ice manual of geotechnical engineering vol 1: geotechnical engineering principles, problematic soils...

Partial Saturation inCompacted Soils

Geotechnique Symposium in Print 2011

Edited by

Domenico GallipoliChair, Geotechnique Symposium in Print 2011 sub-committee

Copyright © ICE Publishing, all rights reserved.

Geotechnique Advisory Panel Sub-Committee for the Symposium in Print 2011:Chair:Professor Domenico Gallipoli, Universite de Pau et des Pays de l’Adour, France

Members:Professor Eduardo Alonso, Universitat Politecnica de Catalunya, SpainDr Lennart Borgesson, Clay Technology AB, SwedenProfessor Federica Cotecchia, Politecnico di Bari, ItalyProfessor Pierre Delage, Ecole Nationale des Ponts et Chaussees, FranceProfessor Cristina Jommi, Politecnico di Milano, ItalyProfessor Claudio Mancuso, Universita degli Studi di Napoli Federico II, ItalyDr John McDougall, Napier University, UKDr Andrew Ridley, Geotechnical Observations, UKProfessor Tom Schanz, Ruhr-Universtat Bochum, GermanyProfessor Alessandro Tarantino, University of Strathclyde, UKProfessor David Toll, Durham University, UKProfessor Simon Wheeler, University of Glasgow, UK

Related titles from ICE Publishing:

Geotechnical Engineering Principles, Problematic Soils and Site Investigation.J. Burland, T. Chapman, H. Skinner and M.J. Brown (eds). ISBN 978-0-7277-5707-4.

UK Specification for Ground Investigation, Second edition (Site Investigation inConstruction Series). Site Investigation Steering Group. ISBN 978-0-7277-3506-5

Rock Engineering. A. Palmstrom and H. Stille. ISBN 978-0-7277-4083-0

Offshore Geotechnical Engineering: Principles and Practice. E.T.R. Dean.ISBN 978-0-7277-3641-3

ISBN 978-0-7277-5775-3

# Thomas Telford Limited 2013Papers extracted from Geotechnique # Authors and Institution of Civil Engineers

All rights, including translation, reserved. Except as permitted by the Copyright,Designs and Patents Act 1988, no part of this publication may be reproduced, storedin a retrieval system or transmitted in any form or by any means, electronic,mechanical, photocopying or otherwise, without the prior written permission of thePublishing Director, ICE Publishing, 1 Great George Street, London SW1P 3AA.

This book is published on the understanding that the authors are solely responsible forthe statements made and the opinions expressed in it and that its publication does notnecessarily imply that such statements and/or opinions are or reflect the views oropinions of the publishers. While every effort has been made to ensure that thestatements made and the opinions expressed in this publication provide a safe andaccurate guide, no liability or responsibility can be accepted in this respect by theauthors or publishers.

Typeset by Keytec Typesetting Ltd, Bridport DorsetPrinted and bound in Great Britain by CPI Group (UK) Ltd, Croydon, CR0 4YY


PrefaceThis book contains the proceedings of the GeotechniqueSymposium-in-Print 2011, which was held on the theme ofPartial Saturation in Compacted Soils at the Institution ofCivil Engineers on 20 June 2011. The Symposium attractedaround 70 delegates, from both industry and academia,representing countries such as Australia, Czech Republic,France, Hong Kong, Italy, New Zealand, Portugal, Spain,Switzerland and UK.

The book contains the two keynote addresses delivered byProf. Eduardo Alonso and Mr Tony O’Brien, respectively,together with the nine papers and one technical note pub-lished in the April and May 2011 issues of Geotechniqueand presented by authors during the Symposium. These ninepapers and one technical note were selected, followingstandard Geotechnique peer-review, from a total of 98articles offered in response to a thematic call in the summer2009. They are grouped in this book under four topicscorresponding to the four Symposium sessions, namelyMaterial Characterization, Experimental Observation andModelling, Benchmarking of Techniques and Models, andApplication to Engineering Problems and Case Studies. Thebook also contains a selection of questions posed by dele-gates after presentations during the Symposium, togetherwith relative answers by presenters.

The idea of devoting the 16th Geotechnique Symposium-in-Print to the theme of Partial Saturation in CompactedSoils was instigated by recent advances in the study ofgeomaterials with multiphase or immiscible pore fluids. Oneapplication of this study relates to the design and analysis ofearth structures/fills (e.g. dams and embankments, clay bar-riers) made of compacted soils that are unsaturated at thetime of placement (i.e. pores are partly filled by water andpartly filled by air). Unlike saturated soils whose pores areentirely filled by water, the presence of two immiscible porefluids gives rise in unsaturated soils to capillary actions onthe solid skeleton that affect both deformation and strength.During service life, the compacted soil will alternate be-tween unsaturated and saturated conditions, or remain per-manently unsaturated, depending on prevailing environmentalactions.

Some key engineering properties of compacted soils (e.g.strength, stiffness and permeability) depend on currentmoisture content, which will change as a consequence of theinteraction with the surrounding environment. It is thereforenot surprising that, since the early 30s, civil engineers havedevoted considerable effort to the study of unsaturatedcompacted soils. Ralph Proctor, from the Los AngelesBureau of Waterworks and Supplies in the USA, performeda wide laboratory campaign of compaction tests on morethan 200 soils, driven by the necessity of developing suitableconstruction protocols to maximise stability and minimizepermeability of earth dams. This resulted in the definition ofthe ‘optimum moisture content’, i.e. the water content thatproduces the highest soil dry density for a given compactioneffort.

Today, routine geotechnical design still relies on Proctor’sdefinition of optimum water content. Nevertheless, modernexperimental techniques, such as Environmental Scanning

Electron Microscopy or Mercury Intrusion Porosimetry, haveprovided new insight into the link between stress-strainbehaviour and material fabric in compacted soils. Some ofthe contributions in this book emphasize the ‘living nature’of soil fabric, which evolves during wetting–drying cyclesand induces corresponding changes of macroscopic mechani-cal properties over time.

Significant improvements of design practice have oftenoriginated from the formulation of general constitutive lawsand principles of soil behaviour, underpinned by an under-standing of material properties at the microscopic scale.During recent years, engineers and scientists have advancedfundamental knowledge of mechanical and retention behav-iour of compacted soils; however, this research momentummust be sustained over time to achieve the detailed under-standing of soil behaviour which is essential to produce astep change in geotechnical models and a leap in analyticalcapabilities.

As highlighted by some of the contributions in this book,there are still significant challenges ahead for unsaturatedsoil modellers. As already known from saturated soil mech-anics, the accurate prediction of irreversible volumetricstrains during shearing is a particularly arduous task forconstitutive modellers and well-known saturated models,such as Modified Cam-Clay, significantly overestimate volu-metric strains at critical state. This assumes even greatersignificance in unsaturated soils where water retention be-haviour is intrinsically linked to volumetric straining. In thiscase, any error in the calculation of volumetric strains willhave an impact on the computed water retention response,i.e. on the predicted variation of degree of saturation and onits effect on strength and deformation.

Moreover, the sensitivity of predictions to parameter cali-bration poses an additional challenge to engineers wishing toapply unsaturated models to practical problems. The adop-tion of different strategies in selecting parameter values,even for the same model and from the same set of experi-mental data, can result in different computations of soilbehaviour, partly because of the large number of parametersinvolved. This raises doubts about the usefulness of develop-ing ever more sophisticated constitutive laws without propos-ing, at the same time, robust methods for parametercalibration.

The book also includes results from two benchmark exer-cises undertaken by several European Universities within the‘‘Marie Curie’’ Research Training Network MUSE (Mech-anics of Unsaturated Soils for Engineering). A salient activ-ity of the MUSE Network has been the comparison ofexperimental/modelling techniques used by researchersacross the world. This activity had the twofold objective ofimproving current procedures and, where possible, formulat-ing accepted standards. In this spirit, these two benchmarksmake use of easily accessible data or commercially availablematerials so that they can be readily repeated by otherresearchers.

I would like to thank all members of the Symposium sub-committee, namely Eduardo Alonso, Lennart Borgesson,Federica Cotecchia, Pierre Delage, Cristina Jommi, Claudio


Mancuso, John McDougall, Andrew Ridley, Tom Schanz,Alessandro Tarantino, David Toll and Simon Wheeler, fortheir contribution in organizing the Symposium and review-ing manuscripts within a very tight timescale. I would alsolike to thank the Geotechnique Advisory Panel, and inparticular the then Chairman, Prof. Chris Clayton, for sup-porting the proposal of a Geotechnique Symposium-in-Printon the theme of Partial Saturation in Compacted Soils andfor their continued help throughout the organization.

This book provides a comprehensive overview of recentadvances in the fast growing area of unsaturated soil mech-

anics, ranging from material testing to modelling and analy-sis of engineering boundary value problems. This knowledgewill contribute to improve design of earth structures/fills bymaximizing the use of locally sourced soils, with consequentgains in safety, cost and sustainability of future buildingpractice. I wish you enjoyable reading and hope that thisvolume will become a valuable reference for engineers andresearchers in future years.

Domenico GallipoliChair, Geotechnique Symposium-in-Print 2011

iv PREFACE


Contents iiiPreface

Keynote speeches

3Compacted soil behaviour: initial state, structure and constitutive modelling

E.E. Alonso, N.M. Pinyol and A. Gens

19The assessment of old railway embankments - time for a change?

A.S. O’Brien

Session 1: Material Characterisation

Papers

35Effects of the maximum soil aggregates size and cyclic wetting-drying on the stiffness of a lime-

treated clayey soil

A.M. Tang, M.N. Vu and Y.-J. Cui

Technical Note

45Some aspects of the behaviour of compacted soils along wetting paths

S. Taibi, J.M. Fleureau, N. Abou-Bekr, M.I. Zerhouni, A. Benchouk, K. Lachgueur and H. Souli

Session 2: Experimental Observation and Modelling

Papers

55An insight into the water retention properties of compacted clayey soils

E. Romero, G. Della Vecchia and C. Jommi

71Hydromechanical behaviour of compacted granular expansive mixtures: experimental and

constitutive study

E.E. Alonso, E. Romero and C. Hoffmann

87Experimental observations of the stress regime in unsaturated compacted clay when laterally

confined

J.L. Boyd and V. Sivakumar

Session 3: Benchmarking of Techniques and Models

Papers

109Benchmark of constitutive models for unsaturated soils

F. D’Onza, D. Gallipoli, S. Wheeler, F. Casini, J. Vaunat, N. Khalili, L. Laloui, C. Mancuso,

D. Masın, M. Nuth, J.-M. Pereira and R. Vassallo

129Benchmark of experimental techniques for measuring and controlling suction

A. Tarantino, D. Gallipoli, C.E. Augarde, V. De Gennaro, R. Gomez, L. Laloui, C. Mancuso,

G. El Mountassir, J.J. Munoz, J.-M. Pereira, H. Peron, G. Pisoni, E. Romero, A. Raveendiraraj,

J.C. Rojas, D.G. Toll, S. Tombolato and S. Wheeler

Session 4: Application to Engineering Problems and Case Studies

Papers

141Hydromechanical behaviour of a heterogeneous compacted soil: experimental observations and

modelling

A. Gens, B. Vallejan, M. Sanchez, C. Imbert, M.V. Villar and M. Van Geet

161Modelling the response of Lechago earth and rockfill dam

E.E. Alonso, S. Olivella, A. Soriano, N.M. Pinyol and F. Esteban

183Physical modelling of wetting-induced collapse in embankment base

L. Thorel, V. Ferber, B. Caicedo and I.M. Khokhar

Selected Questions and Answers 195


Keynote Speeches


Alonso, E. E. et al. Geotechnique [http://dx.doi.org/10.1680/geot.11.P.134]

3

Compacted soil behaviour: initial state, structure and constitutivemodelling

E. E. ALONSO�, N. M. PINYOL�† and A. GENS�

The paper explores the behaviour of compacted soils throughout the (dry density–water content)compaction plane by means of a conceptual framework that incorporates microstructural information.The engineering properties of compacted soils are described by an initial state in terms of a yieldingstress, soil suction and a microstructural state variable. Microstructure is defined by the ratio ofmicrovoid volume to total void volume. The pattern of variation of the microstructural parameterwithin the compaction plane has been determined, for some compacted soils, by analysing mercuryintrusion porosimetry data. The microstructure of wet and dry compaction conditions can then bequantified. To ensure consistency, the framework is cast in the form of a constitutive model defined interms of an effective suction and a constitutive stress that incorporate the microstructural variable.The model is shown to be consistent with a number of experimental observations and, in particular, itexplains the intrinsic collapse potential of compacted soils. It predicts, for a common initial suction, ahigher collapse potential for dry of optimum conditions than for wet compaction. It also predicts in anatural manner the observed evolution of soil compressibility during drained or undrained loading.Model capabilities are illustrated by application to a testing programme on statically compactedsamples of low-plasticity silty clay. The compression behaviour of samples compacted wet and dry ofoptimum and the variation of collapse strains with confining stress have been successfully reproducedby the model.

KEYWORDS: clays; compaction; constitutive relations; fabric/structure of soils; partial saturation; plasticity;suction

INTRODUCTIONA significant proportion of the published research on unsatu-rated soil mechanics concerns compacted soils. It could beinferred that examining the current state of development ofunsaturated soil research would provide detailed informationon compacted soil behaviour. This is only partially true,however. Often, basic research is conducted in silty mater-ials, statically compacted at a low density. These soilsexhibit an open structure, sensitive to suction-inducedeffects. Compacted soils in practice span a much widerrange of grain-size distributions. Proctor (1933) was able toshow the fundamental relationship between attained densityand water content for a given compaction energy. Thisfinding defined the compaction plane (in terms of dry unitweight, ªd, against water content, w) which is a veryconvenient procedure to represent the compaction states of agiven soil. This plane remains the basic representation forinvestigating the properties of compacted soils.

This is also the starting point of the study reported in thispaper. Instead of focusing on a given initial state of acompacted soil, the main objective is to explore the compac-tion plane. This approach will hopefully provide a widerperspective on soil compaction. The properties of compactedsoils (permeability, stiffness, strength) were linked to thecompaction state in some classical contributions published in

the 1950s and 1960s (Leonards, 1955; Lambe, 1958; Seed &Chan, 1959; Lambe & Whitman, 1969). This idea is recov-ered here, but the focus now is to incorporate advances inunsaturated soil mechanics reported in the last two decades.Significant recent contributions involve the search for appro-priate constitutive stress formulations, the development ofelasto-plastic frameworks, and the increasing recognition ofthe role played by soil microstructure.

Microstructure, in particular, was very early identified as akey feature in any explanation of compacted soil behaviourassociating dispersed microstructure with compaction on thewet side (wetter than optimum) and flocculated microstruc-ture with compaction on the dry side (drier than optimum)(Lambe, 1958; Lambe & Whitman, 1969). However, directobservations of soil fabric by means of scanning electronmicroscopy, and the interpretation of mercury intrusionporosimetry tests reported from the 1970s (Sridharan et al.,1971; McGown & Collins, 1975), led to significant changesin this initial microstructural interpretation. It was observed,for instance, that clay tended to form aggregated structuresthat behaved as much larger particles, especially whencompacted on the dry side. It was soon accepted that waterwas trapped inside the clay aggregations, even if the mixtureremained relatively dry. These ideas have been widely con-firmed by subsequent studies (Delage et al., 1996; Romero& Simms, 2008; Lee & Zhang, 2009; Monroy et al., 2010).

The debate on effective stress has been a recurrent topicin unsaturated soil mechanics research since the early intro-duction by Bishop (1959) of an effective stress equation.This topic will be discussed further below. The unavoidablefact is, however, that modern constitutive laws for unsatu-rated soils that attempt to provide a comprehensive descrip-tion of soil behaviour (and not just of a specific property)have been always formulated in terms of two ‘stress states’or ‘constitutive stresses’ (Jommi, 2000; Gens, 2010). One of

Manuscript received 4 November 2011; revised manuscript accepted3 September 2012.Discussion on this paper is welcomed by the editor.� Department of Geotechnical Engineering and Geosciences,Universitat Politecnica de Catalunya, Barcelona, Spain.† International Centre for Numerical Methods in Engineering(CIMNE), Barcelona, Spain.


them is generally a function of soil suction, often the soilsuction itself.

The development of elasto-plastic constitutive modelsprovides an alternative way to characterise the initial state ofcompacted soils by associating model parameters and vari-ables with the pair dry unit weight and water content (ªd,w), which defines the ‘as-compacted’ condition. For instance,the dry density achieved by compaction can be related to theposition of the initial yield surface after compaction. Watercontent, on the other hand, is controlled mainly by thecurrent suction, s, and to a lesser extent by the void ratio. Inthe context of the simple elasto-plastic BBM model (Alonsoet al., 1990), the yield surface is essentially defined by theisotropic yield stress for saturated conditions, p�0: Therefore,as a starting point, the pair ( p�0, s) may provide equivalentinformation to (ªd, w), with one added advantage: theysupply fundamental information for constitutive modelling.

An analysis of a limited number of soil compactiontesting programmes led to the p�0 –ªd relationship given inFig. 1 (Alonso & Pinyol, 2008). It can be noted that thesaturated isotropic yield stress increases rapidly with dryunit weight. Also, for a given dry unit weight, the yieldstress p�0 increases significantly with soil plasticity. The plotin Fig. 1 may help to select p�0 in the absence of specifictests. On the other hand, many s(w, ªd) relationships can befound in the literature (e.g. Gens et al., 1995; Li, 1995;Romero et al., 1999; Tarantino & Tombolato, 2005).

Thus the pair ( p�0, s)as-compacted provides key informationconcerning the compacted state of a given soil, but it doesnot include any information on its microstructure. A reviewof microstructural effects on the compacted soil behaviour,given below, indicates that microstructure is also a signifi-cant aspect that should be introduced in a realistic modellingof compacted soils. The generalisation of techniques (parti-cularly mercury intrusion porosimetry, MIP) to examine theevolution of soil microstructure has provided information

regarding the effects of microstructure on the engineeringbehaviour of compacted soils. The incorporation of sucheffects is a key aspect of the work presented here.

In summary, the goal of this paper is to present a unifiedbut general picture of compacted soil behaviour incorporat-ing recent developments in unsaturated soil mechanics andmicrostructural considerations.

The paper is organised as follows. First, a summary ofsome relevant experimental work devoted to isolate theeffect of microstructure on compacted soil behaviour ispresented. Then a constitutive stress expression that incorpo-rates explicitly soil microstructure is introduced. This isdone through the definition of a single microstructural statevariable. The interpretation of testing programmes providingdata on pore size distribution has allowed this microstructur-al state variable to be mapped onto the compaction plane,and general trends of behaviour to be established. Compres-sibility and its relationship with suction and microstructureare then discussed, because this leads to a consistent descrip-tion of collapse behaviour, one of the key aspects ofunsaturated soil behaviour. To ensure consistency, the de-scription of behaviour is cast in the form of a new constitu-tive framework that incorporates the developments andconcepts described previously. Its predictive capabilities arechecked against experiments.

MICROSTRUCTURE AND COMPACTED SOILBEHAVIOUR

In this paper, information on microstructure is derivedfrom MIP tests performed both after compaction and afterthe application of a given stress–suction path.

Consider, in Fig. 2, the pore size density function of asample of low-plasticity Barcelona silty clay (wL ¼ 28%,IP ¼ 9.3%), statically compacted at a high void ratio(e ¼ 0.82) and then isotropically loaded in a triaxial cell to asubstantially lower void ratio (e ¼ 0.57) (Buenfil et al.,2004). The observation of two dominant pore sizes duringcompaction, especially dry of optimum, is a characteristicfeature, widely observed. These two dominant pore sizeswill be referred to as microporosity and macroporosity.Volumetric deformation upon isotropic loading results in areduction of the size and volume of the macropores. Incontrast, the micropores retain their partial volume and theirsize. Clay aggregates are readily observed in the micropho-tographs for e ¼ 0.82 and, to a lesser extent, for e ¼ 0.57,where it can be observed that the size of the macropores hasclearly reduced.

A second example of the effect of loading and suctionchanges is given in Fig. 3 for Boom clay initially compactedto e ¼ 0.93 and Sr ¼ 0.44 (Romero et al., 2011). The speci-men was wetted at constant volume (swelling pressure path)and then dried at constant vertical stress in an oedometercell. The stress–suction path applied is sketched in Fig.3(a). Pore size distribution tests were performed at points 1,2 and 3, and the results are given in Fig. 3(b). The firstloading–wetting path results in a significant reduction in thesize of the macropores. Further drying reduced the volumeof the macropores, but the microporosity seems to remainlargely unchanged throughout.

Figure 4 shows the change in pore size distribution of asample of Barcelona silty clay, statically compacted on thedry side (sample DD), when it is wetted under a smallconfining stress and taken to the position DW close tosaturation conditions. Some changes are observed, but in thiscase the ‘as compacted’ pore size distribution seems to belargely preserved after wetting.

This is consistent with the results of Thom et al. (2007)testing statically compacted kaolin, dry of optimum, which

Medium-plasticity soil(Honda ., 2003)et al

Medium-low-plasticity soil(Honda ., 2003)et al

Morainic soil(Cuisinier & Laloui, 2004)

Low-plasticity silty clayfrom Barcelona(Barrera, 2002)

Non-plastic silty sand(Balmaceda, 1991)

Measured value ofSan Salvador siltyclay

0

0·2

0·4

0·6

0·8

1·0

1·2

1·4

1·6

1·8

2·0

2·2

1·0 1·2 1·4 1·6 1·8 2·0

Yie

ld s

tres

s,*

: MP

ap

0

Dry unit weight, : g/cmγd3

High-plasticity Boom clay(Romero, 1999)

wL 43%IP 13·4%

�

�

wL 30%IP 12%

�

�

wL 33·5%IP 13·2%

�

�

wL 56%IP 27%

�

�

wL 30·5%IP 11·8%

�

� wL 28%IP 8%

�

�

NP

Fig. 1. Relationship between isotropic yield stress at saturatedconditions and dry density of several soil types. From Alonso &Pinyol (2008). # 2008 Taylor & Francis Group, London, UK.Used with permission

4 ALONSO, PINYOL AND GENS


indicated that the bimodal domain of voids induced by soilcompaction is essentially maintained upon the application ofsignificant stress and suction changes. However, this is notalways the case. Wetting tests reported by Monroy et al.(2010) on statically compacted high-plasticity London Clayindicate that changes in micropores may also be significanton wetting. The higher activity of the clay probably explainsthe enhanced sensitivity of microporosity to suction changes.

The variation of engineering properties of a given soilwhen compacted at different dry densities and water contentshas been often reported (Cox, 1978; Resendiz, 1980; Lawtonet al., 1989, 1991; Alonso et al., 1992; Benson et al., 1992;Fredlund & Rahardjo, 1993; Tinjum et al., 1997; Vanapalliet al., 1999; Simms & Yanful, 2002; Santucci de Magistris& Tatsuoka, 2004; Jotisankasa et al., 2007, 2009). However,it is not feasible to isolate microstructural effects in many ofthese contributions, mainly because compacting dry or wetof optimum implies not only a different microstructure, butalso a different suction. In addition, compacting at differentvoid ratios implies both a variation in macroporosity and achange in the initial yield locus. Hence conventional testingof compacted samples mixes the effects of the initial state( p�0, s) and of the microstructure. Therefore specificallydesigned testing programmes are required to isolate micro-structural effects.

An interesting example of such programmes is providedby Santucci de Magistris & Tatsuoka (2004), who testeddynamically compacted specimens of low-plasticity siltysand (a residual granitic soil) in a triaxial cell. Oncecompacted, all samples were taken to a saturated state beforetesting. Because it is a low-plasticity soil, it was expected

(b)

(c)

0

0·4

0·8

10 100 1000 10000 100000

Por

e si

ze d

ensi

ty fu

nctio

n,/

(log

)�

δδ

eD

Entrance pore size, : nmD

455 nm

e 0·57�

e 0·82�

19 mμ

60 mμ

(a)

Fig. 2. Evolution of microstructure during loading. (a) Pore sizedistribution. (b), (c) ESEM observations, statically compactedlow-plasticity Barcelona silty clay: (b) e 0.82; (c) e 0.57. AfterBuenfil et al. (2004)

As-compacted0·41ew �

Swellingpressure

path

Drying

s

2 σ

1 3

0

0·4

0·8

10 100 1000 10000 100000

Por

e si

ze d

ensi

ty fu

nctio

n,/

(lo

g)

�δ

δe

D

Entrance pore size, : nm(b)

D

Saturation atconstant volume

Dried aftersaturation

(a)

1

2

3

Fig. 3. Evolution of pore size distribution of compacted high-plasticity Boom clay during loading and suction changes: (a)stress path; (b) pore size distribution at three stress–suctionpoints indicated in (a). From Romero et al. (2011)

COMPACTED SOIL BEHAVIOUR: INITIAL STATE, STRUCTURE AND CONSTITUTIVE MODELLING 5


that the as-compacted microstructure would be essentiallypreserved and, naturally, suction at the testing stage waszero. They found that the slope of the virgin compressionline changes moderately with moulding water content fromwet to dry conditions, except for states in the vicinity of themodified Proctor optimum. Similar conclusions were reachedconcerning small-strain stiffness. Drained strength (the limit-ing q/p9 ratio) was not much affected by the attained drydensity, although dry compaction resulted in a moderateincrease in dilatancy when compared with wet compaction.

Wheeler & Sivakumar (1995, 2000), testing staticallycompacted speswhite kaolin, found that the slope and inter-cept of normal compression lines for constant suctiondepend on the compaction water content and compactionstress. A similar result was found for the volumetric critical-state line.

Microstructural effects may be identified directly if speci-mens having a different origin (say, compacted dry and wetof optimum) are tested at a common state. Instead ofselecting a saturated state, Suriol et al. (1998) and Suriol &Lloret (2007) tested samples of statically compactedBarcelona silty clay (wL ¼ 30.5% and IP ¼ 11.8%) in asuction-controlled oedometer cell. The approach adopted isillustrated in Fig. 5(a). Under a vertical compaction stress of

0.6 MPa and a moulding water content w ¼ 14%, the soilequilibrates at point D. A wetter sample (w ¼ 22%) underthe same compaction stress lies on point W. Afterwards,samples W were dried under suction-controlled conditions

1·45

1·55

1·65

1·75

1·85

1·95

8 12 16 20 24

Dry

den

sity

: g/c

m3

Water content: %(a)

Sr 1�

DD

DW

0

0·1

0·2

0·3

0·4

0·5

0·6

ΔΔ

eD

/lo

g

Void size, : nm(b)

D

DD

DW

1 100 10000 1000000

Fig. 4. Statically compacted Barcelona silty clay: (a) compactionstate of samples DD and DW; (b) pore size distributions. AfterSuriol et al. (1998) and Suriol & Lloret (2007)

1·45

1·55

1·65

1·75

1·85

1·95

8 12 16 20 24

Dry

den

sity

: g/c

m3

Water content: %(a)

DD

WD

Sr 1�

0

0·1

0·2

0·3

0·4

0·5

0·6

1 100 10000 1000000

ΔΔ

eD

/lo

g

Void size, : nm(b)

D

5·0

3·0

1·0

�1·0

0 1·0 2·0

Vertical stress: MPa

Ver

tical

str

ain:

%

(c)

WD

WD

DD

DD

DD

D

W

Fig. 5. Statically compacted Barcelona silty clay: (a) compactionstate of samples DD and WD; (b) pore size distributions;(c) measured collapse strains in loading at constant suction andfull wetting. After Suriol et al. (1998) and Suriol & Lloret (2007)



and taken to position D (they are samples WD, which shouldbe read as ‘compacted in position W and tested in positionD’). In contrast, samples DD are both compacted and testedin D. The pore size distributions of samples DD and WDare given in Fig. 5(b). Samples DD exhibit a marked doubleporosity. Samples WD have developed a dominant inter-mediate pore size.

Samples DD and WD were then loaded in a suction-con-trolled oedometer maintaining the suction prevailing at point D(1 MPa) and then saturating the specimens under constantvertical stress. The measured vertical strains (collapse strains)are plotted in Fig. 5(c) in terms of applied vertical stress. DDspecimens collapse more than WD samples, a clear indicationof the effects of microstructure, since the dry density, watercontent and suction are the same in the two cases.

The effect of microstructure on elastic stiffness andstrength will be discussed further once the constitutive stressis defined. However, a final property is discussed here:permeability. As this parameter is essentially controlled bythe pore structure of the soil, it provides a good indicationof microstructural changes if the permeability of samplescompacted at different (ªd, w) values is measured undersaturated conditions. Again, this is especially true if micro-structural changes during the saturation process are minor.Fig. 6 reproduces results published by Mitchell et al. (1965).The soil, a silty clay, was compacted by kneading action.Contours of equal saturated permeability are plotted in thecompaction plane. Clearly, permeability is not uniquelycontrolled by void ratio. In fact, a strong variation is ob-served when, for a given void ratio, the compaction watercontent increases. Thus permeability reveals microstructuraleffects much better than mechanical tests do. The plot inFig. 6 indicates that designs involving compacted soils inwhich permeability is an important issue (e.g. isolatingbarriers) cannot be based only on classical criteria limitingdry density and water content on the basis of mechanicaltests. Adding microstructure in some manner to the constitu-tive behaviour of compacted soils would help to incorporatethe hydraulic properties of compacted soils in a natural way.

These points can be summarised as follows.

(a) It is clear that the microstructure of compacted soils has adistinct effect on their geotechnical properties that is notaccounted for solely by the stress pair ( p�0, s).

(b) Not all the properties are equally sensitive to micro-structure. One extreme case is probably the soil per-meability. Conversely, the drained strength parameters donot seem to be much affected.

(c) Compressibility and, therefore, collapse and swellingpotential are controlled to a significant extent bymicrostructure.

(d ) Stress and suction paths applied to compacted specimensmodify mainly the macroporosity, but in active soils(high-plasticity clays) changes associated with suctionvariations may also modify the microstructural voidvolume.

EFFECTIVE DEGREE OF SATURATION ANDCONSTITUTIVE STRESS

A full description of the pore structure of a soil wouldrequire a large number of parameters, which would preventthe incorporation of such information into a simple constitu-tive formulation. Here, the pore size distribution is describedby a simple state variable: the ratio of the microstructuralvoid ratio, em, and the total void ratio, e. This ratio will beknown here as the microstructural state variable, �m:

�m ¼em

e(1)

If water has access to an initially compacted dry soil, it isexpected that the microstructure will saturate first, becauseof the strong affinity for water of the clay platelets. Oncethe microvoids inside clay aggregates are saturated, anyexcess water will occupy the macropores. Alonso et al.(2010) suggested that only the water partially filling themacropores will have a significant mechanical effect on thesoil. The concept behind this idea is that capillary effectswill be exhibited only by the water forming menisci betweenaggregates and inert soil particles.

The ‘effective’ degree of saturation, in the sense ofcontributing to the constitutive stress, will then take non-zero values only for degrees of saturation in excess of �m:The effective degree of saturation was assumed to varybetween 0 and 1, when Sr spans the macropore space(�m < Sr < 1). This is illustrated in Fig. 7 for the particularvalue �m ¼ 0.4. Sketches showing the assumed distributionof water in the micro and macro volumes illustrates the

90

100

110

120

10 15 20 25

Dry

uni

t wei

ght:

lb/ft

3

Moulding water content: %

ModifiedProctorcurve10�7

10�6

10�5Acceptable zone based ontypical current practice:

0·9 andγ γd d,max�

10�8

w w0–4% wet of� opt

Fig. 6. Contours of permeability, under saturated conditions, forsamples of silty clay compacted by kneading action (100 lb/ft3

approx. 15.7 kg/m3). From Mitchell et al. (1965), with permis-sion from ASCE

0

0·2

0·4

0·6

0·8

1·0

0 0·2 0·4 0·6 0·8 1·0

Sr

Sr

Sr �Sr m� �

1 � �m

Sr m, �

�m

Fig. 7. Definition of effective degree of saturation



location of the water as the degree of saturation increases.The effective degree of saturation is given by the equations

�SSr ¼Sr � �m

1� �m

for Sr . �m (2a)

�SSr ¼ 0 for Sr < �m (2b)

Equation (2a) for the effective degree of saturation wasalready introduced by Romero & Vaunat (2000) and Taranti-no & Tombolato (2005).

It is proposed that the unsaturated soil behaviour bedefined in terms of two independent stress fields.

Constitutive stress:

�� ¼ � � pg þ �SSrs (3a)

Effective suction:

�ss ¼ �SSrs (3b)

where the effective suction and the constitutive stress havebeen made dependent on the void ratio, microstructural voidratio and degree of saturation by means of the effectivedegree of saturation. In equation (3a), � is the total stressand pg is the gas pressure.

Partial aspects (elastic stiffness, drained failure envelopes)of unsaturated soil behaviour were shown to be well pre-dicted by interpreting results in terms of equation (3a)(Alonso et al., 2010).

The piecewise equation (2) was smoothed in Alonso et al.(2010) by a continuous power law (�SSr ¼ (Sr)

Æ; Æ . 1),which provides some advantages in calculation. However,the power function loses a direct reference to the microstruc-tural state variable, �m, which is a variable capable of beingexperimentally measured in MIP tests, as described in thenext section.

Following the procedure put forward by Gesto et al.(2011), an alternative smoothing of �SSr around the cornerSr ¼ �m that maintains �m as a material state variable can beachieved by the equation

�SSr ¼Sr � �m

1� �m

þ 1

nln 1þ exp �nsmooth

Sr � �m

1� �m

� �� (4)

The number nsmooth defines the degree of smoothing aroundthe corner (Fig. 8). It can be noted that equation (4) providesa small effective stress contribution for Sr , �m that is likelyto be more realistic than the sharp transition shown in Fig. 7.

MAPPING MICROSTRUCTURE ON THE COMPACTIONPLANE

The microstructural state variable �m was shown byAlonso et al. (2010) to increase with soil plasticity. However,no indication of the effect of varying compaction variableswas given. A recently conducted experimental programmehas explored, in a systematic manner, the microstructure ofcompacted Boom clay (Merchan, 2011). The attained drydensities covered a wide range (1.40�1.85 g/cm3), spanningthe values typical of practical engineering applications. Fig.9 shows the position of samples tested in the compactionplane. MIP tests in which mercury was first intruded andthen extruded after reaching a maximum pressure of227 MPa were conducted on all samples.

There are alternative procedures available to determine themacropore and micropore volumes. If only the intrusioncurve is analysed, the density distribution of pore sizes maybe interpreted in a straightforward manner by calculating theareas corresponding to the two porosity levels. The densityfunction provides the information to establish the macro-

pore–micropore size boundary. However, this procedurebecomes difficult to apply if the distinct bimodal shape islost.

If the extrusion cumulative curve is available, Delage &Lefebvre (1984) suggested an alternative procedure to distin-guish between porosity levels, as sketched in Fig. 10. Micro-porosity was judged to result in a reversible (elastic) volumeof mercury intruded and extruded from the small voids. Incontrast, mercury would be trapped in the macropores bycapillary effects once the applied pressure is removed. Thisconcept and the resulting procedure to determine the microand macro porosities are adopted here for two reasons. First,it results in a non-ambiguous procedure to identify the twoporosities. Second, the idea of a reversible behaviour of thevolume change induced by pore pressures changes in themicroporosity (also identified as the intra-aggregate space) isconsistent with the behaviour of interacting clay platelets.

0

0·2

0·4

0·6

0·8

1·0

0 0·2 0·4 0·6 0·8 1·0Sr

nsmooth 3�

5

10

Sr

Fig. 8. Effect of parameter nsmooth on effective degree of satura-tion for �m 0.3

1·10

1·30

1·50

1·70

1·90

2·10

0 5 10 15 20 25 30 35 40

Dry

spe

cific

wei

ght,

/γ

γd

w

Water content, : %w

M1

M2

M3

M4

M5

M6

M7

M8

M9

M10

M11

M14

Sr 1�

Fig. 9. Compacted samples of Boom clay tested by Merchan(2011)



This issue was further discussed by Gens & Alonso (1992).In fact, the double structure model proposed there to de-scribe the behaviour of expansive clays highlights the rever-sible behaviour of the microstructure of clayey soils.

Using the Delage & Lefebvre (1984) procedure to identifymacroporosity and microporosity, the microstructural statevariable was determined for all the samples shown in Fig. 9.Fig. 11 shows the variation of the microstructural void ratioin the compaction plane. Interpolated contours show that em

is essentially controlled by the compaction water content.Derived values of the state variable �m are indicated in

Fig. 12 for all the compacted specimens tested. Contours ofequal �m are also plotted. The data show an increase of �m

with density that reflects the progressive reduction in voidratio. Equally relevant is the effect of compaction watercontent. For a given void ratio, �m increases with compac-tion water content, reflecting the fact that samples com-pacted wet develop a larger quantity of micropores.

To illustrate the effect of compaction conditions, consider,for instance, two specimens essentially compacted to thesame dry density and different water contents: M9 (a drierspecimen) and M3 (a wetter specimen). Their microstructur-al states (�m ¼ 0.4 for M9 and �m ¼ 0.6 for M3) and theimplication in terms of effective degree of saturation are

shown in Fig. 13. If both samples are partially filled withwater to a common degree of saturation (say Sr ¼ 0.8), theirdifferent microstructures will result in specimen M9 (drier)exhibiting a higher effective degree of saturation than M3.Therefore, under a common suction, M9 will experience ahigher constitutive stress than specimen M3.

Other recently published data on MIP intrusion–extrusionmeasurements of compacted specimens support the generaltrend of �m variation observed in Fig. 12. This is the casepresented in Fig. 14, which collates data reported by Romeroet al. (2011) on statically compacted Boom clay. Tarantino &de Col (2008) published MIP data on a few statically com-pacted samples of kaolin. The calculation of �m has also beenbased on the intrusion–extrusion diagram of MIP tests (Fig.15). The figure shows the compaction curves for increasingvertical compaction stress, the contours of equal degree ofsaturation, the calculated values of �m for the few samplestested, and the contours for equal �m: The three plots (Figs12, 14 and 15), which correspond to two different soils andthree independent testing programmes, show similar trends.

In order to demonstrate that these conceptual ideas can beintegrated in a consistent manner, a constitutive model ofcompacted soil behaviour has been developed that includesmicrostructural information in terms of the state variable �m:

e0

e

Extrusion

Intrusion

eM

em

D

Constrainedporosity

Macro

Micro

Reversible

Fig. 10. Scheme of intrusion–extrusion stages of an MIP test and interpretation in terms ofmicrovoid (em) and macrovoid (eM) ratios. After a proposal by Delage & Lefebvre (1984)

0·197

1·20

1·40

1·60

1·80

2·00

0 5 10 15 20 25 30 35 40

Dry

spe

cific

wei

ght,

/γ

γd

w

Water content, : %w

M1

M2

M3

M4

M5

M6

M7M8

M9

M10

M11

M14

Sr 1�

Boom clayStatically compacted

samples

0·28

0·325

0·234

0·328

0·296

0·367

0·311

0·335

0·437

0·328

0·35

0·40

0·35

0·30

0·25

0·35 : Microstructural void ratio, em

Fig. 11. Contours of equal microstructural void ratio, em, forcompacted Boom clay. Compaction data taken from Merchan(2011)

0·41

1·10

1·30

1·50

1·70

1·90

2·10

0 5 10 15 20 25 30 35 40

Dry

spe

cific

wei

ght,

/γ

γd

w

Water content, : %w

M1

M2

M3

M4

M5M6

M7

M8

M9

M10

M11

M14

Sr 1�

0·61

0·57

0·33

0·607

0·493

0·56

0·63

0·47

0·47

0·390·377

0·40

0·30

0·61 : Microstructural state variable, �m

0·50

0·60

Fig. 12. Contours of equal microstructural parameter, �m, forcompacted Boom clay. Compaction data taken from Merchan(2011)



The first step is an analysis of compressibility, described inthe next section.

A CONSTITUTIVE MODEL INCORPORATINGMICROSTRUCTURAL EFFECTS

The constitutive stress defined in equation (3a) is capableof predicting strength and elastic stiffness in unsaturatedstates if these properties are known for a saturated state(Alonso et al., 2010). This partial success is, however, farfrom implying that the constitutive stress thus defined canpredict consistently the volumetric behaviour of wet anddry-compacted samples against stress and suction changes.Specifically, thinking in terms of different compaction condi-tions, the challenge is to investigate whether the volumetricbehaviour of compacted samples can be predicted by consid-ering only the initial stress state (say, p�0 and s) and amicrostructural state variable such as �m: The differences incollapse behaviour shown in Fig. 5, which are basicallycontrolled by microstructure, provide an important bench-mark to check the above hypothesis.

A convenient starting point towards this goal is to reviewthe compressibility of samples compacted dry and wet ofoptimum, and to interpret them in terms of effective suction.In fact, compressibility has already been shown to dependon prevailing suction (Alonso et al., 1987, 1990; Wheeler &Sivakumar, 1995). If microstructure affects compressibilityas well, the relationship between a compression coefficientand suction will also depend on the compaction watercontent (for a common initial void ratio). However, as theeffective suction defined above incorporates microstructuraleffects, it is possible that a unique relationship betweencompressibility and effective suction may emerge.

Model formulation for isotropic stress statesThe model is defined in terms of two variables: the

constitutive stress defined in equation (3a) and the effectivesuction defined in equation (3b). Changes in the constitutivestress induce elastic and elasto-plastic strains respectively,according to the logarithmic relationships

dee ¼ ��kkd�pp

�pp(6)

�m 0·4� �m 0·64�

0

0·2

0·4

0·6

0·8

1·0

0 0·2 0·4 0·6 0·8 1·0Sr

Drie

r (M

9)

Wet

ter

(M3)

Sr

M9 M30

0·2

0·4

e

�m �0·20·5

� 0·4 �m �0·320·5

� 0·64

Fig. 13. Microvoid and macrovoid ratios of specimens M9 and M3 after compaction and associatedeffective degree of saturation for a degree of saturation of 0.8

1·30

1·40

1·50

1·60

1·70

10 15 20 25 30 35

Dry

spe

cific

wei

ght,

/γ

γd

w

Water content, : %w

0·4 0·40

0·65

0·60

e em/ 0·6�

0·80

Sr 1�

0·8

Fig. 14. Contours of equal microstructural parameter, �m, forcompacted Boom clay. Compaction data taken from Romero et al.(2011)

0·96

1·12

1·28

1·44

0 8 16 24 32 40

Dry

uni

t wei

ght,

: kN

/mγ d

3

Water content, : %w

0·486 : Experimentally determined values�m

σv 1200 kPa�

σv 900 kPa�

σv 600 kPa�

σv 300 kPa�

0·486

0·85

0·38

0·43

0·38

�m 0·4�

�m 0·5�

�m 0·6

�Sr 1�

Sr 0·9�

MIPtest

0·7

0·50·30·2

Fig. 15. Contours of equal microstructural parameter, �m, forcompacted kaolin. Compaction data taken from Tarantino & deCol (2008)



deep ¼ ��ººd�pp

�pp �ss¼constant

(7)

which introduce the definition of compressibility coefficientsagainst changes in constitutive stress. The elastic index, �kk, isassumed to be constant in this paper.

It is accepted, in view of previous results, that �ºº is afunction of the effective suction, given by

�ºº �ssð Þº 0ð Þ

¼ f �ssð Þ ¼ f �SSrsð Þ (8)

where �SSr is defined in equation (4). Compression linesassumed by this model are plotted in Fig. 16 in terms of theconstitutive stress, for different effective suctions.

The parameter �ºº is conceptually very different from the‘standard’ compressibility coefficient, º, defined in terms ofnet stress. In order to estimate �ºº, the variation of void ratiowith constitutive stress at constant effective suction isrequired. Suction-controlled tests do not provide such infor-mation directly, even if the microstructural void ratio isknown and assumed to be constant, because the void ratio ischanging during loading, and the effective suction is therebyvarying continuously. During loading e reduces and�m ¼ em/e increases. The effect of this increase is illustratedin Fig. 17. If the degree of saturation is maintained duringloading (this is approximately the case of constant-suctionloading), the effective degree of saturation, and therefore theeffective suction will decrease (Fig. 17(a)). As a result, thecoefficient �ºº will increase during loading. The effect of thiscontinuous decrease of �SSrs during loading is illustrated inFig. 18: the actual compression curve ‘travels’ during load-ing across the compression lines plotted at constant effectivesuction (�SSrs) values. This is also the case for undrainedloading (constant water content), for a double reason: thereduction in void ratio and the associated decrease insuction. The net result is that, during undrained loading, thevoid ratio decreases faster in terms of constitutive stress. Ifthe degree of saturation also changes with void ratio (in asuction-controlled test, for instance), an additional change inthe effective degree of saturation and the constitutive stressis introduced. Therefore, �ºº cannot be readily estimated from‘standard’ laboratory compression curves relating void ratioand net stress.

Despite these difficulties, the definition of elasto-plasticcompressibility in equation (7) leads to a powerful andsimple model to predict the isotropic compression of unsatu-rated soils for a wide range of applied stresses. It reconcilesexisting formulations based either on a decrease (as inBBM; Alonso et al., 1990) or on an increase (Wheeler &Sivakumar, 1995) of compressibility with suction, and pre-dicts in a natural way the evolution of collapse strains with

suction, which exhibits a maximum for some intermediateconfining stress. This is discussed again below.

Further analysis requires the proposal of a particularrelationship for the function f (�ss) in equation (8). A diffi-culty in making �ºº dependent on effective suction is that thecompressibility would change when the soil was saturatedand subjected to suctions lower than the air-entry value. Toavoid this shortcoming, the following equation is proposedfor �ºº in terms of the effective suction.

�ºº �ssð Þº 0ð Þ

¼ �rr þ 1� �rrð Þ 1� �ss

�ssº

� �1= 1��ð Þ" #��

(9)

where �rr, �� and �ssº are material parameters. Equation (9) is

λ3

λ2

λ1

Voi

d ra

tio, e

Saturatedcompression

lineλ λ(0) (0)�

Constitutive mean stress, log p

Fig. 16. Definition of compression lines in terms of constitutivestress log �pp ¼ log (pnet þ �SSrs):

1

00 Sr

1

(a)

em

e 1

em

e 2

Sr

⎛⎜⎝ ⎛

⎜⎝ ⎛

⎜⎝ ⎛

⎜⎝

sSr

λ(0)λ3

λ2

λ1

λ

(b)

Fig. 17. (a) Effect of increasing microstructural void ratio oneffective degree of saturation for constant degree of saturation;(b) effect of decreasing effective suction on compressibilitycoefficient, �ºº

Voi

d ra

tio, e Saturated

compressionline

λ λ(0) (0)�

Loading atconstant suction

Undrained loading( constant)w �

λ1

λ2λ3

Constitutive mean stress, log p

Fig. 18. Estimated compression behaviour of unsaturated soil fordrained (constant suction) and undrained (constant watercontent) isotropic loading in terms of constitutive stresslog �pp ¼ log (pnet þ �SSrs):



plotted in Fig. 19 for different values of �rr, �� and �ssº:Parameter �� controls the rate of variation of �ºº(�ss)=º(0) witheffective suction between two limits (1 and �rr). Parameter �rrdefines the maximum stiffness of the soil as effective suctionincreases (�rrº(0)). Suction �ssº defines the air entry value. Infact, it can be taken as being the equivalent parameter of thevan Genuchten (1980) expression for the water retentionrelationship. Note in Fig. 19 that �ºº is essentially constant

when the effective suction is smaller than �ssº: Within thisrange of suctions, effective suction and suction are essen-tially equal.

The constitutive yield isotropic stress for unsaturatedconditions (�pp0) is defined by the loading collapse (LC) yieldcurve

�pp0

�ppc

¼ p�0�ppc

� �º 0ð Þ�k�ºº �ssð Þ�k

(10)

where �ppc is a reference mean constitutive stress correspond-ing to the point where virgin compression lines for differenteffective suctions cross. Since the preconsolidation stressdepends on the effective suction, samples compacted at thesame void ratio but having different microstructural param-eters will exhibit different LC loci. Equation (9) is a uniqueexpression for the entire compaction plane. Differences incompressibility will be the result of differences in currentvalues of suction and the microstructural state variable, �m:

In order to provide an insight into the performance of thecompression model defined, a synthetic example is pre-sented. Several isotropic compression tests, run at differentvalues of suction, which remain constant during loading, aresimulated. Model parameters are listed in Table 1. Thegenerally observed dependence of water retention on voidratio (e) has been introduced in the model in a simple wayby expressing the parameter P0 of the van Genuchten model

0

0·2

0·4

0·6

0·8

1·0

0 0·1 0·2 0·3 0·4 0·5s: MPa

( 0·3 MPa; 0·8; 0·8)s rλ � � ��

(0·1 MPa; 0·8; 0·6)

(0·3 MPa; 0·4; 0·8)

(0·1 MPa; 0·4; 0·8)

λλ

()/

(0)

s

Fig. 19. Variation of isotropic compressibility with effectivesuction for different parameter values

Table 1. Parameters and initial conditions for isotropic synthetic tests and oedometer tests on Barcelona silty clay samples

Parameter Definition Units Synthetic case Barcelona silty clay

Sample dry Sample wet Sample DD Sample WD

Mechanical parameters

�kk Elastic compressibility – 0.006 0.006�ºº(0) ¼ ºð0Þ Plastic compressibility at zero effective

suction– 0.07 0.07

� Poisson’s ratio – 0.3 0.3M Slope of critical-state strength line – – 1.2�ppc Reference stress MPa 0.001 0.001�ssº Minimum value of effective suction

affecting plastic volumetriccompressibility. Equivalent to an airentry value (P0)

MPa 0.25 0.25

�rr Parameter that establishes minimumvalue of compressibility coefficientfor high values of effective suction

– 0.8 0.8

�� Parameter that controls rate of increasein stiffness with effective suction

MPa–1 0.75 0.8

nsmooth Parameter that defines degree ofsmoothing in equation (7)

– 3 3

Water retention curve: van Genuchten model

Sr min Sr � Sr min

Sr max � Sr min

¼ 1þ s

P0

� � 11�º

" #�º

P0 ¼ aVGe + bVG

– 0 0Sr max – 1 1aVG MPa �0.25 �0.25bVG MPa 0.4 0.4º – 0.35 0.35

Initial conditions

e0 Initial void ratio – 0.538 0.538p�0 Yield stress at zero effective suction MPa 0.45 0.45s0 Initial suction MPa 1 1

�m0 Initial microstructural state variable – 0.34 0.38 0.36 0.41



as a linear function of e (Table 1). The model is exploredfor two ‘samples’ of the same soil compacted on the dryside (sample SI-D) or wet side (sample SI-W). It can benoted that the same set of parameters is used for the twospecimens, the only difference being the value of the initialmicrostructural state variable, �m0:

Model calculations are shown in terms of two sets ofstress variables: constitutive stress and effective suction inFig. 20, and net stress and suction in Fig. 21. The plots ofvolumetric strain against constitutive stress (Fig. 20) showthe change in soil compressibility with applied stress, dis-cussed above. Points corresponding to a particular value ofeffective suction may be identified on each compressioncurve, and they are indicated in the figures. Straight linesconnecting these points define the compressibility coefficient�ºº of the underlying model.

The compression curves in Fig. 20 may also be repre-sented in terms of volumetric strain against net stress andsuction (Fig. 21). This plot shows the compression lines thatwould be recorded directly in conventional isotropic tests atconstant suction. The model predicts that those ‘standard’compression lines of an unsaturated soil will first respond ina stiff manner, compared with the saturated compressionline;. but as deformation (and suction changes) accumulates,the compressibility index will eventually become similar toº(0), and then it will reach higher values in order todecrease again later when the saturated compression line isapproached asymptotically. In loose soils, the first stage is

often not observed, and the measured apparent compressionindex for a given suction exceeds the saturated value (thiswas the case presented by Wheeler & Sivakumar (1995)). Indense soils, the entire transition from º , º(0) to º . º(0) isobserved, provided the applied stress is large enough. Thistype of result was shown, for instance, by Romero (2002),testing compacted Boom clay in a suction-controlled oed-ometer; by Jotisankasa et al. (2007), testing a compactedmixture of silt, kaolin and London Clay; and by Benatti etal. (2011), testing a colluvial, collapsible silty clay.

To move from Fig. 21 to Fig. 20 requires knowledge ofthe degree of saturation during loading, and information onthe initial microstructural state variable in order to identifythe variation of �ºº with effective suction through the com-pression lines. Two or more tests, providing the measurementof void ratio, suction and degree of saturation during com-pression, are required to obtain this variation, and to deter-mine the parameters in equation (9).

The shape of the compression lines in Figs 20 and 21indicates that the model is capable of simulating collapsestrains that first increase with confining stress and thenreduce, as frequently observed in collapse tests. Sometimesthe final reduction in collapse is not observed, because theapplied stress is not high enough; in this case, collapseseems to be an increasing function of applied stress. This isthe prediction of BBM. This issue is also discussed bySheng (2011) and Zhou et al. (2012a, 2012b) in the contextof a model in which the plastic compressibility index

s 0·5� MPa

s 0·4� MPa

0·16

0·16

0·12

0·12

0·08

0·08

0·04

0·04

0

0

�0·02

�0·02

0·1

0·1

1

1

10

10

Constitutive mean stress: MPa

Constitutive mean stress: MPa

Vol

umet

ric s

trai

nV

olum

etric

str

ain

(a)

(b)

Saturatedloading

Saturatedloading

1 MPa

1 MPa

3 MPa

3 MPa

s 5 MPa�

s 5 MPa�

2 MPa

2 MPa

0·5 MPa

0·5 MPa

0·3 MPa

0·3 MPa

Saturation

Saturation

0·25 MPa

0·27 MPa

0·35 MPa

0·32 MPa

0·4 MPa

0·22 MPa0·18 MPa

Fig. 20. Synthetic isotropic tests in terms of constitutive stress. Compression responseat different constant suctions: (a) dry compaction, sample SI-D; (b) wet compaction,sample SI-W



depends on the degree of saturation. Their model alsopredicts collapse strains first increasing and then decreasingwith applied confining stress.

Comparison of Figs 21(a) and 21(b) (or Figs 22(a) and22(b)) finally provides the effect of soil microstructure intro-duced by dry and wet compaction. Dry compaction, charac-terised by a lower initial microstructural state variable (�m0),results in a stiffer soil response than wet compaction. Inaddition, collapse strains, for any confining stress, are higherfor dry compaction. This result was also measured in thereal tests that are described below.

Simulation of tests on compacted Barcelona silty clayConsider again the DD and WD tests series on compacted

Barcelona clay reported by Suriol & Lloret (2007) (Figs 4and 5). The microstructure of samples WD and DD wasdiscussed on the basis of the intrusion curve only. The poresize density plot reproduced in Fig. 5(b) was in fact thederivative of the accumulated mercury volume during thepressure-increase phase of the test. Although it was notreported, the authors also recorded an extrusion (pressuredecrease) branch. This is shown in Fig. 22 for the DD andWD samples. The available extrusion curves, performed onseveral samples compacted to the same initial state to checkthe repeatability of results, did not reach a zero rate ofvolume change, but they were close. The plots allow calcu-lation of the microstructural void ratio, as suggested in Fig.10. The following as-compacted (initial) values are derivedfor em0 and �m0: sample DD, em0 ¼ 0.19, �m0 ¼ 0.36; sampleWD, em0 ¼ 0.22, �m0 ¼ 0.41. These values are consistentwith the trends shown in Figs 12, 14 and 15: wet of

optimum compaction results in a higher microstructural voidratio.

Collapse tests represented in Fig. 5(c) for the WD andDD samples were limited to vertical stresses not exceeding2 MPa. However, additional oedometer tests for samples WDand DD were performed later in a high-capacity loadingframe, and the vertical confining stress reached a maximumvalue of 8 MPa. The measured vertical strains for the full0–8 MPa stress range are shown in Fig. 23. Collapse reachesa maximum in both types of sample for a vertical stress ofabout 2 MPa. At higher vertical stresses, collapse decreasesgradually. The distance between the two curves also reducesas the vertical stress increases.

The model is now used to explore in more detail itsability to reproduce the observed effects of compacting dry

0·16

0·12

0·08

0·04

0

�0·02 0·1 1 10Net mean stress: MPa

Vol

umet

ric s

trai

n

(a)

Saturatedloading

Saturatedloading

1 MPa

1 MPa

0·5 MPa

0·5 MPa

0·3 MPa

0·3 MPa

s = 5 MPa

s = 5 MPa

3 MPa

3 MPa

2 MPa

2 MPa

Saturation

Saturation

0·16

0·12

0·08

0·04

0

�0·02 0·1 1 10Net mean stress: MPa

Vol

umet

ric s

tain

(b)

Fig. 21. Synthetic isotropic tests in terms of net stress. Compres-sion response at different constant suctions: (a) dry compaction,sample SI-D; (b) wet compaction, sample SI-W

0

0·1

0·2

0·3

0·4

0·5

0·6

100 102 104 106

Acc

umul

ate

d vo

id r

atio

Equivalent pore size diameter: mm(a)

DD–1

DD–2HOM

DD–9

e0 0·538�

0·34

em 0·194�

� DDm0 0·36�

� WDm0 0·41�

0

0·1

0·2

0·3

0·4

0·5

0·6

100 102 104 106

Acc

umul

ate

d vo

id r

atio

Equivalent pore size diameter: mm(b)

e0 0·538�

0·318

WD–10

WD–11

em 0·22�

Fig. 22. MIP tests on compacted samples of Barcelona silty clay:(a) DD samples; (b) WD samples

5·0

3·0

1·0

�1·0

0 2·0 4·0 6·0 8·0 10·0Net vertical stress: MPa

Ver

tical

str

ain:

%

WD

DD

Fig. 23. Measured collapse strains of samples WD and DD ofcompacted Barcelona silty clay in range 0–8 MPa of confiningtotal vertical stress



and wet of optimum (set of samples WD and DD, in Figs 5and 23). Data are available for constant-suction oedometerloading starting on the common (dry density–water content)state (WD ¼ DD), which corresponds to an initial suctions ¼ 1 MPa.

Dry and wet of optimum compaction are distinguishedexclusively by the initial microstructural state variable, �m0:All the remaining model parameters for the two samplesanalysed take the same value (Table 1). The slopes of thevirgin compression lines for saturated conditions (º(0)) andthe elastic coefficient (�kk) were approximated from testresults. Values of �rr and ��, which define compressibility forunsaturated conditions, have been estimated to fit the col-lapse variation with stress level. The value for parameter �ssºcorresponds to an average value of parameter P0 of the vanGenuchten water retention curves (see below). A low refer-ence stress �ppc was also selected.

The model proposed requires information on the waterretention properties for the calculation of �SSrs: Water reten-tion curves of Barcelona silty clay for several void ratioswere reported by Barrera (2002). Fig. 24 shows the experi-mental data. These curves have been adopted to estimate thewater retention behaviour of samples WD and DD by meansof a van Genuchten (1980) model. The dependence of thewater retention behaviour on void ratio has been introducedin the model in the same simple way as used for thesynthetic cases presented in the previous section, that is,assuming a linear variation of the parameter P0 (whichcontrols the air entry value) with void ratio. The modelapproximation to the reported retention curves is also plottedin Fig. 24. The van Genuchten expression and the values ofthe parameters chosen for calculations are indicated in Table1. The effect of the microstructure on the water retentioncurve has been neglected, and a unique relationship betweensuction, degree of saturation and void ratio has been as-sumed for both samples.

Suriol et al. (1998) reported the compression curves ofDD and WD samples for s ¼ 1 MPa and for saturatedconditions to a maximum vertical stress of 1.50 MPa only(Fig. 25). However, the reported collapse strains at differ-ent confining stresses in the range 0–8 MPa (Fig. 23)allow an estimation of the compression curves of DD andWD samples for the entire stress range. They are shown inFig. 26.

The constitutive model was first used to reproduce thesaturated compression line in the range 0–1.5 MPa (loadingand unloading) (Fig. 25). The comparison between modelcalculations and experiments is quite satisfactory. The calcu-lated saturated compression line was then extended to theentire loading range for the DD and WD compaction condi-tions (Fig. 26). The model reproduces satisfactorily theexperimental results, and predicts correctly the differences incompression behaviour associated with different microstruc-ture. It is recalled that only a single variable, namely theinitial microstructural state (�m0), distinguishes the twocompaction conditions; the rest of the material parametersare identical in both samples. It is remarkable that thecomplexities of the microstructure can be adequately repre-sented by a single state variable.

A further comparison of model performance and experi-mental data is given in Fig. 27, which shows the collapsestrains of both series of tests (DD and WD) plotted in termsof net applied stress. The model correctly predicts the maxi-mum of collapse, its progressive decay towards a zero valueat high applied stresses, and the higher intrinsic collapsibilityof the sample compacted dry of optimum. All these effectsare linked directly with the initial microstructure of the soil

0·01

0·1

1

10

0 0·2 0·4 0·6 0·8 1·0

Suc

tion:

MP

a

Degree of saturation

e 0·53�

0·55

0·64

0·75

0·87

Experimental datae 0·53�

e 0·55�

e 0·64�

e 0·75�

e 0·87�

Fig. 24. Water retention data on compacted Barcelona silty clayreported by Barrera (2002) and model (van Genuchten) approx-imations for different values of void ratio (e)

0·06

0·04

0·02

0

�0·01 Net vertical stress: MPa

Ver

tical

str

ain:

%

(a)

Experimental

Model

Saturation

s 0�

s 1 MPa�

0·06

0·04

0·02

0

�0·01 Net vertical stress: MPa

Ver

tical

str

ain:

%

(b)

Experimental

Model

Saturation

s 0�

s 1 MPa�

1

1

Fig. 25. Oedometric suction-controlled (s 1 MPa; s 0 MPa)compression curves of compacted Barcelona silty clay and modelpredictions: (a) DD samples; (b) WD samples. Results of a fullwetting collapse test at a vertical stress of 1.5 MPa are alsoindicated



and its evolution during the application of a stress–suctionpath.

The initial LC yield locus of DD and WD samples isplotted in Fig. 28(a) in term of constitutive stress and effec-tive suction, and in Fig. 28(b) in term of net stress andsuction. In the first case a unique LC yield curve describesdry and wet of optimum conditions. When transformed tonet stress–suction coordinates, two different initial LCcurves describe dry and wet conditions. The shape of theLC yield curves in Fig. 28(b) was already advanced byAlonso et al. (1987) and Gens (1995) when interpretingcompression and wetting tests on samples compacted wetand dry of optimum. The available model at the time (BBM)

required a different set of parameters (namely r and �) todescribe compaction-induced microstructure. The model de-scribed here confirms the assumption made in the papersmentioned, and provides a more compact and comprehensivedescription of compacted soil behaviour.

CONCLUDING REMARKSUnderstanding the behaviour of compacted soils requires

not only information on the as-compacted density and watercontent (or equivalent information in terms of yield stressand suction, as suggested here) but also a proper considera-tion of microstructure. The review of compacted soil behav-iour presented in the first part of this paper has highlightedthe relevance of microstructure. A conceptual frameworkthat incorporates microstructural information and accountsfor the behaviour of compacted soils throughout the compac-tion plane has been put forward.

Microstructure is quantified by a state variable, �m, theratio of microvoid ratio to total void ratio. The microvoidratio tends to maintain its original ‘as compacted’ valueduring subsequent stress–suction paths in low- to medium-plasticity clayey soils. Changes in total void ratio associatedwith loading and suction changes, however, modify themicrostructural state variable, which becomes a state param-eter with direct influence on the constitutive stress andeffective suction.

The interpretation of pore size distribution data of testingprogrammes on compacted soils, covering a reasonablycomplete range of dry densities and water contents, has ledto the establishment of the pattern of �m0 throughout thecompaction plane. This state variable opens the way for asystematic evaluation of microstructural effects on measur-able ‘macroscopic’ engineering variables, such as elasticstiffness, strength, compressibility, yielding behaviour andpermeability. All of them are influenced to a larger (per-meability) or smaller (drained strength) extent by micro-structure.

The framework is the basis of a constitutive model definedin terms of an effective suction and a constitutive stress thatinclude microstructural information. The proposed constitutivestress was shown previously to be useful in predicting theelastic stiffness and strength of unsaturated soils. The presentwork has concentrated on the analysis of compressibility, and

0·16

0·12

0·08

0·04

0·01

�0·02Net vertical stress: MPa

Ver

tical

str

ain:

%

(a)

0·1 1 10

s 1 MPa�

Data taken from collapsestrain measurements

s 0�

0·16

0·12

0·08

0·04

0·01

�0·02Net vertical stress: MPa

Ver

tical

str

ain:

%

(b)

0·1 1 10

s 1 MPa�

Data taken from collapsestrain measurements

s 0�

Model

Model

Fig. 26. Estimated experimental compression curves of com-pacted Barcelona silty clay in range 0–8 MPa of vertical stressand model predictions: (a) samples DD; (b) samples WD

6

4

2

�1

02 4 6 8 10

Net vertical stress: MPa

Ver

tical

str

ain:

%

WD model

DD model

ExperimentsWD

DD

Fig. 27. Measured and calculated collapse strains of samples DDand WD

0

0·2

0·4

0·6

0·8

1·0

0 1 2 3

Eff

ectiv

e su

ctio

n: M

Pa

Constitutive meanpreconsolidation stress: MPa

(a)

WD

DD

0

1

2

3

4

0 1 2 3

Suc

tion:

MP

a

Net mean preconsolidationstress: MPa

(b)

Fig. 28. Initial LC yield curves of samples DD and WD: (a) interms of constitutive stress and effective suction; (b) in terms ofnet stress and suction



on the associated concepts of collapse and yielding of unsat-urated compacted soils. It has been shown that relevantengineering properties of a compacted soil can be explainedby a set of common material (constitutive) parameters, theinitial stress conditions defined by an initial yield stress, p�0,and a suction, s, and an additional state variable, �m, thatdescribes the microstructure, which varies across the compac-tion plane, Including �m in the definition of constitutive stressand effective suction has the following significant advantages.

(a) It is a very simple option, because only one variable isrequired to quantify microstructure.

(b) It captures the intrinsic volumetric behaviour of com-pacted soils and, in particular, the observed differenceswhen compacting dry or wet of optimum.

(c) It provides a realistic interpretation of experimentalcompression lines for a wide range of applied stressesfor both constant-suction and undrained loading condi-tions. The collapse maximum often observed is explainedin a natural way.

(d ) Existing constitutive models can be enhanced by includ-ing �m with a limited effort.

The model proposed requires information on the waterretention properties. A desirable feature of the analysis inthis case would be to include microstructural information,such as �m, in the formulation of the water retention behav-iour, a proposal already put forward by Romero et al.(2011). In this way, the entire coupled hydromechanicalmodel will require only a reduced number of materialparameters. It has been shown that the model is consistentwith experimental observations.

Naturally, the characterisation of complex microstructureby a single variable, and the assumption that the microstruc-tural void ratio remains unchanged, are simplifications thatshould be reviewed when some other aspects of compactedsoil behaviour involving deviatoric loading or anisotropyeffects are considered. In particular, extension of this modelto expansive clays will require a relaxation of the assump-tion of a constant microstructural void ratio. However, thesuccess of this approach in representing the behaviour ofcompacted soils suggests that it can be a very useful plat-form for subsequent developments and generalisations.

NOTATIONaVG parameter that controls the variation of P0 with ebVG parameter that controls the variation of P0 with e

D pore size diametere total void ratio

eM macrostructural void ratioem microstructural void ratioew water ratioee elastic void ratio

eep elastoplastic void ratioe0 initial void ratioIP plasticity indexM slope of critical-state strength line

nsmooth degree of smoothingP0 parameter of van Genuchten modelp9 effective mean stress�pp constitutive mean stress

pg gas pressurepnet net mean stressp�0 mean yield stress�ppc reference mean constitutive stress�pp0 constitutive yield mean stress for unsaturated conditionsq deviatoric stressr parameter that establishes minimum value of

compressibility coefficient for high values of suction in theBarcelona basic model (Alonso et al., 1990)

�rr parameter that establishes minimum value of

compressibility coefficient for high values of effectivesuction

Sr degree of saturation�SSr effective degree of saturations suction�ss effective suction

�ssº minimum value of effective suction affecting plasticvolumetric compressibility

w water contentwL liquid limit

wopt optimum water contentÆ smoothing coefficient for effective degree of saturation in

the paper by Alonso et al. (2010)� parameter that controls rate of increase in stiffness with

suction in the Barcelona basic model (Alonso et al., 1990)�� parameter that controls rate of increase in stiffness with

effective suctionªd dry unit weight

ªd,max maximum dry unit weight for a given compaction energyªw unit weight of water

�kk elastic compressibility indexº plastic compressibility index in net mean stress–void ratio

plane�ºº plastic compressibility index in constitutive mean stress–

void ratio plane� Poisson’s ratio� total stress�� constitutive stress�v net vertical stress�m microstructural state variable (¼ em/e)�m0 initial microstructural state variable

REFERENCESAlonso, E. E. & Pinyol, N. M. (2008). Unsaturated soil mechanics

in earth and rockfill dam engineering. In Unsaturated soils:Advances in geo-engineering (eds D. G. Toll, C. E. Augarde, D.Gallipoli and S. J. Wheeler), vol. 1, pp. 3–32. Boca Raton, FL,USA: CRC Press.

Alonso, E. E., Gens, A. & Hight, D. (1987). Special problem soils:general report. Proc. 9th Eur. Conf. Soil Mech. Found. Engng,Dublin 3, 1087�1146.

Alonso, E. E., Gens, A. & Josa, A. (1990). A constitutive modelfor partially saturated soils. Geotechnique 40, No. 3, 405�430,http://dx.doi.org/10.1680/geot.1990.40.3.405.

Alonso, E. E., Josa, A. & Gens, A. (1992). Modelling the behaviourof compacted soils upon wetting. In Raul J. Marsal Volume, pp.207–223. Mexico City, Mexico: Sociedad Mexicana de Mecani-ca de Suelos (SMMS).

Alonso, E. E., Pereira, J. M., Vaunat, J. & Olivella, S. (2010). Amicrostructurally based effective stress for unsaturated soils.Geotechnique 60, No. 12, 913�925, http://dx.doi.org/10.1680/geot.8.P.002.

Barrera, M. (2002). Estudio experimental del comportamiento hi-dromecanico de suelos colapsados. PhD thesis, Universitat Poli-tecnica de Catalunya, Barcelona, Spain (in Spanish).

Benatti, J. C. B., Miguel, M. G., Rodrigues, R. A. & Vilar, O. M.(2011). Collapsibility study for tropical soil profile using oedo-metric test with controlled suction. In Unsaturated soils (eds E.Alonso and A. Gens), Vol. 1, pp. 193–198. Boca Raton, FL,USA: CRC Press/Balkema.

Benson, C. H., Zhai, H. & Wang, X. (1992). Estimating hydraulicconductivity of compacted clay liners. J. Geotech. Engng ASCE120, No. 2, 366�387.

Bishop, A. W. (1959). The principle of effective stress. TeckniskUkeblad 106, No. 39, 859�863.

Buenfil, C. M., Romero, E., Lloret, A. & Gens, A. (2004). Experi-mental study on the hydromechanical behaviour of a silty clay.Proc. 2nd Int. Workshop on Unsaturated Soils, Anacapri, 15–29.

Cox, D. W. (1978). Volume change of compacted clay fill. Proceed-ings of the Institution of Civil Engineers Conference on ClayFills, London, pp. 79�86.

Delage, P. & Lefebvre, G. (1984). Study of the structure of asensitive Champlain clay and of its evolution during consolida-tion. Can. Geotech. J. 21, No. 1, 21�35.



Delage, P., Audiger, M., Cui, Y. J. & Howat, M. (1996). Microstruc-ture of a compacted silt. Can. Geotech. J. 33, No. 1, 150–158.

Fredlund, D. & Rahardjo, H. (1993). Soil mechanics for unsaturatedsoils. London, UK: Wiley.

Gens, A. (1995). Constitutive modelling. Application to compactedsoils. Proc. 1st Int. Conf. on Unsaturated Soils, Paris 3,1179�1200.

Gens, A. (2010). Soil–environment interaction in geotechnicalengineering. Geotechnique 60, No. 1, 3–74, http://dx.doi.org/10.1680/geot.9.P.109.

Gens, A. & Alonso, E. E. (1992). A framework for the behaviourof unsaturated expansive soils. Can. Geotech. J. 29, No. 6,761�773.

Gens, A., Alonso, E. E., Suriol, J. & Lloret, A. (1995). Effect ofstructure on the volumetric behaviour of a compacted soil. Proc.1st Int. Conf. on Unsaturated Soils, Paris 1, 83�88.

Gesto, J. M, Gens, A. & Vaunat, J. (2011). Smoothing of yieldsurfaces and a reformulation of multi-surface plasticity. Proc.11th Int. Conf. on Computational Plasticity: Fundamentals andApplications (COMPLAS XI), Barcelona, Spain, 295–306.

Jommi, C. (2000). Remarks on constitutive modelling of unsatu-rated soils. In Experimental evidence and theoretical approachesin unsaturated soils (eds A. Tarantino and C. Mancuso), pp.139–153. Rotterdam, the Netherlands: Balkema.

Jotisankasa, A., Ridley, A. & Coop, M. (2007). Collapse behav-ior of compacted silty clay in suction-monitored oedometerapparatus. J. Geotech. Geoenviron. Engng 133, No. 7, 867–877.

Jotisankasa, A., Coop, M. & Ridley, A. (2009). The mechanicalbehaviour of an unsaturated compacted silty clay. Geotechnique59, No. 5, 415–428, http://dx.doi.org/10.1680/geot.2007.00060.

Lambe, T. W. (1958). The engineering behaviour of compacted clay.J. Soil Mech. Found. Div. ASCE 84, No. SM2, 1�35.

Lambe, T. W. & Whitman, R. V. (1969). Soil mechanics. New York,NY, USA: Wiley.

Lawton, E. C., Fragaszy, R. J. & Hardcastle, J. H. (1989). Collapseof compacted clayey sand. J. Geotech. Engng ASCE 115, No. 9,1252�1267.

Lawton, E. C., Fragaszy, R. J. & Hardcastle, J. H. (1991). Stressratio effects on collapse of compacted clayey sand. J. Geotech.Engng ASCE 117, No. 5, 714�730.

Lee, X. & Zhang, L. M. (2009). Characteristics of dual structurepore size distribution of soil. Can. Geotech. J. 46, No. 2, 129–141.

Leonards, G. A. (1955). Strength characteristics of compacted clays.Trans. ASCE 120, 1420�1454.

Li, Z. M. (1995). Compressibility and collapsibility of compactedunsaturated loessial soils. Proc. 1st Int. Conf. on UnsaturatedSoils, Paris 1, 139�144.

McGown, A. & Collins, K. (1975). The microfabric of someexpansive and collapsing soils. Proc. 5th Panamerican Conf.Soil Mech. Found. Engng, Buenos Aires 1, 323�332.

Merchan, V. (2011). Small strain stiffness and residual strength ofunsaturated Boom clay: a micro-structural insight. PhD thesis,Universitat Politecnica de Catalunya, Barcelona, Spain.

Mitchell, J. K., Hooper, D. R. & Campanella, R. G. (1965).Permeability of compacted clay. J. Soil Mech. Found. EngngDiv. ASCE 91, No. 4, 41�65.

Monroy, R., Zdravkovic, L. & Ridley, A. (2010). Evolution ofmicrostructure in compacted London clay during wetting andloading. Geotechnique 60, No. 2, 105�119, http://dx.doi.org/10.1680/geot.8.P.125.

Proctor, R. R. (1933). Fundamental principles of soil compaction.Engng News Rec. 111, No. 9, 148–156.

Resendiz, D. (1980). Compaction conditions, state variables andengineering properties of compacted clay. Proceedings of theinternational conference on compaction, Paris, Vol. 1, pp.195�202.

Romero, E. (2002). Characterization and thermo-hydro-mechanicalbehaviour of unsaturated Boom clay: an experimental study. PhDthesis, Universitat Politecnica de Catalunya, Barcelona, Spain.

Romero, E. & Simms, P. (2008). Microstructure investigations inunsaturated soils. a review with special attention to mercury

intrusion porosimetry and environmental scanning electron mi-croscopy. Geotech. Geol. Engng 26, No. 6, 705–772.

Romero, E. & Vaunat, J. (2000). Retention curves in deformableclays. In Experimental evidence and theoretical approaches inunsaturated soils (eds A. Tarantino and C. Mancuso), pp. 91–106. Rotterdam, the Netherlands: A. A. Balkema.

Romero, E., Gens, A. & Lloret, A. (1999). Water permeability,water retention and microstructure of unsaturated compactedBoom clay. Engng Geol. 54, No. 1–2, 117�127.

Romero, E., Della Vecchia, G. & Jommi, C. (2011). An insight intothe water retention properties of compacted clayey soils. Geo-technique 61, No. 4, 313�328, http://dx.doi.org/10.1680/geot.2011.61.4.313.

Santucci de Magistris, F. & Tatsuoka, F. (2004). Effects of mould-ing water content on the stress–strain behaviour of a compactedsilty sand. Soils Found. 44, No. 2, 85–101.

Seed, H. B. & Chan, C. K. (1959). Structure and strength character-istics of compacted clays. J. Soil Mech. Found. Div. ASCE 85,No. SM5, 87�128.

Sheng, D. (2011). Constitutive modelling of unsaturated soils: dis-cussion of fundamental principles. Proc. 5th Int. Conf. onUnsaturated Soils, Barcelona 1, 91–112.

Simms, P. H. & Yanful, E. K. (2002). Predicting soil-water char-acteristic curves of compacted plastic soils from measured pore-size distributions. Geotechnique 52, No. 4, 269–278, http://dx.doi.org/10.1680/geot.2002.52.4.269.

Sridharan, A., Altschaeffl, A. G. & Diamond, S. (1971). Pore sizedistribution studies. J. Soil Mech. Found. Div. 97, No. SM5,771�787.

Suriol, J. & Lloret, A. (2007). Cambios en la estructura de sueloscompactados frente a humedecimiento y secado. IngenierıaCivil, Madrid 147, 7–76 (in Spanish).

Suriol, J., Gens, A. & Alonso, E.E. (1998). Behaviour of compactedsoils in suction controlled oedometer. Proc. 2nd Int. Conf. onUnsaturated Soils, Beijing 1, 463�443.

Tarantino, A. & de Col, E. (2008). Compaction behaviour of clay.Geotechnique 58, No. 3, 199�214, http://dx.doi.org/10.1680/geot.2008.58.3.199.

Tarantino, A. & Tombolato, S. (2005). Coupling of hydraulic andmechanical behaviour in unsaturated compacted clay. Geotech-nique 55, No. 4, 307�317, http://dx.doi.org/10.1680/geot.2005.55.4.307.

Thom, R., Sivakumar, R., Sivakumar, V., Murray, E. J. & Mack-innon, P. (2007). Pore size distribution of unsaturated compactedkaolin: the initial states and final states following saturation.Geotechnique 57, No. 5, 469�474, http://dx.doi.org/10.1680/geot.2007.57.5.469.

Tinjum, J. M., Benson, C. H. & Blotz, L. R. (1997). Soil-watercharacteristic curves for compacted clays. J. Geotech. Geoenvir-on. Engng ASCE 123, No. 11, 1060�1069.

van Genuchten, M. T. (1980). Closed-form equation for predictingthe hydraulic conductivity of unsaturated soils. Soil Sci. Soc.Am. J. 44, No. 5, 892–898.

Vanapalli, S. K., Fredlund, D. G. & Rufahl, D. E. (1999). Theinfluence of soil structure and stress history on the soil-watercharacteristics of a compacted fill. Geotechnique 49, No. 2,143�159, http://dx.doi.org/10.1680/geot.1999.49.2.143.

Wheeler, S. J. & Sivakumar, V. (1995). An elasto-plastic criticalstate framework for unsaturated soils. Geotechnique 45, No. 1,35�53, http://dx.doi.org/10.1680/geot.1995.45.1.35.

Wheeler, S. J. & Sivakumar, V. (2000). Influence of compactionprocedure on the mechanical behaviour of an unsaturated com-pacted clay. Part 2: Shearing and constitutive modelling. Geo-technique 50, No. 4, 369�376, http://dx.doi.org/10.1680/geot.2000.50.4.369.

Zhou, A.-N., Sheng, D., Sloan, S. W. & Gens, A. (2012a). Inter-pretation of unsaturated soil behaviour in the stress saturationspace. II: Constitutive relationships and validations. Comput.Geotech. 43, 111–123

Zhou, A.-N., Sheng, D., Sloan, S. W. & Gens, A. (2012b). Inter-pretation of unsaturated soil behaviour in the stress-saturationspace, I: Volume change and water retention behaviour. Comput.Geotech. 43, 178–187.



O’Brien, A. S. (2013). Geotechnique, 19–32 [http://dx.doi.org/10.1680/geot.sip11.ks]

19

The assessment of old railway embankments – time for a change?

A. S. O’BRIEN�

This paper summarises the results of several studies that have been carried out during the last 10 yearson the behaviour of old railway embankments, composed of uncompacted, ‘dumped’ clay fill. Fieldobservations from instrumented embankments, have shown that deformation and delayed failure aresignificantly affected by seasonal climate variations and by the type and spatial distribution ofvegetation on the slope. Pore-water pressures (which largely control deformation and the risk offailure) vary markedly between summer and winter, due to the influence of climate and vegetation,and exhibit complex variations across an embankment. Often the pore-water pressures are negative,and the fill can be unsaturated. The dumped clay fills have a clod–matrix structure which is quitedifferent to modern compacted fills or to natural clay. Laboratory studies indicate that this clod–matrix structure has an important effect on the shear strength, compressibility and permeability of thedumped clay fill. Numerical modelling of progressive failure and of climate–vegetation–pore-waterpressure interactions provides insights into the factors which control behaviour. These studies showthat the ‘steady-state’ assumptions typically assumed for embankment assessments are unreliable.

KEYWORDS: clays; field instrumentation; slopes; pore-water pressures; deformation; permeability

INTRODUCTIONIn the UK, practising geotechnical engineers usually assumethat soils are fully saturated. In particular, when assessingthe stability of earthworks slopes, it is usually assumed thatpore-water pressures increase hydrostatically with depth be-low a zero pressure line which is close to the slope surface.A significant amount (perhaps the majority) of the slopeengineering currently carried out in the UK is associatedwith the assessment of old infrastructure earthworks, ratherthan the design and construction of new earthworks. Infra-structure owners in the UK have been implementing pro-active asset management to improve the conditions of theirearthworks. This has resulted in a growing awareness of thedegradation of embankments as they get older. The taskfaced by Network Rail and London Underground is particu-larly daunting due to the old age (typically more than100 years old) of their infrastructure embankments and toincreasingly intensive use. CIRIA C592 (Perry et al., 2003)outlines the requirements for cost-effective asset manage-ment, which are very different to conventional design, as theprocess is mainly one of comparing relative risks across thenetwork; i.e. does embankment A need more urgent attentionthan embankment B. Table 1 provides a summary of somekey questions for asset management engineers. In this con-text, conventional slope stability analyses are often of littlevalue, as they indicate that the vast majority of old railwayembankments do not comply with modern code require-

ments. Hence, the assumption of fully saturated soil behav-iour does not provide a useful risk assessment tool. Aparticular challenge is to understand the reasons for appar-ently similar clay-fill slopes (with similar height/slope angle,and clays of similar intrinsic characteristics: clay fraction,liquid limit, etc.) behaving very differently. Many embank-ments, composed of high-plasticity clay fill, have been stableat slope angles of 268 to 308 (compared with a typicalcritical state friction angle of 208 to 258) for over a hundredyears, while others move excessively during dry summers,and others are prone to delayed failure (despite havingsurvived many historic wet winters). The intent of this paperis to summarise several applied research studies which havebeen carried out during the last 10 years, and highlight themain practical implications. First, network-wide surveys ofembankment deformation and failure will be described,followed by a series of field and laboratory studies atspecific sites. Numerical modelling of deformation and fail-ure, and modelling of climate–vegetation–slope pore pres-sure interactions will be discussed. Based on these studies,changes to current practice are suggested, and some futureresearch needs are outlined.

NETWORK-WIDE SURVEYSFigure 1 summarises the results of a study on train delays

due to geotechnical causes, carried out by Mott MacDonald

Table 1. Rail embankment asset management – current challenges

Issue Comment

Prioritisation Key issue – which embankments need most urgent attention?Stabilisation costs vs risk How can limited funds be most effectively used?Types of failure/deformation mechanism Deep seated vs shallow failure, or excessive track deformation?Consequence of failure/deformation Train derailment, or delay, or inconvenience to neighbours?Proactive management When are problems likely to develop? Relevant strengths/pore pressures for assessment?

Future degradation rates? Application of site observations?

� Mott MacDonald Ltd, Croydon, UK


(2009) for Network Rail. During the period 2000–2003, itwas estimated that delays due to geotechnical issues costabout £26 million. Figure 1(a) shows that the majority ofthese delays were located in southern England. Correlationwith geological mapping indicated that these problems weremainly located in areas where high-plasticity clays arefound. Statistical analysis of the data demonstrated thatabout 80% of the total delay time was due to a relativelysmall (about 20%) number of incidents. As would beexpected, many of the delays occurred during the winter.Perhaps more surprisingly, for high-plasticity clays, thedelays that occurred during dry summer weather were of asimilar order of magnitude as those that occurred during wetwinters, (Fig. 1(b)). Loveridge et al. (2010) discuss theresults of this survey in more detail. A more recent studyfor the Rail Safety and Standards Board, for the period2003–2009, has confirmed these trends (Mott MacDonald,2011), with about 70% of the failures (excluding those dueto scour or ‘wash-out’) located in Network Rail’s Southernand Western Territories. A significant number of serviceabil-ity limit state (SLS) failures (i.e. excessive deformation)were reported during late summer/early autumn, when theeffects of desiccation on high-plasticity clay fills would beanticipated to be most significant. Of the 95 ultimate limitstate (ULS) failures, about half were classified as deepseated, with failure through the embankment crest. The restof the ULS failures were classified as local crest instability,unravelling of oversteepened ash/ballast at the shoulder(typically most prevalent during the summer), or shallow,translational, failures. Delayed deep-seated failures areusually associated with slopes that have a modest cover ofvegetation, such as grass or small shrubs, whereas SLSproblems are usually associated with large trees beingpresent on the slope. The embankments across the railnetwork are mostly between about 2 m and 8 m high, withslope angles typically between 228 and 308. Hence, theslopes are often smaller and steeper than those which havebeen the focus of previous studies reported in the literature.

FIELD STUDIESSkempton (1996) has described the construction methods

which were used for these old embankments, and theirresulting heterogeneity. Figure 2(a) shows a historical photo-graph of the construction of a railway embankment, wherethe clay fill was tipped onto the natural ground surface,without any systematic compaction of the clay fill. Theseuncompacted clay fills are often known as ‘dumped’ clayfills (Vaughn et al., 2004). Following construction, large‘collapse-type’ settlement of the unsaturated clay fills wouldhave occurred upon rainfall infiltration. These settlementswere compensated by adding readily available granular fill(such as locomotive ash or track ballast). Figure 2(b) showsa sketch of a typical rail embankment composed of clay fill.Major differences are likely to exist between a modernhighway embankment of heavily compacted clay fill and anold embankment of dumped clay fill. In addition to greaterheterogeneity, the old rail embankments often contain ‘hid-den defects’, due to, inter alia: inclusion of old shearsurfaces from historic instability (only partially removed/repaired); local high-permeability sand lenses; and remnantsof topsoil/alluvium along the base of the embankment(typically there was little preparation or improvement of theunderlying natural soils prior to construction). The overallshape of these embankments can be quite variable. Someembankments have retained a relatively steep uniform slope;whereas others have a ‘coat-hanger’ appearance with anoversteepened upper slope and a shallower lower slope.

Several field studies have been undertaken to better under-stand embankment deformation mechanisms. Figure 3, basedon studies by Scott et al. (2007) of a London UndergroundLtd (LUL) embankment, plots vertical deformation at anembankment crest (monitored via extensometers), track de-formation and soil moisture deficit (SMD) against time. Itshould be noted that the SMD values which are used byScott et al. are the values published by the UK’s Met Office,specifically SMD for deciduous trees. Two different embank-ment areas were monitored, one was in an area with a dense

Location of major train delays(a)

Major delays ( 8 hours duration),summer and winter versus plasticity

(b)

�

Key:

Median

25%–75%

A Non-plastic, D High plasticity� �

Non-outlier range

Outliers

Summer Winter2

0

�2

�4

�6

�8

Log

(nor

mal

ised

del

ay m

inut

es)

A B C DPlasticity

A B C DPlasticity

Fig. 1. Network Rail, delays due to geotechnical causes, 2000 to 2003

20 O’BRIEN


cover of very high water demand (VHWD) trees (oaks andpoplars), while the other was predominantly a grass-coveredslope. The two areas were relatively close (less than 100 mapart), and the clay fill (derived from weathered London

clay, with a plasticity index of about 45–50%) was practi-cally identical at both of the monitoring locations. Embank-ment deformation was more than ten times higher in thearea affected by VHWD trees. Deformation in the VHWDtree area was closely correlated with changes in SMD. AsSMD increased during the summer, settlement occurred,while heave occurred when SMD reduced during the winter.In the tree area, the track movement follows a similarpattern to the crest deformation.

Figures 4 and 5 summarise some recent monitoring ofNetwork Rail embankments. One embankment is composedof Gault clay fill, and the other is composed of London clayfill. The index properties for these sites are summarised inTable 2 and can be compared with the LUL fill. Figure 4plots horizontal displacement, via inclinometers in the mid-slope area, versus time for the Gault clay site. It can be seenthat the greatest displacement is at the slope surface, andthere is a seasonal trend in displacement. Displacementswere less than 5 mm at depths greater than about 4 m.Outward displacement is observed between February andJuly, the displacement then moves inwards between August

�45�40�35�30�25�20�15�10

�50

Dis

plac

emen

t (ve

isaw

ay fr

om tr

ack)

: mm

�

18/0

1/07

09/0

3/07

28/0

4/07

17/0

6/07

06/0

8/07

25/0

9/07

14/1

1/07

03/0

1/08

22/0

2/08

12/0

4/08

01/0

7/08

21/0

7/08

09/0

9/08

29/1

0/08

18/1

2/08

0·3 m

1·8 m

2·3 m

(a) Moderate water demand trees on slope

Summer/Autumn Winter/Spring Summer/Autumn

0·3 m

1·3 m

2·3 m

3·3 m

(b) Very high water demand trees on slope�18�16�14�12�10

�8�6�4�2

02

Dis

plac

emen

t (ve

isaw

ay fr

om tr

ack)

: mm

�

Notes:

1. Vegetation on embankment slope classified in accordance with NHBC (2003), except oak and poplar trees classified as Very High Water Demand (VHWD)

Fig. 4. Horizontal deformation versus time, Gault Clay embankment

050

100150200250300350

�50�40�30�20�100102030

SM

D: m

m

Date

Ver

tical

mov

emen

t: m

m

Jan 04 Aug 04 Feb 05 Sep 05 Mar 06 Oct 06 Apr 07

Notes:

1. SMD is Soil Moisture Deficit for deciduous trees.2. NHBC (2003) classification system used to classify tree water demand,

except that an additional class is added, Very High Water Demand(VHWD) for oaks and poplars, based on experience of the significanteffects that these trees have on track deformation.

SMD Grass area

Track displacement Tree area

Fig. 3. Vertical deformation, influence of climate (SMD) andvegetation (after Scott et al., 2007)

Embankment construction(a)

Highpermeabilityballast

Local highpermeabilitylenses

Irregularslope Dessicated

clay fillShearsurfacesPoorly compacted

heterogeneous fill

Topsoil and Alluvium

“Coat-hanger” shape- oversteepened crest- shallower lower slope and toe bulges

Heterogeneity and “hidden defects”(b)

Fig. 2. Historical rail embankment, construction and typicalfeatures

THE ASSESSMENT OF OLD RAILWAY EMBANKMENTS – TIME FOR A CHANGE? 21


and November, followed by another outward/inward cycleduring the subsequent winter/summer, respectively. Duringthe 2-year monitoring period there was a net outward move-ment of between 10 mm and 15 mm per annum. The inclin-ometer located within an area of VHWD trees showed rathersmaller seasonal horizontal displacements than the inclin-ometers located in areas of less dense and lower waterdemand vegetation. Outward movement coincided with peri-ods when SMD was less than about 10 mm. Embankmentdeformation correlated better with SMD than rainfall, solarradiation or evapotranspiration. Inclinometers near the crestdid not show any significant deformation (i.e. greater than5 mm). Figure 5 plots vertical and horizontal deformationagainst time for the London clay site. At this site, in contrastto the Gault clay site, significant vegetation was removed toallow rig access (Fig. 5(a)), and two large oak trees were

removed close to the instruments. Vertical displacement (Fig.5(b)) exhibits an initial settlement during the first summer,followed by sustained swelling which is reasonably deepseated (to a depth of about 4 m). Horizontal displacement(Fig. 5(c)) exhibits negligible movement during the firstsummer, followed by a relatively large outward movementduring the following winter. This movement is relativelyshallow, predominantly within the upper 2 m of the slope.The movements are consistent with a gradual wetting-up ofthe slope following the removal of VHWD trees and rainfallduring the autumn/winter of 2006/2007.

Figure 6(a) plots the weather station data for rainfall atthe Gault Clay site for part of the monitoring period (June2007 to May 2008) compared against long term averages(1972 to 2000) published by the Met Office. Local SMDdata published by the Met Office is plotted on Fig. 6(b).

