains provokes a release of the effective stress according to the elasto-plastic
constitutive model, and obviously a new trend for the evolution of the deduced net stress.
Once suction drops down to the air entry value, the ultimate zone C is entered. The shape of
LC curve defines a recovering of elastic state, Eq. 4.10 being always verified, with the
simplification S r = 1 .
1000
SP1 EXP
Effective stress (a) (b)
SP2 EXP
100 Net stress SP3 EXP

SP4 EXP
100 SP1 MOD
Matric suction (MPa)
A SP2 MOD
10 SP3 MOD
SP4 MOD
Yielding zone B
10
Wetting
1
C
Wetting 1
0.1
Initial LC
Final LC
0.01 0.1
1 100 -2 0 2 4 6 8 10 12
Mean stress p', pnet(MPa) Vertical net stress σv (MPa)
1.2
(c)
1
r
0.8
0.6
0.4
0.2
Exp.
Van Genuchten
0
0.1 10 1000
Figure 4.18 (a) (b) Stress responses to swelling pressure tests (c) Soil Water Retention Curve.
Experimental points: bentonite (Lloret et al. 2004)
The qualitative comprehension of experimental results (Fig. 4.18b) shows to be

satisfactory and presents the advantage of using a single framework for any hydro-
mechanical transformation of the material. However, back predictions still present some
quantitative inaccuracies in plane ( s − σ v ) , mostly due to preliminary assumption of the
independency of air entry suction on mechanical state. As a matter of fact, the introduction
of the latter coupling could contribute among others to adjust the size of zone A (Fig. 4.18a)
and modify judiciously the shape of LC curve so that yielding episodes occur slightly
differently.
Furthermore, as the effective and net stresses are interrelated via the degree of saturation
multiplied by suction, the Soil Water Retention Curve exerts an important influence on the
quantification of the swelling pressure effects. Fig. 4.18c, which shows the Van Genuchten
model actually in use, obviously raises the issue of the accuracy of the retention model.
Experimental SWRC being different according to the initial confining conditions (Fig. 4.18c),
122 CHAPTER 4
further hydro-mechanical coupling should contribute to provide a custom shape for ( S r − s )

model for each set of initial conditions. Indeed, experimental observation tends to evidence
(i) the presence of capillary hysteresis along a wetting drying cycle and (ii) a dependency of
the shape of the SWRC on the mechanical state of the sample. These considerations are taken
into account in chapters 5 and 6.
4.7 Conclusions
A unified constitutive framework for unsaturated soils has been formulated. The
specifications were to demonstrate the capabilities of the advanced framework, knowing that
a minimal number of parameters should be added to the list of parameters of the saturated
reference model. In order to motivate the choice of the generalized effective stress as the
single stress variable for the mechanical part of the model, the main implications of the use
of the unsaturated effective stress have been reviewed. The particular form of the consistency
equation and the definition of hardening variables have been clarified. ACMEG-s is
formulated from a Hujeux’s model and takes advantage of the intrinsic couplings brought by
the generalized effective stress. Non linear elasticity, as well as advanced isotropic and
deviatoric plastic mechanisms, enable satisfactory prediction of conventional features of
behaviour of unsaturated soils such as the response to oedometric unsaturated
compressions. Moreover, the custom shape of the loading collapse curve is favorable to
simulate more complex stress paths, such as wetting-drying cycles under constrained or fully
confined tests. A further level of precision in the prediction of volume change and saturation
could be reached by refining the formulation of the soil water retention model.
4.8 References
Nuth M., Laloui L. (2007). "New insight into the unified hydro-mechanical constitutive modelling of
unsaturated soils." Unsat Asia 2007, Nanjing, pp.
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, pp. 771-801.
Nuth M., Laloui L. (2005). “An introduction to the constitutive modelling of unsaturated soils”,
European Journal of Civil Engineering, vol. 9, no 5-6, pp. 651-670.

40(3), pp. 405-430.
Alonso E. E., Gens, A. & Josa, A. (1999). "A unified model for expansive soil behaviour." Seventh
International Conference on Expansive Soils, pp. 24-29.
Bishop A. W. (1959). "The principle of effective stress." Tecnisk Ukeblad 39, pp. 859-863.
Borja R. I. (2006). "On the mechanical energy and effective stress in saturated and unsaturated porous
continua." International journal of solids and structures 43, pp. 1764-1786.
Casini F., Vsallo R., Mancuso C., Desideri A. (2007). "Interpretation of the behaviour of compacted
soils using Cam-Clay extended to unsaturated conditions." In: Mechanics of unsaturated soils,
Weimar, Springer, pp. 29-36.
Coussy O. (2004). Poromechanics, Wiley.

Estabragh A. R., Javadi A., A. (2007). "Shear strength behaviour of unsaturated silty soil."In:
Mechanics of unsaturated soils, Weimar, Springer, pp. 153-159.
Fernandez A. L., Santamarina J. C. (2001). "Effect of cementation on the small-strain parameters of
sands." Canadian Geotechnical Journal 38, pp. 191-199.
modélisation constitutive. Lausanne, EPFL.
remoulded silt." Soils and Foundations Journal 46(5).
Gens A. (1995). "Constitutive modelling: Application to compacted soils." Unsaturated Soils, Paris,
Balkema, pp. 1179-1200.
193-196.
Hujeux J. (1979). Calcul numérique de problèmes de consolidation élastoplastique. Ecole Centrale
Paris. PhD. Thesis.
Hujeux J. (1985). Une loi de comportement pour le chargement cyclique des sols. Génie Parasismique.
Paris, Les éditions de l'E.N.P.C.: 287-353.
Hutter K., Laloui L., Vulliet L. (1999). "Thermodynamically based mixture models of saturated and
unsaturated soils." Mechanics of cohesive-frictional materials 4, pp. 295-338.
Khalili N., Geiser F., Blight G. E. (2004). "Effective stress in unsaturated soils: review with new
evidence." International journal of geomechanics 4(2), pp. 115-126.
Khalili N., Khabbaz M. H. (1998). "A Unique Relationship for x for The Determination of The Shear
Strength of Unsaturated Soils." Géotechnique 48(5), pp. 681-687.
Kimoto S., Oka F., Fushita T., Fujiwaki M. (2007). "A chemo-thermo-mechanically coupled numerical
simulation of the subsurface ground deformations due to methane hydrate dissociation."
Computers and Geotechnics 34(4), pp. 216-228.
Laloui L., Klubertanz G., Vulliet L. (2003). "Solid-Liquid-Air Coupling in Multiphase Porous Media."
International Journal For Numerical and Analytical Methods in Geomechanics 27, pp. 183-206.
Laloui L., Nuth M. (2005). "An introduction to the constitutive modelling of unsaturated soils." Revue
Européenne de Génie Civil 9(5-6), pp. 651-669.
Laloui L., Nuth M. (2008). "On the use of the generalised effective stress in the constitutive modelling
of unsaturated soils." Computers and geotechnics. doi 10.1016/j.compgeo.2008.03.002.
Laloui L. C., C. (2003). "Thermo-plasticity of clays: an isotropic yield mechanism." Computers and
Geotechnics 30, pp. 649-660.
Lloret A., Romero E., Villar M. (2004). FEBEX II Project: Final report on thermo-hydro-mechanical
laboratory tests, ENRESA.
124 CHAPTER 4
Michalski E., Rahma A. (1989). Modélisation du comportement des sols en élastoplasticité: définition
des paramètres des modèles Hujeux-Cyclade et recherche des valeurs des paramètres pour
différents sols. Orléans, BRGM.
Pietruszczak S., Pande G. N. (1996). "Constitutive relations for partially saturated soils containing gas
inclusions." Journal of Geotechnical Engineering-Asce 122(1), pp. 50-59.
Roscoe K., Burland J. (1968). "On the generalized stress-strain behaviour of 'wet' clay." Engineering
Plasticity, Cambridge, pp. 535-609.
Schrefler B. A. (1984). The finite element method in soil consolidation (with applications to surface
subsidence), University College of Swansea. PhD. Thesis.
of unsaturated soils." Soil Sciences Am. Soc.(44), pp. 892-898.
Yu H.-S. (2006). Plasticity and Geotechnics, Springer.
5. A model for the soil water retention curve
5.1 Foreword
Independently from the formulation of the mechanical stress-strain model developed in
the previous chapter, an adequate and consistent model for the soil water retention curve
must be formulated for ACMEG-s. A non-linear reversible model (Van Genuchten 1980) was
used in chapter 4 to quantify the suction and degree of saturation. Those variables are the
inputs needed for the determination of the generalized effective stress state and to quantify
the capillary effects in the stress-strain model. On the basis of a literature review on the
experimental retention curves of deformable media, this chapter presents the formulation of
a new retention model. The model overcomes the limitations of Van Genuchten’s equation
by featuring a capillary hysteresis and an explicit dependency on porosity variation. The
exact contents of this chapter have been published in the journal Computers and Geotechnics
(Nuth and Laloui 2008a); the paper is reported hereafter with its initial structure. The
experimental tasks led during the research work are discussed in the complementary part
5.7.
5.2 Introduction
Understanding properly the soil water retention behaviour is an essential prerequisite for
a comprehensive description of the behaviour of unsaturated soils. The soil water retention
curve (SWRC) is defined as the relation between the degree of saturation S r , equal to the
volume of liquid water over the volume of voids, and the matric suction that is the difference
between pore air pressure and pore water pressure ( s = pa − pw ) . The terminology ‘retention
curve’ is hereafter preferred to that of ‘characteristic curve’. Indeed, characteristic means
intrinsic, and implies that for a given soil, there would exist a unique and peculiar water
retention curve; but it is the objective of this contribution to show that such a curve depends
on the state of the material and is not unique. So, even if the term ‘soil water characteristic
curve’ has been used for some years in geotechnical and agricultural engineering, it appears
as misleading with respect to the physical evidence of the retention features. Also, instead of
using gravimetric water content, the degree of saturation is the variable to be described
hereafter, in order to take advantage of the normalisation inherent to this particular
volumetric ratio. Often decoupled from the early behavioural stress-strain models, soil water
retention curve has gained more consideration in recent constitutive modelling, mostly due
to the nature of stress variables involving direct couplings between effective stresses, degree
of saturation and matric suction (Nuth and Laloui 2008b). Quantifying with accuracy the
retention relationship also improves mechanical and flow modelling, for instance influencing
the estimation of permeability to water and air. In the following are reviewed some
groundbreaking deterministic models, featuring either a fine description of the capillary
hysteresis, or a direct coupling with soil volume information. However, experimental
126 CHAPTER 5
characterisation of degree of saturation versus matric potential evidences that complex

mechanisms are involved during drying-wetting cycle, i.e. not only dissipation in the
retention contour, but also a parallel dependency on the volumetric state and changes. In
other words, for the same level of suction within a soil, one could determine several different
degrees of saturation, according to the suction path being of the wetting or the drying type,
depending on the maximum (or minimum) level of suction previously reached, and as a
function of the void ratio history and current state.
The objective of the paper is thus to underline the utter necessity to account for both
capillary hysteresis and volume dependency together in retention curve models. An
extensive literature review contributes to the quantification of these aspects. To handle more
efficiently each of the superimposed processes involved in the retention curve, it is proposed
to attempt de-coupling each of them from the others. The analysis of the hysteresis will thus
precede the void-ratio dependency hereafter, and an intrinsic typical shape will be extracted.
The separate ingredients of the soil water retention curve are then integrated into an
advanced unified modelling framework.
5.3 Capillary hysteresis

5.3.1 Conceptual remarks
The hysteretic nature of the soil water retention curve is at the centre of numerous
discussions and experimental works and is agreed to be typical to porous media constituted
of interconnected pores of variable sizes. In spite of an early identification of the occurrence
of capillary hysteresis, modern conceptualization around the dissipations in soil water
retention curves were lately initiated by Fredlund and Rahardjo (1993), Kato et al. (1995)
among others. Focusing on the envelope of the retention curve (Fig. 5.1), a dedicated
terminology is currently in use in geomechanics. The array of ( s, S r ) points is noticed to be
bounded by two main contours; these are named after the corresponding hydraulic
processes. Provided that the highest saturation degrees are covered by a drying path (solid
thick line in Fig. 5.1), the upper outline is called main drying curve (e.g. Pham et al. 2003),
driest curve (Li 2005) or else primary drying curve (e.g. Wheeler et al. 2003). By analogy, the
lower boundary is known as main wetting curve (dotted line in Fig. 5.1). Those two
Figure 5.1 Idealised representation of capillary hysteresis in the soil water retention curve.
MODEL FOR SOIL WATER RETENTION CURVE 127
Figure 5.2 Analogy between dissipations in retention (a) and mechanical consolidation elastoplasticity
(b). e is the void ratio, p′ is the mean effective stress, pc′ is the preconsolidation pressure.
boundary curves are usually observed to possess a similar shape and define the maximum
size of the capillary hysteresis.
Any intermediary point located between the main drying and wetting curves belongs to
one scanning line. Quantified by Romero (1999) among others, the scanning lines can be a
consequence of a suction reversal to a value lower than the maximum suction previously
reached by the material upon drying. Isolating the main drying curve and a number of
scanning lines from Fig. 5.1, it clearly appears in Fig. 5.2a that the material possesses a
memory of the highest suction ever reached. This accounts for the existence of a threshold in
suction below which the changes in degree of saturation are reversible, and above which
larger irreversible transformations in saturation degree are combined with the shifting of the
suction threshold. An evident analogy between mechanical consolidation (Fig. 5.2b) and
retention curve is drawn by comparing the normal compression line with the main drying
curve. Consequently, an elasto-plastic framework would fit adequately the changes in degree
of saturation, as already proposed by Wheeler et al. (2003) and Li (2005). Besides this basic
analogy, the behaviour upon important wetting is also of interest, as another threshold is
detected in Fig. 5.1, point B. Again, the material keeps a memory of the smallest suction
previously sustained on wetting. The domain of reversibility is thus bounded by two given
curves, requiring advanced hardening features in modelling framework. In this context, a
new model based on kinematic hardening is formulated later in part 5.5.
It can be also remarked that models for retention curves based on non linear elasticity,
despite describing possibly the capillary hysteresis (e.g. Pham et al. 2003) would fail to
reproduce the scanning lines. However, accounting properly for these unloading-reloading
lines is essential in modelling complex suction cycles, particularly when residual states of
saturation are not reached upon drying before wetting back.
Along the scanning lines (Figs. 5.1 and 5.2), internal loops can be observed. Yet these
loops are little quantified experimentally and given the low variations in degree of saturation
they imply, they might be conveniently neglected in a conceptual approach. This feature is
another evidence of the non linear nature of the material behaviour, and is usually observed
similarly along mechanical unloading-reloading lines. This indicates that if a refining of the
scanning lines is needed, concepts applied to stress-strain modelling (e.g. bounding surface)
could be possibly extended to retention curve modelling.
128 CHAPTER 5
5.3.2 Influence of capillary hysteresis

A comprehensive description of the behaviour of soils partially saturated in water
requires the understanding of (i) retention properties (ii) stress strain behaviour (iii)
water/air flow. All three specific behaviours are known to be coupled together, and in
particular, the stress strain relationships are explicitly suction-dependent. Recent discussions
on stress frameworks for unsaturated soils (Nuth and Laloui 2008b, Wheeler et al. 2003.
Gallipoli et al. 2003, Jardine et al. 2004) propose to scale down the analysis to the pore scale
to understand the physical implications of saturation. The water retention curve shows that
for a same net stress (defined as the difference between total stress and pore air pressure)
and same amount of matric suction, a specimen on a drying path will own a higher degree of
saturation than a sample along a wetting path. Given that a difference in the degree of
saturation induces a different repartition of water in the soil pores, modified proportions of
bulk water and meniscus water are encountered according to the suction change being on
the wetting or on the drying path.
It is agreed that suction within the meniscus part of the pore water acts only on the forces
at interparticle contacts whereas suction in bulk water is seen as an overall (isotropic) water
pressure deficiency (Wheeler et al. 2003). The skeleton effective stress could thus be
identified as a function of external stresses, interstitial fluid pressures or suction, and
repartition of pore water or degree of saturation (Nuth and Laloui 2008b). In other words,
whenever a capillary hysteresis is identified in a soil water retention curve, it might
significantly affect the state of effective stress within the soil.
5.4 Dependency of retention on mechanical behaviour

5.4.1 Intrinsic shape of the retention curve
Fundamental descriptions of soil water retention curve evidence four main domains of
saturation, namely the complete saturation in liquid water, the funicular state, the pendular
state and the residual state (Fig. 5.3, after Vanapalli et al. (1999)). Along the suction axis, the
boundaries for each of these states are depending on (i) the number of fluid phases filling the
porous space and (ii) the continuous nature of each phase. For instance, the funicular state is
defined over the range of suction for which the soil pores are saturated with liquid water and
gaseous air, under the condition of continuous water phase and continuous gas phase. The
likeliness of having air breaking into to the porous medium at a first step and becoming a
continuous phase at a second step is obviously depending on pore sizes. Indeed, the
saturation phases are narrowly linked to the radii of menisci formed throughout the
medium, as a function of radii of pores themselves (Wheeler and Karube 1995). These
fundamental considerations imply that for a soil with a given granulometry, the repartition
of saturated, funicular, pendular and residual states is different depending on the initial
porosity. Consequently, the whole shape of the soil water retention curve, may it feature an
hysteresis or not, is invariably linked to the porosity of the soil. In the following are reviewed
the most innovative experimental retention curves featuring void ratio information from the
literature. The purpose is to contribute to an effective qualification of the SWRC shape under
different initial compaction states.
Only little evidence of the true intrinsic outline of retention curve is directly available. In
fact, due to the permanent influence of the mechanical state (porosity) of the soil on the
( Sr − s ) relationship, it is a major experimental endeavour to determine the mere retention
aspects alone, that is the retention capacity with constant porosity. In other words, in any
real deformable soil, the shape of the retention curve is anytime linked with both retention
Figure 5.3 Schematic stages of saturation in soil water retention curve. se is the air entry value, S RES is
the residual degree of saturation.
properties and void ratio. For instance, Ng and Pang (2000) explicitly highlighted the volume
change issues when determining the soil water retention curves of a sandy silt and clay
mixture using a modified pressure plate extractor. They proposed to verify the no volume
change assumption throughout the drying wetting processes under different applied stress
states. It was concluded that the conventional assumption of no volume change in pressure
plate test would lead to underestimate the water content at the end of a drying-wetting test.
The study thus confirms the occurrence of non-negligible volume change upon suction
variations.
Based on an extensive literature review, it will be shown that for a given soil, the soil
water retention curve (including drying and wetting paths) owns a typical shape that could
be defined as the intrinsic outline of the ( S r − s ) curve in the material, assuming that no
deformation occurs throughout the suction cycle. The principle is thus to try and uncouple
the effect of mechanical state on retention curve to recover a simplified shape. Later, analysis
on void ratio will intend quantifying the coupled effects between stress-strain and retention.
Vanapalli et al. (1996, 1999) published pioneering results of retention curves of a clayey
silt under different compaction states. The experimental data of major interest are redrawn in
Fig. 5.4. In the present study, the void ratios have been back calculated from the original
work; the obtained volumes for each of the three drying curves were verified to vary less
than 1% of the respective reference values. The covered range of suction and saturation
evidently prevents from observing any sort of residual states and do not show the states of
full saturation in water. Nevertheless, the repartition of available experimental points in
Figure 4 is such that the shape of retention curves is very similar for all void ratios, the
likeness being particularly true for void ratios 0.56 and 0.58. Moreover it appears that the
process of compacting or reducing void ratio is causing an overall shifting of the retention
curve towards higher suctions.
130 CHAPTER 5
100
90
Degree of saturation S (%)

r
80
70
60
50 e = 0.58
e = 0.56
e = 0.52
40
10 1000
Figure 5.4 Soil water retention curves of compacted clay under different constant void ratios
interpretation of data from Vanapalli et al. (1999).
0.9
0.8
r
0.7
0.6
0.5
0.4
e = 0.78
0.3 e = 0.71
e = 0.63
0.2 4
1 100 10
Figure 5.5 Intrinsic shape of soil water retention curves for loam: ( S r : ln s ) relationship at constant
void ratios (Data from Sugii et al. (2002)).
More efforts were recently made to understand the shape of the soil water retention curve
for undeformable soils, in the contribution of Sugii et al. (2002). The principle of their
analysis is to reconstruct virtual soil water retention curves at constant void ratio. To do so, a
large number of water retention curves for a deformable soil has to be processed. Then the
idealized no-volume change retention curve is drawn by isolating all points at a given fixed
void ratio, and recovering the degree of saturation and suction data. Fig. 5.5 gives an
application of the method on loam of Lausebrink Colluvium. It appears that an intrinsic
simple shape is common for each of the void ratios within the range of study, with an array
of quasi-identical slopes. The soil water retention curves at lower void ratios (0.63 to 0.71)
appear to be determined by a simple translation from the parallel reference curve at a void
ratio equal to 0.78.
On the basis of the extensive experimental determination of soil water retention curve of
Boom clay from Romero (1999), the main drying curves are extracted at three levels of initial
void ratio (Fig. 5.6). Whereas Fig. 5.6 already evidences a similar shape for all retention
curves, it is not clear yet if there is a true intrinsic shape for each void ratio. The issue here is
that the material is supposedly deformable, so that the void ratio changes may alter the
shape of the retention curve. In the case where the initial void ratio is higher (1.46), the shape
of the curve is quite noticeably different from that of the two other tests. This change in the
shape is tentatively attributed to changes in soil volume during drying, that are larger for
e0 = 1.46 than for the other cases. In order to improve visibility and set the basis for a
conceptual framework, the constancy of void ratios is verified. On the basis of water content
and degree of saturation, the volume information (available for two retention curves only) is
100
80
r
60
40
20 initial e=0.59
initial e=0.93
initial e=1.46
0
0.001 0.1 10 1000
Figure 5.6 Main drying curves at three void ratios, after Romero (1999).
initial e = .59
initial e=0.93
1
e = 0.92 +/- 3%
0.9
Void ratio e (-)
0.8
0.7
e = 0.64 +/- 3%
0.6
e = 0.59 +/- 3%
0.5
0.001 0.1 10 1000
Figure 5.7 Estimated evolution of void ratios with suction, interpretation of data from Romero (1999).
132 CHAPTER 5
100

80
r
60
40
20 e = 0.59
e = 0.64
e = 0.93
0
0.001 0.1 10 1000
Figure 5.8 Main drying curves at three constant levels of void ratio, interpretation of data from
Romero (1999).
plotted as a function of matric suction (Fig. 5.7). The larger scattering of points for the test at
an initial void ratio of 0.59 shows a variation of void ratio up to 15% of its initial value, while
the test at initial void ratio of 0.93 presents a narrower array of void ratios. The assumption
of no volume change is acceptable only for the latter case.
It was anticipated from the previous analyses that a seemingly intrinsic shape of retention
curve can be determined at a given constant void ratio, and that this shape is duplicated for
other void ratios. It is proposed to verify this hypothesis by selecting a close range of void
ratios (not exceeding ±3% of reference values 0.59; 0.64; 0.92, see Fig. 5.7) for the clays
studied in Romero (1999); to intend drawing the idealized soil water retention curves in
( Sr : ln s ) plane as if the material remained nearly undeformable. From updated Fig. 5.8, and
despite the limited number of points in the analysis, a smooth repartition of points for each
void ratio around a typical shape is noteworthy. Also, Fig. 5.8 shows no clear evidence of a
convergence of curves towards a single line, but rather parallel contours.
In summary, the soil water retention curve is thus understood to own:
(i) an intrinsic shape defined as the outline of the ( Sr − s ) relationship at any given
fixed void ratio,
(ii) a direct dependency on the initial and current porosity, shifting the specific curves
along the ln s axis.
Assuming that the intrinsic shape is not altered by the level of compaction or by
volumetric straining, the dependency of the soil water retention curve on void ratio
apparently sums up to information on the localisation of the retention curve (position along
the suction axis). From the behavioural modelling viewpoint, this implies that once the
intrinsic water retention contour is defined at a given reference void ratio, only the
localisation of the curve along the ln s-axis on the basis of soil volume remains to be
quantified. This determination is the object of next paragraph.
It can also be remarked that if the intrinsic shape of soil water retention curve and the
volume-dependent ordinate are determined, it would be in theory possible to recover a more
complex SWRC involving both suction changes and deformation. Fig. 5.9 illustrates a generic
drying in a deformable idealised medium whose intrinsic retention is a priori determined.
When soil deformability is allowed, the retention curve outline thus diverges from the
reference intrinsic shape, and ( ln s, S r ) points jump from one intrinsic curve to another.
5.4.2 Dependency of air entry value on void ratio

Following the previous discussion, it can now be assumed for further analysis that (i) the
soil water retention curve possesses an intrinsic shape and (ii) it is thoroughly shifted with
void ratio in a ( Sr − ln s ) plane. The consequence on conceptual understanding is that once
the intrinsic shape (featuring capillary hysteresis) is captured, only one parameter setting its
position along the ln s - axis is theoretically necessary. In an effort to keep a physical
visibility in modelling concepts via this parameter, it is proposed to assess the direct response
of the air entry value, also called suction of air entry, with respect to volumetric changes.
The air entry suction was early identified by Aitchison (1960) among others, as the
limiting pressure deficiency in pore water ( pd′′ ) at which the pore drainage occurs. The air
entry value se is hence defined as the suction below which the pores are water filled ( Sr = 1)
and beyond which the pores begin to be drained ( Sr < 1) , i.e. air breaks into the pore space.
This observation was made on the water retention curve of an idealised soil consisting of a
homogenous packing of uniform spheres. A second noteworthy analysis from this pioneer
work was that the range of suction for the condition of full saturation is significantly larger
in a medium consisting of finer particles. In other words, the air entry suction is higher in a
Figure 5.9 Shape of the retention curve of a deformable soil: comparison with intrinsic contour.
134 CHAPTER 5
packing of fine particles (diameter of 0.1 micron) than in an arrangement of larger particles
(e.g. 0.01 mm in diameter). Also, for the same particle dimension, a close packing evidences a
higher air entry value with respect to an open packing. This last feature confirms that the
suction of air entry is closely linked with the size of pores, the growing of the air entry in
finer material being also a direct consequence of pores being of smaller radii.
The explicit dependency of air entry value on the porosity of the soil has been quantified
in a number of relatively recent works reviewed hereafter. Commonly, the principle is to
determine soil water retention curves under different initial porosities, thus under various
volumes of voids. The information on porosity is either directly provided, for instance by
recording void ratio, or indirectly accessible, e.g. by the means of vertical net stress.
Huang et al. (1998) re-analysed data from previous experimental campaigns displaying
retention properties of Touchet silty loam, Columbia sandy loam and sand. The air entry
values were determined for the retention curves drawn for each material at different
densities, prepared by static compaction. Fig. 5.10 redraws the dependency of air entry
suction on initial void ratio. It can be noticed from the obtained relationship that (i) for a
given material, the air entry suction decreases with growing voids and (ii) at a given void
ratio, the air entry value is more than ten times smaller for the sand than for the loams.
On the basis of another extensive experimental program, Vanapalli et al. (1996, 1999) cited
previously, also determined the shape of retention curves of a glacial till under different
preconsolidation states, combining the techniques of pressure plate and desiccators.
Investigations on the influence of the compaction mode on the soil water retention curve of
this same till are also presented. The air entry suctions have been re-determined here on the
basis of reference data using the secant tangents method, leading to values quite different
from those published in the initial work. In an effort to quantify a direct link between air
entry value and void ratio, all obtained data are gathered in Fig. 5.11, all modes of
compaction (optimum, dry side and wet side of optimum) together. In this dataset, only the
initial void ratio (i.e. the void ratio prior to desaturation) is available so that a relationship
between air entry value and void ratio could be drawn essentially with respect to a reference
initial volumetric state. The scattering of points in Fig. 5.11 prevents from determining a
unique function to comprehend the whole ( se − e ) trends. Nevertheless, the global
10
Touchet loam
Columbia loam
8 Unconsolidated sand
Air entry value s (kPa)
e
0
0.5 0.7 0.9 1.1 1.3
Void ratio e(-)
Figure 5.10 Evolution of air entry value with void ratio (after Huang et al. 1998).
120
optimum
100

80
e
60
wet of optimum
40 dry of optimum
20
0
0.2 0.4 0.6 0.8
Void ratio e(-)
Figure 5.11 Evolution of air entry value with void ratio. Air entry values have been re-determined on
the basis of retention curves from Vanapalli et al. (1999).
20
Ng and Pang 2000
Updated points
15
e
10
0
0.6 0.7 0.8
Void ratio e(-)
the basis of retention curves from Ng and Pang (2000).
arrangement of air entry points clearly confirms a clear decreasing tendency with increasing
void ratio.
Data from Ng and Pang (2000) also contribute to draw a direct link between air entry
value and volumetric state for a sandy silt and clay mixture. The available retention curves
have been reinterpreted in order to possibly refine the determination of air entry values. The
original estimations (average values) are plotted along with updated results in Fig. 5.12.
Even though the magnitudes of air entry values are different from a study to the other, the
global trend is closely comparable to those of previously analysed materials. Further works
from Sugii et al. (2002) see Fig. 5.5, corroborate those trends.
136 CHAPTER 5
250
200

e
150
100
50
0
0.84 0.86 0.88 0.9 0.92
Void ratio e(-)
the basis of retention curves from Cabarkapa and Cuccovillo (2006).
Even more recently, Cabarkapa and Cuccovillo (2006) carried out triaxial tests in
unsaturated conditions to evaluate the influence of mean net stress on the shape of the soil
water retention curve, and in particular on the air entry value. Reinterpreting the air entry
values with respect to initial void ratio, Figure 5.13, it is again possible to draw an explicit
dependency of air entry values with void ratio, in total conformity with the trends presented
previously. The reviewed paper also introduces a relation between the air entry value se and
the net stress σ net , Eq. 5.1.
se = 17 × σ net
0.33
(5.1)
This fitting relation is yet another indication that the air entry value increases with
increasing net stress, that is for reducing void ratio. However, the non linearity of the
relationship Eq. 5.1 in ( se : σ net ) plane is not a direct evidence for non-linearity or any evident
functional relationship in ( se : e ) plane.
In the continuity of studies on the behaviour of unsaturated Sion Silt carried out in our
group, (Geiser et al. 2006, Rifa’i 2002, Péron et al. 2007) more retention curves under different
applied stresses were determined1. The drying operations were performed in an oedometric
cell with conventional mechanical axial loading by mass. The oedometric device, called
hydrocon, is modified to cope with partial saturation in water, using the axis translation
technique with controlled air overpressure and water back pressure (Cuisinier and Laloui
2004). Attention has been paid in particular to the quantification of air entry thresholds
under different porosity values. From the synthesis on the array of points in ( se : e ) plane in
Fig. 5.14, it is possible to draw a quasi linear relationship between the air entry and the void
ratio.
As a conclusion, our review of the literature confirms that the air entry suction is directly
dependent on the void ratio and strain history. Independently from the type of material or
1 More details on the experimental procedure and tested material are provided in part 5.7.
110
100

90
e
80
70
60
Geiser 1999
Rifa'i 2002
50 Péron et al. 2007
Nuth and Laloui 2008
40
0.45 0.55 0.65
Void ratio e(-)
Figure 5.14 Evolution of air entry value with void ratio for Sion silt.
compaction method, the air entry threshold grows with smaller pore sizes. Several types of
mathematical functions could fit the ( se : e ) data for the reviewed materials; however, given
the little amount of experimental points, it is proposed to simplify the relationship to a linear
approximation. Within the range of suction studied, and for the panel of reviewed materials,
no information in Figs. 5.10 to 5.14 clearly enables contesting that estimation or call for a
more complex approximation. Nevertheless, the fitting could be possibly refined whenever
more complete experimental evidence is available.
5.5 Soil water retention model

5.5.1 Mathematical formulation
The preceding review of experimental behaviours for water retention of soils evidences
the need in behavioural models to account for both the hysteretic shape of the degree of
saturation versus suction curve, as well as an explicit dependency of fluids repartition on
mechanical state and history. A large number of modelling frameworks have been
formulated to understand the water retention behaviour. Initiated in agricultural soil science,
soil water retention curves model the soil water content as a non-linear reversible function of
water potential or suction. Further studies focus on degree of saturation versus matric
suction (e.g. Van Genuchten 1980), still under similar frameworks of non linear elasticity.
Some attempts on upgrading such models to hysteretic retention predictions proved to be
successful (Pham et al. 2003); still they keep using a fully reversible fitting of data and have a
limited applicability. Nevertheless, non-linear reversible formulations present the advantage
of a straightforward implementation in numerical codes and ease of calibration.
In the footpath of Wheeler et al. (2003) and Gallipoli et al. (2003), a new behavioural
model for the soil water retention curve is proposed, featuring an original combination of
both (i) a hysteresis aperture and (ii) a coupled effect of soil volumetric strain. The
formulation corresponds to the retention part of a comprehensive constitutive model called
ACMEG-s intended for modelling the stress-strain and capillary behaviour of soils partially
saturated in water (Nuth and Laloui 2007).
138 CHAPTER 5
To model accurately the evolution of degree of saturation with suction, the following
principles are adopted:
(i) An elasto-plastic (reversible-dissipative) approach of the retention curve is used to
recover the scanning curves (Fig. 5.15) and large variations of degree of saturation.
(ii) Kinematic hardening contributes to sorting out the main drying curve from the
main wetting curve and thus draws a capillary hysteresis.
(iii) The complete retention curve shape is affected by the mechanical volumetric strain
anytime. The air entry value, one of the shape parameters of the retention model, is
explicitly depending on the skeleton volume information, ε v .
The generic shape of the modelled retention curve is plotted in Fig. 5.15. The increment of
degree of saturation dS r is decomposed into an elastic part dS re and a plastic part dS rp :
dSr = dSre + dSrp (5.2)
The elastic increment of degree of saturation is defined as
ds
dS re = (5.3)
K H × ( s / seH )
seH being the ‘updated air entry value’ and K H an elastic modulus. For particular ranges
of suction and saturation, within the saturated state ( s < seH ) on the one hand and the
residual saturation state ( S r = S res ) on the other hand, the elastic increment remains null,
and the degree of saturation equals either 1 or S res , respectively.
The yield surface delimiting the reversible domain is expressed under a unified following
form for both drying and wetting processes:
Figure 5.15 Proposed model for the soil water retention curve.
1 1
f = ln ( s ) − ln ( sD ) + ⎡⎣ln ( sD 0 ) − ln ( seH ) ⎤⎦ − ⎡⎣ln ( sD 0 ) − ln ( seH ) ⎤⎦ (5.4)
2 2
with sD being the actual ‘drying yield suction’ and sD 0 the reference yield suction. The
‘drying yield suction’ is a threshold that is analogous to preconsolidation pressure in a
mechanical stress-strain model with a yield limit; consequently, sD evolves with the history
of suction, from a given initial value sD 0 (Fig. 5.15). According to conventional lexicon of
1
kinematic yield surface, the center and radius are both defined as ⎡ln ( sD 0 ) − ln ( seH ) ⎤⎦ . The
2⎣
radius of the yield surface controls the aperture of the hysteresis and is assumed to remain
constant.
The flow rule is associated, so the plastic increment of degree of saturation is written
∂f
dS rp = λHp (5.5)
∂s
With the plastic multiplier λH being
ds ⎛ ⎛ s ⎞ 1 ⎛ sD 0 ⎞ ⎞
λH = sign ⎜ ln ⎜
⎜ ⎟ + ln ⎜ ⎟ ⎟⎟ (5.6)
β H sD ⎝ ⎝ sD ⎠ 2 ⎝ seH ⎠ ⎠
β H is the coefficient of compressibility for the plastic part of degree of saturation S rp :
⎛s ⎞
ln ⎜ D ⎟ = β H S rp (5.7)
⎝ sD 0 ⎠
The air entry value seH is updated anytime as a function of the skeleton volume
information. We make the choice to relate the air entry suction to the volumetric
deformation, ε v , directly obtained from stress-strain relationships2 (for instance with model
ACMEG-s (Nuth and Laloui 2007)):
seH = se + π H .ε v (5.8)
se is the reference air entry value, π H is a material parameter. The linear approximation
for evolution of air entry value with volumetric strain is a direct consequence of adequate
linear fitting in terms of void ratios (see Figs 5.10 to 5.14). In order to recover a constant
radius of aperture of the hysteresis, the reference drying yield suction is updated from the air
entry value:
sDI × seH
sD 0 = (5.9)
se
2The implications of the double-way coupling between the stress-strain model and the soil water retention
model are discussed further in chapter 6.
140 CHAPTER 5
1.2

r
0.8
0.6
0.4
0.2
Exp.
Model
0
600 8001000 3000 5000
Figure 5.16 Calibration of the proposed water retention model with experimental points on Hostun
Sand (experimental points from Engel et al. (2003)).
5.5.2 Model response

