Application of FEM On Retaining Walls

THE APPLICATION OF FINITE ELEMENT
ANALYSIS TO THE DESIGN OF

EMBEDDED RETAINING WALLS
RICHARD IAN WOODS
School of Engineering
University of Surrey
Volume I
A Thesissubmittedfor the degreeof

Doctor of Philosophy
September2003
Q Richard Ian Woods, 2003

ABSTRACT
insightinto the behaviourof embeddedretainingwalls canbe providedby

Considerable
numericalmodelling techniques, such as the finite element method. Finite elements
can be
in
used retaining wall design in order to carry out sensitivity to
analyses optimisewall designs
andto identify critical designparameters (the Uhat if' approach),andto obtainrealisticand

accurate quantitativepredictionsof the behaviour of the wall and soil, suchaspredictionsof
the displacementsaround a diaphragm wall (the 'absolute'approach).Much of the recent
motivationfor using finite elementsin retainingwall designhasarisenfrom increased
constructionactivity in urban areas,and the needto demonstrate

that movementsof adjacent
structuresandservicesare acceptable,prior to planningconsentbeinggranted.
Finite elementanalysisof embeddedwalls presentsspecialchallenges.For example,the

geometryis usuallycomplexandit changesincrementally;shortandlong term solutionsare
required;stiffhesscontrastsarehigh, andaccuratepredictionsof soil displacements,
anchor
andprop forcesandwall bendingmomentsarerequired. finite
Commercial elementsoftware
packages make complex, non-linearanalysisusinglarge meshesan apparentlyroutinematter.
it is easyfor noviceusersof thesetools to makeseriousmistakeswithout realizingit, but
thereis little specificadviceon potentialpitfalls or guidanceon good practiceavailablein the
public domain. Five principalareasof concern have been identifiedand examinedin the
(i)
presentstudy: geometricmodellingand (ii)
discretization; constitutivemodellingand
(iii)
parameterselection; modellingof constructioneffects and long term effects;(iv)
difficulties;
computational and(v) obtainingdesignoutput.
Systematicnumericalinvestigationshavebeenconductedinto eachof theseareas.Basedon

the resultsof the analyses,this thesispresents:guidelinesfor describingthe geometryof an
embeddedwall problem;an evaluationof the useof variouslevelsof sophisticationin the
analysis;strategiesfor testingthe 'goodness'of the solutionobtained;a comparisonof
differentways of achievingthe sameobjective;andan examinationof the origin of
'unexpected'results.Particularemphasisis placedon assessing
the reliability of the output
that is mostoften soughtin practicaldesign.The CRISPgeotechnicalfinite elementpackage
hasbeenusedin this study,but the findingsandrecommendations
obtainedarerelevantto the
use of other programs.
ACKNOWLEDGEMENTS
I would like to thankProfessorChrisClaytonfor encouragingme to do this research

andfor supervisingmy efforts. He wasableto wield the stick anddanglethe carrot as
necessary, andgenerallyknewwhich wasappropriateat anygiventime. More recently,
Dr. Vicki Hopegaveup a considerableamountof time in order to readthe final draft of
the thesisandgive me invaluablecomments,andI amextremelygratefulto her.
Many of my other Civil Engineering colleaguesat the University of Surrey have had a
hand in helping me to complete this work. In particular, Mike Gunn, Marcus Matthews,
Mike Huxley and Noel Simons all urged me on and/or helped create the spacefor me to
do the researchand write it up.
A numberof organizationshaveprovidedmewith opportunitiesto be involvedwith

somefascinatingretainingwall projects. I would like to thankmy variouscolleaguesat
Bullen Consultants,Laing TechnologyGroup,MouchelParkman,ParsonsBrinckerhoff
(formerlyKennedy& Donkin), SurreyCountyCouna TarmacConstructionServices,
andthe TransportResearchLaboratory.
Last but by no meansleast,I mustrecordmy deepestgratitudeto my wife Lesleyand

my childrenSarah,Emma,Emily, David andAmy - who haveput up with me doingthis
for a greatdeallongerthan expected.As muchaswaspossible,they createdthe space
for me to do the work whilst preventingme from gettinganynotionsof selfimportance.
CONTENTS
Abstract
Acknowledgments
Contents i
Notation viii
CHAPTER I GENERAL FVMODUCTION

1.1 I-EstoricalBackground 1-1
1.2 Finite Elements in Geotechnical Design 1-3
1.3 Background to Present Work 1-5
1.4 Objectives of Present Work 1-9
1.5 Layout of Thesis 1-11
CHAPTER 2 GENERAL BACKGROUND

2.1 Development
IFEstorical 2-1
2.1.1 Beginnings 2-1
2.1.2 The first decade: 1970-1980 2-3
2.1.3 Milestonesof the 1980s 2-5
2.1.4 Trendsof the 1990& 2-7
2.1.5 Othermethods 2-10
2.2 PresentStateof the Art 2-12
2.2.1 Art v practice 2-12
2,2.2 Categoriesof use 2-13
2.2.3 Motivationsfor FE analysis 2-17
2.2.4 Obstaclesandobjections 2-18
2.2.5 Alternatives 2-22
2.3 Analysisof the Literature 2-23
2.3.1 Usagecategory 2-23
2.3.2 Wall andsupporttype 2-25
2.3.3 Wall section 2-26
2.3.4 Soil type 2-26
2.3.5 Analysistype 2-27
2.3.6 Drainageconditions 2-27
2.3.7 Constitutivemodel 2-28
2.3.8 Wall installationmethod 2-28
2.3.9 Output quantities 2-29
2.3.10 Finite elementcode 2-30
2.3.11 Contributionto mainareasof interest 2-30
2.3.12 Overallrelevance/quality 2-31
2.4 Summary 2-32
Tables 2-34
(i)
CHAPTER 3 THE CRISP FINITE ELEMENT PACKAGE APPLIED TO
RETAINING WALL ANALYSIS
3.1 Developmentof the CRISPPackage 3-1
3.2 Structureof the CRISPPackage 3-4
3.2.1 GEOM program 3-4
3.2.2 MAIN program 3-5
3.2.3 Pre- andpost-processing 3-6
3.3 Featuresof the CRISPPackage 3-7
3.3.1 Elementtypes 3-7
3.3.2 Drainageconditions 3-8
3.3.3 Constitutivemodels 3-11
3.3.4 In-situ stresses 3-12
3.3.5 Boundaryfixities 3-13
3.3.6 Appliedloading 3-15
3.3.7 Changinggeometry 3-16
3.3.8 Solutionscheme 3-17
3.3.9 Summary 3-18
3.4 Selectionof FiniteElementMesh 3-19
3.4.1 Soil 3-19
3.4.2 Retainingwall 3-20
3.4.3 Supportsystem 3-23
3.4.3.1 props 3-23
3.4.3.2 anchors 3-26
3.4.4 Soil-wallinterface 3-27
3.5 Type of Analysis 3-29
3.6.1 Uncoupled 3-29
3.6.2 Coupled 3-29
3.6.3 Otherapproaches 3-31
3.6 Choiceof ConstitutiveModel 3-32
3.6.1 Elasticmodels 3-32
3.6.2 Elastic-perfectlyplasticmodels 3-33
3.6.3 Critical statemodels 3-33
3.6.4 Structuralmaterials 3-34
3.6.5 All continuummaterials 3-35
3.7 Definition of Initial Stresses 3-36
3.7.1 Strataboundaries 3-37
3.7.2 Slopinggroundsurface 3-37
3.7.3 Slopingwater table 3-37
3.7.4 Non-hydrostaticconditions 3-38
3.8 BoundaryConditions 3-39
3.8.1 Displacement 3-39
3.7.2 Drainage 3-39
3.9 AppliedLoading 340
3.10 ConstructionModelling 341
3.10.1 Wallinstallation 3-41
3.10.2 Excavation 342
(ii)
3.10.3 Temporary supports 3-45
3.10.4 Permanentsupports 3-46
3.11 Long Term Equalization 3-47
3.11.1 Drainage boundary conditions 347
3.11.2 Fluctuating groundwater levels 3-49
3.11.3 Time steps 3-50
3.12 Analysis verification 3-52
3.12 Summary 3-53
Figures 3-54
CHAPTER 4 GEOMETRIC MODELLING AND DISCRETIZATION

4.1 Introduction 4-1
4.2 Location of Boundaries 44
4.2.1 Finite elementmesh 4-4
4.2.2 Analysisparameters 4-5
4.2.3 Evaluationof results 4-13
4.2.4 Resultsof analyses 4-16
4.2-4.1 Wall horizontal displacements 4-16
4.2.4.2 Wall bending moments 4-19
4.2-4.3 Excavation heaves 4-21
4.2-4.4 Ground surfacemovements 4-22
4.2.5 Extensionof boundaries 4-24
4.2.6 Summary 4-26
4.3 Mesh Aspect Ratio 4-27
4.3.2 Summary 4-29
4.4 Boundary Conditions 4-29
4.4.2 Summary 4-31
4.5 Number of Elements 4-32
4.5.1 h-refinement: series1 4-33
4.5.4 Summary 4-38
4.6 Sizeof Elements 440
4.6.1 r-refinement series1 440
4.6.2 r-refinement seriesH 4-44
4.6.3 Summary 4-46
4.7 Type of Elements 4-46
4.7.1 Resultsof analyses 448
4.7.2 Summary 4-48
4.8 Discussionand Summary 4-49
Tables 4-53
Figures 4-76
(iii)
CHAPTER 5 CONSTITUTIVE MODELLING AND PARAMETER SELECTION
5.1 GeneralIntroduction 5-1
5.1.1 Nonhomogeneity 5-1
5.1.2 Anisotropy 5-2
5.1.3 Small-strainbehaviour 5-3
5.1.3.1 Back analysis 5-4
5.1.3.2 Triwdal testing 5-5
5.1.4 Non-linearity 5-6
5.1.5 Yielding 5-7
5.1.6 Non-associated flow 5-9
5.1.7 Overviewof numericalstudies 5-11
5.2 Nonhomogeneity 5-12
5.2.1 Descriptionof analyses 5-12
5.2.3 Summary 5-17
5.3 Anisotropy 5-18
5.3.3 Summary 5-22
5.4 Non-finearity 5-23
5.4.3 Summary 5-26
5.5 Yielding 5-27
,
5.5.3 Summary 5-32
5.6 Non-associated Flow 5-33
5.6.3 Summary 5-36
5.7 DiscussionandSummary 5-37
Tables 5-39
Figures 543
CHAPTER 6 MODELLING OF CONSTRUCITONAND LONG TERM EFFECTS

6.1 Introduction 6-1
6.2 Wall Installation 6-2
6.2.1 Previouswork 6-2
6.2.2 Wished-in-placemethod 6-6
6.2.3 Element swoppingmethod 6-7
6.2.4 Completesequence 6-9
6.2.4.1 Multiple overlay method 6-11
6.2.4.2 Applied pressuremethod 6-14
6.2.5 Summary 6-15
OV)
6.3 Bulk Excavation 6-16
6.3.2.1 Numberof incrementblocks 6-19
6.3.2.2 Totalnumberof increments 6-21
6.3.3 Summary 6-22
6.4 TemporaryPropping 6-23
6.4.4 Summary 6-25
6.5 PartialDrainage 6-26
6.5.4 Discussion 6-29
6.5.5 Summary 6-30
6.6 Long-TermEqualization 6-30
6.6.2 Initial studies 6-31
6.6.3.1 Equalizationtime 6-34
6.6-3.2 Numberof equalizationincrements 6-35
6.6.3.3 Weightingof time steps 6-36
6.6.4 Summary 6-37
6.7 Groundwater Fluctuations 6-37
6.7.2 Movingphreaticsurfaces 6-38
6.7.3 Numericalstudies 6-39
6.7.4 Summary 640
6.8 DiscussionandSummary 641
Figures 644
CHAPTER 7 COMPUTATIONAL DIFFICULTIES

7.2 Stiffnessand Aspect Ratio 74
7.2.1 E-Conditioning 7-5
7.2.3.1 equilibrium errors 7-12
7.2.3.2 nodal reactions 7-14
7.2.3.3 normal stressin soil 7-16
7.2.3.4 bendingmomentsin structure 7-17
7.2.4 Discussionand summary 7-18
7.3 Effective StressMethod 7-20
7.3.1 Introduction 7-20
(v)
7.3.4 Discussion andsummary 7-26
7.4 CoupledAnalysis 7-26
7.4.4 Discussion andsummary 7-28
7.5 HorizontalStresses 7-29
7.5.2 Tensileactivestresses 7-30
7.5.2.1 descriptionof analyses 7-30
7.5.2.2 resultsof analyses 7-31
7.5.2.3 discussion andsummary 7-35
7.5.3 Passive pressureconcentrations 7-36
7.5.3.1 descriptionof analyses 7-37
7.5.3.2 resultsof analyses 7-38
7.5.3.3 discussion andsummary 7-40
7.5.4 Lateralstressoscillations 7-42
7.6 SolutionScheme 7-43
7.6.3 Resultsof analyses 746
7.6.4 Discussion andsummary 748
7.7 Discussion andSummary 7-49
Figures 7-51
CHAPTER8 OBTAINING REQUIRED DESIGN OUTPUT

8.1.1 Displacements 8-1
8.1.2 Stressesand pressures 8-3
8.1.3 Internal structuralforces 84
8.1.4 Assessmentv interpretation 8-6
8.2 Displacements 8-7
8.2.1 Graphicalrepresentation 8-7
8.2.2 Wall movement 8-11
8.2.3 Excavationheave 8-12
8.2.4 Ground surfacesettlement 8-13
8.2.5 Horizontal ground movement 8-14
8.2.6 Summary 8-15
8.3 Soil StressDistributions 8-15
8.3.2 Stresssmoothingin CRISP 8-23
8.3.3 Descriptionof analyses:horizontal stress 8-26
8.3.4 Resultsof analyses:retainedside 8-26
8.3.5 Resultsof analyses:excavatedside 8-29
(vi)
8.3.6 Descriptionof analyses: soil-wallshearstress 8-31
8.3.7 Resultsof analyses: soil-wallshearstress 8-31
8.3.8 Summary 8-32
8.4 Wall BendingMoments 8-32
8.4.1 Elementsin bending 8-32
8.4.2 Calculationmethods 8-34
8.4.3 Benchmarking - wall stressbending moments 8-37
8.4.4 Assessment - wall stress bending moments 8-38
8.4.5 Improvement - wall stress bending moments 8-39
8.4.6 Assessment - earth pressure bending moments 8-40
8.4.7 Summary 842
8.5 Wall ShearForces 843
8.5.1 Elementsin shear 843
8.5.2 Calculationmethods 843
8.5.3 Investigation- wall stress shear forces 8-45
8.5.4 Benchmarking - wall stressshear forces 8-46
8.5.5 Assessment - wall stress shear forces 846
8.5.6 Assessment - earth pressure shear forces 847
8.5.7 Summary 848
8.6 Prop andAnchorLoads 849
8.6.1 Tension/compression 849
8.6.2 Bendingandshear 8-51
8.6.3 Summary 8-52
8.7 ConcludingRemarks 8-52
Figures 8-56
CHAPTER 9 CONCLUSIONS AND RECOMMENDATIONS

9.1 PreviousWork andEstablished Practice 9-1
9.2 GeometricModellingandDiscretization 9-2
9.3 ConstitutiveModellingandParameter Selection 9-4
9.4 Modellingof ConstructionandLongTermEffects 9-6
9.5 ComputationalDifficulties 9-8
9.6 ObtainingDesignOutput 9-10
9.7 Recommendations for FurtherWork 9-12
REFERENCES
APPENDICES
A Detailsof casehistories
B Supplementaryplots for numerical studies
C Derivation of equations for stress smoothing
(vii)
NOTATION
A (of
cross-sectionalarea wall, soil sample, etc.)
A, B Skempton'spore pressurecoefficients
A, B, C coefficients in the Jardine et al. non-linearelasticmodel
A matrix linking element stresses to nodal forces
B ratio defined in terms of principal effective stresses a, ', a2', a3'
B matrix relating element strains to nodal displacements
C, coefficientof consolidation (vertical drainage)
D depth of penetrationof wall below formation level
D matrix relating elementstressesto strains
E Young's modulus(drainedV or undrainedEj
E* constrained (oedometric) modulus
Eh Young's modulusin horizontal direction
E. Young's modulusat ground surface
E, Young's modulusin vertical direction
E,, Young's modulusof wall material
F nodal force
F vector of nodal forces
L. vector of nodal forcesrepresentingunloadingdue to excavation
G shearmodulus
G,h shearmodulusin v-h plane
H retainedheight (depth of excavation)
gradientof Hvorslev surfacein q: p' space
layer thickness
IL height by which pore water pressureheadhasbeenlowered (by drainage)
secondmoment of area
yield criterion type
Jacobianmatrix
K finite elementstiffiiessmatrix
K bulk modulus
le elementstiffnessmatrix
K,, coefficientof earthpressure: active condition
Kf bulk modulusof equivalentpore fluid
Ki coefficientof earthpressurepost-installation
K. coefficientof earth pressure at rest (= ciýh/d, )
K. t in
coefficientof earthpressureat rest terms of total stress crda, )
Kp coefficientof earthpressure: passivecondition
K, stiffnessratio betweensoil and structure
K,, bulk modulusof water
L length (of wall, or structuralmember)
L flow matrix in coupledanalysis
LR non-dimensionalgeometricratio defining meshgrading
LR95 value of IR at which parameter is 95% of 'Irue7value
LR99 value of ILR at which parameter is 99% of "true7' value
(Viii)
M wall bendingmoment(+/- prefixsignifies+ve or -vemoment)
Mt bendingmomentat toe of wall
N numberof increments in a block
numberof elements in a mesh
N matrix of interpolation (shape)functions
Ne numberof increments for porewaterpressureequalization in a coupledanalysis
Ni shape function for stresssmoothing
NN numberof nodesin a finite elementmesh
OCR overconsolidation ratio cr,,)
P pointloador force
axialforce in prop (compressive)or anchor(tensile)
P. prestress load (in ananchor)
out-of-balance force at a node
Q shearforce (in wall)
Q, T characteristicmeshdimensionsusedin gradingstudies(r-refinement)
R non-dimensional geometricratio for FE mesh(X(H or Y/W)
Ri diagonaldecayratio
R(i) loadincrementratio (ER(i)= 1.0)
R95 valueof R at whichparameteris 95%of "true" value
RD valueof R at whichparameteris 9911/o of 'Irue7'value
S verticalgroundsurfacemovement(+/- prefixsignifies+veor -ve Y direction)
S global smoothingmatrix
Se elementsmoothingmatrix
To timefor porewaterpressureequalization
T" timefactorfor consolidation (verticaldrainage)
UV averagedegreeof consolidation (verticaldrainage)
V excavationheave
W half-widthof excavation
X distancefrom backof wall to far edgeof mesh
Y distancefrom formationlevelto baseof mesh
YO referenceelevationfor parameter varying with depth
YR yieldratio (growthof Cain-clayyieldlocus)
a, b Henkel'sporepressurecoefficients
a., a,, .. displacementapproximationfunctioncoefficients
a, b, c polynomialfunctioncoefficients
ai lengthof an elementside
a vector of nodaldisplacements
b depthof webon 'T' sectionwall panel
b, bodyforcevector
cl, effectivecohesion
CU undrainedshearstrength
CUO undrainedshearstrengthat groundsurface
cw wall adhesion
OX)
d diameter(of a pile)
depthof webon 'T' sectionwall
d vectorof displacements at a pointwithin anelement
e voidsratio
err equilibriumerror
fb vectorof equivalent nodal loadsfor body forces
f yield function
fd correctionfactor for heavecalculation
fe RHSvectorin smoothingprocess
f, ditto
ft vectorof equivalentnodalloadsfor appliedtractions
9 acceleration due to gravity
h thicknessof prop slab
total head
heightof elementcentroidabovea particularelevationon a wall
intervalsizein finite differencemethod
nodespacingin finite elementmethod
k coefficientof permeability
kh coefficientof permeability in horizontaldirection(or k.)
k, meshgradingratio
k,, coefficientof permeability in verticaldirection(or ky)
k,, normalstifffiess(of aninterface)
k. shearstifffiess(of aninterface)
M. meshgradingelementsidelengthratio
rateof increase of Young's (E'
modulus or Ej with depth
Ma multiplyingvectorcontainingIs or Os
mqmrm. mesh grading element side length ratiosin particularregions
MV coefficientof volumecompressibility
n degreeof anisotropy= Eh/F,
P mean stressinvariant (axisymmetry)
appliedpressure
pC maximumpreviousmeaneffectivestress
q deviatoric; stressinvariant(axisymmetry)
r elementaspectratio
S meanstressinvariant(planestrain)
gradientof tensilecrackfinein q:p' space
spacingof piles(centreto centre)
Sh horizontalspacingor props
S, vector of local consistentstresses
t wall thickness
time
thicknessof aninterfaceelement
deviatoricstressinvariant(planestrain)
te timetakenfor porewaterpressureequalization in a coupledanalysis
t vector of appliedtractions
(x)
u porewaterpressure
in x direction
displacement
ue excessporewaterpressures
Uf finalporewaterpressure
UO initialporewaterpressure
v displacement in y direction
w Gaussian integrationweightingvalue
width of 'T' sectionpanel
intensityof UDL loading
X,y Cartesianco-ordinatedirections
z depthbelowgroundsurface
Z depthbelowformationlevel
ZW depthto groundwater
A. percentage error in computedstress

(D matrix term depending on in
excessp.w. p. gradients coupledanalysis
IF value of output parameter
]F* "true" value of output parameter
A ratio of current stiffnessE,, to that at a referencestrain
U ratio of K,, to K!
fmite differencegrid spacingratio
in
coefficient theJardineet al. non-linearelasticmodel
x degreeof nonhomogeneity of soil stfffness
6 wall horizontaldisplacement
8f effectiveangleof soil-wall.ffiction
5x incrementalx displacement
8y incrementaly displacement
normalstrain
volume strain
C strain vector
angleof internalffiction
arbitraryshapefunction
shearstrain
unit weight
coefficientin theJardineet al. non-finearelasticmodel
Y.. unit weightof concrete
YW unit weightof water
TI localco-ordinate
V Poisson'sratio
0 rotation (at a node)
angleof rotation of stressaxes
Lode angle(stressinvariant)
factorin two-pointfinitedifferenceapproNimation (0:5 0:!ý 1)
angleof inclination(to thehorizontal)of a groundanchortendon
(xi)
a normalstress
a stressvector
Ir shearstress
Tf shearstrength
local ordinate
w angleof dilation
2x2 (reduced)numericalintegrationrule for 2-D elements

3x3 (full) numericalintegrationrule for 2-D elements
subscriptsetc.
a axial
ave average
c centroidal
e excess
equalization
excavated (sideof wall)
h horizontal
In mean
max maximum
min minimum
i initial,post-installation
0 initial(in-situ)
at (ground)sufface,
r retained(sideof wall)
r, s rough,smooth(boundary)
U undrained
v vertical
volumetric
X, Y,z x, y, z direction(Cartesian)
95,99 95,99 %
superscripts
e element
elastic
p plastic
I effective
prefixes
A incrementOarge)
6 increment(small)
d increment(hifiniteshiW)
suffixes
(X) varyingasa functionof horizontaldistance;
a profile
(z) varyingasa fimctionof depth,a profile
(xii)
ABBREVIATIONS
AGS Associationof GeotechnicalSpecialists

APM appliedpressuremethod
AR global aspectratio of finite elementmesh
ASCE AmericanSocietyof Civil Engineers
BEM boundaryelementmethod
BFC3m best-fit curve smoothing(of 3x3 Gausspoint stressesin multipleelements)
BFL2 best-fit line (of 2x2 Gausspoint stresses)
BFL3 best-fit line smoothing(of 3x3 Gausspoint stresses)
BFP3 best-fit planesmoothing(of 3x3 Gausspoint stresses)
BMD bendingmomentdiagram
BRE,BRS BuildingResearch Establishment or Station
CAD/CAM computer-aided design/computer-aidedmanufacture
CIRIA ConstructionIndustryResearch andInformationAssociation
CSSM critical statesoil mechanics
CSB constantstrainbaror beam
CSL criticalstatefine
CST constantstraintriangle
CuST cubic straintriangle
d.o.f. degree(s)of freedom
DTp Departmentof Transport
ECSMFE EuropeanConferenceon Soil MechanicsandFoundationEngineering
EPBM earthpressurebendingmoment
EPSF earthpressureshearforce
EPSRC EngineeringandPhysicalSciencesResearchCouncil
ESP effectivestresspath
FDM firýiitedifferencemethod
FE/FEA/FEM finite element/ finite elementanalysis/ finite elementmethod
FIM full installationmethod(of modellingwall construction)
FL formationlevel
FHWA FederalHighwaysAdministration (US)
GWL groundwaterlevel
HILE homogeneous isotropiclinearelastic
HPA horizontalplananalysis
ICE Institutionof Civil Engineers
IP integrationpoint (Gaussian)
ITNE flag indicatinguniform (=O)or varying(=1) time intervals
LHS left-handside(of an equation)
LSB/LST/LSQ linearstrainbar (beam)/ triangle/ quadrilateral
(Xiii)
LSTpALSQp linearstraintriangle/ quadrilateralwith excesspore water pressured.o.f
LSBF least-squaresbestfit
NUT Massachusets Instituteof Technology
MNR modifiedNewton-Raphson
mom multiple-overlaymethod
NAFEMS NationalAgencyfor Finite ElementMethodsandStandards
NINC total numberof increments
NLR nonuniformload ratios
NOIB total numberof incrementblocks
OCR overconsolidationratio
OoB out-of-balance(load)
PCG preconditionedconjugategradient(solutionmethod)
RHS right-handside(of an equation)
SA3 simpleaveraging
SFD shearforce diagram
SI siteinvestigation
SSI soil-structureinteraction
T[R]RL Transport[andRoad]Research Laboratory
TSP total stresspath
UDL uniformlydistributedload
ULR uniform load ratios
VSA vertical sectionanalysis
WIP wished-in-place
WSBM wall stressbendingmoment
WSSF wall stressshearforce
Computer programs
CRISP CRItical State Program

FLAC Fast Lagrangian Analysis of Continua
FREW Flexible REtaining Wall program
ICFEP Imperial College Finite Element Program
WALLAP WALL Analysis Program
(Xiv)
CHAPTERI
GENERAL INTRODUCTION
1.1 Historical Background
Over the past thirty years, finite elementshave undergone something of a revolution in terms
of their use in engineering design. In the early eighties, engineersrequiring to use finite
elements in a design context would have to run the analysison a mainframe or midi computer;
perhaps located on the premises or at head office, but just as likely by renting time on a bureau
computer.
Finite element analysis(FEA) on a desk top computer was not really feasible with the first
(8086) PCs, owing to excessiveprocessingtime and the constraints of limited memory and
disk capacity. The PC AT with its 80286 chip was stiff not really adequatefor the task, but
the launch of the 80386 in 1987 changedthings permanently. At last engineershad accessto a
hardware platform which brought desktop FEA within the range of even the smallest
company, and not just a single-stepelastic analysiswith a few tens of elements.
Exponentialprogressin the computerindustryhasbroughtthe 486 andPentiumprocessors

with enoughRAM and disk spaceto copewith the analysis
nonlinear of relativelylarge,
meshes.Of course,the power offeredon other platformssuchasthe SunSPARCandDEC
Alpha hasremainedaheadof the PC, but the importantthing is that engineerscannow do
meaningful,realisticanalysison the desktop. And they are doing them.
The late Bruce Irons foresaw this very clearly in 1980, when he wrote:
"The new marketfor finite elementswill be dominatedby the ordinarydesigner.

Every smalldesignteamwill havea local uncostedcomputer,effectivelya
jobs, almostdaily,
modernisticsliderule. The designerwill submithis ten-element
without evenconsultinghis sectionleader. The total numberof suchjobs will be
astronomical,andtherewill be no good engineeringreasonto restrictthem.
Rather,what stiflesthe creativedesigneris the presentemphasison largejobs,
'stress-checking'a big, complicatedschemetoo late for radicaldesignchanges."
(Irons and Ahmad, 1980)
1-1
But clearlythis cannotbe without attendantpitfalls; Irons goeson to say:
"I am never surprised to seeyoung people who are well-versed in the theory, but
less effective as trouble-shooters than somebody with little theory but with a
strong intuitive senseof what is happening physically."