40

28

16

5

�6

�18

�30

Set

tlem

ent:

mm

Phase 1, Section 1Two large oaks removedto allow rig access forinstallation (one of threeremains).

VHWD tree removal(a)

Vertical deformation versus time(b)

1·151 m 1·897 m 3·073 m 4·092 m 4·980 m 7·038 m 8·823 m

01/0

4/06

10/0

7/06

18/1

0/06

26/0

1/07

06/0

5/07

14/0

807

22/1

1/07

01/0

3/08

09/0

6/08

17/0

9/08

�60

�50

�40

�30

�20

�10

0

10

Dis

plac

emen

t(

ve is

aw

ay fr

om tr

ack)

: mm

�

0·2 m

0·7 m1·2 m

1·7 m

2·2 m

Inst

rum

ent

mal

func

tion

Mid slope inclinometer

05/0

6

06/0

6

07/0

6

08/0

6

09/0

6

10/0

6

11/0

6

12/0

6

01/0

7

02/0

7

03/0

7

04/0

7

05/0

7

06/0

7

07/0

7

08/0

7

09/0

7

10/0

7

11/0

7

12/0

7

01/0

8

02/0

8

03/0

8

04/0

8

05/0

8

06/0

8

07/0

8

08/0

8

09/0

8

10/0

8

11/0

8

12/0

8

Horizontal deformation versus time(c)

Notes:

1. Tree removal during mid-March, 2006.2. Depths quoted are depth below slope surface.

Crest extensometer

Fig. 5. Influence of VHMD tree removal on embankment deformation, London Clay embankment

Table 2. Index properties for dumped clay fills

Site Geology Moisture content (%) Liquid limit (%) Plasticity index (%) Clay fraction(%)

Sand/gravel fraction(%)

Mean SD Mean SD Mean SD

Network Rail,Charing

Gault clay 24.9 3.3 57.4 10.9 34.1 9.1 30–60 5–40

Network Rail,Pound Green

London clay 25.1 5.0 53.8 12.5 33.0 9.7 25–50 10–40

LondonUnderground(LUL)(1)

London clay 26.5–35.8 5.3–7.4 67.7–74.4 6.7–15.9 42.7–49.4 6.5–13.9 50–70 0–15

Notes:(1) Ranges for six different LUL sites.(2) SD ¼ standard deviation.

22 O’BRIEN


Figure 7 plots pore-water pressures recorded by geopiezo-meters (Ridley et al., 2003) against the depth below theslope surface. During the monitoring period, the rainfallpattern was unusual for south-east England, with a wetterthan average summer followed by a drier than averagewinter. Most of the piezometers measured small suctions (upto 75 kN/m2 negative pore pressure), especially at depthsgreater than 2.5 m and in the mid and upper slope areas.The piezometers used (Ridley et al., 2003) are capable ofmeasuring suctions in the field of up to about 100 kN/m2:At shallower depths (about 2 m), higher pore-water pressureswere recorded, with a few recording small positive pressures.Typically, the pore-water pressures in the lower slope areawere higher than those towards the crest. These measuredpore pressures need to be considered in the context of theprevailing weather conditions. During summer periods muchlarger suctions have been measured, especially beneath areasaffected by VHWD trees; for example, O’Brien et al. (2004)report summer suctions between 100 kN/m2 and 400 kN/m2,together with seasonal changes of more than 200 kN/m2:Maximum SMD (for deciduous trees) during the monitoringperiod was 70 mm, compared with a value of 279 mmduring the dry summer of 2003 and a national long-termaverage maximum summer SMD of 124 mm. Because of therelatively wet summer followed by a dry winter, the seasonal

change in pore-water pressure at the Gault clay site isrelatively low, 30 kN/m2 or less.

During the winter of 2000/2001, the UK experienced thewettest weather since records began in 1766 (Birch &Dewar, 2002). During the winter/spring period of 2001, spotmeasurements of pore-water pressure were recorded at sev-eral sites across the LUL network (Ridley et al., 2004).These data have been re-analysed by Briggs et al. (2011)and are plotted on Fig. 8. At sites underlain by a low-permeability (typically weathered London clay) foundationsoil (Fig. 8(a)), the piezometer data are bounded by ahydrostatic line to a depth of about 2 m below the clay-fillsurface. Beneath this, the bounding profile reduces to about60% of hydrostatic. Relatively low pressures were recordedat many locations, with about 20% of the readings close tozero pressure, and more than half the pressures being lessthan 20 kN/m2: At sites underlain by a relatively high-permeability (typically chalk or river terrace deposits)foundation soil (Fig. 8(b)), virtually all the piezometersrecord relatively low pressures, with 85% of the data valuesbeing less than 10 kN/m2:

SMD has often been used to assess the risk of slopeinstability (e.g. Ridley et al., 2004). As discussed below, thecomparison between SMD, site-specific pore pressure andrainfall data suggests that some care is required in usingSMD data. For the Gault clay site, from December 2007 toMay 2008 there were 6 months when SMD was zero (orvery close to zero, less than 5 mm), which is only onemonth less than the historically wet winter of 2000/2001.However, winter rainfall is relatively low. This apparentinconsistency can be explained by the preceding wet summer(when SMD was less than 70 mm). Hence, the soil surfacezone requires less rainfall infiltration to return to fieldcapacity. It should be noted that SMD is a measure ofmoisture conditions in the near-surface soil zone, and maynot be representative of pore pressure conditions at depth.Figure 6(a) indicates that the rainfall data measured at thesite weather station is consistently lower than the local MetOffice data (for MORECS square 174). The North Downsare located in this MORECS square and they are likely toinfluence weather patterns, causing differences in weatherpatterns across the 40 km of the MORECS square. PublishedMet Office data (including SMD values) remain a usefultool for considering broader patterns of behaviour, but may

0

20

40

60

80

100

120

140M

onth

ly r

ainf

all t

otal

: mm

Weather station

MORECS data (square 174)

MORECS LTA (square 174)

Monthly rainfall(a)

06/0

706

/07

07/0

707

/07

08/0

708

/07

09/0

709

/07

10/0

710

/07

11/0

711

/07

12/0

712

/07

01/0

801

/08

02/0

802

/08

03/0

803

/08

04/0

804

/08

05/0

805

/08

SMD LTA

SMD Deciduous Trees

0

10

20

30

40

50

60

70

80

90

100

SM

D: m

m

Soil Moisture Deficit, SMD(b)Notes:

1. MORECS Met Office Rainfall and Evaporation Calculation System2. LTA Long Term Average

��

Fig. 6. Monthly rainfall and SMD during monitoring period,Gault clay embankment

5

4

3

2

1

0�100 �75 �50 �25 0 25 50

Pore pressure: kPa

Dep

th: m

m

Upper slope

Mid slopeLower slope

Hydrostatic

Hydrostatic –1 m

Gault Clay site

London Clay site

Notes:1. Gault Clay site, monitoring between January 2007 and January 2008.2. London Clay site, monitoring between April and December 2008.

Fig. 7. Measured porewater pressure at NR sites



be inadequate for assessing local earthworks risks (both forULS and SLS conditions).

DUMPED CLAY FILL PROPERTIESAt several sites, block samples of clay fill have been

obtained and suites of laboratory tests have been carried outto assess shear strength, deformation and permeabilitycharacteristics of saturated test specimens. Field tests tomeasure shear modulus, at very small strain (Go) with aseismic cone, and permeability have also been carried out.Test data for LUL sites have been discussed in O’Brien etal. (2004) and O’Brien (2007). The LUL data are comparedwith the Network Rail data (for the Gault and London claysites) in Figs 9, 10 and 11 for shear strength, compressibilityand permeability, respectively. At the Gault clay site, the insitu Go profile fluctuated between 30 MN/m2 and 70 MN/m2

(occasional hard bands gave higher values), and the labora-tory Go measurements were broadly similar, indicating thatthe block samples are of high quality. In addition to the testson block samples, a series of tests was carried out onreconstituted specimens, which were reconsolidated to a

range of different overconsolidation ratios. For the Gaultclay fill (Fig. 9(a)) shows the effective stress paths forundrained triaxial tests on block samples and reconstitutedclay fill. The stresses have been normalised by the equivalentpressure on the measured oedometric intrinsic compressioncurve (� 9�ve ) (Burland, 1990). This Hvorslev-type normalisa-tion procedure facilitates an appraisal of the potential influ-ence of soil structure on peak strength (Burland et al.,

0

0

1

1

2

2

3

3

4

4

5

5

6

6

7

7

Dep

th b

elow

cla

y su

rfac

e: m

mD

epth

bel

ow c

lay

surf

ace:

mm

�2

�2

�1

�1

0

0

1

1

2

2

3

3

4

4

5

5

6

6

Pressure head: m

Pressure head: m

London Clay foundation

Hydrostatic pressure line

Bounding pressure profile

Low permeability foundation soil(typically weathered London Clay)

(a)

High permeability foundation soil(typically River Terrace Deposits or Chalk)

(b)

Chalk foundationRiver Terrace Depositsfoundation

Hydrostatic pressure line

Bounding pressure profile

Fig. 8. Pressure heads measured across London Underground Ltdnetwork during March/April 2001 (based on data by Ridley et al.,2004)

0

0·0

0·1

0·1

0·2

0·2

0·3

0·3

0·4

0·4

0·5

0 0·2 0·4 0·6 0·8 1·0 1·2 1·4

t/

*�

�σve

s / *� �σve

Triaxial stress paths – block samples

Triaxial stress paths – reconstituted clay

OCR 1�

OCR 12�

OCR 6�

OCR 3�

Block and reconstituted samples(a)

0

0·05

0·10

0·15

0·20

0·25

0 0·1 0·2 0·3 0·4 0·5 0·6 0·7 0·8

t/

*�

�σve

s / *� �σve

Gault Clay fill- block samples

LUL London Clayfill - block samples

Gault Clay fill and LUL London Clay fill(b)

Notes:

t ( )/2; s ( )/2.* equivalent pressure on ICL derived from shear zone moisture

content.ICL intrinsic compression line.

� � � � � � � � ��

�

σ σ σ σσ

v h v h

ve

Fig. 9. Dumped clay fills, normalised triaxial stress paths tofailure

0

0·02

0·04

0·06

0·08

0·10

0·12

0·14

0·16

0·18

0·20

�2·5 �2·0 �1·5 �1·0 �0·5 0

Cs

seca

nt

log ( / )σ σ�v v0�

London Clay, block samples

London Clay, reconstituted

Gault Clay, , block samples

Gault Clay, reconstituted

Notes:

1. is vertical effective stress,σ �v σ�v0 is vertical effective stress atcommencement of swelling. Cs is secant index.

Fig. 10. Dumped clay fills, swelling behaviour, Gault Clay andLUL London Clay

24 O’BRIEN


1996). From Fig. 9(a) it can be seen that the normalisedstrength of the clay fill is similar to that of the overconsoli-dated reconstituted clay. Hence, the in situ clay-fill strengthdoes not appear to be significantly affected by structure. TheGault clay fill tends to mobilise higher shear strength thanthe London clay fill from the LUL sites (Fig. 9(b)); this canprobably be explained by the lower plasticity index andlower clay fraction of the Gault clay fill.

Figure 10 plots the secant swelling index (CS, measuredduring oedometer tests) of reconstituted and block samplesof clay fill, against the ratio of the current (� 9v) to initialvertical effective stress (� 9vo) during unloading (� 9v=� 9vo). Thedata fall into three bands: reconstituted clay fill, blocksamples of clay fill swelled from high stress (in excess of3000 kN/m2) and block samples of clay fill swelled fromlow stress (about 250 kN/m2). The reconstituted samplesexhibit the highest values of CS, and the block samplesswelled from low stress the lowest values of CS: This isindicative of the clay fill retaining some of the ‘structure’(including bonding) from the parent natural clay. The Gaultclay fill is about 30–50% stiffer in swelling then the LULLondon clay fill.

Laboratory and in situ permeability test data are sum-marised on Fig. 11. As discussed later in this paper, per-meability is an important parameter. Permeability variessignificantly, for a particular soil, with the degree of satura-tion. However, unless stated to the contrary, the permeabilityvalues quoted below are for the soil in a fully saturated state.There is a consistent trend for the laboratory values to belower (typically by one or two orders of magnitude) than thein situ values. The in situ values (from variable-head bore-hole tests) cover a wide range, although tests carried outwithin predominantly clay fills (at depths greater than about2 m) vary between 2 3 10–9 m/s and 5 3 10–8 m/s. Com-pared with published data for the in situ permeability ofnatural stiff overconsolidated fissured clays (e.g. Hight et al.,2003), the permeability of the dumped clay fill is about anorder of magnitude higher. Tests in fills affected by silt andsand lenses have measured permeabilities of between10–7 m/s and 5 3 10–6 m/s. Guelph permeater tests were alsocarried out at the slope surface, and are representative of thepermeability (most likely at conditions near ‘field saturation’,i.e. a degree of saturation Sr of ,90%) of a small volume ofsoil. Derivation of the test results accounts for the unsatu-rated nature of the near-surface soils. Measured Guelph test

permeabilities were between about 1.0 3 10–6 m/s and3.0 3 10–6 m/s for the surface clay fills, i.e. about two tothree orders of magnitude higher than the permeabilitiesmeasured at depth within the embankment.

The variable influence of structure and macrofabric on thestrength, compressibility and permeability of dumped clayfills can be readily explained by a simple clod–matrixconceptual model (O’Brien, 2007). The clods represent theintact lumps of parent natural clay (which can be seen in thehistorical photograph shown in Fig. 2(a)). The matrix com-prises both remoulded clay and foreign matter (silt, sand,etc.) introduced during handling, transportation and tippingof the fill. The test data outlined above indicate that com-pressibility/swelling behaviour is largely controlled by theclods, whereas the matrix mainly controls the shear strengthand permeability of the clay fill. An X-ray computed tomo-graphy study at Southampton University (Watson et al.,2007) has confirmed the validity of this simple conceptualmodel.

NUMERICAL MODELLING OF DEEP-SEATEDDELAYED FAILURE

The numerical modelling of the delayed failure of oldrailway embankments has been described by Kovacevic etal. (2001) and Nyambayo et al. (2004), using ICFEP (Im-perial College Finite Element Program), and O’Brien et al.(2004), using FLAC (Fast Lagrangian Analysis Code). TheFLAC modelling simulates the post-peak degradation ofstrength towards post-rupture and then residual strength. Thefailure mechanism commences at the toe, and then, with anincreasing number of seasonally induced pore pressurechanges, the failure surface propagates along a subhorizontalsurface towards the embankment core (Fig. 12). At failure,the clay-fill strength is close to residual in the vicinity of theembankment toe, whereas near the crest it is close to peakstrength. This non-uniform mobilisation of strength hasimportant practical implications for the design of potentialstabilisation measures. Prior to collapse, the simulation ofseasonal changes in pore pressure leads to a ‘ratcheting’-type deformation mechanism being predicted by the numer-ical models, with outward deformation during winter beingonly partially recovered during summer. Field observations(Fig. 4) also suggest that a ratcheting-type deformationmechanism can develop in these embankments, and thisdeformation mechanism also helps to explain the ‘coat-hanger’ shape that is often seen (Fig. 2). This deep-seateddelayed failure mechanism has also been verified by centri-fuge modelling, Take & Bolton (2004). The centrifuge testshowed that clay fill permeability, and rainfall duration andinfiltration were likely to control the development of swel-ling and shear strain at the embankment toe. Numericalmodelling reported by Nyambayo et al. (2004) has indicatedthat embankments with a clay-fill permeability of between10–7 m/s and 10–8 m/s will be prone to delayed failure,whereas embankments with a clay-fill permeability of10–9 m/s or less are relatively stable. The measured clay-fillpermeabilities (Fig. 11) transverse across this intermediaterange of permeability, which leads to embankments beingprone to delayed failure. Fully coupled hydrological/mech-anical modelling by Davies (2010) also showed that the bulkpermeability of the clay fill is critically important, andsignificantly affects the risk of collapse. These numericalmodelling studies were largely based on a characterisation ofthe LUL clay fills. Following the investigations at the Net-work Rail sites, additional numerical modelling has beencarried out. These additional studies used the same method-ology as that described by O’Brien et al. (2004), but thestrength and compressibility characteristics were varied

1 10� �13

1 10� �12

1 10� �11

1 10� �10

1 10� �9

1 10� �8

1 10� �7

1 10� �6

10 100 1000 10000

Per

mea

bilit

y, k

(m

/s)

Vertical effective stress, (kN/m )σ�v2

Note:

1. Slope surface measured permeability, 1 10 to 1 10 m/s(cracking??)

� �� 6 5

Block samples

Reconstituted

London Clay, reconstituted

In situ testsLUL, London Clay

In situ testsNR, Gault Clay fill

Gault clay fill,block sample

London Clay,block samples

Fig. 11. Permeability of dumped clay fill, in situ versus laboratorytest



across a broad range in order to simulate the site-specificdata described above, and the potentially more heteroge-neous conditions likely to be encountered across NetworkRail sites. Some of the results from this parametric study aresummarised in Table 3. Appendix 1 provides a summary ofthe modelling and ‘base case’ parameters. The compressibil-ity and strength of the clay fill (both peak and residualstrength) had an influence on the number of cycles to

failure, as would be expected. However, the risk of collapse(implied by the change in the number of cycles to failure)was highly sensitive to two particular factors:

(i) the inclusion of a thin layer of alluvium below the clayfill

(ii) a relatively small change in maximum winter porepressures in the clay fill.

The presence of a more compressible layer of alluviumled to a concentration of plastic strain within this layer.Once plastic strains are initiated at the toe, the embankmentwith the underlying soft alluvium is less able to mobilise therequired shear resistance elsewhere in the embankment. Incontrast, the clay fill without an alluvial layer (because of itsmore uniform compressibility and shear stiffness) was ableto mobilise a greater proportion of its peak shear strength.The change in winter pore pressure considered two scenar-ios:

(a) Poorly Drained: 100 kN/m2 summer surface suction,reducing linearly with depth to zero at 2 m below theslope; zero winter surface suction.

(b) Well Drained: summer surface suction, same as in (a)above, but zero suction between the surface and 1 mdepth during winter.

For both scenarios, the pore pressure increase below thezero suction line was about 80–90% of hydrostatic. Forscenario (b), the lower winter pore pressure at the toe leadsto a higher mobilised shear strength and lower plasticstrains. Hence, the rate of degradation of strength, due toseasonal pore pressure changes, was far lower for scenario(b) than for scenario (a).

NUMERICAL MODELLING OF CLIMATE–POREPRESSURE INTERACTIONS

VADOSE/w has been used to assess the influence ofclimate on pore pressures in embankment fill (Mott Mac-Donald, 2008). Further detail is given in Briggs (2011). Awide range of scenarios have been considered, including:

(i) changes in climate(ii) changes in clay-fill permeability(iii) changes in vegetation type on the embankment (VHWD

trees versus grass)(iv) changes in slope geometry.

The VADOSE modelling is summarised in Appendix 2.Selected output from some of the modelling is given in Figs

Developmentof shear zone

8·0

7·5

7·0

6·5

6·0

�0·25

�0·25

0·25

0·25

0·75

0·75

1·25

1·25

1·75

1·75

2·25

2·25

Growth of residualshear zone 7·75

7·25

6·75

6·25

( 10 )After 15 shrink-swell cycles

(a)

� 1

( 10 )After 34 shrink-swell cycles

(b)

� 1

Note:

1. Embankment 7·35 m high, with 1:3·3 (V:H) slope gradient.Poorly drained, grass covered slope, intermediate plasticityclay fill.

Fig. 12. Numerical modelling of progressive failure

Table 3. Numerical modelling: influence of an alluvial layer and winter pore pressure on delayed failure

Sitegeometry(3)

Fill properties Groundwaterconditions

Description of parameter change Number of cycles to failure

H ¼ 4.6 mG ¼ 1:1.8(1)

Intermediate plasticity Poorly drained Influence of a thin layer of alluvium belowembankment

33(no alluvium)

7(alluvium)

H ¼ 7.5 mG ¼ 1:3.2(2)


35(no alluvium)

19(alluvium)

H ¼ 8 mG ¼ 1:2.5(2)


15(no alluvium)

8(alluvium)

H ¼ 7.5 mG ¼ 1:3.2

High plasticity Changes simulated Varied from ‘poorly drained’ to ‘well drained’ 14(poorly drained)

.50(Well drained)

H ¼ 4.6 mG ¼ 1:1.8

Intermediate plasticity Changes simulated Varied from ‘poorly drained’ to ‘well drained’ 33(poorly drained)

.50(well drained)

Notes:(1) Alluvial layer assumed to be 0.5 m thick.(2) Alluvial layer assumed to be 1.0 m thick.(3) H ¼ embankment height, G ¼ slope gradient, vertical: horizontal.

26 O’BRIEN


13 to 16. Figure 13 shows a contour plot of pore pressureswithin a relatively large, shallow tree-covered embankmentduring a summer/winter cycle. The Newbury weather data(Smethurst et al., 2006) was used for 2005/2006, whichrepresents a dry summer/wet winter cycle (with SMD vary-ing between zero and 150 mm), although this is not anextreme weather period. In summer, deep zones of suctionare mobilised below the tree-covered slope, and during thesubsequent winter the near-surface layers (upper 2 m or so)wet-up and exhibit positive pore pressures in some locations.If the trees are removed from the upper two-thirds of theslope (Fig. 14), then the depth of summer suctions isreduced, and much larger zones of positive pore pressuredevelop during subsequent winter periods below the grass-covered parts of the slope. Winter suctions are still main-tained at depth below the lower third of the slope, which istree covered. Because of the idealisations within this type ofanalysis, the outputs are not intended to be ‘predictive’ ofthe actual spatial distribution of pore pressures. However,these analyses provide useful insights into overall patterns ofbehaviour, and the relative importance of different variables.The clay-fill permeability has a significant effect, as wouldbe expected, with a critical intermediate zone of permeabil-ity of between 5 3 10–8 m/s and 5 3 10–7 m/s being identi-

fied (Mott MacDonald, 2008). Pore pressures increasedrelatively rapidly during winter rainfall, when clay-fill per-meability was 5 3 10–7 m/s, whereas negligible pore pres-sure change occurred if clay-fill permeability was as low as5 3 10–9 m/s. An important consideration is the ratio ofclay-fill permeability to daily rainfall. At the Gault clay site,during the monitoring period, on days when rainfall events.1 mm occurred, the average daily rainfall was about 5 mm,with a maximum of 25 mm. The rate of rainfall infiltrationis limited by the saturated permeability of the clay fill, withany excess rainfall running off the slope. For a clay-fillpermeability of 5 3 10–9 m/s, the maximum daily rainfallinfiltration is less than 0.5 mm/day (hence most rainfall runsoff the slope), whereas a clay-fill permeability of5 3 10–8 m/s or higher allows most of the daily rainfall toinfiltrate into the slope. Hence, a large number of moder-ately wet winter days will lead to far higher winter porepressures than will a small number of extremely intenserainfall events.

As noted in the Introduction, many of the old railwayembankments are relatively small and steep. It is interesting,therefore, to compare the influence of slope geometry onseasonal changes in pore pressure (Fig. 15). Comparing Figs15(a) and 15(b), the smaller, steeper embankment exhibits

Summer

Trees

Trees

12

10

8

6

4

2

0

Ele

vatio

n: m

m

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28

�50 kM/m suction2

12

10

8

6

4

2

0

Ele

vatio

n: m

m

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28

Positive porewaterpressure

Winter

Trees

Trees

Distance

PWP: kPa�50�40�30�20�10

0�10�20�30

Ash & ballast

Ash & ballast

Embankment fill

Embankment fill

Foundation material

Foundation material

Initial water table

Initial water table

DJI 409

DJI 409

BH409

BH409

DH411

DH411

BH405

BH405

BH212

BH212

�40

�50

�40

�30

�30

�20

�20

�10

�10

0

0

10

10

20

Fig. 13. Hydrogeological modelling of seasonal changes in porewater pressure (tree covered slope)



Summer

Grass

Trees

12

10

8

6

4

2

0

Ele

vatio

n: m

m

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28

�50 kM/m suction2

12

10

8

6

4

2

0

Ele

vatio

n: m

m

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28

Positive porewaterpressure

Winter

Grass

Trees

Distance

PWP: kPa�50�40�30�20�10

0�10�20�30

Ash & ballast

Ash & ballast

Embankment fill

Embankment fill

Foundation material

Foundation material

Initial water table

Initial water table

DJI 409

DJI 409

BH409

BH409

DH411

DH411

BH405

BH405

BH212

BH212

�40�

50

�50

�40

�30

�30

�20

�20

�10

�10

0

0

10

10

Fig. 14. Hydrogeological modelling of seasonal changes in porewater pressure (trees removed fromupper slope)

�50

�40

�30

�20

�10

0

10

20

30

Por

ewa

ter

pres

sure

: kN

/m2

Midslope 2·2 m depth Toeslope 2·2 m depth

03/0

5

06/0

5

08/0

5

11/0

5

02/0

6

05/0

6

07/0

6

10/0

6

01/0

7

04/0

7

06/0

7

09/0

7

12/0

7

03/0

8

06/0

8

08/0

8

11/0

8

02/0

8

04/0

8

Embankment 6·5 m high. slope gradient 1:3·3 (V:H)(a)

03/0

5

06/0

5

08/0

5

11/0

5

02/0

6

05/0

6

07/0

6

10/0

6

01/0

7

04/0

7

06/0

7

09/0

7

12/0

7

03/0

8

06/0

8

08/0

8

11/0

8

02/0

8

04/0

8

Embankment 4·6 m high. slope gradient 1:1·8 (V:H)(b)

�70

�60

�50

�40

�30

�20

�10

0

10

20

30

Por

ewa

ter

pres

sure

: kN

/m2

Midslope 2·2 m depth Toeslope 2·2 m depth

Notes:

1. Grass covered slope.2. Newbury climate data, March 2005 to March 2008.

Fig. 15. Hydrogeological modelling, influence of embankment geometry

28 O’BRIEN


lower pore pressures in general than does the larger, shal-lower embankment. In particular, for the small embankmentthe summer/autumn maximum suctions are relatively high inthe toe slope area, and the winter maximum positive pore-water pressures are relatively low in the midslope area. Forboth embankments the toe area has higher winter porepressures than the midslope area. Modelling by Briggs et al.(2011) has shown that the permeability variations acrossthree zones – near-surface (higher permeability) clay fill,basal clay fill, underlying foundation soil – are critical. Afoundation soil permeability two orders of magnitude higherthan the overlying clay fill was found to be sufficient tocreate underdrainage conditions, comparable to those shownin Fig. 8(b).

From Fig. 7 it can be seen that, even during a relativelywet monitoring period, the slope pore pressures fluctuatebetween moderate suctions and small positive pore pressures.Hence, the variation in permeability with changes in soilsuction is likely to be a critical parameter. Unfortunately,measurements of unsaturated hydraulic properties, both inthe laboratory and, particularly, in the field, are lacking. Thesignificant influence of unsaturated hydraulic properties onseasonal pore pressure change are shown on Fig. 16(b). Theuse of the ‘wetting’ curve in Fig. 16(a) leads to smallchanges in seasonal pore pressure, in contrast to the use ofthe ‘drying’ curve, which leads to large changes in seasonal

pore pressure. The modelling discussed above assumed anintermediate variation of permeability with suction, betweenthe drying/wetting curves shown on Fig. 16(a).

The permeability characteristics of the unsaturated soilwill also be significantly affected by hysteresis effects (Li etal., 2005); i.e. during wetting the permeability will be lower,at a given soil suction, than during drying. Laboratory soil-column experiments (e.g. Lenhard et al., 1991) highlight theimportance of hysteresis effects on pore pressures developedin unsaturated soils. A major practical challenge is the lackof reliable data for the in situ hydraulic properties ofunsaturated soils, in general, and for dumped clay fills (withtheir clod–matrix structure) in particular.

PRACTICAL IMPLICATIONSFigure 17 schematically illustrates the degradation of

mobilised clay-fill strength and pore-water pressure changeswith time within old rail embankments, which have beendiscussed above. Figure 17 emphasises the dynamic natureof the interactions that take place between embankment clayfill, vegetation and climate; which is in stark contrast toconventional ‘steady-state’ assumptions. Two of the mostimportant factors that can affect the risk of delayed failure,or excessive embankment (and rail track) deformation, areclay-fill permeability and the type of vegetation cover on theembankment slope. Both of these factors are usually ignoredduring conventional assessments of embankment behaviour.Hence, it is unsurprising that conventional analyses areunhelpful in the context of asset-management risk assess-ments. During dry summers significant train delays arecaused by excessive track deformation due to VHWD treescausing desiccation of intermediate- and high-plasticity clayfills. Simple solutions for the elimination of seasonal trackdeformation are not available once large trees have becomeestablished across a slope. Tree removal from the crest willbe most effective, but HWD and VHWD trees at the toemay still cause some seasonal track deformation. Treeremoval from the lower half of the slope could lead tocatastrophic collapse during wet winters, once roots havedecayed and pore pressures recovered to equilibrium values,unless there is some stabilising measure to compensate.Hence, a very careful balance is required between minimis-

1 10� �16

1 10� �15

1 10� �14

1 10� �13

1 10� �12

1 10� �11

1 10� �10

1 10� �9

1 10� �8

1 10� �7

1 10� �60·01 0·1 1 10 100 1000 10000

Suction: kN/m2

Hyd

raul

ic c

ondu

ctiv

ity: m

/s

Assumed for VADOSE modelling

Drying curve

Wetting curve

Vegetation wilt point(1500 kN/m )2

Hydraulic conductivity vs. suction(a)

�120

�100

�80

�60

�40

�20

0

20

Por

ewa

ter

pres

sure

: kN

/m2

17/0

2/05

05/0

9/05

24/0

3/06

10/1

0/06

28/0

4/07

14/1

1/07

Drying curveWetting curve

Seasonal changes in porewater pressure(b)

Notes:

1. “Drying” and “wetting” curves based on modified Croney (1977)curves for natural London Clay.

2. Newbury climate data. Pore water pressure at 1·9 m below surface.

Fig. 16. Influence of unsaturated hydraulic properties on seasonalchanges in pore water pressure

Clay fillmobilised

strength High riskconditions

Change inconditions

Low riskconditions

Time: decades

Porewaterpressure

Grass

HWD tree

Dry summer/winter

Wetwinter

Positiveporewaterpressure

Zero

Suction

Assessment Design life

Notes:

1. High risk conditions (for slope instability) e.g. grass covered slope.Alluvium below embankment.

2. Low risk conditions (for slope instability) e.g. high permeabilityfoundation soils, well drained toe, HWD/VHWD trees on slope.

→

→

(Tree felled)

Time: months

Fig. 17. Rail embankments, ‘dynamic’ interactions with changesin conditions



ing seasonal track deformation and minimising the risk ofcatastrophic failures.

The field studies and hydrological modelling have illu-strated the complexity of the pore pressure regime that candevelop in these embankments, both in terms of spatialdistribution of vegetation and climate-induced changes. Theconventional approach of installing a few standpipe piezo-meters and monitoring for several months is unlikely tomeasure pore pressures that are representative of the valueswhich may develop over several decades. Changes inweather patterns could lead to significant differences in porepressure between a particular monitoring period and sub-sequent years.

CONCLUSIONSField observations and numerical modelling have shown

that the deformation and delayed failure of old rail embank-ments, composed of dumped clay fills, is significantlyaffected by seasonal climate variations and by different typesof vegetation. Intermediate- and high-plasticity clay fills areparticularly vulnerable to seasonal shrink–swell deformation,which ultimately can lead to progressive failure. Climate-and vegetation-induced seasonal shrink–swell can eithercause excessive track deformation or deep-seated delayedfailure, depending on the type and spatial distribution of thevegetation and the prevailing weather patterns. Because ofthe original construction method, the fabric and structure ofdumped clay fills is quite different to parent natural clay,reconstituted clay or compacted clay fill. This affects themass permeability, strength and compressibility of thedumped clay fill, but to differing extents. The risk of failurewill be mainly dependent on the maximum pore-waterpressure that may develop, which is predominantly con-trolled by the permeability variations across three differentzones (near-surface clay fill, underlying clay fill and founda-tion soil). The above factors that control behaviour, togetherwith the unsaturated nature of railway embankments, needsto be considered by practising engineers. The variation ofpore-water pressure with depth will seldom be hydrostatic,except perhaps within a near-surface layer, typically affectedby desiccation cracking.

In terms of further research, there is a clear need for thedevelopment of more reliable methods for in situ measure-ment of permeability. Simple, robust equipment for down-

hole tests, with modern, accurate devices to measure pres-sure and volume change are needed. The variation in per-meability between unsaturated and saturated states, includinghysteresis effects, is likely to be an important factor in theseasonal changes in pore-water pressure that occur withinembankments. There is an urgent need for further researchinto, and practical guidance on, this complex subject. Furtherfield studies will be required to better understand the un-saturated hydraulic properties of dumped clay fills, supportedby large-scale laboratory experiments (of sufficient scale tobe representative of the clod–matrix fabric) under carefullycontrolled boundary conditions. The factors affecting desic-cation cracking and associated effects on bulk permeabilityalso need to be better understood.

ACKNOWLEDGEMENTSThe author would like to thank his many colleagues at

Mott MacDonald for their contributions to the studies dis-cussed in this paper, including: James Scott, Ashley Bown,Fleur Loveridge, Graham Taylor, Yu Sheng Hsu and RingoTan. Kevin Briggs carried out VADOSE modelling duringhis EngD research at Southampton University, and theadvice of Professor William Powrie and Dr Joel Smethurstat Southampton University is also acknowledged.

The support of Brian McGinnity, for the LUL studies, andDerek Butcher and Eifion Evans for the Network Rail stud-ies is also gratefully acknowledged.

APPENDIX 1 – FLACA strain-softening strength model was used, as discussed by

O’Brien et al. (2004), with peak, post-rupture and residual strength.Peak strength can be mobilised until a defined plastic strain isdeveloped; once this threshold is exceeded the mobilised strengthreduces rapidly with increasing plastic displacement to residualstrength. Table A1.1 summarises the input parameters. Displace-ments were converted to shear strain by dividing by the characteristicelement thickness. For the Gault clay embankment 2,600 elements(or ‘zones’ in FLAC terminology) were used, with a characteristicdimension of 0.61 m (or about 8% of the embankment height). Forthe smaller, steeper London clay embankment, 1,500 elements wereused, with a characteristic dimension of 0.41 m (or about 9% of theembankment height). An ash–ballast capping 1 m thick was assumedfor all embankments.

Table A1.1. FLAC input parameters

Soil type Bulk unitweight

(kN/m3)

Strength(4) Thresholdplastic strain

for peak

Young’smodulus(kN/m2)

Poisson’s ratio

Peak Post-rupture(3) Residual(3)

High-plasticity clay fill(1) 18.1 c9 ¼ 5�9 ¼ 21

c9 ¼ 2�9 ¼ 21

c9 ¼ 1�9 ¼ 13

6% 75 (p9 + 100)Min. 5000

0.3 load0.2 unload

Intermediate-plasticityclay fill(2)

18.8 c9 ¼ 8�9 ¼ 24

c9 ¼ 2�9 ¼ 22.5

c9 ¼ 1�9 ¼ 13

7% 90 (p9 + 100)Min. 6,000

0.3 load0.2 unload

Alluvium 16.0 c9 ¼ 3�9 ¼ 24

c9 ¼ 2�9 ¼ 22.5

c9 ¼ 1�9 ¼ 13

7% 45 (p9 + 100)Min. 3,000

0.3 load0.2 unload

Ash–ballast 10.5 c9 ¼ 2�9 ¼ 35

N/A N/A N/A 1,000 0.3 (load + unload)

Natural clay foundation 19.6 c9 ¼ 10�9 ¼ 27

c9 ¼ 3�9 ¼ 24

c9 ¼ 1�9 ¼ 13

3% 135 (p9 + 100)Min. 7,500

0.2 (load + unload)

Notes:(1) High plasticity is characterised as a plasticity index of 45–50% and a liquid limit of 70–75%.(2) Intermediate plasticity is characterised as a plasticity index of 30–35% and a liquid limit of 55–60%.(3) Plastic displacement of 5 mm to develop post-rupture strength, and 100 mm to develop residual strength for all strain-softening soils.(4) The failure envelopes are curved, and can be defined either as tangent (c9, �9) values or secant (c9 ¼ 0, �9) values for the particular range ofmean effective stress of interest. For the purpose of this study it was convenient to use tangent fits, and for the slopes considered the meaneffective stresses are less than 100 kN/m2:

30 O’BRIEN


APPENDIX 2 – VADOSE/wVADOSE/w calculates saturated and unsaturated water and heat

flow in response to applied boundary conditions. Daily climate data(temperature, humidity, wind speed, rainfall and solar radiation) areapplied to calculate water infiltration and water removal from theslope surface and from a defined rooting zone. Hence, variations inpore pressure with time, as a result of different weather patterns ofdifferent duration and intensity, can be assessed.

Assumed input parameters are summarised in Table A2.1.Unsaturated hydraulic properties were based on a modified versionof the Croney (1977) soil water retention curve (SWRC) for dryingmeasured for London clay. The dumped clay fill was assigned alower air-entry value and a shallower gradient than the Croney curve,in order to take account of the clod–matrix structure of the fill andthe wide range of pore sizes, including significant sand–gravelfraction. Van Genuchten (1980) parameters were used to develop anappropriate curve fit. The SWRCs were used to define the variationof hydraulic conductivity with suction, based on the Mualem method(Mualem, 1976). Water removal due to evaporation from anunsaturated soil is calculated using the Penman–Wilson equation(Wilson et al., 1994) and transpiration is measures using a rootwater-uptake model (Tratch et al., 1995). The leaf area index (LAI),defining the proportion of solar energy intercepted by the vegetationfor transpiration (Ritchie, 1972), corresponds to full leaf coverduring summer (1 April to 17 October) and zero during the winter,on the basis of typical plant leafing periods (Biddle, 1998).Reduction in root water uptake due to soil drying as the vegetationreduces transpiration was based on the Feddes et al. (1978)relationship.

REFERENCESBirch, G. P. & Dewar, A. L. (2002). Earthwork failures in response

to extreme weather. In: FORDE, Proc. Of the Int. Conf. RailwayEngineering, 2002, London. Edinburgh, UK: Engineering Tech-nics Press.

Biddle, P. G. (1998). Tree root damage to buildings, Vols. 1 and 2.Wantage, UK: Willowmead.

Briggs, K. M. (2011). Impacts of climate and vegetation on railwayembankment hydrology. Doctor of Engineering thesis, Universityof Southampton, UK.

Briggs, K. M., Smethurst, J., Powrie, W. & O’Brien, A. S. (2011).Wet winter pore pressures in railway embankments. Proc. ICE,Geotechnical Engineering (in press).

Burland, J. B. (1990). On the compressibility and shear strength ofnatural clays. Geotechnique 40, 329–378.

Burland, J. B., Rampello, S., Georgiannou, V. N. & Calabresi, G. I.(1996). A laboratory study of the strength of four stiff clays.Geotechnique 46, 491–514.

Perry, J., Pedley, M. & Reid, M. (2003). Infrastructure embankments– condition appraisal and remedial treatment. CIRIA Publica-tion C592. London, UK: CIRIA.

Croney, D. (1977). The design and performance of road pavements.Technical report. London, UK: Her Majesty’s Stationary Office.

Davies, O. (2010). Numerical analysis of the effects of climatechange on slope stability. PhD thesis, Newcastle University, UK.

Feddes, R. A., Kowaklik, P. J. & Zaradny, H. (1978). Simulation offield water use and crop yield. Chichester, UK: Wiley.

Hight, D. W., McMillan, F., Powell, J. J. M., Jardine, R. J. &Allenou, C. P. (2003). Some characteristics of London clay, InProceedings, Characterisation and Engineering Properties ofNatural Soils (eds T. S. Tan, K. K. Phoon, D. W. Hight & S.Leroueil), pp. 851–907. Rotterdam, Netherlands: Balkema.

Kovacevic, N., Potts, D. M. & Vaughn, P. R. (2001). Progressivefailure in clay embankments due to seasonal climate change.Proc. 15th ICSMGE, Istanbul, 3, 2127–2130.

Lenhard, R. J., Parker, J. C. & Kaluarachchi, J. J. (1991). Compar-ing simulated and experimental hysteretic two-phase transientflow phenomena. Water Resources Research 27, No. 8, 2113–2124.

Li, A. G., Yue, L. G., Tham, C. & Law, K. T. (2005). Field-monitored variations of soil moisture and matric suction in asaprolite slope. Canadian Geotechnical Journal 42, 13–26.

Loveridge, F. A., Spink, T. W. & O’Brien, A. S. (2010). The impactof climate and climate change on infrastructure slopes, withparticular reference to southern England. Quarterly Journal ofEngineering Geology and Hydrogeology 43, 461–472.

Mott MacDonald (2008). Seasonal preparedness earthworks (TSERV567). Packages 3 & 4: Deep dive site modelling. CharingEmbankment Monitoring and Modelling Report. London, UK:Network Rail.

Mott MacDonald (2009). Seasonal preparedness earthworks (TSERV567). Final summary report. Recommendations for improvementto practice. London, UK: Network Rail.

Mott MacDonald (2011). The effects of railway traffic on embank-ment stability. Final Report. RSSB 1386 (Revised). London,UK: Rail Safety and Standards Board.

Mualem, Y. (1976). A new model for predicting the hydraulicconductivity of unsaturated porous media. Water Resource Re-search 12, 513–522.

Nyambayo, V. P., Potts, D. M. & Addenbrooke, T. I. (2004). Theinfluence of permeability on the stability of Embankmentsexperiencing seasonal cyclic pore water pressure changes. Pro-ceedings, Advances in Geotechnical Engineering. The SkemptonConference, London 2, 898–910. London, UK: Thomas Telford.

O’Brien, A. (2007). Rehabilitation of urban railway embankments –investigation, analysis and stabilisation. In Proceedings of the14th European Conference on SMGE, Madrid. (eds V. Cuellar,E. Dapena, A. Gens, J. L. de Justo, C. Oteo, J. M. Rodrigez-Ortiz, C. Sagaseta, P. Sola & A. Sorianao E.), Vol. 1, pp. 125–143. Amsterdam, Netherlands: Millpress.

O’Brien, A., Ellis, E. A. & Russell, D. (2004). Old railwayembankment fill – laboratory experiments, numerical modellingand field behaviour. Proceedings, Advances in GeotechnicalEngineering. The Skempton Conference, London 2, 911–921.London, UK: Thomas Telford.

Ridley, A. M., Dineen, K., Burland, J. B. & Vaughan, P. R. (2003).Soil matrix suction: some examples of its measurement andapplication in geotechnical engineering. Geotechnique 53, 241–253.

Ridley, A., McGinnity, B. & Vaughan, P. (2004). Role of pore waterpressures in embankment stability. Proceedings of the Institutionof Civil Engineers, Geotechnical Engineering GE4, 193–198.

Ritchie, J. T. (1972). Model for predicting evaporation from a row crop

Table A2.1. VADOSE/w input parameters

Soil type Permeability (m/s) Van Genuchten constants Root Depth (m)

AEV (kN/m2) Łs Łr m n Tree Grass

Surface clay fill 2.5 3 10–7 30 0.44 0.1 0.167 1.2 3 m 0.5Clay fill 5 3 10–8 30 0.44 0.1 0.167 1.2 3 m N/AFoundation clay 1 3 10–8 33 0.59 0.1 0.13 1.15 N/A N/AAsh–ballast 4 3 10–5 2 0.45 0.01 0.52 2.1 N/A N/A

Notes:(1) Surface clay fill, 1 m thick.(2) Leaf area index ¼ 2.7 for trees and grass during summer.(3) AEV ¼ air-entry value.



with incomplete cover. Water Resources Research 8, 1204–1212.Scott, J. M., Loveridge, F. & O’Brien, A. S. (2007). Influence of

climate and vegetation on railway embankments, In: Proceedingsof the 14th European Conference on SMGE, Madrid (eds V.Cuellar, E. Dapena, A. Gens, J. L. de Justo, C. Oteo, J. M.Rodrigez-Ortiz, C. Sagaseta, P. Sola & A. Sorianao), pp 659–664. Amsterdam, Netherlands: Millpress.

Skempton, A. W. (1996). Embankments and cuttings on the earlyrailway. Construction History 11, 33–49.

Smethurst, J. A., Clarke, D. & Powrie, W. (2006). Seasonal changesin pore water pressure in a grass covered cut slope in Londonclay. Geotechnique 56, No. 8, 523–537.

Take, W. A & Bolton, M. D. (2004). Identification of seasonal slopebehaviour mechanisms from centrifuge case studies. In: Pro-ceedings, Advances in Geotechnical Engineering. The SkemptonConference, London 2, 992–1004. London, UK: Thomas Tel-ford.

Tratch, D. J., Wilson, G. W. & Fredlund, D. G. (1995). An introduc-

tion to analytical modelling of plant transpiration for geotechni-cal engineers. Proceedings 48th Canadian GeotechnicalConference, Vancouver, pp. 771–780. Richmond, Canada: Bi-Tech.

Van Genuchten, M. T. (1980). A closed form equation for predict-ing the hydraulic conductivity of unsaturated soils. Soil ScienceSociety of America Journal 44, 892–898.

Vaughn, P. R., Kovacevic, N. & Potts, D. M. (2004). Then and now:some comments on the design and analysis of slopes andembankments. In: Proceedings, Advances in Geotechnical En-gineering. The Skempton Conference, London, 1, 241–290.London, UK: Thomas Telford.

Watson, G. V. R., Dykes, T. A., Powrie, W. & Blades, A. P. (2007).The use of CT scanning in geotechnical engineering. Privatecommunication.

Wilson, G., Fredlund, D. & Barbour, S. (1994). Coupled soil-atmo-sphere modeling for soil evaporation. Canadian GeotechnicalJournal 31, 151–161.

32 O’BRIEN


Session 1

Material Characterisation


Tang, A. M. et al. (2011). Geotechnique 61, No. 5, 421–429 [doi: 10.1680/geot.SIP11.005]

35

Effects of the maximum soil aggregates size and cyclic wetting–dryingon the stiffness of a lime-treated clayey soil

A. M. TANG�, M. N. VU� and Y.-J. CUI�

Lime treatment is a well-known technique to improve themechanical response of clayey subgrades of road pave-ments or clayey soils used for embankment. Several stud-ies show that lime treatment significantly modifies thephysical and hydromechanical properties of compactedsoils. Nevertheless, studies on the scale effect underclimatic changes are scarce. Actually, wetting–dryingcycles might significantly modify the microstructure oftreated soils, giving rise to changes in hydromechanicalproperties. This modification could be dependent on thesize of soil aggregates before lime treatment. In thepresent work, this scale effect was studied by investigat-ing the stiffness of a compacted lime-treated clayey soilusing bender elements. The studied soil was first air-driedand ground into a target maximum soil aggregates size(Dmax). For each aggregate size, the soil was humidifiedto reach the target water contents wi, then mixed with3% of lime powder (mass of lime divided by mass ofdried soil) prior to the static compaction at a dry densityof 1.60 Mg/m3. Two initial water contents (wi 14 and18%) and four maximum soil aggregates sizes (Dmax

0.4, 1.0, 2.0 and 5.0 mm) were considered. After thecompaction, the soil specimen (50 mm in diameter and50 mm high) was covered by plastic film in order toprevent soil moisture changes. The soil stiffness was thenmonitored at variable time intervals until reaching stabi-lisation. Afterwards, the soil specimen was subjected tofull saturation followed by air-drying to come back to itsinitial water content. The results show that: (a) the soilstiffness after lime-treatment is significantly dependenton the aggregate size: the finer the aggregates the higherthe soil stiffness; (b) the effect of initial water content onthe stiffness is negligible; and (c) the wetting–dryingcycles seem to slightly increase the soil stiffness in thecase of lime-treated specimens and decrease the soilstiffness in the case of untreated specimens. Furthermore,when an intensive drying was applied reducing the soilwater content lower than the initial level, the soil stiffnessdecreased drastically after the subsequent wetting.

KEYWORDS: fabric/structure of soil; laboratory tests; soilstabilisation; stiffness; suction; time dependence

Le traitement a la chaux est une technique bien connuepour ameliorer les reponses mecaniques des sols rou-tiers ou des sols argileux utilises pour des remblais.Plusieurs etudes ont montre que le traitement a lachaux modifie de facon importante les proprietes physi-ques and hydromecaniques des sols compactes. Nean-moins, l’etude sur l’effet d’echelle sous des sollicitationsclimatiques reste rare. Toutefois, les cycles d’humidifica-tion – sechage pourraient modifier significativement lamicrostructure des sols traites, engendrant des modifica-tions des proprietes hydromecaniques. Ces modificationspourraient dependre de la taille des agregats avant letraitement a la chaux. Dans le present travail, cet effetd’echelle a ete etudie par le suivi de la raideur d’unsol argileux traite a la chaux et compacte en utilisantles elements piezo-ceramiques (bender elements). Le soletudie a ete d’abord seche a l’air et broye a une taillemaximale voulue d’agregat (Dmax). Pour chaque tailled’agregat, le sol a ete humidifie pour atteindre lesteneurs en eau voulues, wi, puis melange avec 3% dechaux en poudre (masse de la chaux divisee par massede sol sec) avant le compactage a une densite seche de1,60 Mg/m3. Deux teneurs en eau initiales (wi 14 et18%) et quatre tailles maximales d’agregats (Dmax

0,4 ; 1,0 ; 2,0 ; et 5,0 mm) ont ete considerees. Apres lecompactage, l’echantillon de sol (50 mm de diametre et50 mm de hauteur) a ete couvert d’un film plastiqueafin d’eviter des modifications de la teneur en eau. Laraideur du sol a ete ensuite suivie a des intervalles detemps variables jusqu’a ce que la stabilisation soitatteinte. Par la suite, l’echantillon de sol a ete soumis aune saturation totale suivie du sechage a l’air pourrevenir a la teneur en eau initiale. Les resultats mon-trent que : (a) la raideur du sol apres le traitement ala chaux est fortement dependante de la taille d’agre-gats : plus les agregats sont petits, plus la raideur estimportante ; (b) l’effet de la teneur en eau initiale surla raideur est negligeable ; et (c) les cycles de humidifi-cation – sechage semblent augmenter legerement laraideur du sol dans le cas avec traitement et diminuerla raideur dans le cas sans traitement. De plus, quandun sechage intensif est applique, diminuant la teneur eneau a un niveau inferieur a la valeur initiale, laraideur du sol diminue radicalement apres l’humidifica-tion qui suit.

INTRODUCTIONLime stabilisation is a well-known technique in civil engi-neering applications such as road construction, embank-ments, slab foundations and piles. After Boardman et al.(2001), adding lime to clayey soils has been shown to lead

to various reactions such as cation exchange, flocculation,carbonation and pozzolanic reaction. When quicklime (CaO)is added into a soil–water system, a highly exothermichydration reaction occurs forming calcium hydroxide(Ca(OH)2). The water consumed in the hydration reaction(and that removed from the soil system by way of evapora-tion) can give rise to significant change in soil hydromecha-nical properties. At the same time, hydration reaction resultsin higher concentration of Ca2þ and OH� ions in the soilpore water. The immediate cation exchanges induce then anapparently dried and more friable material. Alongside the

Manuscript received 1 March 2010; revised manuscript accepted18 November 2010. Published online ahead of print 22 March 2011.Discussion on this paper closes on 1 October 2011, for furtherdetails see p. ii.� Ecole des Ponts ParisTech, UMR Navier/CERMES, Paris, France


cation exchange, reaction also occurs between silica andsome alumina of the lattices of the clay minerals. As aresult, the pozzolanic reactions create hydrated cementationand flocculation by bonding adjacent soil particles together.Such pozzolanic reactions are time dependent, the strengthdeveloping gradually over a long period (Bell, 1996), withthe general effect of improving the soil hydromechanicalproperties. Indeed, the treatment reduces the swelling poten-tial (Tonoz et al., 2003; Al-Rawas et al., 2005), increasesthe shear strength (Bell, 1996; Osinubi & Nwaiwu, 2006;Sivapullaiah et al., 2006; Consoli et al., 2009), increases theelastic modulus (Bell, 1996; Rogers et al., 2006; Sakr et al.,2009) and modifies the compaction properties (Bell, 1996;Osinubi & Nwaiwu, 2006; Consoli et al., 2009; Le Runigoet al., 2011). The water retention properties of clays can bealso modified by the lime treatment (Clare & Cruchley,1957).

Microstructural investigations on lime-treated clays showthat the treatment changes the soil fabric significantly (Caiet al., 2006; Russo et al., 2007; Shi et al., 2007; Le Runigoet al., 2009; Sakr et al., 2009). Moreover, the above studieshave shown that the effects of lime treatment depend onlime content, soil water content, soil type, curing time,temperature, stress state and so on.

Even though numerous studies have been performed toanalyse the effect of lime treatment, almost all works haveinvolved soil specimens prepared in the laboratory, whileinvestigations of lime-treated soil specimens taken from thefield still remain rare. Bozbey & Guler (2006) investigatedthe feasibility of using a lime-treated silty soil as landfillliner material by conducting tests on both laboratory andfield scales. They found that the hydraulic conductivitymeasured on the specimens prepared in the laboratory wasone order of magnitude lower than that of undisturbedsamples taken from the field. Kavak & Akyarh (2007)investigated the improvement of road by lime treatmentbased on both laboratory and field California bearing ratio(CBR) tests. They concluded that the soaked CBR valuesobtained in the laboratory increased significantly (16–21times) 28 days after the treatment while that obtained fromfield CBR tests increased slightly (two-fold). Cuisinier &Deneele (2008) performed suction-controlled oedometer testson soil samples taken from an embankment 3 years after theconstruction. They also performed the same tests on un-treated soil and treated specimens prepared in the laboratory.The results show that the swelling potential of the lime-treated samples taken from the field is significantly largerthan that prepared in the laboratory, but still remains lowerthan that of the untreated samples. They attributed the lossof stabilisation efficiency observed in field conditions to theeffects of drying–wetting cycles related to climatic changes.Rao et al. (2001), Guney et al. (2007) and Khattab et al.(2007) also reported a reduced efficiency of lime treatmentwith wetting–drying cycles.

One of the main reasons explaining the difference be-tween lime-treated soil samples prepared in the laboratoryand the treated soil samples taken from the field could bethe difference in soil aggregates size. Indeed, prior tocompaction in the laboratory, the soil is usually ground at afew millimetres and then mixed with lime. On the contrary,in the field, the dimension of clay clods may reach severalcentimetres before the treatment.

The present work aims at investigating the effects of soilaggregates size on the efficiency of lime treatment undercycles of wetting and drying. For this purpose, air-dried soilswere ground at four values of maximum sieve dimensions(0.4, 1.0, 2.0 and 5.0 mm) prior to lime treatment andcompaction. The shear moduli of soil specimens were mon-itored using the bender elements method. When reaching the

stabilisation of the shear moduli, wetting–drying cycles wereapplied in order to simulate the weathering effects.

SOIL STUDIED AND EXPERIMENTAL TECHNIQUESThe soil used in this study was taken at a site near Tours,

a city in central France. Its main geotechnical properties arereported in Table 1. The grain size distribution, as obtainedby dry sieving method after washing (Afnor, 1996) forelements larger than 80 �m and by hydrometer method(Afnor, 1992) for elements smaller than 80 �m, is shown inFig. 1 (curve ‘After washing’). The curve shows that wash-ing has disaggregated the soil particles into small dimen-sions and almost all soil aggregates became smaller than1 mm, with a clay fraction (, 2 �m) of 26%.

To prepare the soil samples, the air-dried soil was firstground and passed through one of the four target sieve sizes(Dmax ¼ 0.4, 1.0, 2.0 and 5.0 mm). The soil aggregateswhich did not pass through the sieve were ground again.The procedure was repeated until all the soil aggregates,except some large stones, passed through the sieve. Thegrading curves (obtained by sieving) of the four soil sub-series having different maximum soil aggregates diameterare also shown in Fig. 1. This procedure allows the soils tobe prepared with the same mineral composition and variousvalues of soil clusters Dmax. Comparison between the curve‘After washing’ and that of ‘Dmax ¼ 0.4 mm’ shows thatwashing preserved the portion of soil aggregates largerthan 0.4 mm (about 20%), while the preparation of‘Dmax ¼ 0.4 mm’ crushed this portion.

Table 1. Geotechnical properties of the studied soil

Soil properties Value

Liquid limit, wL: % 45Plastic limit, wp: % 21Plasticity index, Ip: % 24Value of blue of methylene, VBS 4.86Carbonates content: % 0.35Specific gravity, GS 2.70

0

10

20

30

40

50

60

70

80

90

100

0·001 0·01 0·1 1 10Grain size: mm

Per

cent

pas

sing

by

wei

ght

After washingDmax 5·0 mm (after grinding)�

Dmax 2·0 mm (after grinding)�



Dry sievingHydrometer method

Fig. 1. Grain size distributions of the studied soil prepared byvarious methods

36 TANG, VU AND CUI


After grinding, the soil was humidified by spraying dis-tilled water to reach prescribed initial moisture contents andsealed in a plastic box for at least 48 h for moisture contenthomogenisation. For each Dmax, 14% and 18% water contentswere considered. Prior to compaction, the moist soil wasthoroughly mixed with lime and then poured into a mould of50 mm diameter. The lime content studied was 3% (mass oflime divided by mass of dried soil). Static compaction wasthen carried out to reach a dry density of 1.60 Mg/m3 with afinal height of the soil specimen of 50 mm. After thecompaction, the soil specimen was taken out of the mouldand wrapped in plastic film in order to avoid any moistureexchange between soil and atmosphere. The initial dimen-sions and the basic properties of the compacted soil speci-men are presented in Table 2.

In Fig. 2 the standard Proctor compaction curves of lime-treated and untreated soils are presented. The after-compac-tion conditions studied are also shown in the figure. It canbe observed that the compaction curve of lime-treated soil isquite different from the untreated one. The dry densitychosen for the present study (1.60 Mg/m3) corresponds tothe maximum dry density of lime-treated soil obtained bythe standard Proctor compaction. The mentioned water con-

tent values (14 and 18%) both correspond to the dry side ofthe compaction curve and were chosen in order to preservethe soil aggregates. Indeed, Delage et al. (1996) showed thatcompaction on the dry side leads to a microstructure char-acterised by an assembly of aggregates, whereas compactionon the wet side leads to a more homogeneous microstructurewithout apparent aggregates.

The bender element was used to monitor the small-strainshear modulus. The experimental set-up is shown in Fig. 3.The soil specimen was put in contact with two benderelements: the transmitter one embedded in the top base andthe receiver one embedded in the bottom base. Both benderelements were connected to a control and data-loggingsystem. A triggered sinusoidal signal was then sent to thetransmitter, recording the response of the receiver at thespecimen base. An example of the time-domain recordscollected is reported in Fig. 4, together with the indicationof travel time (˜t ). Considering the travel length, l, assumedequal to the specimen height (50 mm) minus the protrusionsof the bender transmitter and receiver into the soil specimen(2 mm), the shear wave velocity was then calculated asVs ¼ l/˜t. The soil mass density (r) was verified after eachVs measurement by weighing the soil specimen and was usedfor the determination of the small-strain shear modulus:Gmax ¼ rV 2

s . This experimental technique is similar to thatrecently used by Puppala et al. (2006) when monitoring theshear modulus of chemically treated sulfate-bearing expan-sive soils and previously used by several other authors.

Once the stabilisation of Gmax was reached, the soil speci-men was first wetted, adding water with a sprayer andmonitoring the change in the specimen weight until a watercontent of 21% (corresponding to a degree of saturationSr ¼ 82%) was obtained. Adding more water to the soilspecimen would lead to water drainage from its bottom,indicating that the water content of 21% corresponds to themaximum value that the soil specimen can retain. Afterreaching the target water content value and prior to the Gmax

measurements, the soil specimen was wrapped in plastic film(for at least 24 h) for moisture homogenisation. To achievedrying, the soil specimen was air-dried until the target watercontent was reached. Afterwards, it was wrapped in plasticfilm to achieve ‘water content’ equalisation. During wettingand drying, the water content of the soil specimen (w) wascontrolled by monitoring the changes in its mass (m) by theequation w ¼ (1 + wi) 3 m/mi � 1, where wi and mi are re-spectively the initial water content and the initial mass ofthe soil specimen. The changes in the dimensions of thespecimen, as measured by a caliper after the wetting anddrying stages, were found to be negligible and for thisreason were ignored when calculating both Vs and the massdensity. Five wetting–drying cycles were applied for eachtreated soil specimen. For the untreated specimens, the num-ber of cycles was varied from two to five. The soil specimenwas removed systematically from the testing device after

Table 2. Dimensions and basic properties of soil specimens

Dimensions/basic properties Value

Height: mm 50Diameter: mm 50Dry density: Mg/m3 1.60Initial water content: % 14 and 18Lime content: % 3Maximum soil aggregates size: mm 0.4; 1.0; 2.0; and 5.0

1·40

1·45

1·50

1·55

1·60

1·65

1·70

1·75

1·80

10 15 20 25 30Water content: %

Dry

den

sity

: Mg/

m3

UntreatedTreatedSr 80%�Sr 100%�

Conditions studied

Fig. 2. Standard Proctor compaction curves and conditionsstudied

Signal generatorand dataacquisition system

Top platen

Bender element transmitter

Soil specimen

Bender element receiver

Bottom platen

Fig. 3. Experimental set-up – bender element test

EFFECTS OF MAXIMUM SOIL AGGREGATES SIZE AND CYCLIC WETTING–DRYING 37


each measurement of Vs. The bender elements were alwaysinstalled in the same slots during the measurement of Vs.

Summarising, in this study two moulding water contents(14 and 18%) and four maximum soil aggregates sizes(Dmax ¼ 0.4, 1.0, 2.0 and 5.0 mm) were considered. In eachcase, both treated and untreated specimens were tested.Moreover, for each test three identical specimens wereinvestigated for replicate, corresponding to 48 soil specimensin total.

EXPERIMENTAL RESULTSChanges of small strain shear modulus (Gmax) aftercompaction

After compaction, the small strain shear modulus Gmax

was monitored in order to follow its changes with respect totime. In Fig. 5, the mean value of Gmax (measured on threeidentical specimens) is plotted against time. For the un-treated soil passed through a 0.4 mm sieve (Fig. 5(a)),immediately after the compaction Gmax was equal to 65 MPaand 44 MPa for an initial water content of 14% and 18%respectively. The values increased slightly with time andstabilised after 100 h at 73 MPa and 50 MPa, respectively. Inthe case of treated specimens, Gmax was equal to 108 MPafor wi ¼ 14% and 80 MPa for wi ¼ 18%. Comparison withthe values of untreated specimens shows that the limetreatment has a significant effect on Gmax immediately afterthe compaction. With time, Gmax increased and stabilisedafter 200 h at about 120 MPa for both values of wi. The timeincrease of Gmax was more significant in the case of thehigher water content (wi ¼ 18%). Interestingly, the stabilisedGmax value has been found independent of the initial watercontent.

Similar observation can be made from the results of soilground to 1.0 mm (Fig. 5(b)), 2.0 mm (Fig. 5(c)) and5.0 mm (Fig. 5(d)) in terms of:

(a) immediate effect of lime treatment after compaction,characterised by a significant increase of Gmax

(b) slight increase of Gmax with time for untreated speci-mens

(c) increase of Gmax with time for treated specimens,

especially in the case of the higher water content(wi ¼ 18%)

(d ) stabilisation of Gmax about 200 h after the treatment,with similar final values for both water contents.

In order to analyse the effect of maximum soil aggregatessize on Gmax, the mean final values of Gmax and the rangemeasured on three identical specimens are compared in Fig. 6as a function of Dmax. For the treated soil compacted atwi ¼ 14% (Fig. 6(a)), Gmax was found to be decreasing withDmax, showing the highest value of 120 MPa for Dmax ¼ 0.4 mmand the lowest of 103 MPa for Dmax ¼ 5.0 mm. A similarobservation can be made for the treated soil compacted atwi ¼ 18% (Fig. 6(b)), indicating that the larger the maximumsoil aggregate size, the lower the value of Gmax. For theuntreated specimens, the Gmax plotted against Dmax data show aless clear trend for the wi ¼ 14% case and indicate almostconstant Gmax for the wi ¼ 18% case.

Changes of small strain shear modulus Gmax under cyclicwetting–drying

Cyclic wetting–drying was carried out by controlling thewater content of the soil specimen. In Fig. 7, the Gmax

plotted against time data of the soil aggregates ground to0.4 mm are presented. The corresponding water content ateach measurement is also indicated. The starting points(t ¼ 0) correspond to the last points shown in Fig. 5. At theinitial water content wi ¼ 14% (the corresponding degree ofsaturation is Sri ¼ 55%), Gmax was equal to 73 MPa (Fig.7(a)) for the untreated specimen. Wetting to a water contentof 21% (Sr ¼ 82%) decreased Gmax to 28 MPa. The subse-quent drying to the water content of 14% (at t ¼ 100 h)increased Gmax to 50 MPa. On subsequent wetting and dry-ing cycles, cracks progressively developed until the benderelements signal was no longer transmitted through the soilsample. The soil may be then considered as profoundlyfissured after two wetting–drying cycles. The two followingparameters are proposed to characterise the changes of Gmax

under cyclic wetting–drying for untreated soils:

(a) the decrease of Gmax during the first wetting path,˜Gmax1;

Δt

S-wave arrival

Output signalInput signal

Fig. 4. Response of shear wave for the soil sample treated with 3% of lime, Dmax 5 mm,wi 14%, 886 h after the treatment

38 TANG, VU AND CUI


1 10 100 1000 10000Time: h

(a)

40

60

80

100

120

140S

mal

l str

ain

shea

r m

odul

us,

: MP

aG

max

Sm

all s

trai

n sh

ear

mod

ulus

,: M

Pa

Gm

ax

Sm

all s

trai

n sh

ear

mod

ulus

,: M

Pa

Gm

axS

mal

l str

ain

shea

r m

odul

us,

: MP

aG

max

Treated, 14%wi �

Untreated, 14%wi �

Treated, 18%wi �

Untreated, 18%wi �

1 10 100 1000 10000Time: h

(b)

40

60

80

100

120

140

1 10 100 1000 10000

Time: h(c)

40

60

80

100

120

140

1 10 100 1000 10000

Time: h(d)

40

60

80

100

120

140

Fig. 5. Small-strain shear modulus plotted against time after compaction: (a) Dmax 0.4 mm; (b) Dmax 1.0 mm;Dmax 2.0 mm; (d) Dmax 5.0 mm

0·1 1 10Maximum diameter, : mm

(a)Dmax

40

60

80

100

120

140

Sm

all s

trai

n sh

ear

mod

ulus

,: M

Pa

Gm

ax

0·1 1 10Maximum diameter, : mm

(b)Dmax

Sm

all s

trai

n sh

ear

mod

ulus

,: M

Pa

Gm

ax

Treated soil

Untreated soil

40

60

80

100

120

140

Fig. 6. Small-strain shear modulus after stabilisation plotted against maximum aggregate diameter: (a) wi 14%; (b) wi 18%



(b) the number of wetting–drying cycles causing cracking,Nf .

Note that another parameter ˜Gmax1/Gmax i alternative to˜Gmax1 can be used, where Gmax i is the initial valueobtained in the wetting–drying tests. In the case of theuntreated samples having Dmax ¼ 0.4 mm, these parametersare ˜Gmax1 ¼ 45 MPa (˜Gmax1/Gmax i ¼ 62%), Nf ¼ 2.

For the specimen treated at the initial water content of14% (Fig. 7(a)), wetting to a water content of 21% decreasedGmax slightly from 121 to 114 MPa. Nevertheless, when thishigh value of water content was maintained, Gmax wasincreasing and reached 127 MPa after 100 h. The subsequentwetting and drying only induced slight changes of Gmax. Forthe last drying stage (at t ¼ 580 h), the water content wasfinally reduced to 11%. This intensive drying resulted in asignificant decrease of Gmax when the soil was wetted againto a water content of 21%; Gmax decreased from 134 to97 MPa. This can be explained by the development ofmicrocracks observed on the specimen surface. Beside theparameter ˜Gmax1 described above, the decrease of Gmax

during the last wetting path (˜Gmax f ) can also be proposedto describe the behaviour of treated soils under a wetting–drying path. In the case of the treated samples havingDmax ¼ 0.4 mm, these parameters are ˜Gmax1 ¼ 7 MPa(˜Gmax1/Gmax i ¼ 6%) and ˜Gmax f ¼ 37 MPa.

For the soil specimen compacted at the initial watercontent of 18%, microcracks appeared on the untreatedspecimens after three wetting–drying cycles, leading to nobender elements vibration transmission (Fig. 7(b)). For thetreated specimens, the phenomena were similar to that ob-served for a water content of 14%:

(a) slight increase after the first wetting(b) small changes upon cyclic wetting–drying(c) drastic decrease due to development of microcracks

after an intensive drying with the water contentdecreased to 11%.

The changes of Gmax upon wetting–drying for other maxi-mum soil aggregate sizes show similar trends. The mainparameters of all samples are reported in Table 3, including˜Gmax1/Gmax i. The untreated specimens compacted at adrier state (initial water content of 14%) also show diffusedcracking after three wetting–drying cycles. This is not thecase for the untreated specimens compacted at a wetter state(i.e. wi ¼ 18%) except the specimens having Dmax ¼ 0.4 mm(Fig. 7(b)). For the lime-treated specimens, wetting–dryingcycles only induced small changes of Gmax, becoming sig-nificant only on the wetting stage following an intensivedrying. This effect seems to be more significant for the drierspecimen (i.e. wi ¼ 14%). The effect of Dmax upon cyclicwetting–drying was found to be insignificant as the behav-

Treated

Untreated

(a)

Time: h(b)

1821

2121 18 18 2121 18

21 21 1821 2118

21 21 1811

2121

18

2118

2118

21

18

0

0

200

200

400

400

600

600

800

800

0

0

40

40

80

80

120

120

160

160

Sm

all s

trai

n sh

ear

mod

ulus

,: M

Pa

Gm

axS

mal

l str

ain

shea

r m

odul

us,

: MP

aG

max

1421

2121 14 2121

1421 21

1421 21 14 21 21 14 11

2121

14

21

14

21

14

Treated

Untreated

Fig. 7. Changes in small-strain shear modulus upon cyclic wetting–drying for Dmax 0.4 mm(the corresponding water content is given above each point): (a) wi 14%; (b) wi 18%

40 TANG, VU AND CUI


iour of the soil specimens having different Dmax was quitesimilar. Comparison of the ˜Gmax1/Gmax i data with the˜Gmax1 data indicates that the wetting–drying effect is moreclearly evidenced by the ˜Gmax1/Gmax i parameter with˜Gmax1/Gmax I > 38% for untreated specimens and ˜Gmax1/Gmax i < 12% for treated specimens.

DISCUSSIONThe bender elements method is often used to monitor the

changes in shear wave velocity in triaxial cells under stressconfined conditions. In this case, good contact between thebender elements and the soil specimen can be ensured(Leong et al., 2009; Ng et al., 2009). Application of thismethod is much more difficult in the case of the presentstudy where the evolution of Gmax needs to be monitoredduring several days and on a large number of soil speci-mens. For this reason, in this work the bender elements wereput in contact with the soil specimen only during the meas-urement and no confining pressure was applied. In order toanalyse the test scattering, three specimens were tested foreach Dmax and wi. The results showed a good repeatabilityof the procedure used (see Fig. 7).

The changes of Gmax with time for the compacted soilspecimens were monitored until reaching stabilisation. Aslight increase of Gmax with time was observed for theuntreated specimens and is attributed to ageing effects ofcompacted clay soils. It is worth noting that Delage et al.(2006) observed significant changes in microstructure of acompacted expansive soil after compaction. These authorsattributed the mentioned changes to increase in the intra-aggregate porosity caused by exchange of water between theinter-aggregate and intra-aggregate pores. Tang et al. (2008)also observed this phenomenon characterised by a slightincrease of soil suction after compaction. The effect ofsuction on Gmax was also evidenced in the present study asuntreated specimens which have been compacted at lowerwater contents (higher suction) have higher Gmax. Suchbehaviour was also observed by Sawangsuriya et al. (2008).

For the treated specimens, Gmax obtained immediatelyafter the compaction has been found to be significantlyhigher than that of untreated specimens prepared at the samemoulding water content. The values of Gmax of treated speci-mens range between 80 and 130 MPa. This result is similarto that obtained by Rogers et al. (2006) from cyclic triaxial

tests on compacted clay treated with 2.5% of lime. Theimmediate increase of Gmax with lime treatment can bepartly explained by the increase of suction caused by de-crease of water content after treatment due to hydration andevaporation (Boardman et al., 2001), and to the cationexchanges which increase the flocculation of mineral parti-cles.

It is worth noting that the final values of Gmax aftercompaction are independent of the initial values of watercontent considered (14% and 18%), despite the above-men-tioned effect of initial water content immediately after thetreatment (Fig. 5). As described by Bell (1996), pozzolanicreactions, which take place over a long period, inducebonding between adjacent soil particles. The different evolu-tion over time of Gmax at different water contents (shown inFig. 5) can then be explained as follows: the small-strainshear modulus (Gmax) of compacted soil is mainly governedby the contacts between adjacent soil particles; immediatelyafter the compaction, the contacts are mainly governed bythe capillary suction: the higher the suction (or the lower thewater content) the higher the Gmax; over a long period,owing to pozzolanic reactions in lime-treated soil, cementa-tion develops, gradually increasing Gmax; when stabilisationis reached, Gmax is governed mainly by the cementationbonds and the effect of suction (or water content) becomesless significant. The evolution of the Gmax with time can bealso explained from a microstructural point of view. Russoet al. (2007) studied the time-dependency of the microstruc-ture of lime-stabilised soil samples by means of mercuryintrusion porosimetry (MIP) tests. The results show signifi-cant effects of moulding water content on the pore sizedistribution immediately after the compaction. Nevertheless,after a curing time of 28 days the lime-stabilised samplesshow a very similar pore size distribution, irrespective of themoulding water content adopted. This evolution of micro-structure is also similar to that observed on Gmax in thepresent study.

As far as the effect of the maximum soil aggregate size(Dmax) is concerned, the results of the treated specimensshowed a lower value of Gmax for a larger value of Dmax.This effect was not observed for the untreated specimens.Actually, for a smaller Dmax, the total surface of aggregateswas larger and therefore more soil–lime reaction can beexpected. Note that this observation is of importance from apractical point of view, since laboratory tests are usually

Table 3. Results obtained during the wetting–drying cycles (˜Gmax1: decrease of Gmax during the first wetting path; ˜Gmax1/Gmax i;ratio of ˜Gmax1 to the initial value of Gmax during the wetting–drying tests; Nf : number of cycles inducing cracking; ˜Gmax f :decrease of Gmax during the last wetting path)

Dmax: mm wi: % Lime treatment ˜Gmax1: MPa ˜Gmax1=˜Gmax i: % Nf ˜Gmax f : MPa

0.4 14 Yes 7 6 — 370.4 14 No 45 62 2 —0.4 18 Yes 10 8 — 250.4 18 No 19 38 3 —1.0 14 Yes 8 7 — 361.0 14 No 44 70 3 —1.0 18 Yes 4 4 — 131.0 18 No 19 41 . 52.0 14 Yes 13 12 — 382.0 14 No 36 64 3 —2.0 18 Yes 5 5 — 152.0 18 No 31 63 . 3 —5.0 14 Yes 9 9 — 445.0 14 No 47 73 3 —5.0 18 Yes 4 4 — 225.0 18 No 23 49 . 3 —



performed on small soil aggregates size (less than a fewmillimetres) while in the field they may reach severalcentimetres. As a consequence, particular attention should bepaid when using the parameters determined in the laboratoryfor field application design.

As far as the effects of cyclic wetting–drying on Gmax areconcerned, data presented indicate that wetting induced adecrease and drying induced an increase of Gmax. This canbe explained by the effect of suction: wetting decreased thesoil suction and thus the soil stiffness, while drying in-creased the soil suction and thus the soil stiffness. The samephenomenon was observed by Ng et al. (2009) whenperforming measurements of Gmax in a suction-controlledtriaxial cell. Vassallo et al. (2007a) used a suction-controlledresonant column torsional shear cell to study the effect ofnet stress and suction history on Gmax of a compacted clayeysilt. The experimental results show that Gmax depends sig-nificantly on mean net stress and matric suction as well asstress history. Modelling criteria were proposed by Vassalloet al. (2007b) to describe the observed soil behaviour. In thepresent work, comparison between the treated specimens anduntreated specimens shows that the effect of suction changeon Gmax of treated specimens is less significant than that ofuntreated specimens. This shows that lime treatment rein-forces the soil and makes it less sensitive to ‘weathering’.

The significant decrease of Gmax during the last wettingpath after an intensive drying (up to a water content of11%) observed in this work could be explained within theframework of unsaturated soil mechanics, where the soilsuction is usually considered as a stress variable. Actually,after the lime treatment, cementation bonds were progres-sively created under a humidity condition corresponding tothe initial water content (14% or 18%). During the firstwetting–drying cycle, the water content was varied in therange from wi to the maximum value (21%). Thus, the soilsuction remains lower than its maximum value (reached afterthe stabilisation of Gmax). The soil state moves inside the‘elastic’ zone (see Alonso et al. 1990; Cui & Delage, 1996)and the bonds between particles are relatively well pre-served. On the contrary, on intensive drying (water contentdecreased to 11%), the soil suction exceeded its maximumvalue: large elastoplastic deformations and significant dam-age of bonds are thus induced as for the case of Pinyol etal. (2007). This damage of bonds would result in micro-cracking and therefore in decrease of Gmax. On the otherhand, the drying increased the soil suction, thus increasingGmax. The fact that a slight increase of Gmax was observedduring the drying shows that the suction effect prevailed onthe bond damage. During the subsequent wetting, as thesuction effect was removed, the damage effect was finallyevidenced. It is also interesting to note that the decrease ofGmax for treated samples during the last wetting path issimilar to the decrease observed on untreated samples fromthe beginning of the wetting–drying cycles. This seems toconfirm the onset of severe bond damage of the treatedsamples on intensive drying.

For the untreated specimens, it has been observed thatwetting–drying cycles resulted in a decrease of Gmax, espe-cially for drier specimens (see Fig. 7 and Table 3). This canbe explained by the generation of microcracks by cyclicwetting–drying under unconfined conditions (Yesiller et al.,2000). In the works of Vassallo et al. (2007a) and Ng et al.(2009), the generation of microcracks was avoided as thetests were performed under confined conditions. For thetreated specimens, the first wetting path equally induced adecrease of Gmax. Nevertheless, after this decrease, the valueof Gmax increased slightly. The immediate decrease of Gmax

can be explained by the effect of suction, while the subse-quent increase of Gmax after wetting can be attributed to the

onset of various reactions by water addition. This explainswhy Kavak & Akyarh (2007) recommended watering thelime-treated soil one week after the treatment.

CONCLUSIONSThe small strain shear modulus Gmax of compacted lime-

treated soil was investigated using bender elements. Thefollowing conclusions can be drawn.

(a) The lime treatment significantly increases Gmax of thesoil, giving rise to Gmax values independent of themoulding water content about 200 h after lime treat-ment.

(b) For the four maximum soil aggregate sizes Dmax

considered, it has been observed that the larger is thevalue of Dmax the lower is the value of Gmax. Thisobservation is interesting from a practical point of viewfor earthworks involving lime-treated soils. Indeed, theresults obtained show that designing earthworks basedon the parameters determined from laboratory tests canbe misleading, because the maximum aggregate size ofthe soil tested in the laboratory is usually less than afew millimetres while clay aggregates in the field mayreach the dimension of several centimetres.

(c) Owing to the appearance of microcracks, cyclicwetting–drying induced significant decrease of Gmax

of untreated specimens. For treated specimens, thechanges of Gmax during wetting–drying cycles are lesssignificant. Only an intensive drying to water contentvery much lower than the initial value can inducemicrocracks and thus a decrease in Gmax.

(d ) For the treated specimens, only the first wetting induceda decrease of Gmax. On the contrary, the subsequentcycles induced a slight increase of Gmax. If the decreasedue to wetting can be explained by the suction effect,the slight increase by further wetting–drying cycleshould be attributed to the onset of various physico-chemical reactions within the soil.

ACKNOWLEDGEMENTSThe authors would like to express their great appreciation

to the French National Research Agency for funding thepresent study, which is part of the project Terdouest ‘Sus-tainable earthworks involving treated soils’ – ANR-07PGCU-006. The technical support of Mr Dong Jucai is alsogratefully acknowledged.

REFERENCESAfnor (1992). Soil investigation and testing – granulometric analy-

sis – hydrometer method. French standard NF P 94-057 (inFrench).

Afnor (1996). Soil investigation and testing – granulometric analy-sis – dry sieving method after washing. French standard NF P94-056 (in French).

Alonso, E. E., Gens, A. & Josa, A. (1990). A constitutive modelfor partially saturated soils. Geotechnique 40, No. 3, 405–430,doi: 10.1680/geot.1990.40.3.405.

Al-Rawas, A. A., Hago, A. W. & Al-Sarmi, H. (2005). Effect oflime, cement and Sarooj (artificial pozzolan) on the swellingpotential of an expansive soil from Oman. Building Environ. 40,No 5, 681–687.

Bell, F. G. (1996). Lime stabilization of clay minerals and soils.Engng Geol. 42, No. 4, 223–237.

Boardman, D. I., Glendinning, S. & Rogers, C. D. F. (2001).Development of stabilisation and solidification in lime–claymixes. Geotechnique 51, No. 6, 533–543, doi: 10.1680/geot.2001.51.6.533.

42 TANG, VU AND CUI


Bozbey, I. & Guler, E. (2006). Laboratory and field testing forutilization of an excavated soil as landfill liner material. WasteMgmnt 26, No. 11, 1277–1286.

Cai, Y., Shi, B., Ng, C. W. W. & Tang, C. S. (2006). Effectof polypropylene fibre and lime admixture on engineeringproperties of clayey soil. Engng Geol. 87, Nos. 3–4, 230–240.

Clare, K. E. & Cruchley, A. E. (1957). Laboratory experiments inthe stabilization of clays with hydrated lime. Geotechnique 7,No. 2, 97–111, doi: 10.1680/geot.1957.7.2.97.

Consoli, N. C., Lopes, L. S. & Heineck, K. S. (2009). Keyparameters for the strength control of lime stabilized soils. J.Mater. Civ. Engng 21, No. 5, 210–216.

Cui, Y. J. & Delage, P. (1996). Yielding and plastic behaviour of anunsaturated compacted silt. Geotechnique 46, No. 2, 291–311,doi: 10.1680/geot.1996.46.2.291.

Cuisinier, O. & Deneele, D. (2008). Impact of cyclic wetting anddrying on the swelling properties of a lime-treated expansiveclay. Journees Nationales de Geotechnique et de Geologie del’Ingenieur JNGG’08, Nantes, pp. 18–20.

Delage, P., Audiguier, M., Cui,Y. J. & Howat, M. (1996). Micro-structure of a compacted silty clay. Can. Geotech. J. 33, No. 1,150–158.

Delage, P., Marcial, D., Cui, J. Y. & Ruiz, X. (2006). Ageing effectsin a compacted bentonite: a microstructure approach. Geotechni-que 56, No. 5, 291–304, doi: 10.1680/geot.2006.56.5.291.

Guney, Y., Sarib, D., Cetinc, M. & Tuncan, M. (2007). Impact ofcyclic wetting–drying on swelling behavior of lime-stabilizedsoil. Building Environ. 42, No. 2, 681–688.

Kavak, A. & Akyarh, A. (2007). A field application for limestabilization. Environ. Geol. 51, No. 6, 987–997.

Khattab, S. A. A., Al-Mukhtar, M. & Fleureau, J. M. (2007). Long-term stability characteristics of a lime-treated plastic soil.J. Mater. Civ. Engng 19, No. 4, 358–366.

Le Runigo, B., Cuisinier, O., Cui, Y. J., Ferber, V. & Deneele, D.(2009). Impact of initial state on the fabric and permeability ofa lime-treated silt under long-term leaching. Can. Geotech. J.46, No. 11, 1243–1257.

Le Runigo, B., Ferber, V., Cui, Y. J., Cuisinier, O. & Deneele, D.(2011). Performance of lime-treated silty soil under long-termhydraulic conditions. Engng Geol. 118, No. 1–2, 20–28.

Leong, E. C., Cahyadi, J. & Rahardjo, H. (2009). Measuring shearand compression wave velocities of soil using bender–extenderelement. Can. Geotech. J. 46, No. 7, 792–819.

Ng, C. W. W., Xu, J. & Yung, S. Y. (2009). Effect of wetting–drying and stress ratio on anisotropic stiffness of an unsaturatedsoil at very small strain. Can. Geotech. J. 46, No. 9, 1062–1076.

Osinubi, K. J. & Nwaiwu, G. M. O. (2006). Compaction delay

effects on properties of lime-treated soil. J. Mater. Civ. Engng18, No. 2, 250–258.

Pinyol, N., Vaunat, J. & Alonso, E. E. (2007). A constitutive modelfor soft clayey rocks that includes weathering effects. Geotechni-que 57, No. 2, 137–151, doi: 10.1680/geot.2007.57.2.137.

Puppala, A. J., Kadam, R., Madhyannapu, R. S. & Hoyos, L. R.(2006). Small-strain shear moduli of chemically stabilized sul-fate-bearing cohesive soils. J. Mater. Civ. Engng 132, No. 3,322–336.

Rao, S. M., Reddy, B. V. V. & Muttharam, M. (2001). The impactof cyclic wetting–drying on the swelling behaviour of stabilizedexpansive soils. Engng Geol. 60, Nos. 1–4, 223–233.

Rogers, C. D. F., Boardman, D. I. & Papadimitriou, G. (2006).Stress path testing of realistically cured lime and lime/cementstabilized clay. J. Mater. Civ. Engng 18, No. 2, 259–266.

Russo, G., Vecchio, S. D. & Mascolo, G. (2007). Microstructure ofa lime stabilised compacted silt. Expl Unsaturated Soil Mech. –Springer Proc. Physics 112, No. 1, 49–56.

Sakr, M. A., Shahin, M. A. & Metwally, Y. M. (2009). Utilizationof lime for stabilizing soft clay soil of high organic content.Geotech. Geol. Engng 27, No. 1, 105–113.

Sawangsuriya, A., Edil, B. T. & Bosscher, P. J. (2008). Modulus–suction–moisture relationship for compacted soil. Can. Geotech.J. 45, No. 7, 973–983.

Shi, B., Liu, Z., Cai, Y. & Zhang, X. (2007). Micropore structureof aggregates in treated soils. J. Mater. Civ. Engng 19, No. 1,99–104.

Sivapullaiah, P. V., Sridharan, A. & Ramesh, H. N. (2006). Effect ofsulphate on the shear strength of limetreated kaolinitic soil.Ground Improvement 10, No. 1, 23–30.

Tang, A. M., Cui, Y. J. & Barnel, N. (2008). Compression-inducedsuction change in a compacted expansive clay. In Unsaturatedsoils: advances in geo-engineering (eds D. G. Toll, C. E.Augarde, D. Gallipoli and S. J. Wheeler), Proc. 1st Eur. Conf.Unsaturated Soils, Durham, UK 1, 369–374. Leiden: CRCPress/Balkema.

Tonoz, M. C., Gokceoglu, C. & Ulusay, R. (2003). A laboratory-scale experimental investigation on the performance of limecolumns in expansive Ankara (Turkey) clay. Bull. Engng Geol.Env. 62, No. 2, 91–106.

Vassallo, R., Mancuso, C. & Vinale, F. (2007a). Effect of net stressand suction history on the small strain stiffness of a compactedclayey silt. Can. Geo. J. 44, No. 4, 447–462.

Vassallo, R., Mancuso, C. & Vinale, F. (2007b). Modelling theinfluence of stress–strain history on the initial shear stiffness ofan unsaturated compacted silt. Can. Geo. J. 44, No. 4, 463–472.

Yesiller, N., Miller, C. J., Inci, G. & Yaldo, K. (2000). Dessicationand cracking behavior of three compacted landfill liner soils.Engng Geol. 57, Nos 1–2, 105–121.



Taibi, S. et al. (2011). Geotechnique 61, No. 5, 431–437 [doi: 10.1680/geot.SIP11.P.020]

45

TECHNICAL NOTE

Some aspects of the behaviour of compacted soils along wetting paths

S. TAIBI�, J. M. FLEUREAU†, N. ABOU-BEKR‡, M. I . ZERHOUNI§, A. BENCHOUK‡,K. LACHGUEUR‡ and H. SOULI¶

Wetting and oedometric loading tests have been per-formed on several clayey compacted soils. The resultshighlight the influence of compaction water content andcompaction stress on wetting paths. Comparing thechanges in degree of saturation induced by mechanicalloading and hydraulic loading (wetting path under nullstress), it may be noticed that the oedometric path has anopposite curve of that of the wetting path, due to the factthat the void ratio decreases under constant water con-tent when the stress is increased in one case (compres-sion) and increases with water content in the other(swelling).

KEYWORDS: clays; compaction

Des essais d’humidification et oedometriques ont eterealises sur differents sols argileux compactes. Les resul-tats mettent en evidence l’influence de la teneur en eau etla contrainte de compactage sur les chemins d’humidifi-cation. En comparant les variations du degre de satura-tion induites par le chargement mecanique et lechargement hydrique (humidification sous contraintenulle), on note que le chemin oedometrique presente unecourbe opposee a celle du chemin d’humidification, ceciest du au fait que l’indice des vides decroit a teneur eneau constante lorsque la contrainte augmente dans uncas (compression) et augmente avec la teneur en eau dansl’autre (gonflement).

INTRODUCTIONCompacted soils are commonly used in the construction ofsoil structures such as roads, railroad embankments andearth dams. Several investigators have highlighted the influ-ence of the hydromechanical history on the drying–wettingresponse of compacted soils. Also, a large number of factors,which are either not measured or difficult to control, influ-ence the engineering behaviour of compacted soils (Guillotet al., 2001; Alonso & Pinyol, 2008). Several studies havebeen carried out to investigate the influence of compactionstress and compaction water content on the behaviour ofunsaturated clayey soils (Fleureau et al., 2002; Tarantino &Tombolato, 2005; Sun et al., 2006; Brown & Sivakumar,2008; Tang et al., 2008; Birle et al., 2008; Tarantino & DeCol, 2008). Compaction at different water contents results indifferent fabrics of the soil (Ahmed et al., 1974; Gens et al.,1995; Delage et al., 1996; Vanapalli et al., 1999). Afterobserving with a scanning electron microscope (SEM) thearrangement of grains within compacted specimens, Delageet al. (1996) concluded that on the dry side of optimum, awell-developed granular aggregate structure with interaggre-gate porosity is visible. The clayey fraction forms grain jointinfills (aggregate microstructure with bimodal particle sizedistribution (PSD), macro- and micropores). On the otherhand, on the wet side of optimum, a structure of well-developed wetter clay forming a matrix that envelopes thesilt grains and fills the intergranular voids is observed(matrix microstructure with monomodal PSD, mainly micro-pores). Vanapalli et al. (1999) showed that the desaturation

curve of a specimen compacted dry of optimum is notice-ably different from that of a specimen compacted at opti-mum or wet of optimum. At the same suction, the degree ofsaturation of the specimen compacted dry of optimum issomewhat lower than that of the two others, which is inagreement with the more open structure revealed by SEMobservations. Fleureau et al. (2002) showed that for kaolin–sand and sand–clay mixtures, the slope of the line on thedry side of optimum in the (w, log s) coordinate system, wassmaller than the slope on the wet side, as an effect of thedecrease in density below the optimum. Tarantino & Tombo-lato (2005) showed that post-compaction suction of clayeyspecimens compacted at high water contents increased withincreasing degree of saturation. This behaviour was ex-plained in a qualitative fashion by invoking the couplingbetween mechanical and water retention behaviour occurringduring compaction. Sivakumar et al. (2006) showed that therelationship between specific water volume and suction wasunaffected by the compaction effort at suction values higherthan 100 kPa, whereas Birle et al. (2008) showed that thesoil–water retention curves in terms of gravimetric watercontent were independent of the initial dry density. Recently,post-compaction states of samples compacted on the dry sideof optimum over a wide range of water contents and verticalstresses have been investigated by Tarantino & De Col(2008), and three water content regions were identified. Asthe degree of saturation is increased at constant watercontent by the compaction process, post-compaction suctionincreases at higher water contents (region I), decreases atmedium water contents (region II) and remains constant atlower water contents. The authors formulated a coupledmechanical water retention model by combining features ofthe models presented by Wheeler et al. (2003) and Gallipoliet al. (2003b). The water retention model was formulatedaccording to Gallipoli et al. (2003a).

This technical note presents some experimental resultsobtained on free wetting paths and from unsaturated oedo-metric loading tests performed on four compacted clayeysoils coming from cores of earth dams, in relation tocompaction water content and compaction stress.