Among the six parameters to determine, the reference air entry value se and the residual
degree of saturation S res are conventional and trivial (Fig. 5.15). On the contrary the elastic
modulus K H , the coefficient of compressibility β H and the reference drying yield suction
sDI finely govern the shape of the hysteretic part of the curve. Whenever scanning lines are
available, they should be used first to calibrate the elastic slope prior to the adjustment of the
elasto-plastic slope. The initial drying yield suction is lastly chosen to fit the hysteresis shape.
Fig. 5.16 provides an example of accurate fitting of the soil water retention curve of a sand
(Engel et al. 2003), the deformations being here assumed negligible along the drying wetting
path.
Independently, the calibration of the coupling parameter π H requires experimental
evidence of air entry value varying with volumetric strain, on the basis of data from typical
Figs 5.10 to 5.14. An example of fitting for air entry suctions with respect to volumetric strain
is provided in Fig. 5.17a. The retention model then accounts for complex couplings between
volumetric strain and SWRC shape: the air entry value is strain-dependent, and at the same
time, suction changes can induce deformations (Nuth and Laloui 2007). Consequently,
during a drying or wetting path, the air entry value is not only determined with respect to a
given initial soil volume, but is also updated for each suction change, causing the whole
SWRC shape to diverge from the intrinsic outline. The implications of this coupling are
summed up in Fig. 5.17, in which three drying paths previously evocated from Vanapalli et
al. (1999) are modelled or predicted:
• For a given initial void ratio e0 = 0.51 the experimental soil water retention curve is
used to calibrate the model. The evolution of degree of saturation is non linear with
respect to ln s due to the permanent shifting of the curve during suction induced
compression, in conformity with the concept previously developed in Fig. 5.9. For a
matter of comparison, the equivalent retention curve with zero deformation (see
‘model decoupled version’ in Fig. 5.17b), or intrinsic shape, displays constant elastic
70
1.1
(a) ACMEG-s
(b)
60
s =11kPa 1
Air entry suction s (kPa)

e
50 π = 1e3 kPa
r
H
e
0.9
40
0.8 Exp. data:
30 e =.44
0
20 0.7 e =.47
0
e =.51
0
10 Exp. 0.6 Model decoupled version
Model e =.51
0
0
0 0.02 0.04 0.5
10 1000
Volumetric strain εv (-) Matric suction s (kPa)
Figure 5.17 Simulations of retention of a glacial till (experimental data from Vanapalli et al. (1999)). (a)
Calibration of parameters for air entry value and (b) soil water retention curves at different initial
volumes.
1.1 0
A
(a) (b)
A B
1
B Drying -0.005
r
Drying
v
D
0.9 E
Compressing
Wetting
D -0.01
0.8 Compressing C
F -0.015
0.7
Wetting E
C F
0.6 5 6
-0.02 5 6
10 10 10 10
Matric suction s (Pa) Matric suction s (Pa)
Figure 5.18 Qualitative response of the proposed water retention model to drying-wetting cycles
featuring mechanical compression (a) saturation response (b) volumetric response.
and elasto-plastic slopes in ( Sr − ln s ) representation. The intrinsic shape is obtained by

decoupling the retention curve from the mechanical part of the model, setting
parameter π H equal to zero.
• On the other hand, whenever mechanical compressions are applied previously to

drying, the reduction in void ratio causes a clear shifting of the retention curve towards
higher values of suction. The blind prediction of the soil water retention curve for an
initial void ratio e0 = 0.44 appears to be particularly accurate.
142 CHAPTER 5
The proposed framework thus combines an advanced approach for modelling the
capillary hysteresis via elastoplastic concepts, and also a double dependency on soil volume.
The defined soil water retention curve is thus linked to the void ratio history and is updated
in real time with straining due to suction changes. Interestingly, the simple coupling via air
entry value also enables to model changes in degree of saturation in a partially saturated
material under a constant suction and submitted to a mechanical compression, see path DE
in Fig. 5.18. In the case of Fig. 5.18, the volumetric response to drying/wetting cycle is
modelled as fully elastic, according to the mechanical part of ACMEG-s (Nuth and Laloui
2007), but could be issued from any other mechanical model for unsaturated soils. Indeed,
when volumetric straining occurs at constant suction, Eq. 5.8 imposes the air entry value seH
to vary accordingly. As this parameter enters the yield surface formulation, the whole SWRC
is shifted, so that plastic variations in degree of saturation are generated to ensure that the
saturation state remains on the yield limit, under the shape of a vertical path in plane
( Sr − ln s ) . Due to the irreversible nature of the process, the new retention curve (drying path
EF) obtained along a subsequent suction cycle is clearly distinct from the initial one (drying
path AC) in Fig. 5.18a.
5.6 Conclusion
The extensive review confirms that the soil water retention curve is known to feature in
the most general case a capillary hysteresis together with a void ratio dependency. The
analogy between dissipations in degree of saturation along drying-wetting cycles and
elastoplastic straining theory has been drawn, and a new intrinsic outline for the retention
curve is extracted from data that are equivalent to zero deformation retention tests. Lastly,
the dependency of the physical parameter air entry value upon void ratio has been
quantified. All three ingredients are inputted into a new advanced retention model, the use of
which is inseparable from mechanical stress-strain information. In the light of the proposed
innovative modelling framework based on kinematic hardening, simulations of highly
coupled non linear retention curve of fine-grained materials proved to be successful. Such a
retention framework is in theory compatible with advanced models for unsaturated soils
requiring an accurate description of the retention behaviour.
5.7 Complements: experimental tasks

The purpose of this part is to complement the contents of the journal publication disclosed
previously. The theme developed hereafter is the experimental characterisation of the soil
water retention curve of Sion silt under different initial volumetric states. The objective of the
short experimental program was to provide more data for the retention behaviour of a
material already well characterised in our laboratory. The complementary data that has been
obtained was indeed missing to fully calibrate the model for the water retention curve in
ACMEG-s.
5.7.1 Description of the tested soil

The tested soil is a silt from the region of Sion (Switzerland). The reason for choosing this
material is that Sion silt has been already extensively characterised in our laboratory, with
the works of Botu (1994), Geiser (1999), Rifa’i (2002), Péron (2008). The stress-strain
behaviour of Sion silt is thus well characterised under saturated as well as partially saturated
conditions. The retention properties have been also focused on in the most recent research
works of the laboratory. Sion silt was identified as suitable fine-grained material for testing
100
80
Percent passing (%)

60
40
20
0
0.001 0.01 0.1 1
Diameter (mm)
Figure 5.19 Grain size distribution of Sion silt.
in partially saturated conditions due to the shorter testing time and lower suctions to be
imposed in comparison with finer clays. The grain size distribution of Sion silt is plotted in
Fig. 5.19, with a content of 9% clay, 75 % silt and 16% sand (Rifai’i 2002). Soil index
properties are gathered in table 5.1. According to USCS classification, Sion silt belongs to the
CL class that is clayey silt with sand. X-ray diffractometry showed that the smectite mineral
gravimetric content is less than 3%. The material is thus not identified as a potentially highly
swelling material.
Table 5.1 Sion silt index properties
wL ( % ) wP ( % ) IP (%) γ s ( kN / m3 )
22.7 15.5 7.2 27.6
5.7.2 Investigated behaviour and objective

Given that the retention behaviour is highly dependent on the volumetric behaviour, it is
proposed to check the coupling between the void ratio and the soil water retention curve for
Sion silt. As mentioned previously, the mechanical and hydraulic behaviours of that material
have been thoroughly characterised during previous PhD theses in our laboratory. The
features of interest for the retention behaviour only are reviewed hereafter.
The first drying and wetting tests of Geiser (1999) were carried out in Richard’s cell (Fig.
5.20). The suction is imposed by the means of air overpressure (axis translation technique)
while no mechanical stress is imposed to the surface of the samples. The dead weight of the
samples is considered as negligible. The drying tests were performed on samples under the
state of saturated slurry ( w = 1.5wL ). The wetting tests were either carried out directly after
the previous drying tests or otherwise using initially oven-dried samples. Geiser estimated
the air entry value between 50 and 80 kPa. Using the method of the intersection of tangents
in Fig. 5.20a, the air entry value is taken hereafter as equal to 50 kPa. The volume of the
samples was measured by immersing samples in hydraulic equilibrium into oil according to
the fluid displacement technique. The void ratio at air entry suction is measured from Fig.
5.20b, and has a value of 0.70. A second series of data from Geiser (1999) is plotted in Fig.
144 CHAPTER 5
1 1
(a) (b)
0.8
0.9
r
Triaxial
test
Void ratio e(-)

0.6 Other
tests
0.8
0.4
Other
tests 0.7
0.2
Triaxial
test
0 4
0.6 4
1 10 100 1000 10 1 10 100 1000 10
Figure 5.20 Drying paths on Sion silt, from Geiser (1999).
0.9
0.8
r
0.7
0.6
0.5
0.4
Triaxial cell
0.3
Oedometric cell
0.2
1 10 100 1000
Figure 5.21 Drying paths on Sion silt, from Rifa’i (2002).
5.20, with the retention curve being determined in a triaxial cell. Here the samples are tested
under a given confining pressure (400kPa), so that the initial void ratio is lower than that of
the samples in Richard’s cell. It is measured from Fig. 5.20 that in such conditions the air
entry value would be close to 75kPa, and the corresponding void ratio around 0.64.
Rifa’i (2002) carried out oedometer and triaxial tests on Sion silt with different initial
preconsolidation pressures, both devices using the axis translation technique. The principle
of the tests was to desaturate the samples under different levels of mechanical (vertical or
volumetric) stresses; as a result the air entry value can be determined for various initial void
ratios. The curves of degree of saturation versus suction are plotted in Fig. 5.21, from the
oedometric results at vertical stresses of 300kPa, and from the triaxial drying test at a
confining pressure of 600kPa. The void ratios have been calculated from the ( e − σ ) planes.
For a difference of e between 0.72 and 0.67, the air entry value jumps from 25kPa to 55kPa.
1
(a) 0.8
(b)
0.8
r
0.75
0.6
Void ratio e(-)

0.4
0.7
0.2
0 0.65
10 100 1000 10 100 1000
Figure 5.22 Drying paths on Sion silt, from Péron (2008).
Péron (2008) repeated the experiment of Geiser (1999) by drying Sion silt in the pressure
plate extractor, i.e. with no applied mechanical stress. The volume measurement is ensured
by the fluid displacement method using Kerdane, and the rigid plastic rings containing the
silt are coated with Teflon to prevent constraints on free shrinkage (Péron et al. 2007). With
this improved measurement technique, the deduced air entry value that is equal to 60 kPa is
slightly higher than that found in the previous study. The measured void ratio at air entry
point is 0.68.
It can be concluded from the reviewed results that the air entry value increases almost
linearly with decreasing void ratio, see Fig. 5.14 (with an average variation of 632kPa for an
change of void ratio Δe = 1 ). Yet, the data could be complemented with tests at lower void
ratios (of the order of 0.5) to check the consistency of the linear approximation in plane
( se − e ) . In the following are presented the procedure and results from an experimental
campaign meant for determining the air entry value of Sion silt in denser states. The
determination of these supplementary points will help to improve the calibration of
parameter π H from Eq. 5.8.
5.7.3 Experimental procedure

The experimental tests have been carried out with the contribution of Réda Traki (2006), a
student that I supervised, with an oedometric device modified to control suction by the axis
translation technique (Cuisinier and Laloui 2004, Dysli 2007). The oedometric cell called
“hydrocon” (Fig. 5.23) uses a standard mechanical loading ram with weights, while suction
is controlled by the difference between pore air pressure (imposed at the top through the
porous stone) and pore water pressure (imposed in the reservoir underneath the sample).
Both air and water taps are connected to fluid pressure transducers that enable to control
and measure the air and water inflow/outflow to the sample. The air entry pressure of the
ceramic disk is 500 kPa, which is suitable for the range of suctions in the drying paths
imposed here. Vertical displacements are monitored via the position of the piston.
146 CHAPTER 5
Figure 5.23 Layout of the oedometric device with suction control (from Dysli 2007).
Figure 5.24 Moulding tube under load.
The initial dimensions of the silt sample (diameter 63.5mm and height 12mm) are such
that drying is at the origin of significant shrinking. That means that under free shrinkage
conditions (that is drying with no imposed mechanical stress) the sample will not only settle
down vertically but also tend to reduce in radius. This phenomenon is at the origin of either
lateral disjoining between the metal ring and the sample or else desiccation cracking within
the sample if the lateral contact is constrained. Provided that the volume measurement in the
oedometric cell only relies on monitoring the vertical displacement, any loss of lateral contact
would induce errors into the measure by overestimating the total volume. The procedure
followed hereafter ensures that the lateral contact of the silt sample with the non deformable
ring is maintained through the drying process. The principle is to apply a sufficiently high
vertical stress prior to and during the suction increase.
The samples are prepared from remoulded soil, by mixing dry soil powder with de-aired
and demineralised water to the target water content equal to 1.5wL . This value was chosen to
warranty the reproducibility with respect to the previous studies and ensure a saturated
initial sate. The preparation is then poured into a moulding tube including the final
oedometer ring (Fig. 5.24), vibrated and loaded with weights following the procedure in
table 5.2. This static uniaxial compaction addresses several issues for the subsequent
oedometric test. The samples after compaction reach a water content of 25.1%, with a void
ratio of 0.70 and a degree of saturation of 1. This initial state was repeatable and samples
easier to handle for the final placing in the oedometer with respect to the reference slurry.
Furthermore, the initial void ratio being smaller, the time and loads required for
consolidating the sample to the target void ratio (0.5) in the oedometer are significantly
reduced.
Table 5.2 Uniaxial loading procedure for sample preparation

Step Duration Load Operation
1 1 min 1 kg
2 4 min 2.6 kg
3 25 min 4.8 kg Return tube/add ring
4 30 min 4.8 kg Return tube
The test path in the oedometer is summarized in table 5.3. The concept of intrinsic shape
of the retention curve was proposed previously in part 5.4.1 to define the shape of the
( Sr − s ) curve under zero deformation. It would have been theoretically relevant to carry
out a drying test under constant void ratio. However, it would be a major endeavour to
intend controlling the void ratio in the hydrocon cell upon a suction change. The technical
constraints thus imposed to carry out the drying tests with constant vertical net stress
instead. The maximum suctions imposed within the samples are voluntarily moderate to
reduce the testing time. The tests are indeed ended once the air entry value is determined
because the determination of the complete shape of the SWRC down to the residual
saturation state is out of the scope of the experiment. Only one complete retention curve (test
HYDROC-141) was judged exploitable with respect to volume change measurement. The
results of the other tests (under a vertical stress of 300 and 450 kPa) were apparently too
much affected by the friction along the piston and the drops in air overpressure due to
leakage to be taken into consideration into the present analysis.
Table 5.3 Test paths of drying tests in oedometer
Id σ v − net (kPa) Path Name

1 141 s=1,10,30,50,80,100,190,300 kPa HYDROC-141
148 CHAPTER 5
5.7.4 Results
The soil water retention curve obtained after consolidation to a void ratio of 0.51 is plotted
in Fig. 5.25, along with the volumetric behaviour under the effect of drying. The air entry
value for test HYDROC-141 is estimated to 110kPa, with a void ratio close to 0.5. It can be
observed from Fig. 5.25b that the variation of the void ratio with suction, although very
small, is following the conventional trend that is a stabilisation for suctions over the air entry
point.
The obtained point of air entry value is reported into Fig. 5.26 which is Fig. 5.14
completed with all points mentioned in complementary part 5.7. The linear approximation
seems to fit correctly the ( se − e ) relationship while an exponential formulation of the type of
Eq. 5.10 would capture the trend with more accuracy:
1 0.505
(a) (b)
0.9
0.8
r
Void ratio e(-)
0.7
0.6 0.5
0.5
0.4
0.3
0.2 0.495
1 10 100 1000 1 10 100 1000
Figure 5.25 Drying path on Sion silt in unsaturated oedometric cell.
140
Linear fitting
120
Exponential fitting
100
e
80
Geiser (1999) Press. plate
60 Geiser (1999) triaxial
Rifa'i (2002) triaxial
Rifa'i (2002) oedo
40
Péron (2008)
Present study
20
0.2 0.4 0.6 0.8
Void ratio e(-)
Figure 5.26 Evolution of air entry value with void ratio for Sion silt, updated graph.
⎛ ⎛e ⎞⎞
se = se max − ( se max − se 0 ) × exp ⎜⎜η × ⎜ − 1⎟ ⎟⎟ (5.10)
⎝ ⎝ e0 ⎠ ⎠
with se max , se 0 and η being material parameters. The advantage of formulation Eq. 5.10 is
that the air entry value would be limited to a maximum value se max which might meet
physical considerations. In fact, the air entry suction should not be growing infinitely.
However the use of 3 parameters instead of 2 in Eq. 5.8 is a major shortcoming. On the other
hand the void ratio itself does not decrease infinitely, which means that the air entry value
calculated with Eq. 5.8 will have anyway an upper limit due to the material physics. In
consequence, the linear formulation (Eq. 5.8) is judged as more convenient for parameter
determination and numerical implementation, and will thus be adopted as such in this
research.
5.8 References
Nuth M., Laloui L. (2008a). "Advances in modelling hysteretic water retention curve in deformable
soils." Computers and Geotechnics 35(6), pp. 835-844.
Nuth M., Laloui L. (2008b). "Effective stress concept in unsaturated soils: Clarification and validation
of a unified framework." International Journal for Numerical and Analytical Methods in
Geomechanics 32, pp. 771-801.

Aitchison G. D. (1960). "Relationships of moisture stress and effective stress functions in unsaturated
soils." Pore pressure and suction in soils, London, Butterworths.
Botu N. (1994). Détermination des parameters élasto-plastiques d’un limon sableux à l’aide d’essais
triaxiaux, Soil Mechanics Laboratory, EPF Lausanne. Internal report.
Cabarkapa Z., Cuccovillo T. (2006). "Automated triaxial apparatus for testing unsaturated soils."
Geotechnical Testing Journal 29(1), pp. 21-29.
Cuisinier O., Laloui L. (2004). "Fabric evolution during hydromecanical loading of a compacted silt."
International Journal for Numerical and Analytical Methods in Geomechanics 28(6), pp. 483-499.
Dysli M. (2007). Étude expérimentale du dégel d'un limon argileux: application aux chaussées et
pergélisols alpins. Lausanne, EPFL. PhD. Thesis.
Engel J., Schanz T., Lauer C. (2003). "State parameters for unsaturated soils, basic empirical concepts."
In: Unsaturated Soils: Numerical and Theoretical Approaches, Weimar, pp. 125-138.
modélisation constitutive. Lausanne, EPFL. PhD. Thesis.
remoulded silt." Soils and Foundations 46(5), pp. 545-556.
Huang S. Y., Barbour S. L., Fredlund D. G. (1998). "Development and verification of a coefficient of
permeability function for a deformable unsaturated soil." Canadian Geotechnical Journal 35(3),
pp. 411-425.
150 CHAPTER 5
Kato S., Matsuoka H., Sun D. A. (1995). "A constitutive model for unsaturated soil based on extended
SMP." Proc. Unsat'95, Paris, pp. 739-744.
Li X. S. (2005). "Modelling of hysteresis response for arbitrary wetting/drying paths." Computers and
geotechnics 32, pp. 133-137.
NG C. W. W., Pang Y. W. (2000). "Influence of Stress State on Soil-Water Characteristics and Slope
Stability." Journal of Geotechnical and Geoenvironmental Engineering 126(2), pp. 157-166.
Nuth M., Laloui L. (2007). "New insight into the unified hydro-mechanical constitutive modelling of
unsaturated soils." Unsat Asia 2007, Nanjing, pp 109-125.
Péron H. (2008). Desiccation cracking of soils. Lausanne, EPFL. PhD. Thesis.
Péron H., Hueckel T., Laloui L. (2007). "An improved volume measurement for determining soil water
retention curve." Geotechnical Testing Journal 30(1), pp. 1-8.
298.
Rifa'i A. (2002). Mechanical testing and modelling of an unsaturated silt, with engineering
applications. Lausanne, EPFL. PhD. Thesis.
Romero E. (1999). Characterisation and thermo-mechanical behaviour of unsaturated Boom clay: An
experimental study. Barcelona, UPC. PhD. Thesis.
Sugii T., Yamada K., Kondou T. (2002). "Relationship between soil-water characteristic curve and void
ratio." Proc. UNSAT 2002, Recife, pp. 209-214.
Traki R. (2006). Impact of the mechanical state on the hydaulic behaviour, Soil Mechanics Laboratory,
EPF Lausanne. Training period report.
of unsaturated soils." Soil Sciences Am. Soc. (44), pp. 892-898.
Vanapalli S. K., Fredlund D., Pufahl D. E. (1999). "The influence of soil structure and stress history on
the soil-water characteristics of a compacted till." Geotechnique 49(2), pp. 143-159.
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, pp. 379-392.
Wheeler S., Karube D. (1995). "Constitutive modelling." Unsaturated soils, sols non saturés, Paris,
Presses de l'Ecole Nationale des Ponts et Chaussées, pp. 1323-1356.
6. Double-way coupling in ACMEG-s
6.1 Introduction
This chapter is dedicated to discussing the various levels of couplings implied in
unsaturated soils modelling. This discussion follows the two previous chapters where the
stress strain model and the retention model of the Advanced Constitutive Model for
Environmental Geomechanics, unsaturated extension (ACMEG-s) are formulated and
validated individually. As already summarized in Fig. 4.6, it is now relevant to describe the
model as a whole, with all the possible interactions between the two subparts.
Yet the description of the soil behaviour would not be complete without the
understanding of the flow of air and water within the medium. Making the fundamental
assumption that all the studied processes occur under isothermal conditions exclusively (see
conclusions of chapter 2), the constitutive laws are indeed of three types (Fig. 6.1):
(i) model for stress-strain behaviour (mechanical model)
(ii) model for retention behaviour (water retention model)
(iii) model for flow of fluids (hydraulic model)
The state of the art models for the fluid flow (here liquid water and gaseous air) are
reviewed in this chapter. Unlike the stress-strain and retention models, it has been judged
that there was no need for updating such models. The terminology used from now on
requires then a particular attention. The flow models are gathered under the heading
“hydraulic (H)” while the stress-strain model was called “mechanical (M)”. This lexicon is
consistent with the conventional terminology of “hydro-mechanical (HM)” couplings in
Figure 6.1 Complete structure of the proposed modelling framework, updated from Fig. 4.6.
152 CHAPTER 6
saturated soils and in boundary value calculation codes. The third heading “retention (R)”
has to be added in supplement to describe the model of soil water retention.
The purpose of the research work is indeed to provide a complete fully coupled (hydro-
mechanical-retention HMR) constitutive framework and integrate the developed models to
solve boundary value problems like those presented in next chapters 7 and 8. The
prerequisite to the implementation of ACMEG-s into a code for boundary value problems
has been to write a routine for the integration of the model. An insight into the algorithms of
integration is provided in this chapter.
The Finite Element code chosen for the final implementation of ACMEG-s is LAGAMINE,
which was developed in the University of Liège, Belgium (Charlier 1987, Collin 2003). The
procedure of implementation and the features of interest of the finite element code are
presented in part 6.3. A more detailed insight into the flow models is also featured in the
same part.
The last two sections of the chapter provide a qualitative and quantitative analysis of the
implemented coupled model by comparing the model response and predictions with
experimental data from case studies. Some recommendations are formulated in with respect
to the calibration of the parameters of ACMEG-s and handling of the double-way coupling.
6.2 On the integration of the double-way coupled model

This section discusses the numerical integration of ACMEG-s (mechanical and retention
models) only and does not take into account the implementation of the equation for the
transfers of fluids which will be developed in part 6.3. As it is intended to implement
ACMEG-s into a calculation code for boundary value problems, a dedicated algorithm of
integration had to be set up for the constitutive laws. Meanwhile, an independent
programme (driver) based on the same algorithm has been developed for simpler
simulations on a single homogeneous element. The challenges to be faced in the integration
of ACMEG-s are of several natures, and can be summarized in the following points:
(i) The problem is highly non-linear
(ii) There are two mechanisms of plasticity in the stress-strain model
(iii) There are two interdependent sub-models to integrate
(iv) There are interdependencies between the two models
Both the stress-strain behavioural model and retention behavioural model are written in
incremental form (see chapters 4 and 5) and will thus be implemented as such. The
algorithmic data presented hereafter are mostly based on works from the Laboratory of
Mechanics of Soil Structure and Material (LMSS-Mat) of Ecole Centrale Paris (Modaressi et
al. 1989). The driver of ACMEG-s has been indeed generated from a thorough update of an
existing programme from LMSS-Mat. The routines had to be completed to account for
capillary variables and the model for the retention behaviour. The most challenging issue to
be addressed is the management of the double-way coupling; as the mechanical and
retention models interact together anytime, there is not a logical priority to give to any of
them in the integration. Part 6.2.2 features a discussion on the different possible methods to
manage such coupled problems and develops the final choices made in the algorithms. Some
recommendations for the initialization of the problem are also provided.
Many of the options selected for the final algorithm have been indeed imposed
downstream by the specifications of the finite element code (to be presented in part 6.3). On
DOUBLE-WAY COUPLING 153
the other hand, considering the general purpose of the driver and finite element code, it was
chosen here to use the sign convention of the continuum mechanics (that is compression is
negative).
6.2.1 Integration of Hujeux’s model

Provided that ACMEG-s formulation relies on the model of Hujeux (1979), the quality of
simulations from the numerical integration of ACMEG-s will depend on the numerical
handling of the reference model as well as the quality of the numerical schemes. The
obtained results should firstly be an acceptable approximation of the solution of the model
equations while secondly the calculation costs (i.e. time) should remain reasonable. The final
objective is to incorporate the developed constitutive model into a boundary value code
which will call the model in each integration point. The global algorithm of the calculation
code could provide for instance the increment of deformations to the model which in turn
can estimate the corresponding stresses and update the state variables. The sub-algorithm for
the local integration of the constitutive model is thus the focus of this paragraph. As the
scope of this PhD thesis is not to develop or assess the efficiency of numerical schemes, but
rather to provide a working numerical tool, an existing integration algorithm (Modaressi et
al. 1989, Picuezzu, 1991) for the law of Hujeux is reviewed hereafter and used as a reference
for the integration the partially saturated model.
The return-mapping scheme (Ortiz and Simo 1986) used hereafter can be decomposed
into two distinct phases, the elastic prediction (for which the hardening variables and the
plastic deformations are fixed) and the eventual correction to satisfy the consistency
condition.
As presented previously, the model of Hujeux features two mechanisms of plasticity
coupled together by the means of their hardening variables (i.e. the volumetric plastic strain
ε vp , the degree of mobilisation of the isotropic mechanism riso and that of the deviatoric
mechanism rdev ), see chapter 4. The hardening can thus be decomposed into self-hardening
(with respect to each mechanism) and latent hardening (contribution to hardening due to the
dependency between the two mechanisms). The hardening moduli ( H ii , H id , H di , H dd ) are
stored into the matrix H . Let De be the current elastic stiffness matrix. The stress increment
response of the elasto-plastic model to a prescribed increment of strain is thus written
(Mandel 1965):
⎛ 2
⎞
dσ ′ = De : ⎜ d ε − ∑ d ε kp ⎟ (6.1)
⎝ k =1 ⎠
where d ε kp is the increment of plastic deformation due to mechanism k .

pot
At a given time t let N mec be the set of numbers of mechanisms k potentially active.
Those mechanisms are such that:
⎧⎪ f k = 0
⎨ ′ (6.2)
⎪⎩dσ ( k ) > 0
where f k is the yield surface of the mechanism k and dσ (′k ) is the effective stress
increment with respect to the mechanism. The condition of existence for the plastic
154 CHAPTER 6
multipliers λkp is that H is positive definite. So, the set of numbers of active mechanisms
act
N mec is defined as:
act
N mec = {k / k ∈ N mec
pot
; λ pk ≥ 0} (6.3)
pot act
The two sets N mec and N mec are not necessarily equal. The integration algorithm thus
includes a custom process summarized in Fig. 6.2 to search the active mechanisms among
the deviatoric and isotropic yield limits.
Before detailing the local integration routine that calls the searching algorithm presented
in Fig. 6.2, some remarks are formulated on the sub-incremental procedure. The principle is
to subdivide locally the time step Δt into a number ninc of sub-increments facinc defined as:
Δt
facinc = (6.4)
ninc
This subdivision enables to work with bigger time steps while keeping an accurate
integration of the law. The number or size of sub-increments can be either fixed to a constant
arbitrary value. A more efficient alternative is to use a number ninc that is dependent on the
rate of plastic deformations, hardening variables or stresses. The size of sub-increments (Eq.
6.4) is then re-computed for each iteration and adapted to the stress-strain levels. Even
though the sub-increments also tend to reduce the error of integration, their number should
be limited to prevent prohibitive calculation costs.
The local integration algorithm for Hujeux’s model is presented in Fig. 6.3. This routine is
reading first the parameters of the law, the stress state and the state variables at the
beginning of the subincrement. The elastic prediction is then carried out (prescribed stress
increment dσ ′e* ), knowing that the elastic matrix De depends on the new stress state due to
non linear elasticity, and thus recalculated in each iteration. If the predicted stress state
overcomes the elasto-plastic threshold, the prescribed stress increment is corrected so that
the load point returns to the threshold, using the iterative method of Newton-Raphson.
The convergence condition:
⎧ ε pi +1 − ε pi 2 ε pi =1 2 ≤ dtol
⎪ n +1 n +1 n +1
⎨ 2 2
(6.5)
⎪⎩ σ n′ +1 − σ n′i++11 σ n′ +1 ≤ rtol
is used. The condition is written in terms of ratio of stress increments between step n and
step n + 1 of the local integration law. rtol is the convergence criterion in force (usually
equal to 0.001). dtol is the convergence criterion in displacement (usually equal to 0.001).
SEARCH PLASTIC MECHANISMS

Potentially active/non active ?
(loop k)
pot
N mec =0
fk < 0
fk = 0 or
dσ > 0 fk = 0
dσ ≤ 0
pot
N mec = N mec
pot
∪ {k} λkp = 0
df k
act
N mec = N mec
pot
λkp = H −1. dσ
dσ
SEARCH PLASTIC MECHANISMS

Active ?
(loop k)
act
N mec =0
act
N mec ≠ N mec
pot
if λkp > 0 act

N mec = N mec
act
∪ {k}
act
N mec = N mec
pot
UPDATE VARIABLES
EXIT
Figure 6.2 Algorithm for the research and definition of active mechanisms.
The local integration routine of Hujeux is called by the driver of the constitutive model.
The algorithm of this main routine is detailed in Fig. 6.4. The driver can be used as a
standalone routine to perform calculations on one homogeneous element with one
integration point. For the set of tested materials, most of the model parameters (see appendix
A) are calibrated using this independent programme. All simulations and predictions in
chapter 4 and 5 were also computed with the driver only, using the updated version
presented later in part 6.2.2.
156 CHAPTER 6
READ DATA
INITIALIZATION
Stress ′
σ ini
State variables ηini
IMPOSED LOAD INCREMENTS

′ , d ε imp
dσ imp
ELASTIC MATRIX
De (σ end
′i )
ELASTIC PREDICTION
′ + De (σ end
σ ′E ,i = σ ini ′i ) d ε end
i
SEARCH ACTIVE
LOOP PLASTICITY MECHANISMS
ITERATIONS (Fig. 6.2)
i = i+1
YES
k active ?
Plastic strain increment NO

Update hardening
variables
UPDATE
′
Stress σ end
State variablesη end
NO CONVERGENCE
Test (Eq.6.5)
YES
EXIT
Figure 6.3 Algorithm for the local integration of Hujeux’s model (Modaressi et al. 1989).
σ ini
′ = σ n′ ,ηini = ηn
d ε imp
ninc
σ n′ +1 = σ end
′ ,ηn +1 = ηend
Figure 6.4 Algorithm of the driver for the integration routine (Modaressi et al. 1989).
6.2.2 Integration of ACMEG-s

This section presents the generalisation of the previous algorithms to the integration of
ACMEG-s. The logic for modifying the original integration routine is consistent with the way
the model has been developed and presented previously. In other words, the upgraded
elements of the algorithms are introduced in a progressive partially-decoupled approach in
the following. While adding suction and the degree of saturation requires an extensive
rewriting of the arrays for variables, material and state parameters, the most critical issue to
be addressed is the implementation of the double-way coupling.
Adding new variables: suction and degree of saturation

In the previous section 6.2.1 dedicated to the integration of Hujeux’s model, no reference
was made to the pore water pressure pw , as the model describes only the mechanical stress-
strain relations (using the effective stress of Terzaghi (1936)). The initial driver however
includes the pore water pressure as a degree of freedom for the undrained tests. The 6
mechanical degrees of freedom per node are expressed with respect to the three directions of
the space (coordinates x, y, z ): xx, yy, zz , xy, yz , yz . The load increments were either imposed
effective stresses or strains in each of the directions.
158 CHAPTER 6
Concerning the integration of ACMEG-s, the pressure of air pa has to be added in the
degrees of freedom to account for capillary pressure, extending the number of degrees of
freedom to 8. In the following integration algorithm though, the pressures of interstitial
fluids are combined into the matric suction s ( = pa − pw ) exclusively which means that the
degrees of freedom can be considered to be limited to 7 ( xx, yy, zz , xy, xz , yz and s ).
The degree of saturation S r is a supplementary state variable for the material, and
parameters Ω, γ s , se (see chapter 5) have to be added. The relationship between the suction
and the degree of saturation is piloted anytime by the soil water retention model presented
in chapter 5, featuring 5 extra parameters ( K H , β H , sD 0 , S res , π H ). The degree of saturation is
only an output of the retention model, whose algorithm is presented in Fig. 6.5, and cannot
be controlled, which is consistent with the limitations of experimental technology. The
determination of S r does not require any integration over the time step, so the routine of Fig.
6.5 can be called theoretically anytime the value of the degree of saturation is needed. Notice
that the value of the volumetric strain has to be inputted to update the air entry value in Fig.
6.5. In other words, calling the routine of calculation of the degree of saturation (Fig. 6.5) will
produce an update of S r not only in the case of an imposed suction increment, but also upon
changes in volumetric strain.
Implementation of suction and saturation changes in the mechanical model.