(Irons andAhmad, 1980)
Writing nearly 10 years later, Zienkiewicz asserted:
"The revolutionbroughtaboutby the combinationof numericalmethodsandthe

computer,which permitsthe solutionof most complexengineeringproblems,
placesa responsibilityon the professionalengineerto usethis power to produce
betterand saferstructuresandother engineeringartefacts.While muchof this can
be doneby the useof 'canned'automatedprogramsthereis nevera possibilityof
producing'foolproof ones. For this, andindeed,other compelling it
reasons, is
essentialfor the engineerin control to havea full understandingof the theory and
assumptions of the procedureson which he is relying. This understanding must
rangefrom the mathematicalmodelsandthe knowledgeof the underlyingphysics
to familiaritywith the approximationproceduresusedandtheir limitations."
(Zienkiewicz, 1987)
This statementmaybe contrastedwith a morerecentarticleby Clark (1994),which describes

the proliferationof FE packageslinked to computer-aideddraughting(CAD) andsolid
modellingwhich arenow on the market. Someextractsare quotedbelow (with italics added
for emphasisby the writer):
...... finite elementanalysisis no longerthe preserveof stressengineersand

rocket scientists. Systems arenow fast, affordableand safeenoughfor the non-
specialistto use ..... the trend in recentyearshasbeentowards..... providing
sophisticatedstandaloneanalysiscapability for highly qualified(but) non-specialist
designengineer Non-specialistdesignengineerswant the softwareto do the
. .....
meshing- they do not feel comfortablewith this part of FEA. And the chances
are that if they the the
use automaticmeshers, quality will be "
reasonable.
(Clark, 1994)
1-2
Onesoftwarecompanyhasevenintroduced'meshless'analysiswhich:
...... only requires you to specify material properties, loads and boundary
conditions and the software does the rest ..... there is little or no interaction with
the user
ibid.
Someauthoritieshaveexpressedgraveconcernover the releaseof sophisticatedsoftwareinto

the public domain,arguingthat it is easyfor noviceusersto makeseriousmistakeswithout
realizing it. The writer'§ympathizes
with this argument- indeedit wasoneof the motivations
for this thesis- but hasarguedelsewhere(GunnandWoods, 1993)that asfinite elements
"cannotbe uninvented"surelythe right approachis to educateandtrain engineersso that they
usethe tool properly.
1.2 Finite Elements in Geotechnical Design
The earlyfocusof finite elementmethodsin the late 1950swaswithin structuralengineering

(specificallystressanalysisof airframes),beinga naturalextensionof the stifffiessmethod.
Geotechnicswasa relativelate-comerin civil engineering, but wastedno time in establishing
an enthusiasticandindustriouscommunity(or sub-culture,accordingto Peck, 1985). The
internationalconferenceserieswhichbeganat Vicksburgin 1972hascontinuedup to the
presentday,spawninga numberof paralleleventsandassociated journals. Finite difference
andboundary elementmethods have success
enjoyedreasonable in geomechanics, but there
canbe little doubtthat the FEM hasbecomepre-eminentin this particularareaof numerical
modelling.
Most (if not all) reportedapplicationsof geotechnicalfinite elementsin the earlyyears

originatedfrom academicresearchers in North AmericaandEurope. Part of the reasonfor
this was the non-availabilityof suitablecodesfor industrialusersuntil the early 1980s(and
desk-toppackagesanother10yearslater). GeneralpurposecommercialFE codesbeganto
emergethroughthe 1970s,but thesehadlimited applicabilityto modellinggeotechnical,
problems. The major shortcomingswere an inabilityto:
1-3
a) handletwo- or three-phasematerials,
b) offer suitableconstitutivemodels,
C) modelcomplexin-situ stresses,
d) permit changinggeometry(excavationandfilling), and
e) allow for time-dependent
couplingof deformationandpore fluid flow.
CRISP was one of the first conunercially available geotechnical FE packagesin the UK and
(to a limited extent) overseas;its history is describedmore fully in Chapter 3. Computing
houses such as SIA and Strucom.offered FE bureaux services,though not with purpose-
written geotechnical codes; choosing rather to adapt general purpose systemssuch as ANSYS
(a practice which continues to this day in many quarters). Some consultanciesdeveloped and
used their own in-house codes and some universities made IFE software available in modular
form (notably Manchester with the NAG fibrary). But it was, again, the arrival of the 386 PC
which stimulated new offerings and by the turn of the 1990spractising geotechnical engineers
had accessto PLAMS (Delft University), Z_SOIL PC (ZACE Services,Zurich), SAFE (Ove
Arup/OASYS, London), and CRISP-90 (Cambridge University) in Europe.
Someof thesevendors(notablyCambridgeandDelft) recognizedthe needfor educatingusers
- andnot just sellingthem powerful software andorganized training

residential coursesof
-
days'
several duration. But noneof thesewere and
compulsory, no industry watchdogwas in
to
existence dictatetraining standards
andrun a certification to
scheme licenceIFEanalysts.
Of course,it is not merelythe availabilityof the hardwareandsoftwarewhich hasprovided

the motivationfor increaseduseof finite elementin g6otechnicaldesign. Todayengineersare
attemptingto build genuinelymore complexstructures,often to replaceexistingbuildingsin
congested,urbanareas.New structurestendto be deeper,wider andtaller (or thereis little
point in buildingthem),maximizingsiteusageby going ascloseto the limits aspossible.
Thereis naturalconcernaboutpotentialdamageto adjacentstructures(buildings,deep
basements,tunnels,buriedservices,etc.) - concernbroughtinto focusby the increasingly
litigious natureof societyandthe needto demonstrateacceptablemovementsprior to
planningconsentbeinggranted.
Finite element analysishas played an important role in the implementation of the Observational
Method (Peck, 1985). FEA provides an analytical tool for making initial predictions, and
1-4
helpingengineersto set"trigger" levelsfor groundmovementduringconstruction.Observed
behaviourin the earlystagescanbe usedto calibratethe FE modelandreviseit asnecessary.
If alternativeconstructionstrategiesneedto be investigated(perhapsbecausemovementsare
excessive),this canbe donein a convenientmanner.
1.3 Background to Present Work
Since 1989, the writer has been actively involved in the use of finite elements(specifically the
CRISP package) in designing earth-retaining structures. This involvement has beenwith real
projects which were subsequently constructed, or (in one case) with the back analysisof an
existing wall. The majority of analyseswere carried out on behalf of consulting engineers,
either to verify certain aspectsof their own design prior to going out to tender, or to check
alternatives being proposed by contractors. It is not uncommon for contractors to use FE
methods in preparing these alternative designs;indeed the writer has carried out such analyses
in two instances.
Typicalrequirementshavebeento estimatewall deformationsandassociatedground

movementsfor a wall whosedimensionshavebeenestablished by other (simpler)methods.
Sometimes,finite elementshavebeenusedto benchmarkmoreroutinecomputersolutions
which purport to give wall deflections,bendingmoments,etc. It is commonfor the routine
analysesto be carriedout by structuralengineerswith little or no appreciationof the
compleNities of soil behaviour
- the geotechnicalengineers are consultedfor soil parameters
and are often the ones who carry out more rigorouschecks on key cross-sections,but they too
may lack a full understandingof the interactions(thoughfrom the structuralside).
At the outset, the writer was struck by the number and complexity of decisionswhich had to
be made in the course of setting up an FE model for any soil-structure interaction (SSI)
problem, and the paucity of information available to guide engineersin making these decisions.
Furthermore, once results had been obtained there was little guidance on how to determine
their reliability and establishthe overall admissibility of the analysis. Given time, it would be
possible to carry out parametric studies, but commercial pressuresoften dictate that only one
or two analysescan be carried out, and they have to be right first time.
Some assistancewas potentially offered by the National Agency for Finite ElementMethods and
Standards(NAFEMS) in 1984 in their "Guidelines to Finite Element Practice", and there is no
1-5
questionthat greatbenefitwould accrueto anyindividualor organizationseekingto
implementthe generalrecommendations of that document.However,everysingleexample
was drawnfrom pure structuralengineering- nothingrelevantto geotechnicswasincluded.
Certainly,NAFEMS exhortedthe designersto verify everystepof the analysis,andto subject
the resultsto expertscrutiny,but offeredno specificinstructionson what to look for in a
retainingwall analysis.It was nearly 10 yearsafterNAFEMSbeganaddressing benchmarking
andvalidationissuesthat it startedto considergeotechnicalfinite (Peshkarn,
elements 1993).After
another gapof 8 years,
a draft document for was
consultation issued (NAFEMS,2001).
Oneotherpublicationwhich appearedin the sameyear asthe originalNAFEMS guidelines

andwhich wasaimedspecificallyat retainingwall designerswas CIRIA Report 104.
However,despitebeinga comprehensive andauthoritativedocument(evensome17 years
later), finite elementsreceivedonly a brief mention,but werethe subjectof an accurate
prediction:
"... SophisticatedFE programsincorporatingelasto-plasticmodels are

....
researchtools at present andthey arenot commerciallyavailableto designers.In
the future, theseare likely to play an increasinglyimportantpart in the
of designtechniques...
advancement "
(PadfieldandMair, 1984)
A subsequent publicationof relevancewas the Associationof GeotechnicalSpecialistsGuide

on "Validation andVerification of GeotechnicalSoftware"(AGS, 1994). This document
madesomegeneralandusefulpoints aboutFE software(which everyusercould readto
his/heradvantage),but it did not (andwasnot intended)to coverthe subj ect in depth.
At the time of writing, a revisionof CHUA 104(CP96)hasjust beenreleasedto a restricted

audienceof participatingfunders;absencefrom the public domainhaspreventedreviewin this
thesis,but it is claimed(Gabaet al, 2003)to containmoreguidanceon numericalanalysisthan
its predecessor.The Co-operationin ScienceandTechnology(COST) Action C7 hasalso
just producedsomeguidelineson numericalanalysisin soil-structureinteraction(Potts et al,
2002),but reviewhas,again,not beenpossiblein time for this thesis. Nonethelessthese
developments
areto be welcomed.
1-6
There appearto be two fundamentallydifferentwaysin which finite elementscanbe usedto
assistretainingwall design:
(a) the ývhatif'approack wherethe sensitivityof a particularcalculatedvalueis

examined so that a designcanbe improvedin a qualitativeway, or critical
designparameterscanbe identified
(b) the 'absolute'approach,wherevaluesarerequiredto be realisticand accurate,

suchaswhenattemptingto predictthe displacements
arounda diaphragmwall.
There is an increasingtrend towards using FE analysesin the secondway. Programmers are

generally optimistic that FEA will always yield the truth,, but many practising civil engineers
have developed a strong mistrust of FE output. It is certainly true that it is easyfor an
inexperienced engineerto produce results which appearridiculous to any person with real
experience of the problem being solved.
At the time whenthe writer beganto carry out retainingwall analysesin earnest,he already
had over 5 yearscontinuousandintensiveexperiencewith geotechnicalfinite elements,and
CRISPin particular. Yet this wasnot a wholly adequatepreparation,andit took severalmore
yearsto build up specificexpertisewith retainingwalls beforea level
reasonable of confidence
could be claimed. Retaining is
wall analysis oftenparticularlydifficult because:
* geometry is usually complex, and changesincrementally
o both short and long term behaviour is required
e therearehigh contrastsin materialstiffness
accuratepredictionsof groundandwall movements,structuralforces,andprop/anchor

loadsarerequired
In the courseof his consultancywork and otherindustrialcontacts,the writer hasencountered

numerousexamplesof poor FE modellingin retainingwall design. Much of this work tendsto
be doneby youngerengineerswho are both computerliterateandnumerate,but do not have
sufficientexperienceandunderstandingof SSI to be ableto identify erroneouscomputations.
Seniorengineerswho havethe responsibilityof checIdngandsupervisingtheseanalysesare
1-7
generallyableto spotcomputedresultswhich areunexpected,but are oftenunacquainted
with
the softwareor the underlyingtheory. Theymaybe ableto identify inaccuraciesand
inconsistencies without knowing how to correctthem. This unsatisfactorystateof affairsis
fikely to prejudicethemagainstmodemnumericaltechniques.
The way forward is far from clear. Some experts argue for a virtual ban on FE analysis
outside of their own ranks. This is an attractive route for the busy engineerwho wants reliable
computations and has ready access to such an expert. But how win others ever learn if FEA
becomesthe sole province of the few? In any event, FEA is now commonly available to
engineersand will cost increasingly less to perform. What is neededis better education and
the provision of guidelines for different types of application, such as retaining walls.
Whenusingfinite elementsfor retainingwall analysistherearemanyfactorswhich are

potentiallyableto affecta desiredresult. What is requiredis:
a knowledgeof the relevantfactors,andhow they caninfluencethe results
0 of whereandwhen (andwhy) problemsare known to haveoccurred

an awareness
0 strategiesfor checkingfinite elementoutput,to help ensurethat the resultsare reasonable
Another strandto the backgroundof the presentwork camefrom variouscommunications

with other CRISP users in the period 1984-90. During this time, anomaliesandunexpected
resultshad croppedup in the analysisof a variety of problems.
geotcchnical Thesecould have
beendueto:
a) bugswithin the codeitself (programnot worldng asstated)

b) usererrorsin the input data (programnot solvingthe problemintended)
C) inappropriateuseor misunderstanding
of finite (program
elements not beingusedas
envisaged)
d) genuine,thoughunanticipatedresponse(programbehavingcorrectlybut not as
expected)
Obviously,(a) is beyondthe control of the user(thoughhe/sheis not without responsibility),

whereas(d) is beyondhis/herexperienceor knowledge.The majority of problemsin design
1-8
will be classifiedunder(b) or (c), althoughwith (d) thereexiststhe possibilityof confusion
when apparentlyincorrectand/orcounter-intuitiveresultsarereturned.
In order to obtainmoreformal descriptionsof theseanomalousresults,the writer composed

to the world-wide CRISPuser'scommunityin March 1990.
and distributeda questionnaire
From the returns,it becamevery clearthat a largeproportionof the perceivedproblemsarose
in the contextof retainingwall analyses.
Bringing togetherall of thesestrands,it was clearthat an in-depthstudywas requiredin order

to betterunderstandthe natureof the problems,how to recognizethem,andhow to avoid
them. Eachmodellingdecisiontakenin the courseof an FE analysisof a retainingwall
potentiallyintroducesan error in the calculatedquantities,andat presentthereis no body of
guidanceto inform an engineerwhat theseerrorsare,andif they accumulateor canceleach
other out.
1.4 Objectives of Present Work
A significantamountof researchwas conductedin the UK duringthe 1980sand 1990sinto

the numericalmodellingof retainingwall behaviour.However,the emphasiswas not on
providinganswersto the questionsaskedmostfrequentlyby thoseseekingto usethese
techniquesin design. Much of the effort was directedat improvedconstitutivemodelling,to
try to get a betterfit betweenobservedandcomputed'behaviour.Essentially,this is the
pursuit of improvingthe 'absolute'predictions.Whilst the writer recognizesthat this is a valid
line of enquiry,he believesthat other factorscaninfluencethe outcomeof an analysisjust as
much(if not more),andthat theseneedto be investigatedin a comprehensive andconsistent
manner. Thereis little point in usingsophisticatedstress-strainrelationshipsfor the soil
materialsif a poor meshintroducessignificanterrorsfrom a differentsource.
I Theuseof theword 'computed'asopposedtopredicted'is deliberate; as Simpson(1984)has

pointedout,thecomputerprogramproducesa resultbut it is theengineerwhopredictswhatwill actually
takeplace. It maywell bethatthecomputedansweris whollyacceptable andthenbecomes theengineer's
prediction- morelikelycertainaspectsofthe computersolutionwill bemodified,to accountfor known
behaviouror for deficiencies
in theanalytical/mathematical
modelbeingused.
1-9
This thesisattemptsto draw attentionto the majorissuesinvolved;highlightingthe potential
pitfalls aswell asprovidingpracticalstrategiesfor gettingaroundthem. Many of the
examplescited in this paperhavebeenencounteredwhilst usingthe CRISPpackage(Britto
andGunn, 1987)but all FE analysiscould, potentially,havesimilarproblems.Five principal
areasof concernhavebeenidentifiedasfollows:
a) Geometricmodellinganddiscretization
b) Constitutivemodellingandparameterselection
C) Modelling of constructionand long term effects

d) Computationaldifficulties
e) Obtainingdesignoutput
Numericalexperimentshavebeencarriedout in orderto:
0 establishsomeguidelinesfor describingthe geometryof a problem(e.g. sizeof mesh,

numberof nodes,etc.)
0 evaluatethe importanceof includingvariouslevelsof sophistication(e.g. plasticyield and

flow, interfaceelements,etc.)
0 devisestrategiesfor testingthe 'goodness"of the solution(e.g. vary boundaryconditions

to establishadequacyof meshdimensions)
compare different ways of (apparently) achieving the sameobjective (e.g. total v effective
stress approach to undrained analysis,use of wall stressesor earth pressuresto evaluate
bending moments, etc.)
0 try to understandwhat is taking placewhenunexpectedthingshappen(e.g. oscillating

stressdistributions,high out-of-balanceforces,etc.)
Theseexperimentshave(of necessity)beencarriedout on idealizedretainingwall geometries

in order to establishgeneralprinciplesandoutcomes.The principalobjectivehasbeento
provide adviceandguidancein a form suitablefor, andrelevantto, geotechnicaldesigners.
1-10
1.5 Layout of Thesis
ChapterOnehasgiventhe backgroundto the work presentedin the remainderof the thesis.

In summaryform, it hasput forward the argumentsandmotivationfor the work which is
aboutto be reported.
ChapterTwo presentsa reviewof the literaturerelevantto finite elementanalysisof retaining

walls. It beginswith an historicaloverviewof the subject,tracingthe route to presentday
attitudesand identifying
practices, the investigators
who havehad the most significantimpact.
The currentstateof the art is then defined;examiningwhat FE is beingusedfor (andwhy),
obstaclesto its use,andpossiblealternatives.The chapterconcludeswith an analysisof the
literaturein which specificdetailsof all known referenceshavebeenclassifiedandtabulated,
and important trends/biases identified. This hasbeendonein a way which shouldbe most
helpfal to the practitioner,by giving an indicationof the qualityandusefulnessof each
referencein a designcontext.
ChapterThreeprovidesa detaileddescriptionof the CRISPfinite elementpackage,which the

writer hasusedin all of his work - both for jobs,
commercial andfor the more fundamental
studiesreported herein. The emphasisin this chapteris on the modellingof embedded
retainingwalls andthe particulardemands andchallenges which canarisein this context.
Following this, ChapterFour dealswith the first majorareaof enquiry,namelygeometric

modellingand discretization.An is
examination madeof sizeof mesh/domain,boundaryfixity,
numberof elements, mesh grading/density,
andtype of elements (shape/order).
All of this is
the
performedusing element typesand other meshfacilities in
available CRISP,which are
typical of thosefound in manygeotechnicalFE packages.
Next, ChapterFive considersthe decisionsandproblemsinherentin constitutivemodelling

andparameterselection.Featuresof soil behavioursuchasanisotropy,nonhomogeneity, non-
linearelasticity,andplasticyieldingare subjectto scrutiny,in an attemptto identify which
attributesmattermostin a givensituation. A deliberatedecisionhasbeenmadeto concentrate
on thoseconstitutivemodelssuppliedin the commerciallyavailableversionsof CRISP,rather
than in anyof the 'customized'versionsproducedwithin the academiccommunity. Thereare
sufficientuncertaintiesandquestionsfor designersto tacklewith just the standardconstitutive
models.
1-11
Following on from this, ChapterSix presentsan investigationinto the modellingof wall
constructionand long term effects. This includeswall installation,
bulk excavation,the
removal of temporaryprops,partial long-term
drainage, equalisation,
and level
groundwater
fluctuations.
ChapterSevenaddresses the variousnumerical/computationalproblemswhich canoccur in a

retainingwall analysis.Initially, a frameworkof is
understanding laid for the sourcesof such
problems,andtheninvestigationsareconductedto exploretheir influenceon thosecomputed
resultswhich matterin design. Topics covered include stiffnesscontrasts,elementaspect
ratios, effectiveversus total stress time
analysis, stepsin coupledconsolidationmodels,stress
concentrationsandoscillations,andnon-linearsolutionstrategy.
Chapter Eight examinesways of obtaining output which is specifically useful in a design
context. This is a significant issue, as sometimesthe very 3ygyin which output is viewed can
be as important as the output itself. Processing of primary data to derive other quantities is
considered here (e.g. the calculation of bending moments from transversestress distributions
in the wall, or from the external earth pressures).
for
ChapterNine presentsoverallconclusions,andmakesa numberof recommendations
further work.
Becausethe five major areasof inquiry identified in Chapter I are quite distinct, the writer has
decided to begin each of the five relevant Chapters (4-8) with a brief overview of the literature
specific to that arealtopic. This is a departure from normal convention (where all literature is
reviewed at the outset in a single chapter), but was consideredto improve the readability of
the main body of the thesis. The work of other investigators (where it exists) is reviewed
immediately prior to, and clearly distinguished from, the contributions of the writer.
All figuresandtablesfor a givenchapterarepresentedat the endof that chapter. References,

however,havebeengatheredtogetherfrom eachchapterandarefistedafter ChapterNine.
Appendicesare locatedat the backof the thesis.
1-12
CHAPTER2
GENERAL BACKGROUND
2.1 Historical Development
2.1.1 Beginnings
Modem finite elementmethodsevolvedout of discretestiffnessmethodsof structuralanalysis

(e.g. Turneret al., 1956). The essentialconceptwasthat smallportionsor "elemente'of a
continuum could be consideredto behavein a simplifiedmanner. Clough(1960) appearsto
havebeenthe first to usethe term "finite elemenf',andinitial developmentwasvery muchan
intuitive affair in the handsof engineers.
The phenomenal rise of the finite elementmethodin all branchesof engineeringandapplied

scienceis well documented(ZienkiewiczandTaylor, 1989). Today,it is unquestionably the
pre-eminentnumericalmethod;especiallyso in civil andgeotechnical.
engineering,although
explicit finite differencesarecurrentlybecomingmorepopular(see2.1.5). -
The first reportedapplicationof finite elementsto soil mechanicsis dueto Girijavallabhanand

Reese(1968). As their paperincludesan analysisof a modelwall in sand,it constitutesone of
the first known applicationsof finite elementsto retainingwalls. The useof a non-finear
elasticconstitutivelaw written in termsof stressinvariantswas quite advancedfor its time. it
canbe arguedthat this heraldedthe useof non-linearelasticconstitutivemodelswhich have
goneon to dominategeotechnicalfinite elementanalysisin North Americato the presentday.
Shortly after this, key work began to emergefrom the University of California at Berkeley
under the guidance of Duncan. Clough (1969) and Chang (1969) submitted theseson the
finite element analysisof U-frame locks and deep excavations respectively. Both pieces of
work showed encouraging agreementbetween observedand predicted behaviour.

Furthermore, in the caseof U-frame locks (a type of retaining structure) finite elements
corroborated and helped to explain an unexpected and surprising observation, namely that the
walls moved in toward the lock upon flooding. Another landmark at this time was the
presentationby Duncan and Chang (1970) of their "hyperbolic" non-finear elastic stress-strain
model. Subsequently,this model has enjoyed considerablepopularity amongst geotechnical
finite element analysts,used in more than 20% of documented retaining wall analyses.
2-1
In his otherwise excellentstate-of-the-artreview at Mexico City, Peck (1969) fails to make
any mention of the finite elementmethod and its potential use in retaining wall analysis. A
more enlightenedand optimistic attitude was shownby Morgenstern and Eisenstein(1970) at
the ASCE SpecialityConferenceon Earth-RetainingStructuresat Cornell. Speakingof
Winkler spring modelsfor retaining walls basedon the concept of subgrade'reaction(whose
".. deficienciesare well known..") they predicted:
"At least for two-dimensionalproblemsit seemslikely that finite element

methodsof analysiswill eliminatethe needin the future for calculationsbased
upon this concepf'
More that two decadeslater, it is perhapsdisappointingto note the flourishing market for
retaining wall designprogramsbasedon subgradereaction models. Indeed, at the 1990ASCE
Conferenceat Cornell, Kerr and Tamaro.(1990) set out to discredit finite elementanalysisin
favour of Winkler spring models, claiming the latter to be much more useful in design. A
number of their points are valid and will be taken up in detail later on, when the impact which
finite elementshavehad on retaining wall designare discussed/assessed.
Returning to 1970, it is clear that the views of Morgenstern and Eisensteinwere sharedby
some, as the next two years witnessedconsiderableactivity in the area. Three important
conferenceswith retaining wall and/or finite elementthemestook place in 1972; at Madrid,
Purdue, and Vicksburg. The key papersat eachof thesewere presentedby Bjerrum et al.
(1972), Lambe (1972), and Clough (1972a) respectively. The overwhelmingconclusionfrom
reading this early work is that finite elementswere poised to make an enormouscontribution
to the study of retaining walls and other problemsof soil-structure interaction. Bjerrum et al,
(1972), for example,used finite elementsto resolve a major debateover the role of arching in
bendingmoment reduction in flexible walls.
It is unfortunatethat, some13yearsaftertheseconferences,
finite elementmodellingwasstill
beingregardedby someasa "sub-cultural"activity(Peck 1985) andsomesix yearslater,
-
pleaswerestiff beingmadefor this "sub-culture"to be admittedto the mainstream (Ho and
Smith,1991). In the following sections,key developmentsandcontributorsleadingup to the
presentdaywill beidentified,prior to makingan assessmentof the currentstateof the art.
2-2
2.1.2 The first decade : 1970-1980
The contributions of Clough

G.W. Clough went on to establishhimself as the leading authority in finite element analysisof
retaining walls in the 1970s and early 1980s. After leaving Berkeley in 1969, Clough spent the
following 5 years at Duke University and then moved back to California to work at Stanford
University between 1974-1984. I-Escontributions include:
0 backfilledgravity walls (CloughandDuncan,1971),

0 U-frame locks (Duncan and Clough, 1971),
and
anchoredsheet-pile diaphragmwalls (Cloughet al, 1972;CloughandTsui, 1974;
Tsui and Clough, 1974; Murphy et al., 1975),
bracedexcavationsin soft clay (CloughandMana, 1976;Cloughet aL, 1979;Hansen
andClough,1980;CloughandHansen,1981;CloughandSchmidt,1981;Mana and
Clougb, 1981),
sheetpilecofferdams (Hansen and Clough, 1982; Clough and Kuppusamy, 1985),

porepressuredissipationduring construction(OsaimiandClough, 1979),and
the influenceof temporaryberms(CloughandDenby, 1977)
Throughouthis work, Cloughstroveto maintaina balancebetweencomputationalrigour and

calibrationagainstfield measurements.
He was quick to spot andexploit the greatpotential
for conductingparametricdesignstudieswith the finite elementmethod,usingit in both
forward predictionandback-analysis
modes. Clough has probablydonemore to definethe
currentAmericanstateof practicewith regardto finite elementanalysisof retainingwalls than
anyoneelse. He hasbeenrecognizedin academicandindustrialcirclesalle and,although
activepublicationceasedin the early 1990s,his influenceremains.
However,doubt mustbe castover 0 the work publishedby Clough,for two reasons.Firstly,

nearlyall of his work was conductedusingthe "hyperbolic"non-linearelasticmodel. The
deficienciesof this modelhavebeendebatedextensivelyelsewhere;sufficeto sayat this point
that manyinvestigatorshavedemonstratedthe necessityfor plasticyieldingin the constitutive
(e.
model g. Smith, 1970; Fourie, 1984;Azevedo and Ko, 1986). The second(and probably
more reason
serious) is that he almostcertainlyusedan incorrectformulationfor excavation
analysis.His error in this be
respectwould not unique,asmanywereprobablymisledby the
incorrectadvicegivenby ChristianandWong (1973).
2-3
Developmentsin Europe
LeavingNorth Americafor the present,we will reviewparalleldevelopments in Europe,
which principallyhadtheir beginnings

at Cambridge University. During the 1960s,the
Cambridgegroup conductedfundamentalexperimentalwork, inter alia, on earthpressure
problems,mostlywith laboratoryfloor models(Wroth, 1972). It was Simpson(1973) who
pioneeredfinite elementwork at Cambridge,following a ratherdifferentroute to the groupsat
BerkeleyandStanford. Simpsonimplementednon-linearplasticitymodelsfor clay andfor
sand(based on earlierCambridge work), developed an ingeniousautomaticmeshrefinement
technique(whichhasnot, surprisingly,everbeenfollowedup), andwas oneof the first to use
effectivestress for
techniques undrainedloading. The only major criticismof his work is that
his programwasrestrictedto constantstraintriangleelements,knownwidely (eventhen)to
perform ratherpoorly (e.
g. Zienkiewicz,1971).
Despite this early lead, finite element analysisof retaining walls at Cambridge continued only
as far as StJohn (1975), after which it appearedto die out. More powerful FE codes began to
emerge, pioneeredby Zytynski (1976) and culminating in the CRISP package (Britto and
Gunn, 1987), although the emphasishad shifted more towards the analysisof tunnels, shafts
and embankments. Over this period some significant contributions were made to the field by
groups at the Universities of Manchester (Smith, 1973; Smith and Boorman, 1974; Griffiths,
1980a), Sheffield (Al-Shlash, 1979) and Surrey (Rodrigues, 1975; Creed, 1979). Elsewhere in
Europe, workers at Karlsruhe (Egger, 1972) and Darmstadt (Stroh, 1974,1975; Stroh and
Breth, 1976) successfullyapplied FE techniquesto the study of propped and anchoredwalls.
The Norwegian Geotechnical Institute (NGI) used finite elementsto understanddata from
model tests and field measurementson flexible walls (e.g. Bjerrum et al, 1972).
Following his PhD work at Cambridgeandreturnto industry,Simpsoncontinuedto make

importantcontributionsto the finite elementanalysisof retainingwalls (Simpson1980,1981).
The fact that thesewere stimulatedby the demandsof "real" engineeringprojectslent thema
credibilitynot universallyenjoyedby finite elementanalysts.On the constitutivemodelling
front, his advancedelasto-plasticmodelfor LondonClay(Simpsonet al., 1979)wasoneof
the first to implementsmall-strain"threshold" stiffnesseffects,resultingin muchbetter
predictionsof wall deflectionsandassociatedgroundmovementsthanhitherto. More
recently,the "bricks-on-string"modelhasbeenpresented(Simpson,1992),which
incorporatesa form of "proportionalplasticity". Simpsonhasalsobeeninstrumentalin
2-4
devisinga simpHed form of analysisfor walls (Pappinet al., 1986),which incorporates
featuresof both finite elementandWinkler springtechniques,suitablefor routinedesign.
Burland and theBRS