Manuscript received 21 March 2010; revised manuscript accepted22 December 2010. Published online ahead of print 22 March 2011.Discussion on this paper closes on 1 October 2011, for furtherdetails see p. ii.� Laboratoire d’Ondes & Milieux Complexes, Universite du Havre,France† Laboratoire MSS-Mat, Ecole Centrale Paris, France‡ Laboratoire EOLE, Universite de Tlemcen, Algeria§ FONDASOL, France¶ Laboratoire LTDS, Ecole Nationale d’ingenieurs de Saint Etienne,France


MATERIALS AND TESTING METHODSMaterials

The materials come from the cores of four earth dams inthe south of France (Vieux-pre and La Verne dams) andnorthwestern Algeria (Sikkak and Boughrara dams). The LaVerne material is an eroded microshist containing illite,kaolinite and chlorite traces. According to USCS/LCPCclassification, these materials are not very plastic (Ap). LaVerne material is characterised by a small percentage offines (10% of the material is made of particles of diameterbelow 80 �m), whereas the other three materials are primar-ily made up of fines (78% of the material is made ofparticles of diameter below 80 �m). Table 1 summarises theproperties of the four materials.

The materials coming from the cores were passed through asieve of 5 mm (square mesh) in order to remove the largeelements, and then dried in the oven. From these materialpowders, various reconstitutions of samples were carried out.For compacted samples, the powder was wetted to a targetedwater content. The wet powder was packed in a tight bag andcarefully preserved in a moisture-controlled room during 24 h.

The soils were compacted either using the standard Proctorprocedure or, in the case of Vieux-pre material, under quasi-static conditions using three different compaction stresses.Fig. 1 shows the compaction curve and changes in suctionwith water content for La Verne material: suction appears tobe a linear function of water content in a large domain onboth sides of the optimum. Fleureau et al. (2002) showed thatin most tests, the slope of the line on the dry side is smallerthan the slope on the wet side, as an effect of the decrease indensity below the optimum. This is not observed for LaVerne material, probably due to the scarcity of experimentaldata on the dry side (two experimental data only).

Testing methodsSeveral methods were used to control or measure suction

in the samples along the free drying paths (Kassif & BenShalom, 1971; Fleureau et al., 1993; Delage et al., 1992;1998): Sintered glass tensiometric plates were used forsuctions lower than 20 kPa; both air pressure and osmotic

techniques were used for suctions ranging from 0.05 to8 MPa, and salt solution desiccators for higher suctions(from 8 to 1000 MPa). Suction measurements were alsomade by means of the filter paper technique (ASTM D5298-94, ASTM (1995)) and with thermocouple psychrom-eters (Zerhouni, 1995). Oedometer tests were carried out onunsaturated compacted samples in which the piston wasequipped with a psychrometric probe in order to measure

Table 1. Main properties of the soils used in the study

Material wL IP ,80 �m: % ,2 �m: % d10: �m d60: �m ªs=ªw wSPO: % (ªd=ªw)SPO

Vieux-pre 32 13 92 36 — 5 2.75 18 1.72La Verne 35 16 10 ,2 80 700 2.64 16.5 1.79Boughrara 48 28 90 52 0.2 8.5 2.6 21 1.62Sikkak 50 27 78 45 — 40 2.64 16 1.55

Standard NF P94-051(AFNOR, 1993)

NF P94-057(AFNOR, 1992)

NF XP P94-041(AFNOR, 1995)

NF P94-054(AFNOR, 1991)

NF P94-093(AFNOR, 1999)

19

18

17

16

Dry

uni

t wei

ght:

kN/m

3

10 15 20 25

La Vernematerial

35%wL �

Water content: %

Sr 0·6� Sr 0·8�

Sr 1�

Suc

tion:

kP

a

103

102

101

100

10�1

Proctor comp.

Wetting path

10 15 20 25Water content: %

Fig. 1. Compaction curve and changes in suction with watercontent for La Verne material

Table 2. Equipment used and tests done for each material

Material Free wetting paths Oedometertests

Methods used to control suction Methods used to measure suction

Sintered glasstensiometric

plates,s , 20 kPa

Air pressuretechnique,

0.05 , s , 1.5:MPa

Osmotictechnique,

0.05 , s , 8:MPa

Salt solutiondesiccators,

8 , s , 1000:MPa

Filter papertechnique

Thermocouplepsychrometers

Equipped with apsychrometric

probe

Boughrara — — X X X — —La Verne X X X X X — —Sikkak — — X X X — —Vieux-pre X X X X X X X

46 TAIBI, FLEUREAU, ABOU-BEKR, ZERHOUNI, BENCHOUK, LACHGUEUR AND SOULI


the change in suction during the mechanical loading. Table2 summarises the equipment used and the tests carried outfor each material.

RESULTS AND DISCUSSIONWetting paths

To study the influence of compaction water content,wetting tests were performed on dam materials compacted

under standard Proctor conditions at different water contentson both sides of the optimum: wspo, wspo + 2% andwspo � 2% for Sikkak and Boughrara materials, wspo,wspo + 2% and wspo � 3% for La Verne material, with wspo

the standard Proctor optimum (SPO) water content.The results are shown in Fig. 2. In the [log(s), e] plane

(Figs 2(a) (right), 2(b) (right) and 2(c) (right)) there is somescatter in the data but the points generally appear close to astraight line, independently from the initial compaction water

100

100

100

90

90

90

80

80

80

70

70

70

60

60

60

50

50

50

Deg

ree

ofsa

tura

tion:

%D

egre

e of

satu

ratio

n: %

Deg

ree

ofsa

tura

tion:

%

Sikkak dam material

La Verne dam material

Boughrara dam material

Boughrara dam material

La Verne dam material

Sikkak dam material

Comp. SPO 2%�

Comp. SPO 2%�

Comp. SPO 2%�

Comp. SPO 2%�

Comp. SPO 2%�

Comp. SPO 2%�

Comp. SPO

Comp. SPO

Comp. SPO

Comp. SPO

Comp. SPO

Comp. SPO

Comp. SPO 2%�

Comp. SPO 3%�

Comp. SPO 2%�

Comp. SPO 2%�

Comp. SPO 3%�

Comp. SPO 2%�

101 101103 103105 105

Suction: kPa

Suction: kPa

Suction: kPa Suction: kPa

Suction: kPa

Suction: kPa(a)

1·2

1·2

1·0

1·0

0·8

0·8

0·8

0·6

0·6

0·6

0·4

0·4

0·4

0·2

Voi

d ra

tioV

oid

ratio

Voi

d ra

tio

10�1 10�1101 101103 103105 105

(b)

(c)

101101102

102103103104

104

Fig. 2. Influence of compaction water content on the wetting paths of: (a) Sikkak; (b) La Verne; and(c) Boughrara materials in the [log(s), Sr] plane (left-hand side) and [log(s), e] planes (right-handside)

BEHAVIOUR OF COMPACTED SOILS ALONG WETTING PATHS 47


content. For water contents larger than the optimum orslightly smaller, the difference between the three pathsremains very small and it seems that for limited variationsin initial water content, the compaction water content playsno significant role in the relationship between void ratio andsuction.

In the [log(s), Sr] plane (Figs 2(a) (left), 2(b) (left) and2(c) (left)), a difference appears between the wetting paths,the path corresponding to the specimens compacted at theoptimum being located between those of the specimens atwspo � 3% and wspo + 2%. The influence of compactionwater content is well marked for degrees of saturationranging between 70 and 90%. Indeed, for a suction of1000 kPa for example, the difference in the degrees ofsaturation reaches approximately 20% between the samplescompacted at wspo + 2% and wspo � 2%: However, in thecase of Boughrara dam material, this influence is lessimportant. Furthermore, the experiments show that the influ-ence of the compaction water content on the wetting paths isnegligible when the degree of saturation falls below a certainvalue (Sr < 50%). This result is consistent with those ob-tained by Vanapalli et al. (1999) and Thakur et al. (2005).

Interesting remarks can be made about the shape of thecurves and, in particular about the point at which the slopeof the curves changes. A typical wetting curve is presentedin Fig. 3(a), which shows that a variation in slope occurswhen the degree of saturation is higher than 90%. Indeed,the wetting path of unsaturated soils can be described as atwo-stage process, consisting of primary and residual wet-ting. During primary wetting (Sr < 80%), the water and airphases are continuous and the wetting path corresponds tothe filling of the pores by water which replaces the air.When the degree of saturation exceeds 80% (residual wet-ting), the air phase becomes discontinuous. The air formsisolated bubbles within the voids and the degree of satura-tion varies very little. The limit between these two wettingstages corresponds to a suction which can be called ‘suctionof quasi-saturation’. The experiments show that this suctionis influenced by the compaction water content and varieslinearly in a semi-logarithmic plan (Fig. 3(b)).

To study the influence of the compaction stress on thewetting process, wetting tests were performed on Vieux-prematerial compacted at the SPO water content wspo ¼ 18%, to0.5, 1 or 5 MPa. The results are shown in Fig. 4 in the[log(s), Sr] and [log(s), e] planes. In the [log(s), e] plane, itis observed that on wetting paths, the higher the compactionstress, the more significant the swelling of the specimen. Inaddition, it can be noticed in the [log s, Sr] plane that, onthe wetting paths, the suction of resaturation increases withthe compaction stress (Fig. 5).

Fleureau et al. (2002) proposed correlations between thestandard or modified Proctor optimum water content andmaximum density of clayey soils (in the range betweenwL ¼ 25 and wL ¼ 170%), and their liquid limit. At SPO,the relations between the liquid limit and the optimum watercontent (in percent) and the maximum dry unit weight (inkN/m3), are as follows (with the liquid limit in percent)

wspo ¼ 1:99þ 0:46wL � 0:0012w2L (1)

ªdspo ¼ 21� 0:113wL þ 0:00024w2L (2)

Table 3 compares these parameters measured for the testedmaterials and calculated using these equations. There is agood agreement between the data and the predictions ac-cording to equations (1) and (2) for the optimum watercontent and the maximum dry unit weight. Fleureau et al.(2002) also showed that the wetting paths of compacted soilswere generally linear in the [log(s), e] and [log(s), w]coordinate systems, under the conditions investigated. Corre-

lations between the slopes of these lines, also called swellingindices, with respect to void ratio, Cms, and water content,Dms, in relation to liquid limit for specimens compacted atSPO were established by Fleureau et al. (2002). The equa-tions are as follows (with the liquid limit in percent)

Cms ¼�˜e

˜ log(s)½ � ¼ þ0:029� 0:0081wL þ 5:10�6w2L (3)

Dms ¼�˜w

˜ log(s)½ � ¼ �0:54� 0:030wL þ 3:3:10�6w2L (4)

Figure 6 shows the normalised water contents w/Dms (withDms calculated as a function of the liquid limit according toequation (4)) plotted against suction for the studied materi-als. The graph shows the validity of the liquid limit as aprimary classification parameter for fine-grained clayey soils,without excluding the use of other parameters that wouldreduce the scatter of the data.

Mechanical undrained loadingOedometer tests were carried out on unsaturated Vieux-

pre samples, compacted to 1 MPa, corresponding to a degreeof saturation of about 87% (Fig. 7). The variation of suction

Deg

ree

ofsa

tura

tion:

%

100

90

80

70

60

50

Suction ofquasi-saturation

Suction (logarithmic scale)(a)

Com

pact

ion

wa

ter

cont

ent:

%

25

20

15

10

10�3 10�2 10�1 100 101

Suction of quasi-saturation: MPa(b)

Dam materialsSikkakLa VerneBoughrar

CompactedSPO

CompactedSPO

CompactedSPO

Fig. 3. Influence of compaction water content on the suctionof quasi-saturation. (a) Typical wetting path of compacted soil;(b) variation of compaction water content plotted against quasi-saturation suction for the investigated materials



during the mechanical loading was measured with a psychro-metric probe. In the [log(�v), e] plane, the normally con-solidated curve derived from tests carried out on a slurry ofthe same material is also plotted (i.e. specimens prepared ata water content equal to 1.5 times the liquid limit). Twodomains of behaviour are observed: first, the void ratiovaries very slightly, until reaching the yield stress due tocompaction. Indeed, if an analogy is made with the over-consolidated behaviour of saturated materials, the corre-

sponding consolidation total stress, at the intersect betweenthe normally consolidated (NC) consolidation line and thetangent to the loading path, is about 1300 kPa. As soon asthis stress is exceeded, larger compressibility of the materialis observed, with the curve tending towards that of thenormally consolidated reconstituted material. At the sametime, suction decreases significantly during loading (Fig.7(b)), down to a value close to 0. It has already been

Vieux-pré dam materialwetting paths


Compaction stress

Compaction stress

0·5 MPa

0·5 MPa

1 MPa

1 MPa

5 MPa

5 MPa

Slurry

Slurry

1·0

0·8

0·6

0·4

Voi

d ra

tio

Wetting path(on dried slurry)

Wetting path(on dried slurry)

10�1

10�1

100

100

101

101

102

102

103

103

104

104

Suction: kPa

Suction: kPa

100

90

80

70

60

50

Deg

ree

ofsa

tura

tion:

%

Fig. 4. Influence of compaction stress on the wetting paths ofVieux-pre dam material

10·0

1·0

0·1

Com

pact

ion

stre

ss: M

Pa


1 10 100Suction of resaturation: kPa

Fig. 5. Compaction stress plotted against suction of resaturationfor Vieux-pre dam material

Table 3. Comparison between compaction soil properties at SPO and correlations withliquid limits

Material wL: % wSPO: % ªd SPO: kN/m3ªdSPO

Experiment Correlation Experiment Correlation

Boughrara dam 48 21 21.3 16.2 16.1Sikkak dam 50 16 22 15.5 15.9Vieux-pre dam 32 18 15.5 17.2 17.6La Verne dam 35 16.5 16.6 17.9 17.3

25

20

15

10

5

0

Nor

mal

ised

wa

ter

cont

ent:

%

w D/ ms Boughrara (SPO)Sikkak (SPO)La Verne (SPO)Vieux-pré (comp. 1 MPa)

Suction: kPa10�1 101 103 105

Fig. 6. Normalised water contents of soils compacted at theProctor optimum



pointed out that the lower limit for measuring suction with apsychrometer is practically about 300 kPa (Zerhouni, 1995).However, only the suction values above 150 kPa were con-sidered acceptable to interpret the test. Below this value, thedispersion and the discrepancy of the psychrometric meas-urements become too large. In addition, during the cycles,

suction increases during unloading. The slopes of the un-loading–reloading paths do not correspond to those of thesaturated material.

In order to highlight the difference between the twowetting paths (the free wetting and the oedometric paths)performed on the soil from the same initial state, thechanges in void ratio and degree of saturation were plottedin Fig. 8 against the effective vertical stress � 9v: For simpli-city, Terzaghi’s effective stress definition was used, consider-ing that the degree of saturation was larger than 85% inboth cases and that the use of this definition would notresult in too large an error. Fig. 8 shows how the twowetting paths lead the soil to saturation through oppositeloading histories for the soil skeleton. The second conclusionthat can be drawn from this comparison is the importance ofcarrying out undrained oedometric tests, as complements tothe usually performed drained (suction-controlled) tests, asthey provide additional information on the behaviour of thematerial, which can be useful for its modelling.

CONCLUSIONSThe aim of this note was to present some aspects of the

behaviour of laboratory compacted dam materials duringwetting. Some interesting features have been highlighted.Soils compacted at water contents around the optimum seemto follow the same wetting path as those compacted at theoptimum in the [log(s), e] plane. However, a differenceappears between the wetting paths in the [log(s), Sr] plane,the path corresponding to the specimens compacted at theoptimum being located between those of the specimenscompacted at wspo � 3% and wspo + 2%. As a confirmationof existing knowledge concerning the influence of compac-tion stress on wetting paths, it appears that the higher thecompaction stress, the larger the swelling of the soil, andthat the suction of saturation increases with the compactionstress. Correlations between the liquid limit and the mainproperties of the compacted soils show a good agreementfor the tested materials in the normalised water contentsagainst suction plane. Comparing the changes in void ratioinduced by mechanical loading (oedometric loading at con-stant water content) and hydraulic loading (wetting pathunder null stress), the experimental evidence is provided ofhow the oedometric path has an opposite effect to that of

0·8

0·6

0·4

Voi

d ra

tio

101

101

101

102

102

102

103

103

103

Vertical total stress: kPa(a)

Vertical total stress: kPa(b)

Vertical total stress: kPa(c)

NC slurry

Vieux-pré (France)dam material

32% 13%w lL P� � �

0

100

200

300

400

Suc

tion:

kP

a

100

96

92

88

84

80

Deg

ree

ofsa

tura

tion:

%

Fig. 7. Variation of: (a) void ratio; (b) suction; and (c) degree ofsaturation during undrained oedometer loading of Vieux-predam material

100

95

90

85

80

Deg

ree

ofsa

tura

tion:

%

Vieux-pré dam material

Initialstate

Free wetting Oedometer

10 100 1000 10000Terzaghi effective vertical stress: kPa

Fig. 8. Degree of saturation changes, due to oedometer loadingat constant water content and free wetting, plotted againstTerzaghi’s effective stress



the wetting path as it leads to a decrease in void ratio,compared to an increase in the case of the free wetting test,although both tests result in an increase in degree of satura-tion. The results also confirm the interest of carrying outundrained tests with suction measurement (using psychrom-eters) complementing suction-controlled drained tests.

NOTATIONCms swelling coefficient with respect to void ratioDms swelling coefficient with respect to water content

e void ratioIP plasticity indexn porosity

Sr degree of saturations suction (s ¼ ua � uw)

sspo suction at SPO water content and maximum densityua pore-air pressureuw pore-water pressurew water content

wL liquid limitwP plastic limit

wspo standard Proctor optimum water contentªdSPO specific weight at SPO water content and maximum density

ªs specific weight of grainsªw specific weight of water�v total vertical stress

REFERENCESAlonso, E. E. & Pinyol, N. M. (2008). Unsaturated soil mechanics

in earth and rock dam engineering. In Unsaturated soils: ad-vances in geo-engineering (eds D. G.Toll, C. E. Augarde, D.Gallipoli and S. J. Wheeler), pp. 3–32. London: Taylor andFrancis.

AFNOR (1991). Sols: reconnaissance et essais – Determination dela masse volumique des particules solides des sols – Methodedu pycnometre a eau, AFNOR NF P94-054. France: AFNORGroupe.

AFNOR (1992). Sols: reconnaissance et essais – Analyse granulo-metrique des sols – Methode par sedimentation, AFNOR NFP94-057. France: AFNOR Groupe.

AFNOR (1993). Sols: reconnaissance et essais – Determinationdes limites d’Atterberg – Limite de liquidite a la coupelle –Limite de plasticite au rouleau, AFNOR NF P94-051. France:AFNOR Groupe.

AFNOR (1995). Sols: reconnaissance et essais – Identificationgranulometrique – Methode de tamisage par voie humide,AFNOR XP P94-041. France: AFNOR Groupe.

AFNOR (1999). Sols : reconnaissance et essais – Determinationdes references de compactage d’un materiau – Essai Proctornormal. Essai Proctor modifie, AFNOR NF P94-093. France:AFNOR Groupe.

Ahmed, S., Lovell, C. W. & Diamond, S. (1974). Pore sizes andstrength of compacted clay. J. Geotech. Engng Div., ASCE 100,No. GT4, 407–425.

ASTM (American Society for Testing and Materials) (1995). Stan-dard test method for measurement of soil potential (suction)using filter paper (D5298–94). In 1995 annual book of ASTMstandards, vol. 04–08. Philadelphia, PA: ASTM.

Birle, E., Heyer, D. & Vogt, N. (2008). Influence of the initial watercontent and dry density on the soil–water retention curve andthe shrinkage behaviour of a compacted clay. Acta Geotechnica3, No. 3, 191–200

Brown, J. L. & Sivakumar, V. (2008). The changes in stress regimeduring wetting of unsaturated compacted clays when laterallyconfined. In Unsaturated soils: advances in geo-engineering(eds D. G. Toll, C. E. Augarde, D. Gallipoli and S. J. Wheeler),pp. 361–368. London: Taylor and Francis.

Delage, P., Suraj De Silva, G. P. R. & Vicol, T. (1992). Suction

controlled testing of nonsaturated soils with an osmotic conso-lidometer. Proc. 7th Int. Conf. on Expansives Soils, Dallas 1,206–211.

Delage, P., Audiguier, M., Cui, Y. J. & Howat, M. D. (1996).Microstructure of a compacted silt. Can. Geotech. J. 33, No. 1,150–158.

Delage, P., Howat, M. D. & Cui, Y. J. (1998). The relationshipbetween suction and swelling properties in a heavily compactedunsaturated clay. Engng Geol. 50, No. 1–2, 31–48.

Fleureau, J. M., Kheirbek-Saoud, S., Soemitro, R. & Taibi, S.(1993). Behavior of clayey soils on drying–wetting paths. Can.Geotech. J. 30, No. 2, 287–296.

Fleureau, J. M., Verbrugge, J. C., Huergo, P. J., Gomes Correia, A.& Kheirbek-Saoud, S. (2002). Aspects of the behaviour ofcompacted clayey soils on drying and wetting paths. Can.Geotech. J. 39, No. 6, 1341–1357

Gallipoli, D., Wheeler, S. J. & Karstunen, M. (2003a). Modellingthe variation of degree of saturation in a deformable unsaturatedsoil. Geotechnique 53, No. 1, 105–112, doi: 10.1680/geot.2003.53.1.105.

Gallipoli, D., Gens, A., Sharma, R. & Vaunat, J. (2003b). Anelasto-plastic model for unsaturated soil incorporating the effectsof suction and degree of saturation on mechanical behaviour.Geotechnique 53, No. 1, 123–136, doi: 10.1680/geot.2003.53.1.123.

Gens, A., Alonso, E. E., Suriol, J. & Lloret, A. (1995). Effect ofstructure on the volumetric behaviour of a compacted soil. Proc.1st Int. Conf. on Unsaturated Soils (UNSAT 95), Paris, France(eds E. E. Alonso and P. Delage) 2, 83–88. Rotterdam: Balk-ema.

Guillot, X., Al Mukhtar, M., Bergaya, F. & Fleureau, J.M. (2001).Effect of hydromechanical stresses on pore space and waterretention in a clay. Proc. 15th Int. Conf. Soil Mech. andGeotech. Engng (ICSMGE), Istanbul, Turkey 1, 101–105. Rot-terdam: AA Balkema.

Kassif, G. & Ben Shalom, A. (1971). Experimental relationshipbetween swell pressure and suction. Geotechnique 21, No. 3,245–255, doi: 10.1680/geot.1971.21.3.245.

Sivakumar, V., Tan, W. C., Murray, E. J. & McKinley, J. D. (2006).Wetting, drying and compression characteristics of compactedclay. Geotechnique 56, No. 1, 57–62, doi: 10.1680/geot.2006.56.1.57.

Sun, D., Sheng, D. C., Cui, H. B. & Li, J. (2006). Effect of densityon the soil–water-retention behaviour of compacted soil. Proc.4th Int. Conf. Unsaturated Soils, Arizona 1, 1338–1347. Reston,VA: American Society of Civil Engineers.

Tang, A. M., Cui, Y. J. & Bernel, N. (2008). Compression-inducedsuction in a compacted expansive clay. In Unsaturated soils:advances in geo-engineering (eds D. G. Toll, C. E. Augarde, D.Gallipoli and S. J. Wheeler), pp. 369–374. London: Taylor andFrancis.

Tarantino, A. & Tombolato, S. (2005). Coupling of hydraulic andmechanical behaviour in unsaturated compacted clay. Geotechni-que 55, No. 4, 307–317, doi: 10.1680/geot.2005.55.4.307.

Tarantino, A. & De Col, E. (2008). Compaction behaviour of clay.Geotechnique 58, No. 3, 199–213, doi: 10.1680/geot.2008.58.3.199.

Thakur, V. K. S., Sreedeep, S. & Singh, D. N. (2005). Parametersaffecting soil-water characteristic curves of fine-grained soils.J. Geotech. Geoenviron. Engng 131, No. 4, 521–524.

Vanapalli, S. K., Fredlund, D. G. & Pufahl, D. E. (1999). Theinfluence of soil structure and stress history on the soil-watercharacteristics of a compacted till. Geotechnique 49, No. 2,143–159, doi: 10.1680/geot.1999.49.2.143.

Wheeler, S. J., Sharma, R. S. & Buisson, M. S. R. (2003). Couplingof hydraulic hysteresis and stress–strain behaviour in unsatu-rated soils. Geotechnique 53, No. 1, 41–54, doi: 10.1680/geot.2003.53.1.41.

Zerhouni, M. I. (1995). Triaxial testing using psychrometers.Unsaturated soils. Proc. 1st Int. Conf. on Unsaturated Soils(UNSAT 95), Paris, France (eds E. E. Alonso and P. Delage) 2,673–678. Rotterdam: Balkema.



Session 2

Experimental Observation and Modelling


Romero, E. et al. (2011). Geotechnique 61, No. 4, 313–328 [doi: 10.1680/geot.2011.61.4.313]

55

An insight into the water retention properties of compacted clayey soils

E. ROMERO�, G. DELLA VECCHIA† and C. JOMMI†

Experimental data from different testing methodologieson different compacted clayey soils, with dominant bimo-dal pore size distribution, are presented and analysed, toprovide a comprehensive picture of the evolution of theaggregated fabric along hydraulic and mechanical paths.Fabric changes are analysed both from the porous net-work viewpoint, by means of careful mercury intrusionporosimetry investigation, and from the soil skeletonviewpoint, by quantifying swelling and shrinkage of theaggregates in an environmental scanning electron micro-scopy study. The consequences of the aggregated fabricevolution on the water retention properties of compactedsoils are analysed and discussed. A new model for waterretention domain is proposed, which introduces a depen-dence of the intra-aggregate pore volume on water con-tent. The model succeeds in tracking correctly theevolution of the hydraulic state of the different soilsinvestigated along generalised hydromechanical paths.The proposed approach brings to light coupling betweenintra-aggregate and inter-aggregate pores in the retentionproperties of compacted clayey soils. Dependence of theair entry and the air occlusion values on swelling andshrinking of aggregates, besides void ratio, is introducedand discussed.

KEYWORDS: clays; compaction; fabric/structure of soils;laboratory tests; partial saturation; suction

La presente communication expose et analyse des don-nees experimentales obtenues avec differentes methodolo-gies d’essai sur des sols argileux divers a distributiongranulometrique bimodale dominante, pour tracer untableau integral de l’evolution de la structure d’agregatle long de chemins hydrauliques et mecaniques. Onanalyse des variations de la structure du point du reseauporeux, par le biais d’un examen soigneux de la porosi-metrie par intrusion de mercure, et du point de vue dusquelette du sol, en quantifiant le gonflement et le retraitdes agregats dans le cadre d’une etude environnementalea microscopie a balayage electronique. On analyse lesconsequences de l’evolution de la structure d’agregats surles proprietes de retenue de l’eau des sols compactes, eton les discute. On propose un nouveau modele pour laretenue de l’eau, qui introduit le principe de la depen-dance du volume interstitiel intra agregats de la teneuren eau. Le modele parvient a suivre correctement l’evolu-tion de l’etat hydraulique des differents sols examines lelong de chemins hydromecaniques generalises. L’approcheproposee devoile le rapport entre pores intra-agregats etinter-agregats dans les proprietes de retenue de l’eau desols argileux compactes. On introduit la dependance del’entree d’air et les valeurs d’occlusion d’air sur legonflement et le retrait d’agregats, en plus de l’indice devide, et on en discute.

INTRODUCTION AND BACKGROUNDBoth suction and a measure of the amount of water contentof the soil need to be considered to adequately understand,describe and predict the behaviour of unsaturated soils (e.g.Wheeler, 1996; Jommi, 2000; Vaunat et al., 2000; Gallipoliet al., 2003; Sheng et al., 2004; Pereira et al., 2005; Gens etal., 2006; Romero & Jommi, 2008; Laloui & Nuth, 2009).Recently, attention has also been paid to the influence ofhydraulic history on the response of unsaturated soils alonggeneralised hydromechanical stress paths (Sun et al., 2007;Romero & Jommi, 2008; Muraleetharan et al., 2009).

The hydraulic state of a soil is the result of solid–fluidinteraction, and it is evaluated by introducing the so-calledwater retention curve, which is a constitutive law thatsummarises the dependence on suction of the amount ofwater stored in a soil. Solid–fluid interactions are dominatedby capillary and adsorption mechanisms. They are stronglydependent on the mineralogical composition of the soil andon its fabric, on temperature and on pore fluid chemistry.Specific surface and pore size distribution may be chosen todescribe appropriately the role played by the solid phase onthe retention properties of compacted soils.

In the formulation of conceptual water retention models,stored and adsorbed water have been traditionally quantifiedon a mass basis, in terms of water content, w (mass of waterover solid mass), or on a volumetric basis, through degree ofsaturation Sr (volume of water over volume of voids),volumetric water content Łw (volume of water over total soilvolume) or water ratio ew (volume of water over solidvolume). In fact, none of these variables seems to providedefinite advantages in the description of the water retentionbehaviour, as the example presented in Fig. 1 demonstrates.Bentonite-enriched sand, which was used as an engineeredbarrier in the gas migration test at the Grimsel undergroundlaboratory (Olivella & Alonso, 2008), was compacted on thedry side of optimum. Relevant properties of the material aresummarised in Table 1. The stress paths followed in thelaboratory in terms of vertical net stress, (�v � ua), andmatric suction, s (difference between pore air pressure ua

and pore water pressure uw), which replicated the in situtest, are shown in Fig. 1(a). Starting from the as-compactedstate (point A in the figure), the sample was subjected towetting at constant volume (path AB). Afterwards, thesample underwent a series of unloading (BC, EF), drying(CD), loading (DE) and wetting (FG) paths. Changes in voidratio e along the stress paths are reported in Fig. 1(b). Theevolution of the hydraulic state of the material along thehydromechanical paths is reported in Figs 1(c)–1(e) adopt-ing different water content measures. As it can be appre-ciated from the three figures, which plot the evolution of thehydraulic state of the material, retention properties are farfrom being intrinsic, or characteristic. Swelling, shrinkage,loading and unloading all affect the water storage capacity

Manuscript received 2 March 2010; revised manuscript accepted 14December 2010.Discussion on this paper closes on 1 September 2011; for furtherdetails see p. ii.� Department of Geotechnical Engineering and Geosciences,Universitat Politecnica de Catalunya, Barcelona, Spain† Department of Structural Engineering, Politecnico di Milano,Milano, Italy


of the material, but none of the adopted variables allows fullnormalisation of the water retention paths.

Even the concepts of ‘drying’ and ‘wetting’ may beambiguous if they are not referred to a specific water

measure. For example, the compression path DE at constantsuction causes a decrease of water ratio (Fig. 1(c)), ew,which suggests a drying process, whereas the same pathdevelops in Fig. 1(d) with increasing Sr, which entails a

450

5

1

Suc

tion,

: kP

a (lo

g sc

ale)

s

AD

AB: wetting atconstant volume

F E

C B

G

15 95 380 725

Vertical net stress, ( ): kPa(a)

σv a� u

0·50

0·45

0·40

Voi

d ra

tio, e A

D

F E

C

B

G

10 100 1000

Vertical net stress, ( ): kPa(b)

σv a� u

A D

F

E

CB

G

0·20 0·30 0·40 0·50

Suc

tion,

: kP

as

Water ratio,(c)

ew

1000

100

10

1

AD

F

E

CB

G

0·50 0·70 0·90 1·00

Suc

tion,

: kP

as

Degree of saturation,(d)

Sr

1000

100

10

10·60 0·80

AD

F

E

CB

G

0·10 0·30 0·40

Suc

tion,

: kP

as

Volumetric water content,(e)

θw

1000

100

10

10·20

Fig. 1. Hydromechanical paths on sand–bentonite mixture compacted at the dry side of optimum. (a) Stress pathfollowed; (b) volume change response; (c) water retention behaviour in terms of water ratio; (d) water retentionbehaviour in terms of degree of saturation; (e) water retention behaviour in terms of volumetric water content

56 ROMERO, DELLA VECCHIA AND JOMMI


wetting process. In fact, the hydraulic state can be affectedby adding or removing water, hence filling or emptying agiven pore volume, or by changing the volume of the pores.As a way to tackle this conflict, Tarantino (2009) introducedthe concept of hydraulic wetting related to a process whichincreases water volume, in contrast to the term mechanicalwetting, associated with an increase in the degree of satura-tion due to a decrease of void ratio.

In the past few years, much research has been devoted toanalyse the evolution of retention curves with void ratio,both from the experimental and the theoretical viewpoints.Dependence of water retention curves on volumetric statevariables and strain history was underlined by Romero et al.(1999), Vanapalli et al. (1999), Kawai et al. (2000), andmore recently by Tarantino (2009). The hysteretic nature ofwater retention curves was recently addressed by Kohgo(2008), Miller et al. (2008) and Nuth & Laloui (2008), notonly in the light of the well-known pore constriction effect,but also as a consequence of soil deformability.

To provide a reminder of the effects of porosity on waterretention properties, data from wetting paths, performed atdifferent void ratios e, on three different soils, namelyBarcelona clayey silt, Boom Clay and Febex bentonite, areshown in Fig. 2. Relevant properties of the soils are sum-marised in Table 1, which shows that these three materialscover a wide range of clayey soils with different plasticityindex values. All the soils were compacted on the dry sideof optimum, and the as-compacted samples display clearbimodal pore size distributions, due to the formation ofaggregates during compaction. Psychrometric measurements(WP4 dew-point mirror psychrometer), vapour equilibrium,and constant volume wetting and drying paths performed byaxis translation, were adopted to explore a wide range ofsuction, and to provide consistent pictures of the waterretention curves. Here, and in the following, water ratioew ¼ w Gs ¼ Sr e (where Gs is the specific gravity of the soilparticles) is chosen to describe the amount of water incompacted soils (Toll, 1995), basically because it is stronglyindicated to handle adsorption mechanisms. The role playedby void ratio on capillary mechanisms can be naturallyaccounted for.

The data presented in the figure show that two regionscan be distinguished in the water retention curves, assuggested by Romero et al. (1999) and Romero and Vaunat(2000). At low suction values, the amount of water stored inthe soil is high enough to saturate the pores inside the

aggregates – which will be termed ‘intra-aggregate pores’ or‘micropores’ – and to partly fill the pores between theaggregates – ‘inter-aggregate pores’ or ‘macropores’. In thelatter region, capillary storage mechanisms dominate, whichclearly depend on the characteristic dimensions of intercon-nected macropores, hence on porosity. On the other hand, at

Table 1. Properties of soils used in the present study

Soil Mineralogy Liquid limit,wL: %

Plastic limit,wP: %

Density of solidsrs: Mg/m3

% Particles, 2 �m

Total specificsurface: m2/g

Barcelona clayey silt� Illitic clay fraction 32 16 2.66 15 12Boom Clayy Illitic-kaolinitic clay fraction 56 29 2.70 50 53Febex bentonite{ Montmorillonite . 90%

Ca2þ (38%)Naþ (23%)

102 53 2.70 64–70 725

MX-80 bentonite} Montmorillonite82–85%Naþ (82%)Ca2þ (13%)

520 46; 62 2.65 80–90 522; 700

Bentonite-enrichedsand}

80/20 proportion on dry massbasis of quartz sand/Kunigel V1sodium bentonite

36 — 2.65 20 137

� Barrera (2002).y Romero et al. (1999); Romero & Vaunat (2000).{ Lloret et al. (2003); Villar et al. (2005).} Delage (2002); Villar et al. (2005); Delage et al. (2006).} Romero et al. (2002).

Boom ClayAxis translation ( 0·93)e �

Axis translation ( 0·65)e �

Vapour equilibrium (e 0·93)�

Vapour equilibrium (e 0·65)�

WP4 psychrometer ( 0·93–0·99)e �


Febex bentonite (adaptedfrom Lloret ., 2003)Vapour equilibrium

et al

e 0·62–1·78(free swelling)

�

e 0·74�

Barcelona clayey silt(Barrera, 2002)SMI psychrometer

e 0·67�

e 0·47�

e 0·37�

100

10

1

0·1

0·01

Suc

tion:

MP

a

0 0·2 0·4 0·6 0·8 1·0 1·2

Water ratio, ew

Fig. 2. Wetting paths performed on three materials compactedon the dry side of optimum (Boom Clay, Febex bentonite andBarcelona clayey silt) at different void ratios

WATER RETENTION PROPERTIES OF COMPACTED CLAYEY SOILS 57


ew lower than 0.2 for Barcelona clayey silt, 0.4 for BoomClay and 0.6 for Febex clay, the influence of void ratio isfound to be negligible, which indicates a region dominatedby adsorptive storage mechanisms, where water with lowermobility is held inside the intra-aggregate porosity. In amain wetting path, intra-aggregate pores will be saturatedbefore water begins to be stored in the macropores.

The relative contributions of capillarity and adsorptivesurfaces forces had already been highlighted by Or & Tuller(1999), derived from physically based models. The authorsconcluded that capillary forces dominate at low suctionvalues, whereas adsorptive forces rule the hydraulic behav-iour at suctions higher than 1 MPa. The amount of adsorbedwater is strongly linked to the specific surface (Tuller & Or,2005), and it is influenced by the soil intra-aggregatestructure and by the chemical interactions of the solidsurface with the pore fluid. The link between water affinity,specific surface and liquid limit was discussed by De Bruynet al. (1957), Black (1962), and more recently by Aubertinet al. (2003) and Frydman & Baker (2009).

Identical conclusions may be drawn if the retention behav-iour is analysed on drying paths. As an example, data onBoom Clay samples at different void ratios subjected tomain drying are reported in Fig. 3. The experimental datashow that dependence on void ratio may be again appre-ciated only for water ratios above ew ¼ 0.4.

Dependence of water retention properties on void ratiohas been analysed and modelled in previous years mainlyfocusing on the evolution of the air-entry or the air-occlu-sion suction values in drying or wetting processes, respec-tively (see, for instance, Huang et al., 1998; Vanapalli et al.,1999; Kawai et al., 2000; Gallipoli et al., 2003; Nuth andLaloui, 2008; Masın, 2009). Experimental data presented inFig. 4 suggest supporting this approach. The data come fromdifferent wetting paths, performed on samples of Boom Claycompacted at initial void ratio e0 ¼ 0.93. Wetting was per-

formed under oedometer conditions, at the different verticalnet stresses indicated in the figure. The water retention dataare enclosed by the wetting retention curves at constant voidratio e ¼ 0.93 and e ¼ 0.65. The degree of saturation Sr ofthe samples corresponding to the different suctions is re-ported within brackets. The air occlusion suction range,which is bounded by the curves Sr ¼ 0.85 and Sr ¼ 0.90,obtained by interpolation of relevant experimental data, isshown in the figure. This zone clearly progresses towardslarger matric suction values at decreasing void ratio.

Given this picture, an experimental programme wasplanned to analyse in greater detail the evolution of waterretention properties of compacted clayey soils. Besides therole played by void ratio, activity of the clay fraction andsolid–fluid interaction were investigated, with special atten-tion to the inter-playing role of the different structure levels.As will be shown, aggregates will continue exchanging watereven after reaching saturation. The potential for shrinkageand swelling of the aggregates will affect the inter-aggregateporosity and will influence the current retention propertiesof the soil. A new model for the water retention domain, inwhich the clay activity is taken into account by means of aphysically based interpretation, is proposed, to show theimportant effects of the evolution of the compacted soilmicrostructure promoted by hydraulic paths, besides mech-anical paths.

A MICROSTRUCTURAL INSIGHT INTO WATERRETENTION PROCESSES

To analyse the evolution of microstructure along differenthydraulic and mechanical paths, a microstructural study

Drying branchesAxis translation ( 0·93)e �

Axis translation ( 0·65)e �

Vapour equilibrium ( 0·93)e �

Vapour equilibrium ( 65)e � 0·

SMI psychrometer ( 0·50–0·82)e �


100

10

1

0·1

0·01

Suc

tion:

MP

a

0 0·2 0·4 0·6 0·8 1·0

Water ratio, ew

Fig. 3. Main drying paths for compacted Boom Clay at differentvoid ratios

Wetting branchese 0·93�

e 65� 0·

e0 93 at 0·1 MPa� 0·

e0 65� at 0·3 MPa0·

e0 93 at 0·6 MPa� 0·

e0 65� at 1·2 MPa0·

( for 0·65; for 0·93; for 0·93 at 1·2 MPa;0·6 MPa; 0·3 MPa; and 0·1 MPa)S e S e S er r r 0� � �

2

1

0·5

0·2

0·1

0·05

0·02

0·01

Suc

tion:

MP

a

0·40 0·60 0·80 1·00Water ratio, ew

(0·83; 0·59; 0·76;0·67; 0·60; 0·60)

(0·90; 0·70; 0·78;0·78; 0·70; 0·69)

Sr 0·85air-occlusion zone

�

Sr 0·90�

(0·98; 0·84; 0·96;0·92; 0·85; 0·84)

(1·0; 0·97; 1·0;1·0; 0·98; 0·96)

Fig. 4. Air-occlusion zone evolution along wetting paths forcompacted Boom Clay (e0 0.93) in oedometer at differentvertical net stresses



using mercury intrusion porosimetry (MIP) and environmen-tal scanning electron microscopy (ESEM) was undertaken.Details of the techniques adopted can be found in Romero& Simms (2008).

Experimental investigationMercury intrusion porosimeter data are commonly used to

analyse compacted soil fabric (see Delage et al., 1996;Romero et al., 1999; Monroy et al., 2010; among others).The evolution of the pore size density function, PSD, ofcompacted soils along drying or wetting paths was analysedby Cui et al. (2002), Simms & Yanful (2001, 2005), Kolijiet al. (2006) and Delage et al. (2006). Models of the waterretention curve directly based on PSD were also proposed(refer, for example, to Simms & Yanful, 2002, 2004;Romero & Simms, 2008). Here, the experimental investiga-tion was aimed at analysing in detail the evolution of intra-aggregate porosity and at quantifying this evolution alonggeneralised coupled hydromechanical paths.

Samples of Boom Clay were prepared by static compac-tion on the dry side of optimum, at void ratio e ¼ 0.93 andwater ratio ew ¼ 0.41 (Sr ¼ ew/e ¼ 0.44). The samples werethen subjected to different mechanical and hydraulic paths.The PSD of all the samples was evaluated afterwards, bymeans of MIP tests performed on freeze-dried specimens.

Sensitivity of the intra-aggregate pores to hydraulic andmechanical paths is exemplified by the data presented inFig. 5, where the PSD of the as-compacted sample iscompared to those of samples subjected to loading atconstant water content, wetting at constant volume or wet-ting in oedometer at null vertical stress (free saturation). Theas-compacted sample is characterised by a clear bimodaldistribution with dominant pore modes around 13 �m(macropores) and 60 nm (micropores). Loading the sampleat vertical stress of 0.6 MPa shifts the size of dominantmacropores towards a slightly lower value, while the micro-pore volume and its dominant pore size are not affected. Onthe contrary, the two saturation paths affect both inter-aggre-gate and intra-aggregate pores. After saturation, the two

samples display a similar fabric dominated by the singlepeak at around 1.2 �m, which is originated by both expan-sion of the micropores and reduction of macropores. ThePSD function of a sample, which was further subjected to adrying path up to a total suction of 100 MPa (relativehumidity around 50%) after wetting at constant volume, isshown in Fig. 6. It may be observed that the inter-aggregateporosity underwent an important reduction on shrinkage,which is evidenced by the small dominant peak around2 �m. On the contrary, the intra-aggregate porosity is almostcompletely recovered, as similarity of the PSDs of the driedand the as-compacted samples, at entrance pore sizes below1 �m, demonstrates. Therefore, it appears that the aggregatescreated by compaction on the dry side of optimum tend toswell and shrink almost reversibly, and that they may beconsidered a permanent feature of the compacted soil fabric.

A complementary view on the latter observations isqualitatively provided by the ESEM photomicrographs, takenat different hydraulic states, shown in Fig. 7. The aggregatesand the inter-aggregate pore space are clearly visible in theESEM image corresponding to the initial as-compacted state(Fig. 7(a)), whereas intra-aggregate pores are indistinguish-able. Macropores correspond to the dark areas in the figure.Their typical size, 7 �m and 11 �m in Fig. 7(a), is similar tothat estimated from MIP data (Fig. 5). The ESEM photo-micrograph in Fig. 7(b) shows the relatively uniform fabricof the compacted soil sample after wetting at constantvolume. Expansion of the aggregates and occlusion of themacropores, already evidenced by MIP results (Fig. 5), canbe appreciated. Nonetheless, typical aggregated fabric is stillvisible, although the aggregates fused together during hydra-tion. Upon subsequent drying (Fig. 7(c)), a network ofshrinkage cracks appears in the looser zones between den-sely packed aggregates, which recovers the permanent aggre-gated fabric. Similar ESEM photomicrographs, displayingfused aggregates for fully hydrated compacted samples, wererecently shown by Monroy et al. (2010).

Data interpretationTo provide a quantitative measure of intra-aggregate por-

osity for any hydraulic state of the soil, a criterion todistinguish the latter from inter-aggregate porosity must bechosen. Three possible ways, which can be adopted to

As-compacted 0·41ew �

Loaded at 0·41ew �

Saturation at constant volume

Free saturation

1·2

0·8

0·4

0

Por

e si

ze d

ensi

ty fu

nctio

n,/

(log

)�

δδ

ex

1 10 100 1000 10000 100000

Entrance pore size, : nmx

x 1·2 m� μ

Fig. 5. PSD functions of compacted Boom Clay subjected todifferent hydromechanical paths

As-compacted 0·41ew �

Saturation at constant volume

Dried after saturation

1·2

0·8

0·4

0

Por

e si

ze d

ensi

ty fu

nctio

n,/

(log

)�

δδ

ex

1 10 100 1000 10000 100000

Entrance pore size, : nmx

x 1·2 m� μ

Fig. 6. PSD functions of compacted Boom Clay subjected todifferent hydraulic paths



estimate the intra-aggregate porosity are described in Fig. 8,with reference to Boom Clay data. If the intra-aggregateporosity region is defined as the domain in which theretention curve is not sensitive to void ratio, the poreshaving size smaller than 300�400 nm can be identified fromFig. 8(a), according to the Laplace equation. A secondpossibility is to use the as-compacted PSD function, and toseparate intra-aggregate porosity from inter-aggregate poros-ity by an entrance pore size bounded by the as-compacteddominant modes. Data for Boom Clay obtained by MIP fore ¼ 0.93, shown in Fig. 8(b), would suggest a discriminatingpore size ranging from 200 to 400 nm. Although useful, thetwo criteria are fundamentally based on the characteristicPSD of the as-compacted soil, and they may give somewhatambiguous results if swelling and shrinking of the aggre-gates become appreciable as the total water content changes.

Following the proposal of Delage & Lefebvre (1984) andDelage et al. (1996), a third criterion may be adopted, basedon data from a mercury intrusion/extrusion cycle. The firstintrusion fills all the accessible and interconnected pore space,whereas on complete releasing of pressure, mercury is ex-truded only from the non-constricted pores, which are identi-fied with the intra-aggregate pores. The difference betweenthe intrusion and extrusion branches describes the entrapped(constricted) porosity, which is related to the inter-aggregatepore space. According to this criterion, the same MIP data oncompacted Boom Clay would suggest that a slightly largerentrance pore size of approximately 800 nm distinguishedmicropores from macropores for this sample. The lattercriterion would be appropriate to account for active micro-porosity, provided freeze-dried samples were subjected tocareful intrusion/extrusion cycles after any loading-hydraulicpath.

A possible alternative criterion is suggested here, whichwas used to elaborate most of the data discussed in thefollowing. The new criterion stems from the assumption,confirmed by the data presented previously, that the size ofthe inter-aggregate pores is always larger or equal to that ofthe intra-aggregate pores. Therefore, for any compacted soil,a delimiting pore size must exist, which is an upper boundfor the intra-aggregate porosity and a lower bound for theinter-aggregate porosity at the same time. This delimitingpore size can be determined, once and for all, from MIPdata on samples after compaction and saturation at constantvolume, which promotes swelling of the aggregates andreduction of the inter-aggregate porosity. The data presentedin Fig. 6 show that the two mechanisms result in a singlepore mode, from which intra-aggregate and inter-aggregatepores are not distinguishable. As a consequence, in fullyhydrated states the single pore mode can be partially as-signed to the intra-aggregate pore space and partially to theinter-aggregate one. The same conclusion holds indepen-dently of the mechanical constraints imposed during satura-tion, as the data for free-swelling in Fig. 5 demonstrate.Following these observations, the delimiting pore size maybe assumed to coincide with the dominant peak of the PSDfunction determined by MIP after wetting. For Boom Clay,the data in Fig. 5 give a consistent value of 1.2 �m.

Definition of the delimiting pore size allows for separatingthe micropores volume from that of the macropores in astraightforward manner, for any sample. Intra-aggregate (ormicrostructural) void ratio, em (micropores volume over solidvolume), and inter-aggregate (or macrostructural) void ratio,eM (macropore volume over solid volume), are adopted withthis aim, as they provide additive decomposition of void ratioe ¼ em + eM. Integration on PSD data of the pore volumebetween the minimum detectable pore size and the delimitingpore size, normalised by solid volume, gives the microstruc-tural void ratio of the sample under consideration. For

(a)

7 mμ

11 mμ

(b)

Expanded aggregate

(c)

Shrinkage cracksbetween aggregates

Fig. 7. ESEM photomicrographs on compacted Boom Clay atdifferent hydraulic states: (a) as-compacted state; (b) afterisochoric saturation; (c) after isochoric saturation and drying



example, based on the PSD functions presented in Fig. 5,em ¼ 0.38 could be calculated for the as-compacted sampleand em ¼ 0.69 for the sample saturated at constant volume.

Hydromechanical evolution of water retention capacityTo evaluate the influence of compaction water content and

of subsequent hydraulic paths on the evolution of micro-pores, samples of Boom Clay were subjected to differenthydraulic and mechanical paths, namely wetting, drying andloading paths at constant ew, starting from two differentinitial states, A and B in Fig. 9, with ewA ¼ 0.41 andewB ¼ 0.51. At the end of each path, em was evaluated fromMIP data adopting the same delimiting pore size (1.2 �m).The data collected and plotted in Fig. 9 suggest a non-linearevolution of the microstructural void ratio, showing a changein the slope around the shrinkage limit of the soil, whichreplicates the trend of a typical shrinkage curve.

For low water ratios, em increases slowly with ew. Themicrostructural degree of saturation, Sm

r , defined as the vo-lume of water inside the intra-aggregate pores over microporevolume, is lower than one (Sm

r , 1 ), while the macrostructur-al degree of saturation, SM

r , defined as the volume of waterinside the inter-aggregate pores over the macropore volume,is null (SM

r ¼ 0). The data cross the bisecting line (ew ¼ em)at a value of em, which will be denoted by e�m, whichcorresponds to a water ratio that is sufficient to saturate theaggregates (Sm

r ¼ 1 ) but which leaves the inter-aggregatepores empty (SM

r ¼ 0). Starting from this point, an increase inthe soil water content will partially continue to fill thesaturated and still expanding aggregates, and partially bestored in the inter-aggregate pores. Therefore, for water ratiosabove e�m, part of the allowable water will be exchanged bythe swelling (or shrinking, in the drying processes) aggre-gates, at Sm

r ¼ 1. The remaining part will affect the macro-scopic degree of saturation (0 < SM

r < 1). As a firstapproximation, evolution of the microstructural void ratio forew . e�m can be described by a linear interpolation with slopeequal to � (Fig. 9).

A complementary investigation based on ESEM wasundertaken to analyse the evolution of the microstructurefrom the aggregate deformation standpoint, by tracking swel-ling and shrinkage of a single aggregate during wetting anddrying. The ESEM technique jointly with digital imageanalysis has already been adopted by Montes-H et al.(2003a, 2003b), Romero & Simms (2008) and Airo Farullaet al. (2010), to estimate the swelling–shrinkage behaviourof clayey materials at different water potentials at the aggre-gate scale.

The wetting and drying paths imposed on the soil samples,as shown in Fig. 10(a), were controlled by means of relativehumidity. The absolute vapour pressure of the environmentalchamber and the temperature of the sample holder werecontrolled. Microscopic observations were performed on iso-lated aggregates of Boom Clay and Febex bentonite powder,compacted on the dry side of optimum, at relative humidityaround 50%. Images of the same aggregate were recorded atthe different equalisation stages (maintained at least for15 min), starting from point A at relative humidity of 33%,which represents the initial reference state. The volumetricstrain �v of the aggregate at the different wetting and dryingequalisation stages, presented in Fig. 10(b), was quantifiedby means of the expression (Romero & Simms, 2008)

�v ¼ ª�2D ¼ 1:5A� Ao

Ao

� �(1)

where A and Ao are the areas of the aggregate at equalisa-tion and at the initial reference states, respectively. A value

0·8

0·6

0·4

0·2

0

Por

e si

ze d

ensi

ty fu

nctio

n,/

(log

)�

δδ

ex

1 10 100 1000 10000 100000Entrance pore size, : nm

(b)x

e1

1000

100

10

1

0·1

0·01

Suc

tion,

: MP

as

0 0·20 0·40 0·60 0·80 1·00Water ratio,

(a)ew

x 300 nm�

x 400 nm�

e e1 2� e2e1

0·8

0·6

0·4

0·2

0

Cum

ula

tive

intr

usio

n vo

id r

atio

,e


(c)x

e1

Fig. 8. Different criteria to discriminate the intra-aggregate voidratio on compacted Boom Clay: (a) from water retention curvesat different void ratios; (b) from PSD function of the as-compacted material; (c) from separation between constrictedand non-constricted porosity



of ª ¼ 1.5 was adopted to account for isotropic straining inthe direction perpendicular to the image plane, in which �2D

was determined.The data obtained in the ESEM investigation on Boom

Clay are superposed over the data obtained by MIP in Fig.11, which provides an overall picture of the evolution of themicrostructural void ratio for different water ratios. TheESEM data were plotted after converting suction into waterratio by means of the retention data presented in Fig. 2. Thecurrent void ratio was calculated, based on volumetric strain,assuming that void ratio for the lowest attainable value ofsuction after the wetting path AB matched the value detectedby MIP (an initial/reference microstructural void ratio forBoom Clay of 0.31 was used). Taken all together, the datasuggest a value of e�m ¼ 0.38, and an interpolating slope� ¼ 0.40 to track the evolution of the intra-aggregate porevolume at increasing water ratio for ew . e�m.

An equivalent plot is depicted in Fig. 12 for compactedbentonites. Relevant properties of the different materials aresummarised in Table 1. Different data sets are plotted,namely the ESEM data on Febex bentonite already presented(Fig. 10(b)), MIP data on Febex bentonite from Lloret et al.(2003) and Romero et al. (2005), and data on MX-80bentonite from Delage et al. (2006). The separation betweenmicro- and macroporosity for Febex bentonite was definedwith the proposed criterion, applied to MIP results fromwetting at constant volume presented in Romero et al.(2005). For MX-80 bentonite, the criterion illustrated in Fig.8(a) was adopted, based on data of the water retention curveof MX-80 powder and of compacted samples wetted atconstant volume presented by Delage et al. (2006). Few datawere found in the literature to complement the microstruc-tural void ratio information for high values of ew. Data ofsaturated Na-bentonite and Ca-bentonite from Pusch &Karnland (1986; 1996), who gave the amount of microstruc-tural void ratio at given total void ratios, are also presented.As a first approximation, the data suggest a value ofe�m ¼ 0.56. A tentative average value of � ¼ 0.43 can besuggested for the compacted bentonites.

It is worth noting that, owing to the aggregate potentialfor swelling, the intra-aggregate void ratio becomes moreand more relevant to the total porosity as the total void ratiodecreases. The ratio between microstructural void ratio andtotal void ratio is plotted in Fig. 13 for saturated bentonite

and saturated Boom Clay, as a function of void ratio.Microstructural void ratio, inferred from the previous valuesof e�m and �, is compared in the figure with previousliterature data. For compacted Boom Clay the comparison ismade with data from Romero et al. (1999), who reported thevoid volume fraction in which water shows low mobility(water permeability lower than 0.01 of the saturated one, asindicated in the figure). The saturated Na- and Ca-bentoniteliterature data provided by Pusch & Karnland (1986, 1996)were compared with the corresponding microstructural voidratios obtained with e�m ¼ 0.56 and � ¼ 0.43 and assumingfull saturation, for given e. Consistent trends between re-ported data and results of the present study can be observed.

As already suggested by Romero & Vaunat (2000), e�mshould depend on the plasticity of the clay. Good correlationbetween e�m and the total specific surface Ss (reported inTable 1) is shown in Fig. 14, based on the data presented inthis paper and on further data on different soils. Data from

e ew m*�

S mr 1�

0 1 S Mr

e *m

S mr

e

ew

m*� 1

S Mr 0�

1·5

1·0

0·5

00 0·5 1·0 1·5

Water ratio, ew

Mic

rost

ruct

ural

voi

d ra

tio,e

mExperimental results (MIP)

1:1

AB

�

Maximumswelling ofaggregates

Linear model proposed

Fig. 9. Evolution of microstructural void ratio with water ratiofrom MIP tests on compacted Boom Clay: experimental dataand conceptual interpretation

Boom Clay (wetting)

Boom Clay (drying)

Febex bentonite (wetting)

Febex bentonite (drying)

B

B

RH

100%

�

70%

40%

A: RH 33%�

1·5

1·0

0·5

Va

pour

pre

ssur

e,: k

Pa

u v

5 10 15 20Temperature, : °C

(a)T

�15

�10

�5

0

5

Vol

umet

ric s

trai

n,: %

γε2D

B

B

A

3 5 10 20 30 50 100 200 300Total suction: MPa

(b)

Fig. 10. Wetting–drying paths followed at microstructural scale:(a) paths performed under controlled relative humidity inESEM; (b) evolution of volumetric strain along the differentwetting–drying stages



the literature were estimated from PSD and water retentiondata (see, for example, Tarantino (2009) for Speswhitekaolin with Ss ¼ 14 m2/g, and Delage (2002) for FoCa7 claywith Ss ¼ 300 m2/g). Also the slopes � in em: ew, describingthe tendency of the aggregates to swell, appear to be linkedto the plasticity of the clayey soils (see Table 1 and Table2), but further research is needed to propose reliable correla-tions of the two microstructural parameters with traditionalgeotechnical classification properties.

e ew m*�

S mr 1�

0 1 S Mr

S mr

e

ew

m*� 1

S Mr 0�

1·5

1·0

0·5

00 0·5 1·0 1·5

Water ratio, ew

Mic

rost

ruct

ural

voi

d ra

tio,e

m

As compacted

1:1

Wetting

e* 0·380·40

m �

��

MIP

Slight drying

Strong drying

ESEMWetting

Drying

Fig. 11. Evolution of microstructural void ratio with water ratiofrom MIP and ESEM results on compacted Boom Clay

e ew m*�

S mr 1�

0 1 S Mr

S mr

e

ew

m*� 1

S Mr 0�

1·5

1·0

0·5

00 0·5 1·0 1·5

Water ratio, ew

Mic

rost

ruct

ural

voi

d ra

tio,e

m

1:1

e* 0·560·43

m �

��

MX-80 bentonite(MIP results, re-elaboratedfrom Delage ., 2006)et alFebex bentonite(ESEM, drying path)Febex bentonite(ESEM, wetting path)Febex bentonite(MIP, as-compacted)Febex bentonite(MIP, wetting path)Saturated Na-bentonite(re-elaborated from Pusch& Karnland (1986, 1996))Saturated Ca-bentonite(re-elaborated from Pusch& Karnland (1986, 1996))

Fig. 12. Evolution of microstructural void ratio with water ratiofrom MIP and ESEM results on compacted bentonite (Febexand MX-80). Estimated evolution using Pusch & Karnland(1986, 1996) data. Average linear fit proposed for compactedbentonites

1·00

0·80

0·60

0·40

0·20

0

ee

m/

0·6 0·9 1·2

Void ratio, e

Compacted Boom Clay ( 0·01)Romero . (1999)

ket al

r

Saturated Ca-bentonite,Pusch & Karnland (1986, 1996)

Saturated Na-bentonite,Pusch & Karnland (1986, 1996)

Compacted Boom Clay( * 0·38, 0·40)e m � ��

Compacted bentoniteSaturated Febex bentonite( * 0·56, 0·43)e m � ��

Fig. 13. Evolution of the ratio (em/e) with void ratio e fordifferent clayey soils. Results reported in the literature andobtained in the present study

0·80

0·60

0·40

0·20

0

e* m

1 10 100 1000Total specific surface, : m /gSs

2

e S

S

* 0·088 ln 0·05 for 10 m /g

1 m /g

m s2

sr2

� � �

�

S

S

s

sr

⎛⎜⎝ ⎛

⎜⎝

Speswhite

kaolin

Barcelona cla

yey sil

t

Siltycla

y

BoomClay

FoCa clay

MX-80

Febexbentonite

Fig. 14. Correlation between total specific surface and e�m



A WATER RETENTION MODEL ACCOUNTING FORFABRIC CHANGESThe hydraulic model: microstructural and macrostructuraldomains

Exploiting the previous experimental observations, a newwater retention model is proposed for clays with intrinsicand permanent aggregated structure, accounting for swellingand shrinkage of the aggregates induced by total watercontent changes. To model the hydraulic behaviour, theamount of water adsorbed in the micropore volume is distin-guished from that stored in the inter-aggregate void space.Micro- and macropore retention mechanisms may be de-scribed separately, yet ensuring continuity of the wholeretention model and of its first derivatives, as suggested byDella Vecchia (2009). The choice for continuity comes fromthe observation of MIP data, showing PSD functions whichare usually continuous with their first derivatives, and allowsfor taking into account the influence of swelling aggregateson water exchange mechanisms in a straightforward way.The model proposed to describe the water retention domainis written in terms of the work-conjugate variables waterratio, ew, and suction, s (Houlsby, 1997).

For high suction values, water is adsorbed only in themicropores. A simple analytical expression for the mainwetting and drying branches describing the water stored inthe micropores, which does not depend on the void ratio e,may be assumed for ew

ew ¼be�m

ln smax=s�m� � bþ ln smax=s�m

� �bþ ln s=s�m

� � � 1

" #(2)

where e�m is the water ratio corresponding to fully saturatedmicropores and empty macropores, s�m is the suction corre-sponding to e�m, smax is the maximum suction attainable forew ¼ 0 and b is a parameter describing the average slope ofthe curve for high values of suction. In the absence ofspecific data, a value of smax ¼ 1 GPa, corresponding to ovendrying, may be assumed.

For s , s�m, corresponding to ew . e�m, water ratio is thesum of the water stored inside the microstructural porosityand the water which partially fills the macropores. Micro-pores and macropores are assumed to be in local equili-brium, hence subjected to the same suction, s. The key issuein the formulation is that the microstructural water ratio isnot constant for ew . e�m, owing to continuous swelling andshrinkage of the aggregates at changing water content. Itsevolution with water ratio is based on the conceptual frame-work presented previously. For the sake of simplicity, em canbe described by a linear reversible envelope, as a function ofthe two independent parameters, e�m and �

em ¼ e�m þ � ew � e�m� �

for ew . e�m; with 0 < � , 1 (3)

where � quantifies the swelling tendency of the aggregates.The dependence on e of the part of the curve which

describes the predominant capillary mechanisms in the inter-aggregate pores may be achieved by adopting the linearscaling criterion proposed by Romero & Vaunat (2000). Forew ranging between e�m < ew < e, the water ratio reads

ew ¼ em þ e� emð Þ

3 1� ln [1þ (s=sm)]

ln 2

� 1

1þ (Æs)n

� �m (4)

where m and n are independent parameters (van Genuchten,1980), e and em are the current total and microstructuralvoid ratios, respectively. The reference suction sm is calcu-lated, also for ew . e�m, based on the microscopic branch ofthe water retention curve (equation (2))

sm ¼ s�m expbe�m[bþ ln (smax=s�m)]

be�m þ emln (smax=s�m)� b

( )(5)

The two parameters m and n are assumed not to depend one as suggested by Romero & Vaunat (2000), Gallipoli et al.(2003) and Tarantino (2009). A suitable dependence of theair entry value on void ratio e is recovered by simplyimposing the two analytical expressions (2) and (4) to becontinuous, together with their first derivatives, at s ¼ sm.This choice links the value of Æ, which is inversely relatedto the air-entry or air-occlusion values respectively, to theset of independent parameters (m, n and microstructuralparameters)

Æ ¼ H�1=m � 1ð Þ1=n

sm

(6a)

where

H ¼ 2(ln 2)[bþ ln (smax=s�m)]

bþ (e� em)ln (smax=s�m)� b (6b)

The choice to link the air-entry (or air-occlusion) value toboth the parameters describing the microstructure and thosedescribing the macrostructure is strongly justified by theobservation that both a reduction in the total void ratio andan increase in the aggregate volume will reduce the meandimension of macropores, and therefore they will corre-spondingly increase the air-entry (or air-occlusion) value.

The behaviour inside the current main retention domain,which is bounded by main wetting and main drying curves,is assumed to be described by a linear relationship linkingthe suction and the degree of saturation increments

dSr ¼ �ks ds (7a)

where ks describes the slope of reversible hydraulic paths.The previous expression can be rewritten as

Table 2. Micro- and macrostructural parameters for the different soils used in the simulations shown in Figs 15, 16 and 22

Soil e�m � Wetting Drying smax: MPa n

s�m: MPa b s�m: MPa b

Barcelona clayey silt 0.22 0.22 2.0 7.0 — — 1000y 3.50Boom Clay 0.38 0.40 2.4 12 6.7 6.5 500 1.34Bentonite Febex, MX-80 0.56 0.43 25� 12� — — .1000� 1.50�Bentonite-enriched sand 0.25 0.85 0.32 14 0.70 1.5 1000y 1.34

� Febex bentonite.y Oven drying conditions (limited information in the high suction range).



dew ¼ew

ede � e ks ds (7b)

which describes both reversible hydraulic and mechanicalchanges in water ratio in a suitable way.

Model capabilitiesThe capabilities of the proposed model are first illustrated

with reference to the experimental data presented in Fig. 2and Fig. 3. In the simulations, the values for the microstruc-tural parameters e�m and � were taken from the availableMIP and ESEM studies (Fig. 11 and Fig. 12). In the case ofBarcelona clayey silt, water retention data were used toassign the value of e�m, following the criterion described inFig. 8(a), while � was considered part of the model calibra-tion. The suctions s�m D,W (D for drying, W for wetting) wereassigned based on the available retention data in all cases.The remaining parameters, bD,W and n were calibrated bybest fit with experimental data. It is worth noting that, in allcases, the same value of n was adopted for the drying andthe wetting branches. In addition, although the choice wouldnot be necessary, the relationship between the two para-meters n and m

m ¼ 1� 1

n(8)

was exploited, to reduce the number of the independentparameters of the macroscopic part of the model to just one.The whole set of parameters adopted in the simulations issummarised in Table 2.

Experimental data for Boom Clay at constant void ratioalong main drying and wetting paths are compared to modelsimulations in Fig. 15. The macrostructural parameter,n ¼ 1.34, was calibrated on the drying branch correspondingto e ¼ 0.92, in Fig. 3. The remaining branches of the waterretention domain were blind predicted. The four branchesare remarkably well caught, in spite of the adoption of asingle independent macroscopic parameter.

The comparison between experimental data and numericalsimulations of main wetting curves of compacted Febexbentonite (n ¼ 1.5) and Barcelona clayey silt (n ¼ 3.50) atdifferent constant void ratios are shown in Fig. 16. Again,the simulations performed after calibration of a singlemacroscopic parameter allow for tracking correctly the de-pendency of the retention domain on void ratio.

Microstructural effects on air-entry valueTo appreciate the influence of active swelling/shrinking

aggregates on the macroscopic branches of the water reten-tion domain, experimental data from a previous work byRomero et al. (2002, 2003) are analysed.

Mixtures of bentonite-enriched sand, S/B 80/20 (sand/bentonite with 20% bentonite on dry mass basis), wereprepared from Kunigel V1 sodium bentonite, which acts asfilling material, and uniform silica sand (with dominantparticle size around 1 mm), which forms the shieldingskeleton. Two different mixtures were prepared on the dryside of optimum at ew ¼ 0.29 and e ¼ 0.42 (Sr ¼ 0.69). Thefirst mixture was prepared with distilled water, preservingthe swelling characteristics of bentonite. Lead nitrate powderin a concentration of 1% of dry sand/bentonite mass wasadded to the mixing water in the second mixture. The nitrateions can precipitate and cement the stacks of smectite,acting as inhibitor of the swelling potential of the naturalbentonite (Auboiroux et al., 1996; Alawaji, 1999; Romero etal., 2003). The comparison between the behaviour of thetwo compacted materials allows the study of the structural

changes induced by an active microstructure compared to aninhibited, or inert, one.

The PSD functions of the as-compacted samples, andthose of the two mixtures after saturation, are compared inFig. 17 (all the samples were freeze-dried before mercuryintrusion). The as-compacted mixture displays multi-modalporosity, characterised by one pore mode at inter-grain scale(macroporosity, with pore sizes between sand grains around150–250 �m, corresponding to around 20% of the dominantparticle size) and two pore modes at the clay-aggregationscale (microporosity, with pore sizes smaller than 30 �m).The criterion illustrated in Fig. 8(b) was adopted in thiscase, owing to the presence of the shielding sand skeleton,which constrains the macropore dimensions to a minimumvalue of 30 �m (Romero et al., 2003).

On wetting, the saturated mixture with inert bentoniteapproximately preserved the PSD function of the as-com-pacted state. On the contrary, the mixture with activebentonite preserved the two modes of the microporosity,which were shifted towards larger pore sizes as a conse-quence of microstructure swelling. The inter-grain poremode was lost owing to the expansion of bentonite, whichcompletely invaded the macroporosity of the shielding skele-ton.

The observation is qualitatively confirmed by the ESEMimages collected in Fig. 18. The photomicrograph of the as-compacted state, which is displayed in Fig. 18(a), showsclearly detectable sand grains, nearly covered by clay aggre-gations. Interconnecting clay bridges, crossing and hinderinginter-grain pores, can also be distinguished. An average

Wetting ( 0·93)e �

Wetting ( 0·65)e �

Drying ( 0·93)e �

Drying ( 0·65)e �

Drying ( 0·50–0·82)e �

100

10

1

0·1

0·01

Suc

tion:

MP

a

0 0·2 0·4 0·6 0·8 1·0

Water ratio, ew

Model wetting ( 0·93)e �

Model wetting ( 0·63)e �

Model drying ( 0·93)e �

Model drying ( 0·63)e �

Fig. 15. Comparison between water retention experimental dataand model predictions at different void ratios for dry-sidecompacted Boom Clay (wetting and drying paths)



Febex bentonite(experimental data adaptedfrom Lloret ., 2003)et al

e 0·70–1·78(free swelling)

�

e 0·74Model

�

Barcelona clayey silt(experimental data fromBarrera, 2002)

e 0·67�

e 0·47�

e 0·37Model

�

100

10

1

0·1

0·01

Suc

tion:

MP

a

0 0·2 0·4 0·6 0·8 1·0 1·2

Water ratio, ew

1·4

ee

0·74 and1·22

�� e

ee

0·37;0·47

and 0·67

��

�

Fig. 16. Comparison between water retention experimental dataand model predictions at different void ratios for compactedFebex bentonite and Barcelona clayey silt (wetting paths)

300

nm 9 mμ

180

mμ

Macro-porosity

Microporosity (bentonite)

As-compacted

Saturated with active bentoniteSaturated with inert bentonite

0·30

0·20

0·10

0

Por

e si

ze d

ensi

ty fu

nctio

n

1 10 100 1000 10000 100000Entrance pore size: nm

Fig. 17. PSD function of the as-compacted and saturated statesfor two different bentonite-enriched sand mixtures, with activeand inert bentonite

(a)

1

23

(b)

1

4

(c)

2

1

Fig. 18. ESEM photomicrographs of bentonite-enriched sandmixtures: (a) as-compacted mixture; (b) saturated mixture withactive bentonite; (c) saturated mixture with inert bentonite. (1:sand grain; 2: inter-grain macroporosity (100–200 �m); 3: clayaggregations; 4: occluded macroporosity)



inter-grain pore size between 100 and 200 �m is detected,which is in agreement with PSD measurements (Fig. 17).On wetting, invasion of the inter-grain macroporosity byactive bentonite is clearly observed in Fig. 18(b), whereas inthe mixture with inert bentonite the macroporosity is pre-served (Fig. 18(c)). Microstructural void ratio em, associatedwith pore sizes lower than 30 �m, and degree of bentonitefilling F, which is the volume of bentonite divided by totalpore volume, of the different mixtures at different states arereported in Table 3. The latter was calculated with theexpression reported in the table and quantifies the inter-grainporosity occlusion undergone on active bentonite swelling.The different em values at known water ratios were used toestimate the microstructural water retention parameters (e�mand �), which are plotted in Fig. 19. Owing to the lowconstraint imposed by the macroporosity, bentonite is almostcompletely free to develop its swelling potential and fill theinter-grain porosity, which is reflected in a high value ofparameter �.

As a consequence of the microstructural changes under-gone on wetting, the mixtures will display different waterretention properties. Assimilating mercury injection to airintrusion during a desorption path at constant porosity(Romero & Simms, 2008), PSD data were exploited toestimate the branches of the water retention curves that thetwo samples would follow upon drying, if their microstruc-ture were frozen at the end of the first wetting path. Resultsfor the mixtures with active and inert bentonite after under-going saturation are compared to those for the as-compactedmaterial in Fig. 20. The comparison clearly shows that theestimated drying branch of the water retention curve of themixture with inert bentonite previously saturated does notdiffer substantially from the drying branch of the as-compacted mixture that did not undergo a saturation path.On the contrary, the estimated drying branch of the mixturewith active bentonite, which had previously undergone sa-turation, displays a higher water retention capacity with amuch higher air-entry value, due to disappearance of macro-pores. It is worth noting that the curve drawn for activebentonite is an upper bound for the actual drying branch ofthe water retention domain, which in the latter case isaffected by the progressive shrinkage of the aggregates uponsuction increase.

To highlight the capabilities of the proposed model tofollow the evolution of the water retention properties whenthey are affected by microstructure activity, data from com-pacted bentonite-enriched sand with active bentonite arecompared to numerical simulations along a relevant part ofthe stress path already presented in Fig. 1 (ABCDE in Fig.1(a)), describing a path comprising wetting at constantvolume, unloading, drying at constant net stress, and load-ing.

The path was simulated assuming active bentonite, with� ¼ 0.85, and the simulations were compared to the resultsof numerical simulations in which the swelling potential ofthe bentonite was reduced, by assuming � ¼ 0.11, based on

the previous experimental data in Fig. 19. All the othermodel parameters were kept equal in the two simulations(see Table 2). Again a single macroscopic parameter,n ¼ 1.34, was adopted for the macroscopic part of theretention domain.

Fairly good agreement can be observed between theresults of the simulation with active bentonite and theexperimental data in Fig. 21. The wetting path AB inducesswelling of the aggregates, increasing the air-occlusion valueprogressively as the total water ratio increases. On thecontrary, the curve predicted for inert bentonite describes amuch lower water retention capacity along the first wettingpath A9B9, as expected for a less active mixture, in whichthe sand fraction dominates the retention properties at lowvalues of suction. When drying starts, along path C9D9, anair-entry value lower than the experimental one for activebentonite is predicted, owing to the still available macropor-osity, which was not occluded on first wetting.

The previous simulations are exploited to provide a pic-ture of the capability of the model to describe the depen-dence of the air-entry value on the void ratio and on theswelling potential of the aggregates. The values of suctioncorresponding to two degrees of saturation, Sr ¼ 0.90 andSr ¼ 0.95, which may reasonably correspond to the onset of

Table 3. Microstructural characterisation of bentonite-enriched sand at different states

State and mixture em eM ¼ e� em F: %

As-compacted, sand–bentonite mixture e ¼ 0.42 0.28 0.14 77Saturated, active bentonite e ¼ 0.41 0.39 0.02 97Saturated, inert bentonite e ¼ 0.40 0.29 0.11 81

Note: F ¼ [em þ ( f rs=rsB)]=[eþ ( f rs=rsB)], where rs is average density of solid particles,rsB is particle density of bentonite (a constant value was assumed rsB ¼ rs) and f is themass fraction of bentonite in the mixture.

Sand/bentonite 80/20

As compacted

Saturated (bentonite active)

Saturated (bentonite inert)

1·5

1·0

0·5

0M

icro

stru

ctur

al v

oid

ratio

, em

0 0·5 1·0 1·5Water ratio, ew

e ew m*�

S mr 1�

0 1 S Mr

1:1

e* 0·25m �

� � 0·85

� � 0·11

Fig. 19. Evolution of microstructural void ratio with water ratiofor compacted bentonite-enriched sand with active bentonite(� 0.85) and inert bentonite (� 0.11)



air entry, are depicted in Fig. 22, as a function of void ratio.Two sets of curves are plotted for the two values � ¼ 0.11and � ¼ 0.85, corresponding to low and high activity, re-spectively. The figure demonstrates how the model is able topredict an increase in the air-entry value at decreasing e.Increasing � shifts the curves towards higher values ofsuction, and models correctly the contribution of the swel-ling potential of the clay aggregates.

SUMMARY AND CONCLUSIONSExperimental data from different testing methodologies on

different compacted clayey soils, characterised by an aggre-gated bimodal pore size distribution, allowed the analysis ofthe effects of coupled hydromechanical paths on fabricevolution and on water retention properties. Fabric evolutionwas tracked from both the porous network and the soilskeleton viewpoint by means of a combined MIP and ESEMexperimental programme.

The main outcomes of the study may be summarised asfollows. Compaction on the dry side of optimum gives thesoil an aggregated fabric, characterised by bimodal poresize distribution. Starting from the as-compacted state, anysaturation path leads to a fabric dominated by a single poremode. Nonetheless, in a new drying path compacted soilstend to recover a double porosity network. Similarity of thePDSs at intra-aggregate entrance pore size of the driedsample and the as-compacted state suggests that the aggrega-tions created by compaction on the dry side of optimum area permanent feature of the compacted soil fabric.

The fabric of the soil continuously evolves as the voidratio and the water content change. To track this evolution,the inter-aggregate pore volume, eM, was distinguished fromthe intra-aggregate one, em. The physical interpretation ofthe collected data is based on the assumption that a refer-ence intra-aggregate pore volume e�m, corresponding to satu-rated micropores and completely empty macropores, can bedefined. For ew . e�m, increasing water content is partitionedbetween the swelling aggregates, which remain saturated,and macropores, in which the degree of saturation increases.As a first approximation, the amount of swelling of theaggregates for ew . e�m can be described by a reversiblelinear interpolation with slope �. Collected data on differentclayey soils show consistent trends and validate the proposedapproach.

Similarly to the intra-aggregate pore volume, the instanta-neous retention capacity of compacted soils evolves continu-ously with water content. A description of the evolution ofthe intra-aggregate water ratio and of the correspondingsuction is introduced in the model previously proposed byRomero & Vaunat (2000), enhancing the capability ofdescribing the evolving water retention domain. A relativelysimple model for soil retention behaviour is proposed, able

0·1

0·01

0·001

Suc

tion,

: MP

as

0·20 0·25 0·30 0·35 0·40 0·45

Water ratio, ew

As compacted ( 0·42)e �

Saturated (bentonite inert, 0·40)e �

Saturated (bentonite active, 0·41)e �

Fig. 20. Drying branches of water retention curves derived fromMIP data for saturated bentonite-enriched sand mixtures, withactive and inert bentonite, compared to the as-compacted state


Experimental data(bentonite active)

Simulation, 0·85� �

Simulation, 0·11� �1

0·1

0·01

Suc

tion,

: MP

as

E� E

D�

D

A�A

B�B C

0·25 0·30 0·30 0·40 0·45Water ratio, ew

C�

Fig. 21. Water retention data for compacted bentonite-enrichedsand with active bentonite compared to numerical simulationsalong the stress path ABCDE in Figs 1(a) and 1(b), andexpected water retention features for the same mixture withinert bentonite

0·40

0·30

0·20

0·10

0

Suc

tion,

: MP

as

0·4 0·6 0·8 1·0Void ratio, e


� 0·85�

� 0·11�

Sr 0·95�

Sr 0·90�

Sr 0·95�Sr 0·90�

Fig. 22. Evolution with e and � of the air-entry suction zone(Sr 0.90 and 0.95) along drying branches (parameters fromsand–bentonite mixtures)



to account for the dependence of the air-entry value on boththe current void ratio and the intra-aggregate pore space.Remarkably, the microstructural parameters introduced in themodel, e�m describing intra-aggregate pore space, and �,which takes into account the swelling tendency of theaggregates, have a clear physical meaning. A dependence ofthese two parameters on common soil properties, like speci-fic surface and consistency limits, is expected.

Experimental data of main drying and main wettingbranches of compacted Boom Clay and other clayey soilsare compared to model simulations. The model seems tocapture the main features of wetting and drying branches fordifferent water contents, not only in terms of water storagecapacity but also in terms of evolution of air-entry or air-occlusion values. The model capabilities are illustrated bycomparison between numerical simulations and experimentaldata on two compacted bentonite-enriched sand mixtures,prepared with active or inert bentonite. The effects ofmicrostructural changes on retention properties, especiallythose linked to air-entry value changes detected by experi-mental data, are well tracked.

The collected experimental data and their consistentinterpretation highlight that the storage capacity of soil isnot at all a characteristic of the soil itself. In fact, it is theresult of the continuous evolution of both void ratio andthe pore size distribution, which is affected by both hydrau-lic and mechanical loading history. The interaction betweenthe solid and the liquid phases, resulting in swelling andshrinking of the aggregates, plays a definite role in thisevolution.

ACKNOWLEDGEMENTThe experimental work related to the characterisation of

the sand/bentonite was supported by NAGRA (NationaleGenossenschaft fur die Lagerung radioaktiver Abfalle)through GMT Emplacement Project.

REFERENCESAiro Farulla, C., Ferrari, A. & Romero, E. (2010). Volume change

behaviour of a compacted scaly clay during cyclic suctionchanges. Can. Geotech. J. 47, No. 6, 688–703, doi: 10.1139/T09-138.

Alawaji, H. (1999). Swell and compressibility characteristics ofsand–bentonite mixtures inundated with liquids. Appl. Clay Sci.15, No. 3–4, 411–430.

Aubertin, M., Mbonimpa, M., Bussiere, B. & Chapuis, R. P. (2003).A model to predict the water retention curve from basic geo-technical properties. Can. Geotech. J. 40, No. 6, 1104–1122.

Auboiroux, M., Baillif, P., Touray, J. C. & Bergaya, F. (1996).Fixation of Znþ2 and Pbþ2 by a Ca-montmorillonite in brinesand dilute solutions: Preliminary results. Appl. Clay Sci. 11, No.2–4, 117–126.

Baldi, G., Hueckel, T. & Pellegrini, R. (1991). Developments inmodelling of thermo-hydro-geomechanical behaviour of Boomclay and clay based buffer materials, Report EUR 13365.Brussels: Nuclear Science and Technology, Commission ofEuropean Communities.

Barrera, M. (2002). Estudio experimental del comportamiento hi-dro-mecanico de suelos colapsables. PhD thesis, UniversitatPolitecnica de Catlunya, Barcelona, Spain. See http://www.tdx.cat/TDX-0604102-095524 for further details.

Black, W. P. M. (1962). A method of estimating the Californiabearing ratio of cohesive soils from plasticity data. Geotechni-que 12, No. 4, 271–282, doi: 10.1680/geot.1962.12.4.271.

Cui, Y. J., Loiseau, C. & Delage, P. (2002). Microstructure changesof a confined swelling soil due to suction controlled hydration.In Unsaturated Soils (eds J. F. T. Juca , T. M. P. de Campos andF. A. M. Marinho), Proc. 3rd Int. Conf. on Unsaturated Soils,Recife, Brazil 2, 593–598. Lisse: AA Balkema.

De Bruyn, C. M. A., Collins, L. E. & Williams, A. A. B. (1957).

The specific surface, water affinity, and potential expansivenessof clays. Clay Mineralogy Bull. 3, 120–128.

Delage, P. (2002). On the behaviour of engineered clay barriers fornuclear waste disposal. In Environmental geomechanics (eds L.Vulliet, L. Laloui and B. A. Schrefler), Proceedings of theinternational workshop on environmental geomechanics, MonteVerita, Ascona, Switzerland, pp. 161–172. Lausanne: EPFLPress.

Delage, P. & Lefebvre, G. (1984). Study of the structure of asensitive Champlain clay and its evolution during consolidation.Can. Geotech. J. 21, No. 1, 21–35.

Delage, P., Audiguier, M., Cui, Y. & Howat, M. (1996). Microstruc-ture of a compacted silt. Can. Geotech. J. 33, No. 1, 150–158.

Delage, P., Marcial, D., Cui, Y. J. & Ruiz, X. (2006). Ageing effectsin a compacted bentonite: a microstructure approach. Geotechni-que 56, No. 5, 291–304, doi: 10.1680/geot.2006.56.5.291.

Della Vecchia, G. (2009). Coupled hydro-mechanical behaviour ofcompacted clayey soils. PhD thesis, Politecnico di Milano,Milan, Italy.

Frydman, S. & Baker, R. (2009). Theoretical soil-water character-istic curves based on adsorption, cavitation, and a doubleporosity model. Int. J. Geomech. 9, No. 6, 250–257.

Gallipoli, D., Wheeler, S. J. & Karstunen, M. (2003). Modelling thevariation of degree of saturation in a deformable unsaturatedsoil. Geotechnique 53, No. 1, 105–112, doi: 10.1680/geot.2003.53.1.105.

Gens, A., Sanchez, M. & Sheng, D. (2006). On constitutive model-ling of unsaturated soils. Acta Geotechnica 1, No. 3, 31–147.

Houlsby, G. T. (1997). The work input to an unsaturated granularmaterial. Geotechnique 47, No. 1, 193–196, doi: 10.1680/geot.1997.47.1.193.

Huang, S., Barbour, S. L. & Fredlund, D. G. (1998). Developmentand verification of a coefficient of permeability function for adeformable unsaturated soil. Can. Geotech. J. 35, No. 3, 411–424.

Jommi, C. (2000). Remarks on the constitutive modelling ofunsaturated soils. In Experimental evidence and theoretical ap-proaches in unsaturated soils (eds A. Tarantino and C. Mancu-so), Proceedings of an international workshop on unsaturatedsoils, Trento, Italy, pp. 139–153. Rotterdam: AA Balkema.

Kawai, K., Kato, S. & Karube, D. (2000). The model of waterretention curve considering effects of void ratio. In Unsaturatedsoils for Asia (eds H. Rahardjo, D. G. Toll and E. G. Leong),Proceedings of the Asian conference on unsaturated soils,UNSAT-ASIA 2000, Singapore, pp. 329–334. Rotterdam: AABalkema.

Kohgo, Y. (2008). A hysteresis model of soil water retention curvesbased on bounding surface concept. Soils Found. 48, No. 5,633–640.

Koliji, A., Laloui, L., Cuisinier, O. & Vulliet, L. (2006). Suctioninduced effects on the fabric of a structured soil. Transp. inPorous Media 64, No. 2, 261–278.

Laloui, L. & Nuth, M. (2009). On the use of the generalisedeffective stress in the constitutive modelling of unsaturated soils.Comput. Geotechnics 36, No. 1–2, 20–23.

Lloret, A., Villar, M.V., Sanchez, M., Gens, A., Pintado, X. &Alonso, E. E. (2003). Mechanical behaviour of heavily com-pacted bentonite under high suction changes. Geotechnique 53,No. 1, 27–40, doi: 10.1680/geot.2003.53.1.27.

Masın, D. (2009). Predicting the dependency of a degree of satura-tion on void ratio and suction using effective stress principle forunsaturated soils. Int. J. Numer. Analyt. Methods Geomech. 34,No. 1, 73–90, doi: 10.1002/nag.808.

Miller, G. A., Khoury, C. N., Muraleetharan, K. K., Liu, C. &Kibbey, T. C. G. (2008). Effects of soil skeleton deformationson hysteretic soil water characteristic curves: Experiments andsimulations. Water Resources Res. 44, No. W00C06, 1–10, doi:10.1029/2007WR006492.

Monroy, R., Zdravkovic, L. & Ridley, A. (2010). Evolution ofmicrostructure in compacted London Clay during wetting andloading. Geotechnique 60, No. 2, 105–119, doi: 10.1680/geot.8.P.125.

Montes-H, G., Duplay, J., Martinez, L. & Mendoza, C. (2003a).Swelling–shrinkage kinetics of MX80 bentonite. Appl. Clay Sci.22, No. 6, 279–293.

Montes-H, G., Duplay, J., Martinez, L., Geraud, Y. & Rousset-



Tournier, B. (2003b). Influence of interlayer cations on the watersorption and swelling–shrinkage of MX80 bentonite. Appl. ClaySci. 23, No. 5–6, 309–321.

Muraleetharan, K. K., Liu, C., Wei, C., Kibbey, T. C. G. & Chen,L. (2009). An elastoplatic framework for coupling hydraulic andmechanical behavior of unsaturated soils. Int. J. Plasticity 25,No. 3, 473–490, doi: 10.1016/j.ijplas.2008.04.001.

Nuth, M. & Laloui, L. (2008). Advances in modelling hystereticwater retention curve in deformable soils. Comput. Geotechnics35, No. 6, 835–844.

Olivella, S. & Alonso, E. E. (2008). Gas flow through clay barriers.Geotechnique 58, No. 3, 157–176, doi: 10.1680/geot.2008.58.3.157.

Or, D. & Tuller, M. (1999). Liquid retention and interfacial area invariably saturated porous media: upscaling from single pore tosample scale model. Water Resources Res. 35, No. 12, 3591–3606.

Pereira, J.-M.., Wong, H., Dubujet, P. & Dangla, P. (2005). Adapta-tion of existing behaviour models to unsaturated states: Applica-tion to CJS model. Int. J. Numer. Analyt. Methods Geomech. 29,No. 11, 1127–1155.

Pusch, R. & Karnland, O. (1986). Aspects of the physical state ofsmectite-adsorbed water. Stockholm, Sweden: SKB SwedishNuclear Fuel and Waste Management Co., Technical Report TR86-25.

Pusch, R. & Karnland, O. (1996). Physico/chemical stability ofsmectite clays. Engng Geol. 41, No. 1–4, 73–85.

Romero, E. & Jommi, C. (2008). An insight into the role ofhydraulic history on the volume changes of anisotropic clayeysoils. Water Resources Res. 44, No. W00C06, 1–16. doi:10.1029/2007WR006558.

Romero, E. & Simms, P. (2008). Microstructure investigation inunsaturated soils: A review with special attention to contributionof mercury intrusion porosimetry and environmental scanningelectron microscopy. Geotech. Geol. Engng 26, No. 6, 705–722.

Romero, E. & Vaunat, J. (2000). Retention curves of deformableclays. In Experimental evidence and theoretical approaches inunsaturated soils (eds A. Tarantino A. and C. Mancuso), Pro-ceedings of an international workshop on unsaturated soils,Trento, Italy, pp. 91–106. Rotterdam: AA Balkema.

Romero, E., Gens, A. & Lloret, A. (1999). Water permeability,water retention and microstructure of unsaturated Boom clay.Engng Geol. 54, No. 1–2, 117–127.

Romero, E., Knobelsdorf, J., Alonso, E. E. & Vaunat, J. (2002).Hydro-mechanical behaviour of 80/20 sand–bentonite mixture.In Environmental geomechanics (eds L. Vulliet, L. Laloui andB. A. Schrefler), Proceedings of an international workshop onenvironmental geomechanics, Monte Verita, Ascona, Switzer-land, pp. 347–352. Lausanne: EPFL Press.

Romero, E., Castellanos, E. & Alonso, E. E. (2003). Laboratorytests on compacted sand/bentonite/lead nitrate buffer materialfor the GMT emplacement project, Project Report GMT/IR 02–01. Switzerland: NAGRA.

Romero, E., Hoffmann, C., Castellanos, E., Suriol, J. & Lloret, A.(2005). Microstructural changes of compacted bentonite inducedby hydro-mechanical actions. In Advances in understanding en-

gineered clay barriers (eds E. E. Alonso and A. Ledesma),Proceedings of an international symposium on large scale fieldtests in granite, Sitges, Spain, pp. 193–202. Leiden: AA Balkema.

Sheng, D., Sloan, S. W. & Gens, A. (2004). A constitutive modelfor unsaturated soils: thermomechanical and computational as-pects. Comput. Mech. 33, No. 6, 453–465, doi: 10.1007/s00466-003-0545-x.

Simms, P. H. & Yanful, E. K. (2001). Measurement and estimationof pore shrinkage and pore distribution in a clayey till duringsoil-water characteristic curve tests. Can. Geotech. J. 38, No. 4,741–754.

Simms, P. H. & Yanful, E. K. (2002). Predicting soil-water char-acteristic curves of compacted plastic soils from measured pore-size distributions. Geotechnique 52, No. 4, 269–278, doi:10.1680/geot.2002.52.4.269.

Simms, P. H. & Yanful, E. K. (2004). Estimation of soil-watercharacteristic curve of clayey till using measured pore-sizedistribution. J. Environ. Engng 130, No. 8, 847–854.

Simms, P. H. & Yanful, E. K. (2005). A pore-network model forhydromechanical coupling in unsaturated compacted clayeysoils. Can. Geotech. J. 42, No. 2, 499–514.

Sun, D. A., Sheng, D. & Sloan, S. W. (2007). Elastoplastic model-ling of hydraulic and stress–strain behaviour of unsaturatedsoils. Mech. Mater. 39, No. 3, 212–221.

Tarantino, A. (2009). A water retention model for deformable soils.Geotechnique 59, No. 9, 751–762, doi: 10.1680/geot.7.00118.