Matric suction is not a standalone stress variable for the mechanical part (see chapter 3): it
is indeed considered as an equivalent isotropic pore pressure (in the negative range)
influencing the effective stress. Consequently, from the mechanical point of view, an
imposed increment in suction dsimp is considerable as a load increment only through the
′ is thus
increment of effective stress dσ ′ . The imposed increment of effective stress dσ imp
calculated from Eq. 3.38, that states:
′ = dσ net ,imp + d ( Sr simp )

dσ imp (6.6)
with dσ net ,imp being the imposed increment of net stress. Eq. 6.6 entails that the local
algorithm of integration of the mechanical law of Fig. 6.3 is directly applicable to the
integration of ACMEG-s.
In the mechanical stress-strain model of ACMEG-s, the suction-dependent variables are
the plastic compressibility coefficient and the preconsolidation pressure:
⎧ β = β 0 + Ω.s
⎪
⎨ ⎡ ⎛ s ⎞⎤ (6.7)
⎪ pc′ = pc′0 ⎢1 + γ s log ⎜ s ⎟ ⎥
⎩ ⎣ ⎝ e ⎠⎦
If suction is not constant, these two variables have to be updated at each step with the
level of suction at the end of the current step send . The calculation of the preconsolidation
pressure pc′ makes use of the value of air entry value at the beginning of the step seini . Other
implicit dependencies on suction via the effective stress (bulk and shear moduli) are
′ at the end of the step.
accounted for by the update of the effective stress to the value σ end
READ DATA
IMPORT ε
v
INITIALIZATION
Suction sini
Degree of saturation S r ini
Air entry value seH
IMPOSED LOAD INCREMENT

dsimp
ELASTIC PART
dSr = dS re
YES TEST PLASTIC THRESHOLD
f = 0 and dsimp > 0 ?
NO
PLASTIC PART
dS r = dSr + dS r
p
UPDATE THRESHOLD
UPDATE
S r = S r + dS r
CHECK
S r ∈ [ S res ,1]
EXIT
Figure 6.5 Algorithm for the calculation of the degree of saturation in ACMEG-s.
Following this update of variables, the procedure of research of active mechanisms is

carried out the way it was in the reference integration routine of Fig. 6.3. The updated
algorithm of the local integration of ACMEG-s is summarized in Fig. 6.6 whose assumptions
for couplings are discussed hereafter.
Choices for the double-way coupling

In presence of two sub-models (mechanical model and retention model in ACMEG-s), the
problem is fully coupled. The resolution could conventionally rely either on a semi-coupled
approach (e.g. Zienckiewicz and Taylor 2000) or else on the monolithic method (e.g. Collin
2003). In the first approach, the mechanical problem and the retention problem are processed
160 CHAPTER 6
separately with data exchanges at given moments. Thus, different time steps (and also
possibly different meshes) could be used for the two problems. This type of approach lowers
significantly the calculation times for instance in the case of hydro-mechanically coupled
problems (deformation and flow). Yet the terms of coupling are approximate and the method
might converge towards a wrong solution in the case of important couplings.
In the monolithic approach, the mechanical and retention problems are processed
simultaneously with data exchange at each time step. This implies a unique scale of time
increments for the two problems. Moreover, the coupled terms (for instance the stiffness
matrix) will be re-calculated at each time step. This method requires more calculation time
than the semi-coupled approach. Yet, the accounting for couplings is more accurate in the
case of the monolithic process, and the re-formation of auxiliary matrix at each time steps is
anyway necessary to integrate ACMEG-s (see part 6.2.1). In consequence, the fully coupled
monolithic approach is identified as the most appropriate method for processing ACMEG-s
and is adopted in the present work.
The variables to be exchanged between the mechanical and retention models are
s, Sr , se , ε v . The terms of direct couplings within the mechanical part were recalled in Eqs. 6.6
and 6.7. For the retention part, the only equation involving the coupling is Eq. 6.8:
seH = se + π H .ε v (6.8)
At the beginning of the step is calculated the imposed effective stress increment (Fig. 6.6),
which requires the degree of saturation and suction at the beginning of the step S r ini , sini as
well as the imposed suction increment dsimp . The plastic compressibility and
preconsolidation pressure (Eq. 6.7) are updated before calling the routine searching the
(
active mechanisms, on the basis of the updated suction at current step n sn = sini + dsimp ) and
the air entry value at the beginning of the step seini .
The routine for the calculation of the degree of saturation is called only once per step
before entering the loop of integration of the mechanical model. The iterative process and
convergence tests only concern the stress and strains. It has been deliberately chosen not to
introduce further iterations between the retention model and the mechanical model for the
following reasons:
(i) The algorithms would be significantly more complex (extra convergence test).
(ii) The added calculation time is not worth the gained precision.
(iii) Introducing iterations between the sub-models is not compatible with the chosen
Finite Element code formulation.
Concerning the last point, the Finite Element code LAGAMINE needs to call the routine
of calculation of the degree of saturation not only in the driver for the integration routine of
ACMEG-s but also at other moments (permeability calculation, to be explained further in
part 6.3).
The consequences of the algorithmic choices of Fig. 6.6 are that the calculation of the state
of saturation relies on the volumetric deformation taken at the end of previous step ( ε vn −1 ),
with the air entry suction being calculated as:
n
seH = se + π H .ε vn −1 (6.9)
′
σ ini
ηini
S r , se H
dσ imp , d ε imp , dsimp
dσ ′ = dσ + d ( S r s )
β ( send ) , p CR ( seHend )
De (σ end
′i )
′ + De (σ end
σ ′E ,i = σ ini ′i ) d ε end
i
′
σ end
ηend
Figure 6.6 Algorithm for the local integration of ACMEG-s.

162 CHAPTER 6
which also implies that the preconsolidation pressure at current step pc′ n is based on the
air entry suction of previous step sen −1 :
⎡ ⎛ s n ⎞⎤
pc′ n = pc′0 ⎢1 + γ s log ⎜ n −1 ⎟ ⎥ (6.10)
⎣ ⎝ se ⎠ ⎦
6.2.3 Remarks on initialization

Similarly to Cam-Clay model (Roscoe and Burland 1968), Hujeux’s model cannot be used
without setting the initial reference preconsolidation pressure. In a more simple elasto-
plastic model of the Cam-Clay type, the state might be initialized either by providing the
initial preconsolidation pressure or by quantifying the Over-Consolidation Ratio (OCR). Due
to the multiple mechanisms of plasticity and use of degrees of mobilisation in ACMEG-s, the
initialization of the simulation needs the explicit value of preconsolidation pressure (via the
′ 0 ).
critical state pressure pCR
By analogy, in the water retention sub-model, the initial drying yield suction sD 0 is
required at calculation step 1. However, this parameter alone is not sufficient to fully
describe the initial state: what is required is to know whether the point of capillary stress
(suction) is on one of the yield limits or within the elastic domain (scanning lines). Indeed,
unlike the mechanical model where the strain variable (skeleton deformation) always starts
from a zero value (reference state) the retention strain variable (degree of saturation) can
have any value between zero and one. More precisely, for the initial suction s0 , the range of
possible initial degrees of saturation S r 0 goes from a lower bound S r 0 min to an upper bound
S r 0max (Fig. 6.7) due to the capillary hysteresis. The integration routine reads the initial
suction but the initial degree of saturation is not always a priori determined for the
simulation. Consequently, a new parameter RETini , ranging from 0 to 1 has to be defined by
the user:
S r 0 = S r min + RETini ( S r max − S r min ) (6.11)
Figure 6.7 Initialization of water retention model.

6.3 Implementation into LAGAMINE Finite element code

To address the main engineering issues presented in chapter 1, the developed model was
implemented into the finite element code LAGAMINE. Initially, the code was developed in
Liège University (Belgium) to model problems of steel lamination (Charlier 1987). The
formulation is thus written in large deformations, which will be commented further in this
part. LAGAMINE was successively upgraded to simulate a larger range of engineering
problems, including now geomaterials and multi-phasic media. The most recent
LAGAMINE interface developed in Liège is the software CONVILAG which runs on
PC/Windows.
Considering back the specifications for the implementation of ACMEG-s, the numerical
code should have been able to account not only for mechanical degrees of freedom but at
least also for suction. The most relevant argument for choosing LAGAMINE for the
implementation is the recent development of a coupled finite element by Collin (2003). The
element formulation will be described in part 6.3.1. As explained previously, it has been
considered to take advantage of already implemented laws for fluid transfer in unsaturated
conditions, which are reviewed in this part from Collin (2003).
The momentum conservation in an elementary volume and the equilibrium at the surface
are written:
div σ + ρ g = 0 (6.12)
t = nσ (6.13)
where σ is the total (Cauchy) stress tensor with tensile stresses taken as positive, ρ is the
mass density of the medium, g is the gravity vector, t is the vector of stresses applied on
surface and n the surface normal vector.
6.3.1 Multi-phasic flow model

The hydraulic model for the transfer of water and air developed by Collin (2003) and
implemented in LAGAMINE takes into account the variations into the balance between the
liquid and vapour states of the fluids. Modelling the phase change is particularly significant
in the case of non-isothermal processes where liquid water can more easily undergo a phase
change into water vapour. In the present work, the thermal transfers are absent and the
phase change can be considered as out of the scope of this manuscript. However, as
ACMEG-s model is implemented into the fully coupled Hydro-Mechanical Finite element
code, it is necessary to review the definitions of the phases and chemical species:
(i) The gas phase is a mix of dry air and of water vapour.
(ii) The liquid phase is a mix of liquid water and dissolved air.
The only two chemical species are water and air, which are in different concentrations in
the two possible phases. It has been chosen to write the field equations for each chemical
species and not for each phase. This choice implies to write the mass conservation for a given
chemical species whatever the phase it belongs to. For this species, the terms of exchange
between phases are null, which is compatible with the assumption of equilibrium between
phases.
Mass conservation of fluids

The mass conservation of water can be written for the water in the liquid state and in the
vapour state, which added together give:
164 CHAPTER 6
Sw + div ( Vw ) − Qw + Sv + div ( Vv ) − Qv = 0 (6.14)
Where S w , Sv are the quantities of stored water in the liquid and vapour phases
respectively, Vw , Vv the mass fluxes of fluid, Qw , Qv the sources of fluid.
Introducing the mass density of water ρi , the porosity n , the degree of saturation in
water S r ,i ( i being the liquid water w of the water vapour v ):
∂
Si = ( ρi .n.Sr ,i ) (6.15)
∂t
In liquid water, the diffusion of dissolved air is neglected so the mass flux of liquid water
Vw only possesses an advective term :
Vw = ρ wql (6.16)
where q l is the average velocity of the liquid phase with respect to the solid phase. In the
gas phase, the water vapour flux is linked to that of the whole gas phase and to the vapour
diffusion in the gas phase:
Vv = ρv q g + i v (6.17)
q g is the relative velocity of the gas phase and i v the non advective flux of vapour.
The mass conservation of air is written for the dry air in gaseous form (subscript a ) and
also for the air dissolved in the liquid water (subscript a − d ), resulting in the following
combination:
Sa + div ( Va ) − Qa + Sa − d + div ( Va − d ) − Qa − d = 0 (6.18)
Defining the mass density of dry air ρ a , the porosity n , the degree of saturation in gas
S r , g , the time derivative of the quantity of stored dry air Sa is written:
∂
Sa = ( ρ a .n.Sr , g ) (6.19)
∂t
The quantity of dissolved air is calculated by Henry’s law (Weast 1971) that relates the
volume of dissolved air Va − d to the volume of water Vw using Henry’s air solubility
coefficient H :
Va − d
= H ≈ 0.01868 at T = 20°C (6.20)
Vw
which yields, with S r , w being the saturation in liquid water:
∂
Sa − d = ( ρ a − d .n.Sr , w ) (6.21)
∂t
For the calculation of the mass fluxes of air, the diffusion of dissolved air within the liquid
phase is neglected, which removes the non advective term in Eq. 6.23:
Va = ρ a q g + i a (6.22)
Va − d = ρ a .H .ql (6.23)
q g , ql are the relative velocities of the gas and liquid phases and i a the non advective flux
of dry air.
Fluid flow models

The models for fluid flow are a generalization of the laws defined in conditions of full
saturation in water. Darcy’s law, initially written to describe the flow of a liquid fluid in a
porous medium (Bear 1972) can be generalized to liquid water and gaseous air in
unsaturated conditions (Richards 1931, Fredlund and Rahardjo 1993):
Kw
qw = ( grad( pw ) + ρ w .g.grad( z ) ) (6.24)
ρ w .g
Ka
qa = ( grad( pa ) + ρa .g.grad( z ) ) (6.25)
ρ a .g
where q w , q a are the vectors of velocity of water and air phases with respect to the solid
phase, K w , K a the anisotropic tensors of water and air permeability, pw , pa the pore water
and air pressures, z the vertical coordinate taken positive upwards.
The condition of validity of Eqs. 6.24 and 6.25 is to make the permeability dependent on
S
the degree of saturation in water K w = K w r ,w ( ) and the degree of saturation in air
( S ) . In Eq. 6.25, the gravity term (last term on the right hand side) is negligible.
Ka = K a r ,a
This modelled dependency of the permeability on the state of saturation is compatible

with the experimental evidence presented in chapter 2. In a given pore within the medium,
the liquid water flow process is a function of the relative volume of water. A soil that is more
saturated in water will be more permeable to liquid water. Reversely, the relative
permeability to dry air will be increased in dryer materials.
Most of the models describing the evolution of permeability with capillary state use a
direct dependency of the water coefficient permeability kw on the degree of saturation in
water S r , w and not on suction. Fredlund and Rahardjo (1993) reviewed a number of
experimental works from literature characterizing the permeability in various conditions of
saturation in water. The experimental evidence converges towards the conclusion that k w is
a direct function of the volume of water in the soils. In other words, for a given soil, there is a
simple relationship between the permeability and the volumetric water content θ w or degree
of saturation in water S r , w , and this function is “approximately the same for both wetting
and drying” (Nielsen et al. 1972). The experimental interpretation of the variation in
permeability with respect to matric suction s (Liakopoulos 1965), Fig. 6.8a, shows on the
contrary a clear hysteresis. This is interpreted by Fredlund and Rahardjo (1993) as a
duplication of the capillary hysteresis of the soil water retention curve ( S r , w − s , Fig. 6.8b).
166 CHAPTER 6
The proof of that interpretation is provided in Fig. 6.8c, by plotting the permeability
coefficient versus the degree of saturation. In that plane, the hysteresis vanishes, showing
( )
that the k w − S r , w plane is apparently more appropriate for modelling. This analysis also
shows again the importance of accounting for the capillary hysteresis in the model of soil
water retention: only an accurate modelling of the degree of saturation versus suction can
justify a simplified model in plane k w − S r , w . ( )
It is a deliberate choice not to develop a new model for the evolution of permeability to
water and air. Firstly, since the relationship k w − S r , w ( ) is univocal, a simple fitting equation
is sufficient. The number of models developed in the last 60 years accounts for the interest in
attempting to predict this relationship (e.g. Gardner 1958, Brooks and Corey 1964). Several of
5 1
(a) (b)
Permeability coefficient k (x10 m/s)
-6
4 Degree of saturation S (-) 0.8

Drying
r
w
3 0.6
Drying
2 0.4
Wetting
1 0.2 Wetting
0 0
0 10 20 30 40 0 10 20 30 40
5
(c)
Permeability coefficient k (x10 m/s)
-6
4
Drying
w
2 Wetting
0
0 0.4 0.8
Degree of saturation Sr(-)
Figure 6.8 Permeability and capillary variables (after Fredlund et al. 1993). Experimental data from
Liakopoulos (1965). The values of degree of saturation has been here back-calculated on the basis of
volumetric water content and porosity.
these models are already implemented in LAGAMINE, Eqs 6.26 to 6.29. Secondly, the
uncertainty of measures of permeability, combined with the error on measurement of
volume of soil and water might have a significant effect on the variability of experimental
points in the plane (k w − S r , w ) . The existing available formulations have been judged to
approximate accurately enough those experimental points for the tested soils.
kw = kwsat − par1 (1 − Sr , w )
par2
(6.26)
1
kw = (6.27)
1 + ( Sr−.2.429 − 1)
1.1760
w
⎧⎪e par1Sr ,w + par2 Sr ,w if S ≥ S

2
kw = ⎨ r ,w res
(6.28)
sat
⎪⎩k w else
kwsat is the saturated permeablitiy coefficient, pari are material parameters. The coefficient
of permeability to air can be modelled for instance with the following formulation:
⎛ S − S r , w* ⎞
par1
⎛ ⎛ S − S ⎞ par2 ⎞
ka = ⎜1 − r , w ⎟ ⎜1 − ⎜ r , w r , w*
⎟ ⎟ (6.29)
⎜ 1 − Sr , w* ⎠⎟ ⎜ ⎝ 1 − Sr , w* ⎠⎟ ⎟
⎜
⎝ ⎝ ⎠
with S r , w* being a material parameter.
6.3.2 Weak formulation

Eqs. 6.12, 6.14 and 6.17, together with the constitutive elasto-plastic laws constitute a
coupled system of equations to be solved. To be able to resolve this formulation for complex
geometries and boundary conditions, the weak form of the governing equations was
obtained using the principle of virtual works and summarized below after Collin (2003).
For a field of admissible virtual velocities δ v , the sum of the work δ WE exerted by the
external forces and the work δ WI exerted by internal forces equals zero if the local
equilibrium is verified in any point of the domain and thus the global equilibrium is verified.
Defining δ ε as the tensor of velocity of virtual deformation, the external work and internal
work are:
δ WE = ∫ ρ gδ vdV + ∫ tδ vdA (6.30)

V A
δ WI = ∫ σδ εdV (6.31)
V
Here, the integrals are defined in the “deformed” configuration, due to the possibility of
large deformations. Using a Lagrangian formulation actualised at the last configuration at
equilibrium, the stress tensor σ is the Cauchy stress tensor expressed in the deformed
configuration. δ ε is thus the velocity tensor of Cauchy virtual deformation.
The Jacobian of the transformation from the initial configuration (coordinates X ) to the
current configuration (coordinates x ) is written:
168 CHAPTER 6
∂x
F= (6.32)
∂X
In the deformed configuration, the tensor of velocity v and tensor of gradient of velocity
L are written :
dx
v= (6.33)
dt
∂F −1
L= F (6.34)
∂t
The balance equations in volume and surface are introduced into the external work,
which yields:
δ WE = ∫ σ.grad (δ v )dV = δ WI (6.35)

V
Taking into account of the symmetry of the stress tensor, the Cauchy strain tensor is thus
obtained as:
1 ⎛ ∂δ v j ∂δ vi ⎞
δ ε = ⎜ + ⎟⎟
⎜
2 ⎝ ∂xi ∂x j ⎠ (6.36)
1
δ ε = ( L + LT )
2
The mass conservation of fluid can be processed similarly. The external work δ WE due to
the imposed fluxes ( Q in volume, q at the surface) for a field of admissible virtual
perturbation of pore pressure δ q is written:
δ WE = ∫ Qδ qdV + ∫ qδ qdA (6.37)

V A
which, using the local balance equations for volume and surface, give:
δ WE = ∫ Sδ q − Vq grad (δ q ) dV = δ WI (6.38)

V
6.3.3 Coupled finite element

The formulation of the coupled 2D/3D element used in the following is reviewed from
Collin (2003). The coupled finite element called MWAT is isoparametric of the Serendipity
type (Zienckiewicz and Taylor 2000). The geometry and discretized fields in such elements
are defined as a function of their nodal values by the means of interpolation functions N ,
which will not be developed in details here.
Using a plane strain state definition, the element MWAT 2D possesses five degrees of
freedom (d.o.f.) per node:
(1-2) geometric coordinates x, y (mechanical d.o.f.)
(3) Pore liquid pressure pl (hydraulic d.o.f.)

(4) Pore gas pressure pg (hydraulic d.o.f)
(5) Temperature T (thermal d.o.f)

Due to the initial assumption of the isothermal nature of the processes, the fifth degree of
freedom will be practically blocked from now on. The third and fourth degrees of freedom
are of the “flow” or “hydraulic” type. The required degree of freedom in “retention”, defined
in part 6.2.2 as the matric suction needs not to be added as it is a combination of the
hydraulic d.o.f. The developed integration routine is thus compatible with the element
definition.
In element MWAT 2D, coordinates, displacements, velocities, pressures and temperatures
are expressed as function of their nodal values via the interpolation functions N .
xi = N L X Li pl = N L PlL (6.39)
ui = N LU Li pg = N L PgL (6.40)
vi = N LVLi T = N LTL (6.41)
where L ranges from 1 to the number of nodes of the element.

The interpolation functions N are not expressed in the plane of global coordinates
( x1 , x2 ) but in the space of local coordinates (ξ ,η ) . The spatial derivatives of the
interpolation functions are expressed in the deformed configuration:
∂N L ∂N L ∂ξ ∂N L ∂η
= + (6.42)
∂x j ∂ξ j ∂x j ∂η ∂x j
The spatial derivatives of the local coordinates can be obtained from the Jacobian matrix
J transforming the global coordinates to local coordinates (with a non-zero determinant
J ≠ 0) :
⎡ ∂x1 ∂x1 ⎤
⎢ ∂ξ ∂η ⎥
J=⎢ ⎥ (6.43)
⎢ ∂x2 ∂x2 ⎥
⎢ ∂ξ ∂η ⎥
⎣ ⎦
Using Gauss numerical integration, the expressions of the internal work (Eqs. 6.31 and
6.34) thus become:
⎛ 1 ⎡ ∂N L ∂N ⎤ ⎞
δ WI = ∑ ⎜ σ ij ⎢ δ VLi + L δ VLj ⎥ tk J WIP ⎟ (6.44)
⎜ 2 ⎣⎢ ∂x j ∂xi ⎟
IP
⎝ ⎦⎥ ⎠
⎛ ⎡ ∂N L ⎤ ⎞
δ WI = ∑ ⎜⎜ Sq N L − ⎢ Vqi ⎥ tk J WIPδ Q ⎟⎟ (6.45)
IP ⎝ ⎣ ∂xi ⎦ ⎠
with tk being the thickness of the element, IP is the number of integration points and
WIP is the Gauss weight of the integration point.
170 CHAPTER 6
Eqs. 6.44 and 6.45 can be expressed in terms of work equivalent nodal forces FL ,i , relative
to each degree of freedom :
δ WI = FL,1δ VL1 + FL,2δ VL 2 (6.46)
δ WI ,q = FL ,qδ QL (6.47)
with the following expressions:
⎛ ∂N ∂N L ⎞
FL ,1 = ∑ ⎜ σ 11 L + σ 12 ⎟ tk J WIP (6.48)
IP ⎝ ∂x1 ∂x2 ⎠
⎛ ∂N L ∂N L ⎞
FL ,2 = ∑ ⎜ σ 12 + σ 22 − ρ gN L ⎟ tk J WIP (6.49)
IP ⎝ ∂x1 ∂x2 ⎠
⎛ ∂N ∂N ⎞
FL , pl = ∑ ⎜ Sl N L − L V pl1 − L V pl 2 ⎟tk J WIP (6.50)
IP ⎝ ∂x1 ∂x2 ⎠
⎛ ∂N ∂N ⎞
FL , pg = ∑ ⎜ Sg N L − L V pg1 − L V pg 2 ⎟tk J WIP (6.51)
IP ⎝ ∂x1 ∂x2 ⎠
In Eq. 6.49 (vertical direction) an equivalent force corresponding to gravity is added. The
density accounts for the deformations and degrees of saturation in fluids:
ρ = (1 − n) ρ s + nS rw ρ w + nSra ρ a (6.52)
Concerning the time discretization, each time step is defined by two moments, defined
previously as tini and tend . The fluid pressure is assumed to vary linearly over the time step
at any point of the domain:
t = (1 − θ ) tini + θ tend
(6.53)
p = (1 − θ ) pini + θ pend
where θ ∈ [ 0,1] . The time derivative of the pressure can then be reduced to:
dp pend − pini Δp
= = (6.54)
dt tend − tini Δt
The unknowns at each node have been discretized in time and the equivalent nodal forces
FL ,i are quantified at each moment. The governing equations should be verified at all time
steps. The method of weighted residuals is used hereafter for this purpose. The weight
functions Wi vary with time, which yields:
tend tend
∫tini
FLext,i Wi (t )dt = ∫
tini
FLint,i Wi (t )dt (6.55)
Using the collocation method, the weight functions Wi are taken from the family of Dirac
functions δ in the time domain, which means:
⎧1 if t = θ
Wi (t ) = δ (t − θ ) = ⎨ (6.56)
⎩0 otherwise
The equilibrium is thus expressed at a given time t = θ :
FLext,i (θ ) = FLint,i (θ ) (6.57)
For a linear system, the problem is unconditionally stable if θ ≥ 1 2 ; in the simulations

carried out in this research work, the implicit scheme has been used ( θ = 1 ).
6.3.4 Global algorithm of the finite element

The algorithm of the finite element MWAT (Collin 2003) is presented in Fig. 6.9. Due to
the multiple couplings and the monolithic approach, variables are exchanged at each time
step between the mechanical and retention laws and the flow model. The terms of coupling
are thus activated only in the initial calculation phase and in the phase where the integration
routines for the constitutive laws are called. The conventional coupled effects of the
mechanical model over the flow models are due to volumetric deformations that alter the
storage coefficients and permeability. The latter is also directly coupled to the water
retention model.
6.4 Validation of the fully coupled model

6.4.1 Motivations for introducing double-way coupling: a review
The up-to-date experimental evidence from literature (chapter 2) draws the major features
of behaviour of soils partially saturated in water with precision. Some decades ago, it was
usual to quantify with accuracy the soil water retention behaviour on the one hand and the
stress-strain framework under fixed levels of suction or water content on the other hand.
Physicians of soils have a large experience on the water retention of soil (e.g. Brooks and
Corey 1964) whereas the effect of suction on stress-strain response is well quantified by soil
mechanicians (eg. Blight 1967). More recently though, it has been understood that the two
behaviours are highly interdependent and that their intrinsic non linearity is mostly due to
these inherent couplings. Among the first authors to quantify this interdependency could be
mentioned Sivakumar (1993) and Sharma (1998). They quantified the changes in degree of
saturation in response to variations of soil volume and vice-versa. It was observed that any
stress path (mechanical load or suction variation) entails a true alteration of the volumes of
pores (void ratio) and relative volume of water (saturation) (Figs. 6.10 and 6.11).
The stake for the first constitutive models for unsaturated soils was to quantify the stress-
strain behaviour under given levels of suction or to reproduce the volumetric response to
wetting and drying (eg. Matyas and Radhakrishna 1968, Alonso et al. 1990). The Soil Water
Retention Curves (SWRC) used to be simplified to non linear fitting equations (eg. Van
Genuchten 1980). More recently, and following the new experimental evidence on couplings
previously cited, the models for the retention behaviour have been upgraded to account for
the hysteresis loop (eg. Pham et al. 2003, Wheeler et al. 2003) or volume change (eg. Gallipoli
et al. 2003).
172 CHAPTER 6
INPUT DATA
INITIAL CALCULATION
MECHANICAL
Geometric coordinates
Jacobian
Integration weight
Interpolation functions derivatives
Strain velocity gradient
FLOW
Hydraulic/thermal coordinates
Pressure and temperature gradient
Suction
LOOP
INTEGRATION
POINTS CALL CONSTITUTIVE LAWS
Mechanical
Flow
Thermal
NODAL FORCES
Calculation of total stresses
Calculation of density
Calculation of nodal forces
LOCAL TANGENT MATRIX
OUT
Figure 6.9 Algorithm of finite element MWAT (from Collin 2003)
Yet, accounting for the hydromechanical couplings and irreversible processes still needed
a large amount of fitting parameters, often with little physical meaning. Their calibration
would call for particular experimental stress paths, requiring dedicated experimental set-up.
This section presents the added value of ACMEG-s in the unified constitutive modelling
of unsaturated soils by taking benefit of the inputted couplings. The coupled aspects are seen
here as natural consequence of the formulation rather than the results of a heavy calibration.
1.1 0.1
(a) (b)
D Drying D
1 B
E 0.05
Wetting

r
0.9 Wetting A
Wetting 0
Drying Drying
0.8 C
B
C -0.05
0.7 Drying
Wetting
-0.1
0.6
A E
0.5 4 5 6
-0.15 4 5 6
1000 10 10 10 10 10 10
Matric suction s(Pa) Mean effective stress p' (Pa)
Figure 6.10 Effect of suction cycle on specific volume and degree of saturation, from Sivakumar (1993).
2.3 1
(a) (b)
2.2 0.9
Test 9
r
2.1 0.8
Test 10 Test 10
Test 9
2 0.7
1.9 0.6
1.8 0.5
10 100 1000 10 100 1000
Mean net stress pnet(kPa) Mean net stress pnet(kPa)
Figure 6.11 Effect of mechanical loading on the degree of saturation and specific volume, from Sharma
(1998)
6.4.2 Double way coupling: case studies for validation

The performances of ACMEG-s are hereafter assessed on the basis of experimental results
from Sivakumar (1993), Sharma (1998) and Lloret et al. (2004). The interest of these data sets
is the publication of the material response in terms of volume changes and saturation degree,
with complex loading paths involving combined changes in net stress and suction (Nuth and
Laloui 2008). The FEBEX bentonite (Lloret et al. 2004) and kaolin (Sivakumar 1993) modelled
here were already calibrated previously (Appendices C and D) while a new calibration for
bentonite-kaolin mix (Sharma 1998) was necessary.
174 CHAPTER 6
Validation of model on swelling pressure tests

In a swelling pressure test, a fine grained material in an initially dry state undergoes a
humidification while the overall volume is maintained constant. The prevented swelling is at
the origin of the generation of net stresses at the boundaries. Unlike the regular swelling
pressure tests where only the maximum generated stress is recorded, Lloret et al. (2004) have
plotted the complete stress path during wetting with no deformation constrain. The
monitored external stresses were described as a consequence of the stress state being in the
elastic domain or not. Here again, the generalised effective stress will play a major role in the
reversible or elasto-plastic nature of the transformation. The upcoming interpretation is also
aimed at highlighting the role of the soil water retention curve in the predictions. This
analysis complements the preliminary semi-coupled modelling campaign of part 4.6.4.
Fig. 6.12a shows the results of the prediction of swelling pressure stresses computed with
ACMEG-s superimposed with the experimental data on bentonite (Lloret et al. 2004). As in
chapter 4, the model predicts an alternation between an elastic phase (AB) (prevented
swelling), yielding (BC) (prevented collapse) and a residual elastic phase (CD). The plotted
net stress path is indeed deduced from the generalised effective stress equation (Eq. 6.58)
( ) (
Δ σ net ij = Δ σ ij′ − S r sδ ij ) (6.58)
(
As a consequence, the non linearity of the net stress response in the plane s − σ net ) is
entirely piloted by (i) the variations in effective stress and (ii) the variations in the product
suction times the degree of saturation. As detailed in chapter 4 (part 4.6.4) the reason for the
shape of the stress path BC is the generation of plastic straining due to yielding on the
loading collapse curve. In other words, along the wetting path, the effective stress is relieved
upon yielding.
A closer analysis of the soil water retention curves shows an evident plurality of the
retention outlines. The model features a dependency of the air entry value se on volumetric
strain (Eq. 6.8). The consequence of this new degree of freedom is an alteration of the
( Sr − s ) relationship from one wetting path to the other (Fig 6.12c) and a modification of the
loading collapse curve (Eq. 6.7), see Fig. 6.12b. The elastic domain is thus basically reshaped
for each initial dry density, with corresponding amplitude and occurrence of yielding for
each. The whole net stress path (Eq. 6.58) is also directly affected by the values of degree of
saturation and suction. Even though the coupled effects remain subtle in plane s − σ v net( )
(Fig 6.12a), the qualitative interpretation of the experimental trends is satisfactory.
Validation of model on mechanical cycles

Sivakumar (1993) performed a number of laboratory tests on compacted kaolin partially
saturated in water. One of his conclusions was that the degree of saturation could evolve
significantly under the effect of an isotropic compression carried out in drained conditions
with a constant suction (Figs. 6.13a and 6.13b).This phenomenon is attributed to the fact that
the size of the pores is getting smaller while little changes occur for the water content. The
trends for changes in volume and degree of saturation have been first simulated by Gallipoli
et al. (2003). The coupling between the volumetric strain and the air entry value in ACMEG-s
provides a direct interpretation for the change in saturation.
3
A Initial dry density (g/cm
s SP1 ): (MPa)
EXP 1000
100 1.63 (Exp.)
(b)
1.57 (Exp.)
1.50 (Exp.)
B 1.63 (Mod.) LC curves
1.57 (Mod.)
1.50 (Mod.) 100 before wetting

10
Wetting
10
s
e1 C
s 1 s
e1 3
e3
Initial dry density (g/cm )
s 1 1.50
e3
1.57
(a)
D 1.63
0.1
0 2 4 6 8 10 0.1
0.1 1
Vertical net stress σ (MPa)
v Mean effective stress p'(MPa)
s s
e3 e1
1.2
(c)
1
r
0.8 Predicted
3
(ρ =1.63 g/cm )
d
0.6 Predicted
3
(ρ =1.50 g/cm ) Exp. points
d
0.4
0.2
0
0.1 10 1000
Figure 6.12 Prediction of swelling pressure under different initial densities: effects of changes in air
entry value on soil water retention and loading collapse curves.
Starting from the initial state A in Fig. 6.13a, a first isotropic compression (AB) is applied
to the sample. By reference to the mode of preparation of the sample, the initial point A is
located on the main wetting curve in the retention plane ( S r − s ) (Fig. 6.13 c). The volumetric
strain generated by the compression (AB) causes the air entry value to increase from the
initial value seA to the updated level seB , and the whole SWRC loop is shifted towards
higher levels of suction. The point of retention state ( s and S r ) cannot come out of the
hysteresis loop which requires here that point B remains on the main wetting curve; so there
is a generation of plastic variations in the degree of saturation.
176 CHAPTER 6
Along the unloading path BC, the volumetric strain increases slightly so that the air entry
value at point C, seC will verify the double inequality:
seA < seC < seB (6.59)
The complete SWRC is shifted to the left in Fig. 6.13c, without any change in the degree of
saturation nor in suction. Upon further reloading CD, increase in degree of saturation
happens only once the minimum volumetric strain previously experimented (point B) is
exceeded. The complete cyclic process is repeated on path BDE, the response being plotted in
Fig 6.13b. The prediction shows a fairly good comparability with the experimental points,
given the fact that only parameter π h (Eq. 6.8) is used to calibrate the hydro-mechanical
coupling.
0.02 1
A Experimental Experimental
0 Predicted Predicted
0.9
-0.02 E D
r
v
C 0.8
-0.04
B C B
-0.06 0.7
-0.08 A
0.6
-0.1
0.5
-0.12 E
D (a) (b)
-0.14 5 6
0.4 5 6
10 10 10 10
Mean effective stress p' (Pa) Mean effective stress p' (Pa)
s s
eA eB
1
(c)
0.9
Main drying
r
D E curve
0.8
B C
0.7
A Main wetting
0.6 curve
0.5
Initial SWRC
Updated SWRC
0.4 4 5 6 7
10 10 10 10
Figure 6.13 Effect of mechanical loading on the volumetric strain, degree of saturation and reference
air entry value.
Validation of model on suction cycles