Away from the universities,other pioneeringwork wasbeingcarriedout in the UK during the
1970s,led by Burlandat the Building ResearchStation(BRS). A programof instrumenting
and back-analysing deep excavationsin the London Clay was initiated by the BRS, makinguse
of linearelasticfinite to
elements back-figure profilesof (undrained) stiffness. Many of these
excavationsarenow well known - Britannic House, Palace of Westminster undergroundcar
park, CentralYMCA, BarbicanArts Centre,
and the Neasden
LaneUnderpass.This work is
reported by Cole and Burland (1972), Ward andBurland (1973), Burland andHancock
(1977), Sillset al (1977),Burland (1978), andBurlandet al (1979).
The BRS results led to the practice of using non-homogeneousand/or anisotropic linear elastic
soil models for forward predictions on other retained excavations. Despite an awarenessthat
this approach gives poor predictions of settlement/heaveadjacentto the wall, and is only
to
applicable similar types of structures (Simpson, 1981), it has been widely used in practice
up to the present time. Burland then moved to Imperial College in the early 1980s, after
which finite elementwork on retaining walls appears to have stopped at the BRS (renamed the
Building ResearchEstablishment or BRE) - although monitoring work continued (e.g. Bell
Common; Tedd et al, 1984). Once at Imperial however, collaboration with Potts initiated a
seriesof very importantcontributions..
2.1.3 Milestones of the 1980s
Potts and Imperial College
In the UKý difficultiesin designinga V2kmstretchof cut-and-covertunnelat Bell Common

(usingtop-proppedcontiguousboredpile walls) providedthe impetusfor a major finite
elementstudy. Under the guidanceof Burland (now at ImperialCollege)Class A predictions
(Lambe,1973;seeTable2.2) of wall movementsandbendingmomentsat Bell Commonwere
made,using linearelastic-perfectlyplasticsoil models(Potts and Burland, 1983). In parallel,a
fundamentalstudyof proppedand cantileveredwalls in clay wasinstigated(Fourie, 1984),
which producedmanyresultsof practicalimportance.
2-5
It was duringthe 1980sthat Potts establishedhimselfasthe pre-eminentfinite element
modellerof retainingwalls in the LJK. lEs contributionsaretwo-fold:
rigorouscomparisonof FEM with traditional (e.
analyticalmethods g. limit equilibrium
(or
method)andvalidation otherwise)of current design methods,and
applicationof finite to
elements practicalretainingwall design,
making forward
predictionsof behaviour.
The former is exemplifiedby the rigorousstudyof designmethodsfor proppedandunpropped

(Fourie
cantilevers and Potts, 1988,1989; Potts andFourie, 1984,1985,1986),in which the
influenceof earthpressurecoefficientat-rest(K,,), constructionmethod,wall restraint,etc.
were investigated.
The latter includesthe cut-and-covertunnelsat Bell Common (Hubbard et
aL, 1984;Symons et aL, 1985) and George Green (Pottsand Knights, 1985). The formation
of the Consulting
Geotechnical Group (with strongImperialCollege links) led to manymore
contributionsaimedat solvingpracticalproblemsand design issues
in retainingwall analysis.
However,comparisons betweenpredictionsandmeasurementsat Bell Commonwerepoor

(Hubbardet al., 1984;Symonset aL, 1985),promptingfundamentalrethinking. Ideaswere
beginningto take shapeat ImperialCollegeconcerningthe small-strainstiffnessandnon-linear
in
clays,resulting a constitutive
elasticityof overconsolidated modelbased on periodic
logarithmicfunctions(Jardineet al., 1984,1986). Equippedwith this new model,andan a
posteriori knowledge of the actualsequenceof eventson site,Bell Common was tackled
afresh ([-Eggins
et aL, 1989). The predictions(now classCI ratherthan classA; Lambe,
1973)agreedsomewhatbetterwith observations,althoughstill leavingroom for improvement.
An important conclusion was beginning to emerge. The 1970s practice of back-analysing

deep walls and excavationsand using it for forward finite element prediction was being called
into question. Advancesin small-strain measurementswere enabling laboratory tests to
furnish more reliable stiffness parametersfor finite element analyses. This offered the promise
of better results - or at least a more fundamental way of obtaining them, removing the
requirement for previous comparable construction in similar ground conditions.
Americanprogress
The 1980s saw rather different developmentson opposite sides of the Atlantic. Apart from
some promising excursions into hybrid finite element techniques (Lightner, 198 1; Sargand,
2-6
1981; DesaiandSargand,1984),North Americanworkerswere mainlyconsolidatingtheir
experiencewith applyingthe hyperbolicmodel. Notableexceptionswere Eisensteinand
Medeiros(1983)who appreciatedthe importanceof stresspath in soil behaviour;Desaiel al.
(1984)who developeda superiorinterfaceelementsuitablefor walls; andBoda et al. (I 989a,
1989b)who devisedan elegantandpowerful (and correct)finite elementformulationfor
excavation- albeitmanyyearsafter the Cambridgegroup had successfullyaddressedthe
problem(Gunn, 1982). Cloughleft Stanfordin 1984andmovedto the Virginia Polytechnic
Institute and StateUniversityat Blacksburg. Ms output sincethen hasincludedwork on
cellularcofferdams(CloughandKuppusamy,1985),andon constructioninducedmovement
of in-situ walls (CloughandO'Rourke, 1990)
2.1.4 Trends of the 1990s
From the early 1990s,two differentapproaches to reducingthe discrepancybetween

predictionandobservationemerged.The first concernedthe refinementof constitutive
models to includenon-linearity,small-strainstifffiess,history effects,andkinematichardening.
Notableexamplesincludethe "bricle' model(Simpson,1992),the MIT-E3 model(Whittle,
1993),and3SKH (StallebrassandTaylor, 1997). The secondroute wasto improvethe
faithfulnesswith which the constructionprocesswas simulated,with particularemphasisbeing
placedon wall installationeffectsand3D representation.Attemptswere madeto model
excavationunderbentonite slurry support,followed by wet concretebeingtremiedin at the
bottom, eventuallyhardeningto form the wall. Initial attemptsusedplanestrainand
axisymmetry(Kutmen,1986;Gunnet al., 1993). This was followed by partially-coupled
plane stress
strain/plane (De Moor, 1994;Ng et al., 1995),andculminatedin full 3D
representation(Gourvenec,1998;GourvenecandPowrie, 1999,2000).
UK investigatorstook most of the lead,andreflectedthe greateremphasisplacedin Europe

on analytical/theoretical to geotechnicalengineering,comparedwith North
approaches
America. However,therewere signsthat this was changing.Following on from Clough,the
US initiative in finite elementanalysisof retainingwalls wastakenby Boda at Stanford,Finno
at Northwestern(both former studentsof Clough),andWhittle at NET. Boda concentrated
on constitutivemodelling(Boda andKavazanjian,1985),excavationtechniques(Boda et al.,
1989a,1989b;Boda, 1990a)andnumericalprocedures(Boda, 1990b). Finnowas oneof the
first in the US to carryout coupledanalysesof retainedexcavations,and alsoconsideredthe
role of installationeffects(FinnoandHarahap,1991; Finnoet al., 1991; Finno,1992). Whittle
2-7
alsopreferredmodemelastoplasticmodelsin retainingwall analysis(Whittle andHashash,
1993;Whittle et al., 1993;Whittle andKawadas, 1994;HashashandWhittle, 1996;Whittle
andLadd, 1998). It was encouragingthat this new generationof Americaninvestigators
beganto abandonthe earlierempiricismcharacterized by the hyperbolicmodel.
Potts and the IC group continued to make key contributions into the first half of the 1990s.
Fundamental studies switched to sheet pile walls (Day and Potts, 1989,1991,1993; Potts and
Day 1990; Day, 1999), whilst design studies continued with temporary berm support (Potts et
al., 1993), relieving slabs (St John et al., 1993), and anchored earth (Harris et al., 1993).
Involvement with commercial projects included deepbasementsat Aldersgate (Fernie et al.,
1991) and Victoria Street (St John et al., 1992). A useful state-of-the-art summary of finite
element analysisof retaining walls was given at the International Conference on Retaining
Structures at Cambridge (Potts, 1993).
Symonsand the TRRL
The involvementof UK governmentresearchbodieswasnot confinedto the BRSABRE.The

increaseduseof embeddedwalls on trunk in
road schemes the 1980sinevitablyled to the
involvementof the TransportandRoadResearchLaboratory(TRRL), underthe guidanceof
the late I.F. Symons.In additionto establishinga rolling programmeof instrumentingfull-
scalewalls on differentparts of the UK trunk-road network, Symons commissioned centrifuge
modellingat Cambridge, andalso funded the developmentof CRISP to improve its SSI
(see
capabilities Chapter 3). FE wasused by the TRRL mostly in back-analysis
mode in order
to studyfield measurements takenduring andafter construction.Structuresincludedthoseat
the A329(M) in Reading(Carderand Symons,1989),NeasdenLaneunderpass(Carswellet
al., 1991), Al M Hatfield (Symonset al., 1987),A3 Malden Underpass (Symons and Carder,
1991), M25 Bell Common(Symonset al., 1985;Higginset al., 1989,1993),A406
Walthamstow(WatsonandCarder,1994).A406 SouthWoodford(Powrieet al., 1999),and
A331 AldershotRoadunderpass(Carderet al., 1997).
The TRRL carriedout someof their FE work internally,but often employedexternalanalysts

(includingthe writer). Followinga changeto executiveagencystatusin the early 1990s,the
TRRL was renamedthe TransportResearchLaboratory(TRL); field monitoringandback-
analysisof earth-retainingstructurescontinuedunderthe guidanceof D.R- Carder.
2-8
ThecontributionsofPowrie
AnotherimportantUK investigatorin the analysisof retainingwalls hasbeenPowrie,whose

contributionsbeganat Cambridge.Initial (TRRL-funded)investigationson the Cambridge
centrifugehad focusedon the fundamental behaviourof diaphragmwalls before,andat,
collapse(Bolton andPowrie, 1987,1988). Subsequently, whilst at King's College,the
potentialof the centrifugefor benchmarking
finite elementmodelswas realized(Bolton et al.,
1989). Other contributionsto FEA includedwalls singly-proppedat formationlevel (Li, 1990;
Powrie andLi, 1991a,1991b)anddeepexcavationsin Londonclay (Lee, 1990).
Powrie'smoveto QueenMaryCollegein 1989begananinvolvement in full-scalemonitoring

of retainingwalls(possiblya legacyof LA Woodat QMCin the 1980s),augmenting finite
elementandcentrifugemodelling.Investigations includedwallswith relievingplatforms
(Chandler,1995;PowrieandChandler,1998;Powrieet al., 1999),twin proppedwalls
(Richards andPowrie,1994;Richards,1995),andwallswithtemporary berms(Powrieet al.,
1993).Otherusefulcontributions includedimplementationof the"bricle'modelin CRISP
(Chandler, 1995),validationof CSSMmodelsin retainingwallanalysis (Li, 1990;Richards,
1995),anda considerable bodyof experimentalwork oninstallationeffectswhichhadmajor
implicationsfor retainingwall analysis
anddesign(PowrieandKantartzi,1996).
Sincemovingto Southamptonin 1996,Powrie hascontinuedto co-ordinatephysicaland

numerical modelling with field observationson retainingstructures.The Southampton
group
is oneof the few in the world doingthis. More recentFE contributionsincludemodelling
multi-proppedwalls (BattenandPowrie,2000; Powrie and Batten,2000),3D wall installation
effects(Gourvenecand Powrie, 1999),and discontinuoustemporary berms (Gourvenec and
Powrie,2000;PowrieandDaly 2002; Gourvenecet al, 2002).
Contnbutionsfrom theFar East
Two importantinvestigatorsin SE Asia cameto prominenceduringthe 1990s- F.H. Lee at

the NationalUniversityof Singapore(NUS), andC.Y. Ou at the NationalTaiwanUniversity
of ScienceandTechnology,Taipei. Using critical state-based
models,Lee focusedon deep
excavationsin soft clay,which werebeingdug for basementcar parksandshoppingfacilities
well below groundlevel- somereachingup to 30m in depth. The tendencyof 2D analysisto
overpredictgroundmovementsin the vicinity of deepexcavationsled Lee to initiate studies
into the importanceof comer constraints,combiningfield andnumericalwork. The NUS
group developedits own pre/post-processors
to CRISPfor 3D work (Leeet al., 1995)and
2-9
addedproper3D structuralelements.Successfulapplicationto both back-analysis
andforward
predictionhasbeenreportedfor severalkey basementprojectsin Singapore(Lee et al. 1997;
Lee el al., 1998; Chewet al., 1996).
Almost simultaneously,Ou begana similarprogrammeof research,concernedwith deep

in
excavations the interbedded
sands and claysof Taipei. Like Lee, he recognizedthat comer
effectswould be significant
and studied deep
several basement
projectsin Taipei (Ou andLai,
1994;Ou and Shiau,1998;Ou el al., 1996a,1996b,2002).Ou hasmadeimportant
contributionsto 3D analysisof retainingwalls, andis one of the few to haveusedinfinite
domainelementsin this area(Ou and Shiau,1998). Carefulstudyof casehistory dataled to
usefiA expressions
empirical describingsettlementtroughs(Ou et aL, 1993;HsiehandOu,
1998),extendingthe previouswork of Peck(1969)andCloughandO'Rourke (1990).
2.1.5 Other methods
It will be appropriateto trace developmentsin the other computer-basedmethodsfor analysing
retainingwalls, before going on to assessthe current state of the art in finite elements.
Discretespring methods
Engineersandanalystsin the both the US andin Europehavemadewide useof simpler

computationaltools for retainingwall design,with the discretespringtype of analysisbeing
the most prevalent. This is basedon a conceptualmodelof soil-structureinteractionwhere
the soil is replacedby a bed of springs(Winkler, 1867),with springstiffnessdefinedvia a
"modulusof subgradereactiorf' (dimensionsF 1;3). Wall deflectionsandthe structuralforces
in the wall andsupportsystemcanbe computed,but no informationcanbe obtainedon
adjacentgroundmovements.US experiencedatesbackto Haliburton(1968),whenthe so-
called"beam-column7' method first to
appeared,applicable both walls andlaterallyloaded
piles. As recentlyasthe late 1990s,the FederalHighwaysAdministration(FHWA) fundedan
extensiveresearchprogrammeinto anchoredwall behaviourat TexasA&M University
(Briaud andKim, 1998),culminatinginter alia in refinedbeam-columnsoftware. With
FHWA endorsement, this softwarewill be usedwidely for wall designwithin the US
transportationinfrastructure.
The use of Winkler spring programs has been widespread in the UK too, exemplified by the
WALLAP program (Borin, 1988). Examples of such programs used in design are given by
2-10
Wood andForbes-King(1989),Fraser(1992),andOnishiand Sugawara(1999). A significant
improvementwhich allowslimited springinteractionvia pre-storedFEA resultswas developed
by Pappinet al. (1986). A programbasedon this, FREW, hasbeenusedwith reported
success(GroseandToone, 1993;Wallaceet aL, 1993;Phillipset al., 1993). Subsequently,
Vaziri (1996) hasreportedthe applicationof the methodto variousback-analyses.The writer
was involved in one projectin 1993(Al/A406/A598 junction - seeAppendixA) whereFREW
wasused(mostlyby structuralengineers)to designvariousretainingwall geometrieson a
complicatedinterchange
of threemajor trunk roads. CRISPwas usedto benchmarkFREW
resultson key
several cross-sections,
providing (or
assurance otherwise)that the simpler
designsoftwarecouldbe usedwith confidence.
Other continuum methods
Whenthe finite elementmethodfirst arrived,it rapidly displacedfinite differencemethods

(FDM) asthe pre-eminentnumericalmethodfor solvingboundaryvalueproblems.However
(in geotechnicalcirclesat least),explicit finite differencemethodshaveregainedpopularity
over the past 10years. In the late 1980s,whenthe geotechnicalcommunitywas venturingto
do moreFE analysisof retainingwalls, it becameincreasinglyclearthat oneof the main
was
obstacles the lack of a good userinterface.This wastrue of all themaincommercialcodes
(CRISP,PLAXIS, SAFE,etc). Mainstreamprelpost-processing packagessuchasFEMGV or
PATRAN wereavailable,but licencecostscould not bejustified without a guaranteed
throughputof work.
At about this time emergedthe FLAC program (ITASCA, 1991), possessinga user-friendly
interface which could be leamt quite easily and, just as importantly, picked up again by the
intermittent user. It was an immediate success,but the writer suspectsthat understanding (of
the underlying formulation) may have been sacrificed for usability, as the fast-Lagrangian
formulation is not as physically intuitive as the FEM. Users were thus forced to treat FLAC as
more of a "black box7'. In the UK at least, FEM programs lost a significant market shareto
FLAC in the first half of the 1990s. Examples of applications of FLAC to retaining walls are
given by Lorig (199 1), Brooks and Spence(1993), Li el al. (1993), and Ng and Yan (1999).
Boundaryelementmethods(BENI), on the other hand,havehadno appreciableimpacton

embeddedretainingwall analysis.The natureof the problem(especiallythat of changing
andefficienciesof integralequation
geometry)virtually negatesanyof the usualadvantages
2-11
As
methods. with finite differences,
the BEM lacks the strong and
physical intuitive appealof
the FEM (only is
moreso), and often obscured by formidable and
mathematics notation. To
date,the most successfulapplicationsin geotechnicshavemostlybeenin rock mechanics
where linear be
elasticitymay considered a model
reasonable of behaviour.
material
2.2 Present State of the Art
2.2.1 Art v practice
In the early 1990s,threeconferencesprovidedan overviewof the stateof the art andthe state
of practiceof finite elementanalysisof retainingwalls at thosetimes. Thesewere the ASCE
conferenceon Earth RetainingStructuresat Purduein 1990,the loth ECSWE at Florencein

1991andthe ICE InternationalConferenceon RetainingStructuresat Cambridgein 1992.
Therewas definiteevidenceof increasinguseof FE to analyseretainingstructures,andsome
of the potentialoutlined20 yearsearlierat Comell,Madrid andVicksburg had clearlybeen
fulfilled. Basedon reportsat theseconferencesandthe progressmadeup until the present
time, the currentstatusof FE analysisof retainingwalls is summarizedin Table2.1.
Table 2.1 Present status of FE analysisof retaining walls, contrasting acadenk and
industrial practice
State of the Art State of Practice

(Academia) (Industry)
non-linear elastoplasticmodels with small- nonhomogeneous isotropic linear elastic -

strain behaviour and kinematic hardening perfectly plastic
3D analysis,fully coupled 2D analysis,drained or undrained
wall installation fully modelled wall wished-in-place
parametersdeterminedfrom high quality, routine site investigation and lab tests to

specializedlaboratory and field testing determine parameters
comprehensivemonitoring of wall, support partial monitoring (e.g. prop loads or wall

system, and surrounding ground during movement) during construction only
and after construction I
Clearly there will be some practitioners using more advancedtechniquesthan suggestedby

Table 2.1; conversely,not all universities are performing analysesembracingthe full range of
2-12
advancedfeatures.However,the tabledoesdepictwhat would be consideredcharacteristics
of (a) the leadingedgeand(b) the averageindustry-based
finite elementanalysisof retaining
walls. That thereshouldbe sucha gap betweenart andpracticeis not surprising- it is the role
of universitiesto lead,andindustrymaybe up to 20 yearsbehindin embracingnew
technology. In the UK at least,thereseemsto be a willingnessto useFE modelsfor retaining
wall design,encouraged by the exampleof the earlypioneerssuchasSimpsonand St. John,
who have activelypromoted finite elementsin geotechnicalpracticeup to the presenttime. A
credibletrack-recordhasalsobeenestablishedin SE Asia for usingFEA in deepbasement
design. In America,however,severalleadingexperts(e.g. Peck,Tarnaro)haveconsistently
criticizednumericalmodels,advocatinga mix of empiricismandsimplermethodsof analysis,
suchasWinkler springmodels. It hasevenbeenclaimedthat theFEM is confusing,andraises
unrealisticexpectations(Kerr andTamaro,1990- seealso2.2.4). Suchcommentsfrom
highly regardedengineerswill hardlyencouragegreateruseof FE in practice. In the M FE
is not without its critics,but they havebeenlessprominent.
Having briefly definedthe stateof the art andof Practice,the following questionswill now be
addressed in the contextof embeddedretainingwalls:
" what is FEA beingusedfor?

" what arethe reasonsfor usingFEA?
" what arethe obstaclesandobjectionsto the useof FEA?
" what arethe alternativesto FEA?
2.2.2 Categories of use
Finite element analysisof retaining walls can be divided into 3 principal categories:
a) designandforward prediction
b) back-analysis
andcalibration
C) theoreticalandnumericalstudies
Theseare rankedin decreasingorder of severity,in termsof whetheror not the calculated

resultsarelikely to be usedin practice. Clearly,forward prediction(ClassA in Lambe's
terminology;seeTable2.2) is the most exactingtest of a finite elementanalysisand,not
surprisingly,relativelyfew suchanalyseshavebeenpublished.In fact, manyof the so-called
"predictione' of behaviourareretrospective(ClassCI) andaremoreproperlytermedback
analyses.It is not alwaysclearhow manyresultshavebeendiscardeden route to finding
2-13
thosewhich haveactuallybeenpresented.Van Weele(1989)takesthe view that engineers
will tendto publishonly thosepredictionswhich showa good matchwith
andresearchers
observedbehaviour,for everysuccessthere maybe 15 (or more) failures.
Table 2.2 Classification of prediction (after Lambe, 1973)
Predictiontype Whenpredictionmade Resultsat time predictionmade

A Before event Not available
B During event Not yet known
BI During event Already known
c After event Not yet known
ci After event Alreadyknown
a) Design andforwardprediction
This categoryincludesanalysesthat maybe conductedaspart of the designprocessfor an

intendedretainingwall. The analysesmaybe of the "what if' type,wherealterationsto basic
soil parameters,geometry,constructionsequenceetc. areinvestigated,or the "absolute"type
of forward predictionbasedon mostcredibleparameters.In the former case,theremaybe no
intentionto comparethe resultsof suchanalyseswith subsequentbehaviour;indeed,unless
appropriatefield instrumentationis installed there
andmonitored, will be no opportunityto do
so. Nonetheless,analysesof this Idnd canbe extremelyusefulin enablingdesignersto gain
insightsinto variousfeaturesof soil-structureinteraction.
In other cases,instrumentation will be installed before and during construction, giving the
opportunity to assessthe analysesand possibly refine them. For example, if FEA has been
used as part of an observational method approach to construction, early feedback from field
performance can be used to calibrate the analysisand make better forward predictions of the
remaining performance (this was done for QueensberryHouse - seeAppendix A). Examples
of published design studies include:
anchoredsheetpiles- ChannelTunnelterminal,Folkestone(Young andHo, 1994)

anchoredsoldierpile andlaggingwalls - Seattle(Cloughet al., 1972)andHartford,
Connecticut(Murphy et al., 1975)
multi-proppedsheetpile wall - Detroit (Abedi et al., 1993)andRotterdam,(Brassinga
andVan Tol, 1991)
2-14
0 cantileveredcontiguousboredpiles- Dunton Green,Kent (GarrettandBarnes,1984)
cantilevered diaphragmswith berm - A55 Coast Road, Wales (Powrie et al., 1993)
singly-propped diaphragms- Bell Common, Essex (Hubbard et al., 1984) and Barbican
Arts Centre (Stevenset al, 1977)
twin-proppeddiaphragms- GeorgeGreentunnel,London (Potts and Knights, 1985)

multi-proppeddiaphragms- Houseof Commons,London(Ward andBurland, 1973)and
Post Office Squaregarage,Boston (Schoenwolfel al., 1992)
cellularsheetpile cofferdams- Alton, Illinois (CloughandKuppusamy,1985)and
Williamson,West Virginia (Taylor andMeadows,1990)
b) Back-analysis
A significantnumberof backanalyses(classCI predictions)havebeencarriedout on well
documentedcasehistories. At one endof the spectrum,back-analysesmaybe conductedwith
an existingFE codeandrelativelysimplemodellingprocedures,in order (for example)to
establishoperationalsoil stiffnesslevels. Examplesinclude:
0 determiningE. : depthprofile at BritannicHouse,London(Cole andBurland, 1972)
most accuratedescriptionof elasticnon-linearityof E. at ChurchillSquareLRT Station,

Edmonton(EisensteinandMedeiros,1983)
in Singapore,for forward
calibrationof hyperbolicmodelon severaldeepbasements
prediction on a new project (Bromset al., 1986)
refinedcomputationbasedon feedbackfrom initial analysesat Le Havre,France(Felix et
al., 1982)and SaltLake City, Utah (Caliendoet al., 1990)
adjustingE,,/s. to improvethe matchbetweencalculationandobservationfor walls in soft
clay in SanFrancisco(CloughandMana, 1976)andOslo (Athanasiuel al., 1991)
At the other end,the focusis on makingimproved"predictione'by changingsomepart of the

representationsuchasthe constitutivemodel,type of element,or methodof handling
excavation.Someexamplesof this are:
a retrospectivestudyof the influenceof temporarybermsin SanFranciscoandSt. Louis

(Clough andDenby,1977)
improvedcalculationof settlementtrough behindthe Houseof Commonscar park using
the LC constitutivemodel(Simpsonet al., 1979)
2-15
re-analysisof Dunton Greenwith improvedfield datafrom self-boringpressuremeter
(ClarkeandWroth, 1984)
re-analysisof Bell commonwith the Jardinemodel(fEgginsel al., 1989)
usinga coupledformulationto allow partial drainageandmorecloselyreproduce
constructionbehaviourin Singapore(Yong et at, 1989)andTaipei(Hsi and Small,1993)
introductionof sheetpile clutch slip for AIM Hatfield (Day andPotts, 1991)
improved"predictions"by incorporatinga boundingsurfacemodelandinstallationeffects
at the HDR-4 excavation, Chicago (Finno and Harahap, 1991)
re-analysisof VaterlandI subwaystation,Oslowith correctexcavationalgorithmandan
elastoplasticmodel(Ho andSmith, 1991)
improvedreproductionof modelretainingwall test resultsthroughthe useof interface
elements(Van denBerg, 1991)
0 useof "brick" modelandwall installationmodellingat Lion Yard, Cambridge(Ng and
Lings 1995)
,
using neural networks on back-analysesto make forward prediction (Goh et aL, 1995)
adjustment of soil parametersto reproduce observed rotational failure of tie-back walls on
Boston CentralArtery project (ORourke and O'Donnell(1997)
C) Yheoreticalinvestigations
This categorycomprisesthoseanalyseswhich havebeenconductedto studya particular

aspectof the finite elementrepresentation. Comparisonswith measuredvaluesmay neverbe
madesubsequently, althoughthis is quite often donein order to demonstrate
that a proposed
modificationactually works. These analysesare the furthest removed from the designprocess
andcarry the leastrisk, in the that
sense they are rarely held up to scrutinyagainstthe "truth"
of actualperformance.Examplesinclude:
studyingthe effectsof a new constitutivemodelon "predicted"wall response(e.g. Ng

andLings, 1995;Arslanet al., 1981; Jardineet aL, 1986;Simpson,1992;Hansenand
Clough, 1980)
0 investigatingthe influenceof wall fliction on earthpressuredistributionson a backfilled

wall through using interface elements(Nakai, 1985)
comparingdifferentmethodsof calculatingwall bendingmoment(e.g. Gunnand
Ponampalam, 1990)
2-16
sensitivitystudieson the influenceof far boundarylocation,numbersof elements,and
gradingof mesh(e.g. Rodrigues,1975;MorgensternandEisenstein,1970)
validatinga new formulationfor modellingexcavation (e.
sequences g. Ishihara,1970;
Chandrasekaran andKing, 1974;GhaboussiandPecknold,1984;Brown andBooker,
1985;Boýa et al., 1989)
examiningthe influenceof propping,Ko etc. on a hypotheticalembedded
wall (Potts and
Fourie, 1984,1985,1986)
(Pineloand
improvedtie-backwall modellingwith morerigorousanchorrepresentation
Matos Fernandes,1981)
developmentof new elementtypesfor sheetpile wall modelling(Day andPotts, 1993;
Bakker andBeem,1994)andinterfacebehaviour(Ghaboussiet al, 1973)
transientstabilityof excavationsusinga coupledformulation(OsaimiandClough, 1979;
Holt andGriffiths, 1992)
influenceof wall installationandconstructionsequence
on wall bendingmoments(Gunn
andClayton,1992;Gunnet al., 1993;RichardsandPowrie, 1994)
fundamentalstudiesof bracedexcavation(PalmerandKenney,1972;CloughandTsui,
1974;CloughandHansen,1981; Wong andBroms, 1989;HashashandWhittle, 1996)
" temporaryberm studies(CloughandDenby, 1977;Potts et al., 1993)

" studiesof singleproppingat formationlevel (PowrieandLL 1991a, 1991b)
" analysesof 3D effectsandwall installation(Ng et al., 1995;Lee et al., 1998;Ou et al.,
1996a;Ou and Shiau,1998;GourvenecandPowrie, 1999)
0 compactioninducedstresses
on backfilledwalls (SeedandDuncan,1986)
2.2.3 Motivations for FE analysis
What hasbeenthe maindriving force behindengineerschoosingto useFE techniquesin

retainingwall design?Someof the reasonsdeducedfrom the literatureandfrom anecdotal
evidencegatheredby the writer include:
" basicdesignscanbe refinedandmademore efficient