Toll, D. G. (1995). A conceptual model for the drying and wettingof soil. Proc. 1st Int. Conf. Unsaturated Soils (eds E. E. Alonsoand P. Delage), Paris, France 2, 805–810. Rotterdam: AABalkema.

Tuller, M. & Or, D. (2005). Water films and scaling of soilcharacteristic curves at low water contents. Water ResourcesRes. 41, No. W09403, 1–6, doi: 10.1029/2005WR004142.

Vanapalli, S. K., Fredlund, D. G. & Pufahl, D. E. (1999). Theinfluence of soil structure and stress history on the soil-watercharacteristics of a compacted till. Geotechnique 49, No. 2,143–159, doi: 10.1680/geot.1999.49.2.143.

van Genuchten, M. T. (1980). A closed-form equation for predictingthe hydraulic conductivity of unsaturated soils. Soil Sci. Soc.Am. J. 44, 892–898.

Vaunat, J., Romero, E. & Jommi, C. (2000). An elastoplastic hydro-mechanical model for unsaturated soils. In Experimental evi-dence and theoretical approaches in unsaturated soils (eds A.Tarantino and C. Mancuso), Proceedings of an internationalworkshop on unsaturated soils, Trento, Italy, pp. 121–138.Rotterdam: AA Balkema.

Villar, M. V., Romero, E. & Lloret, A. (2005). Thermo-mechanicaland geochemical effects on the permeability of high-densityclays. In Advances in understanding engineered clay barriers(eds E. E. Alonso and A. Ledesma), Proceedings of an interna-tional symposium on large scale field tests in granite, Sitges,Spain, pp. 177–191. Leiden: AA Balkema.

Wheeler, S. J. (1996). Inclusion of specific water volume within anelasto-plastic model for unsaturated soil. Can. Geotech. J. 33,No. 1, 42–57.



Alonso, E. E. et al. (2011). Geotechnique 61, No. 4, 329–344 [doi: 10.1680/geot.2011.61.4.329]

71

Hydromechanical behaviour of compacted granular expansive mixtures:experimental and constitutive study

E. E. ALONSO�, E . ROMERO� and C. HOFFMANN�

Compacted granular mixtures of high-density bentonitepellets have been evaluated as an alternative buffermaterial to fill the empty space around nuclear wastedisposal canisters in horizontal drifts. Despite the obviousbenefits of these compacted mixtures (the backfillingoperation becomes easier and the gaps between the hostrock and the buffer are minimised), there are severalaspects of concern such as the effective blockage of thelarge inter-pellet pores due to granule swelling – thisblockage improves the water permeability properties –and the tendency to develop initial collapses before reach-ing an adequate swelling pressure. Selected test results ofa comprehensive laboratory experimental programme arepresented to gain insight into the hydromechanical re-sponse of this multi-porosity compacted material. Toimprove the information on local transient behaviour,simulation-assisted techniques using a double-structureconstitutive model are used. The paper presents a physi-cally based one-dimensional model to simulate experi-mental results of different transient processes, such asthe progressive loss of permeability during wetting andthe occurrence of concurrent phenomena during fastflooding at constant stress (initial collapse of the granulararrangement and parallel expansion of granules).

KEYWORDS: compaction; clays; fabric/structure of soils;laboratory tests; partial saturation; suction

On a evalue des melanges granulaires compactes depastilles de bentonite a densite elevee, en alternative a lamatiere tampon, pour le remplissage d’un espace videautour de recipients d’elimination de dechets nucleaires,dans des galeries horizontales. En depit des avantagesevidentes de ces melanges compactes (simplification del’operation de remblayage, et minimisation des ecartsentre la roche hote et le rampon), plusieurs aspectssuscitent une certaine inquietude, notamment l’obtura-tion effective de pores inter pastilles de grande taille sousl’effet du gonflement des pastilles – cette obturationrenforcant les proprietes de permeabilite de l’eau – et latendance a la production d’effondrements initiaux avantla realisation d’une pression de gonflement adequate. Desresultats d’essai selectionnes d’un programme experimen-tal integral en laboratoire sont presentes, afin de mieuxcomprendre la reaction hydromecanique de cette matierecompactee a multiporosite. Pour renforcer les informa-tions sur un comportement transitoire local, on fait usagede techniques a simulation faisant usage d’un modeleconstitutif a double structure. La communication pre-sente un modele unidimensionnel a base physique poursimuler les resultats experimentaux de differents proces-sus transitoires, par exemple la perte progressive de lapermeabilite au cours du mouillage, et la presence dephenomenes simultanes au cours d’inondations rapides acontrainte constante (effondrement initial de la configura-tion granulaire et expansion parallele de granules).

INTRODUCTIONCompaction in difficult emplacement conditions and fillingthe empty space around nuclear canisters in horizontal driftsis a major challenge to demonstrate the feasibility of en-gineering barriers. Specifically, the ‘engineered barrier (EB)experiment’ (Mont Terri, Switzerland; Alonso & Hoffmann,2007; Garcıa-Sineriz et al., 2008) was designed to demon-strate the suitability of compacted granular expansive mix-tures. Since these mixtures can be emplaced usingautomated techniques, the backfilling operation becomesmuch easier and the remaining gaps, usually found whenworking with block barriers, can be minimised. In this way,a better contact between canister and backfill, as well asbetween buffer and host geological medium, is ensured andit makes the backfilling operation more accurate.

In spite of the obvious benefits of compacted granularexpansive mixtures there are several aspects of concern inthese types of mixtures. One important aspect is that signifi-cant uncertainties exist on the capabilities of state-of-the-arthydromechanical models to simulate material behaviouralfeatures, which is an obvious consequence of the complex

nature of the material. One of the outstanding characteristicsof this material, which influences most of its properties, isthe multi-modal nature of its porous networks. This multi-modal porosity and the highly expansive nature of the pelletscontrol all the hydraulic and mechanical properties of themixture. Important aspects of the complex nature of hydra-tion under constant volume are the progressive blockage ofthe large pores due to granule swelling – this blockagegreatly reduces the permeability – and the tendency todevelop initial collapse strains before generating the finalswelling pressure.

The paper starts with the description of the geotechnicalproperties and microstructural features of the mixture, to-gether with a presentation of selected test results that willshow the complex nature of the material. However, macro-scopic experiments are insufficient to provide an overallbehavioural picture of the material. Simulation-assisted tech-niques using a physically based double-structure model arerequired to obtain further insight into specific features andlocal phenomena that are beyond the measurable scale. Thenumerical model developed is described in some detail inthe paper, and will be used later to capture the occurrenceof concurrent phenomena during fast flooding (expansion ofgranules and a parallel collapse of the granular arrangement)and the progressive decrease of the water permeability dueto the occlusion of the larger pores. Finally, computedresults and the observed behaviour are compared, and someconclusions are derived.

Manuscript received 4 March 2010; revised manuscript accepted22 November 2010.Discussion on this paper closes on 1 September 2011, for furtherdetails see p. ii.� Department of Geotechnical Engineering and Geosciences,Universitat Politecnica de Catalunya, Barcelona, Spain


BEHAVIOUR OF PELLETS AND PELLET MIXTURESBentonite granules (pellets) were obtained from Febex

bentonite powder (Lloret et al., 2003). Powder was preheatedto 1208C and then compacted using a roller. As a result ofthis process, very high dry density granules (up to a drydensity of 1.95 Mg/m3) with very low water content wereobtained. Basic properties of Febex bentonite (Lloret et al.,2003; Hoffmann, 2005) and the as-compacted state of thepellets, as reported by Hoffmann et al. (2007), are shown inTable 1. The tested material was based on the mixtureactually used under emplacement conditions. The grain sizedistribution was adjusted to a maximum size between 4 and15 mm compatible with the dimensions of the testing cells

and to a minimum size of 0.4 mm to avoid segregationduring preparation. Fig. 1 shows the pellet material for amaximum particle size of 4 mm. Specimens were preparedby dry-side static compaction (at water contents rangingbetween 3 and 4%) to dry densities varying between 1.30and 1.50 Mg/m3. Applied vertical stresses to reach given drydensities are presented in Fig. 2 (this information will belater used to define the initial position of the loading–collapse (LC) yield locus, as proposed by Alonso et al.,1990). Table 1 summarises the main properties of thecompacted mixture. Total suction values for both high-density pellets and compacted mixtures were obtained fromwater retention data presented below.

Table 1. Material properties: Febex bentonite powder, high-density pellets and compacted granular expansive mixtures

FEBEX bentonite

Mineralogy Liquid limit�y,wL: %

Plastic limit�y,wp: %

Density of solids�,rs : Mg/m3

% particles�, 75 �m

% particles, 2 �m

Montmorillonite. 90%Ca2þ (38%)Naþ (23%)

93–102 47–53 2.70 85 64–70

Pellet

Dry density�, rd : Mg/m3 Void ratio�,e0

Water content�:%

Degree of saturation:%

Total suction�:MPa

Max. fabrication stress:MPa

1.95 0.38 3–4 21–28 250–300 50–60

Compacted granular mixtures

Dry density, rd : Mg/m3 Void ratio, e0 Water content:%

Degree of saturation:%

Total suction:MPa

Max. fabrication stress:MPa

1.30 1.08 3–4 8–10 250–300 21.50 0.80 3–4 10–14 250–300 5–6

� Hoffmann (2005); Hoffmann et al. (2007)y Lloret et al. (2003)

5 mm

Fig. 1. Photograph of the granular expansive material with amaximum pellet size of 4 mm

100

80

60

40

20

0

Ver

tical

net

str

ess,

: MP

aσ v

1·2 1·4 1·6 1·8 2·0 2·2

Dry density, : Mg/mρd3

σ v

d

0·00

33ex

p(4·

93)

�

ρ

Fig. 2. Vertical stress applied on dry-side static compaction atwater contents between 3 and 4% and for different target drydensities

72 ALONSO, ROMERO AND HOFFMANN


Mercury intrusion porosimetry (MIP) tests were performedto characterise the multiple-porosity network of a high-density freeze-dried pellet and a pluviated mixture at a drydensity of 1.15 Mg/m3. Fig. 3(a) shows the pore size densityfunction of the high-density pellet, in which two types ofintra-pellet (microporosity) pore modes are identified: alarger inter-aggregate mode at around 3 �m between aggre-gations and an intra-aggregate mode at 13 nm. The porositysize density function (PSD) of the pluviated mixture shownin Fig. 3(b) displays an additional pore mode associated withthe inter-pellet voids (macropores) at around 250 �m. Fig. 1shows pore sizes even larger than this inter-pellet value,which have not been detected by MIP (maximum pore sizesdetected by this technique are around 500 �m). This remark-able multi-modal pore structure is a relevant characteristic ofthe material.

Consider now the macroporosity. It regulates the waterretention and water permeability properties of the medium,which are affected by any hydromechanical change modify-ing this void space. Moreover, the existence of macroporos-ity also allows for the arrangement of the skeleton (collapse)during wetting. Fig. 4 shows the changes undergone by thePSD functions along different stress paths. The largestmacropores not only decrease on loading (Fig. 4(a)), butalso when the mixture is wetted (occlusion of part of themacroporosity on pellet swelling as indicated in Figs 4(b),4(c)). On pellet swelling a new mode emerges at a pore sizearound 400 to 750 nm.

Another relevant property of the pellet mixture is that thehydraulic response of the material on first water injection ischaracterised by the high connectivity of the inter-pellet voidnetwork. Therefore, high inflow rates are expected. However,as the dense pellets hydrate – at the expense of the freecirculating water – they swell and fill the macropores and

the permeability decreases several orders of magnitude. Thisbehaviour is illustrated in Fig. 5(a), which shows the timeevolution of the water permeability during a constant volumeinfiltration test of a mixture at a dry density 1.30 Mg/m3 andsubjected to a water backpressure of 20 kPa. In this infiltra-tion test, the inflow and outflow water volumes are mon-itored. The outflow volume rate is interpreted as a flowthrough the interconnected inter-pellet voids (macropores)and it is used to derive the evolution of the Darcy per-meability of the mixture. The difference between inflow andoutflow defines the water stored in the sample, which is usedto estimate the evolution of the degree of saturation. After ashort initial phase where very high permeability values arecomputed, the permeability decays rapidly and reachesasymptotically the saturated value at the end of the test. Thechange in permeability has also been plotted against thecurrent degree of saturation in Fig. 5(b). The plot shows thefast obstruction of the open pore network when the degreeof saturation changes from 0.6 to 0.7.

From a mechanical viewpoint, the evolution of the PSDfunction has also important consequences. Fig. 6(a) showsthe time evolution of swelling pressure that was monitoredby means of total pressure cells PE3 and PE7 located in thecentral section – in which the compacted pellet buffer isplaced – of the ‘engineered barrier experiment’ tunnel(Garcıa-Sineriz et al., 2008). The injection of water throughan artificial hydration system is also indicated as comple-mentary information of the progression of the hydraulicstate. The swelling pressure evolution shows clear transientdrops, which can be associated with local collapse phenom-ena of the granular fill. Relative humidity values obtained bysensors emplaced near these total pressure cells allow theestimation of the corresponding total suctions by the psy-chrometric law. Fig. 6(b) shows the in situ stress pathsfollowed by these points PE3 and PE7 in a suction–cellpressure plane. Local swelling pressure drops occur at totalsuctions around 100 MPa. An equivalent result was obtainedfrom constant-volume oedometer tests performed at two drydensities (1.30 and 1.50 Mg/m3). Figs 7(a) and 8(a) presentthe development of vertical, horizontal and mean net stresson hydration of both samples. Again, transient stress reduc-tions are detected in both samples at relatively early hydra-tion stages. Changes in degree of saturation were convertedinto time evolution of suction – assuming local equilibrium– using water retention data. In this way, stress pathsfollowed on constant volume hydration could be plotted inFigs 7(b) and 8(b). Note that the peak in swelling pressureat an intermediate suction is only observed in the lightersample. The stress paths presented in Fig. 7(b) recall thesame pattern of behaviour observed under in situ conditionswith local collapse phenomena developing at a slightly lowersuction in the laboratory specimens (around 10 MPa). Waterretention curves on wetting at constant volume for two drydensities of the mixture and the pellet are given in Fig. 9,which shows the important dependence on void ratio.

Another aspect having important mechanical conse-quences, in which the multi-porosity network plays a crucialrole, refers to the application of different wetting rates tosaturate the mixture. Two techniques were used to performwetting under load tests (at a constant vertical net stress of0.3 MPa) following the same stress paths under oedometerconditions of a specimen with an initial dry density of1.30 Mg/m3. The first technique involves saturation withliquid water (fast wetting) under small pressure (20 kPa),whereas in a second procedure, a vapour equilibrium tech-nique was used (slow wetting). In the test involving wettingwith liquid water, the injected water volume was controlledand suction values were derived from water retention dataassuming local equilibrium of moisture. In the second test,

Pluviated pellets Inter-pellet250 mμ

0·4

0·3

0·2

0·1

0

Por

e si

ze d

ensi

ty fu

nctio

nP

ore

size

den

sity

func

tion

(a)

High-density pellet

Intra-aggregatepores

13 nm Inter-aggregatepores

3 mμ

10

1

0·1

0·0110 1000 100000

Entrance pore size: nm(b)

Fig. 3. Pore size density functions of the pellet-based material(Hoffmann et al., 2007): (a) high-density pellet; (b) pluviatedmixture at a dry density 1.15 Mg/m3

HYDROMECHANICAL BEHAVIOUR OF COMPACTED GRANULAR EXPANSIVE MIXTURES 73


Suction

Loading at constant suction

AB

15 MPa

Vertical stress

0·6

0·4

0·2

0

Por

e si

ze d

ensi

ty fu

nctio

n

10 1000 100000

Entrance pore size: nm

13 nm

BA

(a)

5 mμ

21

mμ

ρdf31·70 Mg/m�

ρd031·45 Mg/m�

Suction

Wetting under load

A

B0·5 MPa

Vertical stress

0·6

0·4

0·2

0

Por

e si

ze d

ensi

ty fu

nctio

n

10 1000 100000


13 nm

B

A

(b)

750

nm

20

mμ

ρd031·60 Mg/m�

ρdf31·35 Mg/m�

As-compacted

Suction

Wetting at constant volume

A

B0·7 MPa

Vertical stress

0·6

0·4

0·2

0

Por

e si

ze d

ensi

ty fu

nctio

n

10 1000 100000


13 nm

B

A

(c)

400

nm

40

mμ

ρd031·35 Mg/m�

ρdf31·35 Mg/m�

Suction: 3 MPaA*

Saturation

Saturated

A*

Fig. 4. Effects of stress paths on pore size density functions for the granular expansive material: (a) loading at constantsuction; (b) wetting under constant load; (c) wetting at constant volume

1 10� �7

1 10� �9

1 10� �11

1 10� �13

Wa

ter

perm

eabi

lity:

m/s

1 10 100 1000 10000 100000Time: s

(a)

kws121 10 m/s� � �

1 10� �7

1 10� �9

1 10� �11

1 10� �13Wa

ter

perm

eabi

lity:

m/s

0 0·2 0·4 0·6 0·8 1·0

Degree of saturation(b)

Fig. 5. Evolution of water permeability during a fast infiltration test at constant volume (Hoffmann et al., 2007): (a) timeevolution; (b) evolution in terms of degree of saturation



the wetting path was initially performed in a stepwisemanner by varying total suction using relative humiditycontrol (down to a total suction 3 MPa). The test ends in asoaking stage. The stress paths followed are indicated in Fig.10. A different behaviour was observed in both tests duringthe saturation process. When liquid water was injected (Fig.11(a)), an initial collapse ‘AB’ was observed immediatelyafter inundation. Then, the expanding granules started toswell and a net swelling deformation was measured ‘BC’. Ata final stage, however, when the specimen was close tosaturation, additional compression deformations were regis-tered, ‘CD’. In the vapour transfer case (Fig. 11(b)), a netswelling deformation was measured from the beginning ofthe test and for all the subsequent vapour equilibrium stages‘AB’ until reaching a suction value of around 3 MPa (point3 in Fig. 10(b)). Then, at the final stage, when liquid waterwas injected to bring suction to zero, a sudden collapse‘BC’ was initially observed followed by some final swellingdeformations ‘CD’. Despite the different techniques used,both samples ended at practically the same volumetric strain(point ‘D’ in Figs 11(a) and 11(b)).

Differences in the observed behaviour can be interpretedas follows. Regarding the fast liquid test, the high initialpermeability of the specimen macrostructure explains itsrapid saturation, when water under some pressure is appliedat the specimen boundaries. Water infiltrates fast into thesample, the inter-pellet voids become saturated and theintergranular forces – which maintain the stable arrangementof pellets – are reduced. Then, a global collapse of thegranular structure takes place (‘AB’ in Fig. 11(a)), themagnitude of which depends on the applied stress level.Later, as the expansive pellets progressively hydrate, adelayed swelling response is measured (‘BC’ in Fig. 11(a)).Water is driven from the open connected macroporosity tothe intra-pellet voids until equilibrium is reached. The

macroscopically observed behaviour is the result of bothmechanisms and their interaction. This explanation includesimplicitly the assumption that two different water potentialsco-exist inside the specimen, which is one of the criteria thatwill guide the development of the physical model describedin the next section. The water filling the macropores isconsidered to have a macro potential not necessarily inequilibrium with the water potential inside the aggregates.Equilibrium will be reached when both water potentialsbecome equal.

However, when the vapour equilibrium technique is used,the large pores remain essentially unsaturated, while water isslowly transferred to the pellets. Under these conditions,macro and micro water potentials remain similar or closetogether along the wetting process. Capillary forces holdingthe macrostructural arrangement slowly change and collapseprogressively evolves. In parallel, the pellets gradually swell,which leads to an overall (net) expansion evolution through-out the initial wetting process (‘AB’ in Fig. 11(b)). Aresidual collapse (‘BC’ in Fig. 11(b)) is developed on furthermacrostructural arrangement induced by liquid water transferbefore attaining a similar volumetric deformation at the finalstage in both tests.

1·6

1·2

0·8

0·4

0

Tota

l str

ess:

MP

a

Canister

PE7PE3

PE7

PE3

Injected water

20

16

12

8

4

0

Inje

cted

wa

ter:

m3

0 200 400 600 800Time: days

(a)400

300

200

100

0

Suc

tion:

MP

a

0 0·4 0·8 1·2 1·6

Total stress: MPa(b)

PE3 PE7

Fig. 6. (a) Measured total stresses (swelling pressures) andinjected water during the hydration phase of the engineeredbarrier (EB) experiment for sensors PE3 and PE7. (b) In situstress paths followed during water injection

2·0

1·6

1·2

0·8

0·4

1·0

0·8

0·6

0·4

0·2

0

0·01 1 100 10000Time: min

(a)S

tres

ses:

MP

aD

egre

e of

satu

ratio

n

Horizontal

Mean

Vertical

ρd031·30 Mg/m�

1000

100

10

1

0·1

0·01

Suc

tion:

MP

a

0 0·4 0·8 1·2 1·6 2·0

Stresses: MPa(b)

Vertical Mean

Horizontal

ρd031·30 Mg/m�

Fig. 7. (a) Time evolution of stresses and degree of saturationduring wetting at constant volume (dry density 1.30 Mg/m3).(b) Stress paths followed during the hydration phase



4·0

3·0

2·0

1·0

1·0

0·8

0·6

0·4

0·2

0

0·01 1 100 10000Time: min

(a)

Str

esse

s: M

Pa

Deg

ree

ofsa

tura

tion

Horizontal

Mean

Vertical

ρd031·50 Mg/m�

1000

100

10

1

0·1

0·01

Suc

tion:

MP

a

0 1·0 2·0 3·0 4·0

Stresses: MPa(b)

Vertical

Mean

Horizontal

ρd031·50 Mg/m�

0·001

Fig. 8. (a) Time evolution of stresses and degree of saturationduring wetting at constant volume (dry density 1.50 Mg/m3).(b) Stress paths followed during the hydration phase

1000

100

10

1

0·1

0·01

Suc

tion:

MP

a

0 0·2 0·4 0·6 0·8 1·0

Degree of saturation

ρd31·30 Mg/m�

ρd31·50 Mg/m�

ρd31·90 Mg/m (pellet)�

Fig. 9. Water retention curves along a wetting path at constantvolume of the pellet and the granular mixtures at different drydensities

�0·06

�0·04

�0·02

0·00

0·02

Vol

umet

ric s

trai

n

0 500 1000 1500 2000 2500

Time: min(a)

2

A

B

C

D

4

4C

D

B

3

2A

0 20000 40000 60000Time: min

(b)

�0·06

�0·04

�0·02

0

Vol

umet

ric s

trai

n

Fig. 11. Time evolution of volumetric strains during infiltrationtests at constant vertical stress (mixture at an initial dry densityof 1.30 Mg/m3): (a) fast flooding test with liquid injection;(b) slow wetting test with vapour transfer technique

(a)

12

4

0·3

300

0

Suc

tion:

MP

a


1–2 Loading to 0·3 MPa2–4 Water injection phase

(b)

1 2

4

0·3

300

0

Suc

tion:

MP

a


1–2 Loading to 0·3 MPa2–3 Multi-stage wetting with vapour3–4 Final flooding stage

33

Fig. 10. Stress paths followed during infiltration tests atconstant vertical stress (mixture at an initial dry density of1.30 Mg/m3): (a) fast flooding test with liquid injection; (b) slowwetting test with vapour transfer technique



A WORKING MODELThe objective of this section is to briefly describe a one-

dimensional (1D) working model implemented in a finite-difference computer code, which will be used to gain insightinto the different wetting-induced phenomena previously out-lined. The detailed description of its implementation isdescribed in Hoffmann (2005). A key aspect in the formula-tion is to adopt a physical representation of the real medium,which allows the capture of those transient and local phe-nomena that are far from the measurement scale used inconventional laboratory tests. The granular arrangement ofpellets is substituted by an equivalent porous medium con-sisting of a collection of 1D elements assembled together toemulate the granular packing. In line with the ideas outlinedabove, each element has two structural components repre-senting the macro- and the microstructural level. Structuralparts of each element are connected in a hierarchical wayspecifying that the microstructure is enclosed by the macro-structure. An important consequence of this assumption isthat all the assembled elements are connected through theirmacrostructures and no connection is admitted between

microstructural parts of two adjacent elements. When aspecimen is subjected to liquid water wetting, water flowsthrough the open interconnected macroporosity and pellets(microstructure) become hydrated by exchanging water withthe macrostructure. Fig. 12 shows the real media (Fig. 12(a))and the equivalent 1D one (Fig. 12(b)), in which the macro-structural void ratio eM (interconnected and outside thepellets) is separated from the microstructural one em (insidepellets). In Fig. 12(d), the hydration process is also schema-tically represented by means of a macrostructural flow andwater exchanges between macro- and microstructural levels.

The guiding criteria behind the working model are de-scribed below, giving five main points.

Two different structural levels are considered. A micro-structural level associated with the pore volume inside thepellets and a macrostructural one related to the granulararrangement of pellets. The microstructural void ratio (asso-ciated with the pore volume inside pellets) can be estimatedfrom the dry density of the pellets em ¼ rs=rdp � 1, wherers and rdp are the solid and pellet dry density, respectively.Figs 12(b) and 12(c) show the macrostructural elements and

eM

em

Flow

i N 1� �

UN�1

Micro

Macro

DX

Ui

Xi

i 1�

U1 0�pw

X1 0�

(c)

Stressordisplacement

σv constant�

UN�1 0�

Water pressure, constantDisplacement 0

pU

w

1

�

�(a) (b)

Macro

FlowMacro

i N 1� �

Micro

Water exchange

Macro–micro

i 1�

(d)

DX

Macro

Xi�1 Ui�1

xmi umi

Mic

ro*

DX

Xi Ui

(e)

εvi � �( )

( )

U U

X X

i i

i i

�

�

1

1

�

��

( )U U

DX

i i

i

�1 ��

Δe

e(1 )�

εvMi � �( )

( )

U um

X X

i i

i i

�

�

1

1

�

��

( )U um

DX

i i

i

�1 ��

Δe

e

M

(1 )�

εvMi � �( )

( )

Δum

X X

i

i i�1 �

( )

( )

U U

X X

i i

i i

�

�

1

1

�

�� ε εv vMi i

(f)

Fig. 12. Equivalent 1D double-porosity media used in the working model: (a) real medium; (b) equivalent 1D medium;(c) space discretisation; (d) scheme of the hydration process; (e) nodal displacements; and (f) strain and nodaldisplacement relationships



the embedded microstructural ones. Nodal displacements Ui

and microstructural ones umi are indicated in Fig. 12(e).Strain and nodal displacement relationships for both levelsare also summarised in Fig. 12(f).

Two complementary water potential fields. The discussionin a previous section suggested that a single water potential

field was not enough to interpret the transient behaviour ofexpansive materials with a marked multi-modal porositydistribution. Therefore, in addition to the net stress, twocomplementary water potential fields are assumed to coexistin the working model. One is associated with the wateradsorbed inside pellets –microstructural suction sm – while

sM

Macrostructure

Initial LC0LC1

Final

σ

Microstructure

sm

SD0

SD1

Initial

Final

σ

Fig. 13. Wetting paths at constant stress plotted in the macrostructural suction (sM) and microstructural suction (sm)planes. Activation of macrostructural yield locus LC and microstructural yield locus SD during wetting

Table 2. Micro- and macrostructural hydromechanical parameters used in the simulation (refer to the Appendix)

Mechanical microstructural parameters

km ºm � SDo (initial): MPa

0.015 0.035 0.9 70–100

Hydraulic microstructural parameters

kim: m2 ºp Sresm pwm: MPa ºwm psm: MPa ºsm

1 3 10�22 1.5 0.01 16.6 0.130 3000 0.1

Macrostructural reversible compressibility parameters

Dry density, rd :Mg/m3

kpo Æp kso Æsp Æss: MPa�1 pref : MPa

1.30 0.07–0.09 �0.003 0.001 0.185 0.010 0.0081.50 0.05–0.06 �0.003 0.025 0.120 0.003 0.008

Macrostructural LC yield locus parameters

Dry density, rd :Mg/m3

��vo: MPa � c: MPa º(0) r �: MPa�1

1.30 0.022–0.065 0.08 0.18–0.20 0.45–0.65 0.0071.50 1.5–1.9 0.08 0.16–0.17 0.65–0.85 0.004

Hydraulic macrostructural parameters

Dry density,rd : Mg/m3

kiM: m2 ºp SresM pwM: MPa ºwM psM: MPa ºsM

1.30 1 3 10�16 1.5 0.01 0.95 0.217 1000 0.021.50 1 3 10�16 1.5 0.01 4.15 0.237 2500 0.001



the other is related to the water stored between granules –macrostructural suction sM – which has a dominant capillaryretention component. Lack of local equilibrium between bothsuctions may occur depending on the wetting rate andboundary conditions. Different laboratory protocols lead todifferent histories of macro- and microstructural changes andtherefore to different transient specimen responses. For thepurpose of discussing the experimental behaviour observedunder oedometer conditions, it is convenient to work withtwo stress spaces: the space (�, sM) associated with themacrostructure and the space (�, sm) associated with themicrostructural level (Fig. 13).

Different volume change mechanisms are defined for eachstructural level. The ideas put forward by Alonso et al.(1990) in the Barcelona basic model (BBM) are used todescribe the volume change mechanisms acting at macro-structural level. On the other hand, the volume changeresponse of the microstructure is described with the elasto-plastic model proposed by Gens & Alonso (1992) andAlonso et al. (1999), which extends the BBM framework toexpansive materials. The positions of the yield loci in Fig.13 mark the initiation of activation of irreversible strains:the LC curve in the case of the macrostructural irreversiblecompression and the suction-decrease (SD) curve in the caseof microstructural changes (irreversible expansion). A sum-mary of the model formulation for volume change underisotropic stress state is given in the Appendix. More detailscan be found in Alonso et al. (1999).

Double-structure approach. In order to reproduce thehydraulic response of the bentonite pellet mixture, a double-structure approach was adopted. Advective fluxes of water ineach level are computed using a generalised Darcy’s law(equation (1) below), which incorporates a relative per-meability kr as a function of the micro- and macrostructuraldegrees of saturation (Srm and SrM) and intrinsic permeabilityvalues ki depending on the respective micro- and macrostruc-tural void ratio (em and eM respectively). Srm and SrM arerelated to their corresponding suctions, sm and sM, throughtwo independent modified van Genuchten’s (1980) retentioncurves (equation (2) below), in which pwm and pwM are theair-entry values corresponding to the micro- and macrostruc-ture, respectively; Sresm and SresM are the residual degrees ofsaturation for micro- and macrostructure; and psm, psM, ºsm

and ºsM are experimental parameters. Relative permeabilityis defined as a potential function of degree of saturation(micro or macro) with a common power parameter ºp

kwm ¼ krm(Srm) kim(em);

kwM ¼ krM(SrM) kiM(eM)

with krm ¼ (Srm)ºp

and krM ¼ (SrM)ºp

(1)

Srm ¼ Sresm þ 1� Sresmð Þ 1þ sm

pwm

� �ºwm= 1�ºwmð Þ" #

3 1� sm

psm

� �ºsm

SrM ¼ SresM þ 1� SresMð Þ 1þ sM

pwM

� �ºwM= 1�ºwMð Þ" #

3 1� sM

psM

� �ºsM

(2)

The granular media is conceived as a continuum havingthe properties of the assembly of pellets. Water flows in the

Darcy sense through the (macro) pores of the granularmedia. The mass balance equation of water in the macro-continuum reads

@

@ t�MSrMð Þ � = � kM=jMð Þ ¼ f w

M!m (3)

where M refers to the macro component. The rate of changeof porosity �M with time requires the calculation of volu-metric strains, which is made through the BBM. Theydepend on stress and macro suction changes. The macroretention curve (equation (2)) allows also the calculation ofthe rate of changes of degree of saturation. Macro per-meability has already been defined in terms of void ratioand degree of saturation. It may be calculated for changes insuction and stress.

The connection of the macro analysis with individualpellets is included in the source/sink term f w

M!m whichdescribes the water flow rate, wetting or drying the pellets.Equation (3) may eventually be written in terms of macrosuction (the gravitational component of potential jM is neg-ligible in the tests performed).

kwm

kwM

kw (test)

em

eM

e

(b)

(a)

Srm

SrM

Sr

Sr (test)

1·0

0·8

0·6

0·4

0·2

1·2

1·0

0·8

0·6

0·4

0·2

1 10� �7

0

1 10� �9

1 10� �11

1 10� �13

Deg

ree

ofsa

tura

tion

Voi

d ra

tio, e

Wa

ter

perm

eabi

lity,

: m/s

k

1 10 100 1000 10000Elapsed time: s

(c)

Fig. 14. Fast infiltration test at constant volume (dry density1.30 Mg/m3): (a) simulated evolutions of the different compo-nents of degree of saturation and comparison with overalldegree of saturation results; (b) simulated evolutions of thedifferent components of void ratio; (c) simulated evolutions ofthe water permeability (micro- and macrostructural). Compari-son with test results



Flow of water hydrating and wetting every pellet in themodel is also described as a boundary value problem (BVP).Suction sM provides the boundary condition for the waterflow in pellets. Water mass balance equation inside a pelletis given by

@

@ t�mSrmð Þ � = � km=jmð Þ ¼ 0 (4)

similar to equation (3) but now in terms of microstructuralvariables. Note that no sink or source term is included inequation (4) because the water interchange with the macro-porosity is made at the pellet’s boundaries. Deformationsinside the pellets are calculated through the expansivemodel. Equation (4) may eventually be written in terms ofmicro suction sm by neglecting the gravitational componentof potential jm.

The oedometric tests performed were simulated by the 1Dmodel sketched in Fig. 12. Stress equilibrium is trivial in

this case. Fig. 12(d) provides the definition of micro andmacro deformations in terms of the calculated nodal displa-cements for the two BVPs described.

The calculation process is described as follows:

(a) The macro BVP is first solved for the first timeincrement. Water sinks or sources from pellets areassumed to be zero. Likewise pellet deformations areset to zero. The solution provides a first spatialdistribution of macro-suction, sM(x).

(b) A local BVP is solved for each one of the pelletsincluded in the sample (N pellets in Fig. 12). Theboundary suction for each pellet is given by sM(x) fromthe first step (a). The calculation provides, for eachpellet, the flow in or out ( f w

M!m ) and the pellet globaldeformation (˜um ).

(c) Pellet’s boundary flow and deformations are comparedwith the initial calculation (or the previous calculationin the iterative process). Global pellet expansion is also

1000

100

10

1

0·1

0·01

SM

: MP

a

sM0 300 MPa�

LC0 LC1

Stage I1 2sM1 0·01 MPa�

10 100 1000 10000

Vertical net stress: kPa

1000

100

10

1

0·1

0·01

Sm

: MP

a

sm0 300 MPa�

SD0

Stage I

10 100 1000 10000


1000

100

10

1

0·1

0·01

SM

: MP

a

LC1 LC2

Stage IIsM1 0·01 MPa�

10 100 1000 10000


1000

100

10

1

0·1

0·01

Sm

: MP

a

sm0 300 MPa�

SD1

Stage II

10 100 1000 10000


sm1 287 MPa�

1000

100

10

1

0·1

0·01

SM

: MP

a

LC2 LC3

Stage IIIsM1 0·01 MPa�

10 100 1000 10000


1000

100

10

1

0·1

0·01

Sm

: MP

a

sm1 0·01 MPa�

SD2

Stage III

10 100 1000 10000


sm1 287 MPa�

SD3

0

Fig. 15. Simulated stress paths in the vertical net stress–suction planes: (�v, sM) and (�v, sm) for the fast flooding test(liquid injection) at constant vertical stress (mixture at an initial dry density of 1.30 Mg/m3). Evolutions of LC and SDyield loci



compared with the initial calculation (or the previouscalculation in the iterative process). If differences aresmall, variables are updated, a new time step is addedand the calculation resumes in step (a). Otherwise aninternal calculation loop through steps (a), (b) and (c)is followed until convergence.

The following additional data and remarks define thecalculations made.

(a) Water retention curves for the pellets and the mixtureare given in Fig. 9.

(b) The dependence of permeability on void ratio wasderived from saturated permeability data on samplescompacted at different dry densities. The effect ofdegree of saturation was directly determined by meansof some percolation tests (Fig. 5).

(c) The size of the pellet is given by DX� in Fig. 12(e).DX� describes the pellet radius. Its value is defined bya characteristic size (D50 of grain size distribution curvewas selected). Once DX� is defined the macro volumeof the element is derived from the total density of thespecimen.

(d ) The number of elements (pellets) in the model of anoedometer test derives from the condition that themaximum pellet size (Dmax ) in the pellet grain sizedistribution is no bigger that one-fifth of the specimenthickness. Then, the number of elements is given byN ¼ 5Dmax=D50.

SIMULATION-AIDED INTERPRETATION OF TESTINGRESULTS

This section shows the capability of the physical model tosimulate the tests described in the section headed ‘Behaviourof pellets and pellet mixtures’. The simulations will allowthe capture of transient and local processes not detected atconventional laboratory scale. For example, by providing theevolutions of micro- and macrostructural suctions (or alter-natively the progressive changes of micro- and macrostruc-tural degrees of saturation), the model will assist in thephysical interpretation of such complex phenomena that arebeyond measurable scale. A systematic procedure was usedto obtain the set of parameters required from a comprehen-sive and independent experimental programme describedelsewhere (Hoffmann, 2005; Hoffmann et al., 2007). Materi-al parameters associated with the microstructure were ob-tained from tests performed on compacted samples at thedry density and initial water content of the bentonite pellets(Table 1). Table 2 summarises the set of parameters used inthe computations.

Infiltration test at constant volume. Water permeabilitydecrease due to macropore occlusion

In order to interpret the hydraulic phenomena involved –specifically to account for water permeability variationsassociated with macroporosity changes induced by pelletswelling – the infiltration test under constant volume (Fig.5), performed on a mixture with dry density of 1.30 Mg/m3,was simulated. Water was injected at a constant backpressureof 20 kPa. Calculated evolutions of volumetric variables(void ratio and degree of saturation both at micro- andmacroscale) are plotted against time in Figs 14(a) and 14(b).These tests help to elucidate the expected evolution of soilstructure, which is information not readily available fromconventional laboratory techniques.

The model shows the fast increase of SrM as liquid waterfloods the macropores (only simulated results beyond a time

t ¼ 1 s are plotted, SrM is already close to 0.70). Theextremely open macroporosity helps injected water to flowthrough the macropores, causing this structural level tobecome rapidly saturated (Fig. 14(a)). On the contrary, thedense microstructural level, which undergoes swelling asshown in Fig. 14(b), is hydrated at a much slower rate. Theevolution of the overall degree of saturation Sr, whichfollows closely the experimental results, is calculated fromboth micro- and macrostructural contributions as

Sr ¼em

eSrm þ

eM

eSrM (5)

As shown in Fig. 14(c), the model is able to reproduce ina satisfactory way the evolution on wetting of the macro-scopic water permeability kwM. This water permeability isinitially expected to increase due to the fast increase of SrM

(this condition is not detected in the figure since SrM alreadystarts at a relatively high value). After macropore saturation,the kwM evolution consistently follows the important waterpermeability drop obtained in the experimental data pre-sented in Fig. 14(c). It is seen how kwM is graduallydeclining due to pellet swelling, which causes the progres-sive occlusion of the macroporosity (Fig. 14(b)).

Influence of the wetting rate. Liquid against vapourimbibition

Oedometer results obtained during wetting under constantvertical net stress of 0.3 MPa, performed on low-densitymixtures (dry density 1.30 Mg/m3) are now interpreted usingthe model. A striking characteristic of these two tests withdifferent water absorption rates was the different volumetricfeatures observed on identical samples, despite being sub-jected to the same overall stress–suction path (Fig. 10). The

1000

100

10

1

0·1

0·01

Suc

tion:

MP

a

s0 300 MPa�

sM1 0·01 MPa�

sM

sm

(a)ExperimentaldataSimulation

Sta

ge I

Sta

ge II

Sta

ge II

I

�0·06

�0·04

�0·02

0

0·02

Ver

tical

str

ain

0·01 1 100 10000 1000000Time: s

(b)

Fig. 16. Fast flooding test (liquid injection) at constant verticalstress (mixture at an initial dry density of 1.30 Mg/m3):(a) simulated time evolutions of micro- and macrostructuralsuctions; (b) simulated time evolution of vertical strain andcomparison with experimental results



Experimental data

Simulation

�0·06

�0·04

�0·02

0

0·02

Ver

tical

str

ain

0 500 1000Time: h

(a)

1000

100

10

1

0·1

0·01

Suc

tion:

MP

a

SM

SmVapour transfer

Fast flooding(water injected)

0 500 1000Time: h

(b)

Fig. 18. Slow wetting test (vapour transfer technique) at constant vertical stress (mixture at an initial dry density of1.30 Mg/m3): (a) simulated time evolution of vertical strain and comparison with experimental results; (b) simulatedtime evolutions of micro- and macrostructural suctions

1000

100

10

1

0·1

0·01

SM

: MP

a

sM0 300 MPa�

LC0

LC1

Stage I

sM1 3 MPa�

10 100 1000 10000Vertical net stress: kPa

1000

100

10

1

0·1

0·01

Sm

: MP

a

sm0 300 MPa�

SD0

Stage I

10 100 1000 10000


1000

100

10

1

0·1

0·01

SM

: MP

a

LC1

Stage II

sM1 0·01 MPa�


1000

100

10

1

0·1

0·01S

m: M

Pa

sm0 300 MPa�

SD1

Stage II


sm1 3 MPa�

1000

100

10

1

0·1

0·01

SM

: MP

a LC2

LC1

Stage IIIsM2 0·01 MPa�

10 100 1000 10000


1000

100

10

1

0·1

0·01

Sm

: MP

a

sm2 0·01 MPa�

SD2

Stage III

10 100 1000 10000


sm0 300 MPa�

SD1

SD1sm1 3 MPa�

sM0 300 MPa�

sM1 3 MPa�

sM1 300 MPa�

sm1 3 MPa�

Fig. 17. Simulated stress paths in the vertical net stress–suction planes: (�v, sM) and (�v, sm) for the slow wetting test(vapour transfer technique) at constant vertical stress (mixture at an initial dry density of 1.30 Mg/m3). Evolutions ofLC and SD yield loci



application of the model will provide the evolution of sM

and sm and will help to explain the overall volume changeresponse observed.

Regarding the liquid water test or fast flooding experiment(Figs 10(a) and 11(a)), the extremely open macroporositycontributes to the flow of water along the macropores,causing this structural level to become saturated in a rel-atively short period. A consequence of this phenomenonimplies that the two defined suctions inside the mixtureevolve at different rates. Suction evolutions are characterisedby an important and fast reduction of sM (associated withthe development of collapse) and a delayed decrease of sm

inside the dense pellets inducing swelling. This is indicatedin Fig. 15, in which the evolutions of the two definedsuctions inside the specimen are represented in two verticalnet stress–suction planes: (�v, sM) and (�v, sm). Stage I inFigs 15 and 16(a) represents the initial part of the test,where the macrostructural level becomes essentially saturated

(sM ¼ 0), and the expanding pellets remain at the initialsuction (smo) and, accordingly, do not swell. On macrostruc-tural suction reduction, conditions for a potential collapse ofthe sample develop. In the constitutive model this collapse iscontrolled by the location of the yield surface (LC0) in Fig.15 (left) associated with the granular packing. The initialposition of LC0 is defined by the as-compacted condition interms of the initial suction of the mixture (Table 1) and thevertical stress applied on compaction (Fig. 2). As the LC1

yield surface moves to position LC2, dragged by the decreas-ing macro suction, a collapse is predicted as shown in Fig.16(b). After this initial first stage, a water exchange betweenboth levels starts and the dense bentonite granules begin toswell as the microstructural suction decreases (stage III inFig. 16(b)). The microstructural suction reduces in search ofthe ‘boundary condition’ imposed by sM. During this sm

reduction, both elastic and plastic strains are computed.Plastic strains – associated with sm decrease – are activated

1000

1000

1000

1000

1000

1000

100

100

100

100

100

100

10

10

10

10

10

10

1

1

1

1

1

1

0·1

0·1

0·1

0·1

0·1

0·1

0·01

0·01

0·01

0·01

0·01

0·01

10

10

10

10

10

10

100

100

100

100

100

100

1000

1000

1000

1000

1000

1000

10000

10000

10000

10000

10000

10000







SM

: MP

aS

M: M

Pa

SM

: MP

a

Sm

: MP

aS

: MP

am

S: M

Pa

m

SM

1

0·02M

Pa

�S

M1

0·02M

Pa

�S

M1

0·02M

Pa

�

SM0 300 MPa�

SM0 300 MPa�

SM0 300 MPa�

S 0 300 MPa�m

S 0 300 MPa�m

S 0 300 MPa�m

S 1 130 MPa�m

S 0 2 MPa�m

S 2 2 MPa�m

LC0

LC1

LC2

LC0

LC0

SD0

SD0

SD3

SD2

SD2

Stage I

Stage II

Stage III

Stage I

Stage II

Stage III

0

0

0

1

1

1

1

1

2

2

2

2

0

0

0

Fig. 19. Simulated stress paths followed by the mixture at dry density 1.30 Mg/m3 during wetting at constant volume,using the two vertical net stress–suction planes ((�v, sM) and (�v, sm)). Evolutions of LC and SD yield loci



on dragging the SD yield locus (Fig. 15 (right)). Relevantequations are given in the Appendix.

In the vapour transfer wetting test (slow water absorptiontest) only net swelling deformations were recorded from thevery beginning of the hydration process (Fig. 11(b)). In thiscase, no substantial differences between the two suctioncomponents are expected during the wetting process. This isshown in the simulation reported in Figs 17 and 18(b).Virtually, nearly local equilibrium conditions are prevailingthroughout the process. The stress paths followed in the twovertical net stress–suction planes ((�v, sM) and (�v, sm)) havebeen represented in Fig. 17. During the vapour transferstages (stages I and II in Fig. 17) some collapse deforma-tions (associated with sM changes and the activation of LCyield locus) and important swelling deformations (associatedwith sm changes and the dragging of SD yield locus) takeplace simultaneously. A net volumetric expansion is com-puted, which corresponds to the observed behaviour (Fig.18(a)). The simulation also captures the final volumetriccompression, associated with the activation of the LC curveduring the final flooding stage, as well as the delayedmicrostructural swelling (‘BC’ and ‘CD’ in Fig. 11(b)).

Swelling pressure testsTests results were given in Figs 7 and 8 for mixtures at

dry densities of 1.30 and 1.50 Mg/m3 respectively. Thesimulated response, plotted in the two vertical net stress–suction planes ((�v, sM) and (�v, sm)), is given in Fig. 19and later in Fig. 21. In the case of the sample at a drydensity of 1.30 Mg/m3, the macrostructural suction decayedrapidly and the macropore level became saturated in arelatively short period, because of the extremely open macro-porosity (Fig. 19 (left) and Fig. 20(a)). On the contrary, themicrostructural suction decreased at a much slower rate, as

shown in Figs 19 (right) and 20(a). The macrostructuralstress path in Fig. 19 (left) indicates that on suction reduc-tion no LC yield curve is activated, and the stable granularpacking undergoes an elastic vertical net stress increasenecessary to maintain the constant volume condition. Themicrostructural stress path shows increasing swelling pres-sures on suction reduction. It induced some small finaldragging of the LC yield curve in order to maintain the nullvolume change condition. Despite this final LC yield curveactivation, no decrease of the swelling pressure was detected,against the observed measurements (Fig. 20(b)). The swel-ling pressure increased monotonically to an ultimate valuearound 0.6 MPa, which reproduces well the final experimen-tal value.

In the high-density mixture (1.50 Mg/m3), the macrostruc-tural suction decreased at a lower rate compared to the low-density mixture, because of its lower macrostructural waterpermeability (Fig. 21 (left) and Fig. 22(a)). Fig. 21 (left)shows how the saturation path brings the macrostructuralstress state towards the yield locus LC as a consequence ofthe important swelling pressure developed. The collapsetendency experienced by the macrostructure, when the LCyield locus is activated, is compensated for by the swellingtrend at microscopic scale (Fig. 21 (right)) to maintain thenull volume change condition. Some volumetric strain hard-ening occurred (activation of the LC in Fig. 21 (left)) due tosome reduction in the macrostructural pore space at theexpense of the microstructural swelling. An interestingaspect to remark, which is detected in Fig. 22, is thecalculated suction change during wetting. The macrostruc-tural suction decreases monotonically down to a low value(around sM ¼ 0.05 MPa), but afterwards the decaying trendis reverted. At this point, the still high value of sm demandswater, which is supplied by liquid stored in the macrostruc-ture. This condition originates a slight increase of themacrostructural suction along stage II in Fig. 22, which isfinally reverted towards a monotonic decrease during stageIII. This unusual condition is not easy to detect with conven-tional measuring techniques. But if it is correct it helps tounderstand why the swelling pressure tends to level off andthen re-activate its increasing tendency (stage II in Fig.22(b)).

SUMMARY AND CONCLUSIONSDry-side compacted granular expansive mixtures have

been investigated by different macroscopic tests (fast andslow infiltration tests, swelling under load and swellingpressure tests) to study their complex behaviour on wetting.This response arises from its multi-modal pore networkexhibiting distinct pore sizes, which are controlled by thehydromechanical paths followed during testing.

A most important feature of behaviour is the progressiveloss of water permeability induced by the blockage of thelarge inter-pellet pores due to granule swelling and thedevelopment in time of swelling strains and swelling pres-sures during wetting.

Experimental results and some microstructural observa-tions seem to be consistent with the idea that two differentwater potentials (suctions), not necessarily in equilibrium,may coexist at a given time. Nevertheless, this sub-divisioninto two suction components is far from being measured byconventional laboratory procedures, and alternative techni-ques for the interpretation of the observed phenomena arerequired. Within this context, simulation-assisted techniquesare essential.

The paper presented the rationale behind a physicallybased and double-porosity 1D model, which was furtherused successfully to simulate selected experimental results.

1000

100

10

1

0·1

0·01

Suc

tion:

MP

a

S0� 300 MPa

SM

Sm

(a)

s0·

01 M

Pa

�

1 100 10000 1000000Time: s

(b)

Sta

ge I

Sta

ge II

Sta

ge II

I

ρd031·30 Mg/m�

Experimental data

Simulation

800

600

400

200

0

Ver

tical

net

str

ess:

kP

a

Fig. 20. Swelling pressure test (mixture at dry density 1.30 Mg/m3): (a) simulated time evolutions of micro- and macrostructur-al suctions; (b) simulated time evolution of vertical stress andcomparison with experimental results



However, the most appropriate use of the working modelcame up when assisting in the interpretation of competingand coupled processes and by providing further insight intothe phenomena, which were beyond the measurable scaleused in conventional laboratory techniques.

ACKNOWLEDGEMENTSThe work described has been supported by ENRESA

through the ‘Engineered barrier emplacement experiment inopalinus clay (EB experiment)’ (2000–2003). The authorsalso acknowledge the financial support provided by the ECunder the contract FIKW-CT-2000-00017.

APPENDIX. SUMMARY OF MODEL FORMULATIONFOR ISOTROPIC STRESS STATES

The proposed model integrates two different behavioural featuresfor both structural levels. At macrostructural level (rearrangement ofthe granular structure), the elastoplastic BBM model (Alonso et al.,1990) is adopted. It assumes that volume change for isotropic states

is controlled by two independent stress variables: mean net stress pand macrostructural suction sM. On the other hand, soil fabric atmicrostructural level is assumed to react in a pure volumetric andelastoplastic manner against changes in generalised microstructuraleffective stress pp (Gens & Alonso, 1992; Alonso et al., 1999). In themodel developed, the net mean stress, p, was replaced by vertical netstress � when using the 1D working model for the interpretation ofoedometer results.

Elastic deformationsA non-linear law provides microstructural volumetric changes

d�evm ¼ �

dem

1þ em

¼ km

1þ em

dpp

pp¼ dpp

Km

(6)

where km is a constant compressibility index of the microstructureand pp is the generalised microstructural effective stress defined as

pp ¼ pþ �sm (7)

The elastic behaviour at macrostructural level is expressed by

1000

1000

1000

1000

1000

1000

100

100

100

100

100

100

10

10

10

10

10

10

1

1

1

1

1

1

0·1

0·1

0·1

0·1

0·1

0·1

0·01

0·01 0·01

0·01

0·01

0·01

10

10

10

10

10

10

100

100

100

100

100

100

1000

1000

1000

1000

1000

1000

10000

10000

10000

10000

10000

10000







SM

: MP

aS

M: M

Pa

SM

: MP

a

S: M

Pa

mS

m: M

Pa

S: M

Pa

m

SM1 0·05 MPa�

SM1 0·05 MPa�

SM2 0·07 MPa�

SM0 300 MPa�

SM0 300 MPa�

SM0 300 MPa�

S 0 300 MPa�m

Sm0 300 MPa�

S 0 300 MPa�m

S 1 90 MPa�m

SM0 35 MPa�

Sm2 35 MPa�

LC0 LC1

LC1

LC2

LC2

LC3

SD0

SD1

SD1

SD3

SD2

SD2

SD3

Stage I

Stage II

Stage III

Stage I

Stage II

Stage III

0

0

1

1

1

1

2

2

2

2

0

100000

100000

100000

1

33

Fig. 21. Simulated stress paths followed by the mixture at dry density 1.50 Mg/m3 during wetting at constantvolume, using the two vertical net stress–suction planes ((�v, sM) and (�v, sm)). Evolutions of LC and SD yieldloci



d�evM ¼ �

deM

1þ e¼ d p

KpM

þ dsM

KsM

(8)

where KpM ¼ (1þ e) p=kp and KsM ¼ (1þ e)(sM þ patm)=ks are theelastic bulk moduli with respect to p and sM changes (patm is theatmospheric pressure). Compressibility moduli against p and sm

changes are kp(sM) ¼ kpo(1þ ÆpsM) and

ks(sM, p) ¼ kso 1þ Æspln sM

p

pref

� �� eÆss sM

Plastic deformations and hardening lawsIrrecoverable compressive deformations at macrostructural level

induced on activation of the LC yield curve on loading and wetting(Fig. 13), are given by

d� pvM ¼

º 0ð Þ � kp sMð Þ �

1þ e

d po

po

(9)

po ¼ pc p�opc

! º 0ð Þ � k sMð Þº sMð Þ � k sMð Þ

(10)

º sMð Þ ¼ º 0ð Þ r þ 1� rð Þe��sM

�(11)

where º(sM) is the slope of the virgin compression line at suction sM.The hardening law for LC is

d p�op�o¼ 1þ eMð Þ

º 0ð Þ � kp sMð Þd� p

vM (12)

The SD (suction-decrease) yield locus, corresponding to themicrostructural level, is considered to be parallel to the neutral lineat constant pp (Fig. 13). Irrecoverable expansive deformations atmicrostructural level associated with the activation of the SD yieldlocus on wetting are given by

SD0 � pp ¼ 0 (13)

d� pvm ¼

ºm � kmð Þ1þ em

dpp

pp(14)

Hardening of SD yield surface is governed by the plasticmicrostructural strain

dSDo

SDo

¼ 1þ emð Þºm � km

d� pvm (15)

Parameters of the model (km, �, ºm, kpo, Æp, Æsp, Æss, pref , º(0), r, �,pc) are summarised in Table 2.

REFERENCESAlonso, E. E. & Hoffmann, C. (2007). Modelling the field behav-

iour of a granular expansive barrier. Physics Chem. Earth 32,Nos 8–14, 850–865.


Alonso, E. E., Vaunat, J. & Gens, A. (1999). Modelling the mech-anical behaviour of expansive clays. Engng Geol. 54, No. 1–2,173–183.

Garcıa-Sineriz, J. L., Rey, M. & Mayor, J. C. (2008). The engi-neered barrier experiment at Mont Terri Rock Laboratory.Andra, Sci. Technol. Ser. 334, 65–75.

Gens, A. & Alonso, E. E. (1992). A framework for the behaviourof unsaturated expansive clays. Can. Geotech. J. 29, No. 6,1013–1032.

Hoffmann, C. (2005). Caracterizacion hidromecanica de mezclas depellets de bentonita. Estudio experimental y constitutivo (inSpanish). PhD thesis, Universitat Politecnica de Catalunya,Barcelona. See http://www.tdx.cat/TDX-0518105-163209 forfurther details (accessed 03/09/2010).

Hoffmann, C., Alonso, E. E. & Romero, E. (2007). Hydro-mechani-cal behaviour of bentonite pellet mixtures. Physics Chem. Earth32, Nos 8–14, 832–849.

Lloret, A., Villar, M. V., Sanchez, M., Gens, A., Pintado, X. &Alonso, E. E. (2003). Mechanical behaviour of heavily com-pacted bentonite under high suction changes. Geotechnique 53,No. 1, 27–40, doi: 10.1680/geot.2003.53.1.27.

van Genuchten, M. T. (1980). A closed-form equation for predictingthe hydraulic conductivity of unsaturated soils. Soil Sci. Soc.Am. J. 44, No. 5, 892–898.

1000

100

10

1

0·1

0·01

Suc

tion:

MP

aS0� 300 MPa

SM

Sm

(a)

s f0·

01 M

Pa

�

100 10000 1000000 1000000000Time: s

(b)

Sta

ge I

Sta

ge II

Sta

ge II

Iρd0

31·50 Mg/m�

Experimental data

Simulation

3000

2000

1000

0

Ver

tical

net

str

ess:

kP

a

Fig. 22. Swelling pressure test (mixture at dry density 1.50 Mg/m3): (a) simulated time evolutions of micro- and macrostructur-al suctions; (b) simulated time evolution of vertical stress andcomparison with experimental results



Boyd, J. L. & Sivakumar, V. (2011). Geotechnique 61, No. 4, 345–363 [doi: 10.1680/geot.2011.61.4.345]

87

Experimental observations of the stress regime in unsaturated compactedclay when laterally confined

J. L. BOYD� and V. SIVAKUMAR†

Construction processes often involve reformation of thelandscape, which will inevitably encompass compaction ofartificially placed soils. A common application of fillmaterials is their use as backfill in many engineeringapplications, for example behind a retaining wall. Thepost-construction behaviour of clay fills is complex withrespect to stresses and deformation when the fills becomesaturated over time. Heavily compacted fills swells signifi-cantly more than the lightly compacted fills. This willproduce enhanced lateral stresses if the fill is laterallyrestrained. The work presented in this paper examineshow the stress regime in unsaturated clay fills changeswith wetting under laterally restrained conditions. Speci-mens of compacted kaolin, with different initial condi-tions, were wetted to various values of suction under zerolateral strain at constant net overburden pressure whichallowed the concept of K0 (the ratio between the nethorizontal stress and the net vertical stress) to be exam-ined. Tests were also carried out to examine the tradi-tional concept of the earth pressure coefficient ‘at rest’under loading and unloading and its likely effects on thestress–strain properties. The results have shown that thestress regime (i.e. the lateral stress) changes significantlyduring wetting under laterally restrained conditions. Themagnitude of the change is affected by the initial condi-tion of the soil. The results have also indicated that theearth pressure coefficient ‘at rest’ during loading (underthe normally consolidated condition) is unaffected bysuction and such loading conditions inevitably lead to thedevelopment of anisotropic stress–strain properties

KEYWORDS: retaining walls; stress path; suction

Les procedes de construction comportent souvent unealteration du paysage, comprenant inevitablement le tass-ement de sols places artificiellement. Une applicationcommune des materiaux de remblayage est leur utilisa-tion comme remblai dans de nombreuses applicationsd’ingenierie, par exemple derriere un mur de soutene-ment. Le comportement post-construction des remblaisd’argile est complexe, en ce qui concerne les contrainteset la deformation, lorsque les remblais finissent par etresatures. Les remblais fortement tasses gonflent beaucoupplus que des remblais legerement compactes, ce quiengendre un renforcement des contraintes laterales si leremblai est encastre lateralement. Les travaux decritsdans la presente communication examinent la facon dontle regime de contrainte dans des remblais d’argile nonsatures evolue en fonction de leur mouillage, lorsqu’ilssont encastres lateralement. Pour ceci, on procede aumouillage de specimens de kaolin compacte, presentantdifferentes conditions initiales, avec differentes valeursd’aspiration sous une deformation laterale nulle avec unepression de recouvrement net constante, qui permettaitl’examen du concept de K0 (ratio entre la contraintehorizontale nette et contrainte verticale nette). On aegalement effectue des essais pour examiner le concepttraditionnel du coefficient de pression de la terre « aurepos », en condition de sollicitation et de decharge, etses effets probables sur les proprietes de contrainte-deformation. Les resultats indiquent que le regime decontraintes (autrement dit les contraintes laterales) variede facon significative sous l’effet du mouillage, dans desconditions d’encastrement lateral. L’etat initial du solaffecte l’envergure de ces variations. Les resultats indi-quent egalement que le coefficient de pression de la terre« au repos », au cours de la sollicitation (en condition deconsolidation normale), n’est pas affecte par l’aspiration,et que ces sollicitations donnent lieu inevitablement audeveloppement de proprietes de contrainte-deformationanisotropes.

INTRODUCTIONConstruction processes often involve reformation of thelandscape, which will inevitably encompass compaction ofartificially placed soils. During compaction, soils are gen-erally placed in an unsaturated state for ease of placement.The increasing economic and environmental costs of gran-ular fill materials are causing the construction industry tolook for alternative construction materials, such as site-wonclayey soils (i.e. the material excavated from the site or

nearby location). A common application of fill materials istheir use as backfill in many engineering applications, forexample behind a retaining wall. The post-constructionbehaviour of clay fills is complex with respect to stressesand deformation when the fills become saturated overtime.

The behaviour of unsaturated soils is not covered byengineering standards with respect to geotechnical design.Guidelines do, however, exist for the performance of on-sitecompaction (Design Manual for Roads and Bridges, High-ways Agency, 1995; Specification for Highway Works,Highways Agency, 2004). Difficulties encountered whenunsaturated soils become saturated include loss of shearstrength and volumetric responses ranging from collapsecompression to swelling (Alonso et al., 1990; Charles, 1993;Lu & Likos, 2004; Sivakumar et al., 2010a). Volume changebehaviour is the biggest problem when working with unsatu-rated soils, where collapse compression can lead to excessive

Manuscript received 8 December 2009; revised manuscript accepted7 December 2010.Discussion of this paper closes on 1 September 2011; for furtherdetails see p. ii.� University of Ulster, Newtonabbey (formerly postgraduate studentat Queen’s University Belfast).† School of Planning, Architecture and Civil Engineering, Queen’sUniversity Belfast, Belfast.


settlements while swelling can produce ground heave(Sitharam et al., 1995; Skinner, 2001); both of these canlead to differential settlements (Charles et al., 1993; Skinneret al., 1999; Blanchfield & Anderson, 2000).

Since the early 1990s, research in the topic of unsaturatedsoils has focused on the development of constitutive models(Alonso et al., 1990; Toll, 1990; Wheeler, 1991; Delage &Graham, 1995; Wheeler & Sivakumar, 1995; Cui & Delage,1996; Tang & Graham, 2002; Blatz & Graham, 2003;Gallipoli et al., 2003; Wheeler et al., 2003; Tarantino, 2007;Sivakumar et al., 2010a, 2010b). The basic Barcelona modeluses the classical assumption of elastic behaviour prior toyielding and elasto-plastic behaviour after yielding. Suctionis also considered as one of the stress parameters that affectsyielding.

Saturation of heavily compacted clays produces a greatervolumetric swelling response than lightly compacted clays(Alonso et al., 1990; Wheeler & Sivakumar, 1995; Thom etal., 2007). Fig. 1 shows the results where the heavilycompacted specimens swelled significantly more than thelightly compacted specimens during wetting from initialsuction of 960 kPa (Sivakumar et al., 2010a). This suggeststhat in situ heavily compacted fill is more likely to produceswelling during wetting than lightly compacted fill. Animportant question remains, what effects would the wettinghave on the stress regime within heavily compacted fills andtherefore on the buried or rigid retaining structures?

Consider an element of soil behind the face of theretaining structure, back-filled with compacted clay as shownin Fig. 2. In temperate climates the water table will rise overtime and the fill becomes saturated. This wetting processcould produce volumetric responses of either swelling orcollapse. If swelling takes place under laterally confinedconditions the likely consequence will be the developmentof high lateral stresses acting on the rigid retaining structure.The increase in lateral stresses may cause stability issueswith the wall (Chen, 1987; Carder, 1988). In a limited study,Maswoswe (1985) examined the stress path of unsaturatedcompacted specimens of glacial till during wetting (under K0

conditions) in which the suction was reduced to zero (soak-ing tests). The stress path for the specimens wetted under K0

conditions ended on the saturated, normally consolidated K0

line.

The work presented in this paper examines how the stressregime in unsaturated clay fills changes with wetting underlaterally restrained conditions. Specimens of compacted kao-lin with different initial conditions were wetted to variousvalues of suction under laterally restrained conditions atconstant net overburden pressure. This allowed the conceptof K0 (the ratio between the net horizontal stress and the netvertical stress) given by the following equation to be exam-ined.

K0 ¼�h � ua

�v � ua

(1)

where �v, �h and ua are the total vertical, total horizontaland pore air pressures respectively.

Alonso et al. (1990) presented the Barcelona basic model(BBM), which is defined in terms of four stress statevariables: mean net stress p, deviator stress q, suction s andspecific volume v. The model is intended for use withslightly or moderately expansive soils. Isotropic loadingconditions are modelled using two yield curves, the loadingcollapse (LC) yield locus and the suction increase (SI) yieldlocus, which are shown in Fig. 3. When the state of the soilis inside the yield curve a reduction in suction (path AB)will lead to swelling until the LC yield locus is reached.Further reduction in the suction (path BC) will lead toplastic collapse. An increase in compactive effort will lead

( )ln

0

Cha

nge

in s

peci

fic v

olum

e,Δ

v

patm

s p� atm

Heavy compaction

Light compaction

0 0·2 0·4 0·6 0·8 1·0 1·2 1·4 1·6

0·20

0·15

0·10

0·05

300 kPa200 kPa100 kPa

Fig. 1. Change in specific volume plotted against ln (s + patm)/patm during wetting for kaolin soil, both lightly and heavilycompacted (after Sivakumar et al., 2010a)

σh

σv

Ground surface

Ground surface

Retaining wall

Fig. 2. Compacted fill placed behind a retaining structure

A

B

C

SI yield locus

LC yield locus

s

p

Wetting path

Fig. 3. Loading collapse and SI yield loci in the p–s plane forisotropic conditions

88 BOYD AND SIVAKUMAR


to an enlargement of the yield locus, including a change inshape (Maatouk et al., 1995; Wheeler & Sivakumar, 2000).The size of the yield locus is also affected by the compac-tion water content and the type of compaction (Maatouk etal., 1995; Wheeler & Sivakumar, 2000; Estabragh et al.,2004; Sivakumar et al., 2010a). The work presented in thispaper assesses the effects of compaction effort and compac-tion water content on the development of lateral stresses andtherefore on the earth pressure coefficient K0.

In saturated soil the magnitude of K0 is given by the ratiobetween the horizontal effective stress and the vertical effec-tive stress. The value of K0 is constant for normally con-solidated clays and increases with over-consolidation ratio(OCR). As part of the research presented here an attempt isalso made to assess the concept of K0 resulting from loadingand unloading. The following section describes the experi-mental programme undertaken to meet these tasks.

EXPERIMENTAL PROGRAMMEThe subject of this research is the response of compacted

fill, retained behind a rigid structure when subjected towetting. Obviously research of this calibre requires a sophis-ticated testing system and suitable testing programme. Thefollowing sections give details on the

(a) sampling technique(b) experimental set-up(c) control program(d ) testing programme.

SamplingThe material chosen for the present research was commer-

cially available speswhite kaolin clay supplied by WhitChemLtd. The original intention of the authors was to complementthe current research with work previously reported by theresearch team at Queen’s University Belfast (QUB) (Wheeler& Sivakumar, 1995; Sivakumar & Wheeler, 2000; Wheeler& Sivakumar, 2000; Sivakumar et al., 2006; Thom, 2007;Sivakumar et al., 2010a; 2010b). Initial characterisationperformed on kaolin purchased for the present researchdeviated considerably from the characteristics reported in theabove literature studies. This was caused by significantdifferences between the material used in the present researchand that used in the previous research, which was identifiedto be the percentage of the clay fraction, with a 40% clay-sized fraction in the material used in the present research asopposed to an 80% clay-sized fraction in the material usedpreviously. Table 1 lists the characteristics of the materialused in the current research.

The method chosen for preparing 50 mm diameter,100 mm high specimens was that proposed by Sivakumar etal. (2010a). Dry kaolin powder was mixed with water attargeted moisture contents in a domestic food mixer beforebeing passed through a 1.18 mm aperture sieve. Largeaggregates not passing through the sieve were broken upusing a pestle and mortar and sieved again. The materialwas stored in an air-tight container in a temperature-

controlled environment for 48 h. A 100 mm diameter rubbermembrane was placed around a 100 mm diameter pedestalof a standard triaxial cell and sealed to the base using twoO-rings. A membrane stretcher was placed around themembrane and the top of the membrane was folded at thetop of the stretcher. A 100 mm diameter dry porous disc wasplaced on the pedestal and the sieved material was filledslowly into the membrane. With the membrane full, a topcap of dry porous stone was placed on the moistened kaolin,the membrane stretcher was removed and the top cap wassealed with two O-rings. The height of the uncompressedsample was just over 225 mm and the diameter was about100 mm. The cell was assembled and pressurised to therequired value, based on the target initial specific volume.Any excess air pressure, developed as a result of theapplication of the external pressure, was allowed to drainfrom the top and bottom of the sample. There was noevidence of water drainage during this process. Althoughconsolidation usually took place quickly, the sample was leftin the compression system for 3 days as a standard proce-dure. At the end of this time the confining pressure wasreduced to zero. The size of the sample after compressionwas approximately 70 mm in diameter and 140 mm high. Athin-walled sample tube was used to extract a specimen of50 mm diameter and 100 mm high (Fig. 4). Specimens forthe present research were produced at two different initialwater contents: 23% and 25% (representing dry fill and wetfill) and at two different compression pressures: 375 kPa and750 kPa (representing unengineered fill and engineered fill).The relevant initial specific volumes of these specimens arelisted in Table 2.

Table 1. Characteristics of current research material

Percentage clay-sized fraction Liquid limit Plastic limit Plasticity index Activity Clay mineral identified byXRD�

40% 70% 34% 36% 90% Kaolinite, mica and quartz

� XRD: X-ray diffraction

Thin-walledsample tube

Sample

WykehamFarranceloading frame

Fig. 4. Set-up for specimen cutting

STRESS REGIME IN LATERALLY CONFINED UNSATURATED COMPACTED CLAY 89


Experimental set-upThe experimental system was designed and fabricated in

such a way that it can be used for model wetting, loadingand unloading of unsaturated compacted specimens, whenlaterally confined. A ‘twin cell’ recently developed at QUBwas used in the research (Sivakumar et al., 2006), althoughsome modifications were made in order to meet the require-ments of the present research. Fig. 5(a) shows the cross-sectional view of the modified twin-cell system. The systemwas instrumented with internal strain gauges, inclinometers(Jardine et al., 1984) to measure axial strain and a strainbelt to measure diametric strain (manufactured by ImperialCollege Consultants). In addition, the inner cell was as-sembled under water for accurate measurements of specimenvolume change (Sivakumar et al., 2006).

The testing programme aimed to replicate the in situbehaviour of compacted fill when laterally restrained, forexample behind a retaining wall. Under these conditionsthe overburden pressure will not change (assuming nofurther loading is applied post-construction). Therefore, inthe present research, throughout the wetting stage thecontrol program was used to maintain a constant netvertical pressure. The cell pressure was used to maintainthe zero lateral strain condition. In the event of anincrease in cell pressure (i.e. in order to contain lateralexpansion) a constant vertical pressure was maintained byaltering the lower chamber pressure, thus applying a nega-tive deviator stress. This required a special tension loadingarrangement on the top cap. The supporting base of theinner cell was threaded and it was screwed on to thethreaded pedestal of the Bishop and Wesley stress pathapparatus (Fig. 5(b)).

The tension loading mechanism was based on the designof a manhole key. A photograph of the key is shown in Fig.5(g). The key had one end threaded on to the load cell andthe other end was formed to a bucket shape. Fig. 5(f) showsthe special arrangement made on the top cap to applytension loading. This consisted of a metal bracket which hada small slot in the middle, slightly wider than the keythickness. Fig. 5(e) shows the arrangement during testingwhen the key was engaged in position with the top cap.After slotting in place, the key was turned through 908. Theother end of the load cell was attached to the top plate ofthe outer cell. The load cell end was threaded on to an endcap, which had three threaded bars protruding from it. Astainless steel plate was located on the other end of thethreaded rods as shown in Fig. 5(d). Nuts and washers wereplaced either side of the plate, which allowed manual adjust-ment to be made on the location of the plate if required.The load cell was fastened to the top plate of the outer cellusing a custom-made bolt (50 mm diameter head and120 mm long) threaded through the stainless steel platelocated on the threaded rods (Fig. 5(c)). The bolt could begently turned until its flat face interfaced with the top plateof the outer cell. This bolt was held in position by locatinganother cap which was fastened to the top plate of the outercell. When fastened fully, the underside of this cap flushed

with the top face of the bolt so that any movement of theload cell was restricted.

Control programThe control program Triax version 5.1 (Toll, 1999) was

utilised to control the required testing conditions automati-cally. Suction was controlled using the axis translation tech-nique (Hilf, 1956). Stepper motor-driven regulators wereused for controlling pore air pressure, pore water pressure,cell pressure and deviator stress. K0 control was automati-cally achieved by controlling confining pressure during wet-ting, loading and unloading.

Testing programmeTable 2 lists the tests conducted in the research. Addi-

tional tests were also conducted in order to establish the LCyield locus traced by the initial compression process. Themajority of the tests involved four stages, which aredescribed as follows.

(a) Initial application of relevant confining, pore water andpore air pressures: the initial state of the specimen,after extraction using a sampling tube is represented byO in Fig. 6(a). At this stage the initial mean net stressand deviator stress in the specimen are zero. Thesuction in the specimen refers to the values indepen-dently measured using a Wescor PST-55 thermocouplepsychrometer, which are listed in Table 2. The cell,pore water and pore air pressures were ramped at a rateof 25 kPa/min in order to achieve target suction valuesof 300, 200, 150, 100 kPa or zero at the drainageboundaries and a mean net stress of 50 kPa.. The stateof the specimen immediately after the application ofthese pressures is represented by point A in Fig. 6(a).During this initial application of pressures the conditionof K0 was not activated and the process resulted in asmall influx of water into the specimen (0.60 cm3).

(b) Wetting of the specimen to a pre-selected suction: inthis stage the externally applied suction at the drainageboundaries was allowed to equalise within the speci-men, which usually lasted about 7 days. During thisperiod the condition of K0 was activated with a lateralstrain tolerance of 0.005% (representing 0.0025 mm ona 50 mm diameter specimen). The state of the specimenafter equalisation is represented by points B, C, D, Eand F when the specimen has reached the target suctionvalues of 300, 200, 150, 100 kPa and zero respectively.However, mean net stress at this point is determined bythe response of the soil to meet the condition of zerolateral strain. The authors are aware that the control isbased on the strain performance of the specimen at thecentre, while the wetting begins at the base of thespecimen, where the pore water pressure is controlled.The original intention of the research was to wet thespecimen slowly by reducing the suction from its initial

Table 2. Initial characteristics of compressed fill for K0 testing programme

w: % rb: kg/m3 Relative tostandard Proctor,

Fill name v vw Sr: % s: kPa Applied suction: kPa

rb=rb(St Proc) Zero 100 150 200 300

23 1557 0.95 Dry unengineered 2.088 1.600 55.2 811 — ˇ — — —1701 1.05 Dry engineered 1.925 1.607 65.8 1035 ˇ ˇ ˇ ˇ ˇ

25 1640 0.96 Wet unengineered 2.022 1.666 65.2 704 — ˇ — — —



(a)

(b) (g)

(f)

(e)

(d)

(c)

Loadapplicationbracket

Curved hole forload cell key

Slotted hole

Support bolt

Top cap

Load cellend cap

Threaded bars

Stainlesssteel plate

Outer celltop plate

Clearance

Threadedbars

Stainlesssteel plate

d

c

c

de

b

38·20 5·90

39·0

10·0 37·0 6·5

Supporting baseof inner cell

Pedestal of existingstress path cell

Base plate ofinner cell

Not to scale

Fig. 5. Testing equipment: (a) modified twin-cell system; (b) threaded pedestal of outer cell; (c) arrangement forsupporting load cell end; (d) load cell end arrangement; (e) key and bracket engaged during loading; (f) top cap loadingbracket; and (g) load cell key



value to zero over a long period. However, the plan wasabandoned owing to problems encountered in thecontrol program (i.e. with the older version of Triax,which operated on DOS mode).

(c) Loading of the specimen under constant suction: thedeviator stress was increased at pre-selected rates. TheK0 loading begins from the point where the stress pathterminated during K0 wetting. A typical case for asuction value of 200 kPa is schematically illustrated inFig. 6(b) (i.e. point C). Three different rates wereadopted: 0.6 kPa/h for the first 3 days; 0.4 kPa/h until asign of yielding was noted and 0.2 kPa/h for theremainder of the loading stage. The reason for adoptingthe lower rates at the later stages of the loading was toavoid a rapid increase in confining pressure in order tomeet K0 conditions, particularly when the specimen hasyielded. During the entire loading the suction wasmaintained at a constant value.

(d ) Unloading under constant suction: during this stage thedeviator stress was reduced to zero, under constantsuction, at a rate of 0.6 kPa/h (path C0G in Fig. 6(b))while the horizontal strain was maintained at zero. Inone of the tests the specimen, subsequent to K0

unloading, was taken through isotropic loading until a

clear sign of yielding under constant suction wasachieved (path GH in Fig. 6(b)).

RESULTS AND DISCUSSIONOwing to the complex nature of the experimental pro-

gramme the results are discussed under specific headings tosimplify the presentation:

(a) initial characterisation of the specimen with referenceto the initial position of the yield locus

(b) conceptual modelling(c) K0 in unsaturated soils during wetting and the effects of

initial conditions on its magnitude(d ) effects of bi-modal pore size distribution on the

deformation under K0 conditions(e) K0 during loading and unloading and the effects of

initial conditions on its magnitude( f ) K0 loading-induced anisotropy.

Initial characterisation of the specimen with reference to theinitial position of the yield locus

As part of characterisation the specimens were testedunder isotropic stress conditions (i.e. without K0 control) andsubsequently sheared under drained conditions. Duringequalisation water flowed into the specimen. The intake ofwater was 37.3, 20.2, 12.5 and 9.9 cm3 for suctions of zero,100, 200 and 300 kPa respectively. Fig. 7 shows the changesin specific volume of dry engineered specimens (23H)plotted against suction. A significant amount of swelling wasobserved in the case of the dry-engineered specimen wettedto zero suction, although as expected the swelling waslimited under other values of suction.

Subsequent to the initial equalisation, the specimens weretaken through isotropic compression at constant suction. Fig.8(a) shows the compression lines plotted against naturallogarithm of mean net stress for dry engineered specimens(23H). A simple graphical construction was used to deter-mine the approximate yield stress. Although there are othermethods available to determine yield stresses (Crooks &Graham, 1976) the Casagrande (1936) technique was chosenfor the present purpose. The relevant yield stresses measuredat suction values of zero, 100, 200 and 300 kPa were plotted

(a)

(b)

s

p

p

q

LC yield locus

O A

B

CD

EF

C

C�

C�

G H

Yielding

Fig. 6. Stress paths during experiments: (a) K0 wetting and(b) K0 loading, K0 unloading and isotropic loading

23H

23L100

25L100

0 0·5 1·0 1·5ln [( )/ ]s p p� atm atm

Cha

nge

in s

peci

fic v

olum

e,Δ

v

0·12

0·10

0·08

0·06

0·04

0·02

0

�0·02

Fig. 7. Change in specific volume plotted against suction duringwetting (dry engineered specimens – 23H)



to form the LC yield locus as shown in Fig. 8(b). The formof the LC yield locus is a straight line in s–p space and theshape of it is in close agreement with the previouslyreported data on kaolin by Sivakumar et al. (2010a),although distinctly different from the proposals of Alonso etal. (1990), Wheeler & Sivakumar (1995) and Sivakumar &Wheeler (2000). The form of the yield locus that wasreported by Alonso et al. (1990) and Wheeler & Sivakumar(1995) was a curve, which was obtained on specimens thatwere one-dimensionally compressed before testing. The one-dimensional loading history, inherited by the specimen dur-ing its initial formation, may have led to a degree ofanisotropy. Sivakumar et al. (2010a) reported LC yield locus,which was a straight line in s–p space and it was obtainedusing specimens prepared by isotropic compression beforetesting. This should have led to specimens inheriting isotro-pic stress–strain properties. Further details on this aspectcan be found in Sivakumar et al. (2010a).

The estimated yield stresses at a suction value of 100 kPaon the dry unengineered specimen (23L) and wet unengi-neered specimen (25L), denoted by square and triangulardata points respectively, are also included in Fig. 8(b). Thepositions of the LC yield loci shown using broken lines inFig. 8(b) for these two specimens are only an approximation.The wetting under K0 conditions originated from point A(Fig. 6(a)). If the wetting path remains inside the yield locusthe specimen would swell axially as well as laterally. Thelateral expansion is restricted by an increase in confiningpressure and therefore the position of the LC yield locusshown in Fig. 8(b) is the prelude to most of the discussionpresented in this paper. In order to simplify the presentationof the data, a conceptual view of the anticipated behaviouris presented first and subsequently validated with the ob-served response of the soil.

Conceptual modellingFigure 9(a) shows a three-dimensional schematic view of

the yield surface in q–p–s space. The intersection of theyield surface on a constant q plane is shown in Fig. 9(b).The cross-section of the yield surface parallel to the constantsuction plane is an ellipse and its major principal axiscoincides with the mean net stress plane as shown in Fig.9(c). The cross-section of the yield surface parallel to theconstant p plane is also shown in Fig. 9(d). The initialposition of the specimen is represented by point O, wherethe mean net stress is zero, the deviator stress is zero andthe initial suction in the specimen is listed in Table 2. Uponthe application of initial pressures, the state of the specimenmoves to point A, where q ¼ 0 and p ¼ 50 kPa, the zerolateral strain condition is invoked at this point. The initialposition of each specimen is well within the LC yield locus,thus any reduction in suction will result in radial and axialexpansion, provided the wetting paths do not cross the yieldsurface. Upon lateral swelling the control program is acti-vated to increase the lateral pressure to restrain the lateralexpansion. ABCDEF represents the wetting path duringwhich the lateral strain was maintained at zero. Points B, C,D, E and F represent target suction values of 300, 200, 150,100 kPa and zero respectively. The vertical stress is main-tained at 50 kPa during the wetting and therefore the stresspath will be contained on a plane with a slope of 2/3, asshown in Fig. 9(a). The relevant projections of the stresspaths on s–p , q–p and q– s are also shown in Figs 9(b)–9(d).

Continued swelling, caused by the wetting, instigates afurther increase in lateral pressure. Such an increase in thelateral pressure, together with the continued reduction indeviator stress (an increase in negative deviator stress, Fig.9(d)) also triggers inter-aggregate slippage, creating a ten-dency for collapse settlement. This in theory implies that thestress path approaches the yield surface, initially inherited bythe specimen during its formation, as shown in Fig. 9(a).This particular situation is represented by point X. The pointat which the wetting path approaches the yield surfacedepends on the initial conditions of the specimen. At thepoint when the specimen approaches close to the yieldsurface, the collapse settlement will become imminent andthe control system activates to decrease the lateral stress inorder to maintain the zero lateral stress condition. Thereforethe direction of the stress path will change and the stressstate of the specimen moves slightly back inside the yieldsurface. This process will continue until the suction in thespecimen has fully equalised at the pre-selected suctionvalue. Since the suction applied to the specimen was reducedin one single step from its initial value to the relevant targetsuction, the lateral stress will increase and decrease often

350

300

250

200

150

100

50

0

2·02

2·00

1·98

1·96

1·94

1·92

1·90

1·88

1·86

1·84

Spe

cific

vol

ume,

v

10 100 1000

23H

23L

25L

0 50 100 150 200

Suc

tion,

: kP

as

Log mean net stress log ( )(a)

p u� a

Mean net stress : kPa(b)

p u� a

s zero�

s 100 kPa�

s 200 kPa�

s 300 kPa�

Proposed LCyield locus

Fig. 8. Sample response to isotropic loading of dry engineeredspecimens: (a) specific volume plotted against logarithm ofmean net stress and (b) loading collapse yield loci (LCYL) fordry engineered specimens (23H), including proposed LCYL fordry unengineered (23L) and wet unengineered specimens (25L)



until equilibrium is achieved. This can result in fluctuationsof deviator stress and mean net stress. The fluctuation couldbe avoided if the suction was reduced slowly from its initialvalue. The stress path traversing close to the yield surface istherefore represented by a wavy curve to reflect the nature inwhich the control program attempts to maintain zero lateralstrain conditions. The data presented in the next sectionquantify this conceptual view.

K0 in unsaturated soils during wetting and the effects ofinitial conditions on its magnitude

Five tests were conducted on dry engineered specimenshaving an initial specific volume of 1.925. The specimenswere brought to suction values of zero, 100, 150, 200 and300 kPa from the initial suction of 1035 kPa generated at thetime of specimen preparation. The results obtained fromthese tests are analysed in various forms, culminating withthe assessment of K0 in unsaturated soils subjected towetting. Table 3 lists the phase values after the K0 wettingstage was completed.

Figure 10 shows the change in specific water volumeplotted against time. A specific point on each curve is

marked with a circular point. The relevance of these pointswill be discussed later in this paper. The relevant changes inspecific water volume caused by wetting the specimens tosuctions of zero, 100, 150, 200 and 300 kPa under laterallyconstrained conditions were 0.359, 0.213, 0.175, 0.149 and0.080, respectively. The equalisation phase was terminatedwhen the water flow into the specimen was less than 0.1%/day with respect to the dry mass of the specimen. The influxof water resulted in an increase (or decrease) in the speci-men volume and therefore a change in the lateral strain. Atolerance level of �0.0025 mm (�0.005% in terms of lateralstrain) was set as a trigger level and Fig. 11 shows how wellthis condition was achieved.

Figure 12 shows the lateral and vertical net stressesplotted against time for the tests equalised under suctionvalues of zero, 100 and 300 kPa. The lateral net stressincreased rapidly at the beginning of the equalisation processimplying that the specimen attempted to expand laterally.There were some fluctuations in the net lateral stress as theequalisation stage progressed. The reason for this fluctuationis the fact that the stress path upon wetting under K0

conditions was operating close to the yield surface (Fig. 9).This was also instigated by a steep decrease in suction from

p

p

p

(a) (b)

Yield locus

(c) (d)

Yield surface

C B

q

sB

E

F

DOA

X

C

s

LC yield locus

O A

B

CD

E

F

F

OA

DE

X

q

s

X

O A

F

X

q

Yield locus

Fig. 9. Conceptual modelling in: (a) q–p–s space; (b) s–p plane; (c) q–p plane; and (d) q–s plane



its initial value of 1035 kPa to a much lower value asdiscussed in the conceptual view of the anticipated behav-iour. As shown in Fig. 12 the vertical net stress remainedconstant at 50 kPa in every test during the course of wetting.

It was obviously a concern that the K0 condition was

activated based on the lateral strain measured at the mid-height of the specimen, whereas the wetting front movesupwards from the base. It could be argued that the specimenat the bottom may have swelled laterally before the K0

control became activated, based on the specimen response atmid-height. The height of the specimen used in the presentstudy was 100 mm. The radial strain gauge was located atmid-height of the specimen. The pads attached on the armsof the axial strain gauges were located 23 mm above thebase of the specimen and 37 mm below the top of thespecimen, as shown in Fig. 13. During the equalisationstage, the wetting front moved from the bottom of thespecimen (where the high air entry filter stone was located)to the top where the low air entry filter was located. There-fore, the lower arms of the inclinometers would be the firstto respond to the effects of wetting and soon after that thelateral strain gauge attached at the mid-height. Fig. 14 showssome interesting behaviour where the raw readings from theaxial strain gauge and the lateral pressures are plottedagainst time (during the first 100 min of K0 wetting). Itappears that the time lag between the start of the K0 wettingand the lateral strain response (represented by an increase inlateral pressure) is around 5 min, thus confirming that thewetting front is not merely a line separating two zones withdifferent suctions, but rather it is a moving zone wherechange occurs gradually.

Figure 15 shows the K0 plotted against time for specimensequalised at target suction values of zero, 100, 150, 200 and300 kPa respectively. By default, the response of K0 wassimilar to the behaviour observed in the case of horizontalnet stress as shown in Fig. 12. At the beginning of thewetting, the specimens were subjected to an imposed K0

condition of 1 and this K0 value increased to maximumvalues of 2.3, 3.0, 2.1, 2.3 and 2.0 for suction values of 300,200, 150, 100 kPa and zero respectively. The values of K0

when the suction in the specimen stabilised at the pre-selected values were: 1.5, 1.6, 1.8, 2.0 and 2.0 respectivelyfor suction values of zero, 100, 150, 200 and 300 kPa. In thecase of low suction values, K0 reduced significantly afterreaching a peak. The reason for this response lies in the factthat the wetting path was traversing inside, but close to theyield surface initially inherited by the specimen (see Fig. 9).

At the beginning of wetting the horizontal net stress was50 kPa and it increased to a maximum value of 111, 141,104, 116 and 116 kPa in the case of specimens wetted totarget suction values of zero, 100, 150, 200 and 300 kParespectively. This particular state can be represented by pointX in Fig. 16(a) where suction is plotted against mean netstress. The lateral stresses when the equalisation process wasterminated (i.e. when the externally applied suction equal-ised within the specimen) were 74, 79, 90, 99 and 101 kParespectively and the stress states referring to these values arerepresented by points F, E, D, C and B in Fig. 16 respec-tively for suctions of zero, 100, 150, 200 and 300 kPa. Notehere that the actual value of suction when the lateral stressreached a peak (point X in Figs 16(a) and 16(c)) was not

Table 3. Phase values after K0 wetting

Test v vw Sr: % p: kPa q: kPa

23H0 1.987 1.984 96.9 66 �2423H100 1.973 1.844 86.8 69 �2923H150 1.988 1.782 79.2 77 �4023H200 1.997 1.751 75.3 83 �4923H300 1.947 1.686 72.4 84 �5023L100 2.125 1.839 74.6 54 �825L100 2.034 1.802 77.5 69 �34

0·40

0·35

0·30

0·25

0·20

0·15

0·10

0·05

0

Cha

nge

in s

peci

fic w

ate

r vo

lum

e,Δ

v w

0 50 100 150 200Time: h

s zero�

s 100 kPa�

s 150 kPa�

s 200 kPa�

s 300 kPa�

Fig. 10. Change in specific water volume during K0 wetting ofdry engineered specimens

0·05

0·04

0·03

0·02

0·01

�0·01

�0·02

�0·03

�0·04

�0·05

0

Hor

izon

tal s

trai

n,: %

ε h

0 50 100 150 200Time: h

s 100 kPa�

s 150 kPa�

s 200 kPa�

Fig. 11. Horizontal strain plotted against time for dry engi-neered specimens equalised under suction values of 100, 150and 200 kPa



known. The changes in the specific water volume leading upto these maximum lateral stress state conditions were 0.090,0.055, 0.110, 0.085 and 0.04 respectively for tests carriedout to reach target suction values of zero, 100, 150, 200 and

0 50 100 150

120

100

80

60

40

20

0

Net

str

ess,

: kP

aσ

�u a

Net

str

ess,

: kP

aσ

�u a

Net

str

ess,

: kP

aσ

�u a

160

140

120

100

80

60

40

20

00

0

50

50

100

100

150

150

200

200

160

140

120

100

80

60

40

20

0

Net horizontal stress

Net vertical stress


Net vertical stress

Time: h(a)

Time: h(b)

Time: h(c)

Fig. 12. Horizontal and vertical net stress plotted against timefor specimens equalised under suction value of (a) zero, (b) 100and (c) 300 kPa

Wetting zone

50 m

m

Radial strainbelt

Inclinometer pads

37 m

m40

mm

23 m

m


Net vertical stress

Fig. 13. Strain gauge positioning

0

�0·05

�0·10

�0·15

�0·20

�0·250 50 100

Time: min

Axial strain

Lateral stress

25

50

75

100

125

0

Net

late

ral s

tres

s,: k

Pa

σ 3

Time lag

Axi

al s

trai

n: %

Fig. 14. Initial strain gauge and net lateral pressure responseduring wetting

3·5

3·0

2·5

2·0

1·5

1·0

0·5

0

Coe

ffic

ient

of

eart

h pr

essu

re a

t res

t,K

0

0 50 100 150 200Time: h

s zero�

s 300 kPa�

s � 150 kPa

s 200 kPa�s 100 kPa�

Fig. 15. K0 plotted against time for specimens of dry engineeredspecimens during equalisation



300 kPa (denoted by circular data points in Fig. 10) respec-tively with an average of 0.076. This change in averagespecific water volume corresponds to an approximate suctionvalue of 350 kPa, estimated based on the end condition of

the samples, shown in Fig. 10 when the suctions have fullyequalised within the specimens.