A second type of loading cycles, namely changes in matric suction under a constant level
of net stress, was investigated by Sharma (1998). Sharma’s works showed that a sequence of
wetting and drying cycles imposed to a mix of bentonite and kaolin would generate (i)
irreversibilites in the degree of saturation (Fig. 6.14a) and (ii) permanent volume changes
(Fig. 6.14b). A particular attention should be paid to the phenomenon characterised in Fig.
6.14, namely the yielding at high suctions upon drying. Here, the non linearity of the stress
path in suction versus mean generalised effective stress is an advantage (see chapter 3)
because it makes it possible to reach the yield limit (LC curve) at high suctions along a
drying path (Fig. 6.14c). Similar yielding features were developed by Gallipoli et al. (2003),
with different retention variables and formulation for the LC curve.
The response of ACMEG-s to suction cycles is represented in the two planes ( S r − s ) (Fig.
6.14a) and ( ε v − p′ ) (Fig. 6.14b). The slopes for the retention curve ( K H and β H ) have been
calibrated on paths AB and BC. Along the initial wetting AB, the degree of saturation follows
1.1 0.1
(a) (b)
D D
1
E
0.05 B
r
0.9
0 A
0.8
B C
C -0.05
0.7
-0.1
0.6 Experimental Experimental E
Predicted Predicted
A
0.5 4 5 6
-0.15 4 5 6
1000 10 10 10 10 10 10
Matric suction s(Pa) Mean effective stress p' (Pa)
5
6 10
(c)
5
Stress path
5 10 Initial LC E
Final LC
5
4 10 A C
5
3 10
5
2 10
5
1 10
B
D
0 4 5 6
10 10 10
Figure 6.14. Effect of suction cycle (successive phases of drying and wetting) on specific volume and
178 CHAPTER 6
the main wetting curve. As the product ( S r × s ) becomes smaller, the generalised effective
stress (Eq. 3.38) decreases while remaining in the elastic domain (Fig. 6.14c). The
consequence is an elastic swelling in plane ( ε v − p′ ) . The subsequent drying path (BC)
follows a scanning line in the retention plane. Yet yielding occurs in the mechanical module,
with permanent deformation being generated to reach point C in Fig. 6.14b. In the second
wetting CD, the ( S r − s ) curve draws successively the scanning line, the main wetting curve
(that is different from that of path AB due to plastic changes in void ratio) and the line of full
saturation below the air entry value. On the mechanical side (Fig. 6.14b), the stress state re-
enters the elastic domain which provokes swelling. The last drying path DE is such that the
updated air entry value is exceeded (Fig. 6.14a) and plastic hardening also occurs in the
mechanical model. While the simulation of the evolution of the degree of saturation get
quantitatively close to the experimental data, more noticeable difference is observed for the
prediction of the volumetric response for the second yielding phase. However the model
shows a unique qualitative understanding of the irreversible phases in both the mechanical
and retention plane at once.
6.4.3 Conclusions on the validation of double-way coupling

The proposed model ACMEG-s is formulated with a mechanical module and a retention
part. The key stake is the multi-level hydro-mechanical coupling between the two
constitutive parts. First the mechanical response is influenced by the degree of saturation and
suction via the generalised effective stress. The hysteretic Soil Water Retention model is also
depending on the volumetric strain with the updating of the air entry value. Then the LC
curve is in turn affected by changes in the air entry value. The double-way coupling offers a
qualitative and quantitative understanding of the response of each of the volumetric
variables (i.e. the skeleton strain and the degree of saturation) to each of the loads (i.e. matric
suction and external stress).
6.5 Calibration of ACMEG-s and influence of parameters

The multiple couplings have been studied from the viewpoint of numerical integration
and the cross-effects have been validated on complex stress paths. While the flow models can
virtually be calibrated independently (based on the unsaturated permeability models), the
two subparts of ACMEG-s (mechanical and retention models) are considered as highly
coupled. The calibration procedure should then follow a logical path that is detailed in this
section.
Most of the terms of coupling in ACMEG-s (Eqs. 6.6 to 6.8) feature a simplified dependency
on the counterpart variables and parameters. In other words, linear assumptions help to
reduce the number of parameters. Such a simplifying hypothesis implies that there are no
apparent built-in limits to the coupled variables, they can virtually increase and decrease
infinitely if the suction or volumetric strain grows accordingly. For instance, there is no
mathematical function to define a limit to the air entry suction or plastic compressibility
coefficient. This issue is indeed a false problem: due to the physical nature of the stress and
strain variables defining the state of the material, the coupled variables are naturally
moderated by the physical limits of the stresses and strains. For instance, the volumetric
strain will always reach a limit value which means that the air entry suction also has a
maximum.
From the retention and mechanical stress and strain framework point of view, ACMEG-s
was written using as much as possible experimental variables to provide a direct physical
meaning to the stress and suction imposition as well as response in saturation and
deformation. However, most parameters for coupling have the drawback of not possessing a
true physical meaning on their own. This might be partly explained by the initial assumption
made in chapter 2 that the modelling approach should remain only at a macroscopic level so
that the possible microscopic effects are homogenised within the representative elementary
volume. As a consequence, parameters Ω, γ s and π H from Eqs. 6.6 to 6.8. are only
interpreted as fitting parameters. It will be explained in this section that whereas there are no
theoretical limits to the values of these parameters, it is recommended to remain within a
moderate range of reasonable values. The possible effects of a wrong calibration are
explored in the following on the basis of a parametric study. A particular attention is paid in
conclusion to the model response in undrained conditions.
6.5.1 Remarks on the calibration procedure: a case study

The present section is dedicated to the analysis of results obtained from the participation
of our research group to a Benchmark exercise of the MUSE network (Mechanics of
Unsaturated Soils for Engineering, EU Marie Curie actions). The objective of the
benchmarking exercise is to assess the modelling capabilities of several mechanical and
retention model on the basis of the experimental campaign carried out by Casini (2007). Part
of the experimental results were undisclosed for blind prediction purposes. As the scope of
this paragraph is to discuss the calibration procedure and the influence of parameters in
ACMEG-s, the results of the other models are not reviewed here. The complete comparative
analysis can be found in the final MUSE report on Benchmarks (D’Onza et al. 2008).
The tested material is compacted Jossigny silt. This soil (5% sand, 70% silt, 25% clay) is
classified as a low plasticity silt with clay. The relevance of the experimental program lies on
the fact that all samples were carefully prepared to reach a given identical initial state. This
common initial state is essential for the parameter determination and also for the
initialization of variables in ACMEG-s. The samples were prepared by mixing loose soil with
water to a target water content of 13% prior to undertaking one-dimensional static
compaction. The target dry unit weight for the compaction in layers is 14.5kN / m3 .
Practically, the initial state is taken after compaction. The variations in initial water content
have been observed to attain 9.4% while the initial dry unit weight variations reach 2%.
The provided experimental results are issued from 13 tests carried out in a suction
controlled triaxial cell (8 tests), in a suction controlled oedometer cell (4 tests) and in a
conventional oedometer cell (test EDO-sat). All tests basically feature an equalization phase
to the target suction followed by mechanical loading. Introducing pnet as the mean net stress,
Δqi
ε a as the axial strain and ηi = as the ratio of the variation of deviatoric stress over the
Δpnet ,i
variation of net stress during the triaxial compression ( i = comp ) or equalization phase
( i = equa ) , the test paths are summarized in table 6.1. σ v′ , σ v,net are respectively the vertical
effective stress and the vertical net stress.
In the present case, the large amount of experimental data makes the calibration of
ACMEG-s possible. In particular, the degree of saturation is available anytime (besides the
experimental stress variables net stress and suction) which enable to calculate the effective
stress paths. Also, the test EDO-200 features both mechanical loading cycles and suction
cycles; those cyclic loads can be used for the calibration of the water retention model.
Even though there should be no logical order for the calibration of the retention and
mechanical sub-models in ACMEG-s due to the full coupling, the calibration procedure
180 CHAPTER 6
should anyway follow a logical path, especially for the so-called “terms of coupling”
reviewed previously. Other examples of parameter determination are provided in appendix
B.
Starting with the mechanical model, the elastic parameters are calibrated on the basis of
compressions and shearing for over consolidated samples or along unloading reloading
lines. Even though the determination of elastic parameters is preferably carried out on fully
Table 6.1 Identification and stress paths for provided tests

type Nb ID Description Path
Suction-controlled Equalization s=200kPa, pnet = 20kPa
1 TX03
isotropic compression test Compression s=200kPa, pnet = 200kPa
Equalization s=200kPa,
Suction-controlled pnet = 20kPa,η equa = 0.375
Compressions (triaxial cell)
2 TX04 anisotropic compression

Compression s=200kPa,
test
pnet = 280kPa,ηcomp = 0.375
Suction-controlled pnet = 27 kPa,η equa = 0.750
test
pnet = 370kPa,η comp = 0.750
Suction-controlled pnet = 22kPa,ηequa = 0.875
test
pnet = 370kPa,η comp = 0.875
Equalization s=200kPa, pnet = 10kPa
Suction-controlled triaxial
5 TX01 Shearing s=200kPa, constant cell pressure,
test
ε a = 7.197%
Equalization s=200kPa, pnet = 10kPa
Suction-controlled
Compression s=200kPa, pnet = 20kPa
6 TX02 isotropically consolidated
triaxial test Shearing s=200kPa, constant cell pressure,
ε a = 12.291%
Triaxial tests
pnet = 20kPa,η equa = 0.750

Suction-controlled Compression s=200kPa,
7 TX06 anisotropically
consolidated triaxial test
Shearing s=200kPa, constant cell pressure,
ε a = 28.7%
pnet = 20kPa,ηequa = 0.750
Suction-controlled Compression s=200kPa,
8 TX07 anisotropically
consolidated triaxial test
Shearing s=200kPa, constant cell pressure,
ε a = 25.9%
Table 6.1 (continued) Identification and stress paths for provided tests
type Nb ID Description Path
EDO- Equalization s=0kPa, σ v′ = 1kPa
9 Saturated oedometer test
sat Loading σ v′ = 800 → 100 → 1600kPa
EDO- Suction-controlled oedometer Equalization s=10kPa, σ v , net = 20kPa

10
10 test Loading σ v , net = 800 → 100 → 1200kPa

Oedometer tests
11

12
Equalization s=200kPa, σ v , net = 20kPa

Loading σ v , net = 20kPa, s = 10 → 200kPa
EDO- Suction-controlled oedometer
13
200 test
σ v , net = 800 → 100 → 800kPa,
Loading
s = 200kPa
Loading σ v , net = 800kPa, s = 55 → 200kPa
saturated tests, the bulk modulus K ref and shear modulus Gref have to be determined here at
a non-zero suction level.
The elastic part of the isotropic compression TX03 (Fig. 6.15) (at a suction of 200kPa) is
used to determine the inverse of the slope K ref in the volumetric strain versus mean effective
stress plane. The calibration of the elastic shear modulus is processed with the elastic path of
the compression phase in test TX07 (Fig. 6.16). Note that Figs. 6.15 and 6.16 show not only
the calibrated elastic slopes but also the fitted elasto-plastic responses, which will be
discussed later.
The non-linearity exponent n e (see chapter 4) can be determined in the same stress-strain
planes (Figs. 6.15 and 6.16). However, given the little extent of the elastic ranges reviewed
here, it is difficult to estimate whether the non-linearity of the response in strain needs to be
refined. Consequently the exponent n e is set to a default value of 1.
Unlike the elastic moduli determined in the first place, the plastic compressibility
coefficient of reference β 0 must be calibrated exclusively on the basis of a compression
under fully saturated conditions. In the case where no saturated test is available, the
determination of β 0 would be theoretically possible on the basis of two compression tests at
non zero suction. Indeed a minimum of two tests would be required to be able to calibrate
both β 0 and Ω simultaneously, by fitting the actual non-saturated compressibility
coefficient β ( = β 0 + Ωs ) . In the present case, the calibration relies on oedometric results in
saturated conditions, test EDO-sat, Fig. 6.17.
In Fig. 6.17, the model matches the experimental results in the plastic part because the
reference (at zero suction) preconsolidation pressure p 'c 0 has been adjusted. The saturated
preconsolidation pressure is the product of parameter d by the initial critical state pressure
182 CHAPTER 6
0
Exp.
ACMEG-s
-0.01

v
-0.02
-0.03
-0.04
-0.05 5 6
10 10
Figure 6.15. Calibration of the bulk elastic modulus on test TX03. The parameters of plasticity in
isotropy are also calibrated in this plot. K ref = 1.5e8 Pa
Exp.
5 ACMEG-s
6 10
Deviator stress q(Pa)
5
3 10
0
0 0.3
Axial strain εa (-)
Figure 6.16. Calibration of the shear elastic modulus on test TX07. The parameters of deviatoric
plasticity are also calibrated in this plot. Gref = 1.2e8 Pa
p 'CR 0 that is a variable of initialization. The combination d × p 'CR 0 will thus define the initial
over-consolidation ratio for the material. Given the fact that all samples are equalized and
compressed from the same –target- initial state, it is assumed in the following that the initial
preconsolidation pressure is the same for all samples.
The parameter d = 2 governs the distance between the critical state line and the normal
consolidation line ( p 'c 0 = d × p 'CR 0 ) in plane ( ε v − ln p′ ) (Fig. 4.9 in chapter 4). This
parameter can be determined at any level of suction, so the results of tests TX06 and TX07 are
0
Exp.
ACMEG-s
-0.05

v
-0.1
-0.15
-0.2
-0.25
1000 104 105 106 107
Figure 6.17 Calibration of plastic compressibility coefficient ( β 0 = 17 ) on test EDO-sat.
used here. The critical state pressure of reference is then initialized to a value
p 'CR 0 = 2.7 ×104 by fitting the compression curve in saturated conditions (Fig. 6.17).
All the parameters and variables of initialization calibrated up to now are describing the
saturated stress-strain response exclusively. Further parameter determination will now enter
the domain of partial saturation. Knowing the saturated preconsolidation pressure p 'c 0 , the
preconsolidation pressure at non-zero suction level is calculated with Eq. 6.56:
⎧ pc′0 for 0 < s < se

⎪
pc′ ⎨ ⎡ ⎛ s ⎞⎤ (6.56)
⎪ pc′0 ⎢1 + γ s log ⎜ s ⎟ ⎥ for s > se
⎩ ⎢⎣ ⎝ eH ⎠ ⎥⎦
While γ s is a fitting parameter used to determine the shape of the LC curve, seH is a
physical variable defined as the updated air entry suction. seH must thus be determined
prior to γ s using the soil water retention model. For the first iteration, it is assumed that the
experimental soil water retention curve of the calibration test (here test ED0-200) was
determined at constant volume. In the present case where the calibration uses the main
wetting curve, three parameters of the water retention model have to be pre-determined: the
elastic coefficient K H , the plastic coefficient β H and the air entry suction of reference se (Fig.
6.18).
Assuming that for the initial state seH = se in Eq. 6.56, the parameter γ s is adjusted to fit
the preconsolidation pressure at non-zero suction. One level of suction is sufficient for the
calibration, here 200kPa with the calibration test being TX03, Fig.6.15. Using the same level of
suction and same Figure, one can now determine the parameter Ω to adjust the value of
plastic compressibility coefficient β , Eq. 6.7.
Concerning again the same isotropic compression test, material parameters risoe and c
help to predict a smoother transition between the elastic and elastoplastic behaviours. They
184 CHAPTER 6
respectively govern the beginning of progressive plasticization and the evolution of the
degree of mobilization. Both risoe and c are fitted on test TX03 (see Fig. 6.15).
The deviatoric behaviour has been only characterised for the elastic part up to now; the
plastic and shear resistance parameters can now be determined. The friction angle at critical
sate is calibrated in the plane deviatoric stress versus mean effective stress ( q − p′ ) , Fig 6.19,
on the basis of at least three shear tests (slope M), using the effective stress representation.
It can be noticed from Fig 6.19 that the critical state is overestimated for tests TX01 and
1.2
1
r
0.8
0.6
0.4
Exp.
ACMEG-s
0.2 4 5 6
1000 10 10 10
Fig. 6.18 Preliminary calibration of the initial air entry value on test EDO-200. The elastic and elasto-
plastic slopes had to be also pre-calibrated from this figure.
6 105
TX01 (Exp.)
TX02 (Exp.)
4.8 10
5 TX06 (Exp.)
TX07 (Exp.)
CSL ACMEG-s
3.6 105
2.4 105
5
1.2 10
0
0 1 105 2 105 3 105 4 105 5 105
Figure 6.19 Calibration of the critical state line.

TX02. This is due to the fact that the critical state line is assumed to be unique whatever the
level of suction. Furthermore the model does not feature any cohesion in the effective stress
plane ( q − p′ ) . Given the fact that the most accurate modelled point of critical state is that of
test TX07, the latter will be used to calibrate the parameters of the yield surface of the
e
deviatoric plastic mechanism. Material parameters rdev and a have the same function as their
counterparts risoe and c , but for the deviatoric behaviour. They define the level of deviatoric
stress above which yielding upon shearing starts, and define the shape in deviatoric stress
versus axial strain plane (Fig. 6.16). The material parameter α is eventually adjusted to
refine the volumetric predictions during shearing. It is calibrated in plane volumetric strain
versus axial strain ( ε v − ε a ) for shearing tests on one of same reference shear test, TX07, Fig.
6.20.
Once the parameters for the stress strain model are completely determined, the model for
soil water retention curve can in turn be (re-) calibrated. K h is the slope of the “scanning
lines” of the retention curve, i.e. the slope of the elastic parts between the main wetting and
drying curve. The unique way to calibrate properly K h is to refer to an experimental test
with suction cycles. Test EDO-200 is thus suitable for this determination (Fig. 6.21). The
plastic parameter βh was also calibrated on test EDO-200 along the main wetting curve.
se was determined previously as the air entry value of reference. The effect of mechanical
straining on the updated air entry value seH is fitted by the means of the coupling parameter
π H . Its determination requires at least two retention curves at different initial void ratios.
The value of the parameter is then chosen in order to fit the deviation between the updated
air entry values of the first and second wetting-drying paths of test EDO-200 (Fig. 6.21). sDI
is the initial yield suction, that is the suction of slope change in the unsaturated domain. The
parameter sDI pilots the width of the capillary hysteresis (Fig. 6.21). Lastly, S res is the
residual degree of saturation (facultative). It can be calibrated only if very high suctions are
reached during experimental testing. Here, in absence of data on residual states of
saturation, S res has a default value of 0.01.
Due to the coupling parameter π H , the air entry suction (and the complete retention loop)
are shifted anytime a deformation occur. This is the case during suction cycles (drying and
wetting in a deformable material). Consequently, the response of the retention model is no
more bi-linear in the ( S r − ln s ) plane. Despite this non linearity, the parameters K H , β H still
provide a consistent response. A re-adjustment of retention parameters could be necessary in
other case studies.
Table 6.2 summarizes the list of determined parameters for Jossigny silt.
186 CHAPTER 6
0
Exp.
ACMEG-s
0.02
0.04
0.06
0.08
0.1
0 0.1 0.2 0.3 0.4
Figure 6.20 Calibration of parameter α on test TX07.
0.96
r
0.8
0.64
0.48
0.32 Exp.
ACMEG-s
4 5 6
10 10 10
Figure 6.21 Calibration of the soil water retention curve on EDO-200.

Table 6.2 Jossigny silt parameters for ACMEG-s

Symbol Description Value
K ref Bulk modulus 1.5e8 Pa
Elastic
parameters
Gref Shear modulus 1.2e8 Pa
ne Elastic exponent 1
φ′ Friction angle 31 °
Stress- β0 Compressibility coefficient 17

strain Plastic α Dilatancy coefficient 0.75
model parameters a 0.02
b 0.01
c 0.0001
d 2
Limits of e
rdev Initialisation of dev. mechanism 0.01
elastic
domain risoe Initialisation of iso. mechanism 0.1
se Air entry value 3000 Pa
Kh Elastic coefficient 18
βh Plastic coefficient 10
Retention model πH Coupling parameter 3.5e5 Pa
sDI Drying yield suction 1.5e4 Pa
S res Residual degree of saturation 0.01
6.5.2 Influence of coupling parameters

The logical parameter determination procedure has been presented for the studied case.
However, the procedure is not unique and depends on the amount of available data for
calibration. The “terms of coupling” should in any case receive a particular care in their
determination. Also, some implicit arbitrary choices have been made in the present
parameter determination. In the following, the model predictions using this set of
parameters are confronted with the experimental data. It will be demonstrated that due to
the choices made, some stress-strain paths are more favourably described by the prediction.
In the cases where discrepancies are identified, the influence of involved parameters is
discussed more in depth. The purpose here is not to carry out a complete parametric study
(see appendix C) but rather to investigate the influence of some parameters on the coupled
effects and the global model response.
The unique way to calibrate the soil water retention model including the effect of skeleton
deformation was to use the cyclic path of test EDO-200. The modelled variations in degree of
saturation in Fig.6.21 are large (variations of 102% with respect to the initial sate) which
should entail a strong influence of the retention variables on the stress-strain response. It is
proposed then to assess first the model capabilities in the prediction of the oedometric tests
(EDO-10, EDO-50 and EDO-100) from the mechanical (strain response) and retention
(saturation response) viewpoints. The results of the model predictions to the oedometric
paths are plotted in Figs. 6.22 to 6.24.
188 CHAPTER 6
-0.05
Exp. 1 Exp.
ACMEG-s

ACMEG-s
0
r
0.05
0.8
0.1
0.15
0.4
(b)
(a)
0.2 4 5 6 0 0.1 0.2
10 10 10
Vertical effective stress σv(Pa) Volumetric strain εv (-)
Figure 6.22 Model prediction for test EDO-10 (suction 10 kPa). Even tough the equalization phase was
simulated, only the results of the compression phase are presented above.
The results of back prediction are observed to be accurate enough for the highest suction
level (Fig. 6.24), both for the prediction of volumetric strain and degree of saturation. In the
case of the lowest suction level (Fig. 6.22), the volumetric strain is underestimated. Even
though the elastic and elastoplastic slopes are well captured in plane ( ε v − ln σ v′ ) , Fig. 6.22a,
the preconsolidation pressure is slightly overestimated which has noticeable consequences
on the volumetric prediction.
By comparison, in the prediction of test EDO-50 in Fig. 6.23a, the preconsolidation
pressure appears to be consistent while the experimental data, while the plastic coefficient
cannot fit correctly the plastic slope at this suction level. The predicted degree of saturation
in all three cases shows satisfactory trends, while quantitatively the model misses the final
value of test EDO-50 (Fig. 6.23b). This is again due to the inappropriate plastic slope in Fig.
6.23a, entailing a clear underestimation of the final volumetric strain. Since the updated air
entry value depends only on the generated volumetric strain, the whole degree of saturation
is logically also underestimated (a larger variation of volumetric strain would produce a
higher final degree of saturation, Fig.6.23b).
Overall, the very large variations of the degree of saturation are fairly predicted which
indicate a correct calibration of the parameter π H (coupling between the air entry suction
and volumetric strain. The relatively high value of π H indicates a strong dependency of the
retention model on the volumetric information.
The model prediction to triaxial stress paths is now investigated, mainly from the
viewpoint of the response in saturation. Figs 6.25 to 6.27 show the prediction for some of the
triaxial tests from the database, namely tests TX02, TX08 and TX07.
-0.05
Exp. 1 Exp.
ACMEG-s

ACMEG-s
0
r
0.05
0.8
0.1
0.15
0.5
(b)
(a)
0.2 5 6 0 0.1 0.2
10 10
Figure 6.23 Model prediction for test EDO-50 (suction 50 kPa). Even tough the equalization phase was
simulated, only the results of the compression phase are presented above.
0.9
-0.04 (a) (b)
Exp.
ACMEG-s
0.8
r
0.7
0.04
0.6
0.08
0.5
0.12 Exp.
ACMEG-s
0.4
5
10 10
6 0 0.04 0.08 0.12
Figure 6.24 Model prediction for test EDO-100 (suction 100 kPa). Even tough the equalization phase
was simulated, only the results of the compression phase are presented above.
The prediction of deformation and degree of saturation for test TX08 are satisfactory even
though the degree of saturation at the end of equalization is slightly underestimated (Fig.
6.25). Due to the assumptions made for the calibration of the critical state line on test TX07,
the model overestimates the critical state point for test TX02 in terms of deviatoric stress
(Fig.6.26a). It also appears from Fig. 6.26a that the sample was initially more preconsolidated
than predicted. However, it was assumed that the determined preconsolidation pressure at
the end of compaction was valuable for all samples, so p 'c 0 cannot be adjusted here. For this
triaxial test, the dilatant trend for the volumetric strain is correct (Fig. 6.26b), but not
quantitatively comparable with the experimental results. On the other hand, the predicted
degree of saturation during shearing has a correct average level (Fig. 6.26c).
190 CHAPTER 6
Test TX07 was used to calibrate the elastic and plastic deviatoric parameters of the stress
strain model. The volumetric strain is consequently well fitted (Fig. 6.27a). The predicted
degree of saturation is however highly overestimated. The largest increase in the degree of
saturation predicted by the model occurs during the compression phase where S r moves
from an initial value of 0.41 to a value of 0.61 at the end of the anisotropic compression.
This unexpected increase in the degree of saturation is explained by the fact that the point
of retention state at the beginning of compression is located on the main wetting curve. Due
to the high value of coupling parameter π H , the compression process is at the origin of a
large increase in the modeled air entry value seH . The retention model thus predicts plastic
increments of degree of saturation to remain on the main wetting curve (in a similar manner
to that of Fig.6.13).
These predicted changes are not in agreement with the experimental data. The modelled
response in saturation is however consistent with the simulations of oedometric tests.
-0.04 (b)
(a)
0.7
r
0
0.6
0.04
0.5
0.08
0.4
Exp. Exp.
0.12
ACMEG-s 0.2 ACMEG-s
4 5 6 0 0.04 0.08 0.12

10 10 10
Mean effective stress p'(Pa) Volumetric strain εv (-)
Figure 6.25 Model prediction for the anisotropic compression phase of test TX08.
Indeed, in the case of Fig.6.23, a volumetric strain of the order of 0.1 produces a change of
almost 0.3 for the degree of saturation. Given the order of magnitude of the volumetric strain
during the compression phase of test TX07, one could have expected similar important
changes in the degree of saturation.
Assuming that the parameters of the mechanical model are correctly calibrated, the
unique way to improve the simulation of Fig. 6.27b would be to re-calibrate the coupling
parameter π H . The results of Fig. 6.28 present the simulations of test TX07 using an updated
( )
lower value for parameter π H 100 × 103 . The purpose here is essentially to assess the
influence of this parameter on the prediction of the degree of saturation S r . In the case of test
TX07, S r is modeled with a significantly improved accuracy, while the overall mechanical
response is slightly affected by the parameter update.
(a) -0.04 (b)

2 10
Deviatoric stress q(Pa)
-0.02
v
-0.01
5
1 10
0
Exp. 0.01 Exp.

ACMEG-s ACMEG-s
0
0 0.04 0.08 0.12 0.16 0 0.04 0.08 0.12 0.16
Axial strain εa (-) Axial strain εa (-)
(c)
0.48
r
0.42
Exp.
0.36 ACMEG-s
0 0.04 0.08 0.12 0.16

Figure 6.26 Model prediction for test TX02

192 CHAPTER 6
(b)
(a)
0.7

-0.03
r
v
0.6
0
0.5
0.03
0.4
0.06
Exp. Exp.
ACMEG-s 0.2 ACMEG-s
0.08
5 6 0 0.1 0.2 0.3
10 10
Mean effective stress p'(Pa) Axial strain εa (-)
Figure 6.27 Prediction of the volumetric and retention behaviours for test TX07. This figure should be
read along with previous Figs. 6.16 and 6.20 (calibration).
The new value of π H showed similarly improved prediction for the triaxial tests where
the variations of S r are low (tests TX04, TX09). The parameters of the mechanical model
have been deliberately not adjusted here in order to quantify only the effect of the coupling
parameter π H alone.
However, trying in turn to back-predict the unsaturated oedometer tests, for instance test
EDO-100 in Fig. 6.29, the updated parameter π H is at the origin of poor quantitative
predictions for the degree of saturation (Fig. 6.29b), where the difference in final saturation
between the prediction and the experimental data reaches 0.21. Furthermore, with a more
complex loading path of the type of EDO-200 (Fig. 6.30) the response to suction changes
remains satisfactory along the main wetting curve. However, the updated retention model is
missing the large changes in the degree of saturation upon mechanical loading. Here again,
the predicted degree of saturation at the end of the suction and mechanical cycles is highly
underestimated.
This analysis demonstrates that to account for high variations in the degree of saturation
in response to mechanical loads, the terms of coupling should be strong enough.
The results from the oedometric tests are apparently presenting inconsistencies with those
obtained with the suction-controlled triaxial cell. There is not an a priori best choice between
the two contradictory sets of results for the determination of parameters of the retention
model. However, it happens that the oedometric tests are more exhaustive on the retention
viewpoint (cycles of suction and mechanical cycles) and that the purpose of the present
research work is to valorize as much as possible the terms of coupling. Furthermore, an
overestimation of the degree of saturation is preferable to the underestimation, to remain on
the side of safety.
0.6
(a) (b)
5
0.54 6 10

r
0.48
5
0.42 3 10
0.36
Exp. Exp.
ACMEG-s ACMEG-s
0.3 0
0 0.1 0.2 0.3 0 0.3
Axial strain εa (-) Axial strain εa (-)
0
Exp.
ACMEG-s
0.02
0.04
0.06
0.08
(c)
0.1
0 0.1 0.2 0.3 0.4
Figure 6.28 Model prediction for test TX07 with the second set of parameters
194 CHAPTER 6
0.9
(a) -0.04 (b)
Exp.
ACMEG-s
0.8
r

0
0.7
0.04
0.6
0.08
0.5
0.12 Exp.
ACMEG-s
0.4
0 0.04 0.08 0.12 5
10 10
6
Volumetric strain εv (-) Vertical effective stress σv(Pa)
Figure 6.29 Model prediction for test EDO-100 with the second set of parameters
0.96
r
0.8
0.64
0.48
0.32 Exp.
ACMEG-s
4 5 6
10 10 10
Figure 6.30 Model prediction for test EDO-200 with the second set of parameters
6.5.3 Stress paths

The formulation of the effective stress σ ij′ (Eq. 6.60) is such that the product degree of
saturation times suction ( S r × s ) has a noticeable influence on the state of mechanical stress.
This dependency is particularly noticeable for conditions of extreme drought, where a very
low degree of saturation causes the term of coupling ( S r × s ) to vanish (see chapter 3, part
3.6.3).
σ ij′ = σ net
′ ij + S r sδ ij (6.60)
Furthermore, if very large variations of degree of saturation are occurring at a quasi-

constant level of suction, the effective stress σ ′ will be considerably altered which might
subsequently provoke large straining. On the other side, large mechanical straining is likely
to entail important variations in the degree of saturation, even if the suction remains
constant. In the most general case, both phenomena are simultaneous and the two processes
can feed each other.
The case study of previous part 6.5.2 provides some evidence of this double-way
coupling. It is proposed to investigate the prediction carried out for test TX07 from the point
of view of effective stress path. The area of focus is only the compression phase, which is at
the origin of the larger volumetric strains and variations in degree of saturation. The applied
stresses during compression in test TX07 are net stresses such that the ratio of deviatoric
stress over mean net stress remains equal to ηcomp = 0.75 . The matric suction remains
constantly equal to 200kPa during the anisotropic compression. Virtually, if the degree of
saturation remained constant, the stress ratio deviatoric stress variation over the variation of
mean effective stress should also equal ηcomp = 0.75 . This scenario can be modelled with
ACMEG-s by setting the coupling parameter π H equal to zero (Fig. 6.31a). In the original
parameter determination though, the parameter π H equals 3.5e5. Due to the conditions at
the beginning of the anisotropic compression (i.e. the point is on the main wetting curve),
any positive volumetric strain will provoke an increase in the air entry value seH (Eq.6.8) and
also plastic increase in the degree of saturation (Fig. 6.31b). Modifying the degree of
saturation has an effect on the inputs for Eq. 6.60. As the second term on the right hand side
of Eq. 6.60 only acts isotropically, the mean effective stress will be incremented while the
deviatoric stress remains un-changed with respect to the previous reference case. As a result,
the stress path in Fig. 6.31a diverges from that of the first case. As the overall effective stress
is larger than previously, more volumetric strain has to be produced to sustain the stress
state, and the degree of saturation increases significantly during compression (Fig. 6.31b). A
third test is added in Fig. 6.31 to show the influence of larger values for π H .
5
1.7 10
π =3.5e5 π =1e6
H H
0.72
5
1.4 10 π =0
r
Deviatoric stress (Pa)
0.6
5
1 10
π =3.5e5
η=0.75 π =1e6 0.48 H
H
4
6.8 10
0.36
4
3.4 10 π =0
H
0.24 (b)
(a)
0 5 5 5 0 0.01 0.02 0.03 0.04
1 10 2 10 3 10
Figure 6.31 Influence of parameter π H on the effective stress paths.

196 CHAPTER 6
It has been shown in this paragraph that the generalized effective stress is not an
experimentally controlled variable. Using a fully coupled approach, it might happen that the
stress paths possess non linearity in the conventional deviatoric stress plane. This feature
should be taken into account for the parameter determination, for instance in the
determination of critical state. It is also demonstrated again that the coupling parameter π H
alone might have a considerable influence on the predicted results.
6.5.4 Undrained response

The coupling parameter π H also has a strong impact on the predicted response to an
undrained test. According to the literature review from chapter 2, the behaviour of
unsaturated soils in undrained conditions may vary from one material to the other. The
characteristics of undrained shearing in unsaturated soils can be summarized as follows:
(i) There is a volume change upon undrained shearing due to the air compressibility.
(ii) The variations of pore air pressure are presumably low with respect to the change
in pore water pressure.
(iii) The suction and degree of saturation might increase or decrease, depending on the
sign of the volume change.
Fig. 6.32 shows a parametric analysis of the influence of parameter π H on the prediction
of volumetric strain (Fig. 6.32a), matric suction and degree of saturation (Fig. 6.32b). It can be
observed that the coupling parameter has a strong influence on the sign of suction changes
(either positive or negative) and on the amplitude of the variations of degree of saturation.
This accounts again for the need of an accurate determination of the coupling parameters
using whenever possible reference tests featuring several alternative cycles of stress and
suction loads, such as test EDO-200.
0
(a) (b)
0.69
0.02
π =0
r
0.66 π =3.5e5 H
H
0.04
0.63
π =0 π =3.5e5
H H
π =5e5 0.06
H
0.6
0.08
0.57 π =5e5
H
0.1
5 5 5 0 0.1 0.2 0.3
1 10 3 10 5 10
Matric suction s(Pa) Axial strain εa (-)
Figure 6.32 Influence of parameter π H on the undrained response.