" simplerprogramsandmethodsareconsideredinadequateor overly conservative
" therearegenuineinteractionissueswhich cannotbe addressed
anyotherway
some aspect of the retaining structure is novel or unusual
software/hardware acquisition should be justified by using it
0 it is believedthat FE will alwaysyield the truth
2-17
FE analysisperfonnedby anotherparty hasto be checked
0 the observationalmethodis to be used,andFE is requiredto set"trigger" levels
However,the writer believesthat the currentsituation(in the UK, at least)is probablydriven

by commercialneed:
a) to obtainplanningconsentone needsto demonstratethat only smallmovementsWill be
causedto adjacentbuildings
b) to maximizesiteuse,onehasto go ascloseto the limits aspossible
C) experienceshowsthat movementsare small,generally,but only two typesof analysis
give this result:
FE analysiswith parametersobtainedfrom back-analysis
of previous
excavations(e.g. elasticanisotropic)
FE analysisusingsmall-strainnon-linearlaboratorydatain an appropriate
model(e.g. ImperialCollegetype; Jardineet al., 1986)
So the mainincentivemaynot be technical(betterdesigns)or evenfinartcial(cheaper

construction)- but ratherlegal. The atmosphereis increasinglylitigious,not helpedby high-
damagein urbanareas(e.g. collapseof HeathrowT4
profile casesof excavation-induced
NATM stationtunnelsin 1994). Clientsfor new constructionnow needto commissionmore
comprehensive andaccuratedamagepredictionson adjacentbuildingsbeforeplanning
permissionis granted. This maychangeas examplesof cost savingsattributedto usingFEA in
foundationdesignbecomemorewidely publicized(e.g. Wheeler,1998).
2.2.4 Obstacles and objections
Reportedapplicationsof finite elementsto earth-retainingstructurespresenta relativelyup-

beatpicture of the advantagesandbenefitsgained-However,it would be naiveto assumethat
acceptancehasbeenwidespreadandwholehearted.A variety of objectionsandprejudices
havebeenvoicedasfollows:
previousexperiencehasbeenbad (resultspoor or misleading)

it is very expensive(cost of time, tools andtraining)
only expertsshoulddo it (too manypitfalls for the novice)
high quality parametersanda knowledgeof actualconstructionsequencearerequired
beforeanyusefulanalysescanbe conducted
2-18
0 insufficienttime to do complexanalysis*. a "guesstimate"preferred
.
0 simplerprograms havebeen usedwith successpreviously.-.changeunnecessary
other numericalmethodsarepreferred(e.g. finite difference)
methodsareno substitutefor experienceanda knowledgeof
analytical/numerical
precedent
From occasional remarks made at conferences,it is clear that some engineershave come to
mistrust FE methods, but in the writer's opinion this arisesfrom inappropriate use of FE codes
leading (understandably)to "bad press". Just becausea method can be misusedis no reason
to reject it - rather, one must ensurethat the safeguardsare in place. Another explanation for
the mistrust is that the FEM may have appearedtoo successful:
"The divergencebetweenthe calculatedmovementsandthoseactuallyobserved

mayhelp to give engineersmoreconfidencethanpreviousnumericalstudies
which tendedto give 'good' agreementin all cases"
(Ho and Smith, 1991)
Expensecanbe a genuineissue,but not an insurmountable one (Schad,1985). A top-endPC

is quite likely to existin a typical designoffice nowadays,so theredoesnot alwaysneedto be
a majoroutlay on new hardware. Software can be bought"in perpetuity"(one-off) or licensed
on an annual basis,but it be
maysimplynot worthwhile for a company to buy its own copy if
therean insufficientthroughputof work is expected.It maybe morecost effectivefor
to
companies sub-contract their FE to
work either a bureau or an independent (e.g. university-
based)analystif their requirementis intermittent.
Trainingis the most expensiveaspectof anycomputersoftware(which is why having,for

example,a familiar Windows-styleinterfacecan savemuchtime). This is especiallytrue of FE
programs,and prior to learninghow to useanygiven code,an underpinningknowledgeof
continuummechanicsetc. is essential(Owen, 1991). Companiesoften do themselvesa
disserviceby relegatingFE work to graduateengineers- who havemoreaptitudebut less
likelihoodof stayingput andgiving a good return on the training. If theyareon a graduate
trainingagreementthey will needto spendtime on site,andso will departthe designoffice
taking the expertisewith them. Thereappearsnot to be a careerpathfor the geotechnical
finite elementspecialistexceptin the largestconsultancies
- andyet it is too demanding
to be
left to occasionalusers.
2-19
The issueof "expertsonly" is lessstraightforward.The writer is on recordassayingthat as
FE cannotbe kept out of the handsof practisingengineers,educationis betterthantrying to
prohibit access(GunnandWoods, 1993). However,he doeshavesympathywith the view
that it shouldbe expertsonly, havingseensomemajor (andbasic)mistakesbeingmade. Potts
haslong heldthe view that therearetoo manypitfalls for the novice,andwill not distribute
the programICFEPwhich he hasdevelopedat ImperialCollege(Potts,2003).
Not surprisingly, the literature does not contain many examplesof FEA being used in the
absenceof precise soil parametersand a priori knowledge of exact construction sequence.

Such analysesdo not make for exciting reading. They cannot be used for reliable forward
prediction, but can be helpful in establishingthe importance of particular parameters. In this

way, it would be possible to refine the objectives of SI and testing, becausethe parameters
which need to be measuredmore accurately have been identified. Once refined parameters
become available, they can be run with the existing mesh (which accounts for most of the time
in the initial model set-up).
Time is a real issueif deadlinesare pressingthenFE modellingmaygenuinelybe out of the

question. But asin the previouspoint, usefulanswersmaybe obtainedfrom relativelycrude
modelswhich canbe generatedquickly (Irons andAhmad, 1980).
It canbe hardto convincean engineerthat FE techniquesare neededwhensimplerprograms

haveprovenadequatein the past.However,simplerprogramsmaynot be easilyadaptedto
novel structureswhich arebeyondtheir scope,andthe inherentgeneralityof the FEM makesit
particularlyversatile. Nonetheless,the writer acceptsthe adage"horsesfor courses"- simpler
computeranalysesshouldbe usedwhenthey satisfyall relevantrequirements.FE could be
usedto determinethe requiredpenetrationdepthof a wall (Potts, 1991),but a simplelimit
equilibriumcalculationwill do more or lessthe samething in a fraction of the time. A similar
argumentcanbe madefor linearelasticityversusmore sophisticatedconstitutivemodels.
If other comparablenumericalmethodsare alreadyin use,then it is difficult to justify change-

FEM andFDM schemesshouldgive the sameresults,if properlyformulatedandcorrectly
applied. No responsibleproponentof analytical/numerical
methodsin geotechnicshasever
suggestedthat experienceanda knowledgeof precedenthavebeensuperseded.SomeFE
2-20
be
expertsmay guilty of excessive but
enthusiasm, the advocatesof centrifugemodellingor
field observationcanbe equallybiased.
Somespecificobjectionsraisedby Kerr andTamaro(1990)whichwarrantcloserexamination

are that the FEM:
a) generatestoo muchdatato be usefulasa designtool
b) may seriouslyunderpredictwall bendingmomentsandprop loads
c) requiresinput for
parameters which theremaybe no widely test
accepted methods(or
budgetandtime to run suchtests,if they are available)
With referenceto point (a), it cannotbe disputedthat the FEM generatesmore datathan a
Winkler program- the latter typicallyproducingcolumnsof resultsfor deflection,moment,
and sheardown the wall. But FE resultsarenearlyalwaysdisplayedgraphically,so the user
neednot feet overwhelmed.Point (b) is simplynot true in the writers experience- in fact, the
reverseis often the case(i.e. FE overpredicting).This is why sucha high densityof propping
is often indicated,but a largeproportion canbe omittedonceloadsare measured.Point (c)
dependson the constitutivemodelbeing used- mostof thosein regularuserequirenothing
out of the ordinary. If the projectwarrantsgreateraccuracy,thenmoremoneywin be
availablefor advancedtests.
Kerr and Tamaro also go on to say:
"The soils engineeris generally attracted to this method becausehe may think
he is getting 'precise' estimatesof settlementand other movementsbehind the
wall, but the structural engineer is not happy because the results often run
to
contrary past experience. The contractor is not happy because the analysis
indicates the needto instrument and monitor surrounding structures adjacent
to the site and this makesthe owner uneasyabout the fact that his neighbours
are now being made aware of the potential for damageto their properties from
his construction activity"
(Kerr andTamaro,1990)
Precise estimatesmay have been expected 20 years ago, but more rational and mature thinldng
now prevails. In any event, the program only calculates- it is the engineerwho predicts, so
results can (and should) be queried if they appear wrong.
2-21
The attitude shownin the secondsentenceof the abovequoteis extraordinary.If analysis
showsthat surroundingstructuresneedto be instrumentedandmonitored,would not a
responsiblecontractorwant to know aboutthis soonerthanlater? Most "neighbours"could
not fail to noticea deep taking
excavation placenext door, and somelevel of concernis only
naturalandwould alreadyexist. They canonly feel reassuredby the contractorwho is vigilant
andtaking precautions- ratherthan attemptingto underplay(or hide) potentialrisks.
2.2.5 Alternatives
Alternatives to using finite elementsfor retaining wall analysisare as follows:
Winkler spring-based programs(e.g. WALLAP, FREW)

finite differencecontinuummethods(e.g. FLAC)
empiricalmethodsto estimatewall deflectionandgroundmovement(e.g. Clough and
O'Rourke, 1990;HsiehandOu, 1998) embodiedin software(e.g. ReWaRD)
-
take carefulmeasurements during constructionandbe preparedto alter methods/sequence
if dangerlevelsare approached.
Wmkler springmethodswere describedbriefly in Section2.1.5. In summary,they are well

ableto handlethe wall andcomputeits deflectionandinternalforces,but cannotpredict
adjacentground movements awayfrom the wall. Of on
course, a greenfieldsite awayfrom
little
other structures, be
elsewould required. But giventhat the for
reasons carryingout
someform of analysismaywell be to gain planningpermission,Winkler-springprogramscan
be for
never adequate all situations. They are a useful(thoughlimited) tool.
Other continuumanalyses(FDK BF2vDare an acceptablealternative,asdiscussedin Section

2.1.5. It is the writer's view that FDM programsUkeFLAC havebeensuccessfulpartly
becauseof deficienciesin the userinterfacesto geotechnicalfinite elementprogramsin the late
80sandearly90s(this hasbeenconfirmedby discussionswith engineersin a numberof UK
consultancies).Thusthe argumenthascenteredon usabilityratherthantechnicalmerit.
Commercially,this is entirelyjustifiable,but the writer is concernedthat usersof FLAC will
not havethe sameintuitive feel
engineering for the underlyingtheory andmodellingissues.
BEM programsareunlikelyto makea meaningfulcontributionto the designof retainingwalls
investigators
in the foreseeablefuture;they arenot evenbeingconsideredby university-based
at the presenttime.
2-22
2.3 Analysis of the Literature
Approximately 200 referencesin English-languagejournals and international conferencesdeal

specifically with FE analysisof retaining walls. In order to provide a useful summary of these
references, 12 basic items of information have been abstractedand collated in Table 2.3:
1. usage category 7. constitutivemodel(s)

2. wall and support type 8. method of wall installation
3. wall section 9. output quantities
4. soil type 10. FE code used
5. analysistype 11. contribution to main areasof interest
6. drainage conditions 12. overall relevanceto the designer
Further detailsof the informationrecordedundertheseitemsaregivenin the following

sections,together with comments based on a brief of
analysis the trendsandbiasesin the
reportedcases.
2.3.1 Usage category
Threeprincipalcategoriesof usehavealreadybeenidentified(Section2.2.2),namely:
A) Design/prediction
B) Back-analysis
Q Theoreticalstudy
Further subdivisionshavebeenadoptedasfbHows:
A I checking/calibrating simpler methods g.(e. Winkler spring)
2 making predictions (e.g. for observational method)
3 studying parametric variations (e.g. soil stiffness)
4 investigating alternative designs(e.g. propping sequence)
5 to analysecomplex problem not covered by standardmethods
B 1 full-scale retaining wall

2 centrifuge model wall
3 other model wall
C nI parametric study (e.g. mesh refinement, number of increments)

n2 modelling details (e.g. wall installation, 2D v 3D, element type)
n3 numerical development (e.g. excavation technique, soil model)
n4 comparison with other methods (e.g. numerical, analytical)
n5 new wall type/construction (e.g. berms, relieving platform)
n6 derivation of design charts, rules, correlations, etc.
where "rf' indicates if a comparison has been made with physical data
0= no comparison (purely hypothetical wall)
I= full-scale wall 2= centrifuge model 3= other model
2-23
Finally,for eachprincipalcategory,a suffix is usedto indicatehow muchof the construction
andpost-constructionbehaviouris beingconsidered:
a wall installationonly
b wall installationandexcavation(incl. anypropping/anchoring)
c wall installation,excavationand post-constructionbehaviour
d up to andincludingwall failure/ collapse
The usageis thusfully describedby a compositecodehavingthe format [A-C][0-3][1-6][a-d].

For example,a designstudyin which a parametricvariationhasbeencarriedout during
constructionandinto the long-termwould be denotedMc. A numericalinvestigationof the
influenceof berms,in which the resultswere comparedwith centrifugedata,would be C25b.
It is clearfrom Table2.3 that few (<20%) of the reportedcaseshavebeengenuinepredictions

carriedout at designstage.However,the writer doesnot believethat this low proportion is an
accuratereflectionof the level of useof finite elementsin retainingwall design. Ratherit
reflectsthe difficulty (for confidentialityreasons)in publishingwork on ongoingprojects-
especiallyif problemsoccur andclaimsare in prospect. Whenit might be clearto publish
(possiblyyearslater),the designerswill havemovedon to other projectsanddetailed
recollectionswill be lost. In AppendixA, eight differentcasesarepresentedandonly one of
thesehasbeenreportedin the literature- the writer hasurgedpublication(rarelywith success)
andthis is probablyindicativeof wider experience.The true extentof FE usagein retaining
wall design may neverbe known. Back analysescomprise30% of reportedwork, and
theoreticalstudiesaccountfor just over half (52%).
Another striking featureis that nearly80% of all the analysessummarizedin Table2.3 go only
asfar asthe end of construction,with just 14%continuinginto the long term. The remainder
is madeup almostentirelyof investigationsof failure generallyactiveand/orpassiveearth
-
pressurestudies. Possiblereasonswhy so manyanalysesstop after constructionare:
analystshaveonly hadaccessto programscapableof drainedor undrained.analyses,and

so they havebeenunableto performa continuous(coupled)analysisfrom installationto
final equifibrium
2. the majority of FE usein designis gearedtowardsthe constructionphaseonly, until the
permanentsupportis in place,on the basisthat this is the most critical period
2-24
3. back-analysismustbe basedon field measurements, which mayhaveonly beenobtained
during the constructionphase(possiblybecausethe contractonly paid for instrumentatio
duringthis phase).
In designapplications,well over half (63%) of the casesfocusedon makingpredictionsbased

on bestestimatesof real soil conditionsand expectedconstructionsequences.Parametric
variationsaccountedfor only 5% which is rathersurprisingin view of the tremendous
potentialof the FEM in this area. The largestsinglesubcategoryof theoreticalstudyusewas
numericaldevelopment- mostlycomprisingconstructionmodellingor constitutivemodel
development.
As might be expected,a highproportion of backanalyses(87%) arebasedon full-scalewalls,

with the remainderbeingevenlydividedbetweencentrifugeand Ig models. Only 26% of the
theoreticalstudiesusefield datafor verification,but this accountsfor virtually all of the
verificationattempted,with 63% of the theoreticalstudiesattemptingno comparisonat all.
2.3.2 Wall and support type
TI-dsrefersto the overallwall andsupportsystem,of which sevenbasictypeshavebeen

identified.Abbreviationscomprising3-letters,optionallyfollowedby a further letter or
number,havebeenusedin Table2.3, thus:
BFL backfilled ESP[x] embeddedsingly-propped

EBC embeddedbackfilled cofferdam ESA[x] embeddedsingly-anchored
EUC embeddedunpropped cantilever ENT[n] embeddedmulti-propped
EMA[n] embeddedmulti-anchored
[x] indicates the prop/anchor location t= top level, f= formation level

[n] is the number of props/anchors
Other possible suffixes are b= temporary berm, and r= relieving platform.
Embeddedmultiply-proppedwalls accountfor over a third (34%) of reportedcasesandare

the most prevalenttype, with the most commonnumberof propsbeingthree. Embedded
singly-proppedwalls camenext (17%) with top-proppingoutnumberingformationlevel
propping3: 1. Unproppedandmultiply-anchoredeachcompriseabout 10%of cases.Nearly
2-25
40% of anchoredwalls havejust one anchor,after which the most commonnumberof anchors
is four (29%). Backfilledwalls,whiIst strictly outsidethe remit of this thesis,havebeen
includedin Table2.3 and accountfor about 13% of cases.However,mostof theserelateto
earthpressurestudies- often in conjunctionwith laboratorymodeltests,or simplyverifying
the ability of the FEM to reproducethe predictionsof classicalearthpressuretheories.
2.3.3 Wall section
This refersto the structuralsectionof the wall, determinedby its methodof construction/
installation. Nine different sectionshavebeenidentified,abbreviatedasfollows:
MCG massconcrete gravity PDP plane diaphragm panel

TLC T or L-shaped cantilever TDP T-section diaphragm panel
CBP contiguous bored pile SBP sheet-pile
SBP secantbored pile SPL soldier pile + lagging
CDP contiguous driven pile
111
Concerningwall section/ constructionmethod,planediaphragmwalls arethe most common

(38%), followed by sheetpile walls (25%). All other typesare evenlyrepresentedexceptT-
sectionwalls which are only 1%. Rigid walls outnumber flexiblewalls 2: 1; possiblybecause
to
rigid walls are easier model,not requiringspecial bcarnor interface More
elements. likely it
is becauserigid walls are morelikely to be usedfor excavationsupportin urbansites,andthis
is accountingfor an increasingproportionof embeddedretainingwall applications.
2.3.4 Soil type
Wherethe mainsoil type (or mixtureof soil types)hasbeenidentifiedin the reference,this is

simplyindicatedin Table2.3 asclay (soft fill
or stiff), silt, sand,gravel, - or any combination.
Analyses of walls in clay are more prevalent, accounting for nearly 60% of all cases- stiff clay
being 30%, soft clay and unspecified both about 15%. Sand and gravel account for nearly
30%. There is no reason why any one soil type should be encounteredmore frequently in the
field; the sites concerned are located in different parts of Europe, North America and Asia.
Laboratory (1g) models are more likely to use sand (easier to work with); models using clay
normally require a centrifuge to reduce tune scales. In terms of analysis,there is an argument
that clays can be better describedby simple elastic models - especially overconsolidated clay in
2-26
undrainedconditions- andgreatersuccessmaybe achievedthanwhentrying to modelsands.
Also manyof the recentadvancedconstitutivemodelshavebeendevelopedfor clays.
2.3.5 Analysis type
This refersto the numberof dimensionsmodelled,andcanbe eithertwo-dimensional(usually

denoted2D and3D respectively.Any axisymmetric
planestrain),or three-dimensional,
analyseswill be denoted2D, asthey are only pseudo-three-dirnensional.
In somecases,both
planestrainandplanestresshavebeenused(uncoupled)to analysea problemfrom different
perspectives,andthis is denoted2D+.
Not surprisingly, the overwhelming majority (nearly 90%) of analysesin Table 2.3 are 21),
with only 8% 3D (and virtually all of these as theoretical studies). In many cases,2D is a very
reasonableapproximation - for example, long lengths of retained excavation in a depressed
elevation road corridor. For rectangular basementexcavations,2D is less satisfactory, but at
the time of writing is the only feasible design approach. It is unlikely that 3D analysiswill be
common in design applications for some time; it has only recently become establishedin
university-based studies.
2.3.6 Drainage conditions
The drainage conditions imposed or allowed during the analysiscan be either:
UD undrained(total or effectivestress)- no volumechange

DR drained(effectivestress)- no porewater pressurechange
U+D both typesof drainagepresent(e.g. in stratifiedsoil)
U/D indicatesseparatedrainedandundrainedanalysescarriedout
CP coupledconsolidation- loadinganddrainageratescomparable
Layeredsoil profilesmayrequirea combinationof the abovein the sameanalysis.Further

detailsof coupledconsolidationaregivenin Chapter3- sufficeto saythat it is a way of
permittingdrainageduringthe applicationof loads,but requiresspecificationof time steps,
coefficientsof permeability,anddrainageboundaryconditions.
About 40% of reported analysesare wholly undrained, and 30% drained nearly 20% are fully
-
coupled, evenly spread between the three main categories of usage. The progression UD ->
2-27
DR -+ CP correspondsto an increasein the necessaryquality (andnumber)of parameters,so
it is not surprisingthat most analysesare undrainedandfew are coupled.
2.3.7 Constitutive model
The elastic component of stress-strainresponseis conveniently characterizedby homogeneity,

isotropy and linearity, leading to the abbreviations:
HIN ýhomogeneous/nonhomogeneous IEI/A isotropic/ anisotropic

L/N linear/ non-linear elastic
For example IULE = homogeneousisotropic linear elastic, NANE = nonhomogeneous
anisotropic non-finear elastic, etc. If perfectly plastic behaviour is also modelled, a two-letter
suffix is used to indicate the yield criterion:
DP I Drucker-Prager I VM von Mises

TR Tresca, MC Mohr-Coulomb
If the modelhasa namein commonusage(e.g. hyperbolic,Cam-clay),this is shown.
Purelyelasticanalysesaccountfor aboutonethird of all casesreportedwith nearlytwo-thirds

of theseusingthe hyperbolicmodelor its variants. Nearly 40% of analyseshaveusedan
elastic-perfectlyplasticmodel,with Mohr-Coulombbeingthe mostpopularyield criterionby
far. The remaining30% are specificallynamed(elastoplastic)modelssuchasCam-clay,MIT-
E3 andbrick. The most commonelasto-plasticmodelis nonhomogeneous isotropiclinear
elasticwith Mohr-Coulombyielding(NILE/MC), which combinesrelativesimplicitywith as
high a degreeof sophisticationasmaybe needed(or warranted)in manycases.
2.3.8 Wall installation method
The wall installationprocesscanbe modelledin oneof four differentways:
WIP wished-in-place(installationeffectsignored)
ELS elementswapping(soil removed/ concreteplaced- simultaneously)
FIM full installationmethod(bentonitesupport,tremiedconcrete,etc.)
FCM full constructionmethod(for cofferdams,backfilledwalls, etc.)
2-28
Around 70% of the analysescoveredby Table2.3 commencewith the wall wished-in-place
(MP). If the effectsof includingfull installationmethod(FUVI)areasimportantasmanynow
claim,it maybe concludedthat the majority of reportedretainingwaUanalysesarein error.
However,the FIM is very difficult to justify in routinepractice,and someform of in-situ stress
reductionmaybe morepracticable.
2.3.9 Output quantities
The output from a finite elementanalysisof a retainingwall canbe groupedinto quantities

relatingto the wall, its supportsystem,the surroundingground,andother miscellany.The
quantitiessummarized in Table2.3, togetherwith symbolsadoptedby the writer, are Usted
below.
wall surroundingground
5 wall displacement V ex vation heave
M wall bendingmoment S surfacesettlement/heave
Q wall shear force H surfacehorizontalmovement
A wall axial force C strainlevels
D displacementvectors
I
supportsystem CY earthpressures
F prop load K earthpressurecoefficient
B prop bendingmoment P stresspath
T anchorload. U pore water pressures
I principal directions
I
miscellaneous Y zonesof plasticyield
* nodalforces f factor of safety
* other L stress(or mobilization)level
G groundwaterlevels
The six most commonoutput quantitiesarewall displacement 8 (81%), horizontalor vertical

S (34%), wall bendingmomentM (30%),
soil stressa (45%), surfacesettlement/heave
displacement vectorsD (25%), andprop load F (17%). It is very unusualfor wall deflections
not to be shown(onecasebeingfor rigid wallsundergoingprescribedtranslation/rotation).
Horizontalor vertical stressdistributionsare often requiredin earthpressurestudiesbut not so
often in embeddedretainingwall design(exceptwhencomparingwith classicalearthpressure
2-29
envelopes,and/orattemptingto cross-checkwall bendingmomentsor wall shearforces).
Crestsettlement/heave
is clearlyvital in urbanareas,andhasbecomesomethingof an "acid
test" for retainingwall analyses- especiallythe constitutivemodelcomponent.Wall bending
momentis the most commonlyreportedinternalstructuralforce quantity(for calculating
reinforcingsteelquantities),thoughwall deflectionandground surfacesettlementmayactually
governdesign.
2.3.10 Finite element code
The rangeof FE codesusedfor the retainingwall analysesdescribedin the literatureis wide

andvaried. Somecodessuchas CRISP,PLAXIS, SAFE,andABAQUS are commercially
availableandmuchmore likely to be usedby the industry-based
analyst. Otherssuchas
ICFEP,NEST andEXC" are usedalmostexclusivelyby the university-based group that
developedit. In manycases(mostlyuniversities),the codehasno particularname. Wherethe
codeis identifiedin a givenreference,this is fistedin Table2.3.
CRISPemergesasthe most widely usedcommercialgeotechnicalfinite elementpackagein

publishedretainingwall analysis.This does not necessarilymean that CRISP is usedmore in
practicethan any other program - it could simply be an indicationthat its usersare more likely
to publishtheir endeavours.It is part of the "opeW'philosophywhich the developersof
CRISPhaveencouraged(describedfiulher in Chapter3). ICFEP comesout almostequalin
termsof overallnumbers,but is not commerciallyavailable.As such,other engineersare
unableto verify analysesindependentlyusingthe samecode. Investigatorsusingcodewritten
elsewhereare lesslikely to hesitateto publishresultswhich are at variancewith reality - than
arethe authorsof that code. The latter havea vestedinterestin portrayingtheir programas
accurateandcapable- which is possiblybehindthe commentscited earlierfrom Van Weele
(1989)andHo and Smith(1991).
2.3.11 Contribution to main areas of interest
In Section1.4,five mainareaswereidentifiedwherethe writer considersguidanceto be most

neededand/ormostproblemshavebeenencounteredin finite elementanalysisof retaining
walls. In Table2.3, letters"a!' to "6" areusedasappropriateto indicatein which area(s)a
significantcontributionhasbeenmade:
2-30
a geometricmodellinganddiscretization
b constitutivemodellingandparameterselection
C modellingof constructionandlong-termeffects
d computationaldifficulties
e obtainingrequireddesignoutput
Constitutivemodel(b) andconstructionmodelling(c) receivemost attention,mentionedin

50% of the references.This is followedby geometricrepresentation(a) at 31% andobtaining
designoutput (e) at 21%. Numericalissues(d) were only mentionedin 2% of cases.The
andreflectsthe point madein Section2.1 - that
prominenceof (b) and(c) is understandable
over the pastdecade,investigatorshavebeenrefiningeitherthe constitutivemodelor the
methodof constructionsimulation. The low showingof (d) is that numerical
not an assurance
problemsare not occurring-just that they are not being noticed(and the writer would
contendthat do
mostengineers not know what to look for). Obtainingdesign output (e) is
only highlightedfor if it
a reference makesa usefulcontribution, ratherthan merelyincludes
FE output. This might includehow to computesecondaryquantities(e.g. wall shearforce), or
a novelway to displayresultsthoughtparticularlyhelpfulin a designcontext.
2.3.12 Overall relevance/quality
This final category is the only subjective item in the table. The writer has attempted to place
himself in the position of an industry-based retaining wall designer attempting FE analysis,and
to judge what would be useful to a novice. A number in the range 1-5 has been assignedto
indicate the quality of the overall contribution of a given referenceto finite elementsin
retaining wall design (I =lowest, 5=highest). Used in conjunction with item 11, it would be
possible (for example) to identify which referencesmight be particularly useful to consult
when deciding on an appropriate way to model wall installation, or what might constitute a
"good" finite elementmesh.
Some43% of the publishedcasesare at grade4 or above- 70% at grade3 or better. Some