The initial position of the LC yield locus referring to theinitial compaction is also included in Fig. 16(a). Point Xrefers to the average maximum lateral stress condition foreach test. It appeared that the positions of the stress states atthe end of the K0 wetting were well inside the yield locus,thus implying that any tendency for possible collapse settle-ment upon wetting was minimal. It should be noted that theLC yield locus shown in Fig. 16(a) was plotted on the q ¼ 0plane. A possible position of this LC yield curve for theconditions in positive and negative q space is also illustratedin Fig. 16(a) using a broken line. Under these conditions itcan be argued that the maximum lateral stress condition,together with a deviator stress component, could result in thestress state of the specimen being much closer to the originalyield surface than it appeared on the two-dimensional plot.

The net overburden pressure was kept constant at 50 kPaduring wetting. This implied that the slope of the stress pathdq/dp in the q–p plane was constant and equal to 3/2. Forinformation, the stress paths that the specimens followedduring wetting are shown in Figs 16(b) and 16(c) in q–pand q–s planes. The extreme point on the stress path refersto the maximum lateral stress conditions and the end condi-tions are marked with the points B, C, D, E and F forsuction values of 300, 200, 150, 100 kPa and zero. Path AXwas inside the yield surface and the paths X to B, C, D, Eand F traversed close to the yield surface.

Figures 17(a) and 17(b) show the coefficient of earthpressure ‘at rest’ (K0) for unsaturated soil plotted againstsuction for all five tests. Fig. 17(a) refers to the maximumK0 condition and Fig. 17(b) refers to K0 values achieved atthe end of the equalisation stage. Fig. 17(a) is not anaccurate representation of the soil behaviour, as the actualsuction values corresponding to the K0 maximum valueswere unknown. However, previous discussion on the lateralstresses stated the fact that the amounts of water that flowedinto the specimen when the lateral stress was at a maximumwere generally similar and suggested therefore that the maxi-mum lateral stress conditions occurred at a similar value ofsuction, approximately 350 kPa in all five tests, regardless ofthe targeted suction. Fig. 17(b) shows a more realistic viewof the K0 variation with suction. The values of K0 reducedslowly as the suction reduced to zero.

A limited study was carried out to investigate the effectsof the initial specimen conditions on the lateral stressesrequired to maintain zero lateral stress conditions. A singletest was performed on an unengineered wet specimen and anunengineered dry specimen equalised under a suction valueof 100 kPa. Fig. 18 shows the lateral stress plotted againsttime during the course of wetting for all three specimens, inwhich the suction was reduced from its initial value to100 kPa. The maximum lateral stress observed in the case ofthe engineered dry specimen was 140 kPa, and those for theunengineered dry specimen and wet unengineered specimenwere 93 and 85 kPa respectively. The reduced maximumlateral stress in the case of the unengineered specimensconfirmed the fact that the yield surfaces inherited by theunengineered specimens were well inside the yield surfacefor the engineered specimen (Fig. 8). Fig. 19 shows theposition of the LC yield locus and the stress path duringwetting of dry engineered, wet unengineered and dry unengi-neered specimens. The position of the LC yield locus andthe stress path during K0 wetting has been well establishedfor a dry engineered specimen (as discussed earlier),although this was not the case for dry unengineered and wetunengineered specimens. In these two cases the researchfocused on a suction value of 100 kPa and the LC yield lociare assumed to pass through the yield stresses identified at

1200

1000

800

600

400

200

0

Suc

tion,

: kP

as

0 100 200 300

100

80

60

40

20

0

�20

�40

�60

�80

�100

Dev

iato

r st

ress

, :

kPa

q

0 50 100 150

100

80

60

40

20

0

�20

�40

�60

�80

�100

Dev

iato

r st

ress

, :

kPa

q

0 500 1000 1500

LC yield locus for0q �

A

O

XB

CD

EF

LC yield locus inve or – ve space� q

3

2

O A

F

X

ED

CB

Mean net stress, : kPa(a)

p u� a

Mean net stress, : kPa(b)

p u� a

Assumed yield surface

Wetting path

F

OA

ED

C B

X

Suction, : kPa(c)

s

Fig. 16. Stress paths during K0 equalisation in: (a) s–p space;(b) q–p space; and (c) q–s space



Max

imum

coe

ffic

ient

of

eart

h pr

essu

re a

t res

t,K

0

3·5

3·0

2·5

2·0

1·5

1·0

0·5

00 100 200 300 400

Res

idua

l coe

ffic

ient

of

eart

h pr

essu

re a

t res

t,K

0

2·5

2·0

1·5

1·0

0·5

00 100 200 300 400

Suction, : kPa(b)

s

Suction, : kPa(a)

s

Average value

Fig. 17. Coefficient of earth pressure at rest plotted againstsuction: (a) maximum values and (b) residual values

160

140

120

100

80

60

40

20

0

Net

str

ess,

: kP

aσ

�u a

0 100 200 300Time: h

Dry engineered specimen

Wet unengineered specimen

Dry unengineered specimen

Net vertical stress

Fig. 18. Horizontal and vertical net stress plotted against timefor a dry engineered specimen, a dry unengineered specimenand a wet unengineered specimen

1200

1000

800

600

400

200

0

Suc

tion,

: kP

as

Suc

tion,

: kP

as

Suc

tion,

: kP

as

0

0

0

100

100

100

200

200

200

300

300

300

1000

1000

900

900

800

800

700

700

600

600

500

500

400

400

300

300

200

200

100

100

0

0

Assumedwetting path

Assumed LCYL

A

O

E

Experimentallydetermined yieldstress

Assumed LCYL

Experimentallydetermined yield stress

A

O

E

Wetting path

Experimentallydetermined LCYL

A

O

XB

CD

E

F

Mean net stress : kPa(a)

p u� a

Mean net stress : kPa(b)

p u� a

Mean net stress : kPa(c)

p u� a

Fig. 19. Yielding behaviour of compacted samples in the s–pplane for: (a) dry engineered specimens; (b) dry unengineeredspecimen; and (c) wet unengineered specimen



this suction value. While substantial evidence was gatheredto confirm the position of the K0 stress path during wettingto be slightly inside the LC yield surface in the p–s–qspace, it can be argued that the K0 stress path will nevercross the LC yield surface inherited by the specimen duringits initial formation, regardless of the initial conditions (i.e.compaction water content and compaction effort) since anycontraction initiated by approaching the yield surface willtrigger the zero lateral strain conditions, thus reducing thelateral pressure, and the stress path will be reversed backslightly inside the yield surface.

Figures 20(a) and 20(b) show the K0 values at the timeof peak lateral stress and equalised lateral stress plottedagainst suction obtained on the dry engineered and wetunengineered specimens. Also included on the figures arethe relevant values obtained from the dry engineered speci-men. Peak and ultimate K0 values for both dry and wetunengineered specimens were below (or very close to) thevalues obtained for the dry engineered specimen, confirmingthe fact that the yield surfaces for dry and wet unengi-neered specimens were inside that of the dry engineeredspecimen.

Effects of bi-modal pore size distribution on the deformationunder K0 conditions

Unsaturated soils are constituted of an aggregated struc-ture, with a clear bi-modal pore size distribution (Fig. 21).Ahmed et al. (1974), Gens & Alonso (1992), Delage et al.(1996), Qi et al. (1996), Sivakumar & Wheeler (2000),Lloret et al. (2003) and Cuisinier & Laloui (2004) havedemonstrated the effects of an aggregated structure on thehydromechanical behaviour of unsaturated soils. Thom et al.(2007) and Sivakumar et al. (2010a) have reported experi-mental evidence, using MIP analysis to support the view thataggregates expand into the larger macro-voids during wet-ting. Figs 22(a) and 22(b) show the evolution of bi-modaldistribution before and after wetting of unengineered andengineered specimens, prepared at identical water content(Sivakumar et al., 2010a). The initial macro pore sizedistribution of a lightly compressed specimen was signifi-cantly higher than that of a heavily compressed specimen.However, the wetting resulted in very similar macro poresize distributions. If the aggregates are idealised as sphericalfor simplicity, upon wetting they become distorted as theyexpand into the macro voids where the restriction for expan-sion is less (Fig. 23). This implies that the effective size ofthe pore space will reduce and the evidence for this isexplicitly shown in Fig. 22, particularly in a heavily com-pressed sample. The effect of aggregated structure duringwetting under laterally confined conditions (i.e. under aniso-tropic loading conditions) is remarkable.

Figure 24 shows the axial and lateral strain plotted againsttime for suction values of zero, 100, 150, 200 and 300 kPa.A condition of zero lateral strain was imposed; hence nosignificant change in the lateral strain occurred. However,the vertical strains of the specimens increased significantlydepending on the targeted suction values. The axial strain inthe case of 300 kPa of suction was 0.64%, and this increasedto 3.36% when the suction dropped to zero. However, thereal challenge was to justify the observed response of thespecimen in relation to the axial strain. Fig. 12 clearly showsthat the lateral stresses reached a peak value and then beganto reduce as the wetting progressed. The reductions in thelateral stresses after reaching the peak were significant whenthe targeted suctions were 200 kPa or lower. The net vertical

Dry engineered specimens



Average value




3·5

3·0

2·5

2·5

2·0

2·0

1·5

1·5

1·0

1·0

0·5

0·5

0

0

Max

imum

coe

ffic

ient

of

eart

h pr

essu

re a

t res

t,K

0M

axim

um c

oeff

icie

nt o

fea

rth

pres

sure

at r

est,

K0

0

0

100

100

200

200

300

300

400

400

Suction, : kPa(a)

s

Suction, : kPa(b)

s

Fig. 20. Coefficient of earth pressure at rest plotted againstsuction: (a) maximum value and (b) equalisation value

2 mμ

Fig. 21. The bi-modal pore size distribution of unsaturatedcompacted clay shown by scanning electron microscopy (afterSivakumar et al., 2010a)



stress was kept constant at 50 kPa. Under these conditions,the specimens continued to swell axially under constantvertical stress. The reason for this lies in the structure ofunsaturated soils.

The suction in the aggregates is very high and this wasreduced to a lower value during the wetting process. Theaggregates absorbed water and tried to expand both axiallyand laterally. The expansion of the aggregate can result intwo different phenomena:

(a) the inter-aggregate contacts are pushed apart causingoverall increase in volume

(b) expansion of the aggregates into the larger voidsbetween the aggregates, causing reduced inter-aggregatevoids but marginal changes in the overall volume.

Consider, for example, the case of an engineered speci-men, where the tendency for inter-aggregate slippage ismarginal (Sivakumar et al., 2010a) and the inter-aggregatepore space is limited, therefore the most likely consequenceof wetting would be an increase in overall volume of thespecimen. The question remains, what will be the level ofswelling in the vertical and lateral directions when the speci-men is subjected to anisotropic loading?

At the beginning of the K0 wetting, the loading conditionswere isotropic, whereby the specimens were subjected to50 kPa of net vertical and lateral stress. As the equalisationprogressed the lateral stresses increased substantially in orderto contain the lateral expansion. Therefore, the aggregates,which were expanding or trying to expand in three direc-tions, would now prefer to expand in the vertical direction,

0

0

0·02

0·04

0·06

0·08

0·10

0·1 1 10 100

Pore diameter: mm(a)

Incr

emen

tal p

ore

volu

me:

cm

/g3Initial condition

After saturation

0·02

0·04

0·06

0·08

0·10

0·1 1 10 100

Incr

emen

tal p

ore

volu

me:

cm

/g3

Pore diameter: mm(b)

Initial condition

After saturation

2 mm

2 mm

Fig. 22. Pore size distributions of specimens before and afterwetting (Sivakumar et al., 2010a): (a) lightly compressedsample; (b) heavily compressed sample

Aggregates after swellingAggregates after sample preparation

Wetting

Fig. 23. Aggregated structure subjected to wetting

0·5

0

�0·5

�1·0

�1·5

�2·0

�2·5

�3·0

�3·5

�4·0

Ver

tical

str

ain:

%

0 50 100 150 200Time: h

Lateral strain

s zero�

s 100 kPa�

s 200 kPa�

s 300 kPa�

s 150 kPa�

Fig. 24. Axial and lateral strain plotted against time forspecimens equalised undersuction values of zero, 100, 150, 200and 300 kPa



as opposed to the lateral direction, since the resistance tothe expansion in the vertical direction is less than in thelateral direction; however this does not exclude the aggre-gates expanding into macro voids. Further investigation isnecessary in order to ascertain the validity of this statement.

K0 during loading and unloading and the effects of initialconditions on its magnitude

As part of the research, the specimens were subjected toincreasing deviator stress starting from the stress statecorresponding to the end of K0 wetting (marked by points B,C, D, E and F in Fig. 16(b)). Table 4 lists the phase valuesafter K0 loading and unloading.

Figure 25 shows axial strain plotted against deviator stress(in logarithmic scale) for tests conducted at suction valuesof zero, 200 and 300 kPa. Estimated yield pressures in termsof mean net stress and deviator stress were (80, 0), (158,38), (172, 51), (224, 42) and (266, 93) for suction values ofzero, 100, 150 200 and 300 kPa respectively. Fig. 26 showsthe stress path followed by the specimen equalised at200 kPa during K0 loading. The solid line shows the experi-mental results and the dash-dotted line shows the averagespecimen response.

The previous discussion suggested that the K0 wettingpath traversed near the boundary of the yield surface. It wasalso suggested that the stabilised stress state of the specimenat the end of wetting under K0 conditions was close to theyield surface as indicated by points B, C, D, E and F in Figs16(a), 16(b) and 16(d). The subsequent increase in deviatorstress therefore takes the specimen inside the yield surfaceinitially and on the yield surface upon continued loading(indicated by average paths BB9B99, CC9C99, DD9D99, EE9E99and FF9F99 in Fig. 27 for suction values of 300, 200, 150,100 kPa and zero respectively). The pattern observed in therelationship between deviator stress and mean net stressduring the K0 loading was similar to that of saturated soils.The gradient of the K0 path prior to yielding was high and itgradually reduced to a constant value as the loading pro-gressed. The most interesting observation is that the slope ofthe K0 line after yielding was reasonably constant andunaffected by the suction value. The following equation wasused to calculate the K0 value:

K0 ¼3�

3þ 2(2)

The K0 value after yielding is plotted against suction inFig. 28. The K0 values appear to be unaffected by thesuction value, but more importantly, they are very similar tothe K0 values reported for reconstituted saturated kaolin.Equation (3) relates the gradient of the critical state lineM(s) to the effective angle of internal friction �’ for a soil.The relationship is unaffected by the apparent cohesiongiven by the suction value. The gradient of the critical state

line M(s) is 0.8 (critical state line is not shown here due tothe space limitation). Therefore, the effective angle of inter-nal friction for the soil is 228. Assuming K0 ¼ 1 � sin(�’)(Jaky, 1948), this yields a K0 value of 0.65 under saturatedconditions.

Table 4. Phase values after K0 loading and K0 unloading

Test After K0 loading After K0 unloading

p: kPa q: kPa s: kPa v vw Sr: % p: kPa q: kPa v vw Sr: %

23H0 237 74 0 1.826 1.895 100.0 206 0 1.881 1.893 100.023H100 352 130 100 1.840 1.803 92.6 — — — — —23H150 326 108 150 1.889 1.755 84.9 — — — — —23H200 314 105 200 1.907 1.730 80.3 — — — — —23H300 465 159 300 1.855 1.653 76.3 276 0 1.855 1.651 76.123L100 161 37 100 1.963 1.820 85.2 115 0 1.963 1.824 85.525L100 299 53 100 1.854 1.777 90.9 — — — — —

6

5

4

3

2

1

0

�1

Axi

al s

trai

n,: %

ε a

�1·0 �0·5 0 0·5 1·0 1·5

ln [( )/ ]q p p� atm atm

s 100 kPa�

s 200 kPa�

s 150 kPa�

s 300 kPa�

s 0�

Fig. 25. Axial strain plotted against logarithm of deviator stressfor suction values of zero, 100, 150, 200 and 300 kPa

120

100

80

60

40

20

0

�20

�40

�60

Dev

iato

r st

ress

, :

kPa

q

0 100 200 300 400

Mean net stress, : kPap u� a

K0 line

Yield point

Fig. 26. Stress path during K0 loading of specimen equalised to200 kPa of suction



M sð Þ ¼6 sin�9

3� sin�9(3)

A limited number of tests were performed in order toexamine the effects of initial conditions of the specimen onthe position of the K0 path. Fig. 29 shows the average stresspaths obtained for dry engineered, dry unengineered and wetunengineered specimens. Each specimen was subjected to anincreasing deviator stress from the end point of the K0

wetting stage. The open circular data points on the stresspaths indicate the estimated yield stress from the stress–strain curves. The yield stresses identified in terms of meannet stress, deviator stress were (158, 38), (94, 26) and (150,40) for dry engineered, dry unengineered and wet unengi-neered specimens respectively. As expected the yield stressesin the case of the dry unengineered and wet unengineeredspecimens were smaller than that of the dry engineeredspecimen. Fig. 30 shows the K0 value plotted against suc-tion. The slope of the K0 line was affected by the initial

conditions and therefore the K0 value was also affected;however, further data are needed to confirm this remark.

The specimens were unloaded under K0 conditions. Threeunloading tests were performed, with suction values of zeroand 300 kPa for specimens of dry engineered specimens andwith a suction value of 100 kPa on a specimen of dryunengineered fill. The unloading paths are shown in Fig. 31.The specimens were stiff and overconsolidated, therefore thelateral strain was rarely affected by the reduction in verticalstress. The lateral pressure reduced slightly only at the endof the unloading path. For presentation purposes an approx-imate path was sketched connecting the start of the unload-ing and the end state of the soil based on the average lateralstresses. The behaviour observed was comparable with thebehaviour traditionally observed in saturated soils.

K0 loading-induced anisotropySoil structure inherited either during deposition, or post-

deposition influences stress–strain behaviour and conse-

200

150

100

50

0

�50

�100

Dev

iato

r st

ress

, :

kPa

q

0 100 200 300 400 500Mean net stress, : kPap u� a

s zero�

s 100 kPa�

s 150 kPa�

s 200 kPa�

s 300 kPa�

B

B�

B�

CD

EF

F'

F�

E�

E�

D�

D�

C�

C�

Fig. 27. Stress paths during K0 loading of specimens equalisedto zero, 100, 150, 200 and 300 kPa of suction

0·8

0·7

0·6

0·5

0·4

0·3

0·2

0·1

0

Coe

ffic

ient

of

eart

h pr

essu

re a

t res

t,K

0

0 100 200 300 400Suction, : kPas

Fig. 28. Variation of the coefficient of earth pressure at restwith suction during K0 loading

0·9

0·8

0·7

0·6

0·5

0·4

0·3

0·2

0·1

0

Coe

ffic

ient

of

eart

h pr

essu

re a

t res

t,K

0

0 100 200 300 400Suction, : kPas




Fig. 30. The effect of initial conditions on the variation of thecoefficient of earth pressure at rest against suction

140

120

100

80

60

40

20

0

�20

�40

Dev

iato

r st

ress

, :

kPa

q

0 100 200 300 400Mean net stress, : kPap u� a




Fig. 29. Stress paths during K0 loading for dry engineered, dryunengineered and wet unengineered specimens



quently the yielding characteristics. It is widely accepted thatthe yield locus separating elastic behaviour from elasto-plastic behaviour is elliptical and symmetrical about themean effective stress axis, for soils with an isotropic stresshistory (Graham et al., 1988; Sivakumar et al., 2001). Therotation of the yield locus for unsaturated soils also appearsto be true, despite the distinctly more complex structure,which is largely attributed to the bi-modal pore size distribu-tion (Zakaria et al., 1995; Cui & Delage, 1996; Sivakumaret al., 2010b). The data reported here focuses on whetherthe anisotropic loading history was responsible for makingthe specimen behave anisotropically.

As part of this research one specimen that was takenthrough K0 unloading was subsequently isotropically loaded(the dry unengineered fill equalised under a suction value of100 kPa). Note that the specimens for the present researchwere extracted from samples initially compressed underisotropic stress conditions. Therefore, specimens should haveinherited isotropic stress–strain properties. During the subse-quent isotropic compression the lateral strain experienced bythe specimen was considerably greater than the verticalstrain. This indicates that K0 loading has generated anisotro-pic stress–strain properties as shown in Fig. 32. In the earlystages of loading, the stress–strain behaviour was marked byanisotropy. However, the continued isotropic loading causedthe stress–strain properties to tend toward isotropic behav-iour, indicated by the ratio between the lateral and axialstrain approaching the 458 line as shown in Fig. 32. Thisobservation provides some evidence to support the view thata previous K0 stress history indeed inflicted a degree ofanisotropy in terms of the specimen stress–strain properties.

CONCLUSIONSCarefully controlled tests were conducted using the twin-

cell stress path apparatus in order to examine the effects ofwetting of unsaturated soils in relation to the stress regimeunder lateral confinement. Investigations were also carriedout in order to examine the traditional concept of K0 uponloading and unloading. The following concluding remarkscan be made.

(a) The wetting of compacted clay specimens results in anincrease in lateral stresses and therefore an increase inK0. The magnitude of increase in the lateral stress is

affected by the initial condition of the soil such as theinitial specific volume and compaction water content.

(b) The position of the LC yield locus is an importantfactor that determines the likely response of the soil.During wetting under laterally restrained conditions thestress path traverses inside, but close to the LC yieldlocus. The stress path will never cross the yield locusunder the above testing conditions even for unengi-neered specimens.

(c) The bi-modal pore size distribution has a particularinfluence on the volumetric response during wettingunder anisotropic loading conditions. The work hasshown that the aggregates attempt to expand in avertical direction rather than a lateral direction as theresistance to expansion in the lateral direction is high,due to high lateral pressures.

(d ) The maximum lateral stress condition (or K0) isgenerally affected by the initial condition but unaffectedby the target suction value under wetting. The final K0

value reduced gradually with applied suction.(e) The K0 during loading (under normally consolidated

conditions) is unaffected by suction and its value isapproximately equivalent to that of reconstituted kaolin.

( f ) K0 loading induced a degree of anisotropy in thespecimen that resulted in a significant difference in thestress–strain characteristics of the soil, which inheritedisotropic properties before being subjected to aniso-tropic (K0) loading.

The above conclusions have implications for geotechnicalstructures constructed on or using compacted clay fills. Anexample is the use of heavily compacted fills for backfillingretaining walls. The fill may become saturated owing toenvironmental changes (i.e. rainfall) The saturation of the fillwill be associated with a rapid increase in lateral stresses, asidentified from the present research. This can lead to in-stability of the retaining structures.

ACKNOWLEDGEMENTSFunding for the research was provided by DEL (NI). The

authors would like to thank Dr G. Gallagher (GSI, NI) andK. V. Senthilkumaran (P. J. Carey Ltd) for their support toresearch at Queen’s University Belfast.

180

160

140

120

100

80

60

40

20

0

�20

Dev

iato

r st

ress

, :

kPa

q

0 100 200 300 400 500

Mean stress, : kPap u� a

23H0

23H300

23L100

Fig. 31. Stress paths during K0 unloading

1·4

1·2

1·0

0·8

0·6

0·4

0·2

0

Axi

al s

trai

n,: %

ε a

0 0·2 0·4 0·6 0·8 1·0 1·2 1·4Horizontal strain, : %εh

45º

45º

Fig. 32. Vertical strain plotted against horizontal strain duringisotropic loading



NOTATIONK0 coefficient of earth pressure at rest

M(s) slope of the critical state line in the q–p plane forunsaturated soils

p mean net stressq deviator stresss suction

ua pore air pressurev specific volume slope of the K0 line in the q–p plane

(s) intercept of the critical state line in the q–p plane�h total horizontal stress�v total vertical stress�9 effective angle of internal friction

REFERENCESAhmed, S., Lovell, C. W. & Diamond, S. (1974). Pore sizes and

strength of compacted clay. J. Geotech. Engng Div. ASCE 100,No. 4, 407–425.


Blanchfield, R. & Anderson, W. F. (2000). Wetting collapse inopencast coalmine backfill. Proc. Inst. Civil Engrs – Geotech.Engng 143, No. 3, 139–149.

Blatz, J. A. & Graham, J. (2003). Elastic–plastic modelling ofunsaturated soil using results from a new triaxial test withcontrolled suction. Geotechnique 53, No. 1, 113–122, doi:10.1680/geot.2003.53.1.113.

Carder, D. R. (1988). Earth pressures on retaining walls andabutments. Ground Engng 21, No. 5, 7–10.

Casagrande, A. (1936). The determination of the pre-consolidationload and its practical significance. Proc. 1st Int. Conf. SoilMech., Cambridge, Massachusetts 3, 60–64.

Charles, J. A. (1993). Building on fill: geotechnical aspects. Wat-ford, UK: The Building Research Establishment.

Charles, J. A., Burford, D. & Hughes, D. B. (1993). Settlement ofopencast coal mining backfill at Horsley 1973–1992. In Engi-neering fills (ed. Clarke). London: Thomas Telford, pp. 429–440.

Chen, F. H. (1987). Lateral expansion pressure on basement walls.Proc. 6th Int. Conf. Expansive Soils, New Delhi, India 1, 55–59.

Crooks, J. H. A. & Graham, J. (1976). Geotechnical properties ofthe Belfast estuarine deposit. Geotechnique 26, No. 2, 293–315,doi: 10.1680/geot.1976.26.2.293.


Cuisinier, O. & Laloui, L. (2004). Fabric evolution during hydro-mechanical loading of a compacted silt. Int. J. Numer. Analyt.Methods Geomech. 28, No. 6, 483–499.

Delage, P. & Graham, J. (1995). Mechanical behaviour of unsatu-rated soils: Understanding the behaviour of unsaturated soilsrequires reliable conceptual models. Proc. 1st Int. Conf. Unsatu-rated Soils, Paris, France 3, 1223–1256.

Delage, P., Audigiuer, M., Cui, Y. J. & Howat, M. D. (1996).Microstructure of a compacted silt. Can. Geotech. J. 33, No. 1,150–158.

Estabragh, A. R., Javadi, A. A. & Boot, J. C. (2004). Effect ofcompaction pressure on consolidation behaviour of unsaturatedsilty soil. Can. Geotech. J. 41, No. 3, 540–550.

Gallipoli, D., Gens, A., Sharma, R. & Vaunat, J. (2003). An elasto-plastic model for unsaturated soil including the effect of satura-tion degree on mechanical behaviour. Geotechnique 53, No. 1,123–135, doi: 10.1680/geot.2003.53.1.123.


Graham, J., Crooks, J. H. A. & Lau, S. L. K. (1988). TechnicalNote: Yield envelopes: identification and geometric properties.Geotechnique 38, No. 1, 125–134, doi: 10.1680/geot.1988.38.1.125.

Highways Agency (1995). In association with the Scottish Develop-

ment Department, the Welsh Office & the Department for theEnvironment for Northern Ireland. Earthworks design and pre-paration of contract documents, HA 44/91. In Design manualfor roads and bridges, 4, Geotechnical drainage. London: High-ways Agency.

Highways Agency (2004). In association with the Scottish Develop-ment Department, the Welsh Office & the Department for theEnvironment for Northern Ireland. Manual of contract docu-ments for highways works MCHW1; Specification for highwaysworks, Series 600, Earthworks. London: Highways Agency.

Hilf, J. N. (1956). An investigation of pore water pressure incompacted cohesive soils, Technical Memorandum 654. Denver,Colorado: US Department of Interior Bureau of Reclamation.

Jaky, J. (1948). Pressure in soils. Proc. 2nd Int. Conf. SoilMechanics and Foundation Engineering, Rotterdam, the Nether-lands 1, 130–107.

Jardine, R. J., Symes, M. J. & Burland, J. B. (1984). The measure-ment of soil stiffness in the triaxial apparatus. Geotechnique 34,No. 3, 323–340, doi: 10.1680/geot.1984.34.3.323.

Lloret, A., Villar, M. V., Sanchez, M., Gens, A., Pintado, X. &Alonso, E. E. (2003). Mechanical behaviour of heavily com-pacted bentonite under high suction changes. Geotechnique 53,No. 1, 27–40, doi: 10.1680/geot.2003.53.1.27.

Lu, N. & Likos, W. J. (2004). Unsaturated soil mechanics. Chiche-ster: John Wiley.

Maatouk, A., Leroueil, S. & La Rochelle, P. (1995). Yielding andcritical state of a collapsible unsaturated silty soil. Geotechnique45, No. 3, 465–477, doi: 10.1680/geot.1995.45.3.465.

Maswoswe, J. (1985). Stress path for a compacted soil duringcollapse due to wetting. PhD thesis, Imperial College London,UK.

Qi, Y., Al-Mukhtar, M., Alcover, J. F. & Bergaya, F. (1996).Coupling analysis of macroscopic and microscopic behaviour inhighly consolidated Na-laponite clays. Appl. Clay Sci. 11, No. 2,185–197.

Sitharam, T. G., Sivapullaiah, P. V. & Subba Rao, K. S. (1995).Shrinkage behaviour of compacted unsaturated soils. Proc. 1stInt. Conf. Unsaturated Soils, Paris, France 1, 195–200.

Sivakumar, V. & Wheeler, S. J. (2000). Influence of compactionprocedure on the mechanical behaviour of an unsaturated com-pacted clay, Part 1: Wetting and isotropic compression. Geotech-nique 50, No. 4, 359–368, doi: 10.1680/geot.2000.50.4.359.

Sivakumar, V., Doran, I. G., Graham, J. & Johnson, A. (2001). Theeffect of anisotropic elasticity on the yielding characteristics ofoverconsolidated natural clay. Can. Geotech. J. 38, No. 1, 125–137.

Sivakumar, R., Sivakumar, V., Blatz, J. & Vimalan, J. (2006). Twin-cell stress path apparatus for testing unsaturated soils. Geotech.Testing J., ASTM 29, No. 2, 1–5.

Sivakumar, V., Sivakumar, R., Brown, J., Murray, E. J. & Mack-innnon, P. (2010a). Constitutive modelling of unsaturated kaolin(with isotropic and anisotropic stress history). Part 1: Wettingand compression behaviour. Geotechnique 60, No. 8, 581–594,doi: 10.1680/geot.8.P.007.

Sivakumar, V., Sivakumar, R., Murray, E. J., Brown, J. & Mac-Kinnon, P. (2010b). Constitutive modelling of unsaturated kaolin(with isotropic and anisotropic stress history). Part 2: Perform-ance under shear loading. Geotechnique 60, No. 8, 595–609,doi: 10.1680/geot.8.P.008.

Skinner, H. D. (2001). Construction on fill. Proceedings of theproblematic soils symposium, Nottingham, pp. 127–143.

Skinner, H. D., Charles, J. A. & Watts, K. S. (1999). Grounddeformations and stress redistribution due to a reduction involumes of zones of soil at depth. Geotechnique 49, No. 1,111–126, doi: 10.1680/geot.1999.49.1.111.

Tang, G. X. & Graham, J. (2002). A possible elastic-plastic frame-work for unsaturated soils with high-plasticity. Can. Geotech. J.39, No. 4, 894–907.

Tarantino, A. (2007). Technical Note: A possible critical stateframework for unsaturated compacted soils. Geotechnique 57,No. 4, 385–389, doi: 10.1680/geot.2007.57.4.385.

Thom, R. (2007). Performance of unsaturated soil under monotonicand repeated loading. PhD thesis, Queen’s University of Belfast,UK.

Thom, R., Sivakumar, V., Sivakumar, R., Murray, E. J. & Mack-innon, P. (2007). Technical Note: Pore size distribution of



unsaturated compacted kaolin: the initial states and final statesfollowing saturation. Geotechnique 57, No. 7, 469–474, doi:10.1680/geot.2007.57.7.469.

Toll, D. G. (1990). A framework for unsaturated soil behaviour.Geotechnique 40, No. 1, 31–44, doi: 10.1680/geot.1990.40.1.31.

Toll, D. G. (1999). A data acquisition and control system forgeotechnical testing. In Computing developments in civil andstructural engineering (eds B. Kumar and B. H. V. Topping).Edinburgh: Civil-Comp Press, pp. 237–242.

Wheeler, S. J. (1991). Technical note: An alternative framework forunsaturated behaviour. Geotechnique 41, No. 2, 257–261, doi:10.1680/geot.1991.41.2.257.

Wheeler, S. J. & Sivakumar, V. (1995). An elasto-plastic critical

state framework for unsaturated soil. Geotechnique 45, No. 1,35–53, doi: 10.1680/geot.1995.45.1.35.

Wheeler, S. J. & Sivakumar, V. (2000). Influence of compactionprocedure on the mechanical behaviour of an unsaturated com-pacted clay. Part 2: shearing and constitutive modelling. Geo-technique 50, No. 4, 369–376, doi: 10.1680/geot.2000.50.4.369.

Wheeler, S. J., Sharma, R. S. & Buisson, M. S. R. (2003). Couplingof hydraulic hysteresis and stress–strain behaviour in unsatu-rated soils. Geotechnique 53, No. 1, 41–54, doi: 10.1680/geot.2003.53.1.41.

Zakaria, I., Wheeler, S. J. & Anderson, W. F. (1995). Yielding ofunsaturated compacted kaolin. Proc. 1st Int. Conf. UnsaturatedSoils, Paris, France 1, 223–228.



Session 3

Benchmarking of Techniques and Models


D’Onza, F. et al. (2011). Geotechnique 61, No. 4, 283–302 [doi: 10.1680/geot.2011.61.4.283]

109

Benchmark of constitutive models for unsaturated soils

F. D’ONZA 1, D. GALLIPOLI 1, S . WHEELER1, F. CASINI 2, J. VAUNAT 3, N. KHALILI 4, L . LALOUI 5,C. MANCUSO6, D. MAS IN 7, M. NUTH 8, J. M. PEREIRA 9 and R. VASSALLO1 0

The paper presents a collaborative piece of researchundertaken by seven research teams from different uni-versities within the ‘Mechanics of Unsaturated Soils forEngineering’ (MUSE) network. The objective is to bench-mark different approaches to constitutive modelling ofmechanical and water retention behaviour of unsaturatedsoils by comparing simulations of suction-controlled andconstant water content laboratory tests. A set of 13triaxial and oedometer laboratory tests, covering themechanical and water retention behaviour of an unsatu-rated compacted silty soil under a variety of stress paths,has been provided by one of the seven participatingteams. This data set has been used by the other six teamsfor calibrating a constitutive model of their choice, whichhas been subsequently employed for predicting strainsand degree of saturation in three of the 13 tests used forcalibration, as well as in one ‘blind’ test for whichexperimental results had not been previously disclosed.By comparing predictions from all teams among them-selves and against experimental data, the work highlightsthe capability of some of the current modelling ap-proaches to reproduce specific features of the mechanicaland water retention behaviour of unsaturated soils help-ing to identify potential areas of weakness where futureresearch should focus. It also demonstrates the dispersionof results to be expected when different constitutivemodels, independently calibrated by different teams ofresearchers, are used to predict soil behaviour along thesame stress path.

KEYWORDS: constitutive relations; laboratory tests; partialsaturation; plasticity; suction

La presente communication presente des travaux derecherche realises par sept equipes de recherche dedifferentes universites, au sein du reseau MUSE. L’objec-tif de cette recherche est de comparer differentes meth-odes pour la modelisation constitutive de comportementsmecaniques et de retenue de l’eau de sols non satures, encomparant des simulations d’essais en laboratoire avecaspiration controlee et teneur en eau constante. Une dessept equipes participant a cette recherche a fourni unensemble de 13 essais triaxiaux et sur oedometre enlaboratoire, portant sur le comportement mecanique etde retenue de l’eau d’un sol silteux compacte, soumis aune serie de chemins de contrainte. Cet ensemble dedonnees a ete utilise par les six autres equipes pour lecalibrage d’un modele constitutif de leur choix, qui a eteutilise par la suite pour predire les deformations et ledegre de saturation dans trois des 13 tests utilises pour lecalibrage, ainsi que dans un test anonyme, pour lequelles resultats experimentaux n’avaient pas ete divulguesprecedemment. En comparant les predictions de toutesles equipes entre elles et en fonction de donnees experi-mentales, cette recherche met en lumiere la capacite quepresentent certaines methodes de modelisation actuellespour la reproduction des caracteristiques specifiques ducomportement mecanique et de retenue de l’eau de solsnon satures, afin de contribuer a l’identification de zonesde faiblesse potentielles sur lesquelles on devrait con-centrer les travaux de recherche dans l’avenir. Elle de-montre egalement la dispersion de resultats a prevoirlorsque l’on utilise differents modeles constitutifs, calibresindependamment par differentes equipes de chercheurs,pour predire le comportement du sol le long du memechemin de contraintes.

INTRODUCTIONThis paper presents a collaborative piece of research under-taken by seven universities to benchmark different mechani-

cal and water retention soil models. The objective is todemonstrate the variability of predictions typically obtainedwhen the soil response along a given hydromechanical stresspath is independently simulated by different researchersusing different constitutive models, albeit calibrated from asingle set of experimental data. From an engineering per-spective, this provides an indication of the discrepancies ofpredicted behaviour that can potentially occur in geo-technical design as a result of both the choice of constitutivemodel and the subsequent calibration on the basis ofsuction-controlled laboratory data.

The benchmarking exercise took place in the framework ofa wider scientific programme carried out between 2004 and2008 by the ‘Mechanics of Unsaturated Soils for Engineer-ing’ (MUSE) ‘Marie Curie’ Research Training Network withthe financial support of the European Commission. It in-volved seven teams of researchers at different universities,namely the Universitat Politecnica de Catalunya (UPC) inSpain, the University of Glasgow (UGLAS) in the UK, theUniversita di Napoli Federico II (UNINA) in Italy, the EcoleNationale des Ponts et Chaussees (ENPC) in France, theEcole Polytechnique Federale de Lausanne (EPFL) in Swit-zerland, the University of New South Wales (UNSW) inAustralia and Charles University (CU) in the Czech Republic.

Manuscript received 7 March 2010; revised manuscript accepted 11January 2011.Discussion on this paper closes on 1 September 2011, for furtherdetails see p. ii.1 School of Engineering, University of Glasgow, Glasgow, UK2 Institute for Geotechnical Engineering, ETHZ, Zurich, Switzer-land3 Departamento de Ingenierıa del Terreno, Cartografica y Geofısica,Universitat Politecnica de Catalunya, Barcelona, Spain4 School of Civil and Environmental Engineering, University ofNew South Wales, Sydney, Australia5 Soil Mechanics Laboratory, Ecole Polytechnique Federale deLausanne, Lausanne, Switzerland6 Dipartimento di Ingegneria Idraulica, Geotecnica ed Ambientale,Universita di Napoli Federico II, Napoli, Italy7 Faculty of Science, Charles University, Prague, Czech Republic8 Soil Mechanics Laboratory, Ecole Polytechnique Federale deLausanne, Lausanne, Switzerland9 Universite Paris-Est, Navier – Cermes, Ecole des Ponts ParisTech,Marne-la-Vallee, France10 Dipartimento di Strutture, Geotecnica, Geologia Applicata,Universita della Basilicata, Italy


At the start of the exercise, UPC provided the other sixteams with results from a set of 13 laboratory tests oncompacted silt published in the PhD thesis of Casini (2008).The six teams then chose a constitutive model, calibratedthe relevant parameter values on the basis of the laboratorydata provided and predicted the deformation as well as thewater retention behaviour during three of the 13 calibrationtests plus one blind test for which experimental results hadnot been published.

The exercise was coordinated by UGLAS, which circu-lated spreadsheets with laboratory test results together withan accompanying document containing soil data and adescription of the sample preparation procedure. The sixteams of constitutive modellers were asked to refrain fromlooking at additional information contained in the PhD thesisof Casini (2008) or related publications. Standard returnforms were also circulated for completion by each team withtheir predictions. Teams also had to provide the list ofparameter values used in their simulations, together with abrief description of the procedure followed during calibrationof their chosen model.

Readers interested in replicating this benchmarking exer-cise by using a constitutive model of their choice can down-load electronic copies of the specification document, returnforms and experimental data spreadsheets from the MUSEwebsite (http://muse.dur.ac.uk/) or, alternatively, contact oneof the authors to obtain copies of the relevant documenta-tion.

CONSTITUTIVE MODELSSeven constitutive models have been considered, one for

each team with the exception of UGLAS that returnedpredictions from two models (see Table 1). Hereafter, thetheoretical bases of the different constitutive frameworks arecompared with reference to the water retention behaviour,stress tensor definition, effect of suction on mechanicalbehaviour and nature of irreversible deformation. For thedetailed formulation of each model, the reader is invited torefer to the original articles listed in Table 1.

Water retention modelsAll water retention models assume a relationship between

degree of saturation Sr and suction s ¼ ua � uw (where ua

and uw are the pore-air and pore-water pressures respec-tively) that depends on volumetric strain, by predicting ashift of the Sr –s curve towards the higher suction range asporosity decreases.

Models are divided into two primary groups depending onwhether hydraulic hysteresis is neglected (CU, ENPC,UGLAS-1, UGLAS-2 and UNINA) or accounted for (EPFLand UNSW). In the former group, irreversible changes ofdegree of saturation during wetting–drying cycles are causedby irreversible volumetric strains alone whereas, in the lattergroup, they are attributable to both plastic volumetric strainsand water retention hysteresis.

The five models of the first primary group are furthersub-divided into two categories depending on the form ofthe relationship linking degree of saturation, suction andchanges in pore volume. In the first category, UGLAS-1,UGLAS-2 and UNINA adopt the Van Genuchten waterretention curve (Van Genuchten, 1980)

Sr ¼ 1þ (Æs)nð Þ�m(1)

where m and n are model parameters. The effect of soildeformability is introduced by expressing parameter Æ (re-lated to the air entry suction) as a power function of void

ratio following the Gallipoli, Wheeler and Karstunen waterretention model (Gallipoli et al., 2003).

In the second category, the ENPC and CU models adoptthe Brooks and Corey water retention curve (Brooks &Corey, 1964)

Sr ¼s

se

� ��º) log Sr ¼ �º log

s

se

(2)

described by a line in the log Sr –log s plane with slope ºand intercept se (the latter coinciding with the air entrysuction). The effect of soil deformability is introduced byexpressing the slope º and air entry suction se as functionsof porosity according to different mathematical formulationsin the two models.

In the second primary group, the EPFL and UNSWmodels assume a ‘main’ hysteretic loop, described by a maindrying and a main wetting curve, which bounds the regionof attainable values for degree of saturation and suction. Themajor difference between these two models lies in thedefinition of the main curves, which are parallel straightlines in the log Sr –log s plane for UNSW (i.e. lines describedby equation (2) with se equal to either the air entry or airexpulsion suction depending on whether a main drying orwetting line is considered) and parallel straight lines in theSr –log s plane for EPFL. A family of scanning lines of fixedslope spans the distance between the two main curves tosimulate suction reversals starting from main wetting ormain drying conditions. In both models, volumetric strainsproduce a rigid translation of the main hysteretic loop alongthe suction axis, that is a translation that keeps the slopeand the relative distance of the main lines constant. Aresidual value of degree of saturation is also introduced as alower limit when suction grows large.

Definition of constitutive stress tensorDepending on the model considered, the constitutive stress

tensor �9 is differently defined as a function of net stress� ¼ �tot � ua � 1 (where �tot is the total stress tensor andua � 1 is the isotropic tensor of pore air pressure), suction sand degree of saturation Sr.

The following general definition of the constitutive stresstensor is introduced to help in distinguishing between differ-ent formulations

�9 ¼ � þ �1 � s � 1þ �2 � 1 (3)

where �1 is a factor between zero and one weighing theeffect of suction on the solid skeleton and �2 is an additiveterm measuring energy changes in the phase interfaces(Coussy & Dangla, 2002).

Gens (1996) defined three classes of constitutive stressesdepending on whether (a) �1 ¼ 0, (b) �1 ¼ �1(s) depends onsuction but not degree of saturation for s . se (with �1 ¼ 1for s < se) or (c) �1 ¼ �1(Sr) depends on degree of saturationand possibly suction (with �1 ¼ 1 for Sr ¼ 1). However, Gens(1996) did not consider the additive term �2 measuringenergy changes in the phase interfaces in equation (3), andhis classification is here expanded to introduce a fourth classof constitutive stresses to contemplate this extra case.

Both second and third classes define the constitutive stressas the sum of the net stress tensor plus the product of theisotropic suction tensor multiplied by a scalar coefficientvarying between zero and one. A constitutive stress of thethird class can therefore be recast into a constitutive stressof the second class (and vice versa a constitutive stress ofthe second class can be recast into a constitutive stress ofthe third class) by using the chosen water retention model torelate degree of saturation to suction. Of course, this water

110 D’ONZA ET AL.


Ta

ble

1.

Co

nst

itu

tive

mo

del

su

sed

by

pa

rtic

ipa

tin

gte

am

s

Tea

mW

ater

rete

nti

on

mo

del

Mec

han

ical

mo

del

Dif

fere

nce

wit

hp

ubli

shed

ver

sio

ns

Ref

eren

cesa

tura

ted

mec

han

ical

mo

del

CU

Mas

ın(2

01

0)

Mas

ın&

Kh

alil

i(2

00

8)

Th

ed

efin

itio

no

fth

eco

nst

itu

tive

stre

ssvar

iable

tak

esin

toac

cou

nt

the

dep

end

ency

of

air

entr

yval

ue

on

vo

idra

tio

,as

pre

dic

ted

by

the

ado

pte

dw

ater

rete

nti

on

mo

del

pro

po

sed

by

Mas

ın(2

01

0).

Mas

ın(2

00

5)

EP

FL

Nu

th&

Lal

ou

i(2

00

8)

Nu

th&

Lal

ou

i(2

00

7)

Th

ep

ubli

shed

mec

han

ical

mo

del

use

sth

eV

anG

enu

chte

n(1

98

0)

equ

atio

nto

calc

ula

ted

egre

eo

fsa

tura

tio

nas

afu

nct

ion

of

suct

ion

.H

ow

ever

,an

imp

roved

wat

erre

ten

tio

nm

od

elh

asb

een

use

dh

ere,

wh

ich

inco

rpo

rate

sth

eef

fect

so

fb

oth

hy

dra

uli

chy

ster

esis

and

soil

den

sity

asd

escr

ibed

by

Nu

th&

Lal

ou

i(2

00

8).

Hu

jeu

x(1

98

5)

EN

PC

Bro

ok

s&

Co

rey

(19

64

)P

erei

raet

al.

(20

05

)T

he

infl

uen

ceo

fso

ild

ensi

tyo

nw

ater

rete

nti

on

beh

avio

ur

ism

od

elle

dby

exte

nd

ing

the

Bro

ok

san

dC

ore

yw

ater

rete

nti

on

curv

e(B

roo

ks

&C

ore

y(1

96

4))

toin

corp

ora

teth

efo

llow

ing

dep

end

ency

of

slo

pe,º

and

air-

entr

ysu

ctio

n,

s eo

np

oro

sity

,�

º(�

)¼

º 0ex

p(�

A�(��

ref))

s e(�

)¼

s e0

exp

(�A�(��

ref))

wh

ereº 0

,s e

0,

Aan

d�

ref

are

mo

del

par

amet

ers.

Cam

bo

u&

Jafa

ri(1

98

8)

UG

LA

S1

Gal

lip

oli

eta

l.(2

00

3)

Alo

nso

eta

l.(1

99

0)

No

ne

Ro

sco

e&

Bu

rlan

d(1

96

8)

UG

LA

S2

Gal

lip

oli

eta

l.(2

00

3)

D’O

nza

eta

l.(2

01

0)

No

ne

Wh

eele

ret

al.

(20

03

)

UN

INA

Gal

lip

oli

eta

l.(2

00

3)

Wh

eele

r&

Siv

aku

mar

(19

95

)T

he

pu

bli

shed

ver

sio

no

fth

em

ech

anic

alm

od

elh

asb

een

exte

nd

edby

add

ing

:(a

)a

Hvo

rsle

vsu

rfac

eo

fsl

op

eh

inth

eco

nst

ant

suct

ion

q–

p9

pla

ne

tom

od

elp

eak

stre

ng

tho

nth

ed

rysi

de

of

the

yie

ldlo

cus

(b)

an

on

-ass

oci

ated

flow

rule

foll

ow

ing

the

app

roac

hp

rop

ose

dby

Cu

i&

Del

age

(19

96

)re

lati

ng

incr

emen

tso

fp

last

icsh

ear

stra

ind�p s

and

pla

stic

vo

lum

etri

cst

rain

sd�p v

as

d�p s

d�p v

¼c 1þ

c 2(q=

p0)

1�

[q=

(Mpþ

)]

wh

ere

p0,M

and

hav

eth

esa

me

mea

nin

gas

inth

ew

ork

by

Wh

eele

r&

Siv

aku

mar

(19

95

)w

hil

ec 1

and

c 2ar

ead

dit

ion

alm

od

elp

aram

eter

s(c

2ta

kes

dif

fere

nt

val

ues

dep

end

ing

on

wh

eth

erth

est

ress

stat

eis

on

the

dry

or

wet

sid

eo

fth

ey

ield

locu

s)(c

)an

add

itio

nal

yie

ldli

mit

for

suct

ion

incr

ease

sim

ilar

toth

eex

iste

nce

of

the

SI

yie

ldcu

rve

inth

eB

BM

(Alo

nso

eta

l.,1

99

0).

Un

lik

eo

ther

s,th

ism

od

eld

oes

no

tas

sum

ean

yp

arti

cula

rsa

tura

ted

par

ent

form

ula

tio

n.

UN

SW

Kh

alil

iet

al.

(20

08

)N

on

eK

hal

ili

eta

l.(2

00

5)

BENCHMARK OF CONSTITUTIVE MODELS FOR UNSATURATED SOILS 111


retention model can also incorporate a dependency of therelationship between degree of saturation and suction onvolumetric strains and/or hydraulic hysteresis as relevant. Inspite of such similarities, the distinction between the secondand third class of constitutive stresses is retained in thiswork, consistent with the proposal by Gens (1996), becausethis makes it possible to distinguish between formulationsdepending on the ‘parent’ definition of constitutive stressfrom which multiple ‘child’ definitions of constitutive stressof another class can be obtained by combination with differ-ent water retention relationships.

The UGLAS-1, UGLAS-2 and UNINA models are formu-lated in terms of the net stress tensor, hence they fall in thefirst class for which �1 ¼ 0 and �2 ¼ 0.

The CU and UNSW models adopt a constitutive stress ofthe second class for which �1 depends on suction for s . se

while �2 ¼ 0. In the CU model the factor �1 is expressed as

�1 ¼s

se

� ��ª(4)

where ª is a model parameter (Khalili & Khabbaz, 1998)whereas, in the UNSW model, �1 varies according to ahysteretic law, similar to the water retention model, definedby a pair of main drying and main wetting lines of constantslope in the log �1 –log s plane having the form of equation(4). Scanning lines of fixed slope describe the variation of�1 during suction reversals starting from main wetting ormain drying conditions.

Owing to the particular choice of water retention relation-ship in the CU model, the primary constitutive stress can berecast in an alternative form consistent with the third classby combining equation (2) with equation (4) to yield

�1 ¼ Sª=ºr (5)

The same is true for the constitutive stress adopted byUNSW if the soil state belongs to a main drying or mainwetting curve.

The EPFL model is formulated in terms of a constitutivestress of the third class and assumes �1 ¼ Sr while �2 ¼ 0.

Finally, the ENPC model is formulated in terms of aconstitutive stress of the fourth class with �1 ¼ Sr while �2

provides a measure of the energy change in the phaseinterfaces through the following integral

�2 ¼2

3

ð1

Sr

s(Sr)dSr (6)

where s(Sr) is the inverse of the assumed water retentioncurve (Coussy & Dangla, 2002; Dangla, 2002).

Models adopting a constitutive stress tensor of the firstclass incorporate a suction-induced cohesive term into theircritical strength equation while models adopting a constitu-tive stress tensor of second, third and fourth classes predictcritical strength by means of a purely frictional law with nosuction-induced cohesive term. In addition, if a constitutivestress of the second, third or fourth class is employed, thereis no need for an independent relationship linking elasticstrains to suction.

Effect of suction on mechanical behaviourAll models incorporate suction as a scalar constitutive

variable in addition to the constitutive stress tensor. TheENPC, EPFL, UGLAS-1, UGLAS-2, UNINA and UNSWmodels adopt an elasto-plastic formulation where suctiondefines the expansion of the yield or bounding surfacetogether with the spacing and slope of the constant-suction

normal compression lines. The CU model adopts a hypoplas-tic formulation, where no distinction is made between elasticand plastic strains, but suction is still included in a similarmanner by controlling the size of the bounding surfacethrough the Hvorslev equivalent stress. For the sake ofsimplicity, the term ‘yield’ is used in the following toindicate stress states corresponding to the onset of irreversi-ble deformations in classical plasticity models as well asstress states corresponding to bounding conditions in bound-ing surface plasticity or hypoplastic models.

An increase of suction has a similar effect in all modelsproducing an increase of the mean yield constitutive stressin the absence of irreversible strains. This increase of meanyield constitutive stress corresponds to an expansion of theyield surface in the stress space. In the CU, ENPC, EPFLand UNSW models, this expansion is homologous in theq–p9 plane (p9 is the mean constitutive stress and q is thedeviator stress) at constant Lode angle with the centre ofhomology coinciding with the origin.

Note that, in the elasto-plastic formulations by EPFL andUNSW, suction is introduced as a hardening parameterrather than a stress variable. Nevertheless, in order tosimplify terminology, and given that the practical effects ofsuction are similar in all models, the expression ‘yieldcurve’ is generally used in the following to denote thevariation of yield stress in the s–p9 plane, regardless ofwhether suction is introduced as a stress variable or ahardening parameter.

Models are here distinguished according to the form ofthe yield curve in the s–p9 plane for s . se. A classificationapplicable to all models, regardless of the type of constitu-tive stress adopted, is introduced based on the followinggeneral expression of yield curve in the s–p9 plane

p90(s) ¼ ø1 � p90(se)ø2 þ ø3 (7)

In equation (7), p90(s) is the mean yield constitutive stressfor s . se and p90(se) is the mean yield constitutive stress ats ¼ se, which coincides with the volumetric hardening para-meter in the elasto-plastic models. For s < se, the mean yieldconstitutive stress p90(s) is calculated from the principle ofeffective stresses for saturated soils taking into account thedefinition of constitutive stress. The three symbols ø1, ø2

and ø3 denote three functions of suction governing theincrease of mean yield constitutive stress with increasingsuction.

Three classes of models are thus defined according to thefollowing three cases:

(a) ø1 ¼ ø1(s) is a function of suction (with ø1 ¼ 1 whens ¼ se) while ø2 ¼ 1 and ø3 ¼ 0 are both constant

(b) ø1 ¼ ø1(s) and ø2 ¼ ø2(s) are both functions ofsuction (with ø1 ¼ 1 and ø2 ¼ 1 when s ¼ se) whileø3 ¼ 0 is constant

(c) ø1 ¼ ø1(s) and ø3 ¼ ø3(s) are both functions ofsuction (with ø1 ¼ 1 and ø3 ¼ 0 when s ¼ se) whileø2 ¼ 1 is constant.

If the saturated normal compression line and the volumetricelastic law are given, the choice of one of the above threeclasses of yield curve implicitly fixes the form of theconstant-suction normal compression lines. If the saturatednormal compression line and the elastic compression law areboth represented by straight lines in the v–ln p9 plane (wherev is the specific volume and p9 is the mean constitutivestress), a yield curve of the first class corresponds to afamily of parallel straight constant-suction normal compres-sion lines, a yield curve of the second class corresponds to afamily of straight constant-suction normal compression linesof variable slopes and a yield curve of the third classcorresponds to a family of curved constant-suction normal

112 D’ONZA ET AL.


compression lines (see Appendix for a proof). In the Appen-dix it is shown that the slope and spacing of constant-suctionnormal compression lines are governed by the functionsø1 ¼ ø1(s) and ø2 ¼ ø2(s), respectively.

The EPFL and UNSW models adopt yield curves of thefirst class. The CU, UGLAS-1, UGLAS-2 and UNINAmodels adopt yield curves of the second class (with se ¼ 0for UGLAS-1, UGLAS-2 and UNINA). Note, however, thatthe particular model calibration by UNINA produces parallelconstant-suction normal compression lines in the v–ln p9plane and thus ø2 ¼ 1 (see equation (17)), which changesthe class of yield curve from second to first. Finally, theENPC model assumes a yield curve of the third class,although, again, the particular model calibration by ENPCimplies ø3 ¼ 0 (see equation (18)), which changes the classof yield curve from third to first.

In the second class, the CU, UGLAS-1 and UGLAS-2models assume that the functions ø1 ¼ ø1(s) and ø2 ¼ø2(s)are related through an exponential law as

ø1(s) ¼ p(1�ø2(s))ref (8)

where pref is a reference pressure such that the yield curvereduces to a straight vertical line in the s–p9 plane whenp90(se) ¼ pref , as shown by combining equations (7) and (8).

A classification matrix of the different models, accordingto the type of constitutive stress tensor and yield curve, isgiven in Table 2.

Irreversible mechanical behaviourA first distinction can be made depending on the way

irreversible mechanical behaviour is incorporated in the dif-ferent formulations. In the ENPC, EPFL, UGLAS-1,UGLAS-2, UNINA and UNSW models, irreversibility ofstrains is introduced by making use of standard elasto-plasticprinciples. The CU model is instead formulated in thecontext of the hypoplasticity theory, which does not distin-guish between elastic and plastic strains but describes irre-versible behaviour by means of an incrementally non-linearstress–strain relationship, where material stiffness dependson both stress state and direction of strain vector. Althoughbeing algebraically different, the CU model is based oncritical state soil mechanics, similarly to the other models(Gudehus & Masın, 2009) and incorporates the so-calledswept-out-memory surface (Masın & Herle, 2005) as analternative to the yield surface of elasto-plastic models.

A second distinction can be made between models thataccount for anisotropy of irreversible strains (ENPC, EPFL,UGLAS-2, UNINA and UNSW) and models that do not(CU and UGLAS-1).

Among anisotropic models, two groups are distinguisheddepending on whether anisotropy of plastic strains but notanisotropy of yielding and plastic hardening is taken into

account or, alternatively, anisotropy of plastic strains, yield-ing and plastic hardening are all considered.

The former group, which includes EPFL, UNINA andUNSW, assume constant-suction yield surfaces aligned withthe hydrostatic axis in the principal stress space, whoseevolution is governed by volumetric but not rotational hard-ening. Yielding and plastic hardening therefore depend onlyon the magnitude of the three principal stresses but not ontheir orientation with respect to the external reference sys-tem. Anisotropy of plastic strains is accounted for byintroducing a non-associative flow rule that predicts irrever-sible shear strain during mechanical loading or wetting onthe hydrostatic axis.

The latter group of models, including ENPC and UGLAS-2, adopt constant-suction yield surfaces which are aligned atan angle with the hydrostatic axis in the principal stressspace and whose evolution is governed by both volumetricand rotational hardening. In this case, yielding and plastichardening depend on the value of the three principal stressesas well as on their orientation with respect to the externalreference system. Given the inclination of the yield surface,both associative (as in the UGLAS-2 model) and non-associative flow rules (as in the ENPC model) can predictanisotropic plastic strains during loading or wetting on thehydrostatic axis.

Among the isotropic models, CU and UGLAS-1 assumeconstant-suction yield surfaces aligned along the hydrostaticaxis, with isotropic irreversible strains predicted duringplastic loading or wetting on the hydrostatic axis. In parti-cular, the UGLAS-1 model adopts a non-associative flowrule with no shear component for stress states on thehydrostatic axis.

Models can also be classified depending on whether asmooth or sharp transition between elastic and plastic behav-iour is predicted. CU and UNSW adopt models of the firsttype, based on a hypoplastic formulation in the case of CUand bounding surface plasticity in the case of UNSW. Boththese models predict a gradual build up of irreversiblestrains and show continuous derivatives of the stress–straincurve during monotonic loading. On the other hand, ENPC,EPFL, UGLAS-1, UGLAS2 and UNINA adopt models ofthe second type, which result in a discontinuity of thederivative of the stress–strain curve at the onset of yielding(although, in the case of EPFL, a different choice of modelparameters can also produce a smooth response consistentwith the behaviour of the first group of models).

EXPERIMENTAL DATA SETExperimental data were obtained from suction-controlled

triaxial and oedometer tests on compacted samples of Jos-signy silt (Casini, 2008). Fig. 1 shows the grading curve ofthis soil, which includes 5% sand, 70% silt and 25% clay.The specific gravity is equal to 2.69 with a liquid limit of32% and a plastic limit of 17%, which classifies the soil as

Table 2. Mechanical models classification matrix (parentheses indicate change of class due to calibration)

Yield/bounding curve Constitutive stress

(a) �1 ¼ 0, �2 ¼ 0 (b) �1 ¼ �1(s), �2 ¼ 0 (c) �1 ¼ �1(Sr), �2 ¼ 0 (d) �1 ¼ �1(Sr), �2 ¼ �1(Sr)

(a) ø1 ¼ ø1(s) (UNINA) UNSW EPFL (ENPC)

(b) ø1 ¼ ø1(s) and ø2 ¼ ø2(s) UGLAS-1 UGLAS-2UNINA

CU — —

(c) ø1 ¼ ø1(s) and ø3 ¼ ø3(s) — — — ENPC



silty clay of low plasticity according to the USCS classifica-tion.

Loose soil was initially mixed to a target water content of13%. By assuming that the water content of the wet mixturewas equal to the target value, an appropriate mass of soilwas compacted in layers by one-dimensional static compres-sion to achieve a target dry unit weight of 14.5 kN/m3.Triaxial samples of 7 cm diameter were compacted in fourlayers to a height of 14 cm, while oedometer samples of5 cm diameter were compacted in a single layer to a heightof 2 cm. This required the application of pressures rangingfrom 150 kPa to 200 kPa for each layer. No measurement ofsuction was performed after compaction.

It was subsequently noticed that water content duringcompaction was in some cases slightly different from thetarget value, which caused a variation of the post-compac-tion values of void ratio and degree of saturation amongsamples.

Calibration testsThe data set circulated for calibration of the different

models included the following 13 tests (here identified byusing the same codes as in the PhD thesis of Casini (2008)).

(a) Four compression tests performed in triaxial cells atconstant suction of 200 kPa. Of these, one test involvedisotropic loading (TX03) and three tests involvedanisotropic loading (TX04, TX08 and TX09).

(b) Four compression tests followed by shearing performedin triaxial cells at constant suction of 200 kPa. Ofthese, one test involved isotropic loading prior toshearing (TX02), two tests involved anisotropic loadingprior to shearing (TX06 and TX07) and one testinvolved no loading prior to shearing (TX01).

(c) Five Ko-compression tests (EDO-sat, EDO-10, EDO-50,EDO-100 and EDO-200) performed in an oedometerunder saturated conditions and at constant suctions of10 kPa, 50 kPa, 100 kPa and 200 kPa, respectively. Alltests followed loading–unloading–reloading paths atconstant suction. In addition, the test at a suction of200 kPa included two wetting–drying cycles, beforeloading and after reloading, under a constant verticalnet stress.

All tests are summarised in Table 3, which also lists thepost-compaction values of void ratio and degree of satura-tion, together with the net stress and suction imposed duringinitial equalisation prior to the start of the test. Note that, in

Table 3, p denotes the mean net stress (which is in generaldifferent from the mean constitutive stress p9), �v denotesthe vertical net stress (which is in general different from thevertical constitutive stress � 9v), ua denotes the pore airpressure and denotes the ratio between the increments ofdeviator and mean net stress during loading.

The stress paths for the tests carried out in the triaxialcells (TX01, TX02, TX03, TX04, TX06, TX07, TX08 andTX09) are shown in Fig. 2 (tests with loading only) and Fig.3 (tests with loading followed by shearing). All loadingstages were performed by ramping radial net stress at aconstant rate of 5 kPa/h while the shearing stages wereperformed by imposing a constant axial compression rate of0.2 mm/h. The stress paths for the tests carried out in theoedometer cells (EDO-sat, EDO-10, EDO-50, EDO-100 andEDO-200) are shown in Fig. 4. All loading and unloadingstages were performed by imposing discrete increments ofvertical net stress, with each increment maintained for aperiod between 8 and 16 h to ensure dissipation of excesspore-water pressures. In test EDO-200, the two wetting–drying cycles were performed by changing suction in inter-vals of 50 kPa.

During all tests, water exchange from/to the sample wasmeasured by means of two double-walled burettes and adifferential pressure transducer. One burette was connectedto the pore-water drainage line while the second burette wasisolated to provide a reference constant water level. Thedifferential pressure transducer measured the water levelchange in the first burette with respect to the reference oneand this was translated into a corresponding change of pore-water volume.

Blind testThe stress path for the blind test is shown in Fig. 5. After

equalisation at q ¼ 0 kPa, p ¼ 20 kPa and s ¼ 100 kPa, thesample was isotropically loaded at a constant suction of100 kPa and subsequently sheared at constant water content.It was not known a priori whether suction would increase ordecrease during constant water content shearing, hence theinitial loading stage was performed at a lower constantsuction of 100 kPa compared to the suction of 200 kPaimposed during the triaxial tests used for calibration ofmodel parameters. This ensured that the measured suctionvariation during constant water content shearing would fallbetween a suction of zero and 200 kPa, which is the rangecovered by the calibration data. A constant water contentshearing stage was deliberately chosen for the blind test soas to assess the ability of the different models to predictstrongly coupled hydromechanical soil behaviour.

During shearing at constant water content, suction wasmeasured by means of the axis translation technique. Thepore-water drainage line was isolated so that no pore-waterchanges were allowed while the pressure transducer was keptin communication with the sample. Suction was then meas-ured as the difference between the applied constant pore-airpressure and the pore-water pressure measured by the trans-ducer.

COMPARISON OF PREDICTED AND EXPERIMENTALBEHAVIOUR

All six teams returned predictions for tests TX02, TX07and EDO-200 as well as for the blind test. The initial valuesof void ratio and degree of saturation (i.e. the values of voidratio and degree of saturation corresponding to the equalisa-tion stress state) were also predicted for each test. ‘Incre-mental’ models predicting changes of void ratio, such asthose by CU, EPFL, ENPC and UNSW, calculated the initial

Clay Silt Sand Gravel0·002 0·06 2

100

90

80

70

60

50

40

30

20

10

0

Fin

er: %

0·0001 0·001 0·01 0·1 1 10Diameter: mm

Fig. 1. Grading curve of Jossigny silt

114 D’ONZA ET AL.


void ratio in each test by following a stress path originatingfrom a single reference value of void ratio corresponding tothe equalisation stress state in the blind test, which wasestimated on the basis of available data. On the other hand,‘integrated’ models that incorporate void ratio as a statevariable, such as those by UNINA, UGLAS-1 and UGLAS-2, calculated the initial void ratio directly from the equalisa-tion stress state in each test.

Models were calibrated by the different teams using all 13tests provided and trying to give the best interpretationpossible of the full set of experimental data. The chosenparameter values for all constitutive models are listed inTable 4.

Two general observations can be made here about thecalibration approaches followed by the different teams. First,ENPC and UNSW calibrated their respective water retentionmodels considering the entire set of available data, includingconstant-suction oedometric and triaxial loading stages aswell as the wetting–drying cycles of test EDO-200. On theother hand, CU, UGLAS-1, UGLAS-2 and UNINA used theconstant-suction oedometric and triaxial loading stages butnot the wetting–drying cycles of EDO-200, while EPFL onlyused the loading and wetting–drying stages of test EDO-200.

Table 3. Summary of tests used during calibration of constitutive models

Test code Stress path Post-compactionvoid ratio

Post-compactiondegree of saturation

Initial equalisation

Triaxial loading

TX03 Isotropic load( ¼ ˜q/˜p ¼ 0) @ s ¼ 200 kPa

0.86 0.43 q ¼ 0 kPa, p ¼ 20 kPa, s ¼ 200 kPa

TX04 Anisotropic load( ¼ ˜q/˜p ¼ 0.375) @ s ¼ 200 kPa

0.82 0.42 q ¼ 8 kPa, p ¼ 20 kPa, s ¼ 200 kPa


0.87 0.44 q ¼ 20 kPa, p ¼ 27 kPa, s ¼ 200 kPa


0.81 0.42 q ¼ 19 kPa, p ¼ 22 kPa, s ¼ 200 kPa

Triaxial loading followed by shearing at constant radial net stress

TX01 No load@ s ¼ 200 kPa

0.85 0.44 q ¼ 0 kPa, p ¼ 10 kPa, s ¼ 200 kPa

TX02 Isotropic load( ¼ ˜q/˜p ¼ 0) @ s ¼ 200 kPa

0.83 0.43 q ¼ 0 kPa, p ¼ 10 kPa, s ¼ 200 kPa


0.83 0.40 q ¼ 15 kPa, p ¼ 20 kPa, s ¼ 200 kPa


0.83 0.40 q ¼ 15 kPa, p ¼ 20 kPa, s ¼ 200 kPa

Ko loading

EDO-sat Load–unload–reload @ saturation 0.82 0.42 �v ¼ 1 kPa, s ¼ 0 kPa

EDO-10 Load–unload–reload @ s ¼ 10 kPa 0.82 0.38 �v ¼ 20 kPa, s ¼ 10 kPa



EDO-200 Wet–dry @ �v ¼ 20 kPa 0.81 0.38 �v ¼ 20 kPa, s ¼ 200 kPaLoad–unload–reload @ s ¼ 200 kPaWet–dry @ �v ¼ 800 kPa

600

500

400

300

200

100

0

q: k

Pa

Equalisation

Loading

0 100 200 300 400

p: kPa

TX09

TX08

TX04

TX03

Fig. 2. Stress paths for tests involving triaxial compression:TX03 – isotropic compression until a mean net stress of260 kPa; TX04 – anisotropic compression (� 0.375) until amean net stress of 285 kPa; TX08 – anisotropic compression(� 0.750) until a mean net stress of 370 kPa; and TX09 –anisotropic compression (� 0.875) until a mean net stress of370 kPa



Second, because only one experimental isotropic compressionat s ¼ 200 kPa was available from test TX03, all teams madeuse of additional data from constant-suction oedometric com-pression to calibrate the isotropic part of their mechanicalmodels.

Teams ENPC, EPFL and UNSW, whose models adoptyield curves of the first class in the s–p9 plane, started bydefining the shape of the yield curve based on measuredyield stresses with the spacing between constant-suctionnormal compression lines becoming then fixed as a conse-quence (see Appendix for the relationship between the shapeof the yield curve and spacing of constant-suction normalcompression lines). In the ENPC model, the shape of theyield curve was defined to match the yield stresses from thefive oedometric tests whereas, in the case of EPFL andUNSW, test EDO-200 was replaced by the isotropic loadingof test TX03 at s ¼ 200 kPa. This resulted in a steeperpredicted yield curve between s ¼ 50 kPa and s ¼ 200 kPafor these two models.

Teams CU, UGLAS-1, UGLAS-2 and UNINA, whosemodels adopt yield curves of the second class, startedinstead by defining the spacing between constant-suctionnormal compression lines on the basis of the five oedometrictests, which in turn fixed the shape of the yield curve in thes–p9 plane. A detailed explanation of the calibration proce-

dure followed by UGLAS can be found in the work byGallipoli et al. (2010).

In all seven models, the slopes of constant-suction normalcompression lines were mainly determined from oedometrictests, although ENPC, UGLAS-1, UGLAS-2 and UNINAtook some account of data from the loading stages of thetriaxial tests at s ¼ 200 kPa.

In the following, predictions for tests TX02, TX07 andEDO-200, as well as for the blind test, are compared amongthemselves and with experiments.

Triaxial test TX02Figure 6 compares predicted and experimental data during

the shearing stage of test TX02. Results from the loadingstage are not presented given that the maximum mean netstress applied during this stage was only 20 kPa, with allmodels predicting small and mainly reversible changes ofvoid ratio and degree of saturation over this limited stressrange.

With reference to Fig. 6, three types of results can bedistinguished:

(a) mechanical uncoupled results depending only on themechanical model but not on the water retention model

(b) hydraulic uncoupled results depending only on thewater retention model but not on the mechanical model

(c) coupled results depending on both mechanical andwater retention models.

CU, UGLAS-1, UGLAS-2, UNINA and UNSW all adoptstress tensors not depending on degree of saturation. Thus,for constant-suction tests, predictions in the q–�a and �v –�a

planes (see Figs 6(a) and 6(b)) by these five teams (�a and �v

are the axial and volumetric strains, respectively) representmechanical uncoupled results. The corresponding predictionsby ENPC and EPFL represent instead coupled results as, inthis case, the constitutive stress path is influenced by thechosen water retention model through the dependency of theconstitutive stress tensor on degree of saturation.

Predictions in the Sr –�v plane (see Fig. 6(c)) representinstead hydraulic uncoupled results for all teams. This isbecause all water retention models postulate a link betweendegree of saturation, suction and porosity, which can berecast as a link between degree of saturation, suction andvolumetric strain, once the initial value of porosity at thestart of shearing is taken into account. As a consequence,the shape of the predicted curve in the Sr –�v plane dependson the water retention relationship alone, although the rangeof volumetric strains over which degree of saturation variesdoes depend also on the predicted mechanical response.

Finally, the predicted Sr –�a relationships in Fig. 6(d)represent coupled results, as the shape of these curves comesfrom the combination of the previous Sr –�v and �v –�a

curves in Figs 6(b) and 6(c) respectively.In Fig. 6(a), the q–�a relationships predicted by CU,

UGLAS-1, UGLAS-2, UNINA and UNSW show an initialstiff response, that is in good agreement with experiments,as well as a relatively accurate prediction of critical strength.On the other hand, ENPC and EPFL predict lower values ofinitial stiffness and overestimate critical strength. Smallchanges of degree of saturation are calculated in Fig. 6(c)for both these models, which suggests that their differentpredictions in the q–�a plane are not attributable to theinfluence of the water retention relationship on the constitu-tive stress path (caused by the dependency of the constitutivestress tensor on degree of saturation). Instead the low shearstiffness predicted by the ENPC model is due to the narrowelastic range in the deviatoric plane and the particularkinematic hardening law inherited by the parent saturated

300

250

200

150

100

50

0

q: k

Pa

TX01

TX02

0 100 200p: kPa

(a)

Equalisation

Loading

Shear

600

500

400

300

200

100

0

q: k

Pa TX06

TX07

0 100 400

p: kPa(b)

Equalisation

Loading

Shear

200 300

Fig. 3. Stress paths for tests involving triaxial compressionfollowed by shearing: (a) TX01 – shearing to critical state (nocompression) and TX02 – isotropic compression until a meannet stress of 20 kPa followed by shearing to critical state;(b) TX06 – anisotropic compression (� 0.750) until a meannet stress of 100 kPa followed by shearing to critical state andTX07 – anisotropic compression (� 0.750) until a mean netstress of 200 kPa followed by shearing to critical state

116 D’ONZA ET AL.


model (see Table 1). The kinematic hardening law of theparent saturated model is also responsible for the significantoverestimation of critical strength in the ENPC case. On theother hand, the low initial stiffness predicted by the EPFLmodel is mainly the consequence of the underestimation ofthe elastic stiffness at low stresses.

In Fig. 6(b) all models except ENPC correctly predictdilatant behaviour, although the magnitude of dilation isunderestimated in all cases. For the models predicting dila-tant behaviour, the magnitude of the volumetric strain at the

end of shearing depends on the relative positions of theconstant-suction normal compression and critical state lines.It is therefore not surprising that the most accurate predic-tions are produced by the UNINA model, which offerscomplete flexibility in defining position and slope of con-stant-suction critical state lines in the v–ln p9 plane. On theother hand, in the EPFL and UNSW models, the position ofconstant-suction critical state lines can be varied to fitexperimental data but their slope is constant and equal to theslope of normal compression lines. The least flexibility is

250

200

150

100

50

0

s: k

Pa

0 400 800 1200 1600

σv: kPa(a)

Loading–reloading

Unloading

250

200

150

100

50

0

s: k

Pa

0 400 800 1200 1600

σv: kPa(b)

Loading–reloading

Unloading

250

200

150

100

50

0

s: k

Pa

0 400 800 1200 1600

σv: kPa(c)

Loading–reloading

Unloading

250

200

150

100

50

0

s: k

Pa

0 400 800 1200 1600

σv: kPa(d)

Loading–reloading

Unloading

250

200

150

100

50

0

s: k

Pa

0 400 800 1200 1600

σv: kPa(e)

Loading–reloading

Unloading

Wetting

Drying

Fig. 4. Stress paths for tests involving Ko loading: (a) EDO-sat – loading to a vertical effective stress of 800 kPa, unloadingto 100 kPa and reloading to 1600 kPa; (b) EDO-10 – loading to a vertical net stress of 800 kPa, unloading to 100 kPa andreloading to 1200 kPa; (c) EDO-50 – loading to a vertical net stress of 800 kPa, unloading to 100 kPa and reloading to1226 kPa; (d) EDO-100 – loading to a vertical net stress of 800 kPa, unloading to 100 kPa and reloading to 1080 kPa;(e) EDO-200 – wetting to a suction of 10, drying to 200 kPa, loading to a vertical net stress of 800 kPa, unloading to100 kPa, reloading to 800 kPa, wetting to a suction of 55 and drying to 200 kPa



offered by the CU, UGLAS-1 and UGLAS-2 models, forwhich the distance between constant-suction normal com-pression and critical state lines is fixed in their respectiveconstitutive formulations. Hence, the underestimation of dila-tant behaviour by the CU and UGLAS-1 models in Fig. 6(b)is a consequence of the excessive spacing between constant-suction critical state and normal compression lines in thev–ln p9 plane (similar to excessive spacing between saturatedcritical state and normal compression lines in the modifiedCam-clay model). Rather surprisingly, however, the volu-metric strain predicted at the end of shearing by theUGLAS-2 model provides one of the best matches to theexperimental data. This accurate prediction is caused byrotational hardening of the anisotropic yield locus, whichresults in a smaller distance between constant-suction normalcompression and critical state lines for the UGLAS-2 modelcompared to the CU and UGLAS-1 models.

Figure 6(c) shows hydraulic uncoupled predictions running

1200

1000

800

600

400

200

00 100 200 300 400 500 600

p: kPa

q: k

Pa

EqualisationIsotropic loadingShear at constant water content

Fig. 5. Stress paths for the blind test involving isotropiccompression until a mean net stress of 150 kPa followed byshearing to critical state at constant water content

Table 4. Model parameter values used by participating teams (values below double line represent initial values of state variables)

Team Mechanical model parameters Water retention model parameters

Symbol (units) Value Symbol (units) Value

CU �c(8) 36.0 se0 (kPa) 10º� (–) 0.09 ºp0 (–) 0.25k� (–) 0.0025 e0 (–) 0.7N (–) 0.925 — —r (–) 0.03 — —n (–) 0.055 — —l (–) 0 — —m (–) 2 — —

Saturated Hvorslev stress (kPa) 33.3 — —

ENPC K e0 (kPa) 15 000 se0(kPa) 5.0n (–) 0.6 º0 (–) 3.78� (kPa) 0.35 A (–) 16.0

Kp0 (kPa) 600 �ref (–) 0.465ª (–) 0.8 — —

a (kPa�1) 0.021 — —�0 (–) 7.05 — —�0 (–) �0.08 — —Re (–) 0.1 — —Rc (–) 0.6 — —k1 (–) 0.3 — —k2 (–) 0 — —k3 (–) 0 — —k4 (–) 0 — —

Q0 (kPa) 47 — —X (–) Null tensor — —

EPFL Ki (kPa) 150 000 Kh (–) 18Gi (kPa) 120 000 �h (–) 10ne (–) 1 se (kPa) 3�9 (8) 31.0 �H (kPa) 350�0 (–) 17 sDI (kPa) 15Æ (–) 0.75 Sres (–) 0.01a (–) 0.02 — —b (–) 0.01 — —c (–) 0.0001 — —d (–) 2 — —

redev (–) 0.01 — —

reiso (–) 0.1 — —ªs (–) 1.8 — —� (–) 2 3 10�5 — —

p9CR (kPa) 27 — —

(continued)

118 D’ONZA ET AL.


approximately parallel to the experiments in the Sr –�v plane,which confirms that all water retention models capturereasonably well the variation of degree of saturation withvolumetric strain (with the only exception of EPFL).

The vertical shift between curves in Fig. 6(c) is caused by

the different initial values of void ratio predicted at the startof the loading stage (i.e. after equalisation), which has animpact on the corresponding calculation of degree of satura-tion. As previously mentioned, initial values of void ratio atequalisation are computed by incremental models following

Table 4. (continued )

Team Mechanical model parameters Water retention model parameters

Symbol (units) Value Symbol (units) Value

UGLAS-1 k (–) 0.004 � (kPa�1) 1.318ks (–) 0.006 ł (–) 6.036

G (kPa) 36 000 m (–) 0.146N(0) at p9 ¼ pc (–) �0.4928 n (–) 1.341

º(0) (–) 0.1358 — —r (–) 1.19597 — —

� (kPa�1) 0.00397 — —pc (kPa) 826 699 057 — —

k (–) 0.138 — —M (–) 1.45 — —

p�0 (kPa) 30 — —

UGLAS-2 k (–) 0.004 � (kPa�1) 1.318ks (–) 0.006 ł (–) 6.036

G (kPa) 36 000 m (–) 0.146N(0) at p9 ¼ pc (–) �0.42641 n (–) 1.341

º(0) (–) 0.1358 — —r (–) 1.19597 — —

� (kPa�1) 0.00397 — —pc (kPa) 508 841 924 — —

k (–) 0.138 — —M (–) 1.45 — — (–) 92.047 — —b (–) 1 — —

pm(0) (kPa) 30 — —Æ 0.144 — —

UNINA k (–) 0.003 � (kPa�1) 0.51ks (–) 0.003 ł (–) 4.81ºs (–) 0.09 m (–) 0.22

G (kPa) 5000 n (–) 1.04h (–) 1.2 — —c1 (–) �0.2 — —

c2 wet side (–) 0.5 — —c2 dry side (–) 0.1 — —

º (–) 0.130/0.130� — —N (–) 1.753/1.880� — —M (–) 1.455/1.455� — — (kPa) 0/40� — —� (–) 0.130/0.100� — —ˆ (–) 1.665/1.820� — —

p0(0) (kPa) 52 — —

UNSW k (–) 0.0004 sex (kPa) 5.0� (–) 0.30 sae (kPa) 15.0�9c (8) 35.9 ºp (–) 0.24º (–) 0.16 for s > sex or sae � (–) 0.08N(–) 2.67 for s > 100 kPa

2.62 for s ¼ 50 kPa2.56 for s < sex or sae

� (–) 0.2

N (–) 2.25 — —R (–) 1.40 — —A (–) 1.0 — —km (–) 20.0 — —

Isotropic saturated yield stress (kPa) 98.7 — —

� The two values refer to s ¼ 0 and 200 kPa respectively (linear interpolation was used for intermediate suctions)



a stress path originating from a reference soil state wherevoid ratio had been previously estimated. There also appearsto be a systematic error between predicted and experimentalvalues of void ratio at the start of each test; that is allpredictions appear to make an error of consistent sign,which can be either positive or negative. This is due to thelarge experimental scatter of the post-compaction values ofvoid ratio among the tested samples (see Table 3). Thesedifferences are not entirely erased during equalisation andproduce a discrepancy of consistent sign between experimen-tal and predicted values at the start of each test.

Unlike the uncoupled hydraulic predictions of Fig. 6(c),different models provide distinct mechanical uncoupled pre-dictions (or nearly mechanical uncoupled predictions in thecase of ENPC and EPFL) in Fig. 6(b), with markedly differ-ent forms of variation of volumetric strain with axial strain.The differences between predictions in Fig. 6(b) are alsoreflected in the variability of coupled predictions in Fig.6(d).

Triaxial test TX07Predicted and experimental results are compared for test

TX07 in Fig. 7 (loading stage) and Fig. 8 (shearing stage).Figs 7(a), 8(a) and 8(b) show mechanical uncoupled predic-tions for CU, UGLAS-1, UGLAS-2, UNINA and UNSW,whereas they show coupled predictions for ENPC and EPFL(because of the different definition of constitutive stress inthese two groups of models). Figs 7(b) and 8(c) present

hydraulic uncoupled predictions for all models. Finally, Figs7(c) and 8(d) show coupled predictions resulting from thecombination of the previous two sets of results.

As expected, ENPC, EPFL, UNINA, UGLAS-1 andUGLAS-2 predict a sharp change of stiffness at yielding inFig. 7(a) while CU and UNSW show a gradual transitionfrom elastic to plastic behaviour that better reproduces theexperimental trend. Consistent predictions of yield stressesare obtained by the different models, which are also inreasonably good agreement with experiments. The smallerpreconsolidation stress predicted by ENPC is again theconsequence of the narrow elastic range assumed in thedeviatoric plane by this model.

The prediction by ENPC also shows a stiffer post-yieldresponse compared to other curves, a behaviour that issimilarly observed during constant-suction loading in theblind test and test EDO-200, as will be shown later. It isinteresting to note that the ENPC model predicts the stiffestpost-yield response in the e–ln p plane (where e is the voidratio) despite assuming the smallest plastic stiffness in thee–ln p9 plane, as indicated by the relatively small value ofparameter K

p0 in Table 4. This apparently surprising result is

the consequence of the incorporation of the additive term �2

in the constitutive stress definition of equation (3), whichresults in a comparatively stiffer response when void ratio isplotted against mean constitutive stress rather than mean netstress. It also suggests that the particular choice of constitu-tive stress tensor has an influence on calibration of plasticstiffness.

250

200

150

100

50

0

q: k

Pa

0 2 4 6 8 10 12 14 16 18 20εa: %(a)

CUUNSWENPCUGLAS-1UGLAS-2UNINAEPFLData

0 2 4 6 8 10 12 14 16 18 20εa: %(b)

2

1

0

�1

�2

�3

�4

ε v: %

45

44

43

42

41

40

39

38

37

36

35

Sr:

%

�4 �3 �2 �1 0 1 2

εv: %(c)

2018161412108642εa: %(d)

0

Sr:

%

35

36

37

38

39

40

41

42

43

44

45

Fig. 6. Predicted and experimental behaviour during shearing stage of test TX02: (a) deviator stress (q) plotted against axial strain(�a); (b) volumetric strain (�v) plotted against axial strain (�a); (c) degree of saturation (Sr) plotted against volumetric strain (�v);(d) degree of saturation (Sr) plotted against axial strain (�a)

120 D’ONZA ET AL.


Hydraulic uncoupled predictions in Figs 7(b) and 8(c)confirm that all water retention models capture relativelywell the dependency of degree of saturation on volumetricstrain, with the only exception of EPFL which significantlyoverestimates the increase of degree of saturation during thecompression stage. This is due to the particular waterretention calibration performed by this team and, morespecifically, to the choice of an excessively large value ofparameter �h (see Table 4), which relates degree of satura-tion to volumetric strain and had been selected on the basisof test EDO-200 alone.

With regard to the shape of predicted curves, the relation-ship between degree of saturation and void ratio in Fig.7(b), or between degree of saturation and volumetric strainin Fig. 8(c), shows smaller differences between models thanthe relationship between volumetric strain and axial strain inFig. 8(b) (with the exception of EPFL). This confirms thegreater uniformity of hydraulic uncoupled predictions com-pared to mechanical uncoupled predictions and, hence, in-dicates that greater consistency exists between waterretention models compared to mechanical models at least forpaths that do not involve suction or stress reversals. Simi-larly to test TX02, Fig. 8(b) shows that the prediction of thefinal volumetric strain is least accurate for CU and UGLAS-1, which is again due to the overestimation of the distancebetween constant-suction normal compression and criticalstate lines as previously explained. On the other hand, themost accurate prediction is this time provided by ENPCfollowed by UNINA.

Coupled predictions mirror features of mechanical un-coupled predictions because of their dependency on bothmechanical and water retention models. For example, thesharp mechanical yielding predicted by classic elasto-plasticmodels is also evident in the coupled predictions of Fig.7(c) but not in the hydraulic uncoupled predictions of Fig.7(b). Owing to the greater accuracy of hydraulic uncoupledpredictions compared to mechanical uncoupled predictions,potential errors in the predicted shape of coupled relation-ships are mainly the consequence of inadequacies of themechanical model rather than the water retention model(with the exception of EPFL where the calibration of thewater retention model is responsible for the large deviationfrom experimental results).

Blind triaxial testPredicted and experimental results are compared for the

blind test in Fig. 9 (loading stage at s ¼ 100 kPa) and Fig.10 (shearing stage at constant water content). As previouslydiscussed, Fig. 9(a) shows relatively large errors of consis-tent sign for the predicted initial values of void ratio,although, unlike test TX07, experimental values are nowunderestimated by all teams.

Compared to test TX07, greater inconsistencies exist be-tween predicted yield stresses in Fig. 9(a). In particular,ENPC, UGLAS-1, UGLAS-2 and UNINA underestimateyield stress while EPFL and UNSW overestimate it.

Calibration of the former group of four models took intoaccount all five constant-suction oedometric compressionswhen defining the shape of the yield curve in the s–p9plane. This was done either directly, by fitting yield stressesat different suctions, as in the case of ENPC, or indirectly,by fixing the spacing and slopes of constant-suction normalcompression lines in the v–ln p9 plane, as in the case ofUGLAS-1, UGLAS-2 and UNINA. In all four cases, oncethe shape of the yield curve had been defined, its positionwas adjusted to optimise prediction of yield stresses duringboth triaxial and oedometric tests. Fig. 9(a) shows that theabove procedure predicts a decrease of yield stress withdecreasing suction that slightly underestimates the measuredvalue at s ¼ 100 kPa.

In the UNSW and EPFL models, the shape of the yieldcurve in the s–p9 plane was instead defined after replacingthe yield stress of test EDO-200 with that of the isotropicloading of test TX03 (also at s ¼ 200 kPa), which resultedin a steep variation of yield stress with suction. Fors > 50 kPa, the yield stress remained approximately constantand equal to the value measured at s ¼ 200 kPa in test TX03so that the experimental yield stress at s ¼ 100 kPa is over-estimated (see Fig. 9(a)).

0·86

0·84

0·82

0·80

0·78

0·76

0·74

e

10 100


p: kPa(a)

65

60

55

50

45

40

35

Sr:

%

0·74 0·76 0·78 0·80 0·82 0·84 0·86e

(b)

10 100p: kPa

(c)

65

60

55

50

45

40

35

Sr:

%

Fig. 7. Predicted and experimental behaviour during loadingstage of test TX07: (a) void ratio (e) plotted against meannet stress (p); (b) degree of saturation (Sr) plotted against voidratio (e); (c) degree of saturation (Sr) plotted against mean netstress (p)



Similarly to test TX07, remarkably consistent shapes areobtained for the hydraulic uncoupled predictions in the Sr –eplane for six of the seven models (Fig. 9(b)), with thedifferent simulation from EPFL caused by an inappropriatechoice of value for parameter �h as previously explained.Coupled predictions in Fig. 9(c) are obtained by combiningthe two sets of curves in Figs 9(a) and 9(b) respectively.Coupled predictions therefore show features from both theabove sets of results, such as the occurrence of a sharpyielding point similar to Fig. 9(a), and the large increase ofdegree of saturation predicted by EPFL similar to Fig. 9(b).The initial value of degree of saturation is consistentlyoverestimated by all models in Figs 9(b) and 9(c), whichcorresponds to the systematic underestimation of initial voidratio in Fig. 9(a).

During the subsequent shearing stage at constant watercontent, six out of seven models calculate similar values ofcritical strength within a range of about 50 kPa (Fig. 10(a)).All predictions, however, fall short of the experimentalcritical strength by a margin greater than 100 kPa owing tothe unexpectedly high critical strength measured duringconstant water shearing. This is considerably higher than thestrength recorded in tests TX01, TX02, TX06 and TX07 at asuction of 200 kPa, despite a value of suction lower than200 kPa being measured at critical state during constantwater shearing.

Prediction of volumetric strains in Fig. 10(b) show that,unlike tests TX02 and TX07, the CU and UGLAS-1 modelsprovide the closest match to the experimental data, followed

by UNINA and UGLAS-2. The accurate prediction of volu-metric strain at critical state by the CU and UGLAS-1 modelsis rather unexpected and inconsistent with the previoussimulations of tests TX02 and TX07. This rather surprisingresult is possibly the consequence of the disagreement be-tween measured and predicted values of suction during con-stant water content shearing, as it will be shown later.

Considering all shearing stages in test TX02, test TX07and the blind test, UNINA provides the most accurateprediction of volumetric strains overall, which is an expectedresult given the flexibility of this model in fitting theexperimental volumetric behaviour at critical state. Ratherunexpected is perhaps the good accuracy of UGLAS-2,which confirms that the adoption of an anisotropic yieldlocus (i.e. a yield locus inclined at an angle with respect tothe hydrostatic axis in the principal stress space) improvesprediction of volumetric strains at critical state.