6.6 Conclusions
This chapter has presented how the double-way coupling is managed in ACMEG-s. The
so-called terms of couplings have been reviewed from the constitutive equations of the
mechanical model and retention framework. A third category of models has been introduced
as the “hydraulic part” of the analysis, and focuses on the understanding of the transfer of
fluids.
The management of coupling is essentially piloted by the constraints of numerical
implementation. Consequently, a thorough review of the local integration routines for the
behavioural laws has been proposed, with a particular focus on the algorithm for research of
plastic mechanisms. Starting from the integration routine of the reference saturated model
(Hujeux 1985), the programme has been generalised to partial saturation.
The main choices for the double-way coupling between the mechanical model and the
retention model have been detailed and are justified by the need for an accurate solution of
governing equations while keeping reasonable calculation times. The compatibility with the
finite element code LAGAMINE also determines some of the choices made.
The finite element formulation has been presented and the implemented models for water
and air flow have been discussed. It was chosen to make use of available laws for the transfer
of fluids and evolution of permeability coefficients with saturation.
The model response in a fully-coupled mode has been validated on a number of materials
from literature. The predicted volume and saturation changes show satisfactory trends and
the prediction of the swelling pressure evolution was refined. The complete calibration
procedure has been reviewed on the basis of a case study on which the predictive capability
of the model is assessed again. The chapter ends with a parametric study for some of the
terms of coupling completed with recommendations on the calibration choices.
6.7 References
6.7.1 Publication by the author
Nuth M., Laloui L. (2008) “New insight into the unified hydro-mechanical constitutive modelling of
unsaturated soils”. IWUS08 Trento (In press).

40(3), pp. 405-430.
Bear J. (1972). Dynamics of fluids in porous media, Elsevier.
Blight G. E. (1967). "Effective stress evaluation for unsaturated soils." ASCE Journal of Soil Mechanics
and Foundation Engineering. 93(SM4), pp. 607-624.
Brooks R. H., Corey A. J. (1964). "Hydraulic properties of porous media." Hydrology Papers(3).
Charlier R. (1987). Approche unifiée de quelques problèmes non linéaires de mécanique des milieux
continus par la méthode des éléments finis. Liège, Université de liège. PhD. Thesis.
Collin F. (2003). Couplages thermo-hydro-mécaniques dans les sols et les roches tendres partiellement
saturés. Liège, Université de liège. PhD. Thesis.
D'Onza F., Gallipoli D., Wheeler S. (2008). Comparison of constitutive models for the prediction of
mechanical and water retention behaviour in unsaturated soils. MUSE Benchmark B9.
198 CHAPTER 6
Gardner W. R. (1958). "Some stready state solutions of the unsaturated moisture flow equation with
application to evaporation from a water table." Journal of American Soil Science Society.
Hujeux J. (1985). Une loi de comportement pour le chargement cyclique des sols. Génie Parasismique.
Paris, Les éditions de l'E.N.P.C.: 287-353.
Liakopoulos A. C. (1965). "Theoretical solution of the unsteady unsaturated flow problems in soils."
Bulletin of the International Association of Scientific Hydrology 10, pp. 58-69.
Lloret A., Romero E., Villar M. (2004). FEBEX II Project: Final report on thermo-hydro-mechanical
laboratory tests, ENRESA.
Mandel W. (1965). "Généralisation de la théorie de Koiter." International journal of solids and structures 1,
pp. 273-295.
Modaressi A., Modaressi H., Picuezzu E., Aubry D. (1989). Driver de la loi de comportement de
Hujeux.
Nielsen J. M., Jackson J. W., Cary J. W., Evans D. D. (1972). "Soil water." American Society for Agronomy
and Soil Science.
Ortiz M., Simo J. C. (1986). "An analysis of a new class of integration algorithms for elastoplastic
constitutive relations." International Journal for Numerical Methods in Engineering 23, pp. 353-
366.
298.
Picuezzu E. (1991). Lois de comportement en géomécanique. Modélisation, mise en oeuvre,
identification. Paris, Ecole Centrale Paris. PhD. Thesis.
Richards L. A. (1931). "Capillary conduction of liquids through porous medium." Journal of Physics 1,
pp. 318-333.
Roscoe K., Burland J. (1968). "On the generalized stress-strain behaviour of 'wet' clay." Engineering
Plasticity, Cambridge, pp. 535-609.
Sharma R. S. (1998). Mechanical behaviour of highly expansive unsaturated clays, Oxford. PhD.
Thesis.
of unsaturated soils." Soil Sciences Am. Soc.(44), pp. 892-898.
Weast R. C. (1971). Handbook of chemistry and physics, CRC-Press, Cleveland.
Zienkiewicz O., Taylor R. L. (2000). The finite elment method (Fith edition), Butterworth-Heinemann.
7. Application: numerical modelling of earthdams
7.1 Introduction
This chapter proposes to apply the developed constitutive model to the simulation of
earth dams, with the particular case study of Mirgenbach dam (France). Earth dams are only
one typical example of application for ACMEG-s. Indeed, the range of boundary value
problems that would require such a multi-coupled constitutive framework is large enough,
starting from problems of drainage of natural slopes in movement to pore-collapse and
heave in foundation soils due to variations of the phreatic level and rain. These problems are
in essence highly coupled, with interdependency between volumetric deformations, flow of
fluids and water content. It is thus a major challenge to provide a prediction of the complete
material behaviour in the short and long term, anticipating any plastic behaviour, i.e. large
deformations, and possible zones of failure. The use of ACMEG-s, combined with the finite
element analysis and the fluid flow laws implemented in LAGAMINE, make possible the
required spatial and temporal descriptions of the problems.
It has been seen in the previous chapters that the sensitivity of the couplings implemented
in the constitutive framework is such that the model requires a careful calibration. More
precisely, the use of constitutive laws of ACMEG-s in simulations should practically be
justified only if it is assorted with a methodical characterization of the materials from the
point of view of permeability and geo-mechanical properties in unsaturated conditions. The
supplementary parameter determination is worth the effort only if the suction subscribes to
sufficiently high ranges. Otherwise, the constitutive models for saturated soils should
provide a sufficient approximation, even for the non-zero suctions below the air entry value.
ACMEG-s, used at zero suction, presents the advantage of being a performing model for the
saturated behaviour. Most of the saturated parameters can be calibrated on the basis of a
standard geotechnical characterization.
Considering the above specifications, it is proposed to focus on the particular case of earth
dams, and assess this type of application with respect to the added value of the proposed
constitutive framework. Earth dams, also called embankment dams, are classified in two
categories (Wahlstrom 1974):
(i) Homogenous dams, that are constructed entirely from a more or less uniform
natural material
(ii) Zoned dams, which contain materials of distinctly different properties in various
portions of the dams.
The category of a dam is mostly conditioned by the amounts and types of materials (e.g.
alluvia, weathered bedrock, slide-rock accumulation) available locally for its construction.
Usually, the largest embankments are of the zoned type, with an impermeable blanket or
core (Fig. 7.1), while the low-rise earth dams can be built with a homogenous single fine-
grained material.
200 CHAPTER 7
Figure 7.1 Schematic layout of an earth dam with an upstream sloping core of low permeability
(modified from Hirshfeld and Poulos 1973). The part of the shell above the line of seepage is a zone of
partial saturation, justifying the utmost necessity to account for effects of suction in the conception
and service state of the dam.
Embankment dams represent a favourable application to the constitutive model ACMEG-

s for a number of reasons. Firstly, the earth dams are constituted of a little number of
remoulded materials, which are well characterized via soil mechanics studies prior to the
construction. The different layers can be considered as very homogeneous and have a
reduced probability of random variations of mechanical, retention and hydraulic properties
in space with respect to natural undisturbed soil arrangements.
The fact that the embankment is man-made also indicates that there is a control of the
compaction of the backfill materials and, to a certain extent a control of the water content.
The hydrogeology of the construction site is also usually well characterized, which provides
an excellent picture of the boundary conditions for the problem. Furthermore, the reservoir
dams are by definition designed to retain water (Fig. 7.1) and thus obviously imply water
flow problems, zones of complete and partial saturation in water, and deformations linked to
water level variations.
Even though it is not the scope of the present PhD. thesis to assess the earth dams from
the perspective of risk management, some case histories of dam failure are known to have a
strong impact. On February 12 2005, the failure of Shadi Kor earth dam, in Pakistan, has been
for instance the cause of 60 deaths, five hundred missing and thousands of indirect refugees.
The dam collapsed immediately after weeks of rain, and the failure has been reportedly
attributed to a combination of inappropriate design, poor quality of construction, lack of
controls and unexpected heavy rainfall periods. The sudden failure of the 35m high and 500
wide dam has been at the origin of a massive flood affecting more than 160 km2 of
agriculture fields and swept a number of bridges away. Former Prime Minister Benezir
Bhutto insisted of further investigations to characterize the dam rupture and identify its
origin, but no official report explaining the exact failure mechanism has been published yet.
On June 5 1976, the Teton Dam, USA, also failed in a catastrophic way, with an extremely
short time lapse of 5h between the observation of the first leak and the total collapse of the
dam. A unique series of slides covering the various steps of the event have been published
(Fig. 7.2) and illustrate the mode of rupture of the dam. Some inappropriate remediation
works had been engaged right after the first leakage observation, but these measures were
ineffective and even caused the death of workers. Rexburg, a city 20km away of the dam,
was destroyed by 80% percent of its infrastructure due to the extensive flood caused by the
water of the artificial lake. The water could be finally retained downstream by American
APPLICATION TO EARTH DAM MODELLING 201
Falls Dam which served as a buffer. The dimensions of the dam were 93m of height and
823m of length on the top. The leak was located 40m below the top.
Such embankment failures happen worldwide and at different scales. It appears here that
the design of the structures was not thought to cater for extreme climatic conditions. That
type of external loading has been called environmental loads in the PhD. Thesis. Earth dams
should typically be designed in terms of sustainable development: they should not only
ensure their primary actual function but also be able to sustain millennial rains or
earthquakes in the indefinite future. A dam failure does not only annihilate the primary dam
function that is ensuring retention of water, but it also exerts paradoxically a destructive
action downstream.
With this respect, the advances brought by a fully coupled constitutive framework do not
only enable to verify the dimensioning of the earthen structures, but enables also virtually to
test the effects of any extreme environmental load.
In the first part of the chapter will be briefly reviewed the current practice in the design
and modelling of earthdams. Then, it is proposed to apply the developed coupled model
ACMEG-s to the case study of Mirgenbach earth dam, which raised the need for such an
advanced modelling framework in link with partial saturation.
Figure 7.2 The unique series (partial) of slides of the Teton Dam failure.
202 CHAPTER 7
7.2 Earth dam practice and verification: a brief review

This short section is dedicated to reviewing the practice in modelling embankment dams,
essentially from the geotechnical point-of-view. The advances of modelling with hydro-
mechanical couplings are also mentioned in this part. The purpose is to review the possible
interpretation of flow, deformation and failure in earth dams, and to provide a point of
comparison for the developed model.
7.2.1 Mechanisms of failure

The investigation on the causes of failure of 82 earth dams worldwide (Mallet and
Pacquant 1951), table 7.1, provides a clear picture of the principal factors for embankment
safety. It appears that the mechanical failure is almost invariably linked to hydraulic
problems, or in other words that the loads due to water are preponderant for the dam
verification.
The present paragraph is thus essentially justified by the need for setting the limits of the
present study with respect to the real mechanisms of failure in earth dams. The main causes
of earth dam failure reported in literature are (a) the internal erosion, and (b) the slope
instability (shear resistance).
(a) The internal erosion process, for instance described by Charles et al. (1995), is a
consequence of the flow of water through a porous medium. The flow is at the origin of the
removal of solid material from the clay core or backfills. This elimination of soil particles will
create weak points in the structure, and can entail fractures and deformation in the mass. The
erosion is more likely to start at the exit points of seepage, from which a backwards
alteration is developed in the soil. This phenomenon, called “piping” can reach the water
source: in this case, a sudden breakthrough of water occurs into the pipe and accelerates
internal erosion, which at a given scale can blow the complete structure down. It is very
probable that internal erosion starting from a local point is at the origin of the destruction of
Teton dam presented previously. Other particular modes of more diffuse erosion (suffusion,
hydraulic fracture, hydraulic separation, dispersive soils, see Charles et al. 1995) are not
discussed here. Due to the localized nature of the process of erosion only a specific modelling
context could comprehend the phenomenon, and the conventional continuum mechanics
approach would not be sufficient. This is not the scope of the present PhD. thesis to model
the localized processes (see for instance Péron 2008) consequently, the failure modes due to
erosion will no longer be discussed from now on.
(b) The second and most geotechnical category of failure mode is the slope instability.
Ground movements in the slopes of the embankment can be of the slow type (creep) or else
of the sudden type (slide and mudflow). In the case of the movements due to creep, there is
no characteristic slip (or failure) surface, and the deformation occurs in the mass, with a
visco-plastic behaviour. On the contrary when a slide occurs, the displacements are localised
in a thin layer of material (defining the slip surface, Fig. 7.3) and the process is fast enough.
Table 7.1 Statistics on earth dams failures (from Mallet and Pacquant 1951)
Causes of destruction Number % of the total of destructions
Under-sized spillways 32 39
Internal erosion 15 18
Infiltration along penstocks 14 17
Misc. causes 21 26
Total 82 100
Figure 7.3 possible shape of a sliding surface in a homogeneous embankment.
The slope stability analysis is the historic focus of geotechnics and shall not be thoroughly
reviewed in the following.
7.2.2 Overview of practice in earth dam modelling .

The recommendations for the dimensioning of earth dams from the geotechnical point of
view can be extracted from norms as illustrated in table 7.2. The standard dimensioning of
embankments relies on the type of impervious element and the materials of construction.
The recommended geometry has to be verified with regards to slope stability, by checking
that the safety factor of the slopes (i.e. the ratio of stabilizing forces to the driving forces)
remains anytime greater than 1. Some weighting coefficients are inputted in particular cases
of loading.
Concerning the materials of the core (that have a clay fraction over 5%), the optimum
water content wopt is targeted during placing and compaction, with possible adjustments by
free air drying or watering. The practical tolerance for the target water content is the
optimum water content minus 2% in the lower layers and wopt plus 2% in the upper layers.
The compaction is a mechanical process that makes use of rammers or vibratory rollers and
the generated pore pressures might dissipate more or less rapidly. A particular care is
recommended for the placing of materials at the interface (if any) between the core and the
Table 7.2 recommendations for standard dimensioning (from SIA & DIN 4084)
Impervious Downstream
Fill Upstream slope
element slope
Rock fill Central core 1 :1.8 29.1° 1 : 1.8 29.1°

Inclined core 1 : 2.1 25.5° 1 : 1.8 29.1°
Upstream 1 : 1.5 33.7° 1 : 1.4 35.5°
blanket
Pervious silts Central core 1 : 2.0 26.6° 1 : 2.0 26.6°
Inclined core 1 : 2.3 23.5° 1 : 2.0 26.6°
Finer silts Central core 1 : 3.0 18.4° 1 : 2.5 21.8°

Inclined core 1 : 3.3 16.9° 1 : 2.5 21.8°
Homogeneous 1 : 3.0 18.4° 1 : 3.0 18.4°

dike
204 CHAPTER 7
rock fill, as fines may percolate through the contact.

The stability analysis might be carried out with conventional geotechnical methods which
are deliberately not developed here (e.g. Bishop’s method, Janbu) involving a simplified
analytical approach. The calculated safety factors may vary considerably according to the
method used, the definition of the slip surface, and the simplifying assumptions on the
ground water. The most obvious limitation of such methods is the assumption of non-
deformability of the materials.
Some more performing modelling approaches for embankments that make use of up-to-
date engineering softwares can be found in literature (e.g. Ould Amy and Magnan 1991,
Fredlund and Rahardjo 1993, Burghignoli et al. 2002, Alonso and Pinyol 2008). Two
examples of simulations will be discussed more in detail in next part 7.3 on the particular
issue of Mirgenbach earth dam.
The range and capabilities of available numerical tools for geotechnical dam engineering
are wide enough. One can carry out simulations for seepage alone or two and three-
dimensional coupled deformation and flow analyses. The state-of-the-art modelling tools
often feature partial saturation with various levels of couplings. As mentioned previously in
chapter 6, the processes of mere seepage have been well understood and modelled for a few
decades now and shall not be the focus of the present paragraph. Concerning the fully
coupled analyses, Ould Amy and Magnan (1991) for instance published some applications of
interest using a coupled elastic model of consolidation with the Finite Element code CESAR.
Fig.7.4 provides some typical results of the simulation of the phases of filling of the reservoir.
The analysis accounts for permeability variation with suction, which also implies the
Figure 7.4 Visualization of simulation results for Maurepas-Courance. (a) Contour lines of horizontal
displacements 15 days after reservoir filling (b) Contours of positive pore water pressures at the same
time moment (from Ould Amy and Magnan 1991).
definition of a retention curve. However, the partial saturation only influences the water
flow, and no mention is made to the dependency of the mechanical stress-strain model on
the suction level.
Alonso and Pinyol (2008) propose to predict the behaviour of San Salvador dam during
construction, using the finite element code CODE_BRIGHT, and the Barcelona Basic Model
(BBM) for the mechanical behaviour of the dam materials. The retention behaviour is
modelled by the means of Van Genuchten’s (1983) formulation. A cubic law describes the
relative water permeability with respect to the degree of saturation in water.
The analysis thus features multiple couplings between the mechanical, hydraulic and
retention models (see definitions in chapter 6), which enable to check the evolution of pore
pressures during construction, the deformation of the embankment (see Fig. 7.5) and features
of behaviour that are characteristic of unsaturated soils, like the pore-collapse upon wetting
(Fig. 7.6). Even though that type of simulations is the state-of-the-art practice in numerical
modelling of boundary value problems, the numerical analysis is still susceptible to present
inaccuracy due to the simplifications of the stress-strain model (e.g. transition between
saturated and partially saturated behaviours) and the retention model (no capillary
hysteresis nor dependency on void ratio).
It can be concluded from this overview that a number of efficient modelling tools are now
available on the market and can answer specific design and verification needs. Yet the
simulations are usually assorted with strong simplifying assumptions either on the
geometry, the mechanical and hydraulic behaviours or the partial saturation. The proposed
model ACMEG-s implemented in LAGAMINE finite element code, shall contribute to
answer specific matters due to partial saturation in boundary value problems, as developed
Figure 7.5 Contours of positive pore water pressure and dam deformation, San Salvador dam (from
Alonso and Pinyol 2008)
Figure 7.6 Deformed configuration of a dam because of shell collapse (from Alonso and Pinyol 2008)
206 CHAPTER 7
below with the help of a case study.
7.3 Reproducing the failure of Mirgenbach dam

This section is dedicated to the simulation of the case study of Mirgenbach earth dam,
France. In its initial design and function, this structure had no particular research frame and
was in consequence not specially instrumented with respect to similar works. It has been a
deliberate choice to apply the constitutive model and numerical tool on such a conventional
engineering case study. The objective is indeed to address the issue of parameter
determination on the basis of available data, and besides to show the added value of the
coupled hydro-mechanical-retention framework in such types of problems.
The principal interest of Mirgenbach dam is the failure that occurred during the
construction of the embankment in 1982. As it will be explained later, both the upstream and
downstream slopes failed almost simultaneously, without any water being stored yet in the
reservoir. The hypothesis of failure by erosion (see part 7.2.1) is thus to be left apart while the
mechanisms of generation of excess pore water pressures shall be closely investigated.
The data used for the simulations have been provided by Electricité de France (EDF) in
the reports by Laigle (1993), Fry (2003), EDF (1979) EDF (1982), EDF-REAL (1983), CFGB
(1984) and in the PhD. thesis of Magnin (1995). The presented numerical simulations have
been carried out with the help of Christoph Berger in the framework of a semester project
(Berger 2008).
7.3.1 Presentation of the case study

The Mirgenbach embankement was designed to create an artificial reservoir of water for
the French nuclear power plant of Cattenom (Fig. 7.7) which construction started in 1979.
The cooling system uses water, but the river Moselle in the neighbourhood was identified as
not flowing enough to be used as a cold source, provided that it might have reached a critical
temperature above the legal threshold. The heat of the nuclear power plant is thus mostly
Figure7.7 localization of Cattenom nuclear power plant and satellite view of the present storage lake
and earth dam (Google Earth)
evacuated by the storage lake, which waters are later rejected into the river at acceptable
temperatures. The filling capacity of the lake is 7.2 millions of m3. The lake is not only used
as a cold source for normal service of the plant, but it is also designed to ensure a function of
protection: in the case where a circuit of cooling is short-circuited and the plant is stopped,
the lake alone suffices to absorb the excess heat. As a matter of fact, this nuclear power plant
of Cattenom is on the 3rd place of produced electricity in France in 2006. In this year the four
reactors of 1300MW power each produced 34TWh.
The energy production started in 1985 and is still active today. Actually, the start of
activity of the power plant had to be delayed due to an important failure of the earth dam
implanted at 1 km north-east from the plant. The stream flowing in the small valley cut out
by the dam gives its name (Mirgenbach) to the embankment and lake.
The dam can be categorized into the “homogenous” type, with a fairly low-rise of 22m
and a length of approximately 450m. The dam is constituted of compacted clay, taken from
the excavation of the construction site for the power plant. The foundation soil is marls with
a thin superficial weathered layer of clay. The crest width is 7m, and the maximum width at
the level of the natural ground is 142m (see cut and plane view in Fig. 7.8). The sides were
inclined of 1:3.5 on the lower part to 1:3 on the middle part to 1:2.5 on the highest part. The
drain system was built with a vertical layer of sand in the core connected to a horizontal
layer of sand on the bottom on the downstream side. The volume of the dam body is 325’000
m3. As the radius of curvature of the structure is infinite (straight embankment) a 2-
dimensional analysis is representative.
The schedule of events linked to the double failure of the earth dam has been reported in
internal publications (EDF 1982). On august 25, 1982, There was a first crack observed on the
top of the dam on the upstream side. In the evening of the same day, the moving part of the
Figure 7.9 Geometrical limits of the construction layers (with preliminary meshing) and phases of
construction in time
208 CHAPTER 7
Figure 7.8 Planned geometry of the embankment and failure zones (hatched) in August 1982.
dam had already slid down. The rupture is not described as a catastrophic collapse, but
rather as an ongoing movement which ended four meters lower. Fortunately, there were
neither casualties nor damage on construction material. Two weeks later on September 11, a
similar failure happened on the downstream side producing a symmetrical profile (Fig. 7.8).
The first signs of failure were indeed identified around 400 days after the beginning of the
dam construction which was still under progress. Yet the construction was close to
completion, with only 20’000m3 of clay remaining to be placed over a total of 350’000m3.
The shearing fissures have been reported to be very obvious, and the shape of the sliding
surface confirms that it was a progressive failure that concerned the material of the first
campaign (1981) but not the foundation, which is reportedly unaffected by the accident.
7.3.2 Qualitative interpretation(s) of the phenomena

The preliminary analysis retrieved from EDF archives gives an indication of the probable
origin for the double failure as well as a survey of the working instrumentation prior to and
after the failure. This analysis relies on the major phases of construction of the embankment,
which are summarized in Fig. 7.9.
A stop in the elevation is marked in year 1981 during the rainy season (that is winter
1981). It has been thought (EDF 1982) that due to the rainfalls (an average 6mm per day)
some parts of layer 1 could have been wetter than the average, knowing that the average
water content after the winter break is 26%, that is +3.5% above the assumed optimum. On
the other hand, the foundation is saturated and the fissured weathered shale makes an
aquifer horizon (up to + 3.50m) that has contributed to the saturation of the backfill of the
first campaign. Indeed, slight swellings were measured in layer 1, indicating water
absorption.
It is supposed that the relatively fast construction of the 5 layers in 1982 produced excess
(positive) interstitial pressures that were higher than expected. The water could then have
flowed from the core to the outside where stresses are lower. This could explain unexpected
swelling, and softening of the material at the toe of the sliding mass. The sliding could then
propagate to upper layers along the sliding surface drawn in Fig. 7.8.
An important complementary observation is that the drains were apparently not built
properly, so they have been sealed by clay particles from the embankment. As a result the
drainage network is reported as either fully saturated or even under hydraulic load, with
some meters of hydraulic head, see table 7.3. The non-effective drainage system cannot in
any way contribute to the dissipation of excess pore pressures, as if it was simply not built in.
Among the 17 water pressure cells placed in the embankment, only 2 show a level of pore
pressure that could justify a failure, this remark being formulated on the basis of a slope
stability analysis using Bishop’s method. More detail on the evolution of these pore
pressures will be given later in the text.
In summary, the failure has occurred exclusively in the clayey material of the
embankment and not in the foundation. The failure mode reveals large sliding masses on the
two sides of the embankment as a consequence of slope instability. The excess pore water
pressures are very probably at the origin of the instability.
7.3.3 Previous numerical modelling of Mirgenbach case

Hereafter are briefly reviewed the two existing numerical simulations of Mirgenbach
earthdam. These modelling campaigns can be used as references for comparison with new
simulations and represent tools for decision making on non-acquainted boundary conditions
or material parameters. Such simulations constitute also an appropriate illustration for
previous part 7.2.
P. Magnin (1995) proposes an analysis of the embankment using the model UDAM. This
model basically relies on two types of equations that are the flow of water and air and the
balance equation. The analysis thus provides description of partial saturation in the model
besides the deformation. The choice is made to model only half of the geometry, assuming a
perfect symmetry of the problem between the upstream and downstream sides. The mesh
used, plotted in Fig. 7.10 is relatively rough (238 nodes and 149 elements) and made of
quadrangles. The boundary conditions are blocked displacements at the bottom of the mesh
and no horizontal displacement along the symmetry axis. The slope surface is drained, as
well as the bottom of the mesh (to model the drainage system). The pore water pressures are
adjusted at boundary conditions to introduce suctions.
The suctions are adjusted at the construction of each layer and based on the water content
at optimum. The imposed suctions range from 14 kPa to 165 kPa, for degrees of saturation
between 0.86 and 0.98.
The constitutive laws use the theory of state surfaces (see chapter 2), meaning that the
analysis is fully elastic, both for the void ratio and degree of saturation. The mechanical
parameters are calibrated on oedometric tests carried out in undrained conditions. The
parameters for the soil water retention are fitted by using couples of degree of saturation-
suction available at optimum. The permeability coefficients make the object of a parameteric
study.
Table 7.3 Piezometric measurements in three points of the drainage system

Piezometer id Position (altitude) Measure
A4 161.50 166
B9 158.50 159.50
C4 161 161
210 CHAPTER 7
Figure 7.10 Mesh used by Magnin (1995).
200
150
Pore water pressure p (kPa)
w
100
50
0
Exp. (Cell B6)
Node 34
-50 Node 39
Node 160
-100
0 200 400 600 800
Time t(days)
Figure 7.11 Prediction of interstitial pressures, from Magnin (1995) Nodes 34, 39 and 160 are located in
the neighbourhood of the point of experimental monitoring.
The only published results are that of the prediction of the pore pressure in one of the
pressure cells (Fig. 7.11) in response to the clay elevation in construction steps. The
parametric study indicates that a saturated permeability of 5.3 ×`10−10 provides the most
accurate fitting of the pore pressure evolution. However, the simulation shows inconsistent
prediction of the interstitial pressures in the other cells.
Fry (2003) published a report, internal to Electricité de France, on the numerical modelling
of the failure of Mirgenbach dam. The report provides an interesting insight into the
evolution of the soil properties between the initial geotechnical characterization (1981) and
the material re-analysed after failure in 1982 (comparison made within the first layer). The
comparison shows consistent characteristics between the two experimental campaigns, and
accounts for the fact that the 1982 material is still well compacted and did not seem to evolve
since it had been placed. The goal of the modelling campaign is to analyse the causes of the
failure, but also to provide a numerical methodology and mechanical characterization
necessary to simulate the failure. A new hypothesis is introduced in these works, by stating a
possible effect of frost on the resistance of the material. In this context, joints have been
introduced in the model at the interface between the layer of 1981 and the new layers of 1982
to model the frost-induced foliation.
Figure 7.12 Detail of the mesh used by Fry (2005). The low deformability foundation layer is included
in the model. Horizontal and vertical joints can be easily located from this distorted grid.
The 2D- model represents the highest profile of the embankment (22m) and includes the
foundation. The total number of elements is 2601 with 3244 nodes. The joints (preferential
planes of shearing) are introduced horizontally at the interface foundation-backfill 1981 and
backfill 1981-backfill 1982. Vertical joints are introduced at the location of vertical fissures
observed on site after the failure (Fig. 7.12)
The mechanical displacements are blocked at the boundaries of the foundation mesh in a
conventional U-box way, and the water table is at the level of natural ground (that is at the
interface foundation-layer 1981). The loading scenario includes 37 loading steps based on the
recorded dates of construction. The material parameters are calibrated on the basis of
averaging of available experimental test results, with particular assumptions for the joints.
The mechanical model is of the type elastic-perfectly plastic. The permeability in the backfill
is equal to that taken by Magnin (1995) and no mention is made concerning the eventual
modelling of the drainage system. The calculation was carried out with the software FLAC.
The order of magnitude of settlements in surface is consistent with the in-situ measures,
i.e. a few centimetres at the top of the dam and some meters in the collapsed zones (Fig.
7.13). The kinetics of displacements are also modelled satisfactorily, with a sudden
acceleration of settlements. However, the predicted “failure” comes two months in advance
with respect to the real case. The pore pressure field is different from the measured one,
with only one point of control being comparable between the simulation and field
measurements. This is explained by the author by the insufficient initial degree of saturation
and clay modulus to generate interstitial pressures. The zones of failures are qualitatively
reproduced in the simulation with a symmetric pattern; however, the position of the
horizontal planes of sliding is too high.
212 CHAPTER 7
Figure 7.13 Predicted settlements, from Fry (2005). The vertical displacements are correctly predicted
in the top layers of the embankment only, and underestimated in deeper layers.
The conclusions of the study invalidate the hypothesis of a failure at the interface between
layers 1981 and 1982. The second hypothesis of the failure due to high water contents and
degrees of saturation could not be validated with this approach. Finally, the use of joints can
produce a fair simulation of the openings due to traction and the kinetics of the slide.
However, such an approach can only be carried out a posteriori, with strong assumptions on
the behaviour of such joints.
As a conclusion, different approaches have been proposed to model the behaviour and
failure of Mirgenbach earth dam. A possible way to produce a qualitative picture of the
mechanism of sliding and settlement is to introduce joints in the model (Fry 2005). However,
such an approach is quite artificial and seems to provide an erroneous explanation for the
failure. It is believed that the triggering factor for the failure is mostly the interstitial pore
pressures generated by the construction over a layer that is initially well saturated due to
infiltration and capillary rise. The preliminary simulations of Magnin (1995) show some
effects of the permeability coefficients on the generation of pore pressures. However, no
theoretical description of degree of saturation and deformations of the unsaturated soils a yet
available. The objective of the present application is to complement the reviewed modelling
campaigns with a new finite element analysis of the coupled problem, using ACMEG-s for
the constitutive laws.
7.3.4 Parameter determination and reference simulations.

In this paragraph are summarized the assumptions made for simulations and the choice
of material parameters. The results of a preliminary analysis carried out with Z-Soil finite
element code are also presented as an introduction to the fully coupled analysis carried out
in LAGAMINE with ACMEG-s.
Like in any engineering case of this scale, simplifying assumptions have to be made for
the boundary conditions (BC) and sources and modes of external loading. Provided that the
accent will be laid in the present study on the generation of interstitial pore pressures and
changes in matric suction and degree of saturation, the hydraulic BC should be implemented
with care. On the contrary, when simulating a complete section of the dam in 2D, the
imposition of mechanical restrictions on displacements is standard for this case.
Mesh, boundary conditions and loading scenario

Considering the available data and the assumptions made in the reviewed modelling
campaigns, the geometry of interest is defined as that of the median profile (called “B” in the
project reports) that is the highest profile. The information on water pressures, displacements
and shape of the sliding surfaces is believed to be the most accurate and comprehensive in
this section.
It is chosen to model not only the embankment but also the foundation and to reproduce
the scenario of construction by layers in 6 steps, according to the exact timeline already
presented in Fig. 7.9. The construction phases are modelled by the activation of the
successive layers as a function of time. The upper roof of each constructed deposit gives the
geometrical limit to the different layers in the model, as plotted in Fig. 7.14. As the upstream
side of the embankment has a different geometry from that of the downstream side
(especially at the toe), no symmetry axis at the centre of the dam will be used.
The displacements are fixed in all directions at the bottom of the foundation (arbitrarily
set at 30 meters below the embankment) and vertical displacements are allowed at the left
and right hand boundaries of the mesh (Fig. 7.14). The water table is set up at the natural
ground level that is the top of the foundation layer. The dimensions of the mesh for the
foundation are supposed to be large enough to impose constant hydrostatic pore water
pressures at the left and right hand sides of the foundation. All the boundaries are perfectly
drained, which is also the case of the roof of each layer before the next layer has been placed.
The particular case of pore water and air pressures at the external boundaries of the
embankment will be discussed more in depth later. Regarding the remarks formulated by
the experts concerning the drainage system, see part 7.3.2 and table 7.1, and the assumptions
taken by Fry (2005) it is assumed that the drainage system is not in function. In other words,
the drainage network is not only saturated but it can also develop excess pore water
pressures. As a consequence, the presence of sand channels is neglected here and the
material of the embankment is assumed to be homogeneous compacted clay.
Material parameters
The fully coupled analysis requires the material parameters for the mechanical stress-
strain model, the retention model and the fluids flow model. For this purpose, an extensive
review of the available data has been carried out to verify the homogeneity of published
information. Appendix F summarizes the soil characteristics for which some variability
Figure 7.14 Geometrical limits and boundary conditions to be taken into account for the mesh and
boundary conditions. The elements are not displayed in this figure to improve legibility.
214 CHAPTER 7

0.05
0.1
0.15 Exp. w=0.24

Exp. w=0.19
ACMEG-s w=0.24
ACMEG-s w=0.19
5 7
1000 10 10
Figure 7.15 Calibration of compressibility coefficients on oedometric tests for Mirgenbach clay at
different water contents w .
could be identified between the different publications.