10%arereckonedto be of little (if any)relevanceto retainingwall designers- generally
becausethey dealwith numericalstudiesof earthpressure,verified againstlaboratoryscale
models,ratherthananythingcloselyresemblingan embeddedretainingwall. Justunder 10%
of the referencesareratedat grade5, andin the writer's opiniontheseshouldbe obligatory
readingfor anyoneembarkingon finite elementanalysisof retainingwalls for the first time.
2-31
2.4 Summary
This chapterhas:
tracedthe historicaldevelopmentof FE appliedto earth-retainingstructuresover the

past35 year,identifyingthe major areasof progressandthe key contributors
2) defined both the state of the art and the state of practice; describing the various
categories of FE usage in retaining wall analysis,as well as the motivations for (and
objectionsto) suchusage
3) provideda succinctguideto the importantliteraturein the field of finite elementanalysis

of retainingwalls; abstractingkey itemsof informationfrom eachreference,and
indicatingits usefulnessandrelevanceto the industry-based designer/analyst
With referenceto (1), it hasbeenshownthat mostinput hascomefrom university-based

investigatorsin Europe,N Americaand SE Asia. However,therehasbeensignificant
involvementof practisingengineersfrom the privateandpublic sectorsandthis is important
for the credibilityandacceptanceof FE asa legitimateretainingwall designaid. Field
observationsandphysicalmodellinghaveplayedan importantpart in the developmentof
providingvital calibrationandverificationof numericalmodels.
understanding,
Regarding (2), there is clearly a significant gap between state of the art and the state of
practice. Whilst this is not unusual, the writer believesthat the situation is exacerbatedby the
paucity of modelling guidance available for the industry-baseddesigner/analyst. There are also
conflicting signals from some authorities as to the merits of FE compared with simpler/cheaper
methods of analysis. Added to this, many civil engineering consultants lack a coherent policy
for developing and retaining in-house expertise in geotechnical finite elements. The growth in
FE analysisof retaining walls (predicted, for example, in CHUA 104) has not reaNy
materialized. FEA is attempting to dispel the "sub culture" image attributed to it by Peck, but
progress is slow. Finite elementswere invented in the sameyear (1960) that the annual
Rankine Lecture serieswas established- yet it has been over 40 years until a Rankine Lecture
was wholly devoted to FE analysisin geotechnical engineering (Potts, 2003).
2-32
Objective(3) hasbeenmet by compilinga table summarizingsuchthingsastype of wall and
supportsystem,groundconditions,andwhat modellingassumptions were made. Some40%
of publicationson finite elementanalysisof retainingwalls arerated (by the writer) asbeing
very good or better(ý:grade4), from the viewpoint of usefiilness/relevance
to the novice
analyst. About 30% weregraded2 or less,suggestingthey areof little use(e.g. insufficient
detailof the modelling).Becauseof the complexityof FEA, guidanceon which publications
shouldbe consultedby beginnersis potentiallyinvaluable.The "filtering" of publications
involvedis inevitablysubjective,and so shouldbe performedby a panelratherthan an
individual(asdonehere). SAGE begana similarwork in 1996with the releaseof the CRISP
"PublicationsDirectory", but this was simplya collectionof the originalpapersandstopped
far short of providingany sort of commentaryor critique.
Havingconsideredthe wider contextof finite elementsin retainingwall design,the next

chapterwill focusspecificallyon the CRISPFE package.From the literaturereview, CRISP
wascited morethan anyother commercialcode,accountingfor around30% of retainingwall
analyseswhere the codeusedwas reported. The featuresof CRISPwill be examined in detail
with a specificbiastowardsthe modellingissues for
andrequirements embeddedretaining
walls.
2-33
Summary legend for Table 2.3 (taken from Sections 2.3.1 2.3.12)
-
Usagecategory describedby a compositecodehavingthe format [A-C] [0-3][1-6][a-d]
A I checking/calibratingsimplermethods(e.g. Winkler spring)

2 makingpredictions(e.g. for observationalmethod)
(design/ 3 studyingparametricvariations(e.g. soil stiffness)
predict') 4 investigatingalternativedesigns(e.g. proppingsequence)
5 to analysecomplexproblemnot coveredby standardmethods
B I fidl-scaleretainingwall
(back- 2 centrifugemodelwall
analysis) 3 othermodelwall
C nIparametricstudy (e.g. meshrefinement,numberof increments)

n2modellingdetails(e.g. wall installation,2D v 3D, elementtype)
(T'heor- n3numericaldevelopment(e.g. excavationtechnique,soil model)
etical n4comparisonwith othermethods(e.g. numerical,analytical)
study) n5newwall type/construction(e.g. berms,relievingplatform)
n6derivationof designcharts,rules,correlations,etc.
f
where"n" indicatesif a comparisonhasbeenmadewith physicaldata
0= no comparison(purelyhypotheticalwall)
I= full-scalewall 2= centrifugemodel 3= othermodel
a wall installation only

b wall installation and excavation (incl. any propping/anchoring)
c wall installation, excavation and post-construction behaviour
d- up to and including wall failure / collapse
Wall and suPport type
BFL backfilled I ESP[x]

embedded singly-propped
EBC embedded backfilledcofferdam ESA[x] embedded singlyýanchored
EUC embedded unproppedcantilever EMP[n] embedded multi-propped
EMA[n] embedded multi-anchored
[x] indicatesthe prop/anchorlocation t= top level, f formationlevel
[n] is the numberof props/anchors
Otherpossiblesuffixesare b= temporary berm, and r relievingplatform.
Wall section
MCG massconcretegravity PDP planediaphragmpanel
TLC T or L-shapedcantilever TDP T-sectiondiaphragmpanel
CBP contiguousboredpile SHP sheet-pile
SBP secantboredpile SPL soldierpile + lagging
CDP contiguousdrivenpile I
Soil type clay (soft or stifl), silt, sand, gravel, fill, etc - or any combination
Analysis type 2D (plane strain/axisymmetry), 2D+ (plane strain/stress),
2-34
Drainage conditions
UD undrained(total or effectivestress)- no volumechange
DR drained(effectivestress)- no porewaterpressurechange
U+D both types of drainage present (e.g. in stratified soil)
U/D indicates separatedrained and undrained analysescarried out
CP coupledconsolidation- loadinganddrainageratescomparable
Constitutive model
ý RN I I/A
I homogeneous/nonhomogeneous I isotropic/ anisotropic
elastic
hehaviour L/N linear/ non-linear E elastic
11
D Drucker-Prager vM I von Mises
plasticity
modelsonly TR Tresca MC Mohr-Coulomb
Wall installation method

WIP wished-in-phice(installationeffectsignored)
ELS elementswapping(soil removed/ concreteplaced- simultaneously)
FIM full installationmethod(bentonitesupport,tremiedconcrete,etc.)
FCM full constructionmethod(for cofferdams,backfilledwalls, etc.)
Output quantities
wall surroundi ground

8 wall displacement V excavationheave
M wall bendingmoment S surfacesettlement/heave
Q wall shearforce H surfacehorizontalmovement
A wall axial force E strainlevels
I D displacementvectors
support tem a earthpressures
F prop load K earthpressurecoefficient
B prop bendingmoment P stresspath
T anchorload u porewaterpressures
I principaldirections
miscellaneous Y zonesof plasticyield
* nodalforces f factor of safety
* other L stress(or mobilization)level
I ff
G groundwaterlevels
Finite elementcode I nameof code

computer used(if specified)
Contribution to main areasof interest
a geometricmodelling / discretization I d computationaldifficulties

b constitutivemodelling/ parameterselection e obtainingrequireddesignoutput
c modellingof construction/ long-termeffects
Overall relevance/quality I (very low) 5 (very ýgýh)
2-35
u
C4 en fn N -4 NN cn It -4 -14 en en en en -t C*4 C,4 "te N qt qt en -t
0 c) 00
0 fi (L) Iti t) 0 fi 0
m JD JD JD 0 CU CU JD ce CU CU 0 iD 00Q0M CU ce t) f-) ce
Gn u2
Zw >ý Z: F-4
1-4 K4 rw c> tD tD
ý4- 2 KL4 m >ý > >==
E- ID t n> Ln Ln Ln
tn u2 :i=>b. -4 Gn >.
Co ID ID Co to in r14 0
C, tD ec to w to to Co to to 60 CC to to to to to
.E
1ým11
t114 r-i CN r4 c4 r14 (N c4 c4 C,4 e4 ri m r4 r4 ei ri

4
4,9.119.4
(02) 101.
CN02, p2ý
.
Ici 0 iz
10 00 ýo ID
74
ýmm-8 -0 0 =f2 a0mmm wl
9A. 10 r4
m C> -4 ," -t -4
<<
00
ON
Oo 4el
"z , co
00
g
1
-0 ý? I --
o,
ýý I 9 -
.
2-36
00 JD 0000Q iz
0 vi (1) 0 cu 000u iz C.
) 0 iz cii 9m0000 Co 0
ci
cn
0. CY ng >
0, 92 <
cn
Im
4. -
L u2 93
Co
>m Co >, cn tio U2 >, b tD
9== > ý, ' >
44 >A ID b to bw tD coo Co b Ln to Gn r14
ro
im > ID Z Pý > ID ID W D to
tc cn t0 to (,0 to to to fe to to to ec
JI
III
1ý11 ý
cd
1999921b8b.
ý,
9
I 48.
iý
10
M.
+
,a W.. 04 eý Qn n n m a no 0 m oQ eq n .C-4 a C-4
o Rm
'a
91
C4 en cl cl cl R cl C4 C-4 04 el noel C-4 cl 1mC-4
ed
ý: : -- 4 N.N.10,1 Itj 10
u
k. 4 t
g ,
a
4. k
)
L
7
t P34
4t t
U 01u
ri
1-4 04 $14 Aq 114 1
ýý g ý M-ýpt.
8
iz 0 iz
0Q iz
10 e4
e, - -4
V)
ý
fn
VI ý
< IM
00
cl
C,
m "0 Icý
ý 10
m21A9ý9-18
e4 ,m u2 .1
00
1-.ZMM '
*U.3 j ý
lu AU L)
Je 2-37
Itt -It ý4- -It e4 ýN It -4 t en en eq N
IU
.0u00 .000 .00 0u
cis 0 C'* co .00 .0 .0V0u
cl Mu
10
0
.0 cd cl 'm 4)
Ici
CD
ci
Ol CIO
ý5 ýL, 4-4 am E-4
bb0 ý4 k- b ýý =1 to
t) 0 eo ID 1,4
2, >b 00
4ý Efl En
eAD DO ID U 1,0 M co CIO too co DO vo eo En 0 U) 0 60 b
kkk:
ýz
1-1
"o 999999 eq
lllzl
Cý
CIA C14
Im00mmmm
C14 (14 C14 C*4 C4 C14 (14 C14 C4
0',
U Gn U) U 04 C04 W En .4
. 1.
ýým
.0
4 iz
JD 0
iz
iz
mm
Wý
iz 0 Ici ID
31 -4
<<um
C7% 00
C% 00
CN
rA
00
47%ON CN
II III ýt =, 1 .9, -A. 41.0. - :111

I©I
II
I1I
2-38
- r4 -tr eq - qt tn It 'd, ý eq V) N q1t Iýt
-14 qt en mr -11 (4 It ý1" qt -4
C) 00 C) ei
.4-
cli 0
tu
ci
M0 ID JD 001 t) U0011 4) Q0 CU 80u fi 0 c)
0
: ý'.
0 00 00 fnl4
"I P. a,
09 -N -N 00,0
.
to oM
to >ý
rn :ý 'o En >4 ý4
:3w P,
:ýM %ý-4 ;j
CA rA 4-, ba
0 Z> pZ
to b ;L4 co P to cc 60 co co to to to
C/I
cc
A
58
M
Fj
b t) 898:,!:
-,1s8,95B8bbb982. b9b
in 0a in m 0a0R0ap
C14 C14 C14 C*4 (4 " C4 en cn ima
m C4 anm 9a -4 cq
444 4
,U uuu I 11 41
0 III
-ý -V -V -ý
.
0 0 2.2". 2
ýmm, Pý ý-.
mR Rý,
ý,.
Cz
a2, a2ý
mý:.
c
iz iz Q iz iz 0 10 10 JD 90 ýQ 43
iz iz 0 im iz 10 ,3aam f4 8 rý
91 0 iz ', D -e-3e -
.12 iz -t ', '
rn ell C> (D c"D CD - CD
to
,0.0 r
00 C7,
a,
OCO,
Vo o
00
a a
., .0,.
1
ý-I
00 I
ei
C4
ad
C
ý2
2-39
CD
0 u t) iz (L) u JD ID
. 93
ce 0 iz 0 0 CU 0 jz U JD C.) 0 Co cu
0
ci
lý
Q22
9m p4 fý gl fý
CA
ýý rle 0.4
0m
4
12.4
44 b.-0 '.=0=
Gn 0.4 rn ýZ, t02
4, >
> tD ; Li >>4.
cn.:
w Ln 4 Gn ID :1 >
n Gn w 94 p4
,
tD > 921> : *o > Ln CZ b ID r4
n ID :'0 :, rn > 0,2 Gn > il, > cn Ln cn
00 ectD >, 4-4 to Co
eo to to to to to to to to to to to to Co to
9M 1;12
Irm
01:
00
! IIIiuii1Mih
SS
.M
Iti 0
me
-54 eulg. -g4eiu -u -9 -ý
ý. ä
ýýww 22ýý
Wný 4,:ä; 95 ýýý ýý
45
rw
uý0
ce mm
iz JD
-4 "0
0
m
iz 000
%ýa ýt iz ý
ji
00
43 ,000, en
Uh (n
0
iz
. -4 CD . -4
9
ce
-0 iz 0
wi
Zuuu pz mm -< ri .1ý u 1ý =m
oc
49
2)
e '-, Z ýz 4, u, rn
ýh
ý
;; ý IL)
'ä
-
(2, -0 oý
tfi
2-40
V) r4 VI
clq «* rn c4 em C) -4 rq ý CN er --4 -4 --1 't le
.Mu 0
0 4) C) 0 Q
.0 iz iz 0 0 ce 0 CU jz o o JD ci iz
0 M CU
pi
cu
Ici
CD
; T4
0,
b Ln P. >
Ell Ell :ý b cc >4 ý4 tD tD
8 ýo ý >- En V4 rn cn
to cc to b0
oo
ID bb to t..4 rn (,o too
too to Do 40 to
ZZ
rA
48
ci
9 Ci Iti
- lý
H
u
W P. 4 04 P4 END
W EN
0M iz 0 JD ce
JD 0 iz ID A
00 iz f4 M -4 f4
ýC.
iz 43
10
m ) f4 (D 't e4 0 (D en --4 -4 1-4
e4 ýNý9 (4
u< C) <<uuu 02 C) pz
< PZ
1 9A
C: -
1
Eggl
ýý ý5
-M 1ý
A
ýzeg-ci, :ZU, oý 1
0%
ýg
Ln .9-- Gn >l-
:S5 E (2,
im
241
ýr C*4 C14 -* en I-t en --t in C,4 W'l W) tn
ýt
.M
JD iz 0 C) (L)
cu JD t) 0 JD
0
U cu 0 tu in cu iz JD ce C.) c) 0 iz
ei
10
el --
cn
bb
b
94 -, --
im j ;ý tD
p4 4 v. E- >>
cz
.., :=
,b>
pý > to 2 u2 tD
CD :i cn tn Ln CA ý4
to to to to to to to tD to Co to to v. tc to tc e to
I flU
UL)
Filfifi
2
1:
en CN
g
114
vtt -v
f4 F.,
El.3 E13
1N0
Ici 000 iz
iz aa Q
0 Irl -4 -4
u IZ
0 C,
0
I
-V
00
I
-
Cý r,- r u := ?,
rA
gj v
e .0§ý-
9 -
ý 0:
00
W
I
t
44
66
8
4
242
iz ce Co iz iz ce JD Q ce Co
0
10
ll- 9.1
91
55 pý 12.im.9512.
cn Ký F,
-
mo
Z
>
rn b vo
ID >
to
ci
f4 > C :iXg. >ö Ln tD :m
.3
_M
0bM> tD to ;4 to to
clo to to ID to 00 &0 to ID ',0 to ra4 to
,0=AA
lý
29
Amee
ä M, eem
t) bb lbbBb9888
a
a la A
*ei nn
C14 C-4 la nn C*4 C*4 C14 C4
P4 114 a4 I: Lq 114
141.101. R
Ion.rä-, OR 193-1.
l>
zý
>
912.
822. 12n Hftýý.
4,
ýý
t5 d
e-,
aB ý-.ýý-N
En U)
w
1 31 10
0Q
00 ID JD
iz
ý0ým
0 iz
JD
10
lý 9 -4 4n CD C> e4 CD
Q0 c4 CD -4 CD -4
41 -4 --1 -4
m<u p2 u=umUu L) U<<U
m
C>
I
00
W00ý
0, ,
I
ý 1
t3z
zz,- ;2
,
me2
r4 ;2 -
1a 1 1 1 i
cr
:§me
Z ge, p9
p2
?
g. p. r.
re-,
re-. 2-43
u
c3
",= C-4 f4 ('n mr -mm e4 m rA Ilt c4 (4 c4 In -4 ým
m C,4 m in mM
(L)
.M
0 JD t) 000
0 cu 0 JD 0 JD 0 ce iz Co 00m
iz 0 cu ID ID JD JD iz
A
8i
-ý -2 -g -g -2 -2 2e p4ý
u8)
Gn tn U2 Gn GO 1:4 .ä.
C
ý4 :tD >b0
(10 w c
> Co =0 to
H
;i (D 0 cn
==n ID ID > t) to >
Ln to =VM >
zi
tZ
Co ID U2 ID to CA w to to to > ýo 2 ýo e ZO g$
I
I
V
15
ItE
b9bb9Bb55 519
C-4 C-4 C-4 cl eq en C14 C14 C14 C14 C14 RRRRRRRRRRR9
CN CN f4 r4 clq C,4 CN m
N
1
>"gg9:2 -5el
73 ci j 19
0
9: p4
p4
44 en 0 pn n2.0- ý'n >2 fn'

0
vi
rub) P.
4
Ln
, tnGäa
M
2w
e« lu gD 00 iz 0 JD 00
iz 00 JD V) m CN
ý
rn
-4 ,
0 0
m
10
em
- C-4 -ý -4 C,4 ýý -4 0 rn
ce
QM
< im im u
.
= m u m< u
00
isj
CD ý
I Idhi
c2% 0,
c- -
e4 cr
_GO2
2-44
lz 0 iz JD 0 4) tu ce iz
0
cn U2
0 v c
Ln < <
Z1 t>
a. tz Z$
W ; Le
=>wZ>, tD =
0
to to to Q tc 90
I !!!!!!
3
C114
4)
ý:
ID 00
&l', -i
r4
ý-C.
%0 4
04
M -4 ffl 18
Z1 IM MMZ
L) uu< EL
00
tS C,
mg
.+
245
CfL4,PTER 3
THE CRISP FINITE ELEMENT PACKAGE APPLIED TO RETAINING
WALL ANALYSIS
3.1 Development of the CRISP Package
The suiteof geotechnicalfinite elementprogramsknown asCRISPwaswritten anddeveloped

by variousworkersin the CambridgeSoil MechanicsGroup over the past25 years. The
originscanbe tracedbackto MZSOL, a pair of finite elementprogramswritten by Zytynski
(1976)who drew heavilyon the conclusionsfrom earlierwork at Cambridgeby Simpson
(1973) andThompson(1976). Zytynskiwas responsiblefor the originalsystemdesign,one
exampleof which wasthe decisionto separatethe geometryprocessingandequationsolution
into distinctprograms divisionremainingto the presentday.
-a
Before becoming known as CRISP' in early 1981, the original Zytynski programs were
renamed CRISTINAin 1976, and CRISTINA-1980 in 1980. A variant of CRISTINA, called

GEA (General Earth Analysis), was available for a time through the now-defunct 1-fighways
Engineering Computer Branch (HECB) of the Department of Transport. This early
association with the DTp was not altogether successful,for two reasons. Firstly, the code was
not entirely validated, and one or two bugs were discovered leading to a partial loss of
confidence in the program. Secondly, the BECB was really unable to support the use of a
sophisticated geotechnical finite element code, which representeda big advance on the slope
stability and retaining wall programs previously offered. A more fiuitful and longer lasting
association with the DTp did develop subsequentlythrough the Transport and Road Research
Laboratory (TRRL), who funded a programme of researchat Cambridge - much of which
provided direct financial support for the development of CRISP.
The CRISPpackagewas first releasedgenerallyin 1982,andwas acquiredmostlyby British

CRISP
universities,alongwith someconsultantsandpublicutility companies. 82 (as it came
to be known) comprisedtwo programs;GEOM which carriedout the geometryprocessing,
andMAIN which formedandsolvedthe finite elementequations.This versionofferedonly
'. The CRISP is for"CRItical StateProgram(s)" in

misleading that it
name actiiallyanacronym - slightly
is not restrictedto criticalstateconstittitivemodels(e.g. Carn-clay),
but reasonablein that CRISPwas
probablythefirst commercially availableprogramimplementing suchModels-
3-1
2D analysisin wdsymmetry or plane strain, using linear or cubic strain triangle elements. A
range of constitutive models was incorporated, ranging from simple isotropic linear elasticity
to the elasto-plasticwork hardening Schofield model (Schofield, 1980). Other principal
featuresincluded automatic element numbering optimization ffront squashing"),
computations in effective stressterms, coupled loading and consolidation (Biot, 1941),

changing geometry capability, and an incremental tangent-stiffness solution scheme. Lastly,
prompted by the demandsof another growing geotechnical research activity at Cambridge,
CRISP 82 permitted the modelling of centrifuge tests by varying the gravity field.
A secondversion,CRISP84, emergedtwo yearslater, providingthe additionalfacility of

linearstrainquadrilateralsandfull 3D analysisusinglinearbrick elements.For no apparent
reason(andwithout explanation)the Schofieldmodelandthe front squashingroutinewere
both withdrawn. This versionof CRISPwent on to enjoywidespreaduse especiallyin
-
universitiesandresearchorganizations.
To coincidewith the full publicationof the sourcecodein text book form (Britto and Gunn,
1987),a shortenedform of the program,CRISP-S,was released.What was significantabout
this versionwas that it was distributedon floppy disk andwas the first seriousattemptto put
CRISPonto an EBMPC-compatibledesktop computer. However,therewere some
confusingaspectsof this release- for example,the decisionto unite GEOM andMAIN in one
program,thus doing awaywith the intermediatebinary"link" file and changingthe input data
structure. Someof thesechangeswere motivatedby a desirenot to publishwork already
coveredby others,suchasthe implementationof elastic-perfectlyplasticmodels,describedby
OwenandHinton (1980).
A third major releaseoccurredin 1990,whena 386-PCbasedversion,CRISP90, was made

available.Major advancesin modellingcapabilitywere in the soil-structureinteractionrealm,
offeredthroughthe inclusionof beam,bar andinterfaceelements(albeitfor usein 2D analyses
only - andrequiringmanualinsertioninto the datafiles). A largenumberof pre- andpost-
processingutility programswere also includedto improvethe userinterface;an hitherto
notoriousdeficiencyin CRISP. The Schofieldsoil modelandthe front squashingroutinewere
both reinstated(thoughthe latter now in a separateprogram,ratherthanbeingpart of
GEOM). CRISP90 grew directly out of work commissionedby the TRRL andinitially
incorporatedin a restrictedreleaseversionof the program. Consequently,CRISP90 was
3-2
jointly ownedby CambridgeUniversityandthe TRRL (or TRL asit is now caUed).NEnor
modificationsto the userinterfaceresultedin new versionsbeingreleasedin eachof the years
1992-1995(CRISP93 etc.), known by the generictitle CRISP9x.
Gunn(from 1977)andBritto (from 1980)havebeenresponsiblefor a largenumberof

revisionsandextensionsto the programs,andaregenerallycreditedwith beingthe authorsof
CRISP. Theversionsof CRISPdescribedaboveare thosewhich havebeen"officially"
releasedandsubsequentlysupportedby Cambridge. However, it is importantto realizethat a
very largenumberof customizedversionsare in existence,in which havebeenimplemented
new constitutivemodels,elementtypes,solutionroutines,etc. The openness
with which
CRISPwasoriginallymadeavailableencourageda greatdealof experimentation and
developmentthat maynot otherwisehavebeenpossible.The Cambridgegroup are to be
commended for this - it is possiblyone of their most significantcontributionsto the
geotechnicalcommunity. Variousresearchgroupshavebeenableto makegenuineprogressin
new areas,without havingto "reinventthe wheer' first.
A new departurecamein 1995,with the total rewriting of the userinterfaceto run under
Microsoft Windows. This was undertakenby SAGEEngineeringLtd, andwasthe first time
that a commercialcompanybecameinvolvedwith the evolutionanddistributionof CRISP.
The productwas launchedin September1995and namedSAGE CRISP,though,strictly
speaking,this only referredto the userinterface;the underlying"engin6"wasunchanged.
However,the changein appearance was so radicalthat the new namewasgenerallyassumed
to applyto the whole product (engineandchassis).The SAGE CRISPinterfacewas a major
improvement,but the productwas far from completeon originalreleaseanda seriesof bug
fixeswere issuedin subsequent
months. Eke
Furthermore, its the
predecessor, new interface
was unableto copewith 3D analysis,thuslimiting its usefulness.More significantly,the
CRISPConsortiumwas establishedin 1996,creatingan industrialandacademicpartnershipto
steer(and seekfundingfor) developmentof the CRISP"engine".
At the time of writing, around400 registeredcopiesof CRISPare in usein commercial,

academicandgovernmentalestablishments.GunnandRahim(2003) estimatethat the total
numberof copiesin circulationis at leastdoublethis figure, consideringthat earlierversions
wereunsecured.An independentUser Group (establishedby the writer in 1984)providesa
forum for exchangingexperiences in usingCRISP,andholdsannualmeetingsin the UK. The
3-3
first Developer'sConferencewas heldin 1996andattracteda largegroup of delegatesfrom
aroundtheworld. By virtue of thesenumbers,CRISPis one of the mostwidely usedand
importantgeotechnicalfinite elementprogramsin the world. The literaturesurveyreportedin
Chapter2 certainlyfound CRISPto be prominentin embeddedretainingwall analysis.Other
Programsin this categorycommerciallyavailablein EuropeincludePLAXIS (Delft/Plaxisbv),
SAFE (OveArup/Oasys),VISAGE (VIPS), 2ý_SOIL(ZaceServices),andDIANA (TNO).
It shouldbe notedthat IFEcodeshavebeenwritten anddevelopedat other UK universities,

suchasICFEP(ImperialCollege). This programis particularlydistinguishedin the retaining
wall areaandits successfulapplicationto manysuchproblemshasbeenwidely reportedby
Potts andhis co-workers. Howeverthe programis not commerciallyavailable,and so its use
in the geotechnicalcommunitycannotbe gauged.In fact, the ownersof ICFEP are opposed
to the programeverbecomingcommerciallyavailable,on the groundsthat FE analysisis far
too risky for the averageengineerto attemptby hirn/herself(Potts,2003).
The retainingwall analysesconductedby the writer anddescribedin this thesiswere

performedboth with CRISP84 runningon PrimeandHewlett Packardmainframecomputers,
and CRISP9x on EBM-compatible PCs. A more detaileddescriptionof CRISPwill now be
given, with particularemphasison thoseaspectswhich arerelevantto the modellingof
embeddedretainingstructures. As the computational"engine"and modellingfeaturesof all
the differentversionsof CRISPare essentiallythe same(until very recently),the following
descriptionis applicableto all versionsexceptwherenoted.
3.2 Structure of the CRISP Package
The CRISP"engine" comprisestwo distinctprograms,GEOM (or GP) andMAIN (or NP).

To ensurea high degreeof portabilitybetweendifferenthardwareplatformsand/orcompilers,
thesewere written initially in standardFORTRAN 66, with someFORTRAN 77 extensions
addedlater on. By 2003,the enginehadbeencompletelyrewritten in Fortran90.
3.2.1 GEOM program
GEOM is a "pre-program"which readsthe geometricdetailsof the finite elementmesh

definedby the userin termsof vertexnodeco-ordinates,element-nodeconnectivity,and
34
elementtypes. After carryingout basiccheckson the permissibilityof the data,GEOM
generatesmid-sideand interior nodes,andcreatesa permanentbinary"link" file of geometric
datato be subsequentlyreadin by MAIN.
Caremustbe takenwith meshdatafiles that havebeengenerated(or altered)manually,as

there areseveralbasicmistakesthat GEOM will not detectat present. Oneencounteredby
the writer is the inability to detect"holee' in the mesh- wherean elementmight be completely
missingbut appearsto be presentwhenthe meshis plotted becauseaUneighbouringelements
havebeendrawr?. Another is if a nodenumberoccurstwice in the list of nodal links defining
an element- effectivelycollapsing(say)a quadrilateralinto eithera triangle or an "L" shape-
which will causenumericalerrors(i.e. a zero Jacobian),but only when an attemptis madeto
carry out an analysiswith this mesh. Sucherrorsarefairly easilytrapped;indeedthe writer
has developedsoftwarecapableof doingthis.
As mentionedabove,someversionsof CRISPincludea so-calledautomatic"front squashing"

routine, andit would be appropriateto includea brief descriptionat this point. All versionsof
CRISP usethe frontal solutiontechnique(Irons, 1970),which is a form of Gaussian
eliminationespeciallydevisedfor solvinglargesystemsof simultaneousequations.The frontal
methodreducesthe memory(core storage)requirementsof a finite elementanalysisby
nunmimngthe numberof elementstiffnessmatriceswhich needto be consideredat anytime
during the assemblyor back-substitutionprocedures.This is achievedby assembling the
elementsin a particularorder, andit is the functionof the front squasherto determinethis
order. In CRISP82, this was doneautomaticallywithin GEONLusersof CRISP84 hadto
use a separateprogramcalledSQUASK-CRISP9x includeda utility programcalledSQ9x
which performsthe sametask. The codefor the currentelementassemblyorder optimization
was written by Sloan(SloanandRandolph,1983).
3.2.2 MAIN program
MAIN calculatesthe displacements of a soil body by

caused the loadsbeingappliedto it,
underspecificboundaryconditions. Secondaryquantities,suchas stresses,strainsandpore
pressuresfollow on from the primarysolution. In a coupledanalysis,excessheadswill be
I Of course,it maybetheintentionto introducea holein a meshif tying, for example,to modela

Uumel- but suchopeningsaremorelikelyto beformedby omittingseveralelements fromthemesh.
3-5
computedalongwith displacements, and drainage will be an boundary
additional condition.
User input definesthe materialproperties,in-situ stresses,boundaryconditions,construction
sequence andappliedload steps,to be usedwith the meshprocessedby GEOM andstoredin
the "finle' file.
3.2.3 Pre- and post-processing
Originally,CRISPcamewith no pre- or post-processingcapabilitieswhatsoever.At

Cambridge,a suiteof programswaswritten for usewith the mainframeimplementationof
CRISP;by 1984this hadbeensuperseded
by the useof FEMGEN andFEMVIEW with
suitableinterfacing. Later still CRISP90 offereda suiteof utility programsaimedat mesh
generation,analysisdatapreparation,andresultsvisualization.With eachnew upgradeof
CRISP90 (or 9x), the rangeandcapabilityof the post-processing in
software particular
improvedsignificantly. The latestoffering(SAGE CRISP)hasfurther enhancedthe data
generationandresultsprocessing,but still lagsbehindthe more powerful (andexpensive)
systemslike MYSTRO, PATRAN, or FEMGEN/FEMVIEW. Theseotherpackagesare
generalpurpose,andtheir wider marketjustifiesheavierinvestmentin development.
Commencing in 1984 at City University, the writer developed a suite of pre- and post-
to
processing programs work alongside the GEOM and MAIN programs. In addition to being
used by workers at City and (subsequently)Surrey University, these programs were
distributed to a number of other universities and industrial organizations. Being written in
standard FORTRAN and making use of GINO graphics libraries, they proved to be reasonably
portable (mostly between mainfi-ameinstallations where GINO libraries are more likely to
exist). For the PC, this suite of programs was converted to make use of the NAG PC
Graphics Library in 1989, but was not widely used as CRISP 90 was releasedthe following
year. The suite of programs (Woods, 1990) provided for simple mesh generation within
rectangular regions, undefortned mesh plotting and labelling, extraction/output of specific
numerical quantities, deformed mesh and displacementvector plotting, contour plotting of
cOmPuted/derivedquantities, and XY graph plotting of computed/derived quantities. A
collection of utilities was specifically created for retaining wall analyses,including bending
moment and shearforce diagrams, heaveand settlement profiles, and distributions of
horizontal stressand pore water pressureon the wall.
3-6
3.3 Features of the CRISP Package
3.3.1 Element types
CRISP offers ID, 2D and 3D elements,as summarizedin Table 3.1 and illustrated in Fig. 3.1.
The ID elementswere not present in earlier versions, they were added between 1986-94 (the
2-noded bar and beam being the last to appear).
All elementsareisoparametric,andarenumericallyintegratedwith full integrationrules(e.g.