Similarly to tests TX02 and TX07, the shape of thepredicted curves in the Sr –�v plane (Fig. 10(c)) is closer to theexperimental data compared to predicted curves in the �v –�a

plane (Fig. 10(b)). Note, however, that – unlike the constantsuction tests TX02 and TX07 – in this case the predictedcurves in the Sr –�v plane do not depend on the water retentionmodel but only on the initial values of degree of saturation,Sr0 and void ratio, e0 at the start of shearing. The curves inFig. 10(c) can in fact be calculated by imposing the conditionof zero water content change during shearing while taking intoaccount the relation between void ratio and volumetric strain,that is e ¼ e0 � �v(1þ e0)


550

500

450

400

350

300

250

200

150

q: k

Pa

0 5 10 15 20 25 30εa: %(a)

0 5 10 15 20 25 30εa: %(b)

10

8

6

4

2

0

ε v: %

0 5 10 15 20 25 30εa: %(d)

�2 0 2 4 6 8 10εv: %(c)

70

65

60

55

50

45

40

Sr:

%

70

65

60

55

50

45

40

Sr:

%

�2

Fig. 8. Predicted and experimental behaviour during shearing stage of test TX07: (a) deviator stress (q) plotted against axialstrain (�a); (b) volumetric strain (�v) plotted against axial strain (�a); (c) degree of saturation (Sr) plotted against volumetricstrain (�v); (d) degree of saturation (Sr) plotted against axial strain (�a)

122 D’ONZA ET AL.


Sre ¼ Sr0e0 ) Sr ¼Sr0e0

e0 � �v(1þ e0)) Sr ¼

Sr0 n0

n0 � �v

(9)

where n0 is the value of porosity at the start of shearing.Consistent with equation (9), teams predicting large volu-metric strains in Fig. 10(b), such as CU and UGLAS-1, alsopredict large increases of degree of saturation in Fig. 10(c).

Because the shape of predictions in the Sr –�v plane (Fig.10(c)) does not depend on the water retention model, theshape of predictions in the Sr –�a plane (Fig. 10(d)) is alsoindependent of the water retention model and is governed by

the mechanical model alone. Therefore, unlike tests TX02and TX07, the variability of shapes of predicted curves inthe Sr –�a plane is largely attributable to the variability ofshapes of predicted curves in the �v –�a plane (Fig. 10(b))(the smaller difference between the shapes of predictedcurves in the Sr –�v plane, which is caused by the differentinitial values of degree of saturation and void ratio, has amuch smaller effect on the variability of shapes of predictedcurves in the Sr –�a plane compared to tests TX02 andTX07). In fact, if the Sr –�a predictions are shifted along thevertical axis to start from the same point, a similar distribu-tion of curves as for the �v –�a predictions is obtained (thetwo distributions would look even more similar if the shapeof the predicted Sr –�v relationships in Fig. 10(c) wereexactly the same for all models which, according to equation(9), would only happen if the initial values of degree ofsaturation, Sr0 and void ratio, e0 coincided in all predic-tions).

During constant water shearing, the occurrence of volu-metric strain and the consequent variation of degree ofsaturation induce a change of suction according to thechosen water retention model. The variation of suctioncalculated by the different models during shearing is pre-sented in Fig. 10(e), together with the corresponding experi-mental data.

Figure 10(e) shows significant discrepancies between pre-dictions and even opposite trends. This contradictory rangeof responses during constant water shearing is explained bythe incidence of two contrasting factors, trying to control thevariation of suction in opposite directions. For all waterretention models, an increase of degree of saturation atconstant volumetric strain induces a drop in suction as@s=@Sr , 0 while, on the other hand, a compressive volu-metric strain at constant degree of saturation produces anincrease of suction as @s=@�v . 0. Whether a drop or anincrease of suction is predicted therefore depends on whichof the above two factors is dominant.

Given that all predicted Sr –�v curves in Fig. 10(c) areapproximately parallel, the corresponding increments of de-gree of saturation and volumetric strain are approximatelythe same for all models at any point during shearing. Theprediction of opposite suction gradients in Fig. 10(e) cannottherefore be attributed to differences in the ratio betweenincrements of degree of saturation and volumetric strainpredicted by the distinct models. On the other hand, itdepends on whether the chosen water retention model im-plies greater sensitivity of suction to changes of degree ofsaturation rather than volumetric strain or vice versa; that isit depends on the relative magnitude of the two partialderivatives @s=@Sr and @s=@�v.

In summary, at any point during shearing, the direction ofsuction variation depends on the water retention model alonewhile the magnitude of such variation depends also on themechanical model, which controls the magnitude of volu-metric strain and, hence, change of degree of saturationduring constant water content shearing. This is consistentwith the fact that the absolute values of the final suctionchanges in Fig. 10(e) are largest for CU and UGLAS-1,which are also the two models predicting the largest in-creases of volumetric strain and degree of saturation overthe entire shearing stage (see Fig. 10(c)).

UGLAS-1, UGLAS-2 and UNINA all use the same waterretention model (see Table 1), although UGLAS-1 andUGLAS-2 adopt different parameter values compared toUNINA (see Table 4), but the predicted suction variationfollows opposite directions, with UGLAS-1 and UGLAS-2predicting an increase of suction but UNINA showing aslight drop. This demonstrates that the sign of the suctiongradient in Fig. 10(e) does not necessarily depend on the

0·86

0·84

0·82

0·80

0·78

0·76

0·74

e

10 100


p: kPa(a)

65

60

55

50

45

40

35

Sr:

%

0·74 0·76 0·78 0·80 0·82 0·84 0·86e

(b)

10 100

p: kPa(c)

65

60

55

50

45

40

35

Sr:

%

Fig. 9. Predicted and experimental behaviour during loadingstage of blind test: (a) void ratio (e) plotted against meannet stress (p); (b) degree of saturation (Sr) plotted against voidratio (e); (c) degree of saturation (Sr) plotted against mean netstress (p)



chosen type of water retention model but can also be aconsequence of the particular calibration of one given model.This example also shows that constant water content testscan prove useful in refining calibration of water retentionmodels owing to the high sensitivity of the predictedresponse to the relevant parameter values.

It is also important to highlight here that any error in theprediction of suction during constant water, or partlydrained, shearing will impact on the prediction of otherimportant variables such as, for example, the strength atcritical state.

Oedometric test EDO-200Predicted and experimental results are compared for the

oedometric test EDO-200 in Fig. 11 (first wetting–dryingcycle stage at �v ¼ 20 kPa), Fig. 12 (loading–unloading–reloading stage at s ¼ 200 kPa) and Fig. 13 (second wetting–drying cycle stage at �v ¼ 800 kPa).

Figure 11 compares predicted and experimental data dur-ing the first wetting–drying cycle for the two hystereticmodels of EPFL and UNSW (Fig. 11(a)) and for the fivenon-hysteretic models of CU, ENPC, UGLAS-1, UGLAS-2and UNINA (Fig. 11(b)). Predicted variations of void ratio


600

500

400

300

200

100

q: k

Pa

0 5 10 15 20 25 30εa: %(a)

0 5 10 15 20 25 30εa: %(b)

10

8

6

4

2

0

ε v: %

0 5 10 15 20 25 30εa: %(d)

0 2 4 6 8 10εv: %(c)

70

65

60

55

50

45

Sr:

%

70

65

60

55

50

45

Sr:

%

0

12

75

12

75

140

120

100

80

60

20

s: k

Pa

0 5 10 15 20 25 30εa: %(e)

0

40

Fig. 10. Predicted and experimental behaviour during shearing stage of blind test: (a) deviator stress (q) plotted against axialstrain (�a); (b) volumetric strain (�v) plotted against axial strain (�a); (c) degree of saturation (Sr) plotted against volumetricstrain (�v); (d) degree of saturation (Sr) plotted against axial strain (�a); (e) suction (s) plotted against axial strain (�a)

124 D’ONZA ET AL.


are not presented here owing to their limited interest (theyare very small and largely reversible).

Inspection of Figs 11(a) and 11(b) confirms that incor-poration of hydraulic hysteresis improves considerably theprediction of degree of saturation during the wetting–dryingcycle with both EPFL and UNSW accurately capturing theirreversible change of degree of saturation at the end of thecycle. As expected, CU, ENPC, UGLAS-1, UGLAS-2 andUNINA generally predict reversible changes of degree ofsaturation. Only the ENPC model shows a slight irreversi-bility in the predicted variation of degree of saturationaround the inversion point because of the occurrence of asmall amount of plastic volumetric strains. Predictions byCU and ENPC match relatively well the wetting branch ofthe cycle while predictions by UGLAS-1, UGLAS-2 andUNINA lie close to the experiments at the start of wettingbut then tend to fall below the measured data as suction isreduced.

Predicted and experimental results during the loading–unloading–reloading cycle are compared in Figs 12(a) and12(b). Fig. 12(a) shows larger differences between predictedyield stresses compared to the anisotropic loading stage atthe same level of suction in test TX07 (see Fig. 7(a)). Thisis caused by the variability of the stress paths computed bythe different models under radially constrained conditions,which leads to yielding over different regions of the stressspace. Plastic deformation after yielding is overestimated byall models (with the only exception of ENPC) owing to

underestimation of the yield stress rather than underestima-tion of plastic stiffness. In the case of ENPC, the predictionof a stiffer post-yield response (as already observed duringthe loading stages of test TX07 and the blind test) compen-sates for the underestimation of the yield stress, resulting ina better match to the experimental data compared to othermodels.

In Fig. 12(b), the degree of saturation predicted at thestart of loading by the five non-hysteretic models (i.e. CU,ENPC, UGLAS-1, UGLAS-2 and UNINA) is considerablysmaller than the experimental value because of the signifi-cant underestimation at the end of the previous drying stage.On the other hand, the overestimation of volumetric com-pression during loading in Fig. 12(a) tends to produce acorresponding overestimation of the increase of degree ofsaturation in Fig. 12(b).

Figure 13 compares predicted and experimental data dur-ing the second wetting–drying cycle both in terms ofdeformation (Fig. 13(a)) and water retention (Figs 13(b) and13(c)). At the start of wetting, the predicted values of voidratio and degree of saturation are in all cases lower than theexperiments due to accumulated errors during the previoustest stages.

During wetting, the amount of collapse depends on thespacing between constant-suction normal compression lines.The CU, ENPC, UGLAS-1, UGLAS-2 and UNINA modelspredict a noticeable amount of collapse in Fig. 13(a). Inthese models, the spacing between constant-suction normal

95

90

85

80

75

70

65

60

55

50

45

40

35

Sr:

%

10 100s: kPa

(a)

Drying

Wetting

UNSWEPFLData

95

90

85

80

75

70

65

60

55

50

45

40

35

Sr:

%

10 100s: kPa

(b)

Drying

Wetting

UNINA

ENPC

Data

CU

UGLAS-1UGLAS-2

Fig. 11. Predicted and experimental behaviour during firstwetting–drying cycle of test EDO-200: (a) variation of degreeof saturation (Sr) plotted against suction (s) for hysteretic waterretention models (EPFL and UNSW) and (b) variation of degreeof saturation (Sr) plotted against suction (s) for non-hystereticwater retention models (CU, ENPC, UGLAS-1, UGLAS-2 andUNINA)

0·85

0·80

0·75

0·70

0·75

0·60

0·55

e

10 100


σv: kPa(a)

65

60

55

50

45

40

Sr:

%

1000

70

10 100σv: kPa

(b)

1000

Fig. 12. Predicted and experimental behaviour during loading–unloading–reloading cycle of test EDO-200: (a) variationof void ratio (e) plotted against vertical net stress (�v) and(b) variation of degree of saturation (Sr) plotted against verticalnet stress (�v)



compression lines had been calibrated by taking into accountdata from all five constant-suction oedometric compressioncurves. However, one difference between the CU, UGLAS-1,UGLAS-2 and UNINA models (which largely overestimatecollapse during wetting) and the ENPC model (which pre-dicts instead a smaller and more accurate value of collapse)is that, in the former case, spacing had been directly fixed tomatch the distance between constant-suction oedometriccompression curves whereas, in the latter case, it had beenindirectly fixed by defining the shape of the yield curve inthe s–p9 plane.

As for the other two models, EPFL predicts a negligibleamount of collapse while, in the case of UNSW, collapse isconcentrated during the later stages of wetting when suctionchanges from about 80 kPa to 55 kPa. This is due to the factthat, as mentioned previously, both these models assume ayield curve in the s–p9 plane that becomes very steep whensuction becomes greater than 50 kPa in the case of EPFLand 80 kPa in the case of UNSW. This implies the existenceof very small distances between constant-suction normalcompression lines in the v–ln p9 plane for suctions greaterthan the above values.

The variation of degree of saturation is presented in Fig.13(b) for the two hysteretic models and in Fig. 13(c) for thefive non-hysteretic models. In this case, unlike the firstwetting–drying cycle, all seven models predict some irrever-sible increase in degree of saturation at the end of the cycle,albeit for different reasons. The relatively accurate predictionof the change of degree of saturation by some of the non-hysteretic models is the consequence of the overestimationof volumetric collapse during wetting. On the other hand,the irreversible changes of degree of saturation predicted bythe two hysteretic models are predominantly caused by waterretention hysteresis, given the small or negligible changes ofvoid ratio predicted by these models during the wetting–drying cycle.

CONCLUSIONSThe paper presents the results from a collaborative piece

of research undertaken by seven universities to benchmarkdifferent approaches to modelling mechanical and waterretention behaviour of unsaturated soils. Seven differentconstitutive models have been independently calibrated bydifferent teams of researchers based on the same set of 13suction-controlled triaxial and oedometer tests performed oncompacted silty soil samples. The calibrated constitutivemodels have then been used to predict soil behaviour duringthree of the 13 calibration tests, as well as during one‘blind’ test whose results had not been previously published.

The main features of the seven models are first comparedwith particular reference to water retention behaviour, stresstensor definition, effect of suction on the mechanical re-sponse and nature of irreversible deformation. Through thiscomparison, a model classification matrix has been proposedbased on the adopted type of constitutive stress and yieldcurve in the s–p9 plane (see Table 2). The intrinsic linkbetween the definition of the yield curve in the s–p9 planeand constant-suction normal compression lines in the v–ln p9plane has also been highlighted. The proposed classificationis not necessarily restricted to the models considered in thiswork and could be generally extended to other formulationsin the literature.

Predictions from different models are interpreted in thecontext of the respective analytical formulations and calibra-tion choices made by participating teams. Models withsimilar features appear to produce qualitatively coherentpredictions although quantitative discrepancies are often ob-served. In several cases, these discrepancies are the conse-quence of the chosen calibration approach rather than thespecific features of the model. For example, UGLAS-1,UGLAS-2 and UNINA all used the same water retentionmodel; nevertheless the suction variation predicted byUGLAS-1 and UGLAS-2 during constant water contentshearing in the blind test follows an opposite directioncompared to UNINA. This is due to the selection of differ-ent parameter values by UGLAS-1 and UGLAS-2 comparedto UNINA, which emphasises the difficulties associated withmodel calibration and, especially, the importance of identify-ing the most sensitive aspects of soil behaviour for the

95

90

85

80

75

70

65

60

55

50

45

40

35

Sr:

%

10 100s: kPa

(b)

Drying

Wetting

UNSWEPFLData

95

90

85

80

75

70

65

60

55

50

45

40

35

Sr:

%

10 100s: kPa

(c)

Drying

Wetting

UNINA

ENPC

Data

CU

UGLAS-1UGLAS-2

0·75

0·70

0·65

0·60

0·55

0·50

0·45

e

10 100


s: kPa(a)

Wetting

Drying

Fig. 13. Predicted and experimental behaviour during secondwetting–drying cycle of test EDO-200: (a) variation of voidratio (e) plotted against suction (s); (b) variation of degree ofsaturation (Sr) plotted against suction (s) for hysteretic waterretention models (EPFL and UNSW); (c) variation of degree ofsaturation (Sr) plotted against suction (s) for non-hystereticwater retention models (CU, ENPC, UGLAS-1, UGLAS-2 andUNINA)

126 D’ONZA ET AL.


selection of particular model parameters. In general, theseresults highlight the danger of formulating ever more sophis-ticated constitutive models without dedicating the necessaryattention to the development of robust procedures for select-ing parameter values.

Based on the analysis of all results, it is concluded thathydraulic uncoupled predictions (i.e. predictions governedonly by the water retention model) show smaller differencesamong teams and are generally closer to the experimentaldata than mechanical uncoupled predictions (i.e. predictionsgoverned only or predominantly by the mechanical model).This also confirms that a greater degree of uniformity existsacross water retention models in comparison to mechanicalmodels.

The variation of volumetric strain during shearing appearsparticularly difficult to predict and this is intrinsically relatedto the ability of each model of matching the distancebetween constant-suction normal compression and criticalstate lines in the v–ln p9 plane. Potential inaccuracies in theprediction of volumetric strain during shearing have alsoconsequences on the corresponding prediction of degree ofsaturation because of the assumed dependency of degree ofsaturation on soil porosity. Errors in the calculated relation-ship between degree of saturation and axial strain duringshearing are therefore more likely to be the consequence ofan inaccurate prediction of volumetric strain by the mechani-cal model rather than a deficiency of the water retentionmodel itself.

During shearing at constant water content, the relation-ships between suction and axial strain predicted by thedifferent teams show significant discrepancies and evenopposite trends. This is caused by the strong sensitivity ofthis type of prediction to the chosen water retention modeland its calibration. The direction of suction variation de-pends on the chosen water retention model alone and, morespecifically, on the relative sensitivity of suction variation tochanges in degree of saturation and volumetric strain. Themagnitude of suction variation depends instead on the sizeof the changes of degree of saturation and volumetric strain,which is governed by both mechanical and water retentionmodels.

Any error in the prediction of suction during constantwater content (or partly drained) shearing will impact on theprediction of strength at critical state. Therefore the choiceof an accurate and well-calibrated water retention modelbecomes particularly important during analyses that involvestrong hydromechanical coupling.

ACKNOWLEDGEMENTSThe financial support of the European Commission

through funding of the ‘Marie Curie’ research training net-work MUSE (‘Mechanics of Unsaturated Soils for Engineer-ing’ – contract: MRTN-CT-2004-506861) is acknowledged.Contributions by Dr Claudia Zingariello of the UNINA teamand Drs Michael Habte and Adrian Russell of the UNSWteam are also gratefully acknowledged.

APPENDIXA linear relationship in the v–ln p9 plane is assumed for the

saturated normal compression line corresponding to s ¼ se

v ¼ N � º ln p90(se) (10)

where º is the slope and N is the intercept at p90(se) ¼ 1.By following an elastic path inside the yield locus in the s–p9

plane, the specific volume at a yield stress p90(s) is obtained as

v ¼ N � º ln p90(se)þ ˜vejsse� k ln

p90(s)

p90(se)(11)

where ˜vejs0 is the elastic change of specific volume when suctionvaries from se to s while k is the slope of the elastic compressionline in the v–ln p9 plane.

By substituting for p90(se) from equation (7) into equation (11) andrearranging, one obtains the following expression for the constant-suction normal compression line at suction s

v ¼ N þ º� kø2

ln ø1 þ ˜vejsse

� º� kø2

ln ( p90(s)� ø3)� k ln p90(s)

(12)

which can be rewritten in the following simpler form

v ¼ N (s)� º(s) ln p90(s) (13)

The slope º(s) and spacing N(s) of constant-suction normalcompression lines are therefore given by

º(s) ¼ º� kø2

ln p90(s)� ø3ð Þln p90(s)

þ k (14)

N (s) ¼ N þ º� kø2

ln ø1 þ ˜vejsse(15)

For each of the three classes of yield curve, the slope º(s) andspacing N(s) of constant-suction normal compression lines aretherefore given as follows.

Class (a)

ø2 ¼ 1

ø3 ¼ 0

) º(s) ¼ º

N (s) ¼ N þ (ºþ k) ln ø1 þ ˜vejsse

(16)

Constant-suction normal compression lines have constant slope,while spacing depends on function ø1.

Class (b)

ø3 ¼ 0)º(s) ¼ º� k

ø2

þ k

N (s) ¼ N þ º� kø2

ln ø1 þ ˜vejsse

(17)

Constant-suction normal compression lines have variable slopedepending on function ø2 and spacing depending on function ø1.

Class (c)

ø2 ¼ 1)º(s) ¼ (º� k)

ln ( p90(s)� ø3)

ln p90(s)þ k

N (s) ¼ N þ (º� k) ln ø1 þ ˜vejsse

(18)

Constant-suction normal compression lines are not straight lines inthe v–ln p9 plane (i.e. the secant slope depends on stress state) andspacing depends on function ø1.

REFERENCESAlonso, E. E., Gens, A. & Josa, A. (1990). A constitutive model

for partially saturated soils. Geotechnique 40, No. 3, 405–430,doi: 10.1680/geot.1990.40.3.405.

Brooks, R. N. & Corey, A. T. (1964). Hydraulic properties ofporous media. Colorado State Univ. Hydrol. Pap. 3, 1–27.

Cambou, B. & Jafari, K. (1988). Modele de comportement des solsnon coherents. Revue Francaise de Geotechnique 44, 43–55.

Casini, F. (2008). Effetti del grado di saturazione sul comportamen-to meccanico di un limo. PhD thesis. Universita degli Studi diRoma ‘La Sapienza’, Rome.

Coussy, O. & Dangla, P. (2002). Approche energetique du compor-tement des sols non satures. In Mecanique des sols non satures(eds O. Coussy and J.-M. Fleureau). Paris, France: Lavoisier.




Dangla, P. (2002). Plasticite et hysteresis. In Mecanique des solsnon satures (eds O. Coussy and J.-M. Fleureau). Paris, France:Lavoisier.

D’Onza, F., Gallipoli, D. & Wheeler, S. (2010). Effect of anisotropyon the prediction of unsaturated soil response under triaxial andoedometric conditions. Proc. 5th Int. Conf. Unsaturated Soils,Barcelona (eds E. E. Alonso and A. Gens), pp. 787–794.Rotterdam: Balkema.

Gallipoli, D., Wheeler, S. & Karstunen, M. (2003). Modelling thevariation of degree of saturation in a deformable unsaturatedsoil. Geotechnique 53, No. 1, 105–112, doi: 10.1680/geot.2003.53.1.105.

Gallipoli, D., D’Onza, F. & Wheeler, S. (2010). A sequentialmethod for selecting parameter values in the Barcelona basicmodel. Can. Geotech. J. 47, No. 11, 1175–1186.

Gens, A. (1996). Constitutive modelling: application to compactedsoils. Proc. 1st Int. Conf. Unsaturated Soils, Paris (eds E. E.Alonso and P. Delage), 3, 1179–1200. Rotterdam: Balkema.

Gudehus, G. & Masın, D. (2009). Graphical representation ofconstitutive equations. Geotechnique 59, No. 2, 147–151, doi:10.1680/geot.2009.59.2.147.

Hujeux, J. (1985). Une loi de comportement pour le chargementcyclique des sols. Genie Parasismique, pp. 287–353. Paris: Leseditions de l’ENPC.

Khalili, N., Habte, M. A. & Valliappan, S. (2005). A boundingsurface plasticity model for cyclic loading of granular soils. Int.J. Numer. Methods Engng 63, No. 14, 1939–1960.

Khalili, N., Habte, M. A. & Zargarbashi, S. (2008). A fully coupledflow deformation model for cyclic analysis of unsaturated soilsincluding hydraulic and mechanical hysteresis. Comput. Geo-technics 35, No. 6, 872–889.

Khalili, N. & Khabbaz, M.H. (1998). A unique relationship for �for the determination of the shear strength of unsaturated soils.Geotechnique 48, No. 5, 681–687, doi: 10.1680/geot.1998.48.5.681.

Masın, D. (2005). A hypoplastic constitutive model for clays. Int. J.Numer. Analyt. Methods Geomech. 29, No. 4, 311–336.

Masın, D. (2010). Predicting the dependency of a degree of satura-tion on void ratio and suction using effective stress principle forunsaturated soils. Int. J. Numer. Analyt. Methods Geomech. 34,No. 1, 73–90.

Masın, D. & Herle, I. (2005). State boundary surface of a hypoplas-tic model for clays. Comput. Geotechnics 32, No. 6, 400–410.

Masın, D. & Khalili, N. (2008). A hypoplastic model for mechani-cal response of unsaturated soils. Int. J. Numer. Analyt. MethodsGeomech. 32, No. 15, 1903–1926.

Nuth, M. & Laloui, L. (2007). New insight into the unified hydro-mechanical constitutive modelling of unsaturated soils. Proc. 3rdAsian Conf. Unsaturated Soils, Nanjing, China (eds Z. Z. Yin, Y.P. Yuan, and A. C. F. Chiu), pp. 109–126. Beijing: Science Press.

Nuth, M. & Laloui, L. (2008). Advances in modelling hystereticwater retention curve in deformable soils. Comput. Geotechnics35, No. 6, 835–844.

Pereira, J. M., Wong, H., Dubujet, P. & Dangla, P. (2005). Adapta-tion of existing behaviour models to unsaturated states: applica-tion to CJS model. Int. J. Numer. Analyt. Methods Geomech. 29,No. 11, 1127–1155.

Roscoe, K. H. & Burland, J. B. (1968). On the generalised stress–strain behaviour of wet clay. In Engineering plasticity (eds J.Heyman and F. A. Leckie), pp. 535–609.Cambridge: CambridgeUniversity Press.

Van Genuchten, M. T. (1980). A closed-form equation for predict-ing the hydraulic conductivity of unsaturated soil. Soil Sci. Soc.Am. J. 44, No. 5, 892–898.

Wheeler, S. & Sivakumar, V. (1995). An elasto-plastic critical stateframework for unsaturated soil. Geotechnique 45, No. 1, 35–53,doi: 10.1680/geot.1995.45.1.35.

Wheeler, S., Naatanen, A., Karstunen, M. & Lojander, M. (2003).An anisotropic elastoplastic model for soft clays. Can. Geotech.J. 40, No. 2, 403–418.

128 D’ONZA ET AL.


Tarantino, A. et al. (2011). Geotechnique 61, No. 4, 303–312 [doi: 10.1680/geot.2011.61.4.303]

129

Benchmark of experimental techniques for measuring and controllingsuction

A. TARANTINO1, D. GALLIPOLI 2, C . E. AUGARDE 3, V. DE GENNARO4, R. GOMEZ 5, L. LALOUI 6,C. MANCUSO7, G. EL MOUNTASSIR1, J. J. MUNOZ 8, J. -M. PEREIRA 9, H. PERON 6, G. PISONI 1 0,

E . ROMERO1 1, A. RAVEENDIRARAJ 1 2, J. C. ROJAS 2, D. G. TOLL 3, S . TOMBOLATO1 0

and S. WHEELER2

The paper presents a benchmarking study carried outwithin the ‘Mechanics of Unsaturated Soils for Engineer-ing’ (MUSE) network aimed at comparing different tech-niques for measurement and control of suction. Techniquestested by the eight ‘Mechanics of Unsaturated Soils forEngineering’ research teams include axis-translation(pressure plate and suction-controlled oedometer), high-capacity tensiometer and osmotic technique. The soil usedin the exercise was a mixture of uniform sand, sodiumbentonite (active clay) and kaolinite (non-active clay),which were all commercially available. Samples were pre-pared by one team and distributed to all other teams. Theywere normally consolidated from slurry under one-dimen-sional conditions (consolidometer) to a given vertical stress.The water retention characteristics of the initially satu-rated specimens were investigated along the main dryingpath. Specimens were de-saturated by applying suctionthrough the liquid phase when using an axis-translationtechnique or osmotic method and de-saturated by air-dry-ing, when suction was measured using high-capacity tensi-ometers. In general, the same technique was tested by atleast two teams. The water retention curves obtained usingthe different techniques are compared and discrepanciesare discussed in the paper.

KEYWORDS: laboratory tests; partial saturation; suction

Cette communication presente une etude comparativerealisee dans le cadre du reseau MUSE, dans le but decomparer differentes techniques de mesure et de regula-tion de l’aspiration. Parmi les techniques testees par leshuit equipes de chercheurs de MUSE, indiquons la trans-lation d’axes (plaque de pression et oedometre a succioncontrolee), un tensiometre a capacite elevee, et une tech-nique osmotique. Le sol utilise pour cette tache etait unmelange de sable uniforme, de bentonite de sodium(argile active), et de kaolinite (argile non active), tousdisponibles dans le commerce. Une equipe preparait lesechantillons, qu’elle distribuait ensuite a toutes les autresequipes. Ces echantillons etaient generalement consolidesa partir de boues dans des conditions unidimensionnelles(consolidometre), jusqu’a une contrainte verticale donnee.On recherchait ensuite les caracteristiques de retenue del’eau de specimens initialement satures, le long du che-min de sechage principal, et on procedait a la desatura-tion de specimens par l’application d’une aspiration parsechage a l’air, succion etant mesuree avec des tensio-metres a capacite elevee. En general, la meme techniqueetait testee par un minimum de deux equipes. On proce-da ensuite a une comparaison des courbes de retenued’eau obtenue avec les differentes techniques : les diver-gences sont discutees dans la communication.

INTRODUCTIONSuction plays a key role in the mechanical and hydraulicbehaviour of unsaturated soils and its measurement is there-fore an essential requirement for predictive purposes. Differ-

ent techniques are available for suction measurement andcontrol, which are based on equilibrium through either theliquid or vapour phase. Consistency among different techni-ques and reproducibility of suction measurements amongdifferent laboratories are crucial in the implementation ofunsaturated soil mechanics into routine engineering practice.The paper tackles this problem by comparing four techni-ques for suction measurement and control, namely thepressure plate, the axis-translation oedometer, the high-capa-city tensiometer and the osmotic method, which fall into thecategory of liquid equilibrium. Eight different laboratoriesacross Europe were involved in this ‘round robin’ benchmarkstudy with the aim of estimating confidence level of suctionmeasurement. To cross-check experimental results, testsusing the pressure plate, the axis-translation oedometer andthe high-capacity tensiometer were replicated by two differ-ent laboratories. Only the osmotic method was tested by asingle laboratory.

The eight laboratories participating in this benchmarkstudy were participants in the ‘Mechanics of UnsaturatedSoils for Engineering’ (MUSE) project. This is a majorresearch and training network funded by the European Unioninvolving six European universities as full partners, threeEuropean universities as associated partners and five indus-trial partners. Details about the project can be obtained byvisiting the MUSE website (2005). The academic institutionsinvolved in this benchmarking study were: University of

Manuscript received 1 March 2010; revised manuscript accepted29 November 2010.Discussion on this paper closes on 1 September 2011, for furtherdetails see p. ii.1 Department of Civil Engineering, University of Strathclyde, UK2 Department of Civil Engineering, University of Glasgow, UK3 School of Engineering and Computing Sciences, DurhamUniversity, UK4 Schlumberger, Pau, France5 Departament d’Enginyeria del Terreny, Cartografica i Geofısica,Universitat Politecnica de Catalunya, Barcelona, Spain6 Ecole Polytechnique Federale de Lausanne, EPFL, Switzerland7 Dipartimento di Ingegneria Idraulica, Geotecnica ed Ambientale,Universita di Napoli Federico II, Napoli, Italy8 Universidad Nacional de San Juan Instituto de InvestigacionesAntisısmicas ‘Ing. Aldo Bruschi’, San Juan, Argentina9 Universite Paris-Est, Laboratoire Navier – CERMES, Ecole desPonts ParisTech, Marne-la-Vallee, France10 Dipartimento di Ingegneria Meccanica e Strutturale, Universitadegli Studi di Trento, Italy11 Geotechnical Laboratory, Departament d’Enginyeria del Terreny,Cartografica i Geofısica, Universitat Politecnica de Catalunya,Barcelona, Spain12 Atkins, London, UK


Glasgow in the UK (MUSE coordinator), Durham Universityin the UK, Universita di Trento in Italy, Ecole Nationale desPonts et Chaussees in France, Universitat Politecnica deCatalunya in Spain, Universita di Napoli ‘Federico II’ inItaly, Ecole Polytechnique Federale de Lausanne in Swizter-land (associated partner) and University of Strathclyde in theUK (associated partner). The acronyms used to identify thedifferent academic institutions are given later in Table 2.

The soil used in this experimental programme was pre-pared by mixing sand, kaolinite and bentonite. To ensurethat ‘identical’ specimens were tested, samples were preparedat UNITN and shipped to the MUSE teams, with theexception of one team that only produced samples in itsown laboratory. Each laboratory was then requested todetermine the water retention curve (WRC) starting from thesaturated condition (main drying WRC).

MATERIAL AND SAMPLE PREPARATIONThe soil used in the exercise was obtained by mixing

three different geomaterials, a uniform sand (Hostun sand),an active clay (MX-80 sodium bentonite) and a non-activeclay (Speswhite kaolin), which are all commercially avail-able (Table 1). The mixture made it possible to prepare asoil with a ‘suitable’ WRC by modifying the mass fractionof each component of the mixture. The mixture was requiredto generate a WRC having an air-entry suction not exceed-ing 100–200 kPa, as the suction range of several axis-trans-lation apparatuses used in the benchmarking exercise waslimited to 500–600 kPa. It was also desirable that the slopeof the WRC beyond the air-entry suction was not steep, assmall errors in the values of suction controlled or measuredwould have produced very scattered data in the WRC.Finally, the mixture was expected to generate a WRC devel-oping over a large range of suction so that the same mixturecould be used for matric suction measurement/control (equi-librium by way of liquid transfer) and total suction measure-ment/control (equilibrium by way of vapour transfer), even ifthis benchmark study essentially focuses on matric suction.Preliminary tests were carried out at UNITN by testingdifferent mixtures and it was found that the followingcomposition produced the mixture with the aforementionedfeatures: 70% Hostun sand, 20% MX80 bentonite and 10%Speswhite kaolin.

The majority of samples were prepared at UNITN andshipped to seven of the eight teams involved in this study.One team only produced the samples in its own laboratory(UPC), whereas UNINA tested samples produced both byUNITN and in its own laboratory. Samples used by thedifferent partners are summarised in Table 2. In the follow-ing, the term ‘sample’ will be used only for the soil preparedwithin the oedometer/consolidometer and the term ‘speci-men’ will refer to the material used to determine the WRC.

The procedure for sample preparation at UNITN is brieflyillustrated below. The detailed procedure is described in aMUSE document (Tarantino, 2007) and can be made avail-able on request. The three soils were first dry mixed using aspatula. A small amount of dry mixture was placed in aplastic bowl (60 g) and about 50 g of demineralised waterwas added. The soil and the water were mixed togetherusing the spatula to squeeze lumps. This procedure wasrepeated six times until all the dry soil was mixed withwater. At this stage, the soil–water mixture was quite denseand additional water was added to reach a water content of128%. The slurry was poured into a one-dimensional con-solidometer about 110 mm in diameter and 150 mm high.Vertical pressure was increased at the rate of 3.3 kPa/h and,when the final vertical stress of 101 kPa was reached (after31 h), the vertical pressure was maintained constant for 41 h.At the end of consolidation the sample was unloaded as fastas possible to limit swelling and a water content of 0.53 wasrecorded. The sample was finally put in a plastic bag, in turnput in an airtight plastic container and stored in a high-humidity room until shipping.

UPC consolidated the sample from slurry directly in theaxis-translation oedometer subsequently used to determinethe WRC. The slurry was consolidated by applying a rampfrom 0 to 100 kPa over 1.5 days and by keeping the 100 kPavertical pressure constant for the subsequent 4 days. UNINA,in addition to the sample prepared by the UNITN, tested asample prepared in its own laboratory using the sameprocedure as adopted by UNITN.

APPARATUS AND EXPERIMENTAL PROCEDURESTeams using the same type of equipment (pressure plate,

axis-translation oedometer and high-capacity tensiometer)were invited to use their own experimental procedure. There

Table 1. Index properties of soils used to prepare the mixture

Clay: % Silt: % Sand: % wp wl Gs d50: mm

Hostun sand (De Gennaro et al., 2004) — — 100 — — 2.65 0.38MX-80 bentonite (Tang & Cui, 2005) 60 40 — 35 519 2.76 —Speswhite kaolin (Tarantino &Tombolato, 2005)

80 20 — 32 64 2.61 —

Table 2. Samples used by different teams

Participant Acronym Sample prepared byUniversita di Trento

Sample prepared inits own laboratory

University of Glasgow, UK UGLAS • —Durham University, UK UDUR • —Universita di Trento, Italy UNITN • —Ecole des Ponts ParisTech, France ENPC • —Ecole Polytechnique Federale de Lausanne, Switzerland EPFL • —University of Strathclyde, UK USTRAT • —Universita di Napoli ‘Federico II’, Italy UNINA • •Universitat Politecnica de Catalunya, Spain UPC — •

130 TARANTINO ET AL.


is high variability between procedures adopted across geo-technical laboratories and this variability was therefore partof the benchmark study. All tests were carried out in tem-perature-controlled rooms.

Pressure plate (EPFL, UGLAS, USTRAT)Pressure plate tests were performed by EPFL and UGLAS

to determine the main drying WRC. USTRAT only applied asingle level of suction to three ‘identical’ specimens toinvestigate the influence of contact pressure on the waterretained at a given suction.

Commercial equipment was used by all teams, as sum-marised in Table 3. The porous ceramic plate was saturatedby filling the cell with de-aired water and then applying apositive pressure to force water to flow through the ceramic,while maintaining the pressure underneath the ceramic plateat atmospheric pressure. This procedure was repeated severaltimes until the permeability of the ceramic attained aconstant value. The applied pressures and the numbers ofsaturation cycles adopted by different teams are reported inTable 3.

Sizes of specimens, cut and trimmed from the samplesprovided by UNITN, are given in Table 3. EPFL placedeight specimens on the plate and no vertical pressure wasapplied. UGLAS and USTRAT placed three specimens onthe ceramic plate, and a small vertical pressure was appliedto improve contact (0.6 kPa for UGLAS and 0.3, 4.7 and8.2 kPa respectively for USTRAT). In all pressure plate tests,a tray filled with water remained within the pressure cham-ber to increase the relative humidity and minimise moisturecontent losses due to evaporation.

Each level of suction was maintained for a sufficient timeto allow for moisture equilibration, which was assessedusing different approaches. EPFL regularly weighed speci-mens (twice a week for low suction steps, once a week forhigh suction steps) and equilibrium was considered to bereached once the rate of water content decrease was˜w , 0.2%/day. UGLAS monitored the burette connected tothe water reservoir underneath the ceramic plate. An accep-table rate of burette water volume change was determinedby weighing specimens on successive days during the firstthree suction steps and equilibrium was assumed to bereached when the rate of water content decrease was˜w , 0.04%/day. It should be noted that specimens tested at

EPFL and UGLAS have different heights, 12 and 20 mmrespectively, and this results in different flow rates at a given‘degree of suction equalisation’. For example, Terzaghi’stheory of consolidation for saturated elastic geomaterials(Atkinson, 1993) predicts that the flow rate per unit solidmass at the same degree of consolidation is inverselyproportional to the square of specimen height. When speci-men height is increased from 12 to 20 mm, the flow ratedecreases by 2.8 and this figure is not significantly differentfrom the ratio between rates of water content decreaseadopted by EPFL and UGLAS.

USTRAT did not monitor water content changes as theonly purpose of this test was to investigate the effect ofcontact pressure. Suction was simultaneously applied to thethree specimens uninterruptedly for 12 days.

At equilibrium at a given suction level, EPFL removed asingle specimen from the pressure plate apparatus (with theexception of three specimens for the last suction steps) todetermine water content and also air and total volume byimmersion in Kerdane (Peron et al., 2007). EPFL thereforeadopted the approach of one-specimen–one-point. UGLASused a single specimen to determine the entire WRC andwater content at each suction level was back-calculated fromthe final water content. Measurement of the WRC atUGLAS was made in triplicate.

Axis-translation oedometer (UNINA and UPC)The equipment used at UNINA is a Wissa oedometer

(Wissa & Heiberg, 1969) modified to control matric suction(Rampino et al., 1999) and to measure water contentchanges (Rojas et al., 2007) whereas UPC used an oed-ometer designed and constructed in their own geotechnicallaboratory (Hoffmann et al., 2005) (Table 4).

UNINA applied suction using the water-subpressure tech-nique (water pressure is maintained constant and air pressureis progressively increased to increase suction), whereas UPCapplied suction using the air-overpressure technique (air andwater pressure are initially increased simultaneously, then airpressure is maintained constant and water pressure is de-creased to increase suction) (Romero, 2001). To minimiseair diffusion underneath the high-air-entry ceramic, the waterpressure was raised to values greater than atmospheric(50 kPa for UNINA and in the range 100–490 kPa for UPC)according to Romero (2001).

Table 3. Summary of procedures adopted in pressure plate testing

EPFL (Ecole PolytechniqueFederale de Lausanne)

UGLAS (University of Glasgow) USTRAT (University of Strathclyde)

Apparatus 1500F1 (Soil Moisture Corporation) 1500F1 (Soil Moisture Corporation) 532-132 (ELE)AEV of ceramic plate 1.5 MPa (s . 0.3 MPa)

0.5 MPa (s < 0.3 MPa)1.5 MPa 1.5 MPa

Saturation pressure 0.8 MPa (AEV ¼ 1.5 MPa)0.4 MPa (AEV ¼ 0.5 MPa)

0.8 MPa (AEV ¼1.5 MPa) 1.6 MPa (AEV ¼ 1.5 MPa)

Saturation cycles 6 3 1Ceramic plate measuredpermeability

6.5 3 10�11 m/s (AEV ¼ 1.5 MPa)2.2 3 10�10 m/s (AEV ¼ 0.5 MPa)

N/A 2.5 3 10�11 m/s (AEV ¼ 1.5 MPa)

Specimen size � ¼ 45 mm, h ¼ 12 mm � ¼ 75 mm, h ¼ 20 mm � ¼ 38 mm, h ¼ 20 mmInitial suction applied 10 kPa 20 kPa N/AWater content measurement One specimen per measurement Multiple measurement on same

specimenOne specimen per measurement

Total volume measurement Immersion in Kerdane None NoneEquilibrium (water loss) ˜w , 0.2%/day (by weight) ˜w , 0.04 %/day (burette

connected to drainage line)N/A

Balance resolution 0.01 g (to monitor equalisation)0.001 g (final water content)

0.01 g 0.001 g

Vertical pressure 0 kPa 0.6 kPa 0.3, 4.7 and 8.2 kPa

EXPERIMENTAL TECHNIQUES FOR MEASURING AND CONTROLLING SUCTION 131


The high-air-entry ceramic was saturated using two differ-ent approaches. At UNINA, water was forced to flowthrough the ceramic by increasing air pressure in the cham-ber while continuously flushing the water reservoir under-neath for 1 h. The water reservoir was then closed foranother 1 h and this cycle was repeated until the measuredpermeability attained a constant value (Table 4). At UPC,de-aired water was forced to flow through the ceramic discusing a GDS Instruments pressure/volume controller (withno air–water interface) until permeability attained a value inthe range between 1.0 and 7.0 3 10�11 m/s; that is the rangeof permeability of 1.5 MPa air-entry value (AEV) ceramicsmeasured over 15 years at UPC. In addition, diffusion of airthrough the saturated ceramic disc was measured and it waschecked that the coefficient of diffusion for air was lowerthan 5 3 10�10 m2/s (Romero, 1999; Airo Farulla & Ferrari,2005; Delage et al., 2008).

The pore-water volume change was measured by UNINAusing a system of two double-walled burettes (Rojas et al.,2007), whereas a GDS pressure–volume controller was usedby UPC to monitor the pore-water volume change. Thesaturation of the water drainage line was ensured by periodi-cally flushing the drainage line and the reservoir underneaththe high-air-entry ceramic disc (every 1 h at UNINA and at3-day intervals or when relatively high diffused air volu-metric rates were measured at UPC).

The specimen in the oedometer was initially loaded to100 kPa net axial stress to avoid lateral shrinkage duringdrying. The procedure of air pressurisation at high degreesof saturation is discussed by Di Mariano et al. (2000),Romero (2001) and Delage et al. (2008).

Equilibrium was considered to be attained when nochanges in volume were recorded and water content changedlinearly with time. This is due to either air diffusion towardsthe water reservoir underneath the ceramic plate or waterevaporation into the air pressure line (Airo Farulla & Ferrari,2005). In tests carried out at UNINA, the differential pres-sure transducer recorded a negative water-volume rate, sug-gesting that evaporation was the dominant mechanism, and acorrection was made to account for evaporation effectsaccording to Airo Farulla & Ferrari (2005). In tests carriedout at UPC, evaporation effects were negligible and nocorrection was made.

Osmotic method – ENPCThe WRC at ENPC was determined using the osmotic

method (Cui & Delage, 1996; Delage et al., 1998). Apartially saturated specimen was sealed in a tube-shapedcellulosic semi-permeable membrane having molecular

weight cut-off (MWCO) of 3500 (regenerated cellulose (RC)dialysis membrane, Spectra/Por 3, MWCO 3500) and im-mersed in an aqueous solution of PEG with molecularweight of 20 000. A magnetic stirrer was used in order toimprove the kinetic exchange of water and ensure the homo-geneity of the PEG solution.

The value of PEG concentration was obtained by measur-ing the refractive Brix index (Br) of the PEG solution

cPEG ¼Br

90� Br(1)

where cPEG is the PEG concentration in kg PEG/kg water.To relate the PEG concentration to suctions, two relation-ships can be adopted. The first was proposed by Delage etal. (1998) based on data from Williams & Shaykewich(1969)

s ¼ 11 c2PEG (2)

where s is the suction in MPa. The second relationship isbased on calibration data by Dineen & Burland (1995) andwas recently confirmed by Tang et al. (2010) using the samePEG and semi-permeable membrane used to desaturate theMUSE soil. The two calibration curves are shown in Fig. 1.

A total of eight cylindrical specimens 19.5 mm in dia-meter and 26 � 1 mm high were cut from a sample preparedat UNITN. Each specimen was immersed into a PEG solu-tion for 4 days, removed from the tubular membrane todetermine its mass and volume, and re-immersed in the PEG

Table 4. Summary of procedures adopted in axis-translation oedometer testing

UNINA (Universita di Napoli) UPC (Universitat Politecnicade Catalunya)

Apparatus Rampino et al. (1999),Rojas et al. (2007)

Hoffmann et al. (2005)

AEV of ceramic plate 0.5 MPa (Soil MoistureCorporation)

1.5 MPa (Soil MoistureCorporation)

Saturation pressure 0.65 MPa 2 MPa (AEV ¼1.5 MPa)Saturation cycles 6 1Specimen size � ¼ 79 mm, h ¼ 25 mm

(UNINA)� ¼ 79 mm, h ¼ 15 mm(UNITN)

� ¼ 50 mm, h ¼ 12 mm

Ceramic plate measuredpermeability

2.6 3 10�10 m/s 1 to 7 3 10�11 m/s

0 0·1 0·2 0·3 0·4Concentration: kg PEG/kg water

0

0·4

0·8

1·2

1·6

Suc

tion:

MP

a

Tang (2010)Dineen & Burland (1995)Calibration after Williams & Shaykewich (1969)Calibration after Dineen & Burland (1995)

et al.

Fig. 1. Calibration curve relating PEG concentration to theosmotically generated suction



solution using a new membrane. Once the weight of speci-men attained a stable value, typically after 9 to 12 days,final water content was determined by oven-drying and thetotal volume was determined by immersion in Kerdane.

High-capacity tensiometer (UNITN and UDUR))Trento high-capacity tensiometers (Tarantino & Mongiovı,

2002; 2003) were used to measure matric suction at UNITN,whereas a tensiometer developed by Durham University andWykeham Farrance Ltd (Lourenco et al., 2006, 2008, 2011)were used for measurement at UDUR.

The tensiometers were first calibrated using the proceduresoutlined by Tarantino & Mongiovı (2003) and Lourenco etal. (2008) for the UNITN and UDUR tensiometer respec-tively. The saturation of the UNITN tensiometer porousceramic was checked according to the procedure illustratedby Tarantino (2004).

The experimental procedure adopted at UNITN involvedair-drying samples to a given water content and storage forat least 1 week to ensure moisture equilibration. Specimens80 mm in diameter and 20 mm high were then cut from theair-dried samples and suction measurements were carried outin a suction measurement box (Tarantino & Mongiovı,2002). To improve contact with the specimen a paste madeof Speswhite kaolin was applied on the porous stone of thetensiometer. Water content and degree of saturation weredetermined for each specimen at the end of the suctionmeasurement.

At UDUR, a single specimen was cut and placed in anair-tight box (Lourenco et al., 2011). The tensiometer wasplaced in a hole drilled in the top plate of the box over thecentre of the specimen and was held in place by the ringsealing the hole. When measurement stabilised, the tensi-ometer was removed, the specimen was set on its side toallow air drying and the measurement was repeated. Thewater content at each drying stage was back-calculated fromthe final water content. As shrinkage was not uniform, thatis the specimen ceased to be cylindrical, volume measure-ments required for the degree of saturation were unreliableand water retention data were processed only in terms ofwater content.

EVALUATION OF TECHNIQUES FOR SUCTIONCONTROL AND MONITORING

To better discuss the experimental results, it may beconvenient to examine the path followed by the specimensduring one-dimensional loading and unloading and subse-quent drying. Upon loading in the consolidometer, the speci-men moves along the normal consolidation line. As thevertical stress is increased to its final value � 9vc, the meaneffective stress increases to the value p9c (point A in Fig. 2)given by

p9c ¼1þ 2k0

3� 9vc (3)

where k0 is the coefficient of earth pressure at rest. At thisstage the specimen is saturated and the void ratio e equalsthe water ratio ew (ew being the volume of water per volumeof solids). Even though the total vertical stress is reduced tozero very rapidly, some water enters the specimen and themean effective stress reduces to p90 (point B in Fig. 2). Afterremoval from the consolidometer, the saturated specimentherefore has suction s0 equal to p90.

When using the pressure plate and the osmotic method, aninitial small suction si was applied to the specimen, whichbrought the specimen to point C in Fig. 2(a) along the‘saturated’ unloading–reloading line. Subsequent drying at

zero net stress ( p ¼ 0) first brought the saturated specimenagain to point A along the ‘saturated’ unloading–reloadingline (path C–B–A) and then along the ‘saturated’ normalconsolidation line. At a given suction, s0

AEV, the soil desatu-rated (point D in Fig. 2(a)) and the water ratio ew decreasedmore rapidly than void ratio e. In Fig. 2, it is assumed thatthe AEV s0

AEV is higher than the preconsolidation pressurep9c, which is the case of the MUSE soil. Accordingly, allspecimens are expected to desaturate at the same suction,s0

AEV, irrespective of the initial suction applied si.When measuring suction using the tensiometer, the speci-

mens were directly desaturated from point B (the specimensare not resaturated) and therefore followed the path B–A–D.Again, since the AEV s0

AEV is higher than the preconsolida-tion pressure p9c for the MUSE soil, the soil tested using thetensiometer would be expected to desaturate at the samesuction as the pressure plate and osmotic method (point D).

The case of the axis-translation oedometer is examined inFig. 2(b). After removal from the consolidometer, net stresswas increased at zero suction to match the preconsolidationpressure (pOED ¼ p9c, point A in Fig. 2(b)). Starting from thispoint, suction was progressively increased from zero to theAEV s0

AEV recorded under zero net stress (point D9 in Fig.2(b)). At this suction level, the soil was still saturated butthe void ratio was lower than the void ratio recorded at thesame suction under zero net stress (compare points D andD9 in Fig. 2(b)). As a lower void ratio increases the air-entrysuction (Gallipoli et al., 2003; Tarantino, 2009), the soil istherefore expected to desaturate at a higher suction, s9AEV

(point D99 in Fig. 2(b)).Bearing in mind these paths, the results from the pressure

plate tests (Fig. 3) are now examined. The WRC at UGLAS

A

s p0 0≡ �

s p0 ≡ �

p�c s0AEV

e

p pOED c� � p sOED0AEV�

p sOED AEV� �

si

BC

p s�

e e e≡ w

e e≡ w

eD

AB

e

ew

D

(a)

D�

D�

p s�

(b)

ew

Fig. 2. Hydraulic and mechanical path associated with consoli-dation and drying: (a) pressure plate and osmotic method; (b)axis-translation oedometer



was determined on three specimens simultaneously placedon the pressure plate. These three curves are nearly identi-cal, suggesting that the three specimens had similar suctionat the beginning of the pressure plate test. This point is veryimportant as the success of the benchmark exercise dependson the reproducibility of the initial state of the specimens.

The results from EPFL are also shown in Fig. 3(a) interms of water content and in Fig. 3(b) in terms of degree ofsaturation. The soil remains quasi-saturated up to suctions ofabout 200 kPa. In the range 100–200 kPa, the specimenmoves from the saturated unloading–reloading line (url) tothe normal consolidation line, as discussed in Fig. 2(a). Inthe saturated and quasi-saturated range (s , 200 kPa), EPFLdata slightly overestimate water content with respect toUGLAS data. This may be associated with the differentvertical pressure applied to improve contact, no pressure forEPFL and 0.6 kPa for UGLAS, as discussed later in thepaper. The different water content may also arise from thewater sucked out of the ceramic plate when releasing the airpressure. In contrast to UGLAS, EPFL did not close thewater reservoir drainage valve before releasing the airpressure. However, the amount of water that could move intothe specimen in the few minutes between reducing the airpressure and weighing the specimen is not expected to besignificant because of the low permeability of the high-air-entry value (HAEV) ceramic (Noguchi, 2009). Beyond200 kPa suction, EPFL and UGLAS data appear to beconsistent.

The results from the axis-translation oedometers areshown in Fig. 4. Concerning the tests carried out at UNINA,one curve refers to a sample prepared at the UNITN, andthe other one refers to a sample prepared at the UNINA,following the same procedure adopted at the UNITN. Thedata obtained at UPC also refer to a sample prepared atUPC. It is interesting to observe that these three curves arevery similar, demonstrating that samples tested in this exer-cise can be reproduced if the procedure for sample prepara-tion is carefully followed. Tests carried out within thisbenchmark exercise could therefore be potentially duplicatedby other laboratories as the ‘ingredients’ of the MUSEmixture are all commercially available.

Water retention data from axis-translation oedometer can-not be directly compared to those obtained in the pressureplate because of the different total stress applied. However,as the specimens remain saturated at suctions lower than100 kPa (Fig. 4), data in this range can be analysed in termsof mean effective stress p9 ¼ p + s, p being the mean totalstress and s the suction.

By using equation (3) and assuming k0¼ 1 � sin�9 (�9being the effective angle of friction), the axis-translationoedometer curves in the range 0–100 kPa in Fig. 4(b) canbe re-plotted in terms of mean effective stress p9 in theplane (p + s, w) by tentatively assuming �9 ¼ 208 or�9 ¼ 308 (Fig. 5). Clearly the assumption k0 ¼ 1 � sin�9 isonly valid in the normally consolidated states, which areassumed to be reached as p9 exceeds 70–80 kPa . Data areconsistent with those obtained using the pressure plate andthis essentially confirms the hydraulic and mechanical pathsfigured out in Fig. 2, taking into account that the transitionfrom overconsolidated to normal consolidated states wasassumed to be sharp in Fig. 2 and is indeed smooth.

The water retention data obtained by using the osmotic

0·1

0·2

0·3

0·4

0·5

0·6G

ravi

met

ric w

ate

r co

nten

t,w

UGLAS – pressure plateEPFL – pressure plate

url

10 100 1000Matric suction, : kPas

0·6

0·7

0·8

0·9

1·0

Deg

ree

ofsa

tura

tion,

Sr

EPFL – pressure plate

200 kPa

(a)

(b)

Fig. 3. Results from pressure plate tests (UGLAS and EPFL) interms of (a) water content and (b) degree of saturation

1 10 100 1000Matric suction, : kPas

0·1

0·2

0·3

0·4

0·5

0·6

Gra

vim

etric

wa

ter

cont

ent,

w

UNINA – sample prepared at UNITNUNINA sample prepared at UNINAUPC sample prepared at UPC

––

0·6

0·7

0·8

0·9

1·0

Deg

ree

ofsa

tura

tion,

Sr

σv 100 kPa�

(a)

(b)

Fig. 4. Results from axis-translation oedometer (UNINA andUPC) in terms of (a) degree of saturation and (b) water content

10 100p s: kPa�

0·4

0·5

0·6

Gra

vim

etric

wa

ter

cont

ent,

w

UGLAS – pressure plateEPFL ressure plate– pUNINA xis-translation oedometer– a

φ� � 20°

φ� � 30°

Fig. 5. Comparison of axis-translation data (pressure plate andoedometer) in the saturated range



method (ENPC) are shown in Fig. 6. Two curves are plottedbased on the calibrations by Williams & Shaykewich andDineen & Burland respectively. At a given water content,differences in the suctions predicted by the two calibrationcurves range from ˜s ¼ 30 kPa at s ¼ 200 kPa to˜s ¼ 150 kPa at s ¼ 800 kPa. These differences are relevantand, in the following, only the curve based on the Dineen &Burland calibration will be discussed. Again, the soil startsdesaturating in the range 100–200 kPa.

Figure 7 shows the results obtained using the high-capa-city tensiometers (UNITN and UDUR). Apart from a UDUR

data point at s ¼ 35 kPa and a significant difference insuction measured at a water content of about 0.2, the twodata sets appear to be consistent. Once again, the soilappears to start desaturating in the range 100–200 kPa.

To compare the water retention data using the pressureplate, the osmotic method and the high-capacity tensiometer,experimental results obtained by EPFL, ENPC and UNITNwere analysed respectively. These data allow the comparisonto be made both in terms of water content and degree ofsaturation (water retention data from UGLAS and UDURwere only provided in terms of water content and wereshown to be similar to those obtained by EPFL and UNITNrespectively). The three WRCs are shown in Fig. 8 and areaugmented by total suction measurements using the WP4dewpoint potentiometer performed at UPC. Data are consis-tent at low suctions (in the range where the soil is saturated)and also at high suction (s . 700 kPa). At high suctions,there appears to be continuity with the dewpoint potenti-ometer data. Discrepancies are observed in the mediumrange of suction (shaded area in Fig. 8). This range corre-sponds to the range where the soil starts desaturating asshown in Fig. 9 and in particular in the range where the airphase is present in occluded form in the pore space.

Inspection of Fig. 9 shows that data in terms of degree ofsaturation are not consistent with data in terms of watercontent (the ENPC water retention data lie between theUNITN and EPFL data in Fig. 8(a) and appear to bound theUNITN and EPFL data in Fig. 9). This is associated witherrors in the estimate of total volume, as shown in Fig. 10,where the degree of saturation and void ratio are plottedagainst water content. There is an inevitable data scatteringwhich causes the apparent discrepancy between water reten-tion data in terms of water content and degree of saturationshown in Fig. 8(a) and Fig. 9 respectively. If a uniquefunction is used to relate the degree of saturation to watercontent, as shown by the thick curve in Fig. 10(a), dataclearly regain consistency as illustrated in Fig. 9(b).

The discrepancy observed in the medium range of suction(shaded area in Fig. 8) is perhaps the most remarkable resultof this exercise. It appears that suction in the pressure plateis overestimated with respect to suction measured by thehigh-capacity tensiometer and the osmotic technique (basedon tensiometer calibration) in the range of degree of satura-tion above 0.7–0.8. The suction difference in this range ofdegrees of saturation appears to be significant. For example,suction measured by the tensiometer at w,0.3 is around200 kPa, whereas the suction applied in the pressure plate is400 kPa at the same water content.

Similar effects were previously observed by Chahal & Yong(1965) for both initially saturated and initially unsaturatedspecimens. These authors also observed greater discrepancies


0·1

0·2

0·3

0·4

0·5

0·6

Gra

vim

etric

wa

ter

cont

ent,

w

Dineen & Burland's calibrationWilliams & Shaykewich’s calibration

0·4

0·6

0·8

1·0

Deg

ree

ofsa

tura

tion,

Sr

url

(a)

(b)

Fig. 6. Results from the osmotic technique (ENPC) in terms of(a) degree of saturation and (b) water content


0·1

0·2

0·3

0·4

0·5

0·6

Gra

vim

etric

wa

ter

cont

ent,

w

UDUR – tensiometer

UNITN ensiometer– t

0·4

0·6

0·8

1·0

Deg

ree

ofsa

tura

tion,

Sr

url

(a)

(b)

Fig. 7. Results from the high-capacity tensiometer (UNITN andUDUR) in terms of (a) degree of saturation and (b) watercontent

10 100 1000 10000 100000Matric suction, : kPas

0

0·1

0·2

0·3

0·4

0·5

0·6

Gra

vim

etric

wa

ter

cont

ent,

w

ENPC – osmotic methodUNITN – tensiometerEPFL – pressure plateUPC – dewpoint potentiometer

url

Fig. 8. Comparison between pressure plate (EPFL), osmoticmethod (ENPC), tensiometer (UNITN) and WP4 dewpointpotentiometer (UPC) in terms of water content



as suction exceeded the air-entry suction, that is in the rangewhere the soil starts desaturating and air is discontinuous.Lourenco et al. (2006) also observed that suction in thepressure plate is overestimated with respect to the tensi-ometer.

A possible explanation of this effect was provided byMarinho et al. (2008) considering the capillary tube concep-tual model. A capillary tube representing a quasi-saturatedstate contains an entrapped air cavity at some elevation

below the atmospheric air–water interface (Fig. 11(a)). If, inthe laboratory environment, this same capillary tube issealed and subjected to an elevated air pressure, the rel-atively high compressibility of the air cavity will lead to asignificant reduction in the cavity’s volume. As the gas–liquid–solid junction of the outer meniscus initially remainsfixed, the curvature of the air–water interface will increasebecause of the compression of the entrapped air cavity asshown in Fig. 11(b). As a result, the pressure differentialbetween air and water, which is controlled by the meniscuscurvature, will increase.

Another point to be addressed is the apparent discrepancybetween the osmotic method and the high-capacity tensi-ometer, as shown in Fig. 8. Inspection of this figure showsthat the difference essentially concerns the ENPC data pointat w,0.27. However, the difference between suction appliedby the osmotic method and suction measured by the tensi-ometer is ˜s ¼ ,40 kPa and it is not significantly greaterthan the standard deviation of the error associated with theDineen & Burland’s calibration curve (˜s� ¼ ,25 kPa). Asa result, no conclusions can be drawn about possible differ-ences between the tensiometer and osmotic method andfurther investigation is required.

Another interesting comparison concerns the two techni-ques based on axis-translation, the pressure plate (EPFL)and the oedometer (UNINA). As shown in Fig. 12(c), theair-entry suction for the UNINA specimens (axis-translationoedometer) is lower than the air-entry suction of the EPFLspecimens (pressure plate). This is surprising as the air-entrysuction of the axis-translation oedometer would have beenexpected to be equal to or greater than that recorded in thepressure plate according to the hydraulic path sketched inFigs 2(b) and 2(c). It then appears that the pressure plateoverestimated suction also with respect to the axis-translationoedometer. As a result, the mechanism illustrated in Fig. 11is not sufficient to explain the higher water content measuredin the pressure plate at a given suction.

The main difference between the oedometer and pressureplate is the vertical stress applied to the specimen. Thisaffects the hydromechanical response of the specimens but itis also expected to control the nature of the contact with thehigh-air-entry ceramic plate. To investigate this effect, aspecific test was devised and carried out at USTRAT. Threespecimens cut from the same sample were simultaneouslyplaced in the pressure plate, each specimen loaded with adifferent weight to apply vertical stresses of 0.3, 4.7 and

0·4

0·6

0·8

1·0

Deg

ree

ofsa

tura

tion,

Sr

100 1000Matric suction: kPa

0·4

0·6

0·8

1·0

Deg

ree

ofsa

tura

tion,

Sr

ENPC – osmotic methodUNITN – tensiometerEPFL – pressure plate

(a)

(b)

Fig. 9. Comparison between pressure plate (EPFL), osmoticmethod (ENPC) and tensiometer (UNITN) in terms of degree ofsaturation: (a) raw data; (b) corrected using a single functionw–Sr

0 0·2 0·4 0·6 0·8Water content

0·4

0·8

1·2

1·6

Voi

d ra

tio, e


Sr 1�

0·2

0·4

0·6

0·8

1·0

Deg

ree

ofsa

tura

tion,

Sr


Fig. 10. Relationship between phase variables in pressure plate(EPFL), osmotic method (ENPC) and tensiometer (UNITN) tests

θ θ

θ θAT 0

a w AT a w 0( ) ( )� � �u u u u

ua0 0�

ua0 0�

uw0 0

( ) 0ua AT �

( ) 0uw AT �

(a)

u ua a AT( )

(b)

Fig. 11. Axis-translation technique in presence of occluded air:(a) negative water pressure; (b) positive water pressure



8.2 kPa respectively. A suction of 200 kPa was applied andthe specimens were left to equilibrate for 12 days.

It was found that water content recorded upon the applica-tion of 200 kPa decreases significantly as the vertical stressincreases (Fig. 12(a)). It is worth noting that the verticalstress applied has very little mechanical effect. The degreeof saturation of the three specimens was equal to unitybefore placing them in the pressure plate and a degree ofsaturation equal to unity was measured after the applicationof the 200 kPa suction. Accordingly, their mechanical re-sponses were controlled by the effective stress. The meaneffective stress is given by the sum of the suction applied(200 kPa) plus one-third of the vertical stress applied, 0.1,1.6 and 2.7 kPa respectively. The increase in mean stressassociated with the weight is therefore very small and cannotjustify a change in water content of about 3%. The resultshown in Fig. 12(a) then seems to demonstrate that thecontact pressure has a significant effect on water content.This can tentatively be explained by assuming that air tendsto fill the gaps between the specimen and the plate, thusreducing the average hydraulic gradient between soil waterand water in the high-air-entry ceramic. At the extreme, if acontinuous gap were to form between the specimen and theplate (as if the specimen were suspended over the plate), nowater exchange would occur. These air cavities tend tocollapse as the vertical stress is increased and thus the waterpressure in the porous ceramic tends to be transmitted moreuniformly to the specimen.

CONCLUSIONSThe paper has presented a ‘round robin’ benchmark test

aimed at comparing different techniques for the measure-ment and control of suction. The soil tested consisted of amixture of kaolinite, bentonite and sand, all of which arecommercially available. Samples were prepared by one la-boratory (Universita di Trento) to ensure that ‘identical’samples were tested by all other teams. However, samplesprepared by normal consolidation from slurry at two differ-ent laboratories (Universita di Napoli Federico II and Uni-versitat Politecnica de Catalunya) showed similar responsesto those samples prepared at the Universita di Trento. Itwould therefore be possible for other laboratories to preparetheir own samples to duplicate the tests presented in thispaper and check their own equipment and procedures.

Techniques tested by the eight MUSE research teamsincluded axis-translation (pressure plate and axis-translationoedometer), high-capacity tensiometer, osmotic techniqueand dewpoint potentiometer. The water retention character-istics of the initially saturated samples were investigatedalong the main drying path. Samples were de-saturated byapplying suction through the liquid phase when using axis-translation technique or osmotic method and de-saturated byair-drying when suction was measured using high-capacitytensiometers or dewpoint potentiometer.

When the same technique was tested by two differentlaboratories (pressure plate, axis-translation oedometer andhigh-capacity tensiometer), very similar WRCs were ob-tained, which is a remarkable result in a round robin test. Ittherefore appears that techniques for suction measurement orcontrol are sufficiently reliable to be successfully used inroutine engineering practice.

Differences were observed between the pressure plate andthe tensiometer. In particular, the pressure plate seems tooverestimate suction at given water content in the range ofhigh degrees of saturation, that is where air is in occludedform. This finding is in agreement with experimental evi-dence from the literature.

The pressure plate appears to overestimate suction alsowith respect to the axis-translation oedometer where non-zero total stress is applied to the specimen. It is suggestedthat this difference is associated with the different contactpressure between the specimen and the high-air-entry cera-mic plate. A specifically designed test has demonstrated thatcontact pressure can significantly change the equilibriumwater content at a given applied suction.

ACKNOWLEDGEMENTSThe authors wish to acknowledge the support of the

European Commission by way of the ‘Marie Curie’ ResearchTraining Network contract number MRTN-CT-2004-506861.The helpful discussions with Anh-Minh Tang from the Ecoledes Ponts – ParisTech, UR Navier/CERMES are also grate-fully acknowledged.

REFERENCESAiro Farulla, C. & Ferrari, A. (2005). Controlled suction oedometric

tests: analysis of some experimental aspects. Proceedings of aninternational symposium on advanced experimental unsaturatedsoil mechanics (EXPERUS), Trento, Italy, pp. 43–48. Rotterdam:Balkema.

Atkinson, J. H. (1993). An introduction to the mechanics of soilsand foundations: through critical state soil mechanics. London:McGraw-Hill.

Chahal, R. S. & Yong, R. N. (1965). Validity of the soil watercharacteristic determined with the pressurized apparatus. SoilSci. 99, No. 2, 98–103.

Cui, Y. J. & Delage, P. (1996). Yielding and plastic behaviour of an

0·2

0·3

0·4

0·5

0·6

Gra

vim

etric

wa

ter

cont

ent,

w

EPFL – pressure plateUNINA – axis translation oedometerUSTRAT – pressure plate

0 2 4 6 8 10Vertical stress, : kPa

(a)σv

0·42

0·43

0·44

0·45

0·46

Wa

ter

cont

ent,

w


0·5

0·6

0·7

0·8

0·9

1·0

Deg

ree

ofsa

tura

tion,

Sr

EPFL – pressure plateUNINA – axis translation oedometer

160 kPa200 kPa

(b)

(c)

Fig. 12. Comparison between axis-translation oedometer (UNINA)and pressure plate (EPFL) and effect of contact pressure on watercontent (USTRAT)



unsaturated compacted silt. Geotechnique 46, No. 2, 291–311,doi: 10.1680/geot.1996.46.2.291.

De Gennaro, V., Canou, J, Dupla, J. C. & Benahmed, N. (2004).Influence of loading path on the undrained behaviour of amedium loose sand. Can. Geotech. J. 41, No. 1, 166–180.

Delage, P., Howat, M. D. & Cui, Y.-J. (1998). The relationshipbetween suction and swelling properties in a heavily compactedunsaturated clay. Engng Geol. 50, No. 1–2, 31–48

Delage, P., Romero, E. & Tarantino, A. (2008). Keynote lecture:Recent developments in the techniques of controlling and meas-uring suction. Proc. 1st Eur. Conf. Unsaturated Soils, Durham,UK 1, 33–52.

Di Mariano, A., Airo Farulla, C. & Valore, C. (2000). Retentioncurves and 1-D behaviour of a compacted tectonised unsaturatedclay. In Experimental evidence and theoretical approaches inunsaturated soils (eds A. Tarantina and C. Mancuso), Proceed-ings of an international workshop on unsaturated soils, Trento,Italy, pp. 47–64. Rotterdam: Balkema.

Dineen, K. & Burland, J. B. (1995). A new approach to osmoticallycontrolled oedometer testing. Proc. 1st Conf. Unsaturated Soils(UNSAT’95), Paris (eds E. E. Alonso and P. Delage) 2, 459–465. Rotterdam: AA Balkema.

Gallipoli, D., Wheeler, S. J. & Karstunen, M. (2003). Modelling thevariation of degree of saturation in a deformable unsaturatedsoil. Geotechnique 53, No. 1, 105–112, doi: 10.1680/geot.2003.53.1.105.

Hoffmann, C., Romero, E. & Alonso, E. E. (2005). Combiningdifferent controlled-suction techniques to study expansive clays.Proceedings of an international symposium on advanced experi-mental unsaturated soil mechanics (eds A. Tarantino, E. Romeroand Y. J. Cui), Trento, Italy, pp. 61–67. Leiden: AA Balkema.

Lourenco, S. D. N. (2008). Suction measurements and water retentionin unsaturated soils. PhD thesis, Durham University, Durham, UK.

Lourenco, S. D. N., Gallipoli, D., Toll, D. G. & Evans, F. D.(2006). Development of a commercial tensiometer for triaxialtesting of unsaturated soils. Proc. 4th Int. Conf. UnsaturatedSoils, Phoenix (eds G. A. Miller, C. E. Zapata, S. L. Houstonand D. G. Fredlund) 2, 1875–1886. Reston: ASCE, GeotechnicalSpecial Publication 147.

Lourenco, S. D. N., Gallipoli, D., Toll, D. G., Augarde, C., Evans,F. & Medero, G. (2008). Calibrations of a high-suction tensi-ometer. Geotechnique 58, No. 8, 659–668, doi: 10.1680/geot.2008.58.8.659.

Lourenco, S. D. N., Gallipoli, D., Toll, D. G., Augarde, C. E. &Evans, F. D. (2011). A new procedure for the determination ofthe soil water retention curve by continuous drying using highsuction tensiometers. Can. Geotech. J. 48, No. 2, 327–335.

Marinho, F. A. M., Take, A. & Tarantino, A. (2008). Tensiometericand axis translation techniques for suction measurement. Geo-tech. Geol. Engng 26, No. 6, 615–631.

MUSE website (2005). MUSE ‘Marie Curie’ Research TrainingNetwork. See http://muse.dur.ac.uk for further details (accessed31 August 2005).

Noguchi, T. (2009). Comparison of major suction measurement

techniques used to determine the soil water retention curves.MEng dissertation, Durham University, Durham, UK.

Peron, H., Hueckel, T. & Laloui, L. (2007). An improved volumemeasurement for determining soil water retention curves. Geo-tech. Testing J. 30, No. 1, 1–8.

Rampino, C., Mancuso, C. & Vinale, F. (1999). Laboratory testingon an unsaturated soil : equipement, procedures, and firstexperimental results. Can. Geotech. J. 36, No. 1, 1–12.

Rojas, J. C., Mancuso, C. & Vinale, F. (2007). Stress state andhysteresis influence upon the SWRC of a silty sand using themodified oedometer apparatus. International congress ondevelopment, environment and natural resources: multi-level andmulti-scale sustainability, Cochabamba, Bolivia, 11–13 July2007, vol. 1, pp. 145–153.

Romero, E. (1999). Characterisation and thermo-hydro-mechanicalbehaviour of unsaturated boom clay: an experimental study.PhD thesis, Universitat Politecnica de Catalunya, Barcelona,Spain.

Romero, E. (2001). Controlled-suction techniques. Proc. 48 Simpo-sio Brasileiro de Solos Nao Saturados N SAT’2001 (eds W. Y. Y.Gehling and F. Schnaid), PortoAlegre, Brazil 1, 535–542.

Tang, A.-M. & Cui, Y.-J. (2005). Controlling suction by the vapourequilibrium technique at different temperatures and its applica-tion in determining the water retention properties of MX80 clay.Can. Geotech. J. 42, No. 1, 287–296.

Tang, A.-M., Cui, Y.-J., Qian, L.-X., Delage, P. & Ye, P. W.-M.(2010). Calibration of the osmotic technique of controllingsuction with respect to temperature using a miniature tensi-ometer. Can. Geotech. J. 47, No. 3, 359–365.

Tarantino, A. (2004). Panel report: Direct measurement of soilwater tension. Proc. 3rd Int. Conf. Unsaturated Soils (eds J. F.T. Juca, T. M. P. de Campos and F. A. M. Marinho) 3, 1005–1017.

Tarantino, A. (2007). Task A3: Benchmarking methods of suctioncontrol – Preparation of soil samples. MUSE internal report.

Tarantino, A. (2009). A water retention model for deformable soils.Geotechnique 59, No. 9, 751–762, doi: 10.1680/geot.7.00118.

Tarantino, A. & Mongiovı, L. (2002). Design and construction of atensiometer for direct measurement of matric suction. Proc. 3rdInt. Conf. Unsaturated Soils (eds J. F. T. Juca, T. M. P. deCampos and F. A. M. Marinho) 1, 319–324.

Tarantino, A. & Mongiovı, L. (2003). Calibration of tensiometer fordirect measurement of matric suction. Geotechnique 53, No. 1,137–141, doi: 10.1680/geot.2003.53.1.137.

Tarantino, A. & Tombolato, S. (2005). Coupling of hydraulic andmechanical behaviour in unsaturated compacted clay. Geotechni-que 55, No. 4, 307–317, doi: 10.1680/geot.2005.55.4.307.

Williams, J. & Shaykewich, C. F. (1969). An evaluation of poly-ethylene glycol PEG 6000 and PEG 20000 in the osmoticcontrol of soil water matrix potential. Can. J. Soil Sci. 102, No.6, 394–398.

Wissa, A. E. Z. & Heiberg, S. (1969). A new one-dimentionalconsolidation test. Cambridge, MA: Massachusetts Institute ofTechnology, Research Report 69–9.



Session 4

Application to Engineering Problems and Case Studies


Gens, A. et al. (2011). Geotechnique 61, No. 5, 367–386 [doi: 10.1680/geot.SIP11.P.015]

141

Hydromechanical behaviour of a heterogeneous compacted soil:experimental observations and modelling

A. GENS�, B. VALLEJAN�, M. SANCHEZ†, C. IMBERT‡, M. V. VILLAR§ and M. VAN GEET¶

The paper describes a theoretical and experimental studyof the coupled hydromechanical behaviour of a com-pacted mixture of bentonite powder and bentonite pelletsintended as sealing material in underground repositoriesfor nuclear waste. One of the main advantages of the useof powder/pellets mixtures is the reduction of the com-paction effort required to achieve the value of averagedry density necessary to attain the required swellingpotential. However, the heterogeneous fabric of the ma-terial requires special approaches in order to describeadequately its behaviour during hydration. A doubleporosity formulation is presented to account for thepresence of two distinct structural levels in the material.Hydraulic equilibrium between the two porosities is notassumed; instead a water exchange term between them ispostulated. The formulation is applied to the modellingof a number of one-dimensional swelling pressure testsperformed in the CEA (Commisariat a l’Energie Atomi-que, France) and CIEMAT (Spain) laboratories. A verysatisfactory quantitative description of the experimentalobservations is obtained that includes a number of com-plex behaviour features such as size effects and non-monotonic development of swelling pressures. Somemicrofabric observations using X-ray tomography andmercury intrusion porosimetry lend support to the con-ceptual approach adopted. The formulation is then ap-plied to the analysis of a long-term large-scale sealingtest performed at the Hades underground facility inBelgium, using the same set of hydraulic and mechanicalparameters employed in the modelling of the laboratorytests. Although the field observations exhibit a muchhigher degree of scatter, the basic behaviour of the fieldsealing test is satisfactorily simulated. A formulation thatincorporates basic features of the microfabric of the mix-ture is thus able to span successfully over a large rangeof space and time scales.

KEYWORDS: compaction; expansive soils; fabric/structure ofsoils; full-scale tests; laboratory tests; partial saturation

La presente communication decrit une etude theorique etexperimentale du comportement hydromecanique mixted’un melange compacte de poudre et de granules debentonite utilise comme materiau d’etancheite dans desdepots souterrains de dechets nucleaires. Un des princi-paux avantages de l’utilisation de melanges poudre/granules est la reduction du travail de compactage neces-saire pour realiser la densite seche moyenne requise pourobtenir le potentiel de gonflement specifie. Toutefois, lastructure heterogene du materiau necessite l’emploi demethodes particulieres pour la description adequate deson comportement au cours de l’hydratation. On presenteune formulation a double porosite pour prendre en con-sideration la presence de deux niveaux structurels dis-tincts dans le materiau. On ne suppose pas l’existenced’un equilibre hydraulique entre les deux porosites, maison postule une condition d’echange d’eau entre elles. Onapplique la formulation pour la modelisation d’un certainnombre d’essais de pression de gonflement unidimension-nels effectues dans les laboratoires de la CEA (en France)et de CIEMAT (en Espagne). On obtient ainsi unedescription extremement satisfaisante des observationsexperimentales , y compris un certain nombre de carac-teristiques complexes du comportement, par exemple leseffets de la taille et le developpement non monotones despressions de gonflement. Certaines observations des mi-cro-tissus effectuees en tomographie aux rayons X et enporosimetrie par intrusion de mercure (MIP) soutiennentle principe conceptuel adopte. On applique ensuite laformulation a l’analyse d’un essai d’etancheite a grandeechelle et a long terme, effectuee dans l’installationsouterraine Hades en Belgique, en utilisant le memeensemble de parametres hydrauliques et mecaniques uti-lise pour la modelisation des essais en laboratoire. Bienque les releves sur le terrain presentent une diffusionbien plus elevee, on realise une simulation satisfaisantedu comportement de base de l’essai d’etancheite sur leterrain. On dispose ainsi d’une formulation comprenantles caracteristiques de base du micro-tissu du melangecouvrant avec succes une vaste plage d’echelles spatialeset des temps.

INTRODUCTIONThe construction of deep underground repositories for high-level nuclear waste requires the excavation of shafts, orramps, as well as horizontal drifts that give access to the

waste emplacement area. Much attention has been given toengineered barriers placed around the waste (Gens et al.,1998; 2009b) but the sealing of access shafts and drifts toblock potential preferential pathways for radionuclide migra-tion is of comparable importance (IAEA, 1990; Gens, 2003).Compacted bentonite is often considered as a suitable back-fill material, not only because of its low permeability andfavourable retardation properties, but also for its capability,on hydration, to fill gaps and voids that may arise fromexcavation irregularities or other causes. Therefore, in orderto ensure the long-term safety of the repository, the backfillmaterial must have enough swelling potential upon hydrationto perform the sealing function satisfactorily. As swellingpotential depends essentially on the dry density of thesealing material (Imbert & Villar, 2006), it may be necessary

Manuscript received 3 March 2010; revised manuscript accepted20 October 2010. Published online ahead of print 22 March 2011.Discussion on this paper closes on 1 October 2011, for furtherdetails see p. ii.� Department of Geotechnical Engineering and Geosciences,Universitat Politecnica de Catalunya, Barcelona, Spain.† Texas A & M University, College Station, USA.‡ CEA (Commisariat a l’Energie Atomique), Gif sur Yvette,France.§ CIEMAT, Madrid, Spain.¶ ONDRAF/NIRAS, Brussels, Belgium.


to apply a significant compaction effort in order to achievean adequate density value. Often, this causes operationaldifficulties as it is difficult to ensure a uniform and intensecompaction in enclosed, irregularly shaped spaces. Pre-compacted bentonite blocks are sometimes used but theyleave gaps between them and they are also difficult toaccommodate when shafts and drifts have uneven excavationsurfaces, a feature that is almost unavoidable in many cases.An interesting alternative is to use, as sealing material, amixture of bentonite powder and highly compacted bentonitepellets (Volckaert et al., 1996). In this way, the average drydensity is already quite high on placement due to thecontribution of the high-density pellets. Consequently, onlymodest compaction efforts may be required after putting thematerial in place and, in any case, a minimum density isalways ensured. As an example, Fig. 1 shows the sealing ofan experimental shaft backfilled with a compacted mixtureof bentonite powder and pellets.

The resulting material is obviously highly heterogeneous,consisting of a mixture of relatively low-density bentonitepowder with high-density bentonite pellets. The behaviour ofsuch material upon hydration is likely to be complex andmust be properly understood if a sufficient degree of con-fidence in the design and performance of the seal is to beachieved. A conceptual framework leading to a good under-standing of this behaviour should also be of benefit for otherheterogeneous compacted materials, such as, for instance,the often-encountered case of compacted clayey soils con-taining large lumps of intact material.

In this paper, some basic features of the hydration behav-iour of this heterogeneous compacted soil are presented first.Afterwards, the theoretical formulation required for theanalysis of this type of material is briefly described. Theformulation is then applied to the modelling of a series ofone-dimensional swelling pressure tests performed in thelaboratory. Finally, the analysis is upscaled to the descriptionof a full-scale sealing test performed in an undergroundlaboratory.

BASIC FEATURES OF THE BEHAVIOUR OF POWDER–PELLETS MIXTURES DURING HYDRATION

The behaviour of the powder–pellets mixture duringhydration has been examined by means of laboratory swel-ling pressure tests (Imbert & Villar, 2006). Note that thehydration of a shaft or tunnel seal due to the inflow of hostrock water is akin to a swelling pressure test due to theconfinement provided by the excavation walls. Swelling

pressure tests are, therefore, very appropriate to examine thebehaviour relevant to seal hydration.

Mixtures of 50% bentonite powder and 50% bentonitepellets by dry weight have been tested. FoCa clay, a calciumbentonite from the Paris Basin, has been selected for thestudy. The major component of the clay fraction is aninterstratified clay mineral of 50% calcium beidellite and50% kaolinite. It has also a number of accessory mineralsand the major exchangeable cation is Ca2þ. It also contains,in decreasing amounts, Mg2þ, Naþ and Kþ (Volckaert et al.,2000). Pellets are manufactured by dynamic compaction ofthe powder between two rotating wheels. The dimensions ofthe pellets (Fig. 2) are 25 3 25 3 15 mm and their averagedry density is 1.89 g/cm3. Compaction water content lies inthe range 4–5%.

The swelling pressure tests were performed in the labora-tories of CEA (Commisariat a l’Energie Atomique, France)and CIEMAT (Spain). Samples were statically compacted tothe desired density and were subsequently tested in oed-ometers equipped with load cells. The specimens werehydrated from the bottom, sample deformation was pre-vented and the evolution of swelling pressure recorded.Water intake during the test was also measured. Because ofthe low permeability of the bentonite, tests had to be runover extended periods generally lasting several months.Deionised water was used for hydration as the water fromthe site where the in situ test has been performed has a verylow salinity, thus minimising potential chemical effects onresults. In fact, the influence of the salinity of the solutionon swelling pressure has been considered negligible or smallfor highly compacted bentonites, especially when divalentcations predominate because diffuse double layers are poorlydeveloped (Pusch, 1994; Castellanos et al., 2008). The issueof chemical effects on the hydromechanical behaviour ofbentonite has been discussed in detail in Guimaraes et al.(2007).

Typical results of the oedometer swelling pressure testsare shown in Fig. 3. They correspond to specimens ofdifferent lengths (5, 10 and 12 cm) compacted to a commondry density of 1.60 g/cm3. The diameter of the samples is12 cm in order to accommodate a representative number ofpellets. Several remarks can be made.

(a) The evolution of the swelling pressure is verycharacteristic suggesting a rather complex pattern ofbehaviour. Swelling pressure increases rapidly at thebeginning of the test, until reaching a peak value.Afterwards, the measured pressure drops to a minimum

Fig. 1. Mixture of bentonite pellets and bentonite powder usedas sealing material in an experimental shaft

Fig. 2. Pellets of FoCa clay used in the investigation. Thedimensions of the pellets are 25 3 25 3 15 mm

142 GENS, VALLEJAN, SANCHEZ, IMBERT, VILLAR AND VAN GEET


value to start later a new increase until reaching astationary swelling pressure value at the end of the test.The same pattern of behaviour has been observed in thetests performed at other dry densities.

(b) There is a clear influence of the sample length on therate of development of the swelling pressure. Thelonger the specimen, the slower the hydration kineticsis. This fact should be taken into account whentransferring laboratory test results to field conditionswhere scale and hydration paths are bound to be quitedifferent.

(c) The final swelling pressure at the end of the test issimilar for the three samples, irrespective of length.Indeed, as Fig. 4 indicates, it has been found that finalswelling pressure depends basically on dry density. Itcan be noted that the relationship between dry densityand final swelling pressure for mixtures is very similarto that obtained in the same type of tests performed onsamples of compacted bentonite powder (Imbert &Villar, 2006).

Interactions in the microfabric of the material duringhydration must underlie the observed macroscopic behaviour.Direct observations of the hydration process were obtainedby Van Geet et al. (2005) using microfocus X-ray computedtomography. A 50/50 powder–pellet mixture of FoCa clay at

a dry density of 1.36 g/cm3 and an average water content of5.67% was placed in a methacrylate cylindrical cell. Thesample was 7 cm high and 3.8 cm in diameter. Hydrationwas performed from the bottom of the sample under con-stant volume conditions. During the first 1.5 months, waterwas supplied at a very low pressure; afterwards water wasinjected at a pressure of 0.5 MPa during four additionalmonths. A permeability determination was performed at theend of the test using a 0.6 MPa water pressure. X-raytomography observations could be performed at specifiedtime intervals. From the computation of the linear X-rayattenuation coefficient, the density distribution throughoutthe sample could be determined. Fig. 5 shows the densitydistribution on a vertical slice through the centre of thesample at different times of the experiment. It can be noted:

(a) at the initial state there is a clear difference in densitybetween pellets and powder

(b) during hydration, the density of the pellets reduces(swelling) whereas the density of the powder risesbecause of both soil compression and water contentincrease; the changes move gradually from bottom totop following the progress of hydration

(c) an apparently homogeneous specimen is obtained at theend of the test.

Further information is provided in Fig. 6 that shows theincremental density changes between the stages depicted inFig. 5. It can be observed how hydration rises from thebottom of the specimen towards the top, causing densitychanges due to the swelling of the pellets and to the increasein saturation. Very little density change occurs after 2.5months where the specimen has become much more homo-geneous. It is also interesting to note that, in the initialstages of the test, the surfaces of the pellets away from thehydration front show some swelling due to the local ex-change of water between powder and pellets.

THEORETICAL FORMULATIONBalance equations

The adoption of a double porosity model appears to beeminently suited to describe the heterogeneous nature of thematerial. Probably this concept was first introduced byBarrenblatt et al. (1960) (as quoted in Ghafouri & Lewis,1996) to simulate flow through rigid, fissured porous mediaand it was later enlarged to consider the coupling betweenfluid flow and soil deformation (e.g. Aifantis, 1980; Wilson& Aifantis, 1982; Gens et al., 1993; Khalili et al., 1999;Callari & Federico, 2000). Here, the formulation developedin Sanchez (2004) is adopted.

The overall medium is assumed to consist of two over-lapping but distinct continua. The macrostructure refers tothe large-scale arrangement of soil particle aggregates andthe relatively large pores between them. It is expected that,initially, most of the macrostructural pores belong to thebentonite powder. The microstructure refers to the clayparticles and the micropores and interparticle spaces asso-ciated with them. A large proportion of the micropores lieinitially in the high-density pellets but there will also bemicropores in the clay particle aggregates present in thepowder. Ideally, the microporosity of the powder aggregatesshould be distinguished from that in the pellets but, in thatcase, the number of interactions and parameters multiplyleading to a cumbersome formulation that is difficult toapply in practice. As shown later, a double porosity modelappears to be sufficient to describe the hydromechanicalbehaviour of the material. In the following, subscript M willstand for the macrostructure and subscript m for the micro-structure. Accordingly, macroporosity and microporosity are

4·0

3·0

2·0

1·0

0

Sw

ellin

g pr

essu

re: M

Pa

0 1 10 100 1000

Time: days

RS2B, 5 cm, 12 cmh � �φRS2B, 10 cm, 12 cmh � �φ

RS2B, 12 cm, 12 cmh � �φ

Fig. 3. Evolution of observed swelling pressures of a 50/50powder/bentonite mixture. Tests performed at CEA laboratoryat dry density of 1.60 g/cm3

4·0

3·0

2·0

1·0

0

Sw

ellin

g pr

essu

re: M

Pa

1·20 1·30 1·40 1·50 1·70

Final dry density: g/cm3

FoCa7 small samples

FoCa reseal small samples

Pellets/powder mixtures

1·60

FoCa7 big samples

Fitting for mixtures

Fig. 4. Variation of swelling pressures of FoCa bentonite withdry density from tests performed in the CEA and CIEMATlaboratories. The dashed line corresponds to the relationship for50/50 pellets/powder mixtures and the continuous line corre-sponds to compacted powder

HYDROMECHANICAL BEHAVIOUR OF A HETEROGENEOUS COMPACTED SOIL 143


denoted as jM and jm respectively. Macroporosity andmicroporosity are defined as the volume of macropores andmicropores, respectively, divided by the total volume of thesoil. Thus, total porosity j equals jM þ jm. The degree ofsaturation of the macroporosity, Sw M , is the volume ofmacropores occupied by water over the volume of the

macropores; an equivalent definition holds for the micropor-osity degree of saturation, Sw m.

An important feature is that hydraulic equilibrium betweenthe two continua is not assumed; that is, at each point of thedomain the water potentials in the two continua may bedifferent, leading to an exchange of water between them (Fig.

After 1·5 monthsAfter 0·5 months

After 5·5 months

Initial state

After 2·5 months

Density: g/cm3

0 0·5 1·0 1·5 2·0

Fig. 5. Distribution of densities on a vertical slice through the centre of the sample observed at different stages ofhydration (after Van Geet et al., 2005). The specimen is 7 cm high and 3.8 cm in diameter



7). For simplicity, a linear relationship is assumed (e.g. Wilson& Aifantis, 1982) where water exchange is described by

ˆw ¼ ª (łM � łm) (1)

where ˆw is the water exchange term, ª is a parameter(often called the leakage parameter) and ł is the total water

potential. It is assumed that only matric and gravitationalpotential contribute to the total potential of the macrostruc-ture but an additional osmotic component may also contri-bute to the microstructural potential (Gens, 2010). Here,potential is defined in pressure units. As the water exchangeis local in space, the gravitational potential will be the same

From 0·5 months to 1·5 months


From initial state to 0·5 months


Density change: g/cm3

2 1·5 1 0·5 0 �0·5 �1 �1·5 �2

Fig. 6. Distribution of incremental density changes on a vertical slice through the centreof the sample observed at different stages of hydration (after Van Geet et al., 2005). Thespecimen is 7 cm high and 3.8 cm in diameter



for the two media. Water exchange will therefore be drivenby suction differences alone.

Using the concept of material derivative, the balanceequation for the solid phase can be written, for the case of asingle porosity medium, as:

DjDt¼ 1� jð Þ

rs

Drs

Dtþ 1� jð Þ _�v (2)

where D(.)/Dt denotes material derivative, rs is the soliddensity, _�v is the volumetric strain increment and t is time.For the case of double porosity equation (2) becomes(Sanchez, 2004)

DjDt¼ DjM

Dtþ Djm

Dt¼ 1� jð Þ

rs

Drs

Dt

þ 1� jM � jmð Þ ( _�vM þ _�vm)

(3)

where it has been assumed that the total volumetric defor-mation is the sum of the volumetric deformations of eachmedium.