As mentioned previously in chapter 6, the calibration of the stress strain model
theoretically requires results from tests carried out in both saturated and partially saturated
conditions. Here the soil recognition features 13 oedometric tests undertaken either in fully
saturated conditions or at given water contents, kept constant. Those tests have been used for
the calibration of the plastic compressibility coefficients ( β 0 and Ω ). The bulk and shear
( )
elastic moduli K ref , Gref were adjusted on unloading paths in oedometer, Fig. 7.15, and on
the basis of drained consolidated triaxial tests (Fig. 7.16). The parameters for the critical state
line have been deduced from the results of the shear tests in saturated (drained) conditions,
using also the average of the available data (Appendix F) and a dedicated parametric
analysis detailed later. The parameters of progressive plasticization, namely the initial radii
of mobilization of the isotropic and deviatoric plastic mechanisms and parameters a, b, c, d
are also adjusted on the basis of these oedometric and triaxial tests. The reference
preconsolidation pressure results from the compaction process, and is supposed to be the
same for all the layers at the moment where they are placed.
The parameters for the soil water retention have been calibrated on the basis on the initial
suctions and the degree of saturation measured at the optimum. Only very little information
is available for the retention properties, so the determined retention parameters for
Mirgenbach rely only on a range of low suctions ( s < 165 kPa ).
5
8 10
Exp. σ =1e5 Pa ACMEG-s
c
c
Deviatoric stress q(Pa)

5
6 10 c
5
4 10
5
2 10
0
0 0.1 0.2
Axial strain εa(-)
Figure 7.16 Calibration of deviatoric parameters on the basis of CD triaxial tests, Mirgenbach clay
1.05
1
0.95
r
0.9
0.85
0.8
0.75
0.7 Exp.
ACMEG-s
0.65 5
1000 10
Figure 7.17 Calibration of the retention model for Mirgenbach clay.
Nevertheless, the essential parameter to be determined is the reference air entry value: it
could be calibrated on the available experimental results (Fig. 7.17).
Tables 7.4 and 7.5 summarize the determined or assumed parameters of Mirgenbach clay
to be used in the simulations. The compressibility and perfect plasticity parameters for the
foundation soil were adapted mostly from the report by Fry (2005).
Provided that the foundation layer remains anytime fully saturated, there is no need to
calibrate the parameters of partial saturation for this group of elements.
216 CHAPTER 7
Table 7.4 Table of parameters for clay

Elastic
parameters

strain Plastic α Dilatancy coefficient 0
b 0.008
c 0.001
d 2
Limits of e
rdev Initialisation of dev. mechanism 0. 1
elastic e
domain riso Initialisation of iso. mechanism 0.1
se Air entry value 1e4 Pa
Retention model πH Coupling parameter 0
sDI Drying yield suction 4e4 Pa
Table 7.5 Table of parameters for foundation shale

Elastic
parameters
Gref Shear modulus 3e8 Pa
e
n Elastic exponent 1

b 1
c 0.0001
d 2
Limits of e
rdev Initialisation of dev. mechanism 1
elastic e
domain riso Initialisation of iso. mechanism 1
In the flow models used for the simulation, see previous chapter 6, the permeability
coefficients are evolving with the degree of saturation. Among the different formulations
implemented in LAGAMINE, the two following equations were used:
( S − Sres )
CKW 1
krw = r (7.1)
(1 − Sres )
CKW 2
kra = (1 − se ) (1 − se )
CKA1 CKA 2
(7.2)
Where CKW 1, CKW 2, CKA1, CKA2 are material parameters. In the absence here of
permeability tests in partially saturated conditions, the parameters of evolution of the
permeability for the embankment clay have been taken equal to the default values:
CKW 1 = 4, CKW 2 = 4, CKA1 = 2, CKA2 = 5 / 3 . The parameter remaining to be determined is
the coefficient of saturated permeability for the clay layers. Some inconsistencies seem to
appear from the data on permeability tests in Mirgenbach clay (see appendix F), so that it
was not straightforward to determine the water permeability coefficient. The value
k w = 10−10 m.s −1 used for the simulation was deduced from (i) the analysis of the previous
modeling campaigns and (ii) a thorough parametric study, detailed below.
Reference simulations
The present paragraph presents the preliminary simulation process that was carried out
to verify the permeability coefficient of the clay layers. According to Fry (2005), the
permeability of the clay layers should be of the order of 10−10 m.s −1 , while the parametric
study undertaken by Magnin (1995) ends up with a optimal permeability coefficient of
kw = 5.3 ×10−10 m.s −1 . These two numerical analyses showed however insufficient agreement
with the field of pore pressures measured in-situ, either locally or over the whole mesh.
The purpose of the upcoming preliminary simulations is to address the calibration of the
coefficient of saturated permeability, which is almost the unique parameter that cannot be
determined directly from the reports. The parametric analysis also copes with the variability
observed for the parameters of the critical state line and initial densities (see appendix F).
The reference simulations have been carried out by Christoph Berger, under my
supervision, in the framework of a semester project. All the calculations are made with the
Finite Element program Z-soil (Z-soil 2005). The reasons for choosing this program and not
LAGAMINE for the initial parameteric study are basically the following:
(i) The implementation of ACMEG-s constitutive model in LAGAMINE was still
under validation at the start of simulation.
(ii) The handling of Z-soil is somehow easier for the scenarios of constructions. That
feature contributes to multiplying the number of calculations within the restricted
time of a semester project.
(iii) Z-soil is one state-of-the-art finite element program for consulting engineering
purposes. Yet, it does not feature a tri-phasic approach and proposes constitutive
models that are obviously less exhaustive than ACMEG-s. The reference
simulations can therefore be used as a point of comparison to show the added
value of the new constitutive framework and use of LAGAMINE finite element
code.
218 CHAPTER 7
Figure 7.18 Mesh used for the preliminary simulations
The mesh makes use of 1129 quadrilateral 4-noded elements and a total of 1215 nodes,
plotted in Fig. 7.18. The boundary conditions and limits of layers are those of Fig. 7.14.
It is chosen to adopt a constitutive model of the type elastic-perfectly plastic, with Mohr-
Coulomb criterion, which can indeed be considered as an extreme simplification of ACMEG-
s model. The elastic parameters are those from tables 7.2 and 7.3, and the cohesion and
friction angle are assessed within the parametric analysis. This parametric analysis is indeed
restricted to three families of parameters, listed in table 7.6. Each of the three sets represents
a possible choice for the variations. For instance, set A deals with the stability parameters,
with the two extreme ranges collected in the reports. Set B explores the influence of the target
values for the compaction of the first layer of 1981. Finally, as the permeability is not yet
determined for Mirgenbach clay, set C proposes four different values, knowing that the
( )
lowest permeability k = 10−11 m.s −1 is out of the realistic values for such a material, but still
provides a lower bound for the analysis. These three sets (A,B and C) together create 16
possible combinations, which will practically be reduced down to seven.
The points of reference for the comparison of simulated results with in-situ measures are
Table 7.6 set of parameters and ranges of values for the parametric analysis
Set A Set B Set C
γd, w and Sr
CONTENT Φ’ and c’ Permeability K
of the layer 81
All layers together All layers together
Set X affects… Only layer 81
(82 and all of 81) (82 and all of 81)
(C1) Khorizontal = Kvertical
(B1) γd = 15.9 kN/m3
1e-11 m/s
Max (A1) 21° and 25 kPa w = 23.9%
(C2) Khorizontal = Kvertical
Sr = 94.5%
1e-10 m/s
VALUES (C3) Khorizontal = Kvertical
(B2) γd = 16 kN/m3 1e-9 m/s
Min (A2) 20° and 15 kPa w = 24.1% (C4) Khorizontal = 1e-8
Sr = 97.5% m/s
Kvertical = 1e-9 m/s
Figure 7.19 Location of water pressure cells B2 and B6 in the embankment
principally the two water pressure cells B2 and B6, located both in the clay mass on the
upstream side of the embankment (see Fig. 7.19).
The measured pore pressures and curve of terrain elevation are plotted in Fig. 7.20.
In essence, the strategy for the parametric analysis follows the steps:
1. Fix the most realistic soil characteristics for the choice of set A and set B (A1
and B1)
2. Carry out the parametric analysis for cases C1 to C4 (permeability).
3. Identify the case where the prediction of pore water pressures in points B2 and
B6 is the most accurate, and fix the corresponding permeability.
4. Investigate the four possible combinations of sets A and B.
Before discussing the results of the parametric analysis, some indications on the
management of negative pore water pressure in Z-Soil are formulated (Zimmerman 2005).
Even though Z-Soil formulation does not account explicitly for three-phase problems (i.e.
solid, liquid water and gaseous air), it is still possible to generate negative pore water
pressures.
To model partial saturation, it is assumed that the pore space is filled with a single fluid
F that is different from pure liquid water (this fluid can be seen as a mixture between liquid
water and gaseous air). A saturation coefficient S is introduced as a function of the pore
water pressure pw :
⎧ 1 − Sres
⎪ S res + 2 1/ 2
if pw < 0
⎪ ⎡ ⎛ α pw ⎞ ⎤
S =⎨ ⎢1 + ⎜ ⎟ ⎥ (7.3)
⎪ ⎢⎣ ⎝ γ F ⎠ ⎥⎦
⎪
⎩1 if pw ≥ 0
220 CHAPTER 7
172 120
(a)
170 100

Ground level 80
Ground level z (m)

168
60
166
40
164
20
Upstream failure
w
Pore pressure
162 0
160 -20
0 2 4 6 8 10 12 14
st
Months from 1 August 1981
178 150
(b)
176

174 100
Ground level z (m)
172
170 50
Ground level
168
Downstream failure
Upstream failure
166 Pore pressure 0
164
162 -50
0 5 10 15
st
Months from 1 September 1981
Figure 7.20 Time history of pore pressure measured in situ ((a) cell B2 and (b) cell B6) before the dam
rupture
Where S res is the residual degree of saturation (by default equal to zero), γ F is the proper
weight of the fluid and α a material parameter. Eq. 7.3 can be somehow considered as a
simplified soil water retention curve formulation. Coefficient α is here set to a default value
of 2. The flow equation is Darcy’s law (see chapter 6) in which the water permeability
coefficient kw is a function of the coefficient of saturation S defined previously and of the
saturated permeability coefficient k wsat :
(S − S )
3
res
kw = kwsat (7.4)
(1 − S )
3
res
Z-Soil thus accounts artificially for negative pore water pressures if needed and the
permeability is a direct function of this water pressure. The obtained fields of pore pressure
during the parametric analysis can thus be considered as reasonable approximated, at least
for the calibration of permeability. This calibration will obviously have to be checked when
using the fully coupled framework of ACMEG-s and LAGAMINE.
Recalling that the water table level is assumed to reach the natural ground level that is the
basis of the embankment, it is assumed that the pore pressures in the clay are negative when
a layer is placed. The pore pressure becomes more and more negative as a function of the
altitude, following simply the hydrostatic profile of the underground.
The results of the first step of calibration, making the permeability variable while keeping
the mechanical parameters constant, are plotted in Figs. 7.21 and 7.22, by comparing the
300
K=1e-11 m/s
K=1e-10 m/s
In situ
200 K=1e-9 m/s
w
K =1e-8, K =1e-9
h v
100
0 300 600
st
Days from 1 August 1981
Figure 7.21 Time evolution of water pore pressure at check point B6. K , K h and K v are respectively
the isotropic, horizontal and vertical coefficients of permeability.
200
K=1e-11 m/s
K=1e-10 m/s
In situ
K=1e-9 m/s
w
K =1e-8, K =1e-9
h v
100
0 250 500
st
Figure 7.22 Time evolution of water pore pressure at check point B2

222 CHAPTER 7
predicted excess (positive) pore-water pressures. The effect of the activation (construction) of
the successive layers is obvious as the modeled response in pore pressure shows some peaks
at the corresponding dates. After each loading (= new existing layer, some meters of clay),
the water pressure has 30 days to dissipate. This behaviour is typically due to the numerical
activation of each layer as a whole in one time moment (using a Dirac delta) rather than
reproducing the more progressive construction process (in which the layer is built practically
in 30 days). This day by day construction of layers can be addressed in LAGAMINE, as
shown later in the chapter. So, if the peaks shall not be taken into consideration, the values of
pore pressure at the final time of the construction of each layer are relevant. On both
checkpoints, there is a total dissipation of the excess water pore pressure for cases (C3) and
(C4) (see table7.4), that correspond to the highest coefficients of saturated permeability
( 10−9 m.s −1 and 10−8 m.s −1 ). This is not consistent with the measured values of pore pressure.
On the contrary the values of permeability coefficients taken in cases (C1) and (C2) do
prevent the total dissipation of the excess pore pressure between two layers. Still the lowest
permeability ( 10−11 m.s −1 ) is not only unrealistic with respect to the type of material but is
also at the origin of overestimated pore pressures. The permeability of 10−10 m.s −1 is thus
taken as the most conforming to the measured reality, especially for check point B6 (Fig.
7.21). The monitored pressures of checkpoint B2 cannot be predicted to a satisfactory
measure yet. It seems that the model would predict a faster flow of water out of this zone
which is more at the side of the embankment and thus at a shallower depth with respect to
point B6 (Fig. 7.19). These phenomena are addressed more in depth in the fully coupled
analysis to come.
Figure 7.23 shows an interesting pattern for the distribution of pore water pressures
throughout the mesh. The modeled values are captured at the time of construction
(activation) of the last layer at the top. A clear central pocket of posititve water pressure is
identified at the deep core of the structure while most of the 5 meters of clay below the
surface a partially saturated.
Once the permeability coefficient is fixed, the influence of the mechanical parameters (Sets
A and B) can be in turn assessed. The various tests show little differences in the modelled
pore pressure according to the possible combinations of parameters (e.g. Fig. 7.24). The pore
pressure is thus little affected by the mechanical parameters that are investigated here.
On the contrary, stability analysis should show the significant influence of the Set A on
the safety factor. The stability analysis can be carried out in Z-soil using the method of c, φ
reduction. The probable circular failure surfaces are computed automatically. The safety
factor is indeed observed to decrease every time a new layer is placed (Fig. 7.25). The
influence of parameters (A1) or (A2) is here clearly visible and expected, while no influence
of the sets B is visible. However, the safety factor never decreases down to the critical value
of 1, probably due to the simplifying assumptions made up to now on the partial saturation.
Knowing that failure actually occurred, the set of parameters that is the most unfavourable
for stability is retained (A2).
Figure 7.23 Colour maps of the water pressure atday 398 (construction of the top layer). The lightest
shades correspond to zones in excess (positive) water pressure while the darkest tone identifies the
zones of negative pore water pressure
200
A2-B2
A2-B1
A1-B2
A1-B1
w
In situ
100
0 300 600
st
Figure 7.24 influence of parameters from sets A and B at a fixed permeability coefficient, cell B6, see
table 7.6 for details on parameters combinations.
Fig. 7.26 shows the typical location of the probable failure circles determined during the
stability analysis. Interestingly, the probable failure is located (i) anytime in the clay layers
exclusively and (ii) alternatively (or simultaneously) on the upstream and downstream sides,
indicating that there is not a favourable side for the failure. This is consistent with the
observed symmetrical quasi-simultaneous failure scheme.
7.3.5 Fully coupled analysis

Using the determined parameters and the results of the preliminary calculations carried
out in Z-Soil, it is proposed to assess the capabilities of ACMEG-s and finite element code
LAGAMINE in the modelling of the case study of Mirgenbach dam. This new modelling
224 CHAPTER 7
A1-B2
4 A1-B1
A2-B2
A2-B1
Safety factor
3
1
0 250 500
st
Figure 7.25 influence of parameters from sets A and B on the Safety Factor (SF) (see table 7.6).
Figure 7.26 Stability analysis: probable shapes of the failure surface (colour maps of absolute
displacements).
campaign is intended essentially to provide a more accurate mapping of the pore pressures
generated during the phases of construction, and to investigate the evolution of the degree of
saturation in time and space.
The input of water due to the rain is not taken into consideration for the upcoming
simulations. This assumption is not favourable for the saturation of the layer of 1981, but it
will be seen that the capillary rise from the water table alone already contributes to the
saturation of this layer (1). The only available hydraulic input is in fact the water table level,
which is reasonably assumed stable. The pore water pressures within the foundation layer
can consequently be initialised to the hydrostatic values (Fig. 7.14).
The initial state of saturation, suction and preconsolidation pressure of the 6 clay layers
are piloted by the process of compaction. The suction is thus adjusted in all the layers at the
time of their construction to the value corresponding to the prescribed water content of
compaction (that is s ∈ [5 − 25] kPa ). The pore pressures at the boundaries of a given layer
are obviously free to evolve anytime an upper layer is present. Unlike the preliminary
calculation undertaken with Z-Soil program, the construction by layers is here modelled as a
more progressive elevation of the embankment throughout each month of construction (see
Fig. 7.20). This is made possible by controlling the coefficient of gravity within each layer
independently.
The pore pressures modelled in response to the construction by layers are compared with
the in-situ measures in Figs. 7.27 and 7.28. The qualitative response is very comparable to
that of Figs. 7.21 and 7.22 obtained in the preliminary calculations. However, the quantitative
prediction is significantly improved in the case of control cell B2, with a final pore water
pressure that is 50% more accurate than previously. This improvement shall be mainly
attributed to the performing HMR coupling in the zone of pressure cell B2 (upstream side).
On the other hand, the loading process due to the construction by layers is modelled in a
more realistic way, avoid the peak values of pore pressure obtained previously. The pore
water pressure contours in Fig. 7.29 and evidence a clear zone of partial saturation at the
surface of the embankment. Using the soil water retention model, the repartition of
saturation might be more representative of the reality, and the permeability is also modelled
more accurately in this zone. The drier blanket creates a natural barrier to prevent water
from flowing out in the given time of construction of the layer. Consequently, the water
pressure increases in the pocket created at deeper altitudes, even at the sides of the
embankment. The maximum generation of pore pressures remains however located at the
core of the dam (Fig. 7.29), where the effect of gravity is more sensible. The excess pore
pressures due to the construction of layer 4 with respect to the end of construction of layer 3
are quantified in Fig. 7.30. In the zone of maximum pore pressures, the degree of saturation
can vary up to 7%, indicating a progression of the saturation front.
5
1.2 10
In situ
Predicted
Pore water pressure p (Pa)
4
8 10
w
4
4 10
4
-4 10 7 7
0 2 10 4 10
Time (s)

226 CHAPTER 7
5
1.2 10
In situ
Predicted
Pore water pressure p (Pa)

4
8 10
w
4
4 10
7 7
0 2 10 4 10
Time (s)
Figure 7.29 Colour maps of the water pressure at time 2.50e7s (end of construction of layer 2)
Figure 7.30 Colour maps of excess pore pressure due to the construction of layer 4 (time 2.8e7s).
The observed displacements are consistent with those obtained in the preliminary
analysis, with a slight difference observed at the toe of the slopes. However, given the initial
compaction and the identified material parameters, the limit of plasticity of the materials is
far enough to prevent plastic compressions during construction. This indicates that failure
mode could be reached only with a sudden transition from a fully elastic state to a perfect
plasticity state.
Variation in pore pressure Δp (kPa) 20 0.07
0.06
15
Variation in saturation ΔSr

w
0.05
0.04
Limit of saturation
10
Δp 0.03
w
5 0.02
0.01
0
ΔS
r 0
-5 -0.01
0 10 20 30 40 50
Depth from top (m)
Figure 7.31 Variations in pore water pressure and degree of saturation during the construction of layer
4.
7.3.6 Conclusions for the case study

Mirgenbach earth dam that failed on both upstream and downstream sides constitute an
appropriate application for the numerical tool and constitutive models developed in the
PhD. thesis. The failure mode is typical of slope instability, and the principal cause of
damage identified by the consultants after the accident was the problems of dissipation of
pore water pressure during construction. The factors influencing the water content of the
clay are the malfunction of the drainage system and the winter rains. The embankment has
been simulated previously several times to verify those assumptions, but had shown either
inconsistencies with the measured pore pressures or relied on non realistic joint models.
The new numerical simulations, carried out with Z-Soil and LAGAMINE, show more
consistent understanding of the generation of pore water pressures within the fill during the
stages of construction by layers, compared to the previous numerical studies reviewed from
literature. Once the permeability has been determined by the means of a parametric study,
the fully coupled analysis provides a fairly accurate insight into the pore pressures that are at
the origin of the dam symmetric failure.
7.4 Conclusion
This chapter proposes a brief review of the current practice in dam dimensioning and
construction. It has been observed that embankments shall be able to sustain not only the
average environmental loads, but also should be verified in more critical loading situations.
If the proposed constitutive models cannot embrace all the possible modes of failure (for
instance erosion), it still provides relevant indicators for the triggering of large displacements
228 CHAPTER 7
and failure. The case study of Mirgenbach has been developed, featuring a particular
symmetric mode of rupture. It has been demonstrated, by reference to existing modelling
campaigns, that only a completely coupled hydro-mechanical-retention framework of the
type of ACMEG-s could provide a reasonable understanding of the involved processes. In
particular, the simulations offer an accurate picture of the generation of interstitial pore
water pressures during the construction of the dam in successive layers.
A second typical application for the developed model is presented in next chapter,
dealing with the issue of mass movements in a natural slope.
7.5 References
Alonso E. E., Pinyol N. M. (2008). "Unsaturated soil mechanics in earth and rockfill dam engineering."
First European Conference on Unsaturated Soils, Durham, Balkema, pp.
Berger C. (2008). Numerical simulation of the failure of an earth dam. Studied case: Mirgenbach dam
(France). Lausanne, EPFL.
Burghignoli A., Desideri A., Tamagnini R. (2002). "Numerical modelling of dam using the modified
Cam-Clay extended to the unsaturated condition." Third international conference on
unsaturated soils, Recife, pp. 221-225.
CFGB (1984). Protocol of a technical colloquium.
Charles J. A., Tedd P., Holton I. R. (1995). "Internal erosion in clay cores of British dams." Research
and development in the field of Dams - Swiss National Commitee on Large Dams, Crans-
Montana, pp. 59-70.
EDF (1979). Technical document of the constructor. Lyon, EDF Région d'équipement d'Alpes.
EDF (1982). Incidents ayant affecté le barrage du Mirgenbach. Lyon, EDF Région d'équipement
d'Alpes.
Fry J.-J. (2003). Modélisation de la rupture du barrage de Mirgenbach, EDF.
Hirschfeld R. C., Poulos S. J. (1973). Embankment-dam engineering, Casagrande volume, Wiley.
Laigle F. (1993). Analyse statistique des propriétés de l'argile de Mirgenbach, EDF.
Magnin P. (1995). Modélisation de la consolidation des sols non saturés: amélioration, validation,
justification du logiciel UDAM. Paris, Ecole Nationale des Ponts et Chaussées. PhD. Thesis.
Mallet C., Pacquant J. (1951). Les barrages en terre. Paris, Eyrolles.
Ould Amy M., Magnan J.-P. (1991). Modélisation numérique des écoulements et des déformations dans les
barrages en terre construits sur des sols mous. Paris, Laboratoire Central des Ponts et Chaussées.
REAL E.-. (1983). Barrage du Mirgenbach - comité hydraulique, EDF REAL Service de Chambéry and
MECASOL Paris.
Wahlstrom E. E. (1974). Dams, Dam Foundations, and Reservoir sites, Elsevier.
8. Application: numerical modelling of a shallow
landslide
8.1 Introduction
The purpose of this chapter is to apply ACMEG-s to the case study of a shallow landslide.
The presented work is our contribution to the CCES project TRAMM (Triggering of Rapid
Mass Movements in steep terrain). TRAMM project is an interdisciplinary research network
that intends to provide a global picture of the process of mass movements from the initial
state to the run out. The principle is also to confront, at various scales, the in-situ and
experimental evidence to the physical models and numerical simulations. Within this
context, a real-scale landslide release test has been set up with important instrumentation
and will be the focus of this chapter.
The theme of landslides is indeed a second suitable application for ACMEG-s, provided
that most terrains feature naturally a capillary fringe. In the precise case of steep terrains, the
balance between the infiltration of water and the surface runoff is critical for the water
content of the ground. The understanding of the mechanical stress strain (mass movements)
will consequently have to rely on a performing description of the flow and also the
saturation of the medium.
In consulting engineering, one of the most usual methods to simulate landslides is to
introduce interface elements between the pre-supposed fixed reference layer (bedrock) and
the moving mass (e.g. François et al. 2007). The hydrogeology is then accounted for by
considering the map of pore pressures as an input for the calculation. Due to the interface
elements, the displacements are concentrated in a preferential zone. Such a numerical
modelling approach however requires the plane of sliding to be well determined. In the
present approach, it will be chosen not to concentrate the deformation in a sliding zone for
two reasons: (i) the objective is to investigate the deformations of the mass in all layers and
(ii) no plane of sliding has been identified in the case study. Moreover, the present
calculation is not a stability analysis, but rather an application of the developed stress-strain
model to a case where partial saturation is needed. The interest is to investigate the
predictions in terms of saturation and deformation as a result of water infiltration in natural
conditions.
At the time of the redaction of the present manuscript, the TRAMM project still has a
running period of 2 years, and most data collected from the field is under processing.
Consequently, neither the complete calibration of the parameters nor the comparison with all
field measurements is yet possible. However, the modelling exercise is justified by the need
for blind prediction of the behaviour of the slope, in order to prepare future in-situ landslide
triggering experiment. Considering that the parameters will be calibrated on the basis of
experimental tests and thus will be completely determined, there is no need for a parametric
analysis on the material parameters. The parameters of influence to be considered however
230 CHAPTER 8
in the following analysis are the boundary conditions and loading scenarios with respect to
the imposed sprinkling pattern. It is then proposed to assess qualitatively the effect of rain
infiltration for the stress-strain behaviour of the medium.
8.2 Presentation of Rüdlingen experiment

The site of Rüdlingen (Switzerland), has been identified by TRAMM network as suitable
for the experimental triggering of a shallow landslide at real scale. The experiment involves
currently 9 partners from Swiss universities and research institutes. The first experimental
landslide release was carried out in October 2008 and will be followed by a second triggering
experiment in 2009.
8.2.1 Site view

Fig. 8.1 presents the ground elevation at the test site, which is close to the bed of river
Rhein. The regressive erosion is visible, and the zone shows a number of shallow landslides
that occurred in cohesionless soils. The chosen slope is representative of the natural spots in
Switzerland that are most susceptible to landslides, with an average slope angle of 34°,
located within an area relatively clear of vegetation.
8.2.2 Geology and soil

Two types of material are encountered in the geological profile, namely gray layered
sandstone and brown marl. The boreholes indicate the presence of those rock formations at
depths between 0.5 and more than 4 meters deep from the surface. However, due to the
Figure 8.1 Digital elevation map of Rüdlingen site (GIS Schaffhausen), from Brönniman and Tacher
(2008).
APPLICATION TO LANDSLIDE MODELLING 231
limited depth of the initial boreholes (i.e. 5m), it could not yet be established if the occurrence
of rocks at these depths was characterizing of the roof of the bedrock or not. The geologists
emitted the hypothesis of free rock boulders. This uncertainty on the nature of the
underground layers will be accounted for in the finite element model.
The top layer is constituted of one fairly homogeneous material that is a silty sand, with
an average of 56% of sand, 33% of silt and low fractions of clay and gravel. The material is
identified as a low to medium plasticity soil (Askarinejad et al. 2008). The internal effective
friction angle is equal to 32° in fully saturated conditions. According to the preliminary
analytical slope stability analysis carried out by Askarinejad et al. (2008), the safety factor is
of the order of 1.05 and lower depending on the point of analysis, indicating that the slope
was initially in limit equilibrium.
The water content within the top soil layer is relatively low, indicating that if there is a
water table, it is located deeper in the profile and that it shall not be determinant in the case
of shallow landslides. However, this assumption has still to be verified by the means of
deeper boreholes.
8.2.3 Principle of experiment

The objective of the landslide release experiment is to trigger a mass movement in
conditions that are the closest possible to the natural environmental loads. Consequently, it
Figure 8.2 Sections of the slope (left and right profiles).

232 CHAPTER 8
has been proposed to simulate exclusively the effects of rain infiltration in the soil on the
stability of the slope and on its deformations. The presented documents and results are
coming from the contributions of the teams of TRAMM network.
The dimensions of the test site are approximately 8 meters in width, and 32m in length
with an average slope of 34°, see Figs. 8.2 and 8.3. The slope is located above a road that
requires protection by means of a net during the experiments.
In order to favour the land movement, a number of trees were sawn down to clear the
area of the experiment. In the area likely to be active only three trees were cut (Fig. 8.2) but
their roots were not extracted from the soil.
The principle of the experiment is to water a delimitated surface of the slope by the means
of sprinklers whose hydraulic flow rate is controlled. The water is pumped from a stream in
the neighbourhood of the test site. While the principle objective is to trigger a landslide, all
the displacements and changes in water content prior to the failure of the slope are of interest
which justify the unique instrumentation placed in (see next paragraph 8.2.4).
Small scale infiltration experiments have been carried out prior to the large scale
sprinkling (Askarinejad et al. 2008). It has been determined that the infiltration capacity of
the top soil is greater than 60 mm. The inputted tracer fluid showed very homogeneous
infiltration and evidenced the roof of an impermeable bedrock. These observations were
made in 4 test pits (Fig. 8.2), but are contradictory with other infiltration tests at the centre of
the site where inputted water was fast drained (Brönniman 2008).
Figure 8.3 Plan view of the test site.

8.2.4 Instrumentation and hydraulic inputs

The 5 sprinklers are reparsed along the middle line of the test site at different altitudes
(Fig. 8.4) and equidistant each from each other. The instrumentation is essentially gathered in
three clusters that include piezometers, pore pressure sensors, tensiometers, moisture
sensors, and strain gauges (with inclinometer). Most measurements can be made at the
depths of 15cm, 30 cm, 60cm, 90cm, 120cm and 150cm (or some of these depths). The surface
displacements are monitored by photogrammetry.
The rain gauge enables to measure the effective quantity of water applied to the soil,
knowing that the overland flow is also measured. It was estimated that the optimum
efficiency of sprinkling (infiltration with respect to precipitation intensity) was around 60%.
8.3 List of parameters

Tables 8.1 and 8.2 summarize the parameters used for the simulations analysing the
influence of the boundary conditions and loading scenarios. The simulations are carried out
with hydro-mechanical coupling in finite element code LAGAMINE. The values in bold have
been assumed from databases on similar materials or set to their default values, in the
provisional absence of information. For this first series of simulations, it is assumed that the
two materials have the same mechanical and retention properties. Whenever possible, the
parameters have been calibrated on the basis of in-situ or laboratory test results reported by
Askarinejad et al. (2008), as developed below. The model of permeability is that of Eq. 7.1
presented in previous chapter 7.
Figure 8.4 Setup of the sprinkling devices and clusters for instrumentation (Askarinejad and
Springman 2008).
234 CHAPTER 8
Table 7.1 Table of parameters for silty sand.

Elastic
parameters

model parameters a 1e-4
b 1
c 1e-4
d 2
Limits of e
elastic
domain risoe Initialisation of iso. mechanism 1
Capillary effects Ω Coefficient of compressibility change 0
sDI Drying yield suction 10000 Pa
kw Intrinsic water permeability 1e-10 m/s
Permeability CKW1 2
CKW2 2
Table 7.2 Table of parameters for rock

Elastic
parameters

model parameters a 1e-4
b 1
c 1e-4
d 2
Limits of e
elastic
domain r e
iso Initialisation of iso. mechanism 1
γs Coefficient of LC Curve 2
sDI Drying yield suction 10000 Pa
kw Intrinsic water permeability 1e-10 m/s
Permeability CKW1 2
CKW2 2
The saturated critical state line is calibrated on the basis of 5 experimental points
(Askarinejad et al. 2008), Fig. 8.4. The soil water content being available only in terms of
300
250 Exp.
200
150
100
ACMEG-s
50
0
0 100 200
Figure 8.4 Calibration of the critical state line in saturated conditions.
0.45
Exp.
Volumetric water content θw(-)
0.4
0.35
0.3
0.25
0.2
1 s 100
e
Figure 8.5 Calibration of the reference air entry value.

236 CHAPTER 8
volumetric water content, only the air entry value can be estimated from Fig. 8.4.
8.4 Mesh, boundary conditions and loading scenarios

A choice had to be made concerning the most representative profile of the slope to model
in two-dimensions. Due to the lack of information on the nature of the substratum and the
shape of its roof, a three-dimensional model would not be justified at this point, knowing the
additional cost it would imply. Yet, given the important variability in the geometry through
the along the width, Fig. 8.6, the modelling exercise could later be extended to a third
dimension. However, such a three-dimensional analysis should be justified only if the
important variability of the bedrock roof along x-axis is verified by further boreholes and if
the points of Fig. 8.6 are indeed validated as belonging to the bedrock.
8.4.1 Mesh and mechanical boundary conditions

The finite element mesh models the left section of the geometry (Fig. 8.7), using the
topographic data to draw the surface and the shallow borehole information for the lower
bound. The mesh features 3688 nodes and 1866 6-noded triangular elements.
The two materials characterised previously are affected to the two layers defined in Fig.
8.X. the mesh is refined in the upper part (silty sand layer) to enable a more accurate
prediction of the surface displacements, expected to be of the order of centimetres. On the
contrary, for the lower layer, the mesh is made coarser to reduce the calculation costs. It is a
reasonable assumption to expect that the deformations in the rock-like layer will be less
significant than in the top soil layer; however, given the actual uncertainty on the nature of
the deeper materials, it would be inappropriate to neglect the deeper mass deformations and
discard this layer.
The displacements are neutralized in x and y for all the nodes at the bottom of the mesh,
and the horizontal displacements are blocked on the right hand side (Fig. 8.7). Concerning
the left-hand side, due to the important slope, it is not possible to block the nodes without
generating artificial tensile stresses within the upper coat of soil. The nodes are consequently
0
(a)
Depth below surface (m)
-2
-4
-6
Upper
Middle
Lower
-8
0 2 4 6 8
Distance from left transect (m)
Figure 8.6 Evolution of the Bedrock roof along the width of the site, at three sections. Fig. (b)plots the
left and middle longitudinal transects.
free to move in all directions and a linear load is applied to simulate the earth pressure and
the continuous contact with upper ground.
8.4.2 Hydraulic boundary conditions

The issue of the location of the ground water table has not been solved yet, and it is still
uncertain whether the phreatic level is simply horizontal (with an altitude piloted by the
level of the neighbour Rhein river) or else if there is a water table that is almost parallel to the
Figure 8.7 Finite element mesh and mechanical boundary conditions.
Figure 8.8 Hydraulic boundary conditions. In the case of “partial” drainage, only the nodes effectively
under the level of the water table are considered as drained.
238 CHAPTER 8
surface (Fig. 8.8). From the discussions of the research group TRAMM, no evidence can
account more for one option or the other. Provided that the analysis is an effort to highlight
the effects of partial saturation, it is chosen here to focus exclusively on the case of the
horizontal water table, in which the zone of partial saturation is larger and the maximum
suction levels higher.
As the purpose of the experiment is to study the impact of rain on the eventual triggering
of a landslide, the effect of sprinklers or natural rain can be simulated by an imposed
hydraulic flow at the surface of the ground. To get closer to the conditions of the experiment,
the imposed fluxes are concentrated over the inclined zone only. The imposed hydraulic
flow q f is the variable whose influence is investigated. The loading scenarios are thus
different combinations of time and intensity of sprinkling. The boundary conditions might
also influence significantly the soil response, especially with respect to drainage conditions.
Two types of conditions for drainage are thus tested, the “complete” drainage, that is all the
boundaries are drained (Fig. 8.8) and the “partial” drainage, in which only the nodes at the
boundary that are under the water table are drained (Fig. 8.8). This partial drainage will
cause a channelling of the water from the top (imposed flow) to the bottom right of the mesh
(water table). It reproduces the possible case of the existence of an aquifer at low altitudes
and a true impermeable bedrock that would be parallel to the slope. Lastly, the ratio of
characteristic water permeability coefficients between the top layer and the bedrock might
also have an influence.
Three test programmes including two scenarios each are simulated here and summarized
in table 8.3.
Table 8.3 loading scenarios of the parametric study.