3 x3 for the linear strainquadrilateral).Reducedintegrationcanbe implementedif desired,
althoughthis doesnot seemto havebeendoneby many(if any) CRISPusers. Interestingly,
the useof reducedintegrationseemsto be well establishedin other major geotechnicalFE
codes(e.g. ICFEP, SAFE,VISAGE) - with the exceptionof PLAXIS.
For the so-called"consohdation7'

elements,somenodespossessthe additionaldegreeof
freedom(d.o.f ) of excesspore water pressure.The assumedvariationin this is oneorder less
than that assumedfor the displacementfield; for example,in the cubic straintrianglewherethe
displacementvariationis quartic(order 4), excesspore pressurevariationis cubic(order 3).
Elementsof the sameorder but differenttopology canbe usedtogether(e.g. LST andLSQ) as

can elementsof the sameorder but differentconsolidationd.o.f. (e.g. LST andLSTp). This is
of particularrelevanceandusefulnessin the coupledconsolidationanalysisof a soil-structure
interactionproblem(seenext section).
The interfaceelementavailablein CRISPis of Goodmantype (Goodmanet al., 1968),and

permitsslip andseparation.Its principalusesare in providing relative(shear)movementon
soil-structureinterfaces(e.g. the back of a retainingwall), andin modellinga no-tension
surfacewhich will separateif stressesceaseto be compressive(e.g. a rock joint).
It is not essentialthat a meshbe composedofjust one type of element. Trianglesand

quadrilateralscanbe mixedtogether. Elementspossessing consolidationd.o.f canbe used
with thosewhich do not; bar andbeamelementsmaybe combinedfreely with solid plane
elements.The only restrictionis that elementssharinga commonedgemust havethe same
numberof nodes,which effectivelypreventsthe useof cubic strainandliner strainelementsin
the samemesh. Constantstrainbarsneedonly be connectedto other elementsat eachend,
3-7
whichgivesmoreflexibility in usingthemascompression/tension
members,comparedwith
their linearstraincounterparts.
3.3.2 Drainage conditions
CRISPprovidesfor both fully drained,fully undrainedandcoupledloading-consolidation

(Biot, 1941) analysesto be carriedout. Drained loadingpermitsoverallchangesof volumeto
occur without the generationof excessporewater pressures.This would be appropriatefor
analysesinvolvingrelativelyfree-drainingmaterialssuchassandsandgravels. Undrained
loading,on the other hand,givesrise to changesin pore water pressurewhilst maintaining
constantvolumeof the soil mass.Thesetwo casesrepresentextremesthat arerarely (if ever)
achievedin practice,althoughin somecasesthe loadinganddrainagerateswin be suchthat
one or otheridealizationis sufficientlyvalid. In betweenthe thesetwo limits exist a large
numberof caseswhereloadinganddrainageratesare comparableandit maybecome
importantto "couple" soil deformationsandfluid flow (Woods, 1986).
Drainedandundrainedanalysesare simulatedby meansof the effectivestressmethod,which

appearsto havebeendevisedindependentlyby Simpson(1973) andNaylor (1975). The
essentialsof the method(referredto elsewherein this thesis)will now be outlined.
The stiffnessmatrix for a particularelementK! is obtainedfrom the integral JBTDB over

the volumeof the element,whereB dependsonly on elementgeometryandrelatesincrements
of internalstrainAe andnodal displacement
Aa :
BAa
andD relatesincrementsof internaltotal stressAcyandstrainAe :
Acr =D As = DBAa (3.2)
Termsin the B matrix are obtainedby partialdifferentiationof the shapefunctionmatrix N,

which relatesinternaldisplacements
Ad at a givenpoint to the nodaldisplacements
Aa :
Ad =N Aa (3.3)
3-8
The D matrixis expressedin terms of total stresses,andrepresentsthe combinedresponseof
the soilskeletonandpore fluid. If the effectivestressstiffnessof the soil skeletonis known a
matrixD' maybe defined,which is relatedto D throughthe expression:
D D' + mm T Kf (3.4)
wherem= [I I 10 0 Of andKf is the bulk modulusof the "equivalent"pore fluid. Kf is

relatedto thebulk modulusof the pore waterK, by:
Kf [(I+e)/e] K, (3.5)
wheree is the voids ratio.
In the effectivestressmethodD is obtainedfrom Eqn (3.4). A drainedanalysisis achievedby

settingKf = 0, implyingthat the pore water is infinitely compressibleandall appliedstressis
takenby the sodskeleton. An undrainedanalysisis achievedby settingKf >> 0 (typically 100
x K', the effectivebulk modulusof the soil skeleton)implyingthat the water is virtually
incompressible andcancarry appliedtotal stress. The selectionof a valuefor Kf is not quite
as arbitraryasit mayseem-a valuetoo low impliespartial saturation;too high and spatial
oscillationsin pore pressureprofiles havebeenknown to occur.
Oncestrainshavebeencalculated,effectivestressesandpore pressuresareobtainedfrom:
Aa' = A& (3.6)
Au T'Ag (3.7)
= Kf As, = Kf ni ni
This methodcanbe usedfor both elasticandelastoplasticanalyseswithout restriction.
If a fhfite elementprogramdoesnot work in termsof effectivestressesbut ratherin total

stresses,it is still possibleto conductanundrainedanalysisof an elasticproblem. Soil
Parameters must be in terms of total and
stresses, if theseare availablethey canbe input
directlyto the programý.If only the effectivestressstiffhessparametersare known, these
mustbe convertedto equivalenttotal stressparametersvia the relationships:
3 If total inputto based the Kf = 0,

stressparameters
are a program m effectivestress along,
method "ith
thenthe"effectiv6"stresses andthe"porepressures"
dw calculatedarein &ct thetotal stresses, are
3-9
G G' (3.8)
K Ký + Kf (3.9)
G E/2(1+v) (3.10)
KE/ 3(1-2v) (3.11)
v (3K - 2G) / (2G + 6K) (3.12)
whereG, K andE arethe shear,bulk andYoung's moduhrespectively,andv is Poisson's

ratio. Eqns3.10-3.12arevalid for effectiveor total stress.
Sucha total stressanalysiswill not yield the effectivestressesandpore pressuresexplicitly,

but thelattermaybe assumed
equalto the changein meantotal stressAp with reasonable
accuracy(again,in an elasticanalysisonly - onceyieldingoccursAu = Ap - Ap').
In a coupledanalysis(Bioý 1941),the standardequilibriumequationsareaugmentedby the

continuityequationwhich, for the 2D case,canbe written as:
a2u. a (C,, )
I[k,, O-Ouc ky
YW
-ý2 + 5; T
at
(3.13)
A convenientmeansof implementingcoupledtheorywithin finite elementswas proposedby

Booker andSmall(1975) andadoptedin CRISP. Additional degreesof freedom(d.o.f) for
excesspore water pressureare introducedinto the relevantelements,with internalandnodal
excesspore pressureincrements,Au, andAb respectively,relatedthrough:
Au. =N Ab (3.14)
The interpolationmatrix N canbe (andoften is) differentto N; in CIUSPit is one order lower,
andhencefewer nodesneedhaveexcesspore water pressured.o.f than displacementd.o.f.
For example,the 6-nodeLSTp only hasexcesspore pressured.o.f at the threevertexnodes.
The overall stiffhessequationsstill involvethe conventionalstiffnessmatrix K but now involve
extratermswhich "couple" the deformationandseepagequantitiesover a time stepAt.
calculatedto be zero (becauseKf = 0). This may be cmvenicnt if total stressdistributions are requiredfrom
an undrainedanalysis,but the program gives effectivestressesand pore pressuresseparately.
3-10
FollowingBritto andGunn(1987),the governingfinite elementequationsgenerallytakethe
form:
KL Aa Af
(3.15)
C At Ab AS
-(D.
where
Aa incrementalnodal displacements
Ab incrementalnodalexcesspore water pressures
K skeletonstiffhessmatrix
La "flow" matrix
Af normalincrementalloadterms
As loadscorrespondingto prescribedseepageon boundaries,plus additionalterms
(D dependson excesspore water pressuregradients
the first line in Eqn (3.15) representsthe equilibriumrequirements,whereasthe

In essence,
secondrepresentsthe continuitycondition.
Various schemesfor carryingthe analysisforward in time arein useandthesearemostly

basedon a finite differenceapproximation.In the caseof "two point" formulaethe equations
are solvedat an intermediatepoint t+ OAtbetweentimest andt+ At, where0<0<1.
Stabilityof the computationalschemedependson the valueof 0 chosen;G.-z,V2gives
unconditionalstability(Booker and Small, 1975). The authorsof CRISPhaveadopted0
(seeBritto and Gunn 1987for further details,includingthe derivationof Eqn 3.15).
Elementswith excesspore pressured.o.f mustbe deployedin the appropriateregionsof the

mesh,anddrainageboundaryconditionssetwherenecessary.Additionalparameterswhich
the userhasto specifyarethe coefficientsof permeability(fixed) for the materialsaffectedand
the time step(usuallyvariablethroughoutthe analysis)for eachincrement.
3.3.3 Constitutive models
The following constitutivemodelshavebeenavailablein CRISPsinceits initial release:

* linear elastic,cross-anisotropic,
homogenous
3-11
" linearelastic,isotropic,nonhomogeneous, stiffnessvaryingwith depth(Gibson,1974)
" linearelastic-perfectlyplasticwith Von Mises,Tresca,Drucker-Prager,or Mohr-Coulomb
yield criteria(plus stiffnessand cohesionvaryingwith depthin CRISP9x)
" Cam-clay(Roscoeand Schofield,1963)
" modifiedCam-clay(RoscoeandBurland, 1968)
" Schofieldsoil model('82 & '90 versions)(Schofield,1980)
" specialmodelsfor bar, beamandinterfaceelements
Whilst not totally comprehensive,

this rangeof modelsis adequatefor a largenumberof
retainingwall problemsandtheir relativemeritsarediscussedfurther in Section3.6.
CRISPis amenableto the introductionof new constitutivemodels,with clearguidancebeing

providedin the user'sandprogrammer'sdocumentation.This facility hasbeenexploited
mostlyby researchworkers,leadingto a largenumberof customizedversionsof CRISP
existingin variousuniversitygroups. A representative
Estof modelsimplementedby UK
workers is givenin Table3.2 (seealsothe proceedingsof the I st CRISPDeveloper's
Conference,1996). Versions3 andaboveof SAGE CRISPimplementedsomeof these
models,suchas3-SKH (Stallebrass,1990)andthe power law modelof Gunn(1993).
3.3.4 In-situ sttesses
In CRISP,initial stressescaneitherbe setto zero,definedon a singlevertical profile, or

specifiedat everyintegrationpoint in the mesh.
a) zerostresses
This option setsall initial stressesandpore pressuresto zero throughoutthe mesh.
verticalprofile
This option requiresthe specificationof Cartesianeffectivestresses,pore pressure,andan
optionalmeasureof stresshistory - at two or more elevationsin the mesh. The variationwith
depthcanbe as simple(or complex)asdesired;valuesat intermediateelevationsareobtained
by linearinterpolation. No lateralvariationis providedfor, the stressesat any givenelevation
are appliedacrossthe mesh. The valuesrequiredat eachuser-specifiedelevationare:
a', (yy &Z rxy upfc
3-12
The co-ordinatesystemadoptedby CRISPdictatesthat aY is the vertical effectivestress,a'.
is the horizontaleffectivestress,anda'z is the out-of-planeeffectivestress.Hencea',, = cr'z
K. &Y. The pore pressureis givenby u, andp', is the previousmaximummeanconsolidation
pressure- requiredonly for critical statemodels. In manycases,TXY will be setto zero, as
principalplanesoften coincidewith the Cartesiandirections. (NB: it is alsopossibleto define
non-hydrostaticin-situ groundwaterconditionsif theseexist; seeSection3.7.4.)
C) everyintegrationpoint
This option requiresstressspecificationat everyI.P. in the meshwhich is highly impractical,
unlessgeneratedby a previousanalysis(a so-called"sto'p-restarC'
analysis)2r earlierin the
currentanalysis- suchasforming a slopinggroundsurfacewhich existsin the presentday.
If non-zerostressesare specified,CRISPwill carry out an equilibriumcheckat the beginning

of the analysis.At eachnode,the inferredin-situ stressesareinterpolatedfrom the defined
vertical profile, convertedto forcesandthen comparedwith the resultantsof appliedforces:
fv
BTa dV =v NTbdV +sNTt dA (3.16)
whereall symbolshavetheir standardmeaning(see,for example,ZienkiewiczandTaylor,

1989). The LHS term representsthe resultantsof elementinternalstressdistributionsa; the
two RHS terms representthe resultants of body forces b (usually just self weight) and of
tractions t along element boundaries, respectively (see also Section 3.3.6).
Any imbalanceis normalizedby the largestappliednodalload (horizontalor vertical as

appropriate)andnotified to the userasa percentageequilibriumerror. This is a usefulcross-
checkat the in-situ stage,andnon-zerovaluesgenerallyindicatea grosserror of somesort.
3.3.5 Boundary fixities
a) Displacement boundaries
Nodescanbe fixed in the X and/orY (or Z) directionsalongthe appropriatemeshboundaries,

andthis is normallya straightforwardprocess.Whole elementsidesarefixed in CRISP84 and
9x, ratherthan individualnodepointsaspermittedin CRISP82. The real difficulty lies in
3-13
in the first place;ennuingthat theyarea sufficientdistancefrom
positioningthe boundaries
ther4on of interestsoasnot to influenceits bcbaviour.
Displacement-wntmfled in theX
loadingcanbe appliedthroughprescribingmovements
andlorY (or Z) directions,on the appropriateboundaries.
b) Drainageboundaries
In a coupledconsolidation
analysis,sufficientMtn POre,
",ater ffidties
pressure mustbe
in orderto establishfree-drainingandimpermeable
prescribed boundaries.it is theequivalent
of ensuringthat therearesufficientdisplacementfficitiesso asto avoidcreatinga mesh%itha
mechanism. However,this is potentiallya verydifficult areaandhasprobablycausedmore
confusionthananyothersingleitemin the inputdatato CRISP.
The first problem arisesover the term "excesspore pressur6".which hasa very specific
to i.
meaning most geotecWcal engineers; e. that pall of the total pore pressureMich is above
(or below) the long-term steadystate %-aluethe part which is dissipatingwith time during the
-
(or
processof consofidation swelling). As used in CRISP. it is really the total head h
multiplied by the unit weight of water, y. a definition adopted by Britto and Gunn (1987)4
U. -hy (3.17)
m (U/T.+Z)T. - U+zT.
whereu is theporewaterpressure,
andz is theelevation.
The secondsourceof confusionarisesfrom the fact that (up until 1993) the CRISP manual
identifiedtwo different typesof excesspore pressurefixitY. "incremental" and "absolut6". T'he
first of thesewould be usedat the beginningof an analysisto specifywhich boundariesare to
be regardedas"fixed" throughout the The secondof theseis usedto specifyactual
analysis.
knQ-wm valuesof excesspore water pressureat specificnodes- usually(but not alv%mys)
on the
9 Mutas am not helpedby the fact dw theequ3nonquotedby Britto andGunncortim a

47*9mPhicalC,nx %hichrendm it dancnskxuUyincomct the M'sion quOW h= is the =Trctcd We-
-
3-14
meshboundary.A third fixity wasaddedin 1993andallowstheuserto specifydirectlythose
boundaries
%%Echhavezeroporewaterpressure.
The writer cannot seeany merit in continuing with the phrase"excesspore water pressure"to
describesomethingwhich is more aldn to a total head(albeit multiplied by y. ). The confusion
is understandable.If a drainageboundary is createdat the baseof an excavation,for example,
most geotechnicalengineerswould considerthe surfaceto have zero excesspore pressure(as
well as zero pore pressure)- and yet for manyyears CRISP required the specificationof a
negativequantity (-u. ) with fudty code 2.
Sophisticatednumericalmodellingpackages may well require input,

sophisticated but
descriptivetermsneedto be selected%ithcare- especiaDy if theyareRely to clashwith
establishednomenclature(aUough seeGibsonef aL, 1989).
3.3.6 Applied loading
Applied loads may take the form of,

3) forces applieddirectly at nodes,resolvedinto Cartesi3ncOmPOnelt-%
b) tractions along elementedges,resolvedinto shearand normal components,or
0 self weight body forces (i. e. gravity loading), producedby the processesof excavation
or fifl construction.
Throughout the analysis,appliedloads must be'L"equilibrium

with internal stresses,when the
latter havebeenconvertedto equivalentnodal forces(Eqn 3.16). At the in-situ stage-this
helpsidentify any input errors; during the analysisstageit pro%idesa meansof monitoring the
developmentof "out-ofýbalance"forces. which normally wise due to excessiveyielding in
Plasticitymodels. Out-of-balanceforces at any nodeare expressedas a percentageof the
maximumappliedload in the correspondingX andlor Y directions.
BOdYforces b and sLuf= tractions t are con%vnedto equi-mlentnodal forces using standard
relationships:
fv
fb m NT bdV
(3.18)
fs
ft NT
t dA
3-15
Theseproducesomecounter-intuitiveresultsin higherorder elements.For example,a
uniformnormalstressdistributionof unit magnitudeactingalongthe edgeof a linear strain
elementof unit lengthwill produceforces 116,2/3,1/6at vertex,mid-sideandvertex nodes
Whilst
respectively. this the
neednot concern userunder normalcircumstances, knowledge of
suchthingsbecomes important when attemptingto interpretnodalforce output (or calculating
wall forcesfrom externalsoil pressures).
3.3.7 Changing geometry
CRISPprovidesfor geometricalterationsto be carriedout in anypart of the analysis.In this

way, the full (element
constructionof an embankment addition)or the excavationof a cut
slope(elementremoval)canbe modelledwith relativeease.
Whendefiningthe originalfinite elementmesh,all elementsthat will be requiredat any point

in the analysismustbe specified.This so-calledparent meshis processedthroughGEOM and
presentedto MAIN via the link file, described earlier. In the situ stage,elementsnot present
be
must notified to MAIN these could be partsof a future embanlanent, or perhaps structural
-
components to be installed later on in the analysis.This defines the primary mesh i. e., that
which correspondsto the start of the analysis.At any subsequent point stage,the current
the
meshwill comprise primarymeshplus/minusany elementswhich have been added/
removedsincethe start.
Elements can be overlaid on the samephysical spacein the mesh without conflict. A part of a
mesh which will ultimately comprise a segmentof a concrete bored pile may initially be soil,
followed by bentonite fluid, then wet concrete, 7-day strength concrete, and finally full
strength concrete. In all, five different material types will occupy this particular part of the
mesh. Normally only one element/ material would be required in any given location at a time.
However, it is possible for more than one to be present simultaneously, and care must be taken
that this does not occur inadvertently.
When an elementis removedfrom the mesh,its contributionto the global stiffiiessmatrix is

cancelledimmediately,whereasthe forcesdueto unloadingare spreadover the whole
incrementblock in which the removalis specified.This is equivalentto instantaneously
convertingthe elementto a fluid density
of equivalent but andthen
zero strength/stiffness,
3-16
gradually reducing the density to zero (but with cr. # cry,rather than q. = cry).The correct
equation (as implemented in CRISP) for nodal loads due to excavation is:
fv fs
Fexc ývBT adV - NT b dV - NTt dA (3.19)
(which maybe comparedwith Eqn 3.16). Whenan elementis first addedto the mesh,it is
initially unstressed,but its full stiffhessis immediatelyavailableto the globalmatrix, andthe
forcesdueto selfweight loadingare spreadover the incrementblock concerned.
To someextentgeometricalterationcanbe simulatedthroughextemallyapphedboundary
loads. Many texts describethe processinvolved(e.g. Naylor et al., 1984)andit providesa
fall-backpositionwhenusinganFE programwhich doesnot providefor elementremovalor
addition. However,thereis little point in usingthis altemativeapproachin CRISPwhenthe
explicit facility is alreadyprovided.
3.3.8 Solution Scheme
CRISP uses a tangent stiffhess solution scheme,in which the global stiffnessmatrix K is re-
evaluated in each increment. As no iterative corrections are applied, increments must be kept
small for non-finear materials to prevent excessive"drifr' from the true solution. Updating K
requires every element stiffhess matrix to be evaluatedbecause,for non-linear and elasto-
plastic constitutive models, the contents of the D matrix will depend on stressand/or strain
level. It is also possible that a nodal co-ordinate update analysishas been requested (where
large deformations are expected), in which casethe B matrix will needto be re-evaluated.
The processis ratherinefficientin that matricesfor elementswhich haveundergonevery little

stresschangeor deformationwiU haveto be reformed. However,this is no worse thanthe
inefficiencyof computingthe displacementsof nodeswhich areremotefi7omthe regionof
interest;a well-documenteddrawbackof finite elements.
For elastic-perfectlyplasticmodels,CRISPhasadoptedthe approachdescribedby Owenand

Hinton (1980) for dealingwith elementswhich haveyielded. Briefly this is asfollows:
For eachelement,at the endof the currentloadingincrement,checkthe stressstatein

eachintegrationpoint (I.P.) againstthe currentyield surface
3-17
If the stresspoint is within or on the yield surface,no further actionis taken
If the stresspoint lies outsidethe yield surface,the yield functionhasbeenviolated and

the stresspoint mustbe correctedby bringingit back to the yield surface. This is done
alongthe vector normalto the yield surface(in 3D stressspace)passingthroughthe
stresspoint in question.The resultantload which is takenoff that I.P. (plus all the others
in the element)to do this correctionmustobviouslybe redistributed.CRISPcallsthis an
"out-of-balance"or correctingload andit is appliedin the next increment.
(NB: thereare severaldifferenttechniquesfor correctingstressesbackto the yield surface-

for examplePotts andGens,1985- the oneemployedby CRISPis by no meansunique.)
Geometricnon-linearity(e.g. largedisplacements) canbe accommodated by updatingnodal

co-ordinatesandrevisingthe B matrix (relatinginternalstrainsto nodaldisplacements- Eqn
3.1) in eachload increment,althoughthis maynot be rigorousenoughfor someproblemss.
The decisionon how manyincrementsareto be usedhasa direct bearingon the duration(and

hencecost) of the analysis- it is normallykept to the minimumnumberthoughtnecessaryfor
an acceptableanalysis.However,there is little informationavailableon how manyincrements
are enough,or on what the criteria of acceptabilitymight be. In general,moreincrementswill
give a "better" analysis,but it is not easyto quantify. In a commercialsettingwhere speedof
tum-aroundis paramount,numbersof incrementsare likely to be too few, in academia,it is
often possibleto rerunthe analysisseveraltimesin order to evaluatethe influenceof numbers
of increments.
3.3.9 Summary
(3.3.1 - 3.3.8) havesoughtto reviewthe salientfeaturesof the

The precedingsub-sections
CRISPprogram,without specificreferenceto anyparticularapplication. in the following
sections(3.4 - 3.11), a closerlook is takenat the stagesin creatingandrunningan FE model
of an embeddedretainingwall usingCRISP. The order followed in describingthe features
correspondsto that encounteredwhenphysicallyusingthe program.
IA problemsis providedby Green'sstraintensor

morerigorousformulationfor largedisplacement
VienldewiczandTaylor, 1989),whichincludesthehigher-order termsusuallyneglected.How=, limited
evidenceprovidedby Templeton(2002)suggests thatthenodalco-ordinateupdatingin CRISPcanproduce
resultswhichareverysimilarto thosefrom moresophisticaWschemes in otherFE codes.
3-18
3.4 Selection of Finite Element Mesh
The vast majority of finite element analysesof retaining walls are 2D plane strain; very
occasionally they might be 3D, or possibly 2D axisymmetric (for example,in the caseof a
large diameter shaft fining). The main reasonfor this is cost, but closely allied to it is the extra
complexity of data and results visualization for the user. Most FE "engines" are capableof
performing 3D analyses,but not all users are able to understandthem. Of course, it is
important to question whether a 2D model can ever be a satisfactory representation of a truly
3D problem. In the context of retaining walls, 2D plane strain can be justified for long runs of
wall, where end effects have little influence on the majority of the structure. However, a
retained rectangular excavation (deep basement,etc.) may give rather misleading results if
only 2D sections are considered (St John, 1975). Once again, it is another modelling decision
to be made.
Throughoutthe remainderof this chapter,the discussionwill focuson 2D planestrain

retainingwall analyses,exceptwherenotedotherwise. Therearethreebasiccomponents
which will needto be modelled;the soil, the retainingwall, andanysupportsystem.In some
casesthe interfacebetweenthe soil andthe wall n-dghtbe given specialattention,but this is
not thought to be commonat present.
3.4.1 Soil
In terms of accuracyversuscomputationaleffort, it seemsgenerallyacceptedthat linear strain

elementsoffer the bestvalueon a "per node" basis,(Naylor etal., 1984). Retainingwall
analystsfavour the useof eitherlinearstraintrianglesor quadrilaterals(LSTs or LSQs)to
representthe soil, andin the absenceof anypublishedevidencethat onetopology is
computationallysuperiorto the other,the choiceis largelyoneof convenienceor personal
preference(or evenprejudice).
A meshgeneratedpredominantlyfrom rectangleson a regulargrid will generallybe ableto

copewith most retainingwall geometries,Fig. 3.2. Smalladjustmentsmayhaveto be madeto
copewith slopingstratainterfaces(Fig. 3.2b),bermedexcavations(Fig. 3.2c), moment-free
ground anchors(Fig.
connections(Fig. 3.2d),dishedprop slabs(Fig. 3.2e), andprestressed
3.21). For rathermore effort, greaterefficiencyof meshcanbe achievedif closelyspacedgrid
finesare gradedout prior to reachinga far boundary,Fig. 3.3.
3-19
Oneimportantconsequence
of adoptinga rectangulargrid with quadrilateralsis that the
elementintegration(Gauss)pointswill lie on vertical andhorizontallinesthroughthe mesh.
This hasboth good andbad features. On the positiveside,columns(or rows) of Gausspoints
in the soil immediatelyadjacentto a wall (or slab)will providefor a very convenientway of
plotting distributionsof lateral(or uplift) pressureson structuralelements.This would allow a
designerto assessearthpressuredistributionsobtainedfrom simplermethods(e.g. Rankine
theory). On the negativeside,a numberof contouringpackages(e.g. GINOSURF)havegreat
difficulty with datapointswhich havea "tracked" layout in the region of interest;a more
randomdistributionof points suitsthe contouringalgorithmfar better.
Triangularelementscould be generatedwithin a basicrectangulargrid, by subdividingeach

rectangleinto two or four triangles,Fig. 3.4. Thereis little point in doingthis fforn a
computationalview point, but it mayfacilitatesomeof the featuresillustratedin Fig. 3.2a-d.
A more sensiblemeshingwith triangleswould follow a free-formmeshgrading,suchasthat
adoptedby Powrie andLi (1991aand 1991b). The integrationpoints areno longer"tracked",
which improvesthe contourproductionbut makesearthandpore pressureprofilesmore
Micult to constructfrom basicstressoutput.
3.4.2 Retaining wall
Again, the choiceof solid elementsfor modellingthe wall is usuallyone of quadrilaterals

versustriangles. A columnof LSQs is mostcommonlyused;the 3x3 integrationpoints
provide threehorizontalsectionsper elementat which bendingmomentscanbe derivedfrom
the shapeof the transversestressdistributions,Fig. 3.5, via the equation:
+t/2
d2x/dy2 = cry.x. dx ay 14 1 d4
-EI 22
-t/2
I
,-, t2/4
ayi wi ti
1 (5/9)A(3/5) (5/9). 4(3/5) }