The water mass balance equation for the case of twooverlapping flow media is

@

@ trwSw j� j

� �þ = � jw j

� �� ˆw ¼ f wj ; j ¼ M , m (4)

where Swj is the liquid saturation of medium j, jw j is the

total mass fluxes of water in the liquid phase and fjw is the

external mass supply of water per unit volume in medium j.The possible presence of dissolved air in the liquid phase isneglected for simplicity.

Finally, it is assumed that total stresses for the overallmedium affect equally the macro- and the microstructure.The equilibrium equation is

= � � þ b ¼ 0 (5)

where � is the total stress tensor and b is the vector of bodyforces. In contrast, the total deformation of the medium isobtained from the sum of the deformations of each domain.

To perform numerical analyses, this set of balance equa-tions is incorporated into a computer code by way of asuitable discretisation procedure. The analyses reported in

the paper have been performed with a modified version ofCode_Bright (Olivella et al., 1996) that incorporates thedouble porosity formulation.

Hydraulic constitutive equationsLiquid flow is governed by Darcy’s law

qw j ¼ �K w j =ł j

� �¼ �K w j = pw j � =rw jg

� �; j ¼ M , m

(6)

where q is the mass liquid flow (with respect to the solidphase), K l is the liquid permeability tensor, pl is the liquidpressure, rl is the liquid density and g is the gravity vector.The permeability tensor is expressed as

K w j ¼ k j

kr j

j

; j ¼ M , m (7)

where k is the intrinsic permeability, kr is the relativepermeability that expresses the effect of degree of saturation(or suction) on global permeability and j is the liquidviscosity. Intrinsic permeability depends on many factorssuch as pore size distribution, pore shape, tortuosity andporosity. Here a dependence of intrinsic permeability onporosity is adopted

k j ¼ ko exp b(� j � �o)� �

; j ¼ M , m (8)

where �o is a reference permeability for which the intrinsicpermeability is ko. Relative permeability is given by theempirical relationship

kr j ¼ Sºe j ; j ¼ M , m Se ¼

Sl � Slr

Sls � Slr

(9)

where Se is the effective degree of saturation, Slr is theresidual degree of saturation and Sls is the degree of satura-tion in saturated conditions, normally taken as 1.

Finally the retention curve relates suction (or matricpotential) with degree of saturation. There are a number ofdifferent expressions designed to fit experimentally deter-mined retention curves. A modified form of the VanGenuchten (1980) law has been used here

Se j ¼ 1þ s

Po

� �1= 1�ºoð Þ" #�ºo

1� s

Pd

� �ºd

; j ¼ M , m

(10)

where s is the matric suction and Po, Pd, ºo and ºd aremodel parameters.

Mechanical constitutive lawIt has been postulated that the behaviour of highly expan-

sive clays is better described assuming two levels of struc-ture, a macrostructural level and a microstructural level, akinto the two porosities considered in the formulation justdescribed (Gens & Alonso, 1992). In this work, the mathe-matical model formulated in terms of generalised plasticity(Sanchez et al., 2005) is used.

The model assumes that the physicochemical phenomenaoccurring at the microstructural level are basically reversibleand largely volumetric. Then the deformations arising frommicrostructural phenomena can be obtained from a non-linear elastic model dependent on a microstructural meanstress (pp) defined as

pp ¼ pþ sm (11)

ψ2

Γw1 2( )� �γ ψ ψ

ψ1

Fig. 7. Scheme of a double porosity material with a waterexchange term



where p is the net mean stress and sm is the microstructuralsuction.

In the p–s plane, the line corresponding to constantmicrostructural mean stresses (pp) is referred to as the neutralloading line (FNL) since no microstructural deformation oc-curs when the stress path moves on it (Fig. 8(a)). Theneutral loading line (NL) divides the p–s plane into twoparts, defining two main microstructural stress paths: amicrostructural contraction path (MC, _pp . 0) and a micro-structural swelling path (MS, _pp , 0).

The increment of the microstructural elastic deformationis expressed as a function of the increment of the micro-structural mean stress

_�vm ¼_pp

K m

; K m ¼1þ em

km

pp (12)

where _�vm is the microstructural volumetric strain increment,Km is the microstructural bulk modulus, em is the micro-

structural void ratio and km is a model parameter. Thebehaviour of the macrostructural level is defined by theBarcelona basic model (BBM) (Alonso et al., 1990). Forcompleteness, the main BBM equations are presented in theAppendix.

A fundamental assumption of the framework is thatmicrostructural behaviour is not affected by the macrostruc-ture state but it only responds to changes in the drivingvariables (i.e. stresses and suction) at local microstructurallevel. In contrast, plastic macrostructural strains may resultfrom deformations of the microstructure. It is postulated thatplastic macrostructural strains can be expressed as

_�pvM ¼ f _�vm (13)

where _�pvM is the macrostructural plastic strain arising from

the deformation of the microstructure. Two interaction func-tions f are defined: fc for MC paths and fs for MS paths (Fig.8(b)). In the case of isotropic loading, the interaction func-tions depend on the ratio p/po where po is the yield meannet stress at the current macrostructural suction value. Theratio p/po is a measure of the degree of openness of themacrostructure relative to the applied stress state. When thisratio is low it implies a dense packing of the material, it isthen expected that the microstructural swelling (MS path)will affect strongly the global arrangements of clay aggre-gates, inducing large macrostructural plastic strains. If thep/po ratio comes close to 1, the state of the material ap-proaches the BBM loading–collapse (LC) yield surface andit becomes collapsible. In that case, the multiplying effect ofmicrostructural strains may be small or even negative. In thiswork, the following form of interaction function has beenused

fÆ ¼ fÆ0 þ fÆi 1� p

po

� �nÆ

;

Æ ¼ s (swelling), c (contraction)

(14)

Although this form of formulating the interplay betweenmicrostructure and macrostructure is conceptually quite sim-ple, it has proved able to accommodate a wide range ofdifferent interaction phenomena (Gens & Alonso, 1992).

MODELLING SWELLING PRESSURE TESTSFeatures of the tests and model parameters

The formulation outlined above has been applied to themodelling of a number of swelling pressure tests performedon specimens of 50/50 powder/pellets mixtures compacted totwo different densities (1.45 g/cm3 and 1.60 g/cm3) and witha range of different lengths. The main characteristics of thetests are given in Table 1. The pellets always have the samedensity; the variation in the density of the samples arisesfrom the different initial powder densities.

According to the formulation, it is necessary to estimatethe initial microstructural and macrostructural porosity bothin the pellets and in the powder. Mercury intrusion porosi-metries (MIPs) were performed on the pellets, a typicalresult is shown in Fig. 9. There is always some degree ofarbitrariness in the distinction between the two types ofpores, pore size distribution being a continuum. Here, a poresize of 100 nm has been selected for this purpose based onthe fact that there is a clear change of slope in the pore sizedensity functions determined experimentally. This value isalso consistent with other previous estimations in similarmaterials (Touret et al., 1990; Romero, 1999; Lloret et al.,2003). For the powder, the distinction between porositieswas made on the bases of experimental pore size determina-tions reported previously (Romero, 1999; Marcial, 2003)

s

p

Current state

Neutralloading

Microstructural

compression

Neutralloading

Microstructural

expansion

(a)

0·8

0·6

0·4

0·2

0

�0·2

f

0 0·2 0·4 0·6 0·8 1·0

fc

fs

p p/

(b)0

Fig. 8. Double structure mechanical constitutive law for expan-sive clays: (a) microstructural model; (b) interaction functions



using the same threshold value. From this ensemble ofresults, it was found that the proportion of microporositywith respect to total porosity of compacted samples could beclosely approximated by

vm=vT ¼ 0:0632ª3:89d (15)

where ªd is the dry density in g/cm3, vm is the microstruc-tural pore volume and vT is the total pore volume. Fig. 10shows the assumed variation of the different porosities (andvoid ratios) with as-compacted dry density. It is apparentthat microporosity increases significantly in the highly com-

pacted samples both in absolute and relative terms. Lackingmore specific information, the initial proportions of water inthe microstructure and in the macrostructure are assumed tobe similar to those of the porosity components.

The hydraulic parameters are presented in Table 2. Therehave been quite a number of determinations of the retentioncurve of compacted FoCa clay at different densities. Theyare shown in Fig. 11 together with the retention curvesadopted in this work. The retention curves’ parameters forthe micro- and macroporosity have been estimated from theresults of the higher and the lower density samples, respec-tively, as micropores are dominant in the high-density speci-mens and macropores predominate in the low-densityspecimens. Intrinsic permeability values and their variationhave been derived from laboratory permeability tests oncompacted FoCa clay at different dry densities, the results ofwhich are summarised in Gens et al. (2009a). For relativepermeability, an exponent value of 3 is adopted for themacrostructure, a value that has been obtained in severalinfiltration tests on homogeneous compacted bentonite (Genset al., 2009a). A lower exponent is assumed for the micro-structure as it is conjectured that unsaturation has a lowerimpact on water mobility in the vicinity of clay particlesurfaces. The leakage parameter is sometimes associatedwith some geometric characteristics of the medium, such asthe specific surface of the matrix block (Barrenblatt et al.,1960), the number of fractures and fracture intervals (Warren& Root, 1963) or the average size of clay blocks (Callari &Federico, 2000). Because of lack of relevant geometricinformation in this case, the leakage parameter was obtainedfrom back-analysis and the same value was used in all theanalyses. It should be recalled that, although most of themicroporosity is initially present in the pellets, there is alsosome microporosity in the powder. Therefore, water ex-changes between porosity levels will take place both in thepowder and the pellets with, probably, some differences inthe kinetics. However, a single leakage parameter has beenused in the analyses for simplicity.

The parameters of the mechanical constitutive model arecollected in Table 3. They were determined from back-calculation of laboratory oedometer tests on compacted FoCaclay (Volckaert et al., 2000; Yahia-Aıssa et al., 2000; Cui etal., 2002). It should be noted that only one interaction func-tion is required as only the microstructural swelling mechan-ism is engaged in the swelling pressure test. It should bestressed that although some BBM parameters of the macro-structure depend on the initial density of the powder, theparameters characterising the behaviour of the microstructureremain unchanged throughout all the analyses performed.

Swelling pressure tests at dry density 1.45 g/cm3

Tests MGR7, RS2J and MGR 9 were performed onsamples with a dry density of 1.45 g/cm3. The estimatedinitial porosities and suctions are listed in Table 4. Figs

Table 1. Main characteristics of the swelling pressure tests

Test label Sampleheight:

mm

Diameter:mm

Void ratioof sample

Dry densityof sample:

g/cm3

Initial watercontent ofsample: %

Dry densityof pellets:

g/cm3

Initial watercontent ofpellets: %

Dry densityof powder:

g/cm3

Initial watercontent ofpowder: %

Laboratory

MGR7 50 100 0.84 1.45 8.30 1.89 4.60 1.175 12.00 CIEMATMGR9 100 100 0.84 1.45 5.25 1.89 5.00 1.182 5.80 CIEMATRS2J 50 120 0.84 1.45 4.78 1.89 4.00 1.175 5.00 CEARS2B 50 120 0.67 1.60 8.07 1.89 4.31 1.385 11.83 CEARS2E 100 120 0.67 1.60 4.78 1.89 4.49 1.390 5.07 CEARS2F 120 120 0.67 1.60 4.78 1.89 4.49 1.390 5.07 CEA

0·6

0·4

0·2

0

Por

e si

ze d

ensi

ty fu

nctio

n

100000 10000 1000 100 10 1Entrance pore size, : nmd

Pellet APellet B

Inter-aggregatepores or

macropores

Intra-aggregatepores or

micropores

Fig. 9. Pore size density functions for two pellets

1·0

0·8

0·6

0·4

0·2

00·90 1·10 1·30 1·50 1·70 1·90

Dry density: g/cm3

2·0

1·6

1·2

0·8

0·4

0

Voi

d ra

tio

Total porosityMicro porosityMacro porosityTotal void ratioMicro void ratioMacro void ratio

Por

osity

Fig. 10. Relationship between micro, macro and total porositiesand dry density after compaction



12(a)–12(c) show the computed evolution of swelling pres-sures compared with the measured values. Tests were rununtil macroscopic observations indicated that stationary con-ditions were achieved; test durations were over 200 days forthe 5 cm long samples and 570 days for the 10 cm longsample.

In general, the rate of swelling pressure development atthe start of the tests is underestimated by the analyses butthe magnitude of the first peak and the time required toreach it are adequately reproduced by the calculations. Sub-sequent behaviour is also well modelled, including the ob-served reduction of swelling pressure and the final rise tothe final steady-state value. In spite of having two differentlengths, the pattern of behaviour is very similar in all threetests although, of course, the times required to reach the firstpeak and the stationary condition are longer for the tallerspecimen (MGR9). The swelling pressure rise in test MGR7at day 120 is due to an increase of the injected waterpressure. This pressure increase stage was not applied in theother tests. Figs 12(d)–12(f) indicate that the water intake ofthe tests is also well reproduced although, again, it can benoted that the observed rate of water inflow is larger thanthe computed value at the initial stages of the tests.

As the modelling appears to represent satisfactorily thereal system, interesting information might be obtained byanalysing in more detail some of the results of the analyses.For space reasons, attention is focused on test MGR9, butresults from other tests are used when required. Fig. 13(a)shows the time evolution of macrostructural and microstruc-tural suctions at three different points of sample MGR9:near the bottom hydration boundary (vertical coordinatey ¼ 0.0 m), at the centre of the sample (y ¼ 0.05 m) andnear the top of the sample (y ¼ 0.0975 m). It can be ob-served that, at the bottom boundary, the macrostructural andmicrostructural suctions reduce rapidly, but they differ at thebeginning of the test because of the delay in water transfer

Table 2. Hydraulic parameters

Constitutive law Expression Parameter Micro Macro

Retention curvePo: MPa 378.95 15ºo 0.899 0.064

Se ¼Sl � Slr

Sls � Slr

¼ 1þ s

Po

� �1=(1�ºo)" #�ºo

1� s

Pd

� �ºd

Pd: MPa 800 750ºd 2.243 3.899Sls 1.0 1.0Slr 0.0 0.0

Intrinsic permeabilityHigh-density samples, ªd . 1.60 g/cm3

k j ¼ ko exp (b(� j � �o)) ko: m2 2.9 3 10�20

�o 0.40b 8.00

Low-density samples, ªd , 1.60 g/cm3

k j ¼ ko exp (b(� j � �o)) ko: m2 2.9 3 10�20

�o 0.40b 11.00

Relative permeabilitykr ¼ Sº

e º 1.15 3.00Sls 1.00 1.00Slr 0.00 0.00

Leakage parameterˆw ¼ ª (ł1 � ł2) ª: kg s�1 m�3 MPa�1 5.00 3 10�5

1000

100

10

1

0·1

Suc

tion:

MP

a

0 0·2 0·4 0·6 0·8 1·0

Degree of saturation(a)

CEA, pellets test 1CEA, pellets test 2CEA, 1·87 g/cmρd

3�

CEA, 1·95 g/cmd3�ρ

CEA, 1·97 g/cmd3�ρ

Imbert ., 1997,et al ρd31·90 g/cm�

Volckaert ., 1996,et al ρd31·88 g/cm�

Volckaert ., 1996,et al ρd31·96 g/cm�

Delage ., 1998,et al ρd31·86 g/cm�

Modified van Genuchten fit forhigh-density samples

1000

100

10

1

0·1

Suc

tion:

MP

a

0 0·2 0·4 0·6 0·8 1·0

Degree of saturation(b)

CIEMAT, 1·65 g/cmρd3�

CIEMAT, 1·65 g/cmd3�ρ

CEA, powder of FoCa8Michaux, 1995, ρd

31·52 g/cm�

Michaux, 1995, ρd31·35 g/cm�

Olchitzky, 2002, ρd31·45 g/cm�

Modified van Genuchten fit forlow-density samples

Fig. 11. Retention curves for FoCa clay: (a) samples with drydensities in the range 1.65–2.0 g/cm3; (b) samples with drydensities in the range 1.35–1.65 g/cm3



between the two porosities. Eventually, however, they cometogether and remain in equilibrium for the remainder of thetest. Interestingly, at the other two points the two porositiescome into equilibrium before they are reached by the hydra-tion front, that is before they exhibit any suction reduction.They also maintain this equilibrium condition throughout therest of the test. The analysis suggests, therefore, that non-equilibrium between the two porosity levels is only likely toaffect the early stages of the test. Fig. 13(b) shows the sameinformation for one of the shorter (5 cm) samples: MGR7.The results obtained are qualitatively very similar, althoughhydration occurs earlier as the flow path is now shorter. Thetime required to reach steady-state condition is, as expected,also shorter.

The cause underlying the characteristic temporary drop inswelling pressure is the collapse of the macrostructure that,in the constitutive model, corresponds to reaching the LCyield surface in the BBM. This is illustrated in Fig. 14,where the stress paths (in macrostructural suction – verticalnet stress space) followed by the same three points of testMGR9 are plotted. The initial position of the LC yieldsurface is also shown. In the section of the stress path thatmoves towards the LC surface, stress increase appears to bemore significant than suction reduction. Once the LC yieldsurface is reached, the vertical stress drops to compensatethe tendency of the macrostructure to collapse so that thesample length is kept constant. It can also be noted that the

point of start of the collapse does not coincide exactly withthe plotted LC; this is because, by then, the yield surfacehas moved slightly due to the interaction with the micro-structural strains (MS, microstructure swelling) that developfrom the very start of the test.

The interplay between the various aspects of the problemmay be more readily seen by referring to Fig. 15, whichcontains details of the evolution of test MGR9 for the samethree points considered above. Fig. 15(a) shows the evolutionof vertical swelling pressures that are, naturally, the same forall three points because of equilibrium requirements. Fig.15(b) shows the variations in time of the macrostructuraland microstructural specific volume. There is a monotonicincrease of microstructural porosity (or specific volume), thestart of which is staggered in time as the various pointsrespond to the different arrivals of the hydration front. Thechanges in macrostructural porosity are more complex. Atthe bottom of the specimen, the macrostructural porosityincreases slightly at the beginning of the test because ofsuction reduction. However, on reaching the LC yield sur-face (indicated in the figure), there is a sharp reduction ofmacrostructural porosity due to structural collapse. Thisresults in a translation of the LC yield surface towards theright, indicated by the increase in value of the saturatedpreconsolidation pressure, p�o (Fig. 15(c)). At the other twopoints, there is initially a reduction of the macrostructuralporosity due to the compression applied by the swelling of

Table 3. Parameters of the mechanical constitutive model

Constitutive law Expression Parameter Low-densitysamples

High-densitysamples

BBM

Elastic part

_�ev ¼ �

k1þ eM

d p

p� ks

1þ eM

dsM

sM þ patm

k 0.010 0.010ks 0.022 0.041

Yield locusp�o : MPa 0.14 0.38

p0 ¼ pc

p�0

pc

!(º(0)�k)=(º(s)�k)

pc: MPa 0.005 0.005r 0.65 0.72º(0) 0.23 0.20

º(s) ¼ º(0)[r þ (1� r)e��s ] �: MPa�1 0.025 0.035

BExM

Microstructural behaviour

Km ¼1þ em

km

ppkm 0.045

Interaction function (microstructural swelling)

f s ¼ f s0 þ f si 1� p

p0

� �ns f s0 �2.0f si 1ns 2

Table 4. Initial conditions of the swelling pressure tests

Test label Sample height:mm

Dry density ofsample: g/cm3

Total porosity,nT

Macro-porosity,nM

Micro-porosity,nm

Macrop. suction,sM : MPa

Microp. suction,sm: MPa

MGR7 50 1.45 0.46 0.33 0.13 212 285MGR9 100 1.45 0.46 0.33 0.13 284 321RS2J 50 1.45 0.46 0.33 0.13 298 343RS2B 50 1.60 0.40 0.16 0.24 207 237RS2E 100 1.60 0.40 0.16 0.24 284 314RS2F 120 1.60 0.40 0.16 0.24 284 314



the bottom part of the sample. Eventually, however, thepoints also reach the LC yield surface and collapse ensues.The final stage of reduction of macrostructural porosity isalso partially associated with the increase of vertical stresscaused by the swelling of the microstructure. In this finalstage, the evolution of porosities is the same in all threepoints, consistent with the homogeneity of the samplesobserved at the end of the tests.

Microstructural observationsDetermination of the pore size distribution could provide

a valuable independent confirmation of the capability of theformulation to represent fabric changes of the materialduring hydration. Unfortunately, the pore size distribution ofthe samples after hydration was determined, using the MIPtechnique, only in the case of specimen MGR9. Foursamples from different locations of the specimens werefreeze-dried prior to pore size determination. The relation-ships between the cumulative intrusion porosity and entrancepore size for the four samples are shown in Fig. 16(a). Itcan be observed that there is a large amount of non-intruded

porosity. Using these data, it is possible to derive the poresize density functions depicted in Fig. 16(b). By analogywith other materials (Wan et al., 1995; Monroy, 2005;Romero et al., 2005), it is likely that the large family ofpores in the range of 200–500 nm was not present in theinitial material but it is a consequence of the hydrationprocess. Those pores would be the product of a reduction insize of larger macropores to accommodate the increase ofmicropore volume due to hydration. A direct check is notpossible, however, as there were no MIP tests performed onthe mixture before the test.

The proportions of porosity belonging to different rangesof pore sizes are listed in Table 5. The non-intruded volumehas been ascribed to the ultra-microporosity smaller than6 nm. Maintaining the boundary between microporosity andmacroporosity at 100 nm, the final microstructural porosityfalls in the range of 49–62%. Alternatively, Delage &Lefebvre (1984) suggested that microporosity could be esti-mated from the trapped volumes after mercury extrusion. Inthe present case, this criterion yields microporosity percen-tages ranging from 45% to 61%, quite consistent with thepore size density determination.

�1·2

�0·8

�0·4

0

Sw

ellin

g pr

essu

re: M

Pa

0 50 100 150 200 250

Time: days(a)

Numerical results

MGR7

150

100

50

0

Wa

ter

inta

ke: c

m3

Numerical water intake

MGR7

0·01 0·1 1 10 100 1000Time: days

(d)

�1·2

�0·8

�0·4

0

Sw

ellin

g pr

essu

re: M

Pa

0 50 100 150 200 250Time: days

(b)

Numerical results

RS2J

150

100

50

0W

ate

r in

take

: cm

3


RS2J

0·01 0·1 1 10 100 1000

Time: days(e)

�1·2

�0·8

�0·4

0

Sw

ellin

g pr

essu

re: M

Pa

0 50 100 150 200 250Time: days

(c)

Numerical results

MGR9

150

100

50

0

Wa

ter

inta

ke: c

m3


MGR9

0·01 0·1 1 10 100 1000

Time: days(f)

Fig. 12. Observations compared with computed results for tests on samples of 1.45 g/cm3 dry density: (a)–(c) swellingpressure evolution for tests MGR7, RS2J and MGR9; (d)–(f) evolution of accumulated water intake for tests MGR7,RS2J and MGR9



In turn, the analyses predict that, in this test, microporos-ity increases from the initial value of 0.127 to 0.283 at theend of the test, whereas macroporosity decreases from 0.329to 0.173. Therefore, the final microporosity represents 62%of the total pore volume, a rather satisfactory agreementwith observations. The fact that the proportion of micropor-osity varies monotonically from the bottom to the top of thesample might suggest that, in spite of the long test durationand the constant values of macroscopic observations, thematerial was not completely saturated at the end of the testin the regions further away from the hydration boundary.This would imply the existence of a porosity level with amuch lower leakage coefficient. This possibility, however,

does not seem to prevent the model from achieving a gooddescription of the macroscopic test results.

Swelling tests at dry density 1.60 g/cm3

Three tests performed on samples with a dry density of1.60 g/cm3 (RS2B, RS2E and RS2F) are considered. Theirlengths are 5, 10 and 12 cm respectively. The estimatedinitial porosities and suctions are listed in Table 4. Com-puted evolutions of swelling pressures compared with themeasured values are shown in Figs 17(a)–17(c). Testingtimes ranged from 150 to 500 days depending on specimenlength. The overall behaviour of these tests is similar to thatof the looser samples but, in this case, the final swellingpressures are much higher and the swelling pressure dropafter the first peak is comparatively smaller. At any rate, themodel reproduces very satisfactorily the observed behaviour.The evolution of water intake is also well matched (Figs

300

200

100

0

Suc

tion:

MP

a

0·001 0·01 0·1 1 10 100 1000Time: days

(a)

yM 0·0975�

y 0·0500�M

y 0·000�M

ym 0·0975�

y 0·0500�m

y 0·000�m

300

200

100

0

Suc

tion:

MP

a

0·001 0·01 0·1 1 10 100 1000

Time: days(b)

y 0·0475�M

y 0·0250�M

y 0·000�M

y 0·0475�m

y 0·0250�m

y 0·000�m

Fig. 13. Computed evolution of macrostructural and microstruc-tural suctions at three different points of the sample: (a) testMGR9; (b) test MGR7

1000

100

10

1

0·1

Mac

rost

ruct

ural

suc

tion:

MP

a

0·1 1 10Vertical net stress: MPa

y 0·0975�

LC

y 0·0500�

y 0·000�

Fig. 14. Computed stress path (macrostructural suction–verticalstress) for three different points of sample MGR9. Only theinitial LC yield surface is shown for clarity

�1·2

�0·8

�0·4

0

Sw

ellin

g pr

essu

re: M

Pa

0·1 1 10 100 1000Time: days

(a)

y 0·0975�

y 0·0500�

y 0·000� LC MS�

LC MS�LC

LC

1·60

1·50

1·40

1·20S

peci

fic v

olum

e0·1 1 10 100 1000

Time: days(b)

y 0·0975�M

y 0·0500�M

y 0·000�M

LC MS�

LC MS�

LC

LC

1·30

ym 0·0975�

y 0·0500�m

y 0·000�m

1·0

0·8

0·4

0

Sa

tura

ted

prec

onso

lida

tion

pres

sure

: MP

a

0·1 1 10 100 1000Time: days

(c)

y 0·0975�

y 0·0500�

y 0·000�

LC MS�

LC MS�

LC

LC

0·2

0·6

Fig. 15. Computed evolution of: (a) swelling pressure; (b) macro-structural and microstructural specific volumes; (c) saturatedpreconsolidation pressure for three points of test MGR9. Arrowsindicate the point at which the various yield mechanisms areactivated



17(d)–17(f)); some discrepancies at the end of the test havebeen attributed to small leakages in some of the very long-term tests.

Some results of the 10 cm sample (RS2E) are nowpresented so that they can be directly compared with theresults of test MGR9. The evolutions of macrostructural andmicrostructural suctions at three different points (bottom,middle and top of the specimen) of test RS2F obtained inthe analysis are plotted in Fig. 18. They are similar to thoseof the looser samples (Fig. 13). In this case, establishingequilibrium between the suctions at the two porosities takesa little longer and the onset of collapse is also delayedsomewhat. The stress paths for the same three points arepresented in Fig. 19. The initial LC is now further to theright to account for a denser macrostructure. It is alsoapparent that the drop in vertical stress after reaching the

LC is now smaller if compared with the looser specimens(Fig. 14). Finally, Fig. 20 shows the computed evolution ofswelling pressure, macrostructural and microstructural speci-fic volumes, and saturated preconsolidation pressure for thesame three points. Again the same overall behaviour as intest MGR9 is observed but now the relative variation of themacro and micro specific volumes is more limited than inthe case of the tests on looser specimens. Also, in the finalstages of the test, the evolution of porosities is the same inall three points, corresponding to a homogenised conditionof the sample. The pattern of saturated preconsolidationpressure variation in time is again similar to that of thelooser specimens, but a much higher final value is reachedby the denser samples.

A LARGE-SCALE SEALING TESTDescription of the test

A large-scale sealing test (named the shaft sealing test)has been performed in the Hades underground researchfacility (URF) shown in Fig. 21. The URF is located in Mol(Belgium) and it has been excavated at a depth of 220 m inBoom Clay, an overconsolidated plastic clay of Rupelian age(28–34 million years). The sealing test has been performedin an experimental shaft located at the end of the main testdrift (Fig. 22(a)). To this end, the bottom part of the shaftwas filled with grout and the concrete lining was removed atthe location of the seal. The sealed section is about 2.2 m indiameter and 2.25 m high. The sealing material is a mixtureof 50% powder and 50% highly compacted pellets of FoCaclay, the same mixture as in the laboratory swelling pressuretests. The seal is kept in place with a top concrete lid about1 m thick (Fig. 22(b)). A large number of sensors measuringpore water pressure, total stress, displacement and relativehumidity were installed to follow the hydromechanical evo-lution of the seal and the surrounding host rock. In the seal,most instruments are located on six rods of stainless steelconnected to a central tube. Those rods are located in twogroups of three rods, each at two levels: the instrumentedlevel top (ILT) and the instrumented level bottom (ILB).The ILT is located at 180 cm and the ILB at 65 cm from thebottom of the plug. Some instrumentation is also located atthe hydration level top (HLT) and at the hydration levelbottom (HLB). The instrumented sections are indicated inFig. 22(b). Several filters inside the sealed section enabledartificial hydration to reduce the time required to attainsaturation.

After coating the concrete plug with a resin, the claymixture was installed using approximately 12 tons of pow-der/pellets mixture. The first 60 cm were compacted with avibro-compactor specially designed for the project. Aftercompaction the mixture had a dry density of 1.54 g/cm3.The rest of the mixture of powder and pellets was notcompacted in order to avoid damage of the sensors, resultingin a dry density of 1.39 g/cm3. A top view of the shaftduring backfill installation has been provided in Fig. 1.Some details of the installation are shown in Fig. 23.

0·5

0·4

0·3

0·2

0·1

0

Cum

ula

tive

intr

usio

n po

rosi

ty,n

100000 10000 1000 100 10 1

Entrance pore size, : nm(a)

d

MGR9 – 8·75 cmMGR9 – 6·25 cmMGR9 – 3·75 cmMGR9 – 1·25 cm

Extrusion

Theoretical porosity 0·456�

1·0

0·4

0·8

0·2

0·6

0

Por

e si

ze d

ensi

ty fu

nctio

n


(b)d

MGR9 – 8·75 cmMGR9 – 6·25 cmMGR9 – 3·75 cmMGR9 – 1·25 cm

Inter-aggregateporosity

Intra-aggregateporosity

Fig. 16. Porosimetry results for four samples taken at the end oftest MGR9: (a) cumulative intrusion porosities; (b) pore sizedensity functions

Table 5. Pore size distribution measured in four samples obtained from test MGR9 after hydration

Entrance pore size Percentage of total porosity

y ¼ 8.75 cm y ¼ 6.50 cm y ¼ 3.25 cm y ¼ 1.25 cm

Macroporosity .1000 nm ,21 ,20 ,19 ,17Mesoporosity 100–1000 nm ,30 ,30 ,28 ,21Microporosity 20–100 nm , 5 ,5 , 6 ,5Ultra-microporosity ,20 nm ,44 ,45 ,47 ,57



After backfilling the shaft and closing the seal, a 7-monthperiod was allowed to elapse to achieve steady-state condi-tions in the zone around the test. Afterwards, artificialhydration was applied during 6 years. Some leakage pro-

blems occurred in the early stages of the test and hydrationhad to be stopped and re-started on some occasions. Thewater injection pressure was increased gradually up to avalue of 300 kPa, approximately, measured at the elevation

�3·0

�2·0

�1·0

0

�4·0

�3·0

�2·0

�1·0

0

�4·0

200

200

�3·0

�2·0

�1·0

0

Sw

ellin

g pr

essu

re: M

Pa

0 50 100 150 200Time: days

(a)

Numerical results

RS2B

150

100

50

0

Wa

ter

inta

ke: c

m3


RS2B

0·01 0·1 1 10 100 1000Time: days

(d)

Sw

ellin

g pr

essu

re: M

Pa

0 50 100 150 200Time: days

(b)

Numerical results

RS2E

150

100

50

0

Wa

ter

inta

ke: c

m3


RS2E

0·01 0·1 1 10 100 1000

Time: days(e)

Sw

ellin

g pr

essu

re: M

Pa

0 50 100 150 200

Time: days(c)

Numerical results

RS2F

150

100

50

0

Wa

ter

inta

ke: c

m3


RS2F

0·01 0·1 1 10 100 1000

Time: days(f)

�4·0 200

Fig. 17. Observations compared with computed results for tests on samples of 1.60 g/cm3 dry density: (a)–(c) swellingpressure evolution for tests RS2B, RS2E and RS2F; (d)–(f) evolution of accumulated water intake for tests RS2B,RS2E and RS2F

300

200

100

0

Suc

tion:

MP

a

0·001 0·01 0·1 1 10 100 1000Time: days

yM 0·0975�

y 0·0975�my 0·0500�M

ym 0·0500�

y 0·000�M

y 0·000�m

Fig. 18. Computed evolution of macrostructural and microstruc-tural suctions at three different points of test RS2F

1000

100

10

1

0·1

Mac

rost

ruct

ural

suc

tion:

MP

a

0·01 0·1 1 10Vertical net stress: MPa

y 0·1175�

y 0·0600�

y 0·000�

LC

Fig. 19. Computed stress path (macrostructural suction–verticalstress) for three different points of sample RS2E. Only theinitial LC yield surface is shown for clarity



of the main drift. After seal saturation, various gas andliquid permeability tests have been performed but only thehydration stage is considered in this paper.

Seal hydration and modellingThe hydration of the seal has been simulated using the

theoretical formulation outlined above. The finite-elementmesh used in the coupled hydromechanical analysis isdepicted in Fig. 24 together with the main boundary condi-tions. Advantage has been taken of the basically axisym-metric nature of the test. The initial characteristics of thepellets–powder mixture installed in the shaft are listed inTable 6. The hydraulic and mechanical parameters of thebackfill material are the same as those used in the analysesof the swelling pressure tests (Tables 2 and 3). The chal-lenge is to be able to cover a large-scale span (fromlaboratory to field conditions) using the same formulation,

conceptual model and parameters. The estimated partition ofporosities and the initial suction values are given in Table 7.As the natural Boom Clay parameters have only a limitedimpact on results, they are not given here. Boom Clayparameters and the procedures used for their determinationare reported in Gens et al. (2009a). Some selected resultsconcerning suctions and stresses in the sealing material arenow presented.

Figure 25(a) shows the variation in time of the suctionmeasured in the relative humidity sensors located on therods of the ILT, an instrumentation level located in theuncompacted part of the backfill. There are only two sensorsbecause the one oriented towards the west failed from thestart. The sensor closer to the rock (RH-S-ILT-E) appearedto hydrate only a little faster than the one placed near thecentre (RH-S-ILT-N), suggesting that the natural hydrationfrom the host Boom Clay is not dominant. The computedsuction evolutions are quite close to observations in thiscase and the analysis also shows a somewhat faster hydra-tion in the sensor closer to the rock boundary. The readingsof the relative humidity sensors located at the ILB areplotted in Fig. 25(b). In this case, the instrumented level issituated in the compacted backfill but very close to theboundary between the two zones. Now, there are largedifferences between the readings of the different sensors,significantly larger than those obtained in the modelling.The analysis also suggests that hydration should be slowerat this level because of the lower permeability of thecompacted material. This is borne out by the observationsexcept for the sensor placed quite close to the centralinjection tube (RH-S-ILB-W). It is strongly suspected that adirect connection between hydration tube and the zone ofthe sensor was established during one of the injectionpressure increase episodes.

The evolutions of vertical stresses at 16 points located atdifferent heights of the seal are shown in Fig. 26 andcompared with the results of the analysis. From top tobottom they are located in (a) hydration level top, (b)instrumentation level top, (c) instrumentation level bottomand (d) hydration level bottom. The usual difficulties asso-ciated with trying to measure stresses precisely are apparentin the plots. It can also be noted that stress development isnot axisymmetric; departures from that condition vary de-pending on the instrumentation level considered. The re-sponse of the sensors does not appear to be simultaneouswith hydration; this may be perhaps attributed in part to alack of close contact with the backfill after installation. Inspite of those difficulties, the analysis appears to givecorrectly the rough magnitude of the hydration swellingpressure as well as the rate of stress increase in the laterstages of the test. Naturally, the axisymmetric analysis doesnot discriminate between different directions. It should bestressed that the same set of parameters used in the model-ling of the laboratory tests has been adopted for the analysisof the shaft sealing test.

It is interesting to observe that the model only predicts arather muted drop in the value of the vertical stress in thesecond year of the test. Although the density of the backfillis low, the effect of macrostructural collapse is minorbecause it is averaged out among different zones. This is aconsequence of the very different scale of this test ascompared with the laboratory ones. At small scales, differentstages of macrostructural collapse may be occurring at thesame time in practically the entire sample, leading to a largedrop in swelling stress. At the scale of a field case, thissimultaneity is more unlikely and the collapse in some zonesis at least partially compensated by the non-collapsing be-haviour of other zones. It should be noted that the testobservations do not record any instance of sharp vertical

�1·2

�0·8

�0·4

0

Sw

ellin

g pr

essu

re: M

Pa

0·1 1 10 100 1000Time: days

(a)

y 0·0975�

y 0·0500�

y 0·000� LC MS�

LC MS�

LC

LC

1·60

1·50

1·40

1·20

Spe

cific

vol

ume

0·1 1 10 100 1000Time: days

(b)

yM 0·0975�

y � 0·0500M

yM 0·000�LC MS�

LC MS�

LC

LC

1·30

ym 0·0975�

y 0·0500�m

ym 0·000�

4·0

3·0

2·0

0

Sa

tura

ted

prec

onso

lida

tion

pres

sure

: MP

a

0·1 1 10 100 1000Time: days

(c)

y 0·0975�

y 0·0500�

y 0·000�

LC MS�

LC MS�

LC

LC

1·0

MS

1·10

MS

MS

Fig. 20. Computed evolution of: (a) swelling pressure; (b) macro-structural and microstructural specific volumes; (c) saturatedpreconsolidation pressure for three points of test RS2E. Arrowsindicate the point at which the various yield mechanisms areactivated



stress reduction, although some slowing down in the rate ofincrease (or even some local stress reductions) can be notedin some sensors during the second year of testing. Inaddition, analysis results and observations largely agree inthe steady increase of vertical stress at the end of the test.

CONCLUDING REMARKSThe use of compacted mixtures of bentonite powder and

bentonite pellets as backfill material in underground reposi-tories for nuclear waste is an attractive proposition becauseit reduces significantly the compaction effort that is required

Connecting gallery2001–2002

PRACLAY gallery2007

Test drift1987

First shaft1980–1982

URL1982–1983

Experimental works1983–1984

Second shaft1997–1999

Fig. 21. The Hades underground research facility: the experimental shaft where the sealing test was installed isindicated

Hades URLmain gallery

15 m

19·7

m

Verticalpiezometer

Concreteseal

Bentoniteseal

Radialpiezometers ( 3)�

Displacementsensors ( 3)�

Shaft lining

Boom Clay

HLT

ILT

ILB

HLB

Anchors

Concrete seal

Resin layer

Sand

FoCa clay

Central tube

Flange

Hydration level

ResinConcrete

Instrumented level

Filterplate

Host rock instrumentation

∅140 cm

∅220 cm

�12·90 m/345 cm

�13·95 m/240 cm

�14·11 m/224 cm

187·5 cm

�15 m/135 cm

72·5 cm

�16·35 m/0 cm

Fig. 22. (a) Schematic overview of the location of the seal inside the experimental shaft. (b) Layout of the test includinginstrumentation levels. HLT: hydration level top; ILT: instrumentation level top; ILB: instrumentation level bottom; HLB: hydrationlevel bottom



to achieve the value of dry density needed to provide anadequate swelling potential on hydration. However, the het-erogeneous fabric of the material results in a quite complexbehaviour during hydration.

In the paper, it has been demonstrated that a doubleporosity approach, involving both hydraulic and mechanicalaspects, is able to provide the necessary theoretical frame-work to undertake satisfactorily the hydromechanical model-

(a) (b)

Fig. 23. (a) Rod instrumented with total stress, water pressure, displacement and suction sensors. (b) Installation of a piezometer inthe host medium

σX

l

4·4 MPa2·2 MPa

� ��P

Pl

0·1

MP

a�

CLHydrationlevel top

2·24

Hydrationlevel middle

1·13

Hydrationlevel bottom

0·00

Pl

0·1

MP

a�

Liquid injection pointConstant liquid pressure

50 m

Fig. 24. Axisymmetric finite-element mesh and boundary conditions used in the analysis of the shaft sealing test

Table 6. Main characteristic of the pellets/powder mixture installed in the experimental shaft

Height: cm Void ratio ofmixture

Dry density ofmixture: g/cm3

Initial watercontent ofmixture: %

Dry density ofpellets: g/cm3

Initial watercontent ofpellets: %

Dry density ofpowder: g/cm3

Initial watercontent ofpowder: %

Compacted zone

60 0.734 1.543 6.00 1.894 4.49 1.302 7.51

Uncompacted zone

164 0.929 1.387 6.00 1.894 4.49 1.093 7.51



ling of this material. The formulation (and associated com-puter code) results in close quantitative reproduction of anumber of swelling pressure tests performed on samples ofdifferent densities and lengths. In addition, a detailed exam-ination of the computational results provides valuable in-sights and understanding of the basic processes that underliethe observed macroscopic behaviour. Moreover, microfabricinformation using X-ray tomography and MIP provide addi-tional support for the conceptual approach adopted.

Using the same parameters as in the modelling of labora-tory tests, the application of the formulation to a field-scalelong-term sealing test performed in the Hades URF hasresulted in a quite adequate description of the main behav-iour features of the backfill during hydration. The capabilityof the formulation to span over quite large ranges of spaceand time scales is thus established. In addition, because theformulation is based, albeit roughly, on the actual composi-tion of the material, it is capable of providing importantinsights that are key in the building up of confidence in theeventual repository design. In fact, the conceptual approach

developed in the context of the research reported in thispaper is quite general and it can be readily applied to othertypes of heterogeneous compacted soils. A better under-standing of their hydromechanical behaviour should therebyensue.

ACKNOWLEDGEMENTSThe work reported has been co-funded by ANDRA,

CIEMAT, ONDRAF-NIRAS and the European Commission(EC contract FIKW-CT-2000-00010). The support of theSpanish Ministry of Science and Innovation through grantBIA 2008-06537 is also gratefully acknowledged.

APPENDIXA detailed description of the BBM model can be found in Alonso

et al. (1990). Here, the more relevant equations of the particularversion of the BBM used are presented. The model is defined interms of the following stress invariants

p ¼ 1

3� x þ � y þ � zð Þ (16)

J2 ¼ 0:5 trace s2ð Þ (17)

Ł ¼ � 1

3sin�1 1:5

ffiffiffi3p

det s=J 3 �

(18)

s ¼ � � pI (19)

� ¼ �t � I pf (20)

where pf ¼ pg if pg . pl, otherwise pf ¼ pl. pg is the gas pressureand pl is the liquid pressure. I is the identity tensor.

The LC yield surface (FLC) can be expressed as

FLC ¼ 3J 2 � g Łð Þg �30ð Þ

" #2

M2 pþ psð Þ po � pð Þ ¼ 0 (21)

where M is the slope of the critical state, po is the apparentunsaturated isotropic preconsolidation pressure, g is a function ofLode’s angle and ps considers the dependence of shear strength onsuction. The following relationships complete the definition of theyield surface

ps ¼ ksM (22)

p0 ¼ pc

p�0pc

! º 0ð Þ�kð Þ= º sð Þ�kð Þ(23)

where k is a parameter that accounts for the increase of shearstrength with suction, pc is a reference stress, k is the elasticstiffness parameter for changes in net mean stress, p�0 is thepreconsolidation net mean stress for saturated conditions. º(s) is thecompressibility parameter for changes in net mean stress for virginstates of the soil that depends on macrostructural suction, sM ,according to

º sð Þ ¼ º 0ð Þ r þ 1� rð Þe��sM

�(24)

where r is a parameter which defines the minimum soilcompressibility (at infinity suction) and � is a parameter whichcontrols the rate of decrease of soil compressibility with macro-structural suction.

The hardening law is expressed as a rate relation between

Table 7. Initial conditions of the pellets–powder mixture used in the shaft sealing test

Dry density ofmixture: g/cm3

Total porosity,nT

Macrostructuralporosity, nM

Microstructuralporosity, nm

Macrostructuralsuction, sM : MPa

Microstructuralsuction, sm: MPa

Compacted seal 1.543 0.423 0.278 0.145 272 279Uncompacted seal 1.387 0.482 0.366 0.116 280 308

1000

100

10

1

0

Suc

tion:

MP

a

9/99

3/00

9/00

3/01

9/01

3/02

9/02

3/03

9/03

3/04

9/04

3/05

9/05

3/06

9/06

3/07

Time: days(a)

Beginning ofartificial

hydrationon the shalt

End ofartificial

hydration

Measured RH–S–ILT–EPredicted ILT–EMeasured RH–S–ILT–NPredicted ILT–N

1000

100

10

1

0

Suc

tion:

MP

a

9/99

3/00

9/00

3/01

9/01

3/02

9/02

3/03

9/03

3/04

9/04

3/05

9/05

3/06

9/06

3/07

Time: days(b)

Beginningof artificialhydration

End ofartificial

hydration

Measured RH–S–ILB–WPredicted ILB–WMeasured RH–S–ILB–NPredicted ILB–N

ILT–N

0·49 ILT–E

0·74

ILB–N

0·49

ILB–E

0·74

ILB–W

0·24

Measured RH–S–ILB–EPredicted ILB–E

Fig. 25. Variation of suction with time: computed results andobservations: (a) instrumented level top; (b) instrumented levelbottom



volumetric plastic strain and the saturated isotropic preconsolidationstress, p�0 , as

_p�0p�0¼ 1þ eMð Þ

º 0ð Þ � k� � _� p

vM (25)

where eM is the macrostructural void ratio, k is the elasticcompression index for changes in net mean stress and º(0) is thecompression index for changes in net mean stress for virgin states ofthe soil in saturated conditions.

Regarding the direction of the plastic strain increment, a non-associated flow rule in the plane sM ¼ constant

G ¼ Æ3J2 � g Łð Þg �30ð Þ

" #2

M2 pþ psð Þ p0 � pð Þ ¼ 0 (26)

where Æ is determined from the condition that the flow rule predictszero lateral strains in a K0 stress path.

Finally, the macrostructural elastic volumetric strains are com-puted as

_�evM ¼ �

k1þ eM

d p

p� ks

1þ eM

dsM

sM þ patm

(27)

where k and ks are model parameters and patm is the atmosphericpressure (100 kPa).

REFERENCESAifantis, E. (1980). On the problem of diffusion in solids. Actha

Mechanica 37, No. 3–4, 65–296.Alonso, E. E., Gens, A & Josa, A. (1990). A constitutive model for

partially saturated soils, Geotechnique 40, No. 3, 405–430, doi:1680/geot.1990.40.3.405.

Barrenblatt, G., Zeltov, I. & Kochina, N. (1960). Basic concepts in

the theory of seepage of homogeneous liquids in fissured rocks.Pirkl. Mater. Mekh. 24, 852–864.

Callari, C. & Federico, F. (2000). FEM validation of a double-porosity elastic model for consolidation of structurally complexclayey soils. Int. J. Numer. Analyt. Methods Geomech. 24, No.4, 367–402.

Castellanos, E., Villar, M. V., Romero, E., Lloret, A. & Gens, A.(2008). Chemical impact on the hydro-mechanical behaviour ofhigh density FEBEX bentonite. Phys. Chem. Earth 33, Suppl. 1,S516–S526.

Cui, Y., Yahia-Aissa, M. & Delage, P. (2002). A model for thevolume change behaviour of heavily compacted swelling clays.Engng Geol. 64, No. 2–3, 233–250.

Delage, P. & Lefebvre, G. (1984). Study of the structure of asensitive Champlain clay and its evolution. Can. Geotech. J. 21,No. 1, 21–35.

Delage, P., Howat, M. D. & Cui, Y. (1998). The relationshipbetween suction and swelling properties in a heavily compactedunsaturated clay. Engng Geol. 50, No. 1–2, 31–48.

Gens, A. (2003). The role of geotechnical engineering for nuclearenergy utilisation. Proc. 13th Eur. Conf. Soil Mech. Geotech.Engng, Prague 3, 25–67.

Gens, A. (2010). Soil–environment interactions in geotechnicalengineering. 47th Rankine Lecture. Geotechnique 60, No. 1,3–74, doi: 10.1680/geot.9.P.109.


Gens, A., Alonso, E. E., Lloret, A. & Batlle, F. (1993). Predictionof long term swelling of expansive soft rock: a double-structureapproach. In Geotechnical engineering of hard soils–soft rocks(eds A. Anagnostopoulos, F. Schlosser, N. Kalteziotis and R.Frank), Vol. 1, pp. 495–500. Rotterdam: Balkema.

Gens, A., Garcıa-Molina, A. J., Olivella, S., Alonso, E. E. &

20

16

8

4

0

Ver

tical

str

ess:

bar

9/99

3/00

9/00

3/01

9/01

3/02

9/02

3/03

9/03

3/04

9/04

3/05

9/05

3/06

9/06

3/07

Time: days(a)

Beginning ofartificial hydration

End ofartificial hydration

Measured PT–S–HLT–W (1)

Measured PT–S–HLT–E (2)

Measured PT–S–HLT–S (4)

Predicted HLT (0·71, 2·24)

9/99

3/00

9/00

3/01

9/01

3/02

9/02

3/03

9/03

3/04

9/04

3/05

9/05

3/06

9/06

3/07

Time: days(b)


End ofartificial hydration

12

(1)(3)

(2)(4)

Measured PT–S–HLT–N (3)

20

16

8

4

0

Ver

tical

str

ess:

bar

12

Measured PT–S–ILT–N–02

Measured PT–S– LT–WI –02

Measured PT–S–ILT–E–02

Predicted ILT–2 (0·88, 1·875)

1 234

12

34

1 32 4

PW–SW

PW

–SW

PW

–SW

20

16

8

4

0

Ver

tical

str

ess:

bar

9/99

3/00

9/00

3/01

9/01

3/02

9/02

3/03

9/03

3/04

9/04

3/05

9/05

3/06

9/06

3/07

Time: days(d)


End of artificialhydration

Measured PT–S–HLB–WMeasured PT–S–HLB–E

Measured PT–S–HLB–S

Predicted HLB (0·70, 0)12

WNW

E

SE

Measured PT–S–HLB–N

9/99

3/00

9/00

3/01

9/01

3/02

9/02

3/03

9/03

3/04

9/04

3/05

9/05

3/06

9/06

3/07

Time: days(c)


End ofartificial

hydration

20

16

8

4

0

Ver

tical

str

ess:

bar

12

Measured PT–S–ILB–N–04

Measured PT–S– L –WI B –04

Measured PT–S–ILB–E–04

Predicted IL –4 (0·445, 0·725)B

1 234

12

34

1 32 4

Fig. 26. Variation of vertical stresses with time: computed results and observations: (a) hydration level top, r 0.71 m;(b) instrumented level top, r 0.88 m; (c) instrumented level bottom, r 0.445 m; (d) hydration level bottom, r 0.70 m(r is the radial coordinate)



Huertas, F. (1998). Analysis of a full scale in situ test simulatingrepository conditions. Int. J. Numer. Analyt. Methods Geomech.22, No. 7, 515–548.

Gens, A., Filippi, M., Vallejan, B. et al. (2009a). RESEAL II project– Final report on modelling (WP4), SCK–CEN report ER-80.Brussels: SCK–CEN (Studiecentrum voor Kernenergie–Centred’Etude de l’energie Nucleaire (Belgian Nuclear ResearchCentre)).

Gens, A., Sanchez, M., Guimaraes, L. do N. et al. (2009b). A full-scale in situ heating test for high-level nuclear waste disposal:observations, analysis and interpretation. Geotechnique 59, No.4, 377–399, doi: 10.1680/geot.2009.59.4.377.

Ghafouri, H. & Lewis, R. (1996). A finite element double porositymodel for heterogeneous deformable porous media. Int. J.Numer. Analyt. Methods Geomech 20, No. 11, 831–844.

Guimaraes, L. do N., Gens, A. & Olivella, S. (2007). Coupledthermo-hydro-mechanical and chemical analysis of expansiveclay subject to heating and hydration. Transp. Porous Media 66,No. 3, 341–372.

IAEA (International Atomic Energy Agency) (1990). Sealing ofunderground repositories for radioactive wastes, IAEA TechnicalReport Series No. 319. Vienna: IAEA.

Imbert, V. & Villar, M. V. (2006). Hydro-mechanical response of abentonite pellets/powder mixture upon infiltration. Appl. ClaySci. 32, No. 3–4, 197–209.

Imbert, C., Fleureau, J.-M., Kheirbek-Saoud, S. & Deprez, L.(1997). Caracterisation hydrique des materiaux de l’experiencePRACLAY. Fontenay-aux-roses, France: Commissariat a l’energieatomique (CEA), Technical Report (in French).

Khalili, N., Valliappan, S. & Wan, C. (1999). Consolidation offissured clay. Geotechnique 49, No. 1, 75–89, doi: 10.1680/geot.1999.49.1.75.

Lloret, A., Villar, M. V., Sanchez, M. et al. (2003). Mechanicalbehaviour of heavily compacted bentonite under high suctionchanges. Geotechnique 53, No. 1, 27–40, doi: 10.1680/geot.2003.53.1.27.

Marcial, D. (2003). Comportement hydromecanique et microstruc-tural des materiaux de barriere ouvragee. PhD thesis, EcoleNationale des Ponts et Chaussees, France.

Michaux, L. & Loubignac, J. M. (1995). Evolution de lamicrostructure de l’argile Fo-Ca-7 compactee lors d’un cycled’hydration/deshydration a 258C. Fontenay-aux-roses, France:Commissariat a l’energie atomique, Technical Report (inFrench).

Monroy, R. (2005). The influence of load and suction changes onthe volumetric behaviour of compacted London Clay. PhD thesis,Imperial College, London.

Olchitzky, E. (2002). Couplage hydromecanique et permeabilited’une argile gonflante non saturee sous sollicitations hydriqueset thermiques: courte de sorption et permeabilite a l’eau. PhDthesis, Ecole Nationale des Ponts et Chaussees, Marne LaVallee, Paris, France (in French).

Olivella, S., Gens, A., Carrera, J. & Alonso, E. E. (1996). Numer-

ical formulation for a simulator (CODE_BRIGHT) for thecoupled analysis of saline media. Engng Comput. 13, No. 7,87–112.

Pusch, R. (1994). Waste disposal in rock, Developments in Geo-technical Engineering, Vol. 76. Amsterdam: Elsevier.

Romero, E. (1999). Characterization and thermo-hydro-mechanicalbehaviour of unsaturated Boom clay: an experimental study.PhD thesis, Universitat Politecnica de Catalunya, Spain.

Romero, E., Hoffmann, C., Castellanos, E., Suriol, J. & Lloret, A.(2005). Microstructural changes of compacted bentonite inducedby hydro-mechanical actions. In Advances in understandingengineered clay barriers (eds E. E. Alonso and A. Ledesma),pp. 193–202. Rotterdam: Balkema.

Sanchez, M. (2004). Thermo-hydro-mechanical coupled analysis inlow permeability media. PhD thesis, Universitat Politecnica deCatalunya, Spain.

Sanchez, M., Gens, A., Guimaraes, L. do N. & Olivella, S. (2005).A double structure generalized plasticity model for expansivematerials. Int. J. Numer. Analyt. Methods Geomech. 29, No. 8,751–787.

Touret, O., Pons, C., Tessier, D. & Tardy, Y. (1990). Etude de larepartition de l’eau dans des argiles saturees Mg2þ aux fortesteneurs en eau. Clay Minerals 25, No. 2, 217–223.

Van Geet, M., Volckaert, G. & Roels, S. (2005). The use ofmicrofocus X-ray computed tomography in characterising thehydration of a clay pellet/powder mixture. Appl. Clay Sci. 29,No. 2, 73–87.

Van Genuchten, M. T. (1980). A closed-form equation for predict-ing the hydraulic conductivity of unsaturated soils. Soil Sci. Soc.Am. J. 44, 892–898.

Volckaert, G., Bernier, F., Alonso, E. et al. (1996). Thermal-hydraulic-mechanical and geochemical behaviour of the claybarrier in radioactive waste repositories (model developmentand validation), Nuclear Science and Technology, EUR16744. Luxembourg: Commission of the European Commu-nities.

Volckaert, G., Dereeper, B., Put, M. et al. (2000). A large-scale insitu demonstration test for repository sealing in an argillaceoushost rock RESEAL project – phase I, EUR 19612, 273 pp.Luxembourg: European Commission.

Wan, A. W. L., Gray, M. N. & Graham, J. (1995). On the relationsof suction, moisture content and soil structure in compactedclays. Proc. 1st Int. Conf. Unsaturated Soils, Paris 1, 215–222.Rotterdam: Balkema.

Warren, J. & Root, P. (1963). The behaviour of naturally fracturatedreservoirs. Soc. Petrol. Engrs J. 228, No. 8, 244–255.

Wilson, R. & Aifantis, E. (1982). On the theory of consolidationwith double porosity. Int. J. Engng Sci. 20, No. 9, 1019–1035.

Yahia-Aıssa, M., Delage, P. & Cui, Y. J. (2000). Volumechange behaviour of a dense compacted swelling clay understress and suction changes. In Experimental evidence and theor-etical approaches in unsaturated soils (eds A. Tarantino andC. Mancuso), pp. 65–74. Rotterdam: Balkema.



Alonso, E. E. et al. (2011). Geotechnique 61, No. 5, 387–407 [doi: 10.1680/geot.SIP11.P.013]

161

Modelling the response of Lechago earth and rockfill dam

E. E. ALONSO�, S . OLIVELLA�, A. SORIANOy, N. M. PINYOL� and F. ESTEBAN{

Lechago dam (Teruel, Spain) is a 40 m high zoned earthand rockfill dam sitting on soft continental deltaic depos-its. A relatively narrow central clay core is stabilised bywide rockfill shoulders. The dam was well instrumentedand continuous records of stress development, pore-waterpressures and vertical displacements are available for theconstruction period. Compaction conditions were followedby means of laboratory and in situ control tests. Coreclay material was investigated by means of tests per-formed on compacted specimens of tertiary clays. Rockfillsamples were excavated in outcrops of highly fracturedCambrian quartzitic shale. A testing programme oncompacted rockfill gravels was conducted under relativehumidity control in a large-diameter oedometer andtriaxial cells. A coupled finite-element model has beendeveloped to analyse the tests performed and dam behav-iour during construction. Model predictions, essentiallybased on laboratory tests, are compared with measure-ments during construction. The predicted response of thedam under an assumed programme of impounding is alsogiven. In the future, once impounding occurs, it will bepossible to compare these predictions with actual damperformance. The paper provides an integrated descrip-tion of the dam design, construction and early behaviour.It presents a procedure to interpret available data (la-boratory as well as in situ data) on compacted materialsfrom the perspective of modern constitutive models. Italso provides an evaluation of the capabilities of ad-vanced numerical tools to reproduce the measured dambehaviour.

KEYWORDS: case history; compaction; dams; field instrumen-tation; numerical modelling

Le barrage de Lechago (a Teruel, en Espagne) est unstructure en terre a zones et enrochements, placee surdes depots deltaıques continentaux tendres. Un noyaud’argile central relativement etroit est stabilise par desepaulements a enrochements larges. Le barrage avait etebien instrumente, et on dispose de releves continus dedeveloppement de contraintes, pressions d’eau intersti-tielles et deplacements verticaux effectues au cours de laperiode de construction. On a suivi les conditions decompactage au moyen de tests de controle en laboratoireet « in situ », et examine le noyau d’argile au moyend’essais effectues sur des specimens compactes d’argilestertiaires. Des echantillons d’enrochement ont ete excavesdans des affleurements de schistes quartzites du cam-brien, extremement fractures. On a mene un programmed’essais sur des graviers d’enrochement, avec humiditerelative controlee, dans un oedometre de grand diametreet des cellules triaxiales. On a realise un modele accoupleaux elements finis pour analyser les tests effectues et lecomportement du barrage au cours de la construction.On a effectue une comparaison des predictions de mod-ele, basees essentiellement sur des essais en laboratoire,avec des mesures effectuees au cours de la construction.En outre, la reaction prevue du barrage soumis a unprogramme de retenue suppose est egalement fournie.Dans l’avenir, lorsque l’on procedera a une retenue, ilsera possible de comparer ces predictions avec les perfor-mances effectives du barrage. La communication presenteune description integree de la conception du barrage, desa construction, et de son comportement initial, ainsiqu’une methode d’interpretation des donnees disponibles(donnees relevees en laboratoire et « in situ ») sur desmatieres compactees, sous la perspective de modeles con-stitutifs evolues. Elle fournit egalement une evaluationdes capacites des outils numeriques perfectionnes pour lareproduction du comportement mesure du barrage.

INTRODUCTIONLechago dam, a 40 m high earth and rockfill structure (Fig.1) was built in the period from April 2005 to January 2009.The first impoundment has been delayed to the first monthsof 2011. The flat upstream and downstream average slopesreveal the soft nature of the foundation soil in the centralpart of the dam. This situation was of special concern duringdesign because it implied large total and differential settle-ments caused by the non-symmetrical cross-section of thevalley and the stiffness of abutments, located in ancientshales and quarzitic rocks.

The design called for wide stabilising rockfill shells and acentral clayey core. Fig. 2 is a photograph of the dam core,

filter layer and the rockfill shoulder (shell) materials beingcompacted in situ. The maximum size of the gravel-likerockfill is about 20 cm but the mean particle diameter isclose to 30–50 mm and therefore it may directly be testedin large-diameter laboratory cells, after removing the largestparticles. Fig. 3 shows the grain size distributions of the insitu and tested rockfill material. The dam material hadmaximum sizes close to 9 cm and a somewhat increasedcontent of fines, but otherwise the tested grading is quitesimilar to real conditions. Lechago rockfill material has beenextensively tested during the past decade in the soil mech-anics laboratory of Universitat Politecnica de Catalunya(UPC). The most salient feature of this experimental pro-gramme is the use of relative humidity (RH) control in theoedometer and triaxial tests performed. Some general resultsillustrating the effect of suction (or relative humidity) on therock matrix and rockfill aggregate are collected in Figs 4–6.Brazilian tests on discs of rock equilibrated at differentvalues of humidity were performed. The results (Fig. 4)show the variation of tensile strength with gravimetric watercontent or, alternatively, with total suction. Tensile strengthincreases the drier the specimen. Full wetting of an initially

Manuscript received 4 March 2010; revised manuscript accepted21 December 2010. Published online ahead of print 22 March 2011.Discussion on this paper closes on 1 October 2011, for furtherdetails see p. ii.� Department of Geotechnical Engineering and Geosciences.Universitat Politecnica de Catalunya, Barcelona, Spainy Universidad Politecnica de Madrid, Madrid, Spain{ Confederacion Hidrografica del Ebro, Zaragoza, Spain


dry rock reduces by half its tensile strength. The figureindicates the procedure to achieve a given humidity: equili-brating the rock specimen with salt solutions (indicated) ordrying the specimens at the laboratory relative humidity(50%) (air) or submerging them in water.

Figure 5 provides the change in oedometric stiffness withapplied total suction. It was found (Oldecop & Alonso,2007) that the vertical stress–strain relationship could beapproximated by a linear relationship

� ¼ º˜� (1)

provided � , 8%. Rockfill specimens were compacted to anenergy equivalent to standard Proctor energy. Compressibil-ity also reduces substantially when the rockfill evolves froma very dry state (total suction, �, equal to 255 MPa) to asaturated state.

The long-term deformation coefficient

º t ¼ d�

d ln tð Þ (2)

was also found to be linearly dependent on the compressi-bility coefficient º.

However, the ratio º t=º also depends markedly on currentsuction and it changes from 0.05 to 0.005 when suctionchanges from zero (full saturation) to � ¼ 255 MPa.

Figure 6 indicates the effect of RH on the strengthenvelope determined in a large-diameter triaxial cell onrockfill specimens compacted to standard Proctor energy(Chavez, 2004). The figure shows the expected curvature ofthe envelope. The effect of suction is in this case not sosignificant (RH ¼ 36% is equivalent to � ¼ 130 MPa at208C from the psychrometric relationship).

A number of publications describe the relevance of RH inthe behaviour of rockfill materials, the testing cells built and

Dam crest

Top soilRiver

Soft alluvial soil

900

880

860

840

820

Ele

vatio

n: m

Weathered shale

Intact shale

0 40 m

P-10

(a)

(b)

Drain Filter

895 Rockfill

8742·8:1

1·8:11·8:1

1·8:1 1·8:1

3·2:1869

856846

Weathered shale

0 25 m

Clay coreSoft deltaic soilGranular alluvium

860870

877

MNL 891

(c)

Fig. 1. (a) Valley cross-section; (b) plan view; (c) central cross-section P-10 of Lechago dam

162 ALONSO, OLIVELLA, SORIANO, PINYOL AND ESTEBAN


the experimental results (Oldecop & Alonso, 2001; 2003;2007; Chavez, 2004; Chavez & Alonso, 2003; Chavez et al.,2009).

Piezometers (vibrating wire type), hydraulic settlementgauges and total stress cells were located in five cross-sections of the dam. The most detailed instrumentationcorresponds to the central cross-section P-10, in the positionof the maximum thickness of the soft alluvial deposits underthe dam. Fig. 7 shows the instrument layout in section P-10.Surface topographic marks and levelling points were alsolocated on the dam crest and downstream shoulders once thedam was completed.

Data recorded during dam construction will be later com-pared with the results of the model developed. Modelparameters for the clay core and rockfill shoulders werederived from the analysis of tests performed on compactedclay from the core and the RH controlled triaxial testsreported in Chavez (2004). The foundation soil and, inparticular, the soft alluvial clay and sand sediments werecharacterised at the design stage of the dam in the period1998–1999.

The model has developed benefits from theoretical andapplied contributions in the field of unsaturated soil mech-anics performed by the authors and their colleagues in thepast few years. The finite-element program Code_Bright,

whose basic formulation is described by Olivella et al.(1994; 1996) and DIT-UPC (2002) has been used in avariety of applications, including dam and embankmentanalysis (Alonso et al., 2005) and nuclear waste applications(Olivella & Alonso, 2008; Gens et al., 2009), among others.It handles in a unified manner saturated and unsaturatedstates and transitions between them.

The saturated foundation soils were characterised by amodified Cam-clay model. The central clay core was de-scribed by the so-called Barcelona basic model (BBM) for

Fig. 2. Clay core, filter and rockfill shoulder of Lechago damduring construction

100

80

60

40

20

0

100 10 1 0·1 0·01Grain size: mm

Per

cent

fine

r by

wei

ght

In situ

Tested

Fig. 3. Grain size distribution of Lechago rockfill materialtested in the laboratory

Tens

ile s

tren

gth,

: MP

aσ t

Tens

ile s

tren

gth,

: MP

aσ t

1·6

1·6

1·4

1·4

1·0

1·0

0·8

0·8

0·6

0·6

0·4

0·4

1·2

1·2

Cl Mg2

Cl Mg2

Air

Air

Cl Ba2

Cl Ba2

SO K4 2

SO K4 2

SO K4 2

SO K4 2

H O2

H O2

H O2

H O2

Gravimetric water content, : %(a)

w0 0·5 1·0 1·5

0 1 10 100 1000Total suction, : MPa

(b)Ψ

Fig. 4. Results from Brazilian tests on samples (84 mm dia-meter; 50 mm thick) in terms of (a) gravimetric water content;(b) total suction (Oldecop, 2000)

5 10� �2

4 10� �2

3 10� �2

2 10� �2

1 10� �2

Coe

ffic

ient

of

com

pres

sibi

lity,

: MP

aλ

�1

00 100 200 300

Total suction, : MPaΨ

Fig. 5. Oedometric compressibility plotted against suction

MODELLING THE RESPONSE OF LECHAGO EARTH AND ROCKFILL DAM 163


unsaturated soils (Alonso et al., 1990). For rockfill shouldersand filters the isotropic elastoplastic model initially de-scribed by Oldecop & Alonso (2001) and later generalisedto include deviatoric states in Alonso et al. (2005) was used.

The paper describes first a few relevant features of thedam design and construction, the foundation soils and fieldcompaction data. Then, a back-analysis of large-diametertests on compacted rockfill specimens and conventional testson the compacted core clay is presented. Identified soilparameters are then used to simulate the dam construction.The model has been extended to simulate the first damimpoundment following a protocol that is often used inpractice. The latter part of the paper provides a prediction offuture dam performance, which may later be compared withactual dam performance.

DAM DESIGN AND CONSTRUCTIONThe design of Lechago dam faced a number of challen-

ging situations, namely the soft continental deltaic sedimentsfilling the bottom of the valley, the marked non-symmetry ofthe valley, the sharp transition between highly deformableand rigid substrata and the fractured and pervious shalesubstratum in the entire area.

Earth dam construction on soft soil deposits is not com-

mon but several cases have been described in publishedpapers (Daehn, 1985; Ramırez et al., 1991; Rizzoli, 1991;Tellerıa and Gomez Laa, 1991; Torner and Novosad, 1991;Trkeshdooz et al., 1991). Construction settlements in excessof 1 m have often been reported. In general, upstream anddownstream slopes are controlled by global stability consid-erations and values in the range 4H–5H/1V are often found.This is also the case for Lechago dam.

Lechago has two definite cross-sections. On both sides ofthe central alluvial deposits a conventional zoned dam hav-ing 3.2H/1V and 2.8H/1V slopes (upstream and downstream)was designed. On the valley bottom this design was stabi-lised by wide rockfill berms to improve stability. In Lechagothe project required a further improvement of the undrainedstrength of the soft alluvial deposits by means of a preload-ing operation of the downstream shoulders which would becarried out during construction. The actual constructioninvolved a smaller preloading intensity than originally de-signed, supplemented by a water table lowering. Both opera-tions were included in the model described below.

The actual construction sequence is plotted in Fig. 8 interms of the evolution in time of the embankment height. Inorder to minimise differential settlements the design envi-saged that the central section of the dam, directly foundedon the soft alluvium, would be first built and then extendedto occupy the valley slopes. However, in view of the fastconsolidation of the alluvial soils, it was decided to buildthe dam in horizontal layers.

The three intermediate stops of construction activityshown in Fig. 8 were motivated by the difficulties in build-ing the large spillway, excavated on the right abutment. Thedissipation basin in the lower part encountered deep softdeposits and the construction involved a larger-than-expectedexcavation dewatering by means of deep wells and a pilingfoundation of the base slab. The downstream dam shell wasaffected and therefore the entire dam construction schedulewas delayed. It was also decided to improve the foundationsoils by reducing the water table under the downstreamshell. The third and more extensive water lowering (thirdstop in Fig. 8) involved an average water table lowering of2.5 m. In addition, surcharge preloading was resolved by afill, 3 m high, emplaced on the downstream berms of thedam, at elevations of 869 and 874 m. Fig. 8 also indicatesthe measured piezometric level in the bedrock, below thedam downstream toe. Piezometric variations of about 3–4 mare a consequence of dewatering operations. The upstream

3·00

2·00

1·00

0

Dev

iato

ric s

tres

s: M

Pa

0 0·50 1·00 1·50 2·00

Net mean stress: MPa

RH 36%�

RH 100%�

Fig. 6. Effect of RH on the strength envelope of compacted rockspecimens in large-diameter triaxial cells

7 8

11 12 13

1 21 2 3

857·5

3

4

15

18

920

22 23885

1021

11 12

14 15 16 17 185 619 4

1617

652

3

1 4

Lower alluvium

7

8910

Shale

Deltaic soil

0 25 m

11

12

1314

Granular alluvium

864

Settlement gauge

Total pressure cell

Piezometer

876

9 108

Fig. 7. Instrument layout in cross-section P-10



cut-off wall allowed a limited impoundment (beginning inMarch, 2009) as well, which is also shown in the samefigure.

The dam reached its maximum elevation in March 2008,24 months after the placement of the first compacted layer.The first control of dam surface settlements was performedin April 2009.

FOUNDATION SOILS AND CONSTRUCTIONMATERIALS

The alluvium of the Pancrudo River was identified as asequence of three horizontally layered strata: an upper layerof pervious sandy silt and gravels, 5 m thick, an intermediatesoft clayey and silty stratum (deltaic deposits), 12 m thick,and a lower level of clayey sands and gravels 4 m thickdirectly over the shale substratum. In all levels thin se-quences of impervious/pervious soils explain the fast con-solidation of the entire alluvial stratum.

In the upper alluvium standard penetration test (SPT)values in the range 4–14 were measured. The central softclayey stratum was often classified as low-plasticity clay(CL) or low-plasticity silt (ML), but clayey sands were alsofound. Void ratios range between 0.6 and 0.75. Liquid limitand plasticity index take values in the range 25–40% and5–16% respectively. Critical stability conditions of the damslopes were found under undrained conditions for this layer,whose virgin compression indices Cc ¼ 0.15–0.18 weremeasured in standard oedometer tests. Horizontal consolida-tion coefficients in the range ch ¼ 0.6–6 cm2/min (0.01–0.1 cm2/s) were obtained in CPTU dissipation tests. How-ever, the frequent interlayering of sandy levels results insubstantially higher representative consolidation coefficients.In fact, settlements of the base of the dam during construc-tion increased in parallel with dam height. The lower thinalluvial sand was quite dense (N . 30).

The shale substratum was dissected by several faultswhich crossed the dam site. The upper decomposed andweathered shale level, 5–10 m thick, in contact with thealluvial or colluvial soils of the valley slopes, was moder-ately pervious (k � 10�6 m/s, determined through Lugeon insitu tests).

A low-plasticity clay of Miocene age was selected for thedam impervious core. Dry densities not lower than 102% ofthe standard Proctor optimum value and water contentsinside a � 2% band around optimum were initially specified

for construction. Fig. 9 provides a set of compaction curves,determined during the design stage, on Miocene clayeymaterials located in the vicinity of the dam. Standard Proctor(SP) and modified Proctor (MP) energies were used. Theselected clay for the core was the yellow clay. It reachedoptimum dry unit weights of 18 kN/m3 (SP) and 19 kN/m3

(MP). In all cases optimum conditions correspond reason-ably well to a 5% air content. The clay, after a period ofexposure to natural weathering, was compacted in 25 cmthick layers by means of six runs of a sheep foot roller anda final run of a smooth roller (when the nuclear probecontrol measurements were required). In all cases the con-tinuity of layers was ensured by a scarification of the currentcompacted surface prior to the placement of a new layer.The dry density limit was strictly enforced but water con-

900

890

880

870

860

850

Ele

vatio

n: m

0 3 6 9 12 3 6 9 12 3 6 9 12 3 6 92006 2007 2008 2009

Water elevation in bedrockDownstream toe of dam

Reservoir elevation

3rd dewatering

886

895

874

2nd dewatering

1st dewatering

857

Fig. 8. Construction sequence

γ γs w/ 2·70�

2·50

2·40

2·30

2·20

2·10

2·00

1·90

1·80

1·70

1·60

1·50

1·40

1·30

Dry

spe

cific

wei

ght,

/γ

γd

w

0 4 8 12 16 20 24 28 32 36 40Water content, : %w

Sr 0·8�

1

2

5

6

3

4

1·2%

1 (SP): Brown clay (CH)2 (MP): Brown clay (CH)3 (SP): Sandy and clayey

silt (SM-SC)4 (MP): Sand and clayey

silt (SM-SC)5 (SP): Yellow clay (CL)6 (MP): Yellow clay (CL)

Air content: 5%

6·4%

Fig. 9. Compaction tests on materials considered for the core ofLechago dam (SP: standard Proctor compaction; MP: modifiedProctor compaction)



tents were accepted in a wider range (�8% to +2% aroundoptimum). The average compaction water content remainedslightly below optimum.

Figure 10 provides the dry unit weight–water content insitu determinations during construction. Most of the dry unitweights remain in the range 18.7–19.7 kN/m3 (ec

0 � 0.38–0.45). These values may be compared with the laboratoryresults in Fig. 9. Field compaction in the core reacheddensities in the vicinity of the MP optimum. The plot in Fig.10 shows that the average degree of saturation duringconstruction was around 0.9, a value close to optimum. Asmall proportion of the core volume seems to have degreesof saturation in the range 0.6–0.8, on the dry side ofcompaction.

Oedometer tests for specimens compacted to the range ofªd values reached during construction provided the followingcompression indices: Cc ¼ 0.07–0.12; Cs ¼ 0.03–0.05.Measured coefficients of consolidation ranged between cv ¼1.5 3 10�2 cm2/s and cv ¼ 3.5 3 10�2 cm2/s. Derived perme-ability was in the range (k � 2 3 10�8 to 7 3 10�9 m/s). Themeasured secondary compression coefficient was a smallproportion of Cc (Cc

Æ � 0.013–0.015Cc).Collapse tests were also performed on several clay sam-

ples compacted to several dry specific weights and watercontents. Samples were taken to vertical compression stres-ses of 100 and 400 kPa and were flooded at constant stress.The plot in Fig. 11 shows the measured collapse strains. Theboundaries shown in the figure define three regions: verylow collapse, medium to low collapse and medium to highcollapse. The specimens tested cover a range of Mioceneclayey materials having different grain size distributions andplasticity even if most of them are classified as CL or ML.Note that in all cases the collapse increases substantiallywhen the vertical stress increases from 100 kPa to 400 kPa.The plot in Fig. 11 and the field density and water contentdata given in Fig. 10 indicate that the dam core has a lowcollapse potential (expected collapse strains smaller than 1%as a rough guide). This information was available at designstage and it was necessarily approximate because the objec-tive at the time was to select the appropriate core materialin the vicinity of the dam. More recently, tests were alsoperformed on samples of the core material. The results willbe analysed in the next section.

The quartzitic Cambrian shales used in the shells werehighly fractured and the combined effect of quarry excava-tion, transport and placement in layers (40–50 cm thick)resulted in a gravel-like material (maximum sizes of theorder of 10 cm). Therefore, compaction control tests couldbe performed in the laboratory in modified Proctor testmoulds, removing the bigger fragments when necessary. The

modified Proctor optimum was taken as a reference for fieldcontrol, which was carried out by means of nuclear probesthat were calibrated by direct density and water contentdeterminations at the beginning of construction works. Therockfill was compacted by a 19 ton vibratory roller (six runsfor each layer).

The measured dry unit weights and water contents arecollected in Fig. 12 in the conventional compaction plane.Water contents range between 6% and 7% for dry unit

Dry

wei

ght:

kN/m

3

20·0

19·6

19·2

18·8

18·4

18·0

5 10 15 20 25Water content: %

0·6

0·8

Sr 1�

Fig. 10. Compaction data on core clay; in situ determinations

5·31: Collapse strain in % under100 kPaσv �

(1·06): Collapse strain in % under400 kPaσv �

0·03

0·04

0·04

0·13

(2·24)

0·71

Low to mediumcollapse0·05

(1·06)

Medium tohigh collapse

Incr

easi

ng

Incr

easi

ng

5·31

–2·0 �1·5 �1·0 �0·5 0 0·5 1·0 1·5 2·0 2·5 3·0w w (SP): %� opt

15

16

17

18

19

20

21

Very smallcollapse

γd3: kN/m

0·26

(0·45)

Fig. 11. Oedometer collapse of specimens compacted to thespecified dry unit weight and water content (samples recoveredin trenches at the design stage of the project)

22·2

22·2

22·0

22·0

21·8

21·8

21·6

21·6

21·4

21·4

21·2

21·2

21·0

21·0

20·8

20·8

Dry

uni

t wei

ght:

kN/m

3D

ry u

nit w

eigh

t: kN

/m3

2

2

3

3

4

4

5

5

6

6

7

7

8

8

9

9

10

10

Water content: %(a)

Water content: %(b)

0·4

0·6

0·6

Sr 1�

Sr 1�

0·8

0·8

Fig. 12. Field compaction data for (a) upstream shell; (b) down-stream shell



weights in the range 21–22 kN/m3. The resulting degree ofsaturation varies between 0.6 and 0.8.

Oldecop & Alonso (2001) measured the water retentioncurve of Lechago shale gravel. They found that the porosityof the rock fragments amounts to 3.5–4%. Since the com-pacted rockfill has a global water content in the range 6–7%, if the rock fragments are saturated, some additionalwater partially fills the rockfill pores. This water is probablyassociated with the fines produced by the heavy compactionand/or by the insufficient drainage of the layer just com-pacted in situ at the time of the nuclear probe determina-tions.

The stress–strain behaviour of Lechago rockfill was testedin a RH-controlled large-diameter triaxial chamber, as de-scribed by Chavez (2004) and Chavez et al. (2009). Someisotropic and triaxial compression tests will be reproducedby the model developed in a next section.

TESTS ON COMPACTED MATERIALS AND THEIRINTERPRETATION

The available laboratory tests performed on the rockfillmaterial and the clay core have been simulated as a bound-ary value problem with the help of the finite-element codeCode_Bright with the purpose of determining material para-meters. Table 1 provides a synthetic description of themodels (BBM and rockfill model (RM)) used in the re-presentation of the different materials.

Rockfill materialTriaxial samples, 250 mm in diameter and 500 mm high,

were prepared by repeated compaction, applying an energyestimated at 600 J/m3. The maximum size of particles was40 mm. Samples were subjected to isotropic and triaxialloading at a specified relative humidity, which was con-trolled by imposing a vapour equilibrium technique (Oldecop& Alonso, 2001).

In order to simulate the tests performed, an axi-symmetric

model of the actual triaxial specimens, described by quad-rilateral linear elements, was developed. Stress, strain andflow conditions were applied at the specimen’s boundaries asrequired.

Tests are treated as coupled flow and deformation bound-ary value problems. Both liquid flow and vapour flow wereimposed to simulate the wetting of the samples followingthe test protocol in each case. The list of parametersrequired in the RM (Oldecop & Alonso, 2001) is indicatedin Table 2.

The water retention curve introduced in the calculations,shown in Fig. 13, was based on the curve reported byOldecop & Alonso (2001) for compacted specimens of thesame rockfill. Because of the larger proportion of fines insitu (see Fig. 3) the air entry value was moderately in-creased, and more water was allowed to be retained atmoderate suction values.

The water retention curve was simulated by means of avan Genuchten (1980) equation having a sub-horizontalshape for most of the accessible range of degrees of satura-tion. The smaller voids inside the rock particles retain waterfor low degrees of saturation, at high suction. The largestinter-particle voids provide the air entry value. Relativepermeability was defined by means of the cubic law accord-ing to the following expression

krel ¼ ksat Srð Þ3 (3)

where ksat is the coefficient of permeability for saturatedconditions and Sr is the degree of saturation.

The calibration of the remaining mechanical parametersindicated in Table 2 was obtained by a back-analysis of thelaboratory tests. A comparison of model predictions andmeasured sample responses is shown in Figs 14–17.

Figure 14 shows the measured and simulated response ofan isotropic test at constant suction performed in the triaxialapparatus. The saturated isotropic loading–unloading curve(Fig. 14(a)) shows the linear stress–strain relationship of therockfill-type material and allows the calibration of the

Table 1. Basic relationships for constitutive models used in the analysis of Lechago dam�

Barcelona basic model (BBM)(Alonso et al., 1990)

Rockfill model (RM)(Compressibility part described by Oldecop & Alonso, 2001)

Isotropic elastoplasticvolumetric deformationy d�v ¼

º(s)

1þ e

d p

p

For p < py ) d�v ¼ d�iv ¼ ºid p

For p . py ) d�v ¼ d�iv þ d�d

v

¼ [ºi þ ºd (s)] d p

Volumetric compressibilityindexy

º(s) ¼ º(0)[(1� r) exp (��s)þ r] ºi þ ºd(s)

ºd(s) ¼ ºd0 � Æsln

sþ patm

patm

� �

Hardening law d p�0 ¼(1þ e) p�0º(0)� k

d� pv d p�0 ¼

d� pv

ºi � k

Loading–collapse curve (LC) p0(s) ¼ pc p�0pc

� �[º(0)�k]=[º(s)�k]

For p�0 < py ) p0(s) ¼ p�0For p�0 . py ) p0(s) ¼ py þ

(ºi � k)( p�0 � py)

(ºi þ ºd(s)� k)

Shear strength critical stateparameter

M(s) ¼ M M(s) ¼ Mdry � (Mdry � Msat)Msat

Mdry

� �s=(10 patm)

Tensile strength parameter ps(s) ¼ kssYield surface (triaxial) F( p, q, s) ¼ q2 � M2[ pþ ps(s)][ p0(s)� p] ¼ 0Plastic potential (triaxial) G( p, q, s) ¼ q2 � ÆM2[ pþ ps(s)][ p0(s)� p] ¼ 0

�A common notation was used for equivalent parameters. Material constants are different for the soil and RMs.ySuction, s, refers to matric suction in the case of BBM, and to total suction in the case of RM.



Table 2. Mechanical parameters for rockfill and drains

Definition of parameter Symbol Units Base case for Lechago dam

Shell Drain

(a) Elastic behaviour

Elastic modulus E MPa 140 150Poisson’s ratio � — 0.235 0.3

(b) Plastic behaviour

Plastic virgin instantaneous compressibility (ºi � k) — 0.03 0.025Virgin clastic compressibility for saturated conditions ºd

0 — 0.03 0.028Parameter to describe the rate of change of clasticcompressibility with total suction

Æs — 0.0115 0.0115

Slope of critical state strength envelope for dryconditions

Mdry — 1.8 1.7

Slope of critical state strength envelope for saturatedconditions

Msat — 1.7 1.3

Parameter that controls the increase in cohesion withsuction

ks — 0 0

Threshold yield mean stress for the onset of clasticphenomena

py MPa 0 0.01

Parameter that defines the non-associativeness ofplastic potential

Æ — 0.25 0.3

(c) Initial state for dam model

Initial suction s0 MPa 0.001 0.001Initial mean yield stress (very dry conditions) p�0 MPa 0.25 0.25

1000

100

10

1

0·1

0·01

0·001

0·0001

0 0·2 0·4 0·6 0·8 1·0Degree of saturation

Rockfil model

Clay model

Clay laboratory data

Suc

tion:

MP

a

Fig. 13. Retention curves used in calculations for rockfillmaterial and clay core



0

0

0·2

0·2

0·4

0·4

0·6

0·6

0·8

0·8

1·0

1·0

1·2

1·2

Model

Model

Experimental

Experimental

Vol

umet

ric s

trai

n: %

Vol

umet

ric s

trai

n: %

0

0

2

2

4

4

6

6

8

8

(a)

(b)

Fig. 14. Isotropic test: (a) saturated; (b) dry loading(RH 36%) and saturation at a net mean stress of 0.6 MPa(experimental results by Chavez, 2004)



compressibility parameters (ºi and ºd0 ) and an estimation of

the elastic modulus from the unloading curve.In the second isotropic test (Fig. 14(b)), the sample was

initially loaded at a maintained RH ¼ 36%, which approxi-mately corresponds to a value of total suction of 145 MPaaccording to the psychrometric relationship (Coussy, 1995).Suction was imposed during the first loading and the netmean stress was increased to 0.6 MPa. At this stress levelthe specimen was saturated. In the simulated test, saturationwas reproduced by imposing a suction equal to zero at thetop of the sample. The resulting water and vapour flowinside the sample was calculated by the model. Saturation ofthe sample resulted in a collapse of 2% volumetric strain,which was satisfactorily modelled (Fig. 14(b)). Hydraulicconditions were maintained during the remaining loadingand unloading steps, which were performed under full drain-age.

The first part of the compression curve at an imposedsuction allows the estimation of the º(s) parameter, whichdescribes the variation of compressibility parameter withsuction and therefore the magnitude of collapse.

Figures 15 and 16 show the comparison between labora-tory data and calculated values for triaxial tests performed atsaturated condition and at 36% of RH respectively. The plotsshow the results for two confining stresses (0.1 and0.5 MPa). The slope of the critical-state envelope (M) andits change with applied suction, was calibrated by means ofthe response of the material. A small increment of strengthwith suction was recorded. This is reflected in the values ofMdry and Msat given in Table 2.

The non-associativeness of the plastic potential was esti-mated through the parameter Æ ¼ 0.25. However, the dilatantbehaviour was not well reproduced in all the simulated tests.The best approximation is obtained for the case of the

Model

Model

Experimental

Experimental

Confining stress 0·5 MPa�

0·5 MPa

0·1 MPa

0·1 MPa

0

0

5

5

10

10

15

15

20

20

25

25

Vertical strain: %(a)

Vertical strain: %

Dev

iato

ric s

tres

s,: M

Pa

σσ

13

�

1·6

1·2

0·8

0·4

0

Vol

umet

ric s

trai

n: %

0

2

4

6

8

10(b)

Fig. 15. Saturated triaxial test at two different confiningstresses: (a) deviatoric stress–strain curves; (b) volumetric–axial strain curves (experimental results by Chavez, 2004)

�4

�8

Model

Model

Experimental

Experimental

Confining stress 0·5 MPa�

0·5 MPa

0·5 MPa

0·1 MPa

0·1 MPa

0·1 MPa

0

0

5

5

10

10

15

15

20

20

25

25


Vertical strain: %

Dev

iato

ric s

tres

s,: M

Pa

σσ

13

�

2·0

1·5

1·0

0·5

0

Vol

umet

ric s

trai

n: %

0

4

8

(b)

2·5

Fig. 16. Triaxial test at 36% RH and at two different confiningstresses: (a) deviatoric stress–strain curves; (b) volumetric–axial strain curves (experimental results by Chavez, 2004)

12

16

Model

Model

Experimental

Experimental

0

0

5

5

10

10

15

15

20

20

25

25


Vertical strain: %

Dev

iato

ric s

tres

s,: M

Pa

σσ

13

� 1.2

0.8

0·4

0

Vol

umet

ric s

trai

n: %

0

4

8

(b)

1.6

Fig. 17. Triaxial test at 36% RH and saturation at a confiningstress of 0.5 MPa and constant deviatoric stress: (a) deviatoricstress–strain curves; (b) volumetric–axial strain curves (experi-mental results by Chavez, 2004)



saturated triaxial test at a confining stress of 0.5 MPa (Fig.15(b)). The behaviour exhibited by the drier specimen at aconfining stress of 0.5 MPa (Fig. 16(b)), which initiallycontracted and then dilated without a marked softening inp–q space cannot be properly reproduced by the implemen-ted RM.

Finally, the triaxial test results given in Fig. 17 weresimulated. The specimen was initially loaded in a drycondition (RH ¼ 36%) under �3 ¼ 0.5 MPa and then satu-rated at constant deviatoric stress. The loading stage isresumed under saturated conditions. The parameters cali-brated previously with the help of the tests described abovewere used. The model reproduces this singular test.

Clay coreSome tests were performed on the actual clay (wL ¼ 38%,

plasticity index (PI) ¼ 18.5%) used in the core. Sampleswere compacted statically to densities of standard Proctoroptimum. The optimum corresponds to a dry unit weight of17.6 kN/m3 and a water content of 17.8%.

Figure 13 shows the water retention curve obtained in thelaboratory and the estimated curve introduced in the calcula-tion using the van Genuchten model. The values of para-meters (P0 and º) are given in Table 3.

Clay specimens compacted at standard Proctor optimumwere tested in an oedometer test with suction control. Thetest results, plotted in Fig. 18, correspond to oedometricloading–unloading of a sample initially saturated at lowlevel of stress (0.02 MPa). During saturation, the sampleexpanded. A second sample, compacted approximately at thesame conditions, was first loaded at a constant suction equalto 1 MPa and then flooded at constant vertical stress(0.6 MPa) (Fig. 19). Model calculations are compared withexperimental data in Figs 18 and 19. The clay behaviourwas simulated using BBM (Alonso et al., 1990).

Initial saturation at low stress level allowed an estimationto be made of the parameter ks, which controls the elasticvolumetric strain for suction changes. Elastic and plastic

compressibility parameters (k and º(0)) were estimated fromthe saturated test (Table 4). A comparison between saturatedand unsaturated elastic compression curves indicates that theelastic stiffness increases with suction. This dependence hasnot been considered in the model. The model parameterswhich define the unsaturated compressibility and the shapeof loading collapse (LC) yield surface (in the space netmean stress–suction) are estimated by means the unsaturatedloading curve and the collapse intensity. Derived materialparameters are collected in Table 4. The table also showsthe set of parameters adopted for the natural foundation soil.They are based on field and laboratory data obtained at thedesign stage of the dam.

THE MODELA two-dimensional, plane strain, finite-element model of

cross-section P-10 of Lechago dam was developed. Themesh used in calculations is made of linear quadrilateralfour-node elements (Fig. 20(a)). Nodes have three degrees offreedom (water pressure, horizontal displacement and verti-cal displacement). The horizontal layers introduced to simu-late construction and the dam materials are shown in Fig.20(b).

Three materials define the dam: the clay core, the rockfillshells and the filters/drains. Four natural layers are distin-guished in the foundation. Starting at the bottom, a stiffshale substratum (E ¼ 500 MPa), 12 m thick, having a lowpermeability (3 3 10�9 m/s) is defined. Above this rocksubstratum, three soil layers are represented. Two permeablelayers (4 and 4.9 m thick) of alluvium soil sandwich a lesspermeable soft soil (10 m thick). The upper permeable soillayer was excavated to build the clay core while the othertwo soil layers were cut by an impervious wall, connectedwith the clay core of the dam. The impervious wall wasexcavated by means of a slurry-trench technique and laterfilled with a cement–bentonite mixture, which was simulatedby an elastic material (elastic modulus, E ¼ 100 MPa; Pois-son’s ratio, � ¼ 0.3).

Table 3. Initial void ratio and intrinsic permeability adopted in calculations

Material Initial void ratio, e0 Intrinsic permeability, k: m2 Retention curve, P0: MPa Retention curve, º

Rockfill shell 0.43 1 3 10�12 0.001 0.6Clay core 0.67 1 3 10�15 0.5 0.33Drain 0.43 1 3 10�11 0.01 0.33Foundation soil: bed rock 0.43 3 3 10�16 — —Foundation soil: alluvium soil 0.43 1 3 10�12 — —Foundation soil: soft layer 0.54 1 3 10�14 — —Cut-off wall 0.11 1 3 10�18 — —

Saturation

Model

Experimental

0·01 0·1 0 10Effective vertical stress: MPa

Voi

d ra

tio

0·6

0·5

0·4

Fig. 18. Saturated oedometric test on a compacted clay sampleinitially saturated at constant vertical stress (0.02 MPa)

Voi

d ra

tio

0·6

0·5

0·4

0·01 0·1 0 10Effective vertical stress: MPa

Saturation

Model

Experimental

Fig. 19. Oedometric test on a compacted clay sample. Satura-tion at constant vertical stress (0.6 MPa)



Horizontal and vertical displacements were fixed at thelower plane of the model, which was considered impervious.Horizontal displacements were also fixed to zero and verticaldisplacements were free along the upstream and downstreamvertical boundaries of the foundation.

The phreatic surface is located at the surface of thenatural soil at the upstream side of the dam and 5 m belowthe surface at the downstream side. This jump in water tableis created by the impervious wall and it was recorded by thepiezometers installed. The water level is simulated by impos-

Table 4. Parameters for the mechanical models used for the clay core and the foundation soils

Definition of parameter Symbol Units Base case for Lechago dam

Clay core Alluvium soil Soft layer Bed rock

(a) Elastic behaviour

Elastic modulus E MPa 37 250 — 5000Poisson’s ratio � — 0.3 0.3 0.3 0.3

(b) Plastic behaviour

Elastic compressibility k — — — 0.014 —Virgin compressibility for saturated conditions º(0)� k — 0.046 0.01 0.055 —Parameter that establishes the minimum value of thecompressibility coefficient for high values of suction

r — 0.7 — — —

Parameter that controls the rate of increase in stiffnesswith suction

� MPa�1 2 — — —

Elastic compressibility for changes in suction ks — 0.01 — — —Reference stress pc MPa 0.02 — — —Slope of critical state strength line M — 0.85 1.2 0.85 —Parameter that controls the increase in cohesion withsuction

ks — 0.1 — — —

Parameter that defines the non-associativeness ofplastic potential

Æ — 0.3 0.3 0.3 —

(c) Initial state for dam model

Initial suction s0 MPa 0.3 Saturated Saturated SaturatedInitial yield mean net stress p�0 MPa 0.08 0.25 0.23 —

(a)

(b)

Rockfill shells

Clay core

Dewatering points

Drain material

Bedrock Alluvium soil Soft soil

Fig. 20. Discretisation of cross-section P-10 of Lechago dam: (a) finite-element mesh; (b) materials and layers considered inconstruction simulation



ing a hydrostatic pressure distribution on the lateral verticalboundaries of the foundation.

An initial geostatic total stress distribution on foundationsoils was defined by a K0 coefficient equal to 0.5. Apreconsolidation mean effective stress ( p�0 ) equal to0.25 MPa has been defined for the pervious granular soil ofthe foundation and a slightly lower value, p�0 ¼ 0.23 MPa,for the soft soil.

Construction was simulated by adding layers to the initialgeometry of the foundation soils. The weight of each layeris applied in a ramped manner during the specified timeperiod of its construction. The combination of steps andramp loading gives a sufficiently realistic modelling. Eachlayer contains clay core elements, rockfill elements and drainelements, which were placed in between the core andshoulders. The dam was built in six steps (Table 5). Theinitial yield value of mean stress for dam materials shouldtake into account the compaction stresses. For the clay core,the nominal value of the applied stress of the roller compac-tor, was used to define the initial mean stress, assumingK0 ¼ 0.5. The RM requires the specification of the yieldstress for very dry conditions. The selected value,p�0 ¼ 0.25 MPa (Table 2), was approximated from the back-analysis of triaxial tests. The same parameter values wereselected for filter materials.

Concerning initial suction the measured water content insitu (Fig. 10 for the clay core and Fig. 12 for the rockfill)and the water retention curves (Fig. 13) provide an approx-imate initial mean suction: 0.3 MPa for the clay core and0.001 MPa for the rockfill.

In order to simulate the water level drawdown that wasperformed during construction to increase the preconsolida-tion stress of the soft deltaic deposits, two sink points (Fig.20(b)) were included in this layer with the purpose ofsimulating the imposed drawdown by means of a waterpressure reduction.

Before the construction of the two uppermost layers of thedam the preloading was simulated on the downstream side.Preloading was modelled through a vertical stress acting onthe downstream slope surface. The preload was applied andremoved, also using a ramp loading procedure to reproducethe construction and excavation process. During the 4months of preloading time, dewatering in the lower per-meable layer within the natural soil was simulated by meansof 2.5 m water level decrease in the two sink points justbelow the preloading zone.

Granular drains are simulated by means of a RM and thematerial parameter values are given in Table 2. Comparedwith the rockfill properties, the sandy filters are assumed tobe slightly stiffer and their frictional strength somewhatdecreased. The downstream drain is simulated by means of acontinuous layer which limits the clay core and continues

below the rockfill shoulder to filter possible foundation waterleaks and to avoid pressure development in the downstreamshell.

BEHAVIOUR DURING CONSTRUCTIONComparison between model performance and dam behav-

iour during construction will be based on stress measure-ments from total stress cells, vertical settlements determinedby hydraulic gauges and pore pressure from piezometerreadings. Hydraulic settlement gauges require a referencecontrol position which was installed at the downstream edgeof the horizontal berms at elevations 866 and 876 m. Thesepoints also experience settlements due to water table lower-ing and dam construction. The settlement gauges record thedifference in elevation between a given gauge location andthe reference position. The comparison made below refers tothis difference in settlements. The reference control positionswere levelled also by topographic procedures but the resultswere considered unreliable.

The total stress evolution recorded in some cells at somepoints has been compared with calculations. The comparisonrefers to several cells installed at the elevation 864–867 m.Fig. 21 shows the time records of measured stresses andtheir comparison with model predictions. The sudden reduc-tion of the stress in the measurements at one of the gaugesin the downstream rockfill (Fig. 21(c)) was attributed to anerror in the instrument performance. Stress cells seem toprovide in this case a fairly accurate indication of the damconstruction history. In fact, the periods of rest betweenperiods of construction activity are precisely detected. Nosignificant delayed effects are shown by the readings. Ver-tical stresses within the clay core at the end of damconstruction reach an average value of 0.4 MPa (two deter-minations). Recorded stresses in neighbouring points withinthe rockfill shells, close to the granular drains, show highervalues, demonstrating the presence of arching effects due tothe lower stiffness of the compacted clay core. Note that thedam geometry imposes a rapid change of height of com-pacted soil away from the core and arching effects develop.The model captures reasonably well the entire set of meas-urements. Of course, the large recorded stress variationswithin points close to each other (see, for instance, therecords for the two stress cells within the clay core) cannotbe reproduced by the model. In other respects the model isremarkably accurate. The model does not predict any signifi-cant delayed behaviour of stress development in the rockfillshells, as recorded. However, arching effects induced by theconsolidation of the lower part of the clay core result in acalculated decrease of the vertical stresses, a phenomenonnot observed in the field. Pore pressures in the core will bediscussed below.

Table 5. Construction stages

Model time Dates Modelling interval

Equilibrium of initial conditions Time 0–100 days 1Construction until elevation 874 m (see Fig. 1) Time 100–280 days 1 July 2006–30 December 2006 2

3Stop construction: Time 280–460 days 1 January 2007–30 June 2007 4Construction until elevation 886 m (see Fig. 1) Time 460–520 days 1 July 2007–30 September 2007 5

6

Phreatic level red. andpre-loading

Pre-loading rampPre-loadingUnloading

Time 520–560 daysTime 560–620 daysTime 620–640 days

1 September 2007–30 December 2007 789

Final construction (until elevation 895 m) (see Fig. 1) Time 640–680 daysTime 680–700 days

1 January 2008–30 February 2008 1011



867

Model

Model

Model

Model

Model

Model

02/0

7/20

0602

/07/

2006

02/0

7/20

06

01/0

9/20

0601

/09/

2006

01/0

9/20

06

01/1

1/20

0601

/11/

2006

01/1

1/20

06

01/0

1/20

0701

/01/

2007

01/0

1/20

07

03/0

3/20

0703

/03/

2007

03/0

3/20

07

03/0

5/20

0703

/05/

2007

03/0

5/20

07

03/0

7/20

0703

/07/

2007

03/0

7/20

07

02/0

9/20

0702

/09/

2007

02/0

9/20

07

02/1

1/20

0702

/11/

2007

02/1

1/20

07

02/0

1/20

0802

/01/

2008

02/0

1/20

08

03/0

3/20

0803

/03/

2008

03/0

3/20

08

03/0

5/20

0803

/05/

2008

03/0

5/20

08

03/0

7/20

0803

/07/

2008

03/0

7/20

08

02/0

9/20

0802

/09/

2008

02/0

9/20

08

02/1

1/20

0802

/11/

2008

02/1

1/20

08

02/0

1/20

0902

/01/

2009

02/0

1/20

09

04/0

3/20

0904

/03/

2009

04/0

3/20

09

04/0

5/20

0904

/05/

2009

04/0

5/20

09

Date(a)

Date(b)

Date(c)

�0·8

�0·8

�0·8

0

0

0

�0·2

�0·2

�0·2

�0·4

�0·4

�0·4

�0·6

�0·6

�0·6

864

864

Ver

tical

str

ess:

MP

aV

ertic

al s

tres

s: M

Pa

Ver

tical

str

ess:

MP

a

Fig. 21. Comparison of measured and calculated vertical stresses at elevation 867–864 m: (a) upstream rockfill near core; (b) clay core; (c) downstream rockfill nearcore



Figure 22 shows the calculated distribution of verticalstresses in the dam at the end of construction. Archingeffects reduce the reference geostatic distribution of stressesin the core.

Calculated vertical displacements at elevation 876 m arecompared with measurements in Fig. 23. In the upstreamzone, near the core, measured displacements are somewhatoverestimated by the calculated absolute displacements inthe corresponding points. The reference for measurementswas a fixed point in the slope of the valley. In contrast, inthe case of the clay core and the downstream shoulder, thecalculated values had to be corrected by substracting thesettlement in the location where the reference measurementsystem was installed, as described above. Obviously this isnot the best reference point but it had practical advantages.This situation can lead to measured values which apparentlyindicate an uplift of the soil, although all points experiencesettlements at all times. This trend is observed in the meas-ured settlement records and it is more exaggerated in thecalculations made. However, the magnitude of the measuredvalues is in general well captured. Settlement records alsoreact to the step-wise history of loading, but in a moreprogressive manner than the total stresses owing to consoli-dation effects in the foundation and the compacted structure.

Pore-water pressures are calculated and compared withmeasurements in Fig. 24. Piezometers could only measuresmall pressure deficiencies below the atmospheric value. Inpractice only positive values are measured. Consider first thepiezometers located in the saturated foundation soil invertical borings (Figs 24(a) and 24(c)). The measured timerecords reflect, in general, a hydrostatic distribution of porepressures, controlled by the position of the water table. Thedewatering period, especially the third one (final months of2007), is reflected in all measurements. The model alsoreacts in a similar way.

The piezometer close to the base of the clay core, up-stream of the cross-sectional axis (elevation 854.5 m) isinteresting because it shows a steady increment of porepressure due to the pore pressures accumulating slowly inthe compacted core. The model calculations follow a sharperrate of pressure accumulation: pressure starts in a negativevalue (suction) but evolves rapidly to a positive pressure,which increases slowly and also follows the dewateringepisodes.

The response of the clay core is plotted in Fig. 24(b).Piezometers located in upper levels (elevations 875.4 and885 m) cannot record the prevailing suction. The modelcalculates suction values not shown in the figure because ofthe pressure scale of the graph. The piezometer located inthe vicinity of the foundation (elevation 857 m) soon startsto record positive pore-water pressures. Model calculationsstart at the compaction suction and they show a progressivedecrease in suction and, at some time during construction,the development of positive pore pressures. The model reactsto the rapid increase in total stresses by increasing the pore

pressure (or by reducing suction). One period of rapidaccumulation of stress was June–September 2007. It can befollowed in the calculated response of the model but also onthe reaction of the piezometers at lower elevation (857 m).The fast reductions in suction and the transient periodtowards a new equilibrium predicted by the model do notcompare well with measurements. Interestingly, the piezo-meter at a higher elevation (864 m) begins to record positivepore-water pressures some time after the completion of thedam, in November 2008. This is not well captured by themodel, which predicts an earlier development of positivepore pressures. The recorded increase of pore pressures atthe end of the time period represented in Fig. 24 reflects thewater level elevation upstream (6 m), which took place inMarch–April 2009.

Figure 25 shows the calculated distribution of pore-waterpressures in the core at the end of construction. The modelpredicts a limited development of positive pore-water pres-sures in the lower third of the clay core. This result may becompared with the average value of recorded pore pressuresin four cross-sections of the dam. The measured porepressures in all four sections, in February 2008, once thedam was completed, have been plotted together by super-imposing, averaging and smoothing the measurements (Fig.25). The scatter of results in a given section and the limitednumber of piezometers installed makes this approach areasonable one to derive an integrated and global picture ofthe core hydraulic behaviour. Piezometers were located inthe downstream half section of the core and therefore noinformation on the upstream half is available. The calculatedresponse is plotted in Fig. 25 for two dates: February 2008and April 2009. The model predicts higher pore pressures inthe lower part of the core, although it correctly predicts thatthe upper two-thirds of the dam core remain unsaturated.Once the dam was completed, calculated excess pore pres-sures dissipate at a relatively fast rate. This dissipation andthe associated core settlement was the reason for the transi-ent stress reduction shown in Fig. 21(b). Measured porepressures were lower and their recorded dissipation was veryslow. This discrepancy may be explained by a lower-than-assumed compacted water content in shells and core andalso by errors in the assumed water retention and permeabil-ity parameters of both materials. However, no attempt tomodify them in order to fit the actual measurements wasmade.

PREDICTION OF DAM BEHAVIOUR DURINGIMPOUNDMENT

The protocol for dam impoundment has not yet beendefined for Lechago dam. The assumption made here is thatimpoundment takes place at a maintained rate of 0.1 m/dayuntil the maximum elevation (892 m). Figs 26–28 show thecalculated response of the dam in terms of vertical stresses,vertical displacements and pore-water pressures. The plots

Syy stresses

0�0·11�0·22�0·33�0·44�0·55�0·66�0·77�0·88�0·99�1·10

Fig. 22. Calculated contours of vertical stress at the end of dam construction



Model

Model

Model

Model

Model

Model

Model

Model

02/0

7/20

0602

/07/

2006

02/0

7/20

06

01/0

9/20

0601

/09/

2006

01/0

9/20

06

01/1

1/20

0601

/11/

2006

01/1

1/20

06

01/0

1/20

0701

/01/

2007

01/0

1/20

07

03/0

3/20

0703

/03/

2007

03/0

3/20

07

03/0

5/20

0703

/05/

2007

03/0

5/20

07

03/0

7/20

0703

/07/

2007

03/0

7/20

07

02/0

9/20

0702

/09/

2007

02/0

9/20

07

02/1

1/20

0702

/11/

2007

02/1

1/20

07

02/0

1/20

0802

/01/

2008

02/0

1/20

08

03/0

3/20

0803

/03/

2008

03/0

3/20

08

03/0

5/20

0803

/05/

2008

03/0

5/20

08

03/0

7/20

0803

/07/

2008

03/0

7/20

08

02/0

9/20

0802

/09/

2008

02/0

9/20

08

02/1

1/20

0802

/11/

2008

02/1

1/20

08

02/0

1/20

0902

/01/

2009

02/0

1/20

09

04/0

3/20

0904

/03/

2009

04/0

3/20

09

04/0

5/20

0904

/05/

2009

04/0

5/20

09

Date(a)

Date(b)

Date(c)

�0·8

�0·8

�0·8

0

0

0

�0·2

�0·2

�0·2

�0·4

�0·4

�0·4

�0·6

�0·6

�0·6

876

876

876

Ver

tical

dis

plac

emen

t: m

Ver

tical

dis

plac

emen

t: m

Ver

tical

dis

plac

emen

t: m

Fig. 23. Comparison of measured and calculated differences of vertical displace-ments between the measurement gauge and the reference location, elevation 876 m:(a) upstream rockfill near core; (b) clay core; (c) downstream rockfill near core



02/0

7/20

0602

/07/

2006

02/0

7/20

06

01/0

9/20

0601

/09/

2006

01/0

9/20

06

01/1

1/20

0601

/11/

2006

01/1

1/20

06

01/0

1/20

0701

/01/

2007

01/0

1/20

07

03/0

3/20

0703

/03/

2007

03/0

3/20

07

03/0

5/20

0703

/05/

2007

03/0

5/20

07

03/0

7/20

0703

/07/

2007

03/0

7/20

07

02/0

9/20

0702

/09/

2007

02/0

9/20

07

02/1

1/20

0702

/11/

2007

02/1

1/20

07

02/0

1/20

0802

/01/

2008

02/0

1/20

08

03/0

3/20

0803

/03/

2008

03/0

3/20

08

03/0

5/20

0803

/05/

2008

03/0

5/20

08

03/0

7/20

0803

/07/

2008

03/0

7/20

08

02/0

9/20

0802

/09/

2008

02/0

9/20

08

02/1

1/20

0802

/11/

2008

02/1

1/20

08

02/0

1/20

0902

/01/

2009

02/0

1/20

09

04/0

3/20

0904

/03/

2009

04/0

3/20

09

04/0

5/20

0904

/05/

2009

04/0

5/20

09

Date(a)

Date(b)

Date(c)

Model

Model

Model

Model

Model

Model

Model

Model Model

835

857864

840

845

875

847851

854

885

854

0·4

0·4

0·4

0·3

0·3

0·3

0·2

0·2

0·2

0·1

0·1

0·1

0

0

0

�0·1

�0·1

�0·1

Wa

ter

pres

sure

: MP

aW

ate

r pr

essu

re: M

Pa

Wa

ter

pres

sure

: MP

a

Fig. 24. Comparison of measured and calculated pore-water pressures in thelocations indicated: (a) foundation soils, under the core; (b) clay core; (c) foundationsoils, downstream



refer to the position of instruments in the central cross-section of the dam, given in Figs 21, 23 and 24. The timeorigin in Figs 26–28 is the final calculation time for theconstruction phase. This origin may change when the actualimpoundment starts.

Figure 26(a) shows a moderate increase of stresses atelevation 867 m, upstream of the core, due to the increase intotal unit weight of the shell during its saturation. Inside thecore (close to saturated conditions) and downstream theexpected change in stresses is very small. Collapse-inducedsettlements calculated at elevation 876, are very small and aslight heave is predicted due to reduction of effective stressduring impoundment (Fig. 27(a)). Downstream the expectedsettlement is even smaller. The maximum response concernsexpected water pressures. The points within the foundationfollow essentially the hydrostatic increase in pressure (Fig.28(a)). The core is progressively saturated and pore pres-sures change from negative (suction) to positive values inthe manner indicated in Fig. 28.

CONCLUSIONSLechago dam integrates widely different materials: a soft

saturated clay foundation, wide rockfill shells to ensurestability and a core of compacted, medium-plasticity Mio-cene clay. Material properties could be derived from differ-ent sources. The dam provision process followed standardpractice, which provided initial information on the propertiesof the foundation soils, the clay core and the rockfill.

Rockfill behaviour, however, was investigated in moredetail and large-diameter tests on compacted specimens ofthe quartzitic shale used in the shells became available.These tests were performed under RH control, followingrecent developments which highlight the relevance of thisvariable to control the constitutive behaviour of coarsegranular aggregates. The highly fractured quartzitic shaleled, after quarrying, to a coarse gravel-like granular materialwhose grain size distribution can be reasonably well reducedto be tested in a 25 cm diameter triaxial cell.

Triaxial tests on rockfill have been simulated by discretis-ing the specimens and following the actual sequence of RHchanges and load increments. The RM used reproducessatisfactorily the isotropic stress–volumetric strain as well asthe deviatoric stress–strain records. Difficulties are found incapturing the measured dilatancy.

Conventional wetting under load oedometer tests wassimulated in a similar way to derive material parameters forthe clay core.

The central cross-section of the dam was discretised andthe actual construction sequence was applied. Specific fea-tures of Lechago construction are the preloading of thedownstream shell to improve the undrained strength of thesoft foundation deposits and the dewatering of the founda-tion soils for the same purposes. These operations resultedin a staged construction of the dam, which lasted 2 years.

The comparison between dam performance and modelcalculation involved three types of measurements in time:vertical total stress in core and rockfill, settlements of thedam at different levels and (positive) pore-water pressures inthe clay core. These measurements are available at differentelevations and at several vertical profiles in the naturalfoundation soils.

Measured vertical stresses were remarkably consistentdespite the difficulties often encountered with these instru-ments. They indicated some arching phenomena in the softercore, which was captured by the model. Stress developmentsalso closely matched calculations in most cases.

Settlement includes the response of the soft foundationand the compacted structure. Despite the difficulties inmatching actual field conditions with the laboratory experi-ments available, the model performance is reasonably good.Lechago experienced large total settlements close to 1 m inthe central section.

Pore-water pressures in the core cannot be determinedunless they reach positive values. This prevents the properevaluation of model performance in the unsaturated regions.However, the development in time of positive pore pressures(starting at a condition of partial saturation) is an indirectand useful check of the calculations. It has been shown thatthe lower third of the core develops positive pore pressuresin the range 0.01–0.06 MPa (the latter in the base of thedam core). The model also predicts positive pore-waterpressure developments in the lower part of the dam. This isa check and a further validation of the model and the set ofparameters selected.

However, there are difficulties in practice which cannot beoverlooked. At the design stage properties of compactedmaterials are established on the basis of a few laboratorytests performed in a variety of specimens with the purposeof selecting the most appropriate soil type. Later, field

Model, April 2009

0 20 m

885·00

875·00

864·00

857·00

Measured, February 2008 Model, February 2008

0·01 MPa

0·02

0·05

0·06

00·11 MPa

0·14

0·18

0·22

00·03 MPa

0·067

Fig. 25. Average measured pore pressures in cross-sections P-8, P-10 and P-12 in February 2008 andcomputed results in February 2008 and April 2009



compaction techniques and specifications may experiencesignificant changes if compared with the project specifica-tions. Lechago is not an exception.

Attained densities, reference compaction energies andwater content limits were modelled to a certain extent. The

natural variability of quarries adds further uncertainties.Therefore, the reasonable response of the model and itssatisfactory comparison with some measurements may in-clude some fortuitous coincidences.

The model was used to explore the future behaviour of

0

0

0

�0·2

�0·2

�0·2

�0·4

�0·4

�0·4

�0·6

�0·6

�0·6

�0·8

�0·8

�0·8

Ver

tical

str

ess:

MP

aV

ertic

alst

ress

: MP

aV

ertic

alst

ress

: MP

a

Model

Model

Model

Model

Model

Model

06/0

7/20

0906

/07/

2009

06/0

7/20

09

05/0

9/20

0905

/09/

2009

05/0

9/20

09

05/1

1/20

0905

/11/

2009

05/1

1/20

09

05/0

1/20

1005

/01/

2010

05/0

1/20

10

07/0

3/20

1007

/03/

2010

07/0

3/20

10

07/0

5/20

1007

/05/

2010

07/0

5/20

10

07/0

7/20

1007

/07/

2010

07/0

7/20

10

06/0

9/20

1006

/09/

2010

06/0

9/20

10

06/1

1/20

1006

/11/

2010

06/1

1/20

10

06/0

1/20

1106

/01/

2011

06/0

1/20

11

Date(a)

Date(b)

Date(c)

867

864

864

Fig. 26. Calculated vertical stresses at elevation 867–864 m during dam impound-ment: (a) upstream rockfill near core; (b) clay core; (c) downstream rockfill nearcore



Model

Model

Model

Model

Model

Model

Model

Model

876

876

876

0·05

0·05

0·05

0

0

0

�0·05

�0·05

�0·05

�0·10

�0·10

�0·10

�0·15

�0·15

�0·15

Ver

tical

dis

plac

emen

t: m

Ver

tical

dis

plac

emen

t: m

Ver

tical

dis

plac

emen

t: m

07/0

4/20

0907

/04/

2009

07/0

4/20

09

07/0

6/20

0907

/06/

2009

07/0

6/20

09

07/0

8/20

09

08/0

8/20

1008

/08/

2010

07/1

0/20

09

07/0

8/20

0907

/08/

2009

07/1

2/20

09

07/1

0/20

0907

/10/

2009

06/0

2/20

10

07/1

2/20

0907

/12/

2009

08/0

4/20

1008

/04/

2010

08/0

4/20

10

08/1

0/20

10

08/1

0/20

1008

/10/

2010

08/1

2/20

10

08/1

2/20

10

07/0

2/20

1107

/02/

2011

08/1

2/20

10

08/0

6/20

1008

/06/

2010

08/0

6/20

10

08/0

8/20

10

06/0

2/20

1006

/02/

2010

Date(a)

Date(b)

Date(c)

Fig. 27. Calculated differences of vertical displacements between the measurementgauge and the reference location during dam impoundment, elevation 876 m:(a) upstream rockfill near core; (b) clay core; (c) downstream rockfill near core



07/0

4/20

0907

/04/

2009

07/0

4/20

09

07/0

6/20

0907

/06/

2009

07/0

6/20

09

07/0

8/20

0907

/08/

2009

07/0

8/20

09

07/1

0/20

0907

/10/

2009

07/1

0/20

09

07/1

2/20

0907

/12/

2009

07/1

2/20

09

06/0

2/20

1006

/02/

2010

06/0

2/20

10

08/0

6/20

1008

/06/

2010

08/0

6/20

10

08/1

0/20

1008

/10/

2010

08/1

0/20

10

08/1

2/20

1008

/12/

2010

08/1

2/20

10

07/0

2/20

1107

/02/

2011

07/0

2/20

11

08/0

8/20

1008

/08/

2010

08/0

8/20

10

08/0

4/20

1008

/04/

2010

08/0

4/20

10

Date(a)

Date(b)

Date(c)

Wa

ter

pres

sure

: MP

aW

ate

r pr

essu

re: M

Pa

Wa

ter

pres

sure

: MP

a

0·5

0·5

0·5

0·4

0·4

0·4

0·3

0·3

0·3

0·2

0·2

0·2

0·1

0·1

0·1

0

0

0

854

864

854

845

857

847851

835

875885

840

Model

Model

Model

Model

Model

Model

Model

Model

ModelModel

Fig. 28. Calculated pore-water pressures in the locations indicated during damimpoundment: (a) foundation soils, under the core; (b) clay core; (c) foundationsoils, downstream



the dam during impounding and some predictions of stress,settlements and pore pressure developments at the positionof installed instruments have been given in the paper inorder that they could be compared in the future with actualmeasurements.

REFERENCESAlonso, E. E., Gens, A. & Josa, A. (1990). A constitutive model

for partially saturated soil. Geotechnique 40, No. 3, 405–430,doi: 10.1680/geot.1990.40.3.405.

Alonso, E. E., Olivella, S. & Pinyol, N. M. (2005). A review ofBeliche Dam. Geotechnique 55, No. 4, 267–285, doi: 10.1680/geot.2005.55.4.267.

Chavez, C. (2004). Estudio del comportamiento triaxial de materi-ales granulares de tamano medio con enfasis en la influencia dela succion. PhD thesis, Universitat Politecnica de Catalunya,Barcelona, Spain.

Chavez, C. & Alonso, E. E. (2003). A constitutive model forcrushed granular aggregates which includes suction effects. SoilsFound. 43, No. 4, 215–227.

Chavez, C., Romero, E. & Alonso, E. E. (2009). A rockfill triaxialcell with suction control. Geotech. Testing J. 32, No. 3, 219–231.

Coussy, O. (1995). Mechanics of porous continua. Chichester, UK:Wiley.

Daehn, W. W. (1985). Behaviour of a rolled earth dam constructedon a compressible foundation. Proc. 5eme Congres des GrandsBarrages, Paris Q18, R7, 171–191.

DIT-UPC (2002). Code_Bright. A 3-D program for thermo-hydro-mechanical.analysis in geological media. User’s guide. Barcelo-na, Spain: Centro Internacional de Metodos Numericos enIngenierıa (CIMNE).

Gens, A., Sanchez, M., Guimaraes, LdN. et al. (2009). A full-scalein situ heating test for high-level nuclear waste disposal: ob-servations, analysis and interpretation. Geotechnique 59, No. 4,377–399, doi: 10.1680/geot.2009.59.4.377.

Oldecop, L. (2000). Compresibilidad de escolleras. Influencia de lahumedad. PhD thesis, Universitat Politecnica de Catalunya,Barcelona.

Oldecop, L. & Alonso, E. E. (2001). A model for rockfill compres-sibility. Geotechnique 51, No. 2, 127–139, doi: 10.1680/geot.2001.51.2.127.

Oldecop, L. & Alonso, E. E. (2003). Suction effects on rockfillcompressibility. Geotechnique 53, No. 2, 289–292, doi:10.1680/geot.2003.53.2.289.

Oldecop, L. & Alonso, E. E. (2007). Theoretical investigation onthe time dependent behaviour of rockfill. Geotechnique 57, No.2, 289–301, doi: 10.1680/geot.2007.57.2.289.

Olivella, S. & Alonso, E. E. (2008). Gas flow through clay barriers.Geotechnique 58, No. 3, 157–176, doi: 10.1680/geot.2008.58.3.157.

Olivella, S., Carrera, J., Gens, A. & Alonso, E. E. (1994). Non-isothermal multiphase flow of brine and gas through salinemedia. Transp. Porous Media 15, No. 3, 271–293.

Olivella, S., Gens, A., Carrera, J. & Alonso, E. E. (1996). Numer-ical formulation for simulator (CODE_BRIGHT) for coupledanalysis of saline media. Engng Comput. 13, No. 7, 87–112.

Ramırez, J. L., Soriano, A. & Serrano, C. H. (1991). Design andconstruction of Barbate dam. Proc. 17th ICOLD, Vienna, AustriaQ67, R8, 129–163.

Rizzoli, J. L. (1991). Observations du comportement de fonda-tions compressibles: barrages des retenues Seine, Aube etdique de Lazer. Proc. 17th ICOLD, Vienna, Austria Q66, R8,129–163.

Tellerıa, J. & Gomez Laa, G. (1991). Arbon dam: A didacticexperience about problems of a dam built on a deformablefoundation. Proc. 17th ICOLD, Vienna, Austria Q66, R9, 135–156.

Torner, V. & Novosad, S. (1991). The effect of difficult foundationconditions on the conceptual solution and construction of theSlezska Harte dam. Proc. 17th ICOLD, Vienna, Austria Q66,R40, 719–730.

Trkeshdooz, R., Nafissiazar, F. & Rastegari, Y. (1991). Effects offoundation conditions on the design of Taleghan embankmentdam. Proc. 17th ICOLD, Vienna, Austria Q66, R75, 1415–1488.

van Genuchten, M. T. (1980). A closed-form equation for predictingthe hydraulic conductivity of unsaturated soils. Soil Sci. Soc.Am. J. 44, 892–898.



Thorel, L. et al. (2011). Geotechnique 61, No. 5, 409–420 [doi: 10.1680/geot.10.P.029]

183

Physical modelling of wetting-induced collapse in embankment base

L. THOREL�, V. FERBER�†, B. CAICEDO‡ and I . M. KHOKHAR�

The relevance of the oedometer tests used for the predic-tion of wetting-induced deformations in embankments isexamined. Single and double oedometer tests are carriedout. A comparison is made between laboratory tests andgeotechnical centrifuge modelling at 100g conducted toexamine an inundated embankment made of a sand–claymixture. A 20 cm high embankment model is built andinstrumented. The material is compacted on the ‘dryside’ of the optimum Proctor curve at a low compactionrate in order to emphasise settlement phenomena. Theinundation simulation is conducted in two successivesequences during centrifuge flight up to a water table of5 cm. The results prove that the prediction of the drydensity after settlement due to inundation is good.

KEYWORDS: centrifuge modelling; collapsed settlement; com-paction; embankments; settlement

On examine la pertinence de tests sur oedometre effec-tues pour la prediction de deformations induites parl’imbibition dans les talus. On effectue des essais oedome-triques simples et doubles, ainsi qu’une comparaisonentre des essais en laboratoire et une modelisation encentrifuge geotechnique a 100 g, realisee pour etudier untalus inonde constitue d’un melange de sable et d’argile.On realise une maquette du talus de 20 cm de haut,dument instrumentee. Le sol est compacte du cote sec del’optimum Proctor, pour obtenir une faible compaciteafin de magnifier le phenomene de tassement. L’inonda-tion est simulee en deux phases successives, au cours dela centrifugation, jusqu’a atteindre un niveau de nappede 5 cm. Les resultats permettent d’affirmer que la pre-diction de la densite seche apres un tassement du a uneinondation est bonne.

INTRODUCTIONThe use of dry soil materials for the construction of roadwayand railway embankments is limited to small embankments,which, however, must plainly comply with compaction re-quirements (LCPC, 2003). The compaction of dry soils isenergy consuming and does not prevent settlement fromhappening in the case of soaking. A survey of the literaturereveals that the compaction rate is one main parameteraffecting long-term deformations of embankment bases(Lawton et al., 1992; Auriol et al., 2000).

This problem is crucial for countries combining naturallydry soils and notable flood hazard, where highways construc-tion programmes involving the construction of high embank-ments are developed (e.g. in Morocco: ADM, 2006).Therefore, the understanding of dry compacted soil behav-iour must be improved in order to optimise embankmentcompaction for better stability in relation to settlement risks.

Frequently, the wetting-induced behaviour of compactedsoils results in ‘collapsing’. Settlement mechanisms dependon the following factors.

(a) The compaction water content (or the suction) and thedry density after compaction. Collapse occurs mainly indry and loose soils (Lawton et al., 1992). Thecompaction of dry soils can induce low dry densitiesif the compaction energy is not high enough. Accordingto Ferber et al. (2008), the dry density after wettingunder a given vertical stress is affected by the initialdry density and not by the compaction water content.

However, some authors have observed some effectsrelative to the water content (Basma & Tuncer, 1992).

(b) The vertical stress applied during the inundation.Collapse settlement, in fact, increases with increasingvertical stress (Lawton et al., 1992); that is, thecollapse potential increases within the embankmentdeeper layers.

(c) The intrinsic properties of the soil, such as critical claycontent (Lawton et al., 1992; Rollins et al., 1993) orplasticity index (Lim & Miller, 2004), which dependson the non-clayey fraction of the soil. This critical claycontent ranges between 10 and 40% (El Sohby &Rabbaa, 1984), depending on the vertical stress andprobably on the grain size distribution of the non-clayeyfraction (Basma & Tuncer, 1992).

(d ) The stress anisotropy during the inundation (Lawton etal., 1989).

(e) The water content increase (or the suction decrease).The settlement increases non-linearly with the suctiondecrease (El-Ehwany & Houston, 1990; Pereira &Fredlund, 2000).

Regarding soil physical properties, the dry density aftercompaction is the predominant parameter affecting wetting-induced deformations under vertical stress (Lawton et al.,1992; Rao & Revanasiddappa, 2000; Estabragh et al., 2004;Ferber et al., 2008). Moreover, the long-proposed experimen-tal methods for the prediction of wetting-induced deforma-tions, like the single or double oedometer tests (Jennings &Knight, 1957), although remaining pertinent (Basma &Tuncer, 1992), need to be compared with observations likephysical models. Physical modelling has already been usedto study the collapse of embankments (Miller et al., 2001)without, however, ever focusing on the relationship betweeninitial dry density and settlement.

This paper presents a study conducted on a small-scaleembankment made of a dry and loosely compacted sand–clay mixture (SCM) using the Laboratoire Central des Pontset Chaussees (LCPC) geotechnical centrifuge. The objectiveof the comparison with oedometer tests is to examine andquantify the effect of a low compaction rate on settlement in

Manuscript received 8 March 2010; revised manuscript accepted20 December 2010. Published online ahead of print 22 March 2011.Discussion on this paper closes on 1 October 2011, for furtherdetails see p. ii.� L’Universite Nantes Angers Le Mans, Laboratoire Central desPonts et Chaussees, Centre de Nantes (Geotechnical Engineering,Environment and Risks Department, Physical Modelling inGeotechnics Unit), Bouguenais, France.† Entreprise Charier, Montoir de Bretagne, France.‡ Civil and Environmental Engineering Department, Universidad delos Andes, Bogota, Colombia.


case of inundation, and to assess the relevance of laboratorytest-based predictions. As the model embankment is com-pacted in loose conditions, which do not represent actualfield conditions, the results are presented at the model scaleonly. The extrapolation to prototype scale has to be carriedout using the classical scaling laws for displacement, timeand flow velocity (Garnier et al., 2007).

The laboratory tests carried out to measure soil geo-technical characteristics and hydromechanical performancesare first presented. Then, the centrifuge test is described andthe results discussed. Finally, a comparison is made betweencentrifuge and laboratory results.

LABORATORY RESULTSThe SCM components used for the centrifuge tests are

presented first. The oedometer test results are discussed in aseparate subsection.

MaterialsThe SCM consists of a mixture of Speswhite kaolin and

NE34 Fontainebleau sand. The sand is characterised by avery narrow particle size range (Fig. 1) and a negligible finecontent (Table 1). Because of these unusual properties, thecompaction of the sand is not very sensitive to watercontent, which accounts for the ‘flat’ shape of the Proctor

curve (Fig. 2). The absence of fines makes it difficult todensify, as illustrated by the low maximum dry density(MDD ¼ 1.51 t/m3, Table 1).

The clay is a purified kaolin made of 77% clay particles(Table 1, Fig. 1) and characterised by a plasticity index ofalmost 23. The optimum moisture content is 29% forMDD ¼ 1.42 t/m3 (Fig. 2). The clay fraction is mostlycomposed of kaolin minerals, among which some traces ofillite have been detected.

The SCM used for the tests is composed of 40% Spes-white kaolin and 60% Fontainebleau sand. Therefore, thematerial characteristics are a combination of those of theclay and the sand. The clay content places it within therange of collapsible soils (El Sohby & Rabbaa, 1984). Thecompaction characteristics and the shape of the standardProctor curve are similar to those of a clayey sand materialand the plasticity index is low (Table 1).

The SCM water retention curve, obtained using the osmo-tic method, is presented as a function of suction and heightof water above the water table (Fig. 3).

Single and double oedometer testsCollapse is a wetting-induced phenomenon, which must

be studied by describing accurately the effects of bothinundation and vertical stress. In this section the proceduresand the results of the laboratory tests involving both satura-tion and vertical stress variation are presented.

0

10

20

30

40

50

60

70

80

90

100

0·001 0·01 0·1 1Particle diameter: mm

Pas

sing

: %

Speswhite kaolin

40% kaolin–60% sand

Fontainebleau sand

Fig. 1. Grain size distribution of the Fontainebleau sand, theSpeswhite kaolin and the SCM

Table 1. Geotechnical characteristics of the soils

Characteristic Fontainebleau sand Speswhite kaolin 40% kaolin–60% sand

C400 �m: % 100 100 100Fines content: % 0.01 100 40Clay content: % , 2 �m 0 77 31Liquid limit, wL: % — 55.1 25Plastic limit, wP: % — 32.3 14Plasticity index, IP: % — 22.8 11Methylene blue absorption value: g/100 g 0.1 1.54 0.6Density of solid particles, rs : g/cm3 2.64 2.65 2.65Std Proctor optimum water content: % 7 29 13.5Std Proctor optimum dry density MDD: g/cm3 1.51 1.42 1.87Std Proctor optimum degree of saturation: % 24.7 88.7 85.8

1·2

1·3

1·4

1·5

1·6

1·7

1·8

1·9

2·0

0 5 10 15 20 25 30 35 40w: %

Fontainebleau sand40% kaolin–60% sandSpeswhite kaolin

Sr 100%�

ρ d3

: t/m

Fig. 2. Standard Proctor curves on the Fontainebleau sand, theSpeswhite kaolin and the SCM

184 THOREL, FERBER, CAICEDO AND KHOKHAR


Procedures. The tests are carried out in oedometer cells70 mm in diameter and 19 mm thick. After the sand and clayare mixed, tap water is added to reach the required watercontent. The material is kept in sealed bags for at least 24 h.

After the curing time, the specimens are compacteddirectly in the oedometer cells using a miniaturised dynamichammer (Ferber et al., 2008). Thus, the dry density isdirectly related to the number of blows of the hammer, inother words, to the compaction energy. The method has theadvantage of reproducing the compaction phenomena asso-ciated with the Proctor test compaction. Two tests, duringwhich the same compaction energy is applied on the mater-ial at constant water content, give similar dry densities.

The tests consist of loading and/or soaking the specimensaccording to three different paths (Fig. 4):

(a) loading unsaturated specimens at a constant watercontent (path 1), simulating the construction of anembankment

(b) loading saturated specimens (path 2) according to theconventional oedometer test

(c) soaking specimens after loading at a given verticalstress (path 3) to simulate the inundation of compactedfills occurring by way of, for instance, flooding.

In the case of path 1, the water content is kept constantduring the test by protecting the cells from the ambient airusing a plastic membrane (Delage & Fry, 2000). The in-crease in vertical stress is obtained by loading the framewith increasing masses. The specimen mass is measuredbefore and after the test to determine water content varia-tions, showing a decrease of 0.5 points at the highest watercontent (e.g. from 11 to 10.5%). The degree of saturationcorresponding to a water content of 11% is 39% for aninitial void ratio of 0.75 and 43% for e ¼ 0.68.

Path 2 corresponds to a conventional oedometer testcarried out on a soil compacted at a given water content anda given compaction energy, then loaded under a 3 kPavertical stress, and finally soaked. Very small deformationsresult from this saturation since the soil is not expansive.The mechanical load is applied when all deformations arecompleted. Paths 1 and 2 correspond to the elementary testsof the double oedometer test procedure.

Path 3 is performed on specimens compacted at differentdry densities but at the same water content. Six differentvertical loads are applied on the specimens, between 3 and800 kPa. These vertical stresses correspond to the stressundergone by a material found at the base of an approxi-mately 40 m high embankment. The initial void ratio rangebetween 0.39 and 0.74 corresponds to initial dry densitiesbetween 1907 kg/m3 and 1526 kg/m3.

Test results. The unsaturated and saturated oedometer testresults are plotted together in Fig. 5. Because of their low drydensity (high void ratio), no preconsolidation stress isobserved for the specimens compacted for the saturatedoedometer tests. The resulting compression index is approxi-mately 0.2. The four saturated test results do not merge,showing a discrepancy, which illustrates an inherent un-certainty. A void ratio amplitude of approximately 0.4 isobserved. This uncertainty is highly dependent on the initialvoid ratio because the tests reveal that the void ratio at all thestresses increases with increasing initial void ratio (3 kPa)(Fig. 5).

The unsaturated tests present a substantial preconsolida-tion stress (400–800 kPa). It is the limit between the elastic(for low stresses) and plastic (for high stresses) zones, whichis typical of the unsaturated behaviour of low-plasticitycompacted soils (Alonso et al., 1990). The elastic compres-sion index, which corresponds to the slope of the curve fora vertical stress lower than the preconsolidation stress, is0.01, whereas the plastic compression index determined fromthe data obtained at 800 and 1400 kPa, is 0.25. Finally, thevoid ratio difference for the three tests is only 0.025 onaverage. As for the saturated tests, this uncertainty isprobably controlled by the initial void ratio.

The unsaturated curves are located ‘above’ the saturatedcurves for vertical stresses higher than 30–60 kPa. It showsthat, above this threshold, the specimen is subject to settle-ment if water is added. On the contrary, below 30 kPa,swelling occurs with the addition of water.

The vertical stress influence on the deformation type andamplitude is confirmed by the results of the single oed-ometer tests (Fig. 6) according to path 3 (Fig. 4). Swelling

0·01

0·1

1

10

100

1000

5 10 15 20 25 30Water content: %

Suc

tion:

kP

a

0·001

0·01

0·1

1

10

100

5 10 15 20 25 30Water content: %

Mod

el h

eigh

t abo

ve w

ate

r ta

ble:

cm

Interpolated curveData

Fig. 3. SCM water retention curve

Watercontent Path 2

Path 1

Path 3

s 0� w

w

( 100%)Sr �

( 100%)r S

Verticalstress

Suction

Fig. 4. Hydromechanical paths

PHYSICAL MODELLING OF WETTING-INDUCED COLLAPSE IN EMBANKMENT BASE 185


occurs within specimens inundated with a vertical stress of3 kPa, but, when the vertical stress is higher than 100 kPa,settlements occur all the more extensively since the initialvoid ratio is high. Therefore, the void ratio after inundationdoes not depend on the void ratio after compaction. Forexample, the void ratio after inundation of specimens inun-dated with an 800 kPa vertical stress is about 0.38 for thefour loosely compacted specimens (void ratios within therange 0.5–0.7). Similar characteristics are observed for 200and 400 kPa. This is consistent with the observations fromthe literature on low-plasticity compacted soils (Ferber et al.,2008).

When the final void ratio after collapse of loose speci-

mens is plotted against vertical stress (Fig. 7), the data arefound on the compression curve of the saturated oedometertest. This means that the conventional compression curve ofthe saturated oedometer test can be used to predict the voidratio after collapse and, consequently, the collapse deforma-tions.

CENTRIFUGE TEST PREPARATION AND PROCEDUREA two-dimensional SCM embankment scale model, repre-

sented on the 1/100 scale, is inundated during a centrifugeflight at 100g. A description of the experiment is first

30

35

40

45

50

55

60

65

70

75

80

1 10 100 1000 10000

Vertical stress: kPa

Voi

d ra

tio: %

Unsaturated – 10·5%wi �

Saturated

ρd ·� 1 58 g/cm1

3

rS �

ρd3

r

1 52 g/cm1

�

�

·S

ρd ··

� 1 67 g/cm0 49

3

rS �

ρ ··

d3

r

1 79 g/cm0 60

�

�S

ρd � 1·92 g/cm1

3

rS �

ρd � 1·98 g/cm1

3

rS �

Fig. 5. Double oedometer tests on the compacted SCM

0·30

0·35

0·40

0·45

0·50

0·55

0·60

0·65

0·70

0·75

0·80

0·3 0·4 0·5 0·6 0·7Void ratio after compaction

Voi

d ra

tio a

fter

inun

datio

n

1:13 kPa100 kPa200 kPa400 kPa800 kPa

0·8

Fig. 6. Void ratio after inundation in single oedometer tests onthe compacted SCM (wi 11%)

30

35

40

45

50

55

60

65

70

75

80

1 10 100 1000 10000Vertical stress: kPa

Voi

dra

tio:%

Saturated oedometer

Unsaturated, 10·5%wi �

Void ratio after collapse

Fig. 7. Comparison between the results of single and doubleoedometer tests



proposed, then some information on the model post-con-struction SCM characteristics is given to characterise theinitial state.

Experimental set-up and preparationFor the purposes of the experiment, a two-dimensional

half cross-section of the embankment, which is assumed

symmetrical, is built (Fig. 8) inside a strongbox consistingof (Fig. 9):

(a) a main compartment, 20 cm wide and 60 cm long, inwhich the material is compacted

(b) some lateral compartments communicating with themain compartment, in which the height of water can bemeasured

Wa

ter

tabl

e

3 5 cm

2 0 33°

Laser(3)

LVDT(7)

Markersfor imageanalysis

Colouredsand interface

Geotextilesandgeotextile

Drainageplate

60 cm

28·5 cm3

cm

Fig. 8. Cross-section of the embankment model in the strongbox

Water tank(Mariotte bottle)

Waterdistributionsystem

Lateralcomponents

Maincompartment

Embankmentplatform

20 cm

Embankmentslope foot

Embankmentslope crest

CameraTransparentwall

28·5 cm

Water pressure transducer

LVDT displacement transducer

Laser displacement transducer

I

II

III 5 6 7

1 2 3 4

Fig. 9. Experimental set-up and instrumentation



(c) a transparent wall in the main compartment, which isused to film the model during the test.

A 2 cm thick sand layer, surrounded by a geotextile, islaid on the bottom of the strongbox. This draining layer isused to simulate the water table rise below the embankmentbase before the inundation event. The SCM, with a watercontent of 10.5%, is compacted on the ‘dry side’ of theProctor curve in seven successive layers, each approximately3 cm thick (Fig. 2).

Concerning the friction on the walls, previous researchwork conducted on the bearing capacity of strip footingsbuilt on a dry and dense sand model (Bakir, 1993) showsthat failure occurs with an average value 8% higher on atwo-dimensional model than on three-dimensional models,whereas settlement is 16% lower. If those results are linkedto the model geometry only, they come within the range ofthe variability due to uncontrolled soil properties, and conse-quently Bakir (1993) neglected this effect. In the presentstudy, the compaction process is simple, without using thewall effect reduction technique. Side friction may appear onthe glass window and reduce settlement. To evaluate suchbehaviour linear variable differential transducer (LVDT)sensors and laser sensors have been implemented on severalrows. As those displacement sensors are located at differentdistances from the walls (Fig. 9), the consequence of frictionof the SCM on the walls should be differential settlementsbetween sensors. This behaviour was not observed (see Fig.12 later), except at the final stage of phase 3 and near thecrest where the settlement is affected by the slope. This lowfriction is the result of the roughness of the glass. Ifthe maximum roughness corresponds to a micrometre, andthe friction developed at the SMC–glass interface followsthe behaviour of a sand–glass interface, Uesugi & Kishida(1986) have identified a surface friction angle of about 68.This would lead to a surface friction equivalent to 5% of thevertical stress. As a consequence, the side friction effect hasnot been taken into account in the present analysis.

Each layer is compacted with 96 blows of a conventionalstandard Proctor hammer, evenly distributed on the surface.Loosely compacted material is needed here to emphasise thedisplacement phenomena caused by wetting.

At the interface between each layer, circular pins areplaced close to the transparent wall as geometrical referencepoints. The image analysis makes it possible to follow thepoint displacements using a specially dedicated LCPC pieceof software (Thorel et al., 2000). The results correspond tothe mean value of the vertical displacement of the pins ofeach row.

After compaction, the structure is carved manually toform a slope of 338, a classical inclination for actual roadembankments filled with the same material. The final lengthsof crest and base are 28.5 cm and 60 cm, respectively (Fig.8). The corresponding prototype embankment would bescaled by a factor 100. Finally, a plastic sheet is placed ontop of the model to stop evaporation between the end of thepreparation phase and the centrifuge test.

The water level (0 and 5 cm from the base of theembankment) is controlled using a Mariotte bottle (Thorel etal., 2002) connected to the main compartment (Fig. 9). Thetest is conducted according to three successive sequences:

(a) no hydraulic connexion(b) water table at the bottom of the embankment(c) water table up to 50 mm above the base of the

embankment.

The total displacement of the embankment crest is calculatedusing some laser transducers placed above the crest, at a few

centimetres from the slope crest and an LVDT, regularlyapplied above the central part of the crest (Fig. 9).

Geotechnical properties of the embankment modelA ‘reference’ embankment, which is not subjected to

increasing g level, is devoted to the characterisation. It isprepared identically and in the same conditions as the modeldevoted to centrifugation. Characterisation consisted inmeasuring:

(a) the water content and dry density after compaction(b) the soil resistance using cone penetration tests (CPTs)

and a hydraulic penetrometer (Thorel et al., 2008).

In each layer, 28 specimens are sampled: 12 for watercontent measurements and 16 undisturbed ones for drydensity measurements. The water content is measured ac-cording to the oven-drying procedure and the dry density byhydrostatic weighing. In total, 196 specimens were sampledin the reference model after compaction.

The results (Fig. 10) show the following.

(a) The average water content is 10.4%, with a standarddeviation of 0.2%, which demonstrates the good

0

5

10

15

20

25

10

10

10·1

10·1

10·2

10·2

10·3

10·3

10·4

10·4

10·5

10·5

10·6

10·6

10·7

10·7

10·8

Water content: %

Pro

port

ion

ofsa

mpl

es: %

Average: 10·4%

Std dev.: 0·2%

0

5

10

15

20

25

30

35

40

1·42

1·42

1·44

1·44

1·46

1·46

1·48

1·48

1·5

1·5

1·52

1·52

1·54

1·54

1·56

1·56

1·6

Dry density: t/m3

Pro

port

ion

ofsa

mpl

es: %

Average: 1·53 t/m3

Std. dev.: 0·02 t/m3

Fig. 10. Water content and dry density histograms of thecompacted SCM in the embankment model



homogeneity (0.2% being approximately equivalent tothe uncertainty). The soil is very dry in proportion toearthworks as a whole (LCPC, 2003). A water contentof 10.4%, indeed, corresponds to 77% of the optimummoisture content (Table 1 and Fig. 2).

(b) The average dry density is 1530 kg/m3 (82% of theMDD), with a standard deviation of 20 kg/m3, which isalso a good index of homogeneity. This very lowcompaction ratio (generally, a minimum compactionrate of 95% is required in conventional embankments)is the result of both the water content and thecompaction energy applied. With such a density,substantial deformations are bound to occur.

Thirteen CPTs are carried out at 1g on the ‘reference’embankment on three rows: one row of five along thelongitudinal axis (Fig. 11(a)), and two rows of four on eitherside of the first row, at 5 cm from the compartment walls(Fig. 11(b)). These two rows are intended for assessing thewall effect.

The results show that the resistance is homogeneous bothvertically and horizontally, which confirms the water contentand dry density results. There is no discernible wall effect

during the lateral CPTs, although a larger range of resistanceis noted.

Water content, dry density and mechanical property resultsconfirm the good homogeneity of the model soil.

Test sequencesThree sequences simulating the different stress states that

embankment materials are susceptible to undergo, are de-scribed as follows:

(a) stress state 1 – embankment construction with increas-ing vertical stress

(b) stress state 2 – a possible water level rise beneath theembankment causing capillary rises

(c) stress state 3 – a possible inundation event, leading towater infiltrations from both the slope and the base ofthe embankment.

These three stress states are simulated by three differentsequences during the test:

(a) phase 1 – g level application to the embankment modelto increase the vertical stress in the soil

(b) phase 2 – saturation of the sand layer under theembankment by maintaining the water table just abovethe sand–embankment interface level

(c) phase 3 – inundation of the strongbox with a watertable located 5 cm above the sand–embankment inter-face.

CENTRIFUGE TEST RESULTSThe objective of the centrifuge tests is to examine the

wetting-induced settlements of the embankment model be-fore comparison with laboratory results. The settlements aremeasured according to two different principles (Fig. 8):

(a) for each layer, according to the reference points filmedby a camera placed in front of the transparent wall(Fig. 9); image analysis allows for the monitoring ofthe lowest four layers throughout a height of 12 cm

(b) at the crest level.

The total crest settlements (for which the zero valuecorresponds to the measurements before centrifugation) andthe crest settlement due to inundation (for which the zerovalue corresponds to the beginning of the first inundationevent) are presented as measurements continuous with time(Fig. 12) or as average values at the end of the three mainsequences of the test (Fig. 13 and Fig. 14):

(a) before inundation (phase 1), that is when the settle-ments caused by the g level increase are stabilised

(b) after inundation with a water table of zero (phase 2),that is when the settlements due to capillary rises fromthe sand layer are stabilised

(c) after inundation with a water table of 5 cm (phase 3),that is when the settlements due to the saturation of thesoil are stabilised.

Two acceleration/deceleration cycles are performed (Fig.12) at the beginning of the test (first hour) because ofunexpected technical problems. These unexpected eventshave generated a succession of positive and negative displa-cement readings: the model heave due to deceleration (Fig.12) caused irreversible loading-induced settlements, prevent-ing any possibility of returning to the initial value.

The settlement due to the centrifuge acceleration rangesfrom 0.7 to 4 mm (1.8 mm on average). Part of thesedifferences may come from the use of contact (LVDT) andnon-contact (laser) sensors. Laser transducers, for instance,

0

20

40

60

80

100

120

140

160

180

Cone resistance: MPa

Dep

th: m

m

0 0·2 0·4 0·6 0·8 1·0 1·2

0

20

40

60

80

100

120

140

160

180

0 0·2 0·4 0·6 0·8 1·0 1·2

Cone resistance: MPa

Dep

th: m

m

Fig. 11. CPT profiles along the longitudinal axis of the referenceembankment (a) and at 5 cm from the strongbox face (b)



show lower settlement values than LVDTs. This probablycan be accounted for by the load applied by the rod of theLVDT sensors, which generates a stress producing a smalloverestimation of the settlement values.

These settlements due to loading are not observed onfield-compacted soils, because they occur progressively dur-ing the construction of the embankment. On full-scaleembankments, measurements focus on settlements occurringafter their construction.

0

�18

�16

�14

�12

�10

�8

�6

�4

�2

0

2

0 2 4 6 8Elapsed time: h

Tota

l cre

st s

ettle

men

t: m

mLaser I

Laser II

Laser III

LVDT 1

LVDT 2

LVDT 3

LVDT 4

LVDT 5

LVDT 6

LVDT 7

Phase 2

Phase 3�16

�14

�12

�10

�8

�6

�4

�2

0

2

0 2 4 6 8Elapsed time: h

Cre

st s

ettle

men

t due

to in

unda

tion:

mm

Laser I

Laser II

Laser III

LVDT 1

LVDT 2

LVDT 3

LVDT 4

LVDT 5

LVDT 6

LVDT 7

Phase 2

Phase 3

0

20

40

60

80

100

120

0 1 2 3 4 5 6 7 8Elapsed time: h

Acc

eler

atio

n (9

·81

m/s

)2

2

4

6

8

10

Wa

ter

tabl

e: c

mAcceleration

Water table

Phase 2

Phase 3

Fig. 12. Cumulative crest settlement, acceleration and water table measurements: focus on the inundation sequences (model time)

�14

�12

�10

�8

�6

�4

�2

0

Beforeinundation(reference)

Water table0 cm

� Water table5 cm

�

Pla

tform

set

tlem

ent:

mm

Fig. 13. Cumulative crest settlement (model scale) at the end ofthe two main inundation sequences (average displacementsensor measurements; the error bar represents the standarddeviation)

�6

�5

�4

�3

�2

�1

0


Water table0 cm

� Water table5 cm

�

Laye

rs s

ettle

men

t: m

m

Layer 1

Layer 2

Layer 3

Layer 4

Fig. 14. Layer settlement (variation in height) at the end of thetwo main inundation sequences measured by image analysis(model scale)



After this self-weight-induced settlement phase, the twoinundation sequences taking place 3 h and 5 h after thebeginning of the test, respectively (Fig. 12), generate twodistinct crest settlement periods. Each phase begins whensettlement has reached an asymptotic value. The first inunda-tion event (phase 2) generates homogeneous settlements onthe embankment crest. The second event (phase 3) generatesmore extensive settlements near the slope crest (laser sen-sors) because of a combination of lateral displacement. Thislateral displacement combined with settlement has to belinked with the natural rotation of the stresses in the vicinityof the slope. Apparently, this phenomenon is not observedon the crest, far from the slope. This gives an indication thatthere is probably no rotation of the stresses in this zone, onwhich the study is focused.

During the test, the settlements are recorded on two (forLVDTs) or three (for laser transducers) locations on thetransversal axis (Fig. 9). The results show, for both types ofmeasurement, no gradient of settlement in the transversedirection. This confirms that the assumption of a translatorysymmetry of the vertical plane of the embankment, which isparallel to the glass, is acceptable. In other words, theassumption of low interface friction on the walls of thecontainer is admissible, even if their natures (glass oraluminium) are different.

The average crest settlement shows that (Fig. 13):

(a) the saturation of the sand layer generates a 5 mmsettlement

(b) the inundation with a water table of 5 cm generates anadditional settlement of 7 mm, reaching a total of12 mm.

The image analysis gives more details on the settlementdistribution within the different layers (Fig. 14) as follows.

(a) The settlement of each layer increases with eachinundation phase.

(b) The individual settlement of each layer increases withthe depth of the layer. For instance, layer 1, located atthe very base of the embankment presents a moreextensive settlement than the three other layers.Conversely, the extent of the settlement of layer 4 isthe less significant. Layer 3 appears to be an exception,since its settlement is more extensive than thesettlement of layer 2.

(c) The total settlement (Fig. 15) is 3.4 mm and 14.6 mmafter the first and the second inundation sequences,respectively. These values are similar to the averagevalues of 4.8 mm and 12 mm measured on the crest(Fig. 13).

The coefficient of variation (standard deviation divided bythe mean) ranges between 6% and 17% during phase 2(water table ,0) and between 1 and 13% during phase 3(water table ,5 cm). As the measurements are performed atthe model boundary, the glass window may affect the results.Nevertheless, all the results appear consistent.

The inundation of the embankment generates extensivesettlements due to the wetting-induced collapse behaviour ofloosely compacted soils. The maximum crest settlement afterthe last inundation sequence is higher than 12 mm (modelscale), which corresponds to a 1.2 m settlement in a full-scale embankment. This value, probably due to the initialloose soil compaction rate (82% of the MDD), is notacceptable for a transportation embankment.

The settlement values (measured using transducers orimage analysis) are consistent and increase with increasingdepth and increasing water table height. This result iscoherent considering that:

(a) the vertical stress amplifies the deformation due tocollapse

(b) the rise of the water table causes saturation, andconsequently, the collapse of an increasing thickness ofcompacted material.

DISCUSSIONThe void ratio before and after inundation is modelled on

the basis of the unsaturated and saturated oedometer tests,characterised by the following parameters.

(a) Saturated compression curve: the compression index(Cc) is equal to 0.2 (Fig. 16).

(b) Unsaturated compression curve: the elastic compressionindex is equal to 0.011 and the plastic compressionindex is equal to 0.25 (Fig. 16). The variation withincreasing suction of these compression indices isknown to be negligible (Chen et al., 1999). On theother hand, the preconsolidation stress depends on theinitial void ratio (Fig. 16). For instance, with an initial

�16

�14

�12

�10

�8

�6

�4

�2

0


Water table0 cm

� Water table5 cm

�

Cum

ula

ted

laye

rs s

ettle

men

ts: m

m

Fig. 15. Total layer settlement measured by image analysis atthe end of the two main inundation sequences (model scale)

0·3

0·4

0·5

0·6

0·7

0·8

0·9

1 10 100 1000 10000Vertical stress: kPa

Voi

d ra

tio

Law unsaturated oedometer

Law saturated oedometer

Cc saturated 0·2�

Cc elastic, unsaturated 0·011�

Cc plastic,unsaturated 0·25�

e3 kPa 0·5890% MDD

�

e3 kPa 0·7482% MDD

�

Fig. 16. Oedometer tests and compressibility used for displace-ments prediction



void ratio of 0.58 or 0.74 (corresponding to thecentrifuge embankment model), the preconsolidationstress would be 800 kPa or 200 kPa, respectively.

To compare void ratios obtained with the centrifuge testand with the oedometer tests, respectively (Fig. 17), thefollowing assumptions are made.

(a) In the embankment, only the zone far from the slope istaken into account. It has been shown (Fig. 12), that themeasured settlements of the crest were very close toeach other, and comparable to the settlement deducedfrom image analysis. In this zone it is assumed thatthere is no rotation of the stresses, and consequentlythat the principle of the comparison with oedometertests is relevant.

(b) The dry density after compaction is equal to 1530 kg/m3

(Fig. 10) in the embankment.(c) The vertical stress, �(z), at a given depth, z, in the

embankment, is related to the material height above thispoint by the initial density of the material. With a drydensity after compaction of 1530 kg/m3 and a watercontent of 10.4% (Fig. 10), the wet bulk density, r,used for calculations is 1530 3 1.104 ¼ 1690 kg/m3.The vertical stress is given by equation (1) and thecalculations are made using the centrifuge accelerationvalue measured during the test (Fig. 12), namely100 3 g ¼ 981 m s�2, with g being acceleration due toEarth’s gravity.

(d ) The void ratio for each test sequence is determined, inthe four layers, as a function of the initial state aftercompaction and of the variation of the layer heightprovided by the image analysis (equation (2)).

� (z) ¼ rgz (1)

hf � hi

hi

¼ ef � ei

1þ ei

, ef ¼hf � hi

hi

1þ eið Þ þ ei (2)

where ei and ef are the void ratio after compaction and afterthe test phase, respectively, and hi and hf are the layerheight after compaction and after the test phase, respectively.

The consequences of the inundation may be considered interms of cumulative displacement as a function of the heightin the embankment (Fig. 18). This parameter is calculatedfrom the following results.

(a) The oedometer tests: the displacement is calculatedwith the variations of the void ratio in the elementarylayers (equation (3)), assuming that their thickness islow enough to neglect the vertical stress variation. Thecumulative displacement at a given height is the sum ofvariations of the lower elementary layer height(equation (4)).

(b) The displacement measurements on the embankment:the cumulative displacement is deduced from the imageanalysis information or from the crest displacement.

hkf � hk

i ¼ hki

ekf � ek

i

1þ eki

(3)

where eki and ek

f are the void ratio of the kth elementarylayer after compaction and after the test phase, respectively,and hk

i and hkf are the kth elementary layer height after

compaction and after the test phase, respectively.

Hf � Hi ¼Xn

k¼1

hkf � hk

i (4)

where Hi and Hf are the embankment height after compac-tion and after the test phase, respectively, and n is thenumber of elementary layers.

The variations of the void ratio observed during the differ-ent sequences (Fig. 17) reveal the following.

(a) After the first inundation event (water table ¼ 0 cm),the void ratio decrease is relatively small, and, exceptfor the first layer, the value is very close to that of thevoid ratio predicted by the unsaturated oedometer test.Considering the high values obtained for crest settle-ment during this sequence (Fig. 12), layer 1 wasprobably already partly saturated because of capillaryrises, which explains why the void ratio decrease islarger than the unsaturated oedometer test prediction (adifference of 0.2 mm is observed);

(b) After the second inundation event (water table ¼ 5 cm),the void ratio decreases in the three lowest layers

0

3

6

9

12

15

18

21

0·3 0·4 0·5 0·6 0·7 0·8

Void ratio

Dep

th: c

m

Model after compaction – measuredModel after inundation (0 cm) – measuredModel after inunation (5 cm) – measuredUnsaturated oedometer – calculatedSaturated oedometer – calculated

Layer 1

Layer 2

Layer 3

Layer 4

Fig. 17. Comparison of void ratio measurements obtained withthe centrifuge and oedometer tests (centrifuge model scale)

Model after centrifugation – measuredModel after inundation (0 cm) – measuredModel after inundation (5 cm) – measuredUnsaturated oedometer – calculatedSaturated oedometer – calculated

0

3

6

9

12

15

18

21

�2�1·5�1�0·50Cumulated displacement due to inundation: cm

Dep

th: c

m

Layer 1

Layer 2

Layer 3

Layer 4

Fig. 18. Comparison of the cumulative vertical displacementdue to inundation obtained with the centrifuge test andoedometer tests (model scale)



because of the increasing water level. In the fourthlayer, a limited decrease is observed, which can beexplained by the fact that the capillary rise does notreach the full height of the fourth layer. The saturatedoedometer test results give a good prediction of thefinal void ratio in layers 1 and 3 but a discrepancyoccurs regarding the second layer.

The cumulative displacements as a function of depth inthe model (Fig. 18) show the following.

(a) The application of the centrifuge force generates a crestsettlement (2 mm) less extensive than the one predictedwith the unsaturated oedometer tests (3.5 mm). Thisrelatively low value for the model scale would representa difference of 15 cm on a full-size embankment, whichis quite significant. However, this settlement wouldoccur during the construction of the embankment and,therefore, would not be actually observed.

(b) The first inundation sequence presents a cumulativedisplacement profile similar to the unsaturated oed-ometer profile. The small shift between both curvesmight be the consequence of the saturation of thelowest layer, most probably by capillary rise. Thedifference between both curves on the crest alsoreaches 1.5 mm at model scale.

(c) The second inundation sequence presents a profile whichis very close to the saturated oedometer profile in thethree first layers. Above, the fourth layer, only, is partlyaffected by the inundation. Extensive settlements,consistent with the predictions, are observed in the thirdand in part of the fourth layer. This suggests that theinundation event comes with notable and very fastcapillary rise, causing the saturation of the soil at aheight of approximately 5 cm in a few hours. Previousresearch studies have shown that, in sand materials,capillary rises occur very rapidly under the effect ofmacro-gravity because of an increase of the capillaryvelocity by N times (Depountis et al., 2001), which canexplain the fast capillary rises observed in layers 3 and 4.

A reasonable consistency is observed between oedometerand centrifuge tests. Assuming that the oedometer tests arerepresentative of embankment settlements far from the slope(no horizontal displacements observed or allowed), thesetests are satisfactory for prediction. In the shallower zonesand under the slopes, the stress ratio may noticeably modifyperformances (Lawton et al., 1992), and this approachshould not be used.

The prediction is associated with uncertainty, since modelscale discrepancies of 1 to 2 mm are observed betweenlaboratory and centrifuge tests, corresponding to differencesbetween 10 cm and 20 cm at full scale, that is 10–20% ofthe settlement. This gives the order of magnitude of theprediction uncertainty obtained with the oedometer tests.

CONCLUSIONAn inundation is simulated during a centrifuge test in

two-dimensional geometry in order to examine the collapsebehaviour of embankments. The test embankment model isbuilt with a loosely compacted (82% MDD) clay–sand mix-ture to increase the collapse phenomenon. This material isstudied by carrying out single and double oedometer tests.

The maximum water height reached during the inundationevents is 5 cm, which causes 12 mm crest settlements atmodel scale (at full scale, a 20 m high field embankmentsubjected to a 5 m high inundation would settle by approxi-mately 1 m).

It must be emphasised that the quality of the predictiondepends mainly on the two following parameters:

(a) the height of the capillary rises, that is the saturationfinal height

(b) the dry density profile before inundation.

The analysis of the laboratory results shows that thesettlements far from the slope (where there are no rotationsof the stresses) can be roughly predicted on the basis of thedouble oedometer test. The method presented could also beused to define dry density objectives during the placementprocess depending on the embankment height. These objec-tives are given by the saturated oedometer test. The inunda-tion simulation study conducted at the same time on a full-scale embankment confirms the transposition potential ofthis approach (Vinceslas et al., 2009). This method, however,would not be relevant for high-plasticity clays (Ferber et al.,2008), knowing that such soils are generally banned fromcommon compacted fills.

ACKOWLEDGEMENTThe authors would like to thank the LCPC technical staff

involved in this study, particularly Jean-Pierre David.

REFERENCESADM (Autoroutes du Maroc) (2006). Autoroutes du Maroc: Rap-

port d’activites 2006, 44 pp. See www.adm.co.ma for furtherdetails (accessed 21/09/2010).


Auriol, J. C., Havard, H., Mieussens, C. & Queyroi, D. (2000).Resultats d’enquetes sur la pathologie des remblais en service.Routes/Roads, No. 306, 57–74.

Bakir, N. (1993). Investigation on centrifuged models of bearingcapacity of shallow foundations. PhD thesis, Nantes University,France (in French).

Basma, A. A. & Tuncer, E. R. (1992). Evaluation and control ofcollapsible soils. J. Geotech. Engng 118, No. 10, 1491–1504.

Chen, Z. H., Fredlund, D. G. & Gan, K. M. (1999). Overall volumechange, water volume change, and yield associated with anunsaturated compacted loess. Can. Geotech. J. 36, No. 2, 321–329.

Delage, P. & Fry, J. J. (2000). Comportement des sols compactes:apports de la mecanique des sols non satures. Revue Francaisede Geotechnique 92, 17–29.

Depountis, N., Davies, M. C. R., Harris, C. et al. (2001). Centrifugemodelling of capillary rise. Engng Geol. 60, No. 1–4, 95–106.

El-Ehwany, M. & Houston, S. L. (1990). Settlement and moisturemovement in collapsible soils. J. Geotech. Engng 116, No. 10,1521–1535.

El Sohby, M. A. & Rabbaa, S. A. (1984). Deformational behaviourof unsaturated soils upon wetting. Proc. 8th Regional Conf. forAfrica Soil Mech. Found. Engng 1, 129–137. South AfricanInstitution for Civil Engineers.

Estabragh, A. R., Javadi, A. A. & Boot, J. C. (2004). Effect ofcompaction pressure on consolidation behaviour of unsaturatedsilty soil. Can. Geotech. J. 41, No. 3, 540–550.

Ferber, V., Auriol, J. C., Cui, Y. J. & Magnan, J. P. (2008). Wetting-induced volume changes in compacted silty clays and high-plasticity clays. Can. Geotech. J. 45, No. 2, 252–265.

Garnier, J., Gaudin, C., Springman, S. M. et al. (2007). Catalogueof scaling laws and similitude questions in geotechnical centri-fuge modelling. Int. J. Phys. Modelling in Geotechnics 7, No. 3,1–24.

Jennings, J. E. & Knight, K. (1957). The prediction of total heavefrom the double-oedometer test. Transactions of a symposium onexpansive clays, pp. 13–19. Midrand, South Africa: SouthAfrican Institution of Civil Engineering.

Lawton, E. C., Fragaszy, R. J. & Hardcastle, J. H. (1989). Stress



ratio effects on collapse of compacted clayey sand. J. Geotech.Engng 117, No. 5, 714–730.

Lawton, E. C., Fragaszy, R. J. & Hetherington, M. D. (1992).Review of wetting-induced collapse in compacted soil. J. Geo-tech. Engng 118, No. 9, 1376–1394.

LCPC (Laboratoire Central des Ponts et Chaussees) (2003). Practi-cal manual for the use of soils and rocky materials in embank-ment construction (Techniques et methodes des laboratoires desponts et chaussees), 60 pp. Bouguenais, France: LCPC.

Lim, Y. Y. & Miller, G. A. (2004). Wetting-induced compression ofcompacted Oklahoma soils. J. Geotech. Geoenviron. Engng 130,No. 10, 1014–1023.

Miller, G. A., Muraleetharan, K. K. & Lim, Y. Y. (2001). Wetting-induced settlements of compacted-fill embankments. TranspnRes. Rec. 1755, 111–118.

Pereira, J. H. F. & Fredlund, D. G. (2000). Volume change behaviorof collapsible compacted gneiss soil. J. Geotech. Geoenviron.Engng 126, No. 10, 907–916.

Rao, S. M. & Revanasiddappa, K. (2000). Role of matric suction incollapse of compacted clay soil. J. Geotech. Geoenviron. Engng126, No. 1, 85–90.

Rollins, K. M., Rollins, R. L., Smith, T. D. & Beckwith, G. H.

(1993). Identification and characterization of collapsible gravels.J. Geotech. Engng 120, No. 3, 528–542.

Thorel, L., Noblet, S., Garnier, J. & Bisson, A. (2000). Capillaryrise and drainage flow through a centrifuged porous medium.Proceedings of the international symposium on physical model-ling and testing in environmental geotechnics, La Baule, 15–17May 2000, pp. 251–258.

Thorel, L., Favraud, C. & Garnier, J. (2002). Mariotte bottle in acentrifuge: a device for constant water table level. TechnicalNote. Int. J. Phys. Modelling in Geotechnics 2, No. 1, 23–26.

Thorel, L., Rault, G., Garnier, J. et al. (2008). Macro-gravity meas-urements on reduced-scale models of geotechnical structures.Bull. de Liaison des Ponts et Chaussees 272–273, SpecialMetrologie, 93–131.

Uesugi, M. C & Kishida, H. (1986). Frictional resistance at yieldbetween dry sands and steel. Soils Found. 26, No. 4, 139–149.

Vinceslas, G., Khay, M., Sagnard, N. & Ferber, V. (2009). Moisturestate variations on embankments located in flood zones: Instru-mentation and behavioral monitoring of an experimental em-bankment. Bull. des Laboratoires des Ponts et Chaussees 274,5–30.



Selected Questions and Answers


Partial Saturation in Compacted Soils, 197–201 [http://dx.doi.org/10.1680/geot.sip11.qa]

197

Keynote Speeches

Questions posed by Professor Eduardo E. Alonso (Depart-ment of Geotechnical Engineering and Geosciences, Univer-sitat Politecnica de Catalunya, Barcelona, Spain) to MrAnthony O’Brien (Mott MacDonald, Croydon, UK) in rela-tion to the keynote speech ‘‘The assessment of Old RailwayEmbankments. Time for a change?’’ by A.S. O’Brien:

(a) In an embankment poorly compacted collapse phenom-ena induced by rain infiltration are likely. Since collapsestrains are far from being homogeneous, they will resultin shearing strains throughout the embankment mass,especially during the early stages of operation (severalyears?). These shearing strains will degrade the strengthof the compacted soil and will contribute, as anadditional mechanism, to instabilities.

(b) Details of the numerical analysis by means of thecomputer code FLAC are requested. In particular, whatis the driving mechanism in FLAC to simulate swelling(or collapse) strains?

(c) It appears that the main mechanism leading to theeventual failure of a given embankment is a progressiveprocess. This mechanism is especially relevant in plasticclays because of the large drop in strength from the initialas compacted conditions to residual conditions. However,in low plasticity materials (involved in many failures),this process of strength reduction is of secondaryimportance. What is then the explanation for the delayedfailures observed in embankments built with lowplasticity soils?

Response from Anthony O’Brien:(a) There can be several different triggers which may initiate

the development of non-uniform strains across anembankment. Shortly after construction of these oldrailway embankments, collapse settlement usually devel-oped and the resulting drop in crest level was made goodby tipping locomotive ash over the clay fill. Thisphenomenon was quite widespread, and is now reflectedin the thickness of ash which is commonly observedwhen drilling through the crest of these embankments.An ash thickness of about one metre is fairly typical, bothacross embankments which have remained stable andthose which have been prone to delayed failure. Hence,additional factors also need to be considered in order toexplain the delayed failures and the differing rate of longterm degradation of embankments composed of plasticclay fills. These other factors include, inter alia: failuresduring construction; vegetation/climate interactions; clayfill permeability; presence of underlying soils which mayincrease or decrease the risk of progressive failure (due todiffering permeability, compressibility or strength charac-teristics); effectiveness of drainage (especially in vicinityof embankment toe). Some embankments suffered fail-ures during construction, especially larger embankments(in excess of 6m height), as previously discussed bySkempton (1996). Although construction was eventuallycompleted successfully, sometimes by novel techniquessuch as forming ‘‘burnt clay’’, there is little doubt that insome locations the old shear surfaces from the construc-tion failures were not fully removed. Hence, these

remaining zones now act as ‘‘hidden defects’’. Asdiscussed in my paper these old embankments ‘‘breathe’’seasonally, and the seasonal shrink-swell movements(which have been monitored across many embankments)are not fully reversible. The plastic deformation at theend of each shrink-swell cycle leads to a ‘‘ratcheting’’mechanism with a net outward movement at theembankment toe. After several decades, these movementswill inevitably lead to a substantial loss of strength in thebase of the embankment, if the embankment fill iscomposed of ‘‘brittle’’ plastic clay (with a substantialpost-peak drop in strength towards residual). Themagnitude of seasonal shrink-swell movement, and theassociated risk of instability depends on several factors asoutlined in my paper. The permeability of the clay fill is akey factor since this will fundamentally affect thevulnerability of the embankment to seasonal changes inclimate. Many of the dumped (or poorly compacted) clayfills have an intermediate range of permeability (between10�7m/s and 10�9 m/s), between what could be consid-ered to be practically impermeable and free draining,(when compared with the rainfall and evaporation rateswhich are usually experienced in the UK). Within thisintermediate permeability range, rainfall can infiltrate butnot easily drain, and therefore the clay fill can experiencerelatively large seasonal changes in pore water pressure.In contrast, well compacted clay fills (typical of modernCivil Engineering construction) would have a far lowerpermeability and would experience smaller seasonalchanges in pore water pressure.

(b) The soil model used in the FLAC code is a bespokemodel which uses specific ‘‘FISH’’ functions to simulateseasonal shrink-swell behaviour, together with post-peakstrain softening towards residual strength. The model hasbeen described in detail by Ellis and O’Brien (2007),where it was used to analyse progressive collapse ofcuttings in stiff plastic clays. Referring to Appendix 1 ofmy paper, the compressibility and swelling behaviour ofthe clay fill, and underlying natural soils, are described bya linear elastic-perfectly plastic soil model. Prior to yield,the elastic Young’s Modulus is dependent on the meaneffective stress. Hence, the modulus decreases whenswelling towards low effective stress, and increases whenconsolidating towards high stress. Figure 1 shows com-parisons between FLAC simulations, and oedometer tests(with unload-reload cycles) on hand cut block samples. Itis appreciated that this type of stress-strain model over-simplifies the soils non-linear small-strain behaviour(which is important for many practical soil-structureinteraction problems). However, for delayed failureanalyses it is more important to correctly capture thesoil’s large strain behaviour, and this type of model issatisfactory for this purpose. A similar soil model hasalso been successfully used by Nyambayo et al. (2004),for simulating the progressive failure of embankments.Comparisons between FLAC simulations and observedseasonal embankment deformations showed that this typeof soil model could predict shrink-swell movements of asimilar order of magnitude as those observed (althoughgiven the complex range of factors which influencedseasonal behaviour precise matches could not be


achieved). More rigorous soil models could be imple-mented, however, this additional complexity would haveprecluded the comprehensive suite of sensitivity studies(some of which are outlined in the paper) which werecarried out. These sensitivity studies were essential inorder to assess the relative importance of the differentfactors which can affect the risk of delayed failure.

(c) In the context of UK Railway embankments, the cause ofthe delayed failures which affect embankments built withlow plasticity clays was very clear (based upon thesurveys which have been reported across the network,following each failure). The terms used to describe thesefailures are ‘‘wash-out’’ or ‘‘flow’’ failures. Typically,they are relatively shallow (about one or two metresdeep), and are triggered by intense rainfall (Winter,2006), and are usually associated with faulty drainage, ornon-uniform ground surface topography, which leads toconcentrated flows of water across the embankmentsurface. These failure mechanisms are quite different tothe deeper failure mechanisms which can be seen in more

plastic clay fills. For low plasticity soils, the mostimportant factors are to understand the overall hydro-logical regime, and to ensure the drainage systems areadequate. For high plasticity clays, the causative factorsare more complex, as discussed in the paper.

REFERENCESEllis, E.A. & O’Brien, A. S. (2007): Effect of Height on Delayed

Collapse of Cuttings in Stiff Clay, Proc. of the I.C.E., Geotech-nical Eng., GE2, p. 73–84, April 2007

Nyambayo, V.P. Potts, D.M., & Addenbrooke T.I. (2004): The Influ-ence of Permeability on the Stability of Embankments Experien-cing Seasonal Cyclic Pore Water Pressure Changes, Proc. TheSkempton Conf., London, Thomas Telford, Vol 2, p. 898–910.

Skempton, A.W. (1996): Embankments and cuttings on the earlyrailway, I.C.E. Proc., Construction History 11: p. 33–49

Winter, M. (2006): Unstable Natural Slopes, Climate Change andImpacts on Scottish Infrastructure, Presentation to 2nd CLIFFSworkshop at http://cliffs.Iboro.ac.uk/.

0

0·5

1·0

1·5

2·0

2·5

10 100 1000

Ver

tical

str

ain:

%

σ�v: kPa

Range of FLACload/reload gradient fromelement simulations

Range of observedload/reload gradientfrom Oedometer tests

Key:Oedometer test– TP105

Oedometer test– TP607

FLACE 75(p 100)� � �

FLACE 90(p 100)� � �

FLACE 60(p 100)� � �

Note:Quoted FLAC stiffnessprofiles are for theload/reload loop notthe initial loading loop

Figure 1. Compressibility and swelling of a dumped clay fill, flac simulation and oedometer data(block samples)

198 SELECTED QUESTIONS AND ANSWERS


Material Characterisation

Question posed by Professor Domenico Gallipoli (Labora-toire SIAME, Universite de Pau et des Pays de l’Adour,France) to Dr Anh Minh Tang (Ecole des Ponts ParisTech,UMR Navier/CERMES, Paris, France) in relation to thepaper ‘‘Effects of the maximum soil aggregates size andcyclic wetting–drying on the stiffness of a lime-treatedclayey soil’’ by A.M. Tang, M.N. Vu and Y.-J. Cui

The results from this experimental campaign show that theshear stiffness of lime-treated soils is almost insensitive tovariations of water content and suction (provided that ex-cessive drying is prevented) suggesting that, within theexplored suction range, cementation dominates the behaviourof inter-granular contacts rather than suction. If this wasproven to be a general result, the effect of suction oncemented soils could possibly be disregarded at least withina limited range of water contents, thus enabling extension ofmodels developed for saturated cemented soils to the case ofunsaturated cemented soils. Of course, this also requires thatinter-granular cementation stays relatively undamaged, andhence soil strains remain relatively small, throughout servicelife of the structure. In spite of these restrictions, the abovehypothesis has the potential of introducing significant simpli-fications in constitutive modelling of unsaturated cementedsoils and is probably worth further exploration. I would beinterested to know the Authors’ opinion on this issue andwhether they have performed additional tests to explore therelative importance of water capillarity and inter-granularcementation on other soil properties such as, for example,normal compression behaviour and strength, which mightcorroborate or refute the above hypothesis.

REFERENCESTang, A. M., Vu, M. N. & Cui, Y.-J. (2011). Effects of the

maximum soil aggregates size and cyclic wetting–drying on thestiffness of a lime-treated clayey soil. Geotechnique 61, No. 5,421–429, doi: 10.1680/geot.SIP11.P.005

Response from Anh Minh Tang:In the authors’ opinion, we cannot say that the shear

stiffness of lime-treated soils is almost insensitive to varia-tions of water content and suction because the results showclear increase of Gmax when water content is decreasing andclear decrease of Gmax when water content is increasing. Amore precise statement is: the shear stiffness of lime-treatedsoils is almost insensitive to suction cycles when excessivedrying is prevented. It is actually interesting to investigatethe relative importance of water capillarity and inter-granularcementation on soil mechanical properties. Nevertheless,from the experimental point of view, some difficulties re-main in performing such a study. Actually, in the workpresented by the authors, the values of Gmax of lime-treatedsoils are compared to that of untreated soils. The differenceincludes not only the effect of inter-granular cementationbut also others effects of lime treatment. The best optionwould be performing suction-controlled mechanical test. Butthe results from this kind of test are difficult to analysebecause of the evolution of mechanical properties due to thepouzzolanic reactions during the test.

SELECTED QUESTIONS AND ANSWERS 199


Experimental Observation and Modelling

Question posed by Professor Domenico Gallipoli (Labor-atoire SIAME, Universite de Pau et des Pays de l’Adour,France) to Dr Enrique Romero (Department of Geotechni-cal Engineering and Geosciences, Universitat Politecnica deCatalunya, Barcelona, Spain) in relation to the paper ‘‘Aninsight into the water retention properties of compactedclayey soils’’ by E. Romero, G. Della Vecchia and C. Jommi

In equation (2) the suction s�m is introduced as a soilparameter corresponding to the suction for which the waterratio ew becomes equal to e�m and micropores are fullysaturated while macropores are empty. However, when thewater ratio ew increases above e�m and s drops below s�mequation (2) is no longer used to calculate the microstruc-tural void ratio em (which coincides with the microstructuralwater ratio, given that micropores are fully saturated) as afunction of suction. The model assumes instead that themicrostructural void ratio em is calculated by equation (3)as a function of total water ratio ew. This assumption leadsto the consequence that the reference suction sm calculatedby equation (5) when s , s�m has no physical meaning but itis only a mathematical parameter, which corresponds to thesuction value at which the two branches of the waterretention curve given by equations (2) and (4) join together.The Authors are invited to comment whether they haveexplored the possibility of using equation (2) instead ofequation (3) to calculate the microstructural void ratio em

when s , s�m, in which case equation (5) is not needed andthe value of suction s�m can be used in equation (4) insteadof sm.

REFERENCESRomero, E., Della Vecchia, G. & Jommi, C. (2011). An insight into

the water retention properties of compacted clayey soils. Geo-technique 61, No. 4, 315–330, doi: 10.1680/geot.2011.61.4.315

Response from Enrique Romero:The microstructural void ratio under saturated conditions

(for s , s�m, corresponding to ew . e�m) cannot be directlyestimated using equation (2) (microstructural branch of thewater retention curve). There are two main reasons for this.To begin with, the extrapolation into the microstructuralsaturation range of this equation significantly over-estimatesthe amount of microstructural void ratio, compared to meas-ured values using mercury intrusion porosimetry and digitalimage analysis of photomicrographs in which equation (3) isbased. Particularly for s!0, the extrapolation of equation(2) predicts higher microstructural values than soil voidratio. Conversely, the physically based equation (3) alwaysresults in a microstructural void ratio lower than void ratio.Secondly, since equation (2) does not depend on void ratio,its extrapolation into the saturation range will predict thesame microstructural void ratio for densely and looselycompacted clayey soils. This is not the case, since low-density samples display greater capacity for microstructuralexpansion due to the larger amount of water stored atsaturation. The proposed water retention model onlyconsiders the extrapolation of equation (2) into themicrostructural saturated range (ew . e�m) to determine thereference suction sm when em is known (equation (5)).This reference suction, which presents a different value onmain drying and main wetting to tackle hydraulic hysteresis,is used to smooth the transition between the microstructuraland macrostructural water retention domains. The incursionof sm into microstructural saturation (equation (5)) generallyextends over a limited suction range in most clayey soils,which depends on the activity of the clay (parameter � inequation (3)) and the amount of water stored (ew in equation(3)).

200 SELECTED QUESTIONS AND ANSWERS


Benchmarking of Techniques and Models

Question raised by Professor Eduardo E. Alonso (Uni-versitat Politecnica de Catalunya, Barcelona, Spain) to Pro-fessor Alessandro Tarantino (University of Strathclyde,Glasgow, UK) in relation to the paper ‘‘Benchmark onexperimental techniques for measuring and controlling suc-tion’’ by A. Tarantino, D. Gallipoli, C.E. Augarde, V. DeGennaro, R. Gomez, L. Laloui, C. Mancuso, G. El Mountas-sir, J.J. Munoz, J.-M. Pereira, H. Peron, G. Pisoni, E.Romero, A. Raveendiraraj, J.C. Rojas, D.G. Toll, S. Tombo-lato and S. Wheeler

Some of the discrepancies reported may be explained ifenough time for equilibrium is allowed during tests. It seemsthat time is the missing parameter in the explanation given.Simulating the different tests as boundary value problems,including water transfer mechanisms, may help to reducediscrepancies.

REFERENCESTarantino, A., Gallipoli, D., Augarde, C., De Gennaro, V., Gomez,

R., Laloui, L., Mancuso, C., McCloskey, G., Munoz, J., Pereira,J.M., Peron, H., Pisoni, G., Romero, E., Raveendiraraj, A.,Rojas, J.C., Toll, D., Tombolato, S. & Wheeler, S. (2011).Benchmark of experimental techniques for measuring and con-trolling suction. Geotechnique 61, No. 4, 305–314, doi:10.1680/geot.2011.61.4.305

Response from Alessandro Tarantino:We agree that equilibrium time and hydraulic boundary

conditions are critical when measuring or applying suction.Nonetheless, we have concluded that this was not theexplanation for the discrepancies observed between differenttechniques.

Tensiometer measurement was carried out with the sampleplaced in a small closed cell and we assumed that equili-brium between liquid pore-water and vapour water surround-ing the sample was reached when the water pressure

measured by the tensiometer attained a constant value.Suction and water content measured were therefore asso-ciated with a ‘static’ equilibrium condition.

For the case of the osmotic technique, the specimens wereenveloped by a semi-permeable membrane, which was sub-merged into the osmotic solution. Under these conditions,we assumed that water vapour transfer was negligible andthat water was only transferred via liquid phase. Equilibriumwas considered to be reached when water content attained aconstant value (checked by successive weighing). Suctionand water content measured were therefore associated with a‘static’ equilibrium condition.

Equilibrium is more complex in the axis translation tech-nique because samples are placed in an open system, withwater vapour possibly evaporating into the air supply system.In the axis translation oedometer, liquid water exchangedwith the sample could be measured. It was therefore possibleto check whether significant evaporation was taking place(water volume exchanged with the water supply systemtends to change linearly with time) and, if it was the case, acorrection was made to the water content measurement.

For the pressure plate, it was difficult to assess whetherwater vapour lost by evaporation was significant. It is clearlypossible that a dynamic equilibrium was achieved, with therate of water withdrawn from the porous ceramic discequating the rate of water evaporating into the air pressurechamber/air supply system. However, if this was the case,water content should have been underestimated whereas thepressure plate seems to overestimate the water content withrespect to the tensiometer at given suction. This is thereason why we have been looking for other explanations andfound that the contact pressure may in part explain theobserved discrepancies. In this respect, it is worth noticingthat the weight applied at the top of the samples tested byUSTRAT to improve contact had negligible effect on thewater flow process. The change in total isotropic stress onthe initially saturated samples and, hence, the excess pore-water pressure generated, was very small compared to theinitial suction of the sample.

SELECTED QUESTIONS AND ANSWERS 201


ice manual of geotechnical engineering vol 1: geotechnical engineering principles, problematic soils...

Documents