Programme Id Imposed flow Drainage BC Permeability
1a Fig. 8.9 Partial Default
1
1b Fig. 8.9 Complete Default
2a Fig. 8.10 Partial Default
2
2b Fig. 8.11 Partial Default
k w− rock = 10 −10 m / s
3a Fig. 8.11 Partial
k w− soil = 10 −9 m / s
3
k w− rock = 10−9 m / s
3b Fig. 8.11 Partial
k w− soil = 10 −10 m / s
Figure 8.9 Imposed surface flow, case 1. The intensity of rain and estimated infiltration for the test site
are plotted as a reference.
0.06
Imposed water flow q(m/s)
0.04
0.02
0 5 5 5
0 1 10 2 10 3 10
Time t(s)
Figure 8.10 Imposed surface flow, case 2a
8.5 Discussion on results of the numerical simulation

Hereafter are compared the responses to the 6 different loading scenarios presented
previously. The calculations presented hereafter have been carried out with the help of John
Eichenberger. The objective is to identify the most critical configuration (with respect to load
and boundary conditions) for pore water pressures and deformations in the soil mass.
240 CHAPTER 8
0.08
Imposed water flow q(m/s)
0 5 5 5
0 1 10 2 10 3 10
Time t(s)
Figure 8.11 Imposed surface flow, case 2b.
Figure 8.12 Flow vectors in the medium at time 2.59e5 s.
8.5.1 Loading scenario 1

The first modelling programme investigates the influence of the conditions of drainage
that is either partial or complete (Fig. 8.8). The modelled evolution of imposed flow is that of
the reference case (Fig. 8.9). The flow vectors (Fig. 8.12), provide a representative picture of
the infiltration process, evidencing in the case 1a (partial drainage) an obvious channel for
the flow of water that is parallel to the slope.
The hydraulic flow is however not the only point of interest of the test, as the pore water
pressures are affected by the imposed flow. Fig. 8.13 shows the state of saturation and pore
water pressure before the start of the water injection. Due to the initial choice of a low level
Figure 8.13 Initial state prior to sprinkling (time 0) (a) Contours of pore water pressures (b) Contours
of degree of saturation.
horizontal water table, the matric suction reaches high levels at the top of the slope (235 kPa).
The computed degree of saturation is a result of the water retention model. Provided that the
air entry value is very low, the saturation drops below 1 immediately above the water table.
The overall effect of the imposed flow is to generate excess pore pressures, mostly at the
neighbourhood of the surface but also deeper below (Fig. 8.13a). At checkpoint 2, according
to the preliminary data from Askarinejad et al. (2008), the pore pressure sensor at a depth of
1.5m indicates a maximum variation in pore pressure of 7kPa. The predicted variation in
pore pressure of 6.8 kPa at point 2 (depth 1.5m below the surface) is consistent with the in-
situ measurement, and enables to validate the imposed flow condition. Over the whole
slope, the maximum variation of pore pressure predicted is of the order of 14 kPa and is
measured at the top of the slope, where the initial suction is the highest. The last part of the
curves in Fig. 8.14b, that is the evolution pore pressures once the imposed flow is steady
(time 176’400s, see Fig. 8.9) will be discussed later in the chapter.
Given the initial levels of suctions, the excess pore pressures are not sufficient to turn the
pore pressures in the positive range. Provided also that the air entry value is very low, the
material remains partially saturated and no indication of local saturation out of the water
table is observed.
In the partially saturated zone, the change of pore pressure and degree of saturation
causes the effective stress to decrease and consequently swelling is observed (Fig. 8.15). The
modelled displacements, although qualitatively satisfactory are too small to be considered as
significant (of the order of 1e-5m). The null displacements are however consistent with the
results of the first sprinkling experiment where hardly no displacements have been
monitored.
242 CHAPTER 8
8
(b)
Variation of water pressure Δpw(kPa)

6
2
Point 1
Point 2
Point 3
0 5 5 5
0 1 10 2 10 3 10
Time t(s)
Figure 8.14 Variation of pore water pressure at three points close to the depth (time 2.59e5s) . The
coordinates of the points in the local reference system are: Point 1 (4.98;-4.34), Point 2(4.98;-5.84), Point
3(4.98;-7.34).
Figure 8.15 Displacements due to the infiltration (time 2.592e5s) (a) horizontal displacements and non
deformed mesh (b) vertical displacements (deformed mesh x50’000)..
The alternate boundary conditions featuring a complete drainage have a significant

influence on the generation of pore pressures. The close field of the surface is affected by the
changes but the excess pore pressures are fast dissipated with depth (Fig. 8.16). In this case,
the variation of pore pressure at check point 1 is less than half the one in the partially
drained case, and thus gets out of the target based on the experimental measurements.
8
(b)

6
2
Case 1a
Case 1b
0 5 5 5
0 1 10 2 10 3 10
Time t(s)
Figure 8.16 Effect of complete drainage on the generated pore water pressures. (a) Variation of pore
pressures at time 2.592e5s with respect to initial state (b) Comparison of generated pore pressures in
point 1 between the two cases of drainage.
For the following calculations, only the most defavourable drainage conditions for the
slope stability are maintained, that is the partial drainage at the bottom right of the mesh.
This implies that the assumption of a deeper impermeable rock layer and a low aquifer is
accepted. Some preliminary conclusions can be formulated at this point:
(i) The predicted excess pore pressures due to imposed surface flow are consistent
with the measured pore pressures within the soil.
(ii) The initial suctions are very likely overestimated at the top of the slope. However,
even at the bottom of the slope where the initial suctions are low, the inputted
water flow is still not sufficient to saturate the material
(iii) The mechanical response to the infiltration test is almost insignificant but
consistent with the observed trends. The maximum displacements are of the order
of a few centimetres to the maximum. It is thus very likely that the flow process is
correctly simulated. The overall trend is a swelling due to excess pore pressure
that relieves the effective stress. The probability of pore-collapse is very low given
the level of stress.

In the loading programme 2 are tested alternate durations and intensity for the imposed
flow. As the parametric study is aimed at providing trends for the real case experiment, the
tested values of flow are conditioned by the technical limitations of the experiment. It is
indeed considered, following discussions with TRAMM partners, that an inflow could not
technically go beyond twice the value already tested, due to (i) the water supply to the
sprinklers and (ii) the capacity of infiltration of the soil.
It was observed in Fig. 8.14b that once the water inflow stabilizes to its maximum value of
0.04 m/s, the pore pressure first decreases and then continues to increase, but at a lower rate
than in the previous steps. It is proposed to verify the behaviour in pressure with the loading
scenario of case 2a, where the maximum flow (0.04m/s) is imposed in a short period of 1
244 CHAPTER 8
hour (Fig. 8.10) followed by constant water flow. The analysis of the evolution of the pore
water pressure for point 1 to that type of load is presented in Fig. 8.17a and compared to the
prediction of case 1a. The maximum value of pore pressure reached in the case 2a is
comparable to that of case 1a (8.3 kPa). However, the stabilization of the water inflow
induces also a stabilization of the pore water pressure at node 1. As a consequence, the pore
water pressure at the end of the test is higher in case 2a than in case 2b, with the same value
of flow at the surface.This indicates that faster rate of increase of the inflow would be more
favourable to the generation of pore pressures and consequently to mass deformations. The
predicted displacements for case 2a remain however of the same order of magnitude than
previously and will consequently not be discussed here.
In case 2b, the water inflow is augmented linearly to a maximum value that is twice that
of case 2a, but then it is lowered down to zero after 128’700s. In that case, the maximum pore
pressure quasi-linearly increases with time to 16.4 kPa and recovers its reference value at the
final state where the inflow is null (Fig. 8.17b). The maximum displacements obtained in this
case remain very small (5.2e-5m, fig. 8.18).
20
(a) (b)
Variation of water pressure Δp (kPa)
8
w
10
0
Case 1a Case 1a
Case 2a Case 2b
0 5 5 5 5 5 5
0 1 10 2 10 3 10 0 1 10 2 10 3 10
Time t(s) Time t(s)
Figure 8.17 Effect of duration and intensity of water inflow on the generation of pore pressures at
point 1. the results are compared with those of case 1a (a) Configuration Fig. 8.10 (case 2a) (b)
Configuration Fig. 8.11. (case 2b).
Figure 8.18 Displacements at time 128’700s (peak flow) – deformed mesh x50’000.
If it is possible practically to increase rapidly the flow up to its final value (case 2a), the
loading profile to be preferred is that of Fig. 8.10. On the other hand, the response in pore
pressure and displacements is obviously sensitive to the intensity of the flow. It can be
deduced from the above analysis that the realistic ranges of inflow tested to model the
natural infiltration are apparently insufficient to generate significant displacements at the
surface and within the mass.

Here, the conditions of drainage and flow input of case 2b are selected as the most
relevant for the case study. This case is now re-assessed with respect to differential
permeability coefficients between the two layers. The qualitative responses in terms of
variation of pore water pressure and displacements are plotted in Figs. 8.19 and 8.20. When
the intrinsic permeability of the rock layer is smaller than that of the soil (case 3a, Figs. 8.19a
and 8.20a), the infiltration process is very similar to that of case 2b, with however smaller
displacements. In the case of a smaller permeability for the soil, it the contours of excess pore
water pressure a sensibly different, with a faster dissipation within the rock mass, and a
smaller amount of displacements. It the new boreholes shall reveal the existence of a
permeable layer underneath the soil, it is very probable that the case 3b would be the most
representative, with little chance for deformations in the mass.
246 CHAPTER 8
Figure 8.19 Maps of pore water pressures excess at time 128’700s (a) case 3a (b) case 3b.
Figure 8.20 Contours of displacements at time 128’700 (a) case 3a (b) case 3b.
8.5.4 Comments on the possible mechanisms of failure.

It has been evaluated from the studied loading scenarios that the natural infiltration is at
the origin of only little perturbation of the state of pore pressure and thus of the state of
stress. The realistic input for hydraulic flow is thus insufficient to generate deformations in
the mass or plasticity.
It was chosen to model the ground as two deformable materials to investigate the possible
localization of stresses and verify the assumption of homogenous deformation within the
mass. In complement to the present study, an alternate model should feature an interface
between the rock and the soil layers to localize the deformations. Such an approach is
however not the scope of the present PhD. thesis and is not presented here.
8.5.5 Conclusions
In this part have been presented the results from the analysis of the influence of the rain
intensity and duration on the hydro-mechanical behaviour of Rüdlingen slope. It can be
noticed that the qualitative analysis provides a good picture of the pattern of infiltration
within the medium. The assumption of a deformable rock-like layer seems to hold, but it
must be noticed that the predicted displacements of the deeper layer are relatively small. The
prediction evidence two contradictory trends within the soil mass. On the one hand, the
input of water causes a reduction in matric suction and an increase in the saturation degree.
In fully elastic conditions, this combination of capillary variables causes the effective stress to
reduce and thus induces a swelling of the soil layer. On the other hand, the global trend due
to wetting is a settlement combined with a displacement of the ground downwards the
slope. Those two volumetric trends are not balanced and the overall response is an apparent
settlement.
The loading scenario that is the most favourable to movement is the case 2b. The involved
hydraulic input is however beyond the edge of our technical capability due to the constraints
of (i) limited supply of water from a stream and (ii) infiltration rate within the soil (the water
is not injected under pressure).
8.6 Conclusion
This chapter covers the application of ACMEG-s and LAGAMINE to the case study of a
shallow landslide that is the object of a real-scale site experiment. The finite element model is
a representative profile of the geometry of the problem and the two-dimensional analysis is
appropriate for this case. The parameters could be partly calibrated on the first available
results from the geotechnical characterization of Rüdlingen soil. The missing material
constants are complemented with standard values for silty sands and shall be verified with
supplementary laboratory tests.
The rain, which is controlled in the site experiment by the means of sprinklers, is
modelled through an imposed hydraulic flow at soil surface. The objective of the predictions
presented in the chapter is to assess the influence of the rain intensity and duration on the
triggering of the mass movements.
It was chosen not to implement interfaces elements as it is, to our present knowledge, not
sure that there is a preferential plane of shear nor if the bedrock could be considered as
stable. The whole mass is thus deformable, which provides a field of displacements over the
two modelled layers. The overall displacements are very low due to the little pore water
pressure change. The predicted water pressures within the soil are consistent with the first
measurements of the phase 1 of the in-situ experimental programme, as well as the
displacements.
The perspectives of the study are to assess the technical capability for sprinkling and
investigate further the infiltration process of the soil in place. The characterization of the
hydro-geology of the site, as well as the laboratory testing of soil samples have to be
completed.
8.7 References
Askarinejad A., Kienzler P., Casini F., Springman S. (2008) . “Rüdlingen landslide experiment”,
internal TRAMM presentation.
Askarinejad A., Springman S. (2008). “Surveying results Rüdlingen sprinkling experiment”, internal
TRAMM report.
Bronniman C., Tacher, L. (2008). “Geology and hydrogeology in Rüdlingen”, internal TRAMM
presentation.
248 CHAPTER 8
François B., Tacher L., Bonnard C., Laloui L. (2007). "Numerical modelling of the hydrogeological and
geomechanical behaviour of a large slope movement: The Triesenberg landslide
(Liechtenstein)." Canadian Geotechnical Journal 44, pp. 840-857.
9. Conclusion
The PhD. thesis addresses the issue of the modelling of partially saturated soils, from the
constitutive point of view. The management of multiple-way couplings between the
mechanics and hydraulics is also at the centre of the presented research. The objectives have
been (i) to identify the critical issues in the knowledge of the behaviour of unsaturated soils
and (ii) to propose a new modelling tool that is applicable to a wide range of engineering
cases where partial saturation is essential.
The method that has been followed consisted first in defining the stress and strain
framework in partially saturated conditions. Then, the constitutive models for mechanical
and retention behaviours have been formulated separately before being re-coupled together.
The activation of double-way coupling between the two models was then processed. The
model has been then validated on the basis of experimental data and applied to case studies.
Considering these initial objectives, the PhD. work leads to a number of contributions that
are reviewed hereafter and conclude the manuscript.
9.1 Achievements
9.1.1 Clarification of effective stress for unsaturated soils and establishment of
a new theoretical framework
The definition of an appropriate and consistent stress and strain framework is a necessary
condition for the formulation of a constitutive model for unsaturated soils. Given the number
of stress frameworks published in literature and still subject to debate, it has been proposed
to critically review the historical developments of the effective stress. The starting point for
the review is the concept of effective stress of Terzaghi, formulated for saturated soils, and
relating the experimental effective stress to the external stress and the pressure of pore water.
The extension of the effective stress principle and unique stress state variable to
unsaturated soils has been discussed. The conclusion is that a second stress variable is
necessary to describe the behaviour of the soil. On the other hand, the two stresses are not
necessarily independent, and with this regards, it is possible to define a unique stress to
describe the mechanical stress-strain behaviour. The stress variables are necessarily used
within a complete elasto-plastic framework.
The possible effective stresses have been reviewed and classified into three different
types. Each type of stress is assessed with respect to the stress path, the transition between
saturated and partially saturated states and the inclusion of couplings due to capillarity.
Among the consistent stress variables, the generalized effective stress, defined as the sum
of the net stress and the product of degree of saturation by suction was selected as the most
convenient for constitutive modelling purposes. The implications of the use of the
generalised effective stress have been then investigated in preparation to the formulation of
250 CHAPTER 9
the constitutive model. In particular, it has been obtained that the critical state line was
uniquely determined whatever the level of suction, if represented with the effective stress.
This property was verified for a number of fine-grained soils like kaolin or Sion silt.
A new constitutive model for unsaturated soils has been formulated and validated.
ACMEG-s stands for Advanced Constitutive Model for Environmental Geomechanics,
unsaturated extension. The developed model belongs to a family of models developed by L.
Laloui and Co-workers.
Using the advantages of the generalized effective stress, a reference stress-strain model
has been extended to partially saturated soils. The key steps performed within this task were:
(i) Choosing an appropriate constitutive model to adapt.
(ii) Replacing Terzaghi’s effective stress with the generalized effective stress
(iii) Define the capillary variables and the retention model
(iv) Introduce the missing effects of capillarity.
The reference model is that of Hujeux, which belongs to the category of Cam-clay models
in the sense that it is an elasto-plastic model with hardening plasticity, defined within the
critical state framework. The two mechanisms of plasticity (one isotropic and one deviatoric)
have been preserved, which provide a more accurate management of the transition between
elastic and elasto-plastic behaviours. The use of mechanisms of plasticity provides results
that are equivalent to those of bounding plasticity. The model also features non linear
elasticity.
If some effects of capillarity or partial saturation are accounted for by the effective stress,
for instance the increase in apparent cohesion with suction or the capillary hysteresis, other
features of behaviour required a specific formulation on the basis of the reference model. The
major specification for the formulation of the model was to aim at adding a minimal number
of new parameters compared to the list of parameters of the saturated reference model.
The particular form of the consistency equation has been clarified, along with the nature
of the hardening variables. A key element in the definition of the yield locus is the
dependency of the preconsolidation pressure with suction, also known as the loading
collapse curve. Using provisionally the water retention model of Van Genuchten, the model
is validated on a number of materials and stress paths taken from literature. The most
noticeable advance of the model is the capability to predict the stress response in swelling
pressure tests.
The second main focus of the constitutive modelling tasks is the formulation of a new
water retention model. The relationship between the degree of saturation and suction will
have considerable influence on the state of stress and also the permeability coefficients. An
experimental campaign, motivated by the review of the up-to date experimental
programmes from literature, had the objective to study the influence of the initial void ratio
on the air entry value of Sion Silt. This material was previously thoroughly characterized in
our laboratory. It was established that (i) the air entry value is dependent on the volume of
the soil and that (ii) the soil water retention curve presents a hysteresis.
The model for the water retention behaviour makes use of kinematic hardening to
reproduce the plastic and elasto-plastic change in degree of saturation versus suction.
Besides, the air entry value is the parameter that causes the whole curve to shift towards
higher levels of suction if the material undergoes compression. Very few constitutive models
feature both the hysteresis and the void ratio dependency; consequently, the performances of
the model on highly coupled processes such as constant water content tests are unique.
CONCLUSION 251
It has been lastly demonstrated that the stress-strain and retention models are coupled
together in double-way. In essence, the effective stress requires the information of the degree
of saturation and suction anytime while the air entry value depends on the volumetric strain.
The “terms of coupling” have been thus highlighted from the constitutive equations of the
two models, and recommendations are formulated for their calibration.
9.1.2 Numerical modelling and Boundary Value Problems

The integration of ACMEG-s raises a number of issues, related (i) to the reference stress-
strain model itself and (ii) to the double-way coupling. Concerning the stress-strain model,
integration algorithms have been taken from existing works in order to keep an optimal
numerical management of the two mechanisms of plasticity. Due to the coupling, some
choices had to be made concerning the priority and updates to give to each variable at each
step.
Using the developed integration routines, the model ACMEG-s has been implemented in
the finite element code LAGAMINE in the objective of extending the analysis to boundary
value problems. The field equations have been reviewed and the formulation of the multi-
phasic coupled element MWAT has been developed. A particular attention is paid to the
fluid flow models, already implemented in the finite element code and considered as valid
for our analyses. The problem takes then a third dimension of coupling, with the mechanical,
retention and hydraulic models that are interconnected together.
Two applications are proposed for the model, namely the construction of Mirgenbach
earth dam and the infiltration tests of Rüdlingen landslide test. It is in particular
demonstrated in the first case that only an adequate partially saturated approach can
reproduce with accuracy the generated pore pressures within the soil layers. The model
enables thus a natural transition between the saturated and partially saturated media, which
is of particular interest for the analysis of the influence of the hydraulic boundary conditions.
9.2 Perspectives
Regarding the contributions of the PhD. thesis, it appears that more insights into the
mechanics of unsaturated soils are still possible and necessary.
The experimental characterization of the behaviour of unsaturated soils will continue to
provide needed evidence. In particular, the characterization of the soil water retention curves
still requires more characterization with respect to the coupled mechanical effects. Three
main issues are not clearly addressed at present, which are:
(i) The shape of the retention curves under very low void ratios
(ii) The cyclic water retention behaviour at high levels of suction
(iii) The evolution of the capillary hysteresis with mechanical cycles.
The issue of effective stress has been presented here mainly from the conceptual point of
view. It would however be of particular interest to intend to demonstrate experimentally the
nature of the effective stress (generalized), in terms for instance of null tests. The
experimental control of the effective stress is thus a key point. More analysis could also be
carried out concerning the quantification of the effective stress at residual degrees of
saturation, where the water phase is strongly discontinuous and thus the pore water
pressure uneasy to quantify.
At the constitutive level, a fully coupled stress-strain and retention model has been
proposed. Some laws had to be simplified in the model, in particular for the “terms of
252 CHAPTER 9
coupling”. The effects of capillarity on the compressibility coefficients or else the dependency
of air entry value on void ratio could be more accurate than the actual simplified linear
relationships. However, such a model improvement will require more experimental evidence
to justify the increased complexity in the parameter determination.
While the applications to a model for unsaturated soils are numerous, some research level
full-scale experiments are still missing. The specifications for such tests would be not only to
model physically an engineering problem (e.g. embankment, tunnel, road) but also to
provide a complete set of data with respect to stress-strain and water retention behaviour in
cyclic conditions.
Lastly, it is very likely that the unsaturated soil mechanics will progressively take the
place of the saturated soil mechanics. The couplings between capillarity and stress-strain
behaviour being once fully understood and predictable, there will be room for other levels of
interaction. The unsaturated soil behaviour is promised to be considered as the reference
state to introduce the effects of soil structure, time, temperature, chemistry and the
associated couplings. From the modelling point of view, a model such as ACMEG-s should
then be complemented with those various environmental effects.
A. Appendix A: Conventions, notations
A.1 Sign convention

The sign convention of soil mechanics is adopted throughout the manuscript, i.e. the
compression is positive while the traction is negative.
A.2 General conventions

A.2.1 Functional
f ( x ) f is a function of variable x .
A.2.2 Variables
x, x scalar.
x, X vector.
x, X tensor.
A.2.3 Operators
d (.) increment
x rate of x
d ./ dx derivative with respect to variable x
∂./ ∂x partial derivative with respect to variable x
grad(.) gradient
div(.) divergence
Δ ( .) variation
A.3 Stresses and strains

The stress tensor σ ij and strain tensor ε ij are written in three dimensions:
⎡σ 11 σ 12 σ 13 ⎤ ⎡ε11 ε12 ε13 ⎤

σ ij = ⎢⎢σ 21 σ 22 σ 23 ⎥⎥ ε ij = ⎢⎢ε 21 ε 22 ε 23 ⎥⎥ (A.1)
⎢⎣σ 31 σ 32 σ 33 ⎥⎦ ⎢⎣ε 31 ε 32 ε 33 ⎥⎦
254 APPENDIX A
Those two tensors are symmetric σ ij = σ ji ; ε ij = ε ji
The invariant of the stress tensor are written :
First invariant J1 = tr (σ ) = σ 11 + σ 22 + σ 33 (A.2)
1 1
Second invariant J 2 = σ ijσ ji = tr (σ )
2
(A.3)
2 2
1 1
Third invariant J 3 = σ ikσ kmσ mi = tr (σ )
3
(A.4)
3 3
The stress tensor can be decomposed into two symmetric tensors that are the tensor of
deviatoric stress tij and the tensor of volumetric stress pδ ij :
σ ij = t ij + pδ ij (A.5)
p is the mean pressure defined as
J1 σ 11 + σ 22 + σ 33
p= = (A.6)
3 3
The invariants of the deviatoric stress tensor t ij are written:
First invariant J1D = 0 (A.7)
1
Second invariant J 2 D = t ij t ij
2
1
Third invariant J 3 D = t ik t km t mi (A.9)
2
q is the deviatoric stress defined as
( )
1/ 2
2⎡ 2 2⎤
q = 3J 2 D = ( σ − σ ) + ( σ − σ ) ( σ − σ ) + σ + σ + σ
2 2 2 2 2
6 (A.10)
2 ⎢⎣ ⎥⎦
11 22 22 33 33 11 12 13 23
The same quantities are defined for the strain tensor. Only the invariants of the strain
tensor of interest are mentioned below:
First invariant of the strain tensor I1 = ε11 + ε 22 + ε 33 (A.11)
1
Second invariant of the deviatoric strain tensor I 2 D = Eij Eij (A.12)
2
1
With Eij = ε ij − ( ε11 + ε 22 + ε 33 ) δ ij
3
The volumetric ε V and deviatoric ε d strains are defined as:
εV = I1 = tr (ε ) (A.13)
CONVENTIONS, NOTATIONS 255
2 1/ 2
εd = ⎡⎣(ε11 − ε 22 ) 2 + (ε 22 − ε 33 ) 2 +(ε 33 − ε11 ) 2 ⎤⎦
3 (A.14)
256 APPENDIX A
B. Parameter determination for Sion silt
B.1 Foreword
This appendix delivers complementary information on the determination of parameters
for ACMEG-s model for the material called Sion silt, essentially with respect to chapter 4.
Only the simplified retention model is featured here. The modelled material has been
studied in LMS- EPFL. The reported experimental results are taken mainly from the
references: Geiser (1999), Rifa’i (2002), Péron (2008) (see list of references of chapter 4). In the
last part B4 are plotted some results of predictions concerning the deviatoric behaviour and
the soil response to suction cycles (free deformation or wetting pore-collapse).
B.2 Table of parameters

Table B.1 Parameters for Sion silt (ACMEG-s version 1)
K ref Bulk modulus 170.5 MPa
Elastic
parameters
Gref Shear modulus 56.7 MPa
n e
Elastic exponent 1
Reference φ′ Friction angle 32.3 °
stress- β0 Compressibility coefficient 42.6
model Plastic
parameters a 0.0012
(Hujeux
b 0.8
1985)
c 0.0055
d 2
Limits of e
rdev 0.01
elastic
domain r e
iso 0.01
αs 4e-6
Retention model
ns 1.25
ms 1
258 APPENDIX B
B.3 Parameter determination

B.3.1 Determination of retention parameters
The Van genuchten parameters α s , ms , ns are determined on the basis of a few
experimental points (minimum 3). The shape of the retention curve has been voluntarily
slightly flattened to avoid numerically induced inconsistencies at high suctions. The air entry
suction se is determined with the tangents intersection method on the experimental points.
1.2
1
r
0.8
0.6
0.4
0.2
Exp.
Van Genuchten
0 5 7
1000 10 10
Matric suction (Pa)
Figure B.1 Calibration of retention model
B.3.2 Determination of mechanical parameters

The evolution of preconsolidation pressure with suction is modelled on the basis of 3
levels of saturation, as plotted in Fig. B.2. This determination lies on the verification of an
identical initial state for the 3 tests. The LC curve is partly determined by the saturated
preconsolidation pressure and by the air entry suction se . The last parameter γs serves to fit
the experimental points.
SION SILT PARAMETERS 259
5
4 10
Exp.
ACMEG-s
5
3 10

5
2 10
5
1 10
s
e
p'
c0
0 5 5 5 5 6
2 10 4 10 6 10 8 10 1 10
Figure B.2 Calibration of the Loading Collapse yield curve
The isotropic modulus K ref , the saturated compressibility parameter β 0 ,the isotropic
radius riso and the other isotropic parameters are obtained by fitting a saturated isotropic
test, Fig. B.3.
-0.01
Exp.
ACMEG-s
0
0.01
0.02
0.03
0.04 5
10
Figure B.3 Calibration of the isotropic mechanical parameters.
Parameter Ω, accounting for the evolution of compressibility with suction, is determined

by adding one unsaturated isotropic test to Fig. B.3, leading to Fig. B.4.
260 APPENDIX B
0.01

0.02
0.03
s=0 Pa Exp.
0.04 s=0 Pa ACMEG-s
s=1e5 Pa Exp.
s=1e5 Pa ACMEG-s
0.05 5
10
Figure B.4 Fitting of the elasto-plastic slopes under two levels of suction.
With the assumption of a unique critical state line whatever the level of suction, the
friction angle φ ′ is determined on the basis of saturated critical state results (Fig. B.5)
2000
Exp.
1500
ACMEG-s
1000
500
0
0 400 800 1200 1600
Figure B.5 Critical state line determined under saturated conditions.
The shear modulus Gref , deviatoric radius rdev the other deviatoric parameters a, d and
the dilatancy parameter α are determined via a curve fitting of a conventional saturated
shearing test, as illustrated in Fig. B.6. In this test, the initial state of the material is normally
consolidated to an isotropic stress of 600 kPa. The suction remains at all times equal to zero.
6
2 10
(a)
6
1.6 10

6
1.2 10
5
8 10
5
4 10
Exp.
ACMEG-s
0
0 0.15 0.3
Axial strain ε1(-)
0
Exp. (b)
ACMEG-s
0.006
0.012
0.018
0.024
0.03
0 0.15 0.3
Axial strain ε1(-)
Figure B.6 Calibration of deviatoric parameters on a saturated shearing test carried out at a confining
pressure of 600 kPa.
B.4 Predictions
Here are presented some results of back predictions of the stress-strain behaviour of Sion
silt. For all the presented tests, an adjustment of the saturated preconsolidation pressure has
been necessary, considering the different initial states available from the experimental data.
B.4.1 Mechanical loading at various levels of suction

The prediction of the volumetric response to an isotropic mechanical loading paths
carried out under a constant suction of 200 kPa is plotted in Fig. B.7.
262 APPENDIX B
0.01
Exp.
s=100kPa
Exp.
0 s=200kPa

v
-0.01
-0.02 ACMEG-s
Exp.
s=0
-0.03
-0.04 5 5 6
4 10 7 10 10
Figure B.7 Prediction of isotropic mechanical compression at a suction level of 200 kPa
Three drained shear tests are simulated with the determined set of parameters. They all
correspond to normally consolidated states with a confining pressure of 600 kPa (Fig. B8).
6
2 10
(a)
6
1.6 10

6
1.2 10
5
8 10
Exp. s=0 kPa
ACMEG-s s=0 kPa
5 Exp. s=100 Pa
4 10 ACMEG-s s=100 Pa
Exp. s=200 Pa
ACMEG-s s=200 Pa
0
0 0.15 0.3
Axial strain ε1(-)
0
Exp. s=0 kPa
ACMEG-s s=0 kPa
0.006 Exp. s=100 kPa
ACMEG-s s=100 kPa

Exp. s=200 kPa
ACMEG-s s=200 kPa
0.012
0.018
0.024
(b)
0.03
0 0.15 0.3
Axial strain ε1(-)
Figure B.8 Prediction of deviatoric stress-strain behaviour in response to shearing in partially

saturated conditions (confining pressure equal to 600 kPa).
The brittleness observed in experimental data is not accounted for in the present
modelling framework. The deviatoric behaviour is also modelled under a confining pressure
of 400 kPa. Experimental results are reported for two level of suction, s = 0 KPa and s = 100
kPa. (Fig. B9).
264 APPENDIX B
6
1.5 10
(a)
6
1.2 10

5
9 10
5
6 10
Exp. s=0 kPa

5 ACMEG-s s=0 kPa
3 10
Exp. s=100 kPa
ACMEG-s s=100 kPa
0
0 0.15 0.3
Axial strain ε1(-)
0
Exp. s=0 kPa (b)
ACMEG-s s=0 kPa
0.006 Exp. s=100 kPa
ACMEG-s s=100 kPa
0.012
0.018
0.024
0.03
0 0.15 0.3
Axial strain ε1(-)
Figure B.9 Test in deviatoric conditions under a confining pressure of 400 kPa.
In Fig. B.9, while the saturated shear test is fairly well captured, some discrepancy is
observed between the simulation and experimental points at s = 100 kPa, probably due to the
critical state observed experimentally being not consistent with the trend of Fig. B.8.
B.4.2 Suction loading under different levels of net stress.