;tý t2/4 ay. + cry,.
; -_ t2, A (5/9) q(315) (-ay,, - ayj (3.20)
3-20
It is stiff far ftom proventhat Eqn (3.20) is the bestway of computingbendingmomentsin a
wall, but it is quite widely adopted. To manyengineers,it would be the obviousway.
Occasionallyshearforce distributionsare alsosought,andthesecanbe obtainedfrom:
f +t/2 +1
d3
-El x/dy3 = dx
-ry,,,. -ry. 1dt
2
-t/2
;tý t/2 Z Tyi wi
t/2 { (5/9) Ty, + (8/9) Ty.,b + (5/9) 'ry,
-- t/18(5, rym+8Tyxb+5Tyxc) (3.21)
However, shearforce distributionsobtainedin this mannerareoften quite erratic,if sections

through everytriplet of Gausspointsareused. Distributionsaremuchsmootherif only the
centraltriplet is used. Better sA someform of stresssmoothingshouldbe carriedout first
on the raw Gausspoint values,asdescribedlater in Section8.5.
Other questionsfor quadrilateralelementsusedto modelthe structuralwall relateto maximum

permissibleaspectratio, andwhethertwo or morecolumnsof elementswould be a significant
improvementon the more commonlyusedsinglecolumn.
Triangularelementscould be usedto form the wall, but thereis no obviousadvantageover

quadrilaterals.A numberof studieshavebeenpresentedin the literatureconcerningthe
relativeperformanceof thesetwo differenttypesof elementin bending. Onestudyreported
by Cook et at (1989) hasexaminedthe performancein bendingof somesimplemeshesfor a
cantileverbeam,comprisingeitherquadrilateralsor triangles. The generalconclusionwas that
quadrilateralswould performbetterthan triangleswhenthe formerwererectangular,but the
reversewas true whenthe quadrilateralsbecametrapezoidal. The difficulty in obtaining
bendingmomentsvia someequivalentof Eqn (3.20) makesthe useof trianglesunattractive-
raw Gausspoint datawould haveto be interpolatedto producehorizontalsections.
Beamelements,with bendingdegreesof freedom(i.e. rotation) couldbe usedto modelthe

wall; they would be bestsuitedto very slenderwalls (suchas sheetpiling) whereconventional
2D solid elementswould needvery high aspectratios.
3-21
As notedearlier,the vast majority of FE retainingwall analysisare 2D planestrain,wherethe
wall is effectivelybeingmodelledasa panelof uniform thicknessin the out-of-planedirection.
If the wall is of conventionaldiaphragmtype, this representationis quite accurate,andthe FE
modelthicknesswould be the sameasthat of the true wall. A very commonstructuralform
for a retainingwall is that of a contiguousboredpile wall, Fig. 3.6. If the pilesare of diameter
V anda centre-to-centrespacingof T, thenthe secondmomentof inertiaI per metrerun of
wall is givenby:
4.1
d (3.22)
64 s
The equivalentthicknessof wall It' to usein a planestrainFE analysisis then obtainedfrom:
3 412-1
(3.23)
One drawbackwhich is relevantwhen consideringdrainageconditionsis that, in reality, a

contiguouspile wall is not truly impermeable- water can seepthroughthe gaps(althoughwith
low permeabilitysoiLthis maybe a smalleffect)- whereasa solid equivalentpanelwould not
allow this. If consideredimportant,the wall materialcanbe given a small,finite permeability.
If a secantwall is used,the secondmomentof inertia canbe calculatedwith Eqn (3.22), if the

secondary(female)pilesmakelittle or no contribution. In the SouthwarkStationTicket Hall
(Case4A, AppendixA), the femalepileswereunreinforcedandwere constructed20mmoff-
centrefrom the malespiles;consequentlytheir contributionto the flexuralrigidity of the wall
was ignored. If maleand femalepilesareof a similarstiffness,then is
computation very much
more complicated,though bounded by Eqn (3.22) and e112 (seeFig. 3.6).
T-sectiondiaphragmpanelsrepresentyet anotherway of constructingan embeddedwall, in

situationswherehigh stiffhessis calledfor, Fig. 3.7. For a 'T' section:
3+
Wt3 + Wt (d + t12_yc)2 + bd bd (yc- d12Y
12 12 (3.24)
here y, wt (d+t/2) + bd2/2

bd + wt
3-22
Equivalentwall thicknessis then obtainedfrom Eqn (3.23). The T-sectionis probablythe
most difficult wall to modelrealisticallyin 2D, becauseit deviatesso significantlyfrom a wall
of uniformthickness.In field situations,the steppedprofile (in plan) of the rear faceof the
wall will give rise to activeearthpressuredistributionswhich arenonuniformalongthe run of
the wall, andthesewill not be modelledin a 2D FE idealization. (A further point of relevance
is that carehasto be takenin the interpretationof field measurementsof earthpressureswhich
may be madeon the retainedsideof T-sectionwalls.)
3.4.3 Support system
Supportscomprisepropsactingin axialcompression(possiblywith somebending)andground

6
anchorsactingin axialtension. In practicethe formerare generallyusedeitherto provide
temporary(construction)or permanent(long term) supportto the wall, whereasthe latter
nearly alwaysprovide supportcontinuouslyfrom the momentthey areinstalled. Propsand
anchorsbehavevery differently,andneedto be modelledquite distinctlyin anFE analysis.
3.4.3.1 Props
In general,props arenot prestressed

andwill be installedaspassivestructuralelements,
developingload only asthe wall attemptsto move. Formationlevel propswill nearlyalways
be in the form of a continuousslab,andso the planestrainrepresentationis fairly realistic.
Permanentpropsnot at formationlevel, suchasfloors in a multi-levelbasementor the roof of
a cut-and-covertunnel,will alsotend to be continuousslabs. In contrast,temporaryprops (in
the form of hollow steeltubes,universalbeamsections,or timbers)areusuallyplacedat
discreteintervalsalongthe lengthof the wall. The correct a)" stiffnessmustbe modelled
within CRISPby ensuringthat the FE representation
givesthe equivalentvalueper unit run of
wall.
Take,for example,hollow steeltubesof 600mm.externaldiameter,10nimwall thickness,and

a horizontalspacingshof 3.5m. This givesan axialstiffhessEA/L = 200 x 7c(.
602-.582)/4=
3.707ALGN/rn for eachprop. If in the FE meshsolid elementsof depth0.8m(but of the same
length)wereto be usedfor the prop, the requiredYoung's moduluswould be E* = 3.707 x
0 Soilnailingandotherformsof soil reinforcematarerarelyusedwith structurulwalls(exceptfor

remedialpurposes), in thisthesis.
andarenot considered
3-23
1/3.5 x 1/0.8 = 1.324 GPa. If 2- or 3 bar elementsare used, the true E for steel,
-node
together with the area of steel per metre run (=A/sh), are supplied directly to the program.
The connectionbetweenprop andwall may,in reality, allow a full, partial or zero moment
transferconnection.All of theseconditionscanbe simulatedin CRISP:
a) full momenttransferis affectedby usingsolid elementsfor the wall and slab,with a full
connectionbetweenthe nodesalongthe commonedge,Fig. 3.8(a)- e.g. Southwark
StationTicket Hall (Case4A, AppendixA)
b) partial momenttransferrequiresa numberof elementsto createa "necle'whosewidth

canbe varied,Fig. 3.8(b) - e.g. A406 Walthamstow (Case1, AppendixA)
C) zero momenttransferis modelledby havingthe wall andslabmeetonly at a common

node,ratherthan a commonedge,Fig. 3.8(c-f) - e.g. RegentsParkRoadJunction
(Case6, AppendixA). The point of contactintroducesa stressconcentration,which is
undesirablebut probablynot critical asit will normallyoccurwithin materialswhich
are wholly elastic. The mainquestionfor the analystis whetherthe slabshouldbe
taperedto a point in line with the top edge,bottom edge,or neutralaxisof the slab.
The first two, strictly speaking,introducean eccentricloading(andhencemoment)in
the slab;a problemwhich doesnot arisein the third method.
Prop slabs are normally straight but may occasionally be cranked, Fig. 3.9. - e.g. Regents Park
Road Junction. As the cranking is usually asymmetric, it is not possible to considerjust a half-
section of the excavation. The action of "axial" forces at the ends of a cranked slab will cause
rotation of the slab,which mustbe resistedby shearforcesat the ends.
In addition, prop slabsare normallyhorizontalbut mayoccasionallybe slopingor dished,Fig.

3.9. (also, seeA406 Walthamstow andA331 Alder-shotRoadUnderpass,CasesI&7,
AppendixA). Thereis no particularproblemin modellingthis, andit may still be possibleto
considerjust a half-sectionasthe dishingis often symmetricalaboutthe centreline.Another
featureof prop slabsis that they maybe uniform or of varyingthickness(althoughthe latter is
not very common),but againthis presentsno particularobstaclesto CRISP.
3-24
The choiceof elementin CRISPis betweensolid 2D elementswith the appropriatethickness
(depth)andelasticmodulus,or ID bar elementsin which E andA areinput as"material
Parametere'.For eitherchoice,the remoteendof the prop mustbe fixed to give the required
reactionand(whereapplicable)rotationalrestraint. If a 2-nodebar elementis usedto model a
prop at formationlevel,it will be unconnectedto the soil below it and so will not prevent
heaveoccurringbetweenthe endsof the bar. The 3-nodebar shouldnot be used,unlessit is in
contactwith anotherelementalongits entirelength(e.g. soil surfaceat formationlevel).
Often, the axialload in a prop needsto be estimated,andfor a solid 2D elementthis canbe
obtainedon anyvertical sectionthrougha triplet of Gausspointsfrom:
+W2 f +1
dTj
EA du/dx a,,.dy a. h
-h/2 -1 2
IV2 I a,,, wi
--
h/2 1 (5/9) cr, + (8/9) cr.b + (5/9) cr.

.,,
%t; 18 (5a,,,, +8 cr,,b +5a,, ) (3.25)
(Shearforcesandbendingmomentscanbe obtainedfrom Eqns3.20 and3.21, merelyby

transposingx andy subscriptsandsubstitutingt with h). If a permanentprop slabis inclined
at an angle 0 to the horizontal,Eqn (3.25) does not yield the true axial force, though for
relatively shallowinclination(say,0 :520') the equationis sufficientlyaccurate.
If ID bar elementsare used, CRISP will output directly the load in the prop at any given
increment of the analysis. (More precisely it will output the changein load subsequentto
installation if there is any prestressload, this must be added to give the actual load in the
-
prop. This latter point is more relevant to the modelling of ground anchors).
A perfectlyrigid prop canbe simulatedby fixing the appropriatenodeon the wall against
(usually)horizontalmovement.The fixing of a singlenodewas possibleis CRISP82 and9x,
but not in CRISP84, whereonly a whole elementsidecould be fixed. To overcomethis
restriction,the prop could be modelledby a single(very stiff) triangularelementwhich had
onevertexnodein contactwith the appropriatewall node,andthe oppositeedgefixed against
horizontalmovement(thuspreventingsheartransfer,Fig. 3.10).
3-25
In someinstances,it maybe preferableto applya prop forcewhich caneitherbe constant
(easy)or varying(moredifficult) throughoutthe analysis.This is achievedby applyinga point
force of the desiredmagnitudeat the appropriatewall node;appliedforcesare discussedmore
thoroughlybelow.
3.4.3.2 Anchors
Ground anchorsmaybe eitherprestressedor passive.Prestressed anchors comprise a free

length (which is unbondedwith respectto the surroundingsoil) anda fixed length(which is
fully bonded- normallyby somekind of grouting),Fig. 3.11. Passiveanchorshaveonly an
unbondedlength,connectingthe wall with somerelativelyfixed object, suchasa concrete
block or a pile.
Both typesof anchorrequirea substantialheaddetail,to makesurethat anchorloadsare

transmittedproperlyto the structuralwall. An individualanchoris madeup of several
tendons,eachcomprisingmanystrandsof steelwire. PrestressloadsWill be appliedvia the
anchorheadwhenthe grout aroundthe fixed lengthis sufficientlystiff to restrainthe distalend
of the tendons. For a prestressed
anchor,stresstransfertakesplacealongthe fixed length-
betweenthe steeltendonsandthe grout, andbetweenthe grout andthe soil. This is a very
complexinteraction,andalthoughrigorousfinite elementmodellingis feasibleandcanbe
justified for a singleanchor(e.g. BarkhordarL1998),somesimplificationis necessaryfor a full
earth-retainingsystem,asdiscussedbelow.
Two aspects of geometry are the spacing (vertical and horizontal) and the inclination of the
anchor. Vertical spacing will be tied in with the magnitude of the excavation lifts, and the FE
mesh will normally be detailed to cope with this. Horizontal spacing is determined by such
things as pile spacing (contiguous walls) or panel widths (diaphragm walls). Operational
anchor cross-section areasand prestressloads will have to be factored by the inverse Of Sh,the
horizontal spacing, to give a correct plane strain representation.
Thereare four principalwaysin which the anchorcouldbe modelled:
a) refening to Fig. 3.12(a),a single2-nodedbar elementlinking point A to B, C or a

point midway between B and C. This is perfectly for
adequate the proximal end at A,
but is quite incorrectat the distal end- load transferoccursover some(if not all) of the
3-26
groutedlength,andnot at a singlepoint asthis form of modellingwould imply. The
attractionof this approachis the simplicitywith which it canbe incorporatedinto the
mesh-a regularmeshwill usuallyneedonly minor adjustmentto accommodatethe
distal.end. The concentratedloadtransferwould result in excessiveyieldingof the
soil, leadingto an overpredictionof groundmovement.
b) referringto Fig. 3.12(b),a single2-nodedbar elementlinks pointsA andB, plus a

numberof additional bars modelling the lengthB to C. This providesa muchbetter
representationof load transferat the distalend,with a more gradualtransferof load.
However,it requiresconsiderableadjustmentto a regulargrid mesh,as exemplifiedby
the meshusedby Creed(1979)for the NeasdenLaneUnderpass,Fig. 3.13.
C) an extensionof method(b) would be to.usea single2-nodedbar to link A andB, plus

a numberof solid 2D elementsmodellingthe lengthB to C. This is probablyas
accuratea representationasis possible,but with rathermore adjustmentto a mesh
thanwith method(b).
d) a very crudeapproximationdispenses with anyform of anchorelementaltogether.

Point loadsequivalentto the initial prestressare appliedat the nodepoints
correspondingto the distal,andproximalends. Whilst very simple,this providesno
"tying" of the loadsanddoesnot allow for changesin anchorload (which canbe
for
expected further excavationstagesandlong term it
equilibration)- yet requiresthe
sameamountof meshadjustmentasmethod (a).
3.4.4 Soil-wall interface
Thereare two differentaspectsof interfacebehaviourwhich are of interestto the retaining

wall modeller;slip andseparation.The former mayoccur (in reality) on eithersideof the wall;
on the retainedsidedueto self-weightcompressionin back filled walls, andon the excavated
sidedueto swellingandheavein embeddedwalls. Separationmayoccur with unpropped
cantileverwalls, asthe wall movesawayfrom the soil underinitial (undrained)conditions.
This thesisis not concernedwith backfilledwalls, andthereforethe settlementof fin neednot

be consideredfurther. Swellingandheaveare,however,very relevantin the caseof
embeddedwalls usedto supportexcavations.The soil immediatelyin front of the wall, at
3-27
formationlevel, is likely to heavequite considerably;undrainedreboundin shortterm,
followed by swellingin the longerterm. The level of movementexpectedwill be more than
sufficientto mobilizefull wall adhesionand/orfriction, so relativeslip is almostcertainto
occur in the soil-wall interfaceover the first few metresbelow formationlevel.
In the FE model,this slip is effectivelypreventedif standardcontinuumelementsareused,as

full connectivityof the soil andwall elementswill be maintainedthroughoutthe analysis.
However,this neednot imply that excessiveshearstressesaremobilizedat the interface. It is
quite commonto usesomeform of yield criterion in the soil constitutivemodel(e.g. Mohr-
Coulomb)which would help ensurethat the soil shearstrengthis not exceeded.This maybe
somewhathigherthan the interfacevaluesencounteredin practice,wherec', < c' and5' <
is often thought appropriate.
The most seriousconsequences

of theseexcessiveshearstressesis that the bendingmoments
in the wall will be increased(or reduced)dependingon the positionbeingconsidered.Fig.
3.14 showsan highly idealizedlateralearthpressuredistributionon a cantileverwall, together
with a simpleuniform shearstressdistributionactingupwardson the passivesideand
downwardson the activeside. About the neutralaxisof the wall, the shearstresseson the
active sidewill be actingin the oppositesenseto the lateralpressures,whereason the passive
side they will be in
acting the samesense. 7
As an illustration, 10 kPa of shearstressactingover a heightof 8m in aIm thick wall would

changethe bendingmomentby 10 x8x1x 1/2= 40 kN-m /m run. This might not be
consideredexcessiverelativeto the momentsof severalhundred(or thousand)kN-m
encounteredwith diaphragmwallsin stiff clay. But if the excessshearstrengthon the
interfacewere nearer20-50 kPa, significanterrorsin bendingmomentcould arise.
Next, possibleseparationbetweensoil andwall shouldbe considered.In practicethis could

conceivablyhappenover the upperfew metreson the retainedside,aftertop level proppingis
removed. If interfaceelementsarenot used,andthe soil is fully elastic,the retainedsoil can
I Thepresence for thedifferences

is partly responsible
of theseshearstresses observed
sometimes
wbcncomparingbendingmomentsfiom finiteelemat andWinklerspringcalculations.In Winklermodels
(e.g. FREW,WALLAP) thewall usuallyhaszerothickness, as ID beamelemmtsareemployed.Soevenif
wereallowed,theywouldhaveno leverarmabouttheneutralaxis.
non-zeroshearstresses
3-28
restrictforward movementof the wall. This canbe partly mitigatedby the incorporationof
plasticyield in the soil model,but the soil andwall elementswill remainconnected.
The mainconsequence
of "no-separation7'
is likely to be reducedwall deflections.To estimate
the magnitudeof the effectconsidera concretediaphragmwall of thicknessIm retaininga
heightof 8m. TakeEI = 15 x 106x 14/12 125 x 104kN-mýandassumevalidity of the
standardcantileverdeflectionexpression5 P12/ 3EI. If a 10 kN "tensile" force were to act
at the top of the wall, the normaloutwarddeflectionwould be reducedby 1.4mm. Of course,
the lower part of wall will not be fully fixed, andforcesmuchhigherthan 10 kN might develop
in the FE analysis both of which would tend to reducemovementevenfurther.
-
3.5 Type of Analysis
3.5.1 Uncoupled
If it is only the construction(short term) situationwhich is beinganalysed(e.g. an

investigationof differenttemporarybermgeometriesor proppingsequences), an undrained
analysisin termsof total or effectivestresseswill suffice. Time stepswill not be relevant.
However,the actualperiodduringwhich swellingtakesplaceis uncertainandmaybe quite

short if the soil containssilt or sandlensesandthe groundwaterlevel is high. It maytherefore
be prudentalwaysto carry out a fully-coupledanalysisin termsof effectivestress,with
realisticcoefficientsof permeabilityandtime steps,evenif it is only the constructionstage
which is of interest.
Long term behaviour(alone)couldbe investigatedby carryingout a fully-drainedanalysisin

termsof effectivestress.However,this is likely to permit excessivemovementprior to
installationof the permanentsupportsystem,andso is probablynot satisfactoryfor fine-
grainedsoils.
3.5.2 Coupled
Modelling the short term (construction)andlong term (equflibration)aspectsof a retaining

wall in one continuousanalysisis muchmore elegant,andis madepossiblein CRISPby the
coupledconsolidationfacility (oneof the first commerciallyavaable FE codesto provide
3-29
this). After convertingthe elementtypesin the affectedregions,the only additionalinput data
requiredarethe time stepsfor eachincrement,the permeabilitiesof the relevantmaterials,and
the locationof drainageboundaries.It is thenpossibleto checkthe assumptionof undrained
behaviourduring construction,to tracethe developmentof wall displacements,
bending
moments,prop forces,excavationheave,etc. with time, andto investigatethe effectsof
changingpermeability(essentialin view of the inherentlyhugevariabilityof this parameter).
One further advantageof the fully-coupledformulationis that seepage

forcesare explicitly
taken into account. Theseforcescanconstitutea significantloadingon the soil skeleton
(highly relevantto excavationbasalstability). Furthermore,whenthe full three-dimensional
theory is appliedto consolidationproblems,it is observedthat in certainregionsthe excess
pore pressurerisesbeforeit decays.The phenomenon is known asthe Mandel-Cryereffect
(MandeL 1957;Cryer, 1963)andhasbeenobservedexperimentally.The authorhas
demonstratedthe importanceof this effectby simulatingthe loadingandconsolidationof
triaxial test sampleswith CRISP(Woods, 1986). Radialnon-uniformitiesin water content,
strengthand stifffiesswere predicted,dependingon the drainageconditionsandthe rate of
applicationof total stress. Similar had
non-uniformities been observedexperimentallyby
Atkinson et al. (1985). The implicationsof Mandel-Cryerfor embeddedretainingwall
analysisare not fully understood,but couldbe very significant.
For stratawherepore pressurechangesarenot expectedduring the period of time being

modelledin the analysis(e.g. unconfinedsandsandgravels),it is inefficientto use
consolidationelements.The introductionof excesspore pressured.o.f increasesthe sizeof
the global systemof equations(seeEqn 3.15);unnecessarydeploymentof consolidation
elementswill leadto a gratuitousincreasein storagerequirementsandprocessingtime.
However,theremaybe occasionswheremanipulationof pore pressurein unconfined
sand/gravellayersis required,suchasrising groundwaterlevels.
An examplewhich occursquite often is that of an unconfinedfree-draininglayer over much

lesspermeablematerial,suchasgravelsoverlyingclay (commonin London). In this case,
consolidationelementsmustbe usedfor the clay,but eitherkind could be usedfor the gravel.
Nonconsolidationelements(togetherwith drainedmaterialproperties)would maintaininitial
groundwaterconditionsin the gravelsthroughoutthe analysis(whichmaybe exactlywhat is
required). A drainageboundarywould thenbe locatedat thejunction of the gravelandclay
3-30
just at the top of the clay cannot
with conditionue= 0, indicatingthat the porepressures
changeduring the analysisandwill be maintainedat the hydrostaticheadin the gravel.
Alternatively,consolidationelements(togetherwith high permeability)couldbe usedfor the

gravel,with a drainageboundary(of conditionue= 0) locatedat the initial groundwater level.
This would leavethe gravelfree to experiencepore pressurechanges,but thesewould be
unlikely with normalratesof construction.
The wall itself will often be consideredto be an impermeableboundary,modelledby non-

consolidationelements.However,for contiguousboredpile walls, exposedstripsof soil
betweenpiles at the excavatedfaceprovidea drainagepath to the atmosphere,andit may be
necessaryto permit somedrainagethroughthe wall elements.Furthermore,in the full wall
installationprocessof Gunnet al.(1993),water shouldbe permittedto flow from the wet
concreteinto the surroundingsoil, necessitating the useof consolidationelements.(Bentonite
slurry forms a filter cake,so suchflow is not possibleandnon-consolidationelementswould
be sufficient.) For further discussionon the location andtype of boundaryconditionsneeded
for a coupledanalysis,seeSection3.8.2.
3.5.3 Other approaches
Drained,undrainedandcoupledanalysesprovidefor most conceivableshort- andlong-term

situationsencounteredin retainingwall modelling. One exceptionto this is whenattempting
to follow the recommendations of CIRIA Report 104(PadfieldandMair, 1984)for a mixed
total andeffectivestressanalysis.This is straightforwardwith Winkler springprogramslike
WALLAP (Borin, 1988)andpseudo-continuum analysesEkeFREW (Pappinet al., 1986).
However,in CRISPit would involve a so-called"Stop-Restarf' analysis,in which total stress
undrainedparametersareusedfor the constructionstageand effectivestressdrained
parametersusedfor the long term. This "switclf', which mustbe carriedout in a single
incrementblock (immediatelyfollowing the restart),causesa "shocle'throughoutthe meshas
excesspore pressuresarereleasedsuddenly,andmassiveeffectivestresschangesoccur.
In a sense,it is completelyunnecessary
to do this in an FE codewhich hasfully-coupled
consolidationfacilities. Non-coupledprogramscould modelthe transitionfrom short-to
long-termin a ratherbetterfashion,througha "known-change-of-pore-pressure" analysis
(Naylor et al., 1984). Indeed,in the writer's exTerience,the needfor sucha "switch" hasonly
3-31
arisenwhenattemptingto benchmarkWinkler springanalyses(which havefollowedthe
CIRIA 104recommendations) againsta more rigorouscontinuumanalysis.
3.6 Choice of Constitutive Model
3.6.1 Elastic models
The elasticmodelshavethe mainadvantages of simpficityandrobustness.Their principaluse

is in giving an initial indicationof behaviour,without the userneedingto be concernedwith
plasticyielding,incrementsize,etc. Obviouslyin somecasesit WHInot be possibleto advance
beyondsimpleelasticmodelsbecausethe availabledatamay not warrantorjustify it.
The linearelastic,cross-anisotropic,
homogenousmodelrequires:
Young's modulusin horizontaldirection Eh

Young's modulusin verticaldirection F,
Poisson'sratio in v-h plane VA
Poisson'sratio in h-h plane Vhh
shearmodulusin v-h plane Gvh
The linear elastic,isotropic,nonhomogeneous

modelparametersare:
Young's modulusat referenceelevation Eo

y ordinateof referenceelevation YO
rate of changeof Young's moduluswith depth M
Poisson'sratio V
(theseparametersmaybe enteredin consistenteffectivestressor total stressterms- see

Section3.3.2. for further explanation)
A major problem(not uniqueto thesemodels)is the selectionof appropriatestiffhessvalues.

With only one stiffnesspermittedin anygivendirection,the usermustanticipatethe strain
level fairly accurately.The selectionof Poisson'sratio (andindeedthe valuesof at leasttwo
of the five requiredparameters)for cross-anisotropy
is not a trivial exerciseif attemptingto
work in termsof total stress(seeBishopandIlight, 1977).
3-32
3.6.2 Elastic-perfectly plastic models
The incorporationof somesort of soil (shear)strengthlimitation providesan extralevel of

realismto a retainingwall analysis.The elastic-perfectlyplasticmodelsarethe most
commonlyusedin practicalretainingwall analyses,possiblybecauseshearstrengthparameters
are asreadilyinferredfrom typical siteinvestigationdataas stifffiessparameters.They are
alsoreasonablyrobustmodels,unlikelyto experiencemajor numericalproblems.
The parametersrequiredfor a linearnonhomogeneous

elastic-perfectlyplasticmodelarc:
Young's modulusat referenceelevation E,,

Poisson'sratio V
cohesioninterceptat referenceelevation co
angleof internalfriction
y ordinatefor referenceelevation YO
rate of changeof Young's moduluswith depth ME
rate of changeof cohesionwith depth nic
yield criteriontype
Von Mises,Tresca,(total stress)
Drucker-Prager,Mohr-Coulomb(effectivestress)
(NB: prior to CRISP 88, the linear elastic-perfectly plastic model did not incorporate
nonhomogeneity of stiffness or strength)
The load incrementsin CRISPneedto be reasonablysmall,to avoid problemsassociatedwith

excessiveyieldingin anyone increment.However,automaticload correctionis normally
carriedout, which reducespotentialerrorsconsiderably.
Excessivedilation (or negativepore pressurechange)may occur with Drucker-Pragerand

Mohr-Coulombmodels,becausethe flow rule is associated.This will leadto the soil havinga
stfffer and strongerresponsethancould be realisticallyexpected.More recently,a non-
associatedMohr-Coulombmodelhasbeenimplementedin SAGE CRISPto addressthis
deficiency.In additionto the aboveparameters,
the angleof dilation W' mustbe specified.
3.6.3 Critical state models
The Cam-clayfamily of modelsin CRISPoffer a more sophisticatedrepresentationof soil

behaviour,andhavebeenparticularlysuccessfulwhenappliedto embankment-on-soft-clay
3-33
problems(e.g. PicklesandWoods, 1989). Their usein retainingwall analysisis relatively
rare,andmostreportedcaseshavebeenfor hypotheticalstructures,suchascentrifugemodels
(e.g. Bolton et al., 1989;Powrie andLi, 1991a; RichardsandPowrie, 1994).
The parametersrequiredfor all critical statemodelsare:
gradientof normalcompressionline in v: ln(p') %

gradientof swelling-reloadline in v: ln(p') IC
specificvolumeon normalconsolidationline at p'=l N
specificvolumeon critical stateline at p'=l IF
gradientof critical stateline in q: p' M
shearmodulusor Poisson'sratio G or V
effectivepreconsolidationpressure pC
The Schofieldmodeladditionallyrequires:
gradientof Hvorslevsurfacein q p' H

gradientof tensilecrackline in q p' s
The Cam-claymodelsin CRISPdo not havestresscorrectionsat the endof an increment,due

to the difficultiesin correctingbackto a yield surfacewhich maybe hardeningor softening
(Britto and Gunný1987). As a consequence, the incrementsneedto be kept very smallif
reliableanswersare to be obtained. Someinvestigators
(notablyPotts et al, 1990)claimthat
suchstresscorrectionscanandshouldbe done,andthat it is potentiallymisleadingto rely on a
tangentstiffnessschemewithout any corrections.
Generalizationof the critical statemodelsin 3D stressspaceis achievedby simplerotation

about the isotropic axis. This means that the failure conditionswill be incorrect for several
orientationsin the deviatoric (p) plane. For in
example, strength triaxial extensionwould
the
be overestimatedif M is basedon a ý' measuredin compression.
3.6.4 Structural materials
Constitutivemodelswhich canbe usedwith structural(ID) elementsare,of necessity,

relativelyunsophisticated.SimpleHooke's law is availablefor the bar andbeamelements,
with the additionof elasticmoment-rotationrelationshipsin the latter.
3-34
The constitutiverelationshipsbuilt into the interfaceelementare a little moreinteresting,
basedon thosedevelopedby Goodmanet al. (1968). The normalandshearload-deformation
characteristicsare definedin Fig. 3.15.
The parametersrequiredfor bar andbeamelementsare:
Young's modulus E
Poisson's ratio v
cross-sectional area A
second moment of area (beam only) I
For the interfaceelement:
cohesionintercept c
angleof internalfriction
normal stiffness k,,
shearstiffness k.
residualshearstiffness krres
thickness(height)of slip element t
In general,the useof bar andbeamelementsin retainingwall problemsis straightforward.

One difficulty which hasarisenconcernsthe (erroneous)useof 3-nodebar elementsasprops
spanning between two points. The endnodes will generallybe fixed to structuralcomponents
(wall, prop slab,etc.), but the mid-sidenodeis usuallyleft free andthis lack of restraintcan
causea fatal numericalerror.
Interfaceelements,on the other hand,havebeenfar from trouble-free. Powrie andLi (1990b)

reporteda numberof difficultieswith the slip and separationperformance. When the shear
stressexceedsthe limiting value,slip occurs but the shearstressis not correctedback to the
limiting value. Whena tensilestressis developedbetweensoil and structure,they separate-
but tensilestressis carriedthroughto successive incrementsandis not zeroed. An exhaustive
evaluationof the CRISP slip element,andvariousextensionsto it, hasbeencarriedout by Ng
et al (1997) in the contextof buriedpipes.
3.6.5 All continuum materials
Additional parametersrequiredby soil modelsusedfor 2D and3D elementsare:
3-35
bulk unit weight y
bulk modulusof the pore fluid K,,
Coupledconsolidationanalysisadditionallyrequires:
unit weight of water TW

permeabilityin the horizontaldirection k.
permeabilityin the horizontaldirection ky
Bulk unit weight hasits conventionalmeaning,andK,,,hasbeendiscussedin Section3.3.2.
The coefficients of permeability k,, and ky play a paramount role in governing drainage rates.
In the standard versions of CRISP, k,, and ky are constant throughout the analysis,regardless
of changesin voids ratio. Some researchers(e.g. Almeida, 1984) have modified CRISP to
incorporatea form of voids ratio dependency
for k. and/orky, asthis canbe particularly
importantwhen dealingwith highly compressiblesoils. A cruderway of modifyingk would be
through a "stop-restart"analysis,asoutlinedin Section3.5.
A major difficulty whendealingwith stiff fissuredclaysis estimatinghow k will be affectedby

cracksandfissuresopeningdueto stressrelief, especiallybeneaththe excavatedregion. A
more fundamentalissueconcernsthe selectionof k valuesfor normalintact ground,when k is
known to vary over severalordersof magnitude morethan any other engineeringmaterial
-
parameter.Laboratorymeasurements are often completelymisleading,andfield
determinationscanbe highlylocation-specific.In view of thesedifficulties,it would seem
mostappropriateto selectupperandlower boundsfor k, both to gaugethe sensitivityof the
analysisto k, andto bracketthe likely behaviour.
3.7 Definition of Initial Stresses
The zero in-situ stressoption hasvery little practicalapplicationin retainingwall analysis,asit

canonly be usedfor linearelasticanalysiswherechanginggeometryis simulatedby the
manipulationof boundarystresses.Non-linearor elasto-plasticmodelsrequirethe
specificationof startingstresslevels;excavationor filling by the removal/additionof elements
requiresthoseelementsto possessself weight.
3-36
Thevertical profile specificationoption canaccommodate fairly complexvariationsof K. with
depth. However, becausethe stressdistributionsare in effect beingdefinedon a singlevertical
section,there canbe no horizontalvariationof stressesacrossthe meshusingthis option.
3.7.1 Strata boundaries
At somestrataboundaries(e.g. madegroundoverlyingstiff clay) theremaybe a changeof K.

leadingto a discontinuityin the distributionof horizontaleffectivestress.This is besthandled
by defting two pointswith a smallvertical separation(say0.01mapart),usingthe appropriate
K. valuesat each. CRISPwill not accepttwo or more in-situ pointsat the sameelevation.
Strataboundariescanbe steppedor slopedin CRISP,asthereis no requirementfor element

boundariesto be horizontal(or vertical). However,thereWill be problemsif thereis a switch
in unit weight y and/or&,, Fig. 3.16.
3.7.2 Sloping ground surface
Onefeatureof real siteswhich arisesoccasionallyis that of a slopinggroundsurface. This

impliesthat directionsof principalstressareno longerhorizontalandvertical,andthat -ry can
no longer be assumedzero. The pictureis actuallyquite complexand,of course,therewin be
lateralvariationsin stresslevelsalonganygiven elevationdueto the changingoverburden.
The only satisfactory way of modelling a ground surface which is originally sloping prior to
retaining wall construction is to create it as part of the analysis. The analysiswould start with
a level ground surface and a uniform variation of stresswith depth. Then, sufficient elements
would be removed to create the desired sloping surface profile (obviously the mesh would
have been devised to allow this). The time
steps and increments used to achieve this would
have to be selectedwith care, to
create conditions which are properly in equilibrium.
3.7.3 Sloping water table
Curvedphreaticsurfacestend to be be treatedin
associatedwith side-longground,andmust
the sameway. If a slopedgroundsurfaceis beingcreatedin the initial part of the analysis,an
initially horizontalphreatic
surfacecanalsobe drawndown by the useof appropriateexcess
pore pressureboundaryconditions,Fig. 3.17.
3-37
3.7.4 Non-hydrostatic conditions
Underdrainage andartesianconditionsarisenot infrequentlyandit is possiblethat an FE

analysiswill alsohaveto copewith somefuture changein pore water pressureregime. One
exampleof this encounteredby the writer wasfor the A406 Walthamstow (seeCase1,
AppendixA), whereporewater pressureon site prior to constructionshowedstrongevidence
of underdrainage, andthe designhadto anticipatefuture rising groundwater levels,restoring
profilesto full hydrostaticvalues. The actualpore water pressureprofile prior to construction
is definedwith asmany"in-situ points" asnecessatyto captureit faithfully.
Pore pressurefixing codesareinitially setaroundthe boundariesin sucha way asto maintain

long term
this under-drainedprofile throughthe early stagesof constructionandsubsequent
equilibrationof pore water pressure.Then,anyfuture groundwater level rise is simulatedby
increasingexcessheadson the boundaries(baseandside)andcausingthis to moveinward and
upwardsthroughthe mesh. In Fig. 3.18, area[a] representsa distributionof total headwhich
impliesheadlossverticallydownwards.Area [b] showsthe additionaltotal headrequiredto
provide a full rechargeto the hydrostaticcondition. (In an initially artesiansituation,the
reversemanipulationwould be carriedout.)
This approachto a movingphreaticsurfaceis not totally rigorous,but thereis no betterway at

presentof approximatingit with CRISP8.Anotherqueryrelatesto the correctnessof starting
with anunderdrainedprofile which impliessteadystateseepagedownwards;prior to SAGE
CRISPv3.02, an analysiswasintendedalwaysto start with hydrostaticconditions?.However,
asfar ascanbejudged, correcteffectshavebeenachieved(includingfuture rechargeto ffill
hydrostatic)wherethis procedurehasbeenfollowed in earlierversionsof CRISP.
Artesianconditionsare encounteredlessfrequently,to the writer's knowledgeandin his

experience.Artesianpressureswere encounteredat depthat the AldershotRoadunderpass
site (Case7, AppendixA), but the proximity of the River Blackwaterensuredhigh hydrostatic
groundwaterlevels;no attemptwas madeto includeartesianpressuresin the analysis.
a Between1993-97thewriterjointly developed for modellingrising/falling

more rigoroustechniques
in CRISP),Gum et al. (1997).
underanEPSRCcontract(notyet iniplerriented.
groundwater
aspart of thein-situ
In SAGECRISPv3.02andlaterversions,excessheadscanbeprescribed
conditions- pennittingananalysisto cumnencewith steadyseepage.
3-38
3.8 Boundary Conditions
3.8.1 Displacement
Typicaldisplacement
boundaryconditionsfor a retainingwall problemare straightforwardand
are shownin Fig. 3.19;the bottomboundaryis fixed againstverticalmovementandthe side
boundariesare fixed againsthorizontalmovement.It is conceivablethat the remotevertical
boundarycouldbe load controlled,but the potentialadvantagesof this areunclear.
It is difficult to think of a situationin a practicalretainingwall analysiswheredisplacement

controlledloadingwould be of particularuse,althoughmore fundamentalFE studiesof earth
pressuregenerationhavemadeuseof the facility to causea wall to rotateabouttoe or top, or
to translate(e.g. EvginandEisenstein,1985)
3.8.2 Drainage
Initial drainageboundaryconditions canonly be locatedalongelementedgeswhere

excesspore pressured.o.f exist. For an embeddedretainingwall, the locationof possible
drainageboundariesis shownin Fig. 3.20. BoundaryA maycorrespondto:
a) the ground surface- if low permeabflitysoil extendsto the top of the meshandthe
GWL is high
b) the interfacebetweentwo layersof contrastingpermeability- if the upper(more
permeable)layeris water-bearingandis modelledwith non-consolidationelements
C) the phreaticsurface- if this is within a layer of consolidationelements
The normalrequirementwill be for pore pressuresto remainat their initial values(ue= 0,

fixity code 1) throughoutthe analysis.In cases(a) and(c), it is alsoexpectedthat u=0.
BoundaryB is nearlyalwaysartificial, in the sensethat it rarely corresponds
to a real physical
featurein the ground. It is normallylocatedat a sufficientdistancefrom the wall that it will be
reasonableto assumepore pressuresarealwaysgovernedby regionalgroundwaterconditions.
The usualrequirementwill be for pore pressuresto remainat their initial valuesthroughthe
analysis(u. = 0, -fixitycode 1), althoughin generalu#0.
BoundaryC, unlike B, maywell correspondto a true physicalfeature. It maybe a relatively

rigid layer (e.g. very densegravelor rock) which, from the point of view of displacement
conditions,would form a logical lower boundaryto the mesh. The drainagecharacteristics
3-39
couldrangefrom impervious(e.g. freshgranite)to pervious(e.g. jointed chalk). The former
caseis a flow boundaryandit would be unnecessaryto defineany excesspore pressure
boundaryconditions pore pressuresarethenfree to changeasthey wish throughthe analysis.
-
The latter caseis a seepage
boundary(at constantheadif conditionsin the basallayerare not
changing),andthe excesspore pressureswould be setto zero (fixity code 1).
BoundaryD will only existfor symmetricalexcavations.It is anotherartificial boundaryin

that respect,but onewhich is, in effect,a flow boundary. Pore pressurechangesaregenerally
(andwould in fact be incorrect)to defineanyexcesspore
unrestricted,so it is unnecessary
pressureconditionson this boundary.
BoundariesE andF do not existat the in-situ stage,but only comeinto beingduringbulk
excavation.They are dealtwith in Section3.11.
3.9 Applied Loading
Applied loadsin CRISPcantake the form of point, line, patchand strip loads. True point and
patchloadsare only possiblein afull 3D analysis,Fig. 3.21(a-b). Giventhat most retaining
wall analyseswill be 2D planestrain,the two typesof loading be
which can appliedarefine
and strip, Fig. 3.21(c-d). (Which,in a planestrainsection,appearasa point andpatch
respectively).A "fine' load is definedby the nodeat which it acts,andits horizontaland
vertical components,Fig. 3.21(e). A "strip" load is definedby the elementside(s)alongwhich
it acts,andthe magnitudeof the normalandshearcomponentsat eachnodeon the element
side(s)- thus allowing for non-uniformdistribution,Fig. 3.21(f-h).
In CRISP,all appliedloadsare incremental;they are consideredactiveuntil increased,reduced

or cancelledaltogether. For example,the SouthwarkStationTicket Hall analysis(Case4A,
AppendixA) requireda shortterm constructionplant surchargeof 12 kPaextending16m
back from the wall, anda long term 7 kPa footpathload from 04m togetherwith a 12 kPa
road pavementsurchargefrom 4-16m. This was achievedby specifyinga normalstressof 12
kPaover the appropriateelementsright from the start of the analysis,followed by the
applicationof -5 kPa over the 4m adjacentto the wall.
3-40
Surfacetractionscanalsobe usedto simulateexcavationandfill construction,thoughthis has
no advantageover elementremoval/addition.On onesectionof the cut-and-covertunnelsat
the RegentsPark RoadJunction(Case6A, AppendixA), the rotationalstiffnessat the top of
the wall wasrequiredto checkthe assumedstiffnessinput to anotherprogram. This was
achievedby applyinga pure momentto the upperedgeof the top elementrepresentingthe
wall anddividingthis by the resultingrotation. The pure momentwas createdwritha double-
triangularstressdistribution(+q, 0, -q, kPa),but could equallywell havebeengeneratedwith
a couple- equalandoppositeforcesappliedto the endnodes.
CRISPprovidesthe option to specifythe proportionof load appliedin eachincrementof the

block. The defaultis a uniform weighting,so that if a block containsN increments,the load(s)
introducedin that block will be appliedIIN per increment.Alternatively,individualratiosR(i)
could be specified,e.g. 0.40,0.25,0.20,0.10,0.05"(E = 1.00)- forcing moreload to be
appliedat the beginning.Anotherpossibilitymight be to specify0.25,0,0.25,0,0.25,0,
0.25,0 - in which the loadsare effectivelyappliedover four increments,with the intervening
increments(whereR(i) = 0) beingusedpurelyto helpredistributeout-of-balanceloads.
3.10 Construction Modelling
3.10.1 Wall installation
The wall canbe instaUedin oneof threeways:

a) wishedin place
b) swappingconcretefor soil elements
C) hardening.
full replicationof boring, concreteplacement(by tremie)andsubsequent
If installationeffectsarethoughtto be minimal,option (a) is quite reasonable.To overcome

an initial lack of stressequilibrium,the unit weight for concretewould haveto be equalto the
unit weight of soil. The problemis that it is not known,aptiori, if the installationeffectsare
going to be minimal.
Option (b) would appearto be an improvementandis probablywhat is donein mostpractical

retainingwall analyses.Little is known aboutthe effect of numbersof incrementsin this
context,andthe writer is unawareof anypublishedstudieswhich haveaddressed
the issue.
341
The elementstiffnessmatrix K! is basedon the originalmesh(unlessco-ordinateupdateis
invoked);i.e. the componentsof B (Eqn 3.1) are calculatedusingundeformedgeometry. In
CRISPelementsareinitially unstressedwhenplaced(a problemin modellingfill construction),
so in the installationincrementblock the wall is simplypresentprovidingverticalnodalloading
throughself-weighteffects. Options(a) and(b) canbe usedin a coupledor non-coupled
analysis
The mostfaithful modellingof the installationprocessvia CRISP- option (c) - hasnot been
usedby the writer in anydesignscenario,but hasbeeninvestigatedby Gunnet al. (1993). In
their work they first replacedeachsoil elementwith equivalentpressuredistributionsto
replicatethe bentoniteslurry support. Thenworldng from the bottom upwards,new elements
of heaviermaterial(still without strength)were placedto representfreshconcretebeing
tremiedin. Theseconcreteelementssubsequently underwentfurther transitionsto acquirethe
stiffnessandstrengthof fully-curedstructuralconcrete. This ldnd of representationnormally
requiresa coupledanalysis;it is discussedin greaterdetail in Chapter6, as a considerable
amountof investigationhasbeencarriedout by others.
3.10.2 Excavation
Usersof CRISPwill generallymodelexcavationin an explicit manner,wherebyelements

representinglayersof soil will be removedfrom the meshin someappropriatesequence.For a
numberof years,CRISPwas the only commerciallyavailablegeotechnicalFE programwhich
calculatedforcesdueto excavationcorrectly. Erroneousadvicegivenby ChristianandWong
(1973)waspropagatedwidely, andled to solutionswith an apparentdependency on number
Ofexcavationsteps- evenin an elasticanalysis,which Ishihara(1970) hadprovencould not
be the case. Gunn(1982) identifiedthe problemand explainedhow it hadarisen.
Layer thicknesses
Severalaspectsof the modellingwill havebeenpre-determinedby FE discretizationin the
zoneconcerned.In CRISP-90,usingthe ADG prograni,minimumexcavationthicknessare
basedon superelements,which is not particularlyconvenient- SAGE CRISPallowsremoval
of individual"domain7'elementsthroughthe graphicalinterface.The excavationof soil layers
mustbe carriedout over a sufficientnumberof incrementsto ensurereasonableaccuracy.
3-42
Clearlythis is a rather subjectiveassessment
and,in commonwith manyaspectsof retainirig
wall analysis,little work hasbeendoneto establishfirm guidance.
Takefor examplea 2m layer of soil which is to be excavated.In the mesh,this might have
beenrepresented
as a singlerow of elements2m high,or two rows of elementseach1rahigh,
Fig. 3.22. The questionfor the analystis whetherexcavatingthe 2m thick row of elements
over, say,20 incrementsgivesthe sameresultasexcavatingthe Im thick elementsover the
samenumberof increments.In addition,in the secondmesh,both Im rows could be taken
out in oneincrementblock comprising20 increments,or eachrow couldbe takenout
sequentiallyin separateincrementblockscomprising10 incrementseach. All threeschemes
might be expectedto give similarresults,asthe sameoverallthicknessof soil is beingremoved
in the sametotal numberof increments.But is this the case? Certainlyin an elasticanalysis
the answermustbe the affirmative,or the FE codeis simplywrong. In an elasto-plastic
analysis,theremight well be variationsbetweenthe differentschemesoutlinedabove,but this
will requirenumericalinvestigation.
Geometricissues
Another issueconcernsthe implicationsof a planestrainidealization. Removingan elementor

row of elementsimpliesthe removalof a strip of soffwhich is infinitely long in the out-of-
planedirection. This is mostunrealistic(evenif the wall structureitself is very long) as
excavationwill normallytake placein bays. This win generallyleadto overestimationof wall
deflectionsduringconstruction,but thereis no easyway of avoidingit in a 2D analysis.
It is becomingincreasinglycommonin retainingwall constructionto usetemporarybermsasa

meansof reducingshortterm wall deflectionsandmoments.This is relativelyeasyto
incorporatein an FE analysis,simplyrequiringmanipulationof the meshin the zoneof
excavationto accommodatethe trapezoidalshapesusuallyemployed,Fig. 3.23. The typical
procedureis to excavateto maximumdig level acrossthe site,leavingin a soil bermresting
againstthe wall. A centralgroundslabwill thenbe cast,allowingthe top of the wall to be
struttedoff the edgeof the slab- thuskeepingthe centreof the excavationrelativelyclear.
Thereare no specialdifficultiesor provisionsin CRISPconcerningthe modellingof soil berms
(see,for example,Powrie et al., 1993;PowrieandDaly, 2002; GourvenecandPowrie,2000).
343
Ratesof excavation
Giventhe time scalenormallyassociatedwith excavationin clay,the conventionalwisdomis

that conditionsremainlargelyundrained.Typical time scalesfor (say)a dual caniageway
underpass in stiff claywould be for constructionto take 18-24months,andthe intended
designlife to be 120years. During construction,no more than 5-10% of the total excavation-
inducedexcesspore water pressureswould be expectedto dissipate,leaving90-95%to occur
post construction.An uncoupled,undrainedanalysiswould, therefore,be adequate.However,
in-situ drainagein stiff clay can often be controlledby fabric featuressuchaslenses/partings
of muchmorepermeablematerial(silt, sand),and fissureswhich canopenup dueto stress
relief This could acceleratedissipationso that the ratio (betweenconstructionandlong term)
might be nearer40:60 or even70:30. In thesecircumstances,the assumptionthat no drainage
takesplacecanbe n-dsleadingandunsafe,andit would be more appropriateto modelthe
excavationaspartiallydrained. Onepossibleapproachto this would be to usea valueof v or
K,, which was in betweenthe drainedandundrainedlimits (e.g. v=0.4 or Kv;:tsI OK'). This
hasthe attractionof beinguncoupled,but would requirecalibration- and, strictly speaking,it
would deliverpartial saturation(B<I) ratherthan partial drainage. -
A moreelegantapproachis to usea coupledanalysisto specifyexpectedexcavationtimesand

drainagecharacteristics.This hasthe advantagethat the post-excavationphasecanalsobe
examinedasa part of the sameanalysis.Whenperforminga coupledanalysis,time stepsare
allocatedaccordingto the expectedfield duration. This cannormallybe estimatedwith
reasonableaccuracywhencarryingout forward prediction(barring delays)
unforeseen - in
actualdurationswill be known andcanbe used. Time stepsare discussed
back-analysis,
further in 3.11.3in the contextof long-termequalization.
In orderto capturea truly undrainedresponsein a coupledanalysis(perhapsto comparewith

an undrained/uncoupled model),time At be
steps must sufficientlysmall- but how small?
Britto and Gunn(1987)warn againstmalcingAt too smallwhen changingexcesspore water
pressurefixities at a drainageboundaryasit will leadto numericalinstabilities- this hasbeen
confirmedindependentlyby Woods (1986). In the contextof bulk excavation,however,
drainageboundaryconditionsare not normallychanging- the upper surface(original ground
level)is the only drainageboundaryandthis is specifiedfrom the in-situ stage,so any excess
pore water pressuresinducedby wall installation(and subsequently by bulk excavation)are
alwaysableto dissipatetowardsit. Even so, theremight still be a lower limit on At when
3-44
removingelementsto simulateexcavation;the noviceusermight think "the faster,the better"
if theywereattemptingto achieveundrainedconditions,but this could leadto problems.
3.10.3 Temporary supports
Temporarysupportscanbe installedat anytime duringthe analysis;whethertheybe

compression(prop/strut)or tension(tietanchor)members.The variouswaysof representing
temporarysupportsfor retainingwalls was describedat lengthin an earliersection;it is
aspectsof installationandremovalwhich arerelevanthere.
Before a prop canbe installed,soil mustbe excavatedto the appropriatelevel. The only major
decisionwhich hasto be takenis whetherprop installationand soil excavationshouldbe
handledsequentiallyor concurrently.This is illustratedin Fig. 3.24. SchemeA andB have
exactlythe samefirst incrementblock, in which soil is excavatedto someconvenientdistance
below the intendedtemporaryprop location. SchemeA then goeson to installthe prop in an
incrementblock of its own, followed by the secondstageof excavationin a separateincrement
block. SchemeB on the other handinstallsthe prop andexcavatesthe secondlift in the same
block (in CRISPthis meansthat the elementchangelist includesthe numbersof elementsto
be excavatedalongwith the prop elementnumber).
The two schemesoughtto be identical(in a non-consolidationanalysis),for reasonsthat have

beendiscussedin the contextof wall installationandbulk excavation.Installingthe prop in
SchemeA, step(ii) hasno effect on the rest of themesh,asit is installedunstressed(like any
other elementwould be) - load is put into the prop (andwall deformationwill be inhibited)
whenthe next stageof excavationproceeds,step(iii). In SchemeB, step(ii), element
stiffnessesareassembledor ignored(as appropriate)at the beginningof the block, so the soil
is "out" andthe prop is "in" prior to the commencement
of excavation.
In a consolidationanalysis,all incrementblockshaveto be givennon-zerotime steps. It is

conceivablethat somedifferencescould be introducedbetweenthe two schemes, because
schemeA would experiencethe passageof time (albeitbrief) with just the prop in placeand
beforeany excavation.Howeverthis differenceis likely to be very small,unlessthe time
intervalwas largeenoughfor significantpore water pressurechangeto take place.
345
Oneof the mostimportantaspectof temporarypropsis their removal,andhow this is
modelledin the analysis.Temporarypropswill be removedat somepoint in the construction
sequence,usuallyafter permanentprops/anchorshavebeeninstalledand(in the caseof
anchors)prestressed.A high-levelprop will be carryinga significantproportionof the lateral
thrust from the retainedsoil, dependingon the ability of the wall to redistributeloadsto soil
finther downon the passiveside. It thereforefollows that the removalof sucha prop can
causea lateralunloading(on the wall) which is of a similarmagnitudeto the original
excavation.
The selectionof the numberof incrementsover which the prop removaltakesplacemust,

therefore,be accordeddue care. Whennon-linearor elasto-plasticconstitutivemodelsare
beingusedin CRISPit is most importantthat suchstresschangesare appliedin small
increments.Whena top prop is removedfrom, say,an 8m retainedheight,it maybe carried
out in 10-20incrementsat most,whereasthe excavationwill probablyhavetaken40-50
increments.However,prop removalmaycausethe equallysignificantload changeson the
wall-soil system.
Thereis no publishedguidanceon this matter,andthe writer is unawareof any detailedstudy

in the public domain. Theincrementblock load ratio featuremight be usedto advantagehere,
providinga meansof havinga non-uniformdistributionof load changein the prop removal
incrementblock(s). The questionfor the analystis, given a fixed numberof incrementsin a
block, what is the bestway to organizethe load distribution? If smallerratios areusedat first,
will this be undoneif largerratios areusedlater on? Perhapsthe only satisfactoryapproachis
simplyto usemore incrementsin the block concerned.
3.10.4 Permanent supports
Similarcommentsasthosemadein Section3.10.3apply,exceptthat permanentsupportsare

not removedafter installation.
Propsareusuallypassivewheninstalledandmuststrainbeforeload is generatedin them.

Someanchors(e.g. "deadmarf'anchors)are alsopassive,andtheir installationinto the mesh
simplycompriseselementadditionconnectingthe relevantwall and anchorblock nodes.
Prestressed however,
anchors, will requirethe applicationof forcesto the in
mesh, additionto
3-46
elementinstallation.Usinga singlebar element,prestressingis carriedout in the following
manner(seeFig. 3.25):
calculate the horizontal and vertical componentsof the prestressload P as Ph -`ýP cosO
and P, =P sinOrespectively, where 0 is the angle of inclination
H) applytheseloadsto the nodescorrespondingto the distal andproximalendsof the

anchor,in sucha way asthe two nodesarebeingpulledtogether
iii) installthe bar elementrepresentingthe anchor,which will thenbe put into tensionby the
requiredamountof prestressby virtue of the loadsappliedin step(H).
As the prestressis oftenlarge,attentionmustbe paidto the numberof incrementsover which

it is applied,asthe groundaroundthe distalendmayexperiencesignificantdeformation.
3.11 Long Term Equalization
Following bulk excavationandcompletionof the retainingstructure,the ground arounda

retainingwall will go througha processof adjustmentasexcessporewater pressures(positive
and/ornegative)dissipateandlong term conditionsare approached.Dependingon the
permeabilityof the wall andanypermanentsupportsin contactwith the ground,the new
constructionwill alter the originaldrainageboundaryconditions,thusaffectingthe equilibrium
pore pressuredistribution. The natureof the soil obviouslyplaysa crucialrole; in fine-grained
soils equilibrationmaytake manytensof years;in coarse-grained materials,very smallexcess
pore pressuresmaybe left to dissipatepost construction. The only way to do this is with a
coupledanalysis
3.11.1 Drainage boundary conditions
The drainageboundaryconditionsarethe sameasthoseidentifiedin Fig. 3.20 anddiscussedin

Section3.8.2, exceptthat bulk excavationandinstallationof structuralelementsmaycreate
new boundariesE andF.
Boundary A will not normally undergo any changein specified excesspore pressuresafter the
start of the analysis. One possible exception would be if the upper strata were dewatered,
which would effectively require A to move downward at the appropriate stage in the analysis.
347
The specifiedconditionsat nodesalongthe originalline of A could be left in placewith a
secondrow of nodesdefinedlower down alongwhich u=0 (u. = -ywHL,whereHL is the
amountby which the headis lowered). The upperrow at A are effectivelyredundant,but
would not causeerrorsif left in.
Furthermanipulationof excesspore pressureson boundariesB andC maybe specifiedif

attemptingto modela rechargingaquifer,for example(see3.11.2). BoundaryE is of
particularinterest,andhasprobablycausedthe most misunderstandingamongCRISPusers.
It is a realboundary,correspondingto the excavatedsurface,but it doesnot comeinto
existenceuntil excavationbegins. During excavation,it is not normalto specifyanyparticular
drainageboundaryconditionson this movingsurface,but ratherto wait until final dig level is
reached.The net effectof this is to preventthe soil from drainingto this boundaryat any
intermediatestage which could leadto an unsaferesultif partial drainageis anticipated,and
-
bulk excavationproceedsover a significanttime period.
Whenfinal dig level is reached,boundaryE will eitherbecomean impermeable

boundarywith
u=u, (e.g. undera deepbasementfloor) or a long term drainageboundarywith u=0 (e.g. a
permanentprop slabwith under-drainage).The former caseis straightforward- the structure
will eventuallyexperienceuplift (buoyancyeffects),andthe userneednot specifyany
particularconditionson boundaryE oncethe structuralelementsarein place. The latter case
involvesthe establishment seepage