The volumetric variation due to drying path under zero net stress is modelled and
illustrated in Fig. B.10.
-0.02
Exp.
ACMEG-s
0

v
0.02
0.04
0.06
0.08 4 5 6
10 10 10
Figure B.11 shows the volumetric variations induced by drying and wetting back a
material under different mean net stresses. Depending on the mechanical stress state, the
phase of wetting at constant mean stress, may induce swelling behaviour (test SY34-54) or
plastic collapse (test SY34-55), which is correctly predicted by the model.
-0.01
SY34-54
SY34-54 Mod.
0 SY34-55
SY34-55 Mod.
v
0.01
0.02
0.03
0.04 5 5 5 5
1 10 3 10 5 10 7 10
266 APPENDIX B
C. Parameter determination for Kaolin
C.1 Foreword
This appendix is dedicated to the determination of parameters for ACMEG-s for kaolin
that has been characterized by Sivakumar (1993) and Wheeler and Sivakumar (1995) (see list
of references of chapter 4). The determined parameters are those defined in chapter 4 only,
that is to say that the version of the retention model is that of Van Genuchten.
The soil is described by Sivakumar (1993) as obeying the critical state concept under
saturated conditions. It features a 75% clay fraction, is commercially available in a uniform
and homogeneous form, has a relatively high rate of consolidation and its behaviour under
partially saturated conditions has been previously investigated thoroughly by several
research teams.
C.2 Table of parameters

Table C.1 Parameters for Kaolin (ACMEG-s version 1)
K ref Bulk modulus 9e7 Pa
Elastic
parameters
Gref Shear modulus 5e7 Pa
n e
Elastic exponent 1
stress- β0 Compressibility coefficient 16
strain α Dilatancy coefficient 2.5
model Plastic
parameters a 0.0012
(Hujeux
b 1
1985)
c 0.0055
d 2
Limits of e
rdev 0.1
elastic
domain re
iso 1
Coefficient of compressibility
Capillary effects Ω change
3.5e-5

268 APPENDIX C

αs 5e-7
Retention model
ns 1
ms 4
C.3 Parameter determination

C.3.1 Determination of retention parameters
The Van genuchten model (parameters α s , ms , ns ) is fitted on the available experimental
points (Fig. C.1). xperimental points (minimum 3). The deduced shape of the water retention
curve also gives an estimation of the air entry value se .
1.2
1
r
0.8
0.6
0.4
0.2
Exp.
Van Genuchten
0 5 7
1000 10 10
Matric suction (Pa)
Figure C.1 Calibration of retention model
C.3.2 Determination of mechanical parameters

The loading collapse curve is determined on the basis of isotropic compression tests
carried out at four different levels of suction, for which the initial saturated preconsolidation
pressure pc′ 0 is reportedly equal. The air entry value se was determined previously. The last
parameter γ s is calibrated to obtain the fitting of Fig. C.2.
KAOLIN PARAMETERS 269
5
4 10
Exp.
ACMEG-s
5
3 10

5
2 10
5
1 10
s
e
0 5 5 5 5
0 p' 1 10 2 10 3 10 4 10
c0
Figure C.2 Calibration of the Loading Collapse yield curve
The isotropic modulus K ref , the saturated compressibility parameter β 0 ,the isotropic
radius riso and the other isotropic parameters are obtained by fitting a saturated isotropic
test, Fig. B.3.
0
Exp.
ACMEG-s
0.05
0.1
0.15 4
10
Figure C.3 Calibration of the isotropic mechanical parameters.
Parameter Ω is determined by adding one unsaturated isotropic test to the previous figure
(at a suction of 200 kPa), Fig. C.4.
270 APPENDIX C
-0.05

0.05
0.1
0.15 s=0 Pa Exp.

s=0 Pa ACMEG-s
s=2e5 Pa Exp.
s=2e5 Pa ACMEG-s
0.2 4
10
Figure C.4 Fitting of the elasto-plastic slopes under two levels of suction.
The unique critical state line is calibrated via the angle φ ′ on the basis of saturated critical
state results (Fig. C.5).
200
150
Exp.
100
50 ACMEG-s
0
0 100 200
Figure C.5 Critical state line determined under saturated conditions.
The deviatoric parameters ( Gref , rdev , a, d , α )are calibrated on the basis of drained shear
tests undertaken under fully saturated conditions. In the present case, data from shear test at
suction of 100 kPa have also been used for the calibration, Fig C.6.
5
3 10 -0.04
(a) s=0 kPa Exp.
s=0 kPa ACMEG-s
5
2.4 10 s=100 kPa Exp.
0 s=100 kPa ACMEG-s

v
5
1.8 10
0.04
5
1.2 10
4 s=0 kPa Exp. 0.08

6 10 s=0 kPa ACMEG-s
s=100 kPa Exp.
s=100 kPa ACMEG-s (b)
0 0.12
0 0.15 0.3 0 0.15 0.3
Axial strain ε1(-) Axial strain ε1(-)
Figure C.6 Calibration of deviatoric parameters on a saturated shearing test carried out at a confining
pressure of 150 kPa.
C.4 Predictions
Several types of behaviour are simulated and compared to experimental results. The
reference initial state for each test enables defining the suction level and preconsolidation
pressure.
C.4.1 Wetting pore collapse

Most of the stress paths tested by Sivakumar (1993)] are complex and involve
combinations of net stress and suction loading. The first process simulated here is the
equalization stage: the initial state of the kaolin is relatively dry (s = 3e5 Pa). An external load
of 4e4 Pa is applied while the material is wetted to the wanted suction. In the presented case
of Fig. C.7 for instance, a final suction equal to zero is reached. The three presented
experimental tests have the same initial state and wetting rate: the experimental points thus
describe the same test carried out three times. The wetting process is at the origin of swelling
(reversible process), followed by plastic pore collapse and then a residual swelling. The
trends and amplitudes are well predicted by the model, even though the start of yielding
could be adjusted for better results. The volumetric response at high suctions is highly
dependent on the soil water retention parameters and thus on the water retention curve
formulation.
272 APPENDIX C
0.02
(c)
B
0.01

A
0
-0.01 Wetting
D
-0.02
C Exp. 1
Exp. 2
-0.03
Exp. 3
ACMEG-s
-0.04 5 5
0 2 10 4 10
Figure C.7 Prediction of wetting pore collapse during equalization to the null target suction.
C.4.2 Isotropic mechanical behaviour under different levels of suction

In some cases presented by Sivakumar (1993), the equalization process is preceded or
followed by an isotropic compression phase carried out at constant suction. Fig. C.8
compares the numerical prediction to the experimental data along these stress paths,
including both the equalization steps and the subsequent mechanical consolidation. Fig C.8
shows a satisfactory description of the evolution of preconsolidation pressure with suction.
However, it appears that the compressibility coefficient of model is realistic only for the
levels of suction up to 200 kPa.
0
v
-0.04
-0.08 s=0 kPa Exp.

s=0 kPa ACMEG-s
s=100 kPa Exp.
s=100 kPa ACMEG-s
s=200 kPa Exp.
s=200 kPa ACMEG-s
-0.12 s=300 kPa Exp.
s=300 kPa ACMEG-s
5
10
Figure C.8 Prediction of volumetric response to isotropic compressions after equalization to 4 different
levels of suction.
C.4.3 Deviatoric behaviour under constant levels of suction

Shear tests under 4 different levels of matric suction (0 kPa, 100 kPa, 200 kPa, 300 kPa) are
simulated with AGMEC-S (see previous part C.3 to identify the tests used for calibration).
Figures 3.3.a and 3.3.b below give the superimposition of experimental and numerical
results. The confining pressure for all tests is 150 kPa
5
4 10 -0.04
s=0 kPa Exp.
s=0 kPa ACMEG-s
s=100 kPa Exp.
5 s=100 kPa ACMEG-s
3 10
s=200 kPa Exp.

0

s=200 kPa ACMEG-s
s=300 kPa Exp.
v
s=300 kPa ACMEG-s
5
2 10
0.04
5 s=0 kPa Exp.
1 10 s=0 kPa ACMEG-s
s=100 kPa Exp.
s=100 kPa ACMEG-s 0.08
0 s=200 kPa Exp.
s=200 kPa ACMEG-s
s=300 kPa Exp.
(a) s=300 kPa ACMEG-s (b)
0.12
0 0.2 0.4 0 0.15 0.3
Axial strain ε1(-) Axial strain ε1(-)
Figure C.9 Prediction of deviatoric stress-strain response to shearing in partially saturated conditions
(confining pressure 150 kPa)
274 APPENDIX C
D. Parameter determination for Bentonite
D.1 Foreword
In this appendix are determined the parameters of the bentonite of FEBEX studied by
Lloret et al. (2004) (see list of references of chapter 4). Here are provided not only the
parameters for Van Genuchten water retention model but also those of the version of the
retention model with capillary hysteresis and volume dependency.
D.2 Table of parameters

Table D.1 Parameters for Bentonite
Elastic
parameters
n e
Elastic exponent 1
stress- β0 Compressibility coefficient 42.6
model Plastic
parameters a 0.001
(Hujeux
b 0.8
1985)
c 0.005
d 2
Limits of e
rdev 0.5
elastic
domain risoe 0.5
se 0 Reference air entry value 1.5e6 Pa
αs 1e-7
Retention model
ns 3.5
ms 0.2
276 APPENDIX D

KH Elastic modulus 6.5
βH Plastic coefficient 12
Retention model
sD Drying yield suction 1.78e7 Pa
(ACMEG-s)
πH Parameter of coupling 3e5
D.3 Parameter determination

D.3.1 Determination of retention parameters
The hydraulic parameters are determined on the basis of several information sources.
Most of the available information on water retention is linked to different material densities.
As the first version of the model is not capable of accounting for such volume-dependent
features (Van Genuchten model, parameters α s , ms , ns ), an averaged soil water retention
curve is determined, see Fig. D.1.
1.2
1
r
0.8
0.6
0.4
0.2
Exp.
Van Genuchten
0 5 7 9
1000 10 10 10
Matric suction (Pa)
Figure D.1 Calibration of Van Genuchten retention model
In the complete version of ACMEG-s model, the retention model features a capillary
hysteresis and a dependency on void ratio, Fig. D.2. The retention model has to be calibrated
for one initial volume of reference. Here the drying-wetting test carried out at an initial
density of 1.50 g .cm −3 was used. Once the air entry value se is calibrated, the elastic slope
K H , drying yield suction sD and plastic coefficient β H are adjusted, followed by the
residual degree of saturation S res . A second set of experimental points ( S r , s ) determined at
a different initial density is then necessary to calibrate the parameter π H .
BENTONITE PARAMETERS 277
s s
e ref e updated
1.2

r
0.8 ACMEG-s
3
(ρ =1.63 g/cm )
d
0.6 ACMEG-s
3
(ρ =1.50 g/cm ) Exp. points
d
0.4
0.2
0
0.1 10 1000
Figure D.2 Calibration of ACMEG-s water retention model.
D.3.2 Determination of mechanical parameters

Although several œdometric compression results are available, the apparent
preconsolidation pressures cannot be compared and grouped in the same plot, considering
that the initial states are different (sample underwent either a hydraulic loading, a
mechanical loading or both). Moreover, isotropic compression tests are preferred to
œdometric ones, in the sense that they enable a direct interpretation. γ s is determined via the
fitting of the swelling pressure tests, see chapter 4. The determination of elastic moduli
( K ref , Gref ) and compressibility parameter ( β 0 ) is based on the fitting of a saturated
oedometric compression test and swelling tests upon wetting. The parameters for the size of
the elastic domain and the isotropic and deviatoric mechanisms ( a, b, c, d , rdev , riso ) are also
calibrated at this point (Fig. D.3).
Exp.
Exp.
-0.3 ACMEG-s -0.3
ACMEG-s
-0.2
-0.2
-0.1
-0.1
0
(a) (b)
0 4 7
10 10 5 8
10 10
Vertical net stress σvnet(Pa) Matric suction s(Pa)
Figure D.3 Calibration of the isotropic mechanical parameters.
No clear evidence could justify the use of parameter Ω for bentonite, so it was set to zero.
278 APPENDIX D
With the assumption of a unique critical state line whatever the level of suction, the
friction angle φ ′ is determined on the basis of saturated critical state results, Fig. D.4. The
reference critical state lines are taken for a given density of 1.65 g .cm −3 .
2000
Exp.
Deviatoric stress q (kPa) 1500
1000
500 ACMEG-s
0
0 400 800 1200 1600
Figure D.4 Critical state line determined under saturated conditions.
D.4 Predictions
Here are presented some predictions of the model along drying-wetting paths and
mechanical loading paths carried out under oedometric conditions. The predictions of the
swelling pressure tests, already thoroughly discussed in chapters 4 and 6, are not presented
here.
Lloret et al. (2004 report the generalized stress paths for 5 tests for which the loading
paths are combinations of wetting, drying, mechanical compression and unloading. Fig. D.5
recapitulates the stress paths for five tests 1 to 5:
9
10
Test 1
Initial Test 2
8
10 point
Test 3
7
10
Test 4
6
10
Final
5
point Test 5
10
4 6 8
10 10 10
Vertical stress σv (Pa)
BENTONITE PARAMETERS 279
Figure D.5 Stress paths for tests S1 to S5. The horizontal axis represents the vertical net stress (from
Lloret et al. 2004)
In the following, a distinction is made between the suction loading paths and the
mechanical steps under a constant level of suction. Fig. D.6 plots the volumetric variations
during the wetting/drying phases of the tests. Notice that the mechanical compressions also
appear on the graph in order to get the comprehensive history of the material.
-0.2
Exp. test 1 Exp. test 3
ACMEG-s test 1 ACMEG-s test 3
Exp. test 2 Exp. test 4
-0.15
ACMEG-s test 2 ACMEG-s test 4
-0.1
-0.05 Initial
point
Wetting
0.05 5 7 9
10 10 10
Figure D.6 Prediction of volume variation in response to stress paths of Fig. D.5.
To get a better view of individual results for each single test, Figs D.7 and D.8 use a
separate plot for each test. The major discrepancy between numerical simulation and
experimental data is observed for test 2 However, the volumetric variations under the
conditions of test 2are not consistent with those of similar tests (wetting under mechanical
load, e.g. tests 1 and part of test 3).
-0.2 -0.2
Exp. test 1 ACMEG-s test 1 Exp. test 2 ACMEG-s test 2
-0.15 -0.15
-0.1 -0.1
-0.05 Initial -0.05 Initial

point point
Wetting
0 0
Wetting
0.05 5 7 9
0.05 5 7 9
10 10 10 10 10 10
Figure D.7 Prediction of volume change for tests 1 and 2.

280 APPENDIX D
0.2
0.2 Exp. test 4 ACMEG-s test 4
Exp. test 3 ACMEG-s test 3
0.15

0.15
v
0.1
0.1
0.05 Initial
0.05 Initial
point
point
Wetting
0 0
Wetting
-0.05 -0.05 5 7 9
5 7 9
10 10 10 10 10 10
Figure D.8 Prediction of volume change for tests 3 and 4.
Simulations of the subsequent or prior mechanical loading under constant levels of

suction are gathered in Fig. D.9. Notice that during the drying/wetting paths, the constant
vertical net stress is not constant during numerical simulations. This is attributed to the very
important variations in suction (from 0 to 1000 MPa) causing significant variations of the
effective stress and thus of the net stress. However, subsequent oedometric compressions are
correctly predicted with the determined parameter set. Once more, the ‘initial sate’ before
mechanical compression is narrowly linked to the wetting swelling, itself being highly
dependent on SWRC formulation.
0.4
(c) Exp. 1
Num. 1
Exp. 2
0.3 Num. 2
Exp. 3
Num. 3
0.2 Exp. 4
Num. 4
Exp. 5
Num. 5
0.1
-0.1 4 5 6 7 8
10 10 10 10 10
Figure D.9 Prediction of volume during oedometric compression at constant suctions.

E. ACMEG-s: influence of parameters and stress paths
E.1 Foreword
The objective of this appendix is to discuss further the qualitative response of the
developed constitutive model, with respect to the influence of parameters and their domain
of definition. Some remarks are also formulated on the stress paths. This appendix is a
complement to chapters 4, 5 and 6. The aspects covered here are the loading collapse yield
limit, the mechanical response to suction changes and the model for the soil water retention
behaviour.
E.2 Loading collapse yield curve

The qualitative shape of the yield limit called Loading Collapse (LC) curve is investigated
in the following, which is one of the key aspects of the ACMEG-s model. The Loading
Collapse yield curve being determined in plane ( s − p′ ) , a preliminary discussion on the
possible stress states in this plane is featured hereafter.
E.2.1 Domain of attainable stress states

The use of the effective stress induces a certain number of particularities in stress path
representation. As explained in details in chapter 3, a wetting path has a non-linear
representation in ( s − p′ ) plane, Fig. E.1. Conceptually, independently from the adopted
formulation for the retention part of the model, as displayed in Fig. E.2, a wetting line should
not reach the hatched zone in plane ( Sr − s ) , called domain 1. In other words, domain 1
corresponds to the saturated state, where the degree of saturation is uniquely equal to unity.
Furthermore, the domain of possible stress paths can be deduced in complementary plane
( s − p′ ) . In Fig. E.1, the unattainable domains (1 and 2) are either hatched or shadowed. The
zone of possible stress paths is thus bounded on its left by the ‘standard wetting path’, that is
the wetting at a zero net stress. Due to the limitation of the analysis to compressive net
stresses only, the attainable stress states remain on the right hand side of the standard
wetting line (white zone in Fig. E.1).
282 APPENDIX E
7
10

6 unattainable
10 domain 2
s
e
5
10 standard
domain 1 wetting path
4
10 4 6
10 10
Figure E.1 Domain of possible stress states in ( s − p′ ) plane (white zone).
7
10
6
10
s
e
5
10
unattainable domain 1
4
10
0.2 0.6 1
r
Figure E.2 Domain of existence of water retention curves (white zone).
E.2.2 Qualitative parametric analysis

The influence of parameters pc′0 , se , γ s on the shape of the loading collapse is studied
hereafter. Notice that parameter γ s is the only one to remain a fixed material constant,
provided that the saturated preconsolidation pressure pc′0 is modified upon yielding
(mechanical model) and that the air entry value se depends on volumetric strain (retention
model).
The saturated preconsolidation pressure pc′0 reflects the mechanical history of the
material and sets the origin of the Loading Collapse curve. Increase in saturated
preconsolidation pressure does not only shift the reference point along mean effective stress
axis p′ but also accentuates curvature of the loading collapse curve in the partially saturated
domain (Fig. E.3), due to the definition Eq. 4.18 given in chapter 4.
INFLUENCE OF PARAMETERS 283
5
4.5 10

5
3 10
5
1.5 10
p' (Pa) =
c0
1.5e5
2.2e5
3.5e5
0 5 5
0 4 10 8 10
Preconsolidation pressure p' (Pa)
c
Figure E.3 Influence of parameter pc′0 on the shape of Loading Collapse curve, se = 105 Pa, γ s = 3 .
The air entry value se sets the boundary between the saturated and the unsaturated
domain and is inherited from the soil water retention model. In plane ( s − p′ ) , the air entry
value pilots the size of the linear part of the LC curve, Fig. E.4.
6
1.6 10
6
1.1 10
s (Pa) =
5 e
5.3 10 1e3
1e4
5e4
1e5
2e5
4e5
0 6 6
0 1 10 2 10
c
FigureE.4 Influence of parameter se on the shape of Loading Collapse curve, pc′0 = 1.5 × 105 Pa, γ s = 3 .
Parameter γ s is used to set up the shape of the non-linear part of the loading collapse
curve. Fig E.5 gives the typical response to variations in γ s .
284 APPENDIX E
5
4.5 10

5
3 10
γ =
s
5
1.5 10 2
3
4
6
8
10
0 5 5 6
0 4 10 8 10 1.2 10
c
Figure E.5 Influence of parameter γ s on the shape of Loading Collapse curve,

pc′0 = 1.5 × 105 Pa, se = 105 Pa .
E.3 Stress-strain response

The qualitative response of the model to the stress path of the mechanical type is
investigated first (that is change of mechanical external stress under a constant level of
suction). The accent is indeed mostly laid on the effect of suction levels on the modelled
response. Then, the response to suction loading-unloading is discussed.
E.3.1 Isotropic stress states

The plastic compressibility is imposed to depend on the suction level in a linear fashion
(see chapter 4). Below are plotted numerical results from various isotropic simulations under
constant levels of suction. The parameters used for Figs. E.6 and E.7 are the following:
pc′0 = 105 Pa, se = 105 Pa, γ s = 6, β 0 = −38.85

(E.1)
K ref = 100 × 106 Pa, Ω = 3.5 × 10−5
The reader may notice that progressive mobilisation of plastic mechanisms is

provisionally neutralised in order to improve legibility of figures. Here is shown an extreme
example where the slope of the plastic compressibility could become infinite or even reverse.
The integration process does thus include a test in order to prevent such inconsistencies in
the values of calculated β m .
6
1.2 10
LC
path 0
path 4e5
5
9 10 path 1e6

5
6 10
5
3 10
0 5 5 5 6
0 3 10 6 10 9 10 1.2 10
Figure E.6 Stress paths for isotropic compression
0
path 0
path 4e5
path 1e6
0.04
v
0.08
0.12
4 5 6 7
10 10 10 10
Figure E.7 Simulated volumetric response to isotropic compression
E.3.2 Oedometric stress states

The integration routine is capable to simulate oedometric compression tests in saturated
and unsaturated states. Below is plotted an array of results for oedometric compressions
under three constant levels of matric suction, starting form a given identical initial state. The
vertical parts of the mechanical response in plane ( ε v − ln σ vnet ) obviously correspond to the
wetting process under constant vertical net stress. The qualitative trends obtained for the
oedometric simulations are satisfactory.
Parameters of interest for this simulation are the following:
pc′0 = 3 × 105 Pa, se = 6 × 104 Pa, γ s = 5, β 0 = −15

(E.2)
K ref = 50 × 106 Pa, Ω = 1 × 10−5
286 APPENDIX E
v
-0.083
-0.17
s (kPa) =
0
250
500
-0.25 5 7
10 10
Vertical net stress σ (Pa)
vnet
Figure E.8 Simulated volumetric response to oedometric compression
E.3.3 Volumetric response to suction changes.

The yield locus defined in chapter 4 is such that in the saturated domain, the
preconsolidation pressure is not modified with suction/degree of saturation (vertical line in
plane ( p′ − s ) ), while in the unsaturated domain, where the capillary effects are activated,
the shape of the yield curve is modified, with the apparent preconsolidation stress increasing
faster than the mean effective stress upon suction increase.
The resulting stress-strain behaviour, represented in chapter 4, featured conventional
elastic and plastic slopes according to the stress state being or not inside the elastic domain.
Conventionally, the volumetric response of drying wetting process is represented as a
function of matric suction rather than effective stress. The (ε v − ln s ) plane can be
superimposed with the mechanical plane ε v − ln p′ , Fig. E.10. For an initial net stress of 10 ( )
KPa, a strong non linearity of the curve (ε − ln s )
v can be noticed, even though the two
responses converge at the neighbourhood of the air entry value.
Effective stress
Suction
-0.015
v
-0.03
-0.045
s
e
-0.06
104 105
Mean effective stress p' (Pa), suction s (Pa)
Figure E.9 Volumetric response to drying wetting path as a function of mean effective stress and
suction. The air entry value for this generic material is 100 kPa.
E.4 Water retention model: Van Genuchten formulation

Here is briefly investigated the influence of each of the three parameters in the Van
Genuchten formulation on the shape of the non-linear reversible Soil Water Retention Curve.
The reference values for Van Genuchten parameters used in the following are:
α s = 4.10 −6 ; ns = 1.25; mr = 1 (E.3)
1
(a)
0.8
Degree of saturation Sr (-)
0.6
0.4 α =
s
4e-7
1.8e-6
0.2 4e-6
1.8e-5
4e-5
0 5 7
1000 10 10
1.2
(b)
1
0.8
0.6
0.4
n=
s
0.2 0.5
0.9
1.25
0 1.9
5
-0.2 5 7
1000 10 10
Figure E.10 (a) Influence of parameter α s ; ns = 1.25; mr = 1 , (b) influence of parameter ns ;

α s = 4.10 −6 ; mr = 1 .
288 APPENDIX E
0.8

0.6
0.4
m=
r
0.2
0.1
0.5
0 1
10
20
-0.2 5 7
1000 10 10
Figure E.11 Influence of parameter mr ; α s = 4.10 ; ns = 1.25 .

−6
E.5 AMEG-s soil water retention model, influence of parameters

The advanced version of the hydraulic model presents more features that the previous
simple reversible version. The number of parameters is also superior. Below is investigated
the influence of the parameter values on the shape of the soil water retention model.
Particularly, the aperture of the hydraulic hysteresis is of major interest.
The reference values for the parametric study are the following:
se = 700Pa; K H = 6 ×106 Pa; β H = −1; sD = 1600Pa; SrRES = 0.07 (E.4)
In Figs. E.12 to E.14 below, the indices (d) and (w) respectively refer to drying and
wetting. Reader may notice that due to the coupling between retention curve and stress-
strain model, the air entry value is likely to be modified due to mechanical straining. This
will affect the shape of the soil water retention curve in the fashion of the parametric analysis
of Fig. E.12a.
1.2 1.2
(a) (b)
1 1

0.8 0.8
0.6 0.6
s (Pa)= K (Pa)=
e H
0.4 300 (d) 0.4 2e6 (d)
300 (w) 2e6 (w)
700 (d) 6e6 (d)
0.2 700 (w) 0.2 6e6 (w)
1500 (d) 6e7 (d)
1500 (w) 6e7 (w)
0 4
0 4
100 10 100 10
Figure E.12 (a) Influence of parameter se , (b) influence of parameter K H .
1.2 1.2
(a) (b)
1 1
0.8 0.8
0.6 0.6
β = s (Pa)=
H D
0.4 -3 (d) 0.4 800 (d)

-3 (w) 800 (w)
-1 (d) 1600 (d)
0.2 -1 (w) 0.2 1600 (w)
-0.1 (d) 3200 (d)
-0.1 (w) 3200 (w)
0 4
0 4
100 10 100 10
Figure E.13 (a) Influence of parameter β H , (b) influence of parameter sD .

290 APPENDIX E
1.2

0.8
0.6
S =
rRES
0.4 0.007 (d)

0.007 (w)
0.07 (d)
0.2 0.07 (w)
0.7 (d)
0.7 (w)
0 4
100 10
Figure E.14 Influence of parameter S rRES .

F. Mirgenbach earth dam: review of published
characteristics
F.1 Foreword
The following tables summarize the information collected from all the reports internal to
Electricité de France (EDF) concerning the case of Mirgenbach earth dam. The list of reports
can be found in chapter 7. As mentioned in chapter 7, some differences were identified
between the characteristics published in the various reports. The synthesis proposed here
serves as a basis for the decision making of chapter 7.
F.2 Summary of characteristics
Table F.1 Values collected from EDF reports.
w Sr φd cd φu cu γd γ d max DC K
Part of the construction wopt [%]
[%] [%] [°] [kPa] [°] [kPa] [kN/m3] [kN/m3] [%] [m/s]
Technical document of the Constructor, 1979
Clay in ground before excavation for plant. Page 17 170 16.4
Clay of ground of excavation for plant. Page 18 21.5 22 16.3
Clay in deposit January 1979. Page 58 24.5
22 - 26
Clay in deposit June 1979. Page 59 23.7
19.5–26.5
Clay in deposit 1979. Page 63 21.5 16.4
Clay in deposit 1979. Page 68 21 25
Clay in ground before excavation for plant. Page 68 24 25
[%] [%] [°] [kPa] [°] [kPa] [kN/m3] [kN/m3] [%] [m/s]
Technical report. 1982. EDF
Clay in dam. Layer 81. After winter brake 81/82. 21.5 26
page 3
Technical report. 1983. EDF - REAL
Clay in dam. Layer 81. page 1.9 22.5 23.9 15.9 16.3 97.8
Clay in dam. Layers of 82. page 1.9 22.6 22.9 16.2 16.4 98.8
Clay after failure. Layer 81. page 1.11 60
Clay after failure. Layers of 82. page 1.11 100
On slide plane. Page 1.11 28
Colloque technique CFGB 1984.
Foundation. Page 139 1e-5
Clay in dam. Layers 81. page 147 22.5 23.9 16.6
Clay in dam. Layers of 82. page 147 21.6 22.9
Groupe de Travail du C.F.G.B . Compactage des sols argileux humides.1998
Clay putting in dam. Page 5 95
Clay putting in dam during September 1981. Page 5 97
Clay in dam. Layers of 82. page 6 W - Wopt 95 96
1.3
Clay in dam. Layer 81. Away of zone of failure. page 6 W - Wopt 97.5 73
2.6
Clay in dam. Layer 81. In zone of failure. page 6 W - Wopt 96 57
3.3
Clay in dam. Layer 81. On slide plane. page 6 W - Wopt 99 35
5.6
294 APPENDIX F
[%] [%] [°] [kPa] [°] [kPa] [kN/m3] [kN/m3] [%] [m/s]
Modélisation de la rupture du barrage de Mirgenbach. Electricité de France. 2003
Clay. Campaign 1980 (90 checks). Page 7/21 22.5 23.9 94.5 15.9 16.1 99
Clay. Campaign 1981. Second half of September 22.5 24.1 97.5 16 16.1 98
(27 checks). Page 7/21
Clay. Campaign 1981. Not second half of September 22.5 23.9 94.5 15.9 16.1 97.5
(210 checks). Page 7/21
Clay. Campaign 1982 (317 checks). Page 7/21 21.6 22.9 95.5 100 16.2 16.4 99
Clay in dam. Page 8/21 and 9/21 21 25 kH
1e-8
kV
1e-9
Clay in dam after accident. Page 9/21 20 10
Clay. Numerical modeling. Page 16/21 23 25 16.7 1e-10
Foundation. Numerical modeling. Page 16/21 35 1000 24.5 1e-14
Mathieu NUTH
French, born 10/12/1981
Engineer – Ecole Centrale Nantes

Laboratory of Soil Mechanics (LMS-ENAC)
Ecole Polytechnique Fédérale de Lausanne (EPFL)
EPFL-LMS Station 18, 1015 Lausanne, Switzerland

+ 41.21.693.23.32 (Office)
+ 41.79.839.71.23 (Mobile)
mathieu.nuth@epfl.ch
EDUCATION
Since Jan. 2005 Ph.D. Candidate, Laboratory of Soil Mechanics – EPFL.

Thesis title: 'Advanced modelling of unsaturated soils, constitutive and hydro-
mechanically coupled finite element analysis'', Supervisor Prof. L. Laloui
Teaching assistant for Civil Engineering section
2004 Ecole Centrale Nantes (France) Engineer diploma (M.Sc.) – 'Mention Bien'
April – Sept. 2004 Master thesis, LMS-EPFL : 'Numerical modelling of heat exchanger piles'
2001-2004 Undergraduate student at Ecole Centrale Nantes - Civil engineering and
environment, 'Construction' section
1999-2001 ‘Classes préparatoires aux grandes écoles', Physics & Technology. Angers (F)
1999 Scientific French Baccalauréat, 'Mention Très Bien'
INDUSTRY
December 2003 Participant to Bouygues Construction Challenge
June- Sept 2003 Field engineering training period – 'GTB Construction' Bouygues, Angers (F).
Housing building operation
June- Aug. 2002 Programmer, database managing – Vecteur Plus, computer services, Phnom
Penh, Cambodia
2001-2002 Visitors' technical guide – Arcelor Packaging Basse-Indre (F)
LANGUAGES
French Mother tongue
English Fluent. First Certificate of Cambridge University
Spanish Read, written, spoken. Diploma Basico de Español
COMPUTER TOOLS
Programming C/C++, Fortran, Visual Basic
Modelling Petal, Cesar LCPC, Algor/Superdraw3, GefDyn, GID, ZSoil, Lagamine
Database MySQL, Access
Internet Jahia, Typo3, Flash
Graphics 3D Studio, Photoshop, Première, After effects
1/3
AWARDS
December 2008 Article “Effective Stress Concept in Unsaturated Soils: Clarification and
Validation of an Unified Framework” (IJNAMG) rated in the top 5 most read
papers of the journal.
June 2008 Best PhD thesis in Civil Engineering – AUGC René Houpert Prize, French
association for civil engineering
June 2006 Best poster award of EPFL-ENAC PhD Students Day 2006 : "Unified stress
framework for advanced constitutive modelling of unsaturated soils"
April 2004 Ecole Centrale Nantes Special Grant for diploma work abroad
PUBLICATIONS
Peer-reviewed journal papers
Nuth M., H. Peron, L. Laloui. Intelligent realisation of ground energy. GeoDrilling International,
no 141 April (2008), p. 28-29.
Nuth M., Laloui. L., “Advances in modelling hysteretic water retention curve in deformable soils”,
Computers and Geotechnics, (2008), vol 35, num 6, p 835 844.
Laloui L., Nuth M., “On the use of the generalised effective stress in the constitutive modelling of
unsaturated soils”. Computers and Geotechnics, 2008, in press.
Nuth M., Laloui L., “Effective Stress Concept in Unsaturated Soils: Clarification and Validation of
an Unified Framework”. Int. Journ. of Numerical and Analytical Methods in Geomechanics, no
32(2008), p. 771-801.
Nuth M., Laloui L., “Unified stress framework for modelling unsaturated subsoil behaviour”. Int.
Journal of Road Materials and Pavement Design, no 8(4)(2007), p. 767-781.
Laloui L., Nuth M., “Numerical Modeling of Some Features of Heat Exchanger Pile”. In ASCE
Geotechnical Special Publication: Foundation analysis and design-Innovative methods, (2006) p.
189-194.
Laloui L., Nuth M., Vulliet L., “Experimental and numerical investigations of the behaviour of a
heat exchanger pile”. International Journal for Numerical and Analytical Methods in
Geomechanics, vol. 30, no 8 (2006) p. 763-781.
Laloui L., Nuth M., “An introduction to the constitutive modelling of unsaturated soils”, European
Journal of Civil Engineering, vol. 9, no 5-6 (2005) p. 651-670.
Laloui L., Nuth M., “Numerical modelling of the behaviour of a heat exchanger pile”. European
Journal of Civil Engineering, vol. 9, no 5-6 (2005) p. 827-839.
Book chapter
Laloui L., Nuth M., Vulliet L. “Experimental and numerical investigations of the behaviour of a
heat exchanger pile”. In: Ground Improvement - Case Histories. Elsevier (2005).
2/3
Conference proceedings
Nuth M. ,Un nouveau cadre constitutif couplé pour la modélisation avancée des sols non saturés:
XXVèmes Rencontres AUGC (2008) XXVèmes rencontres AUGC, Nancy, June 4-6, 2008.
Nuth M., Laloui L., “New insight into the unified hydro-mechanical constitutive modeling of
unsaturated soils”. IWUS08 Trento (2008), in press.
Nuth M., Laloui L., “Advanced hydro-mechanical coupling for unified constitutive modelling of
unsaturated soils”. In 1st European Conference on Unsaturated Soils. CRC Press (2008).
Laloui L., François B., Nuth M., Péron H., and Koliji A., “A thermo-hydro-mechanical stress-strain
framework for modeling the performance of clay barriers in deep geological repositories for
radioactive waste”. In 1st European Conference on Unsaturated Soils. CRC Press (2008).
François B., Nuth M., Laloui L., “Mechanical constitutive framework for thermal effects on
unsaturated soils”. In Numog X (2007), p. 9-13.
Nuth M., Laloui L., “New insight into the unified hydro-mechanical constitutive modeling of
unsaturated soils”. Invited lecture In: Proc. of the 3 rd Asian Conference on Unsaturated Soils.
(2007), p. 109-126.
Nuth M., Laloui L., “Implications of a generalized effective stress on the constitutive modelling of
unsaturated soils”. In: Mechanics of unsaturated soils, Weimar, Germany (2007), p 75-82.
3/3

Nuth - 2009 - Constitutive Modelling of Unsaturated Soils With Hydro-Geomechanical Couplings

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Nuth - 2009 - Constitutive Modelling of Unsaturated Soils With Hydro-Geomechanical Couplings

Transféré par

Droits d'auteur :

Formats disponibles

Constitutive Modelling of Unsaturated Soils

with Hydro-Geomechanical Couplings

THÈSE NO 4358 (2009)

ÉCOLE POLYTECHNIQUE FÉDÉRALE DE LAUSANNE

POUR L'OBTENTION DU GRADE DE DOCTEUR ÈS SCIENCES

acceptée sur proposition du jury:

Prof. I. Botsis, président du jury

Table of Contents ...............................................................................................................I

3.3 Conceptual principle of effective stress .................................................................... 58

4. Stress-strain model for unsaturated soils ........................................................... 95

4.6.4 Swelling pressure .................................................................................................................. 120

5. A model for the soil water retention curve ....................................................... 125

6. Double-way coupling in ACMEG-s .................................................................. 151

6.4 Validation of the fully coupled model .................................................................... 171

7. Application: numerical modelling of earthdams .................................................. 199

8.4.2 Hydraulic boundary conditions ............................................................................................. 237

Keywords: Unsaturated soils, constitutive modelling, effective stress, capillarity,

Mots-clefs: sols non saturés, modélisation constitutive, contrainte effective, capillarité,

fiso isotropic yield limit

f dev deviatoric yield limit

g dev plastic potential of the deviatoric mechanism

giso plastic potential of the isotropic mechanism

G shear elastic modulus

H matrix of hardening moduli

kw intrinsic water permeability

kr relative water permeability

kwsat satured water permeability

K bulk elastic modulus

K0 coefficient of earth pressure at rest

M slope of the critical state line

ne non-linear elasticity exponent

pnet mean net stress

p0′ initial mean effective stress

pc′ preconsolidation pressure

pc′ 0 initial preconsolidation pressure

pa pore air pressure

pv pore water vapour pressure

pw pore water pressure

riso degree of mobilization of the isotropic mechanism

sI wetting yield suction

se reference air-entry suction

se H updated air-entry suction

Sr ,w degree of saturation in water

S res residual degree of saturation

Vα volume of the constituent α

W work input per unit volume

β0 plastic compressibility modulus at zero suction

γd dry unit weight

γs parameter for LC curve

λisop plastic multiplier of the isotropic mechanism

ε total strain tensor

εe elastic strain tensor

εp plastic strain tensor

ε ve volumetric elastic strain

ε de deviatoric elastic strain

ε vp volumetric plastic strain

ε dp deviatoric plastic strain

Ω parameter for evolution of compressibility

φ′ friction angle at critical state

ρa bulk dry density

ρd bulk density of gas

ρg bulk density of solid

ρw bulk density of water

σ* = Dep ( ε, σ, dσ, ξ ) dε (2.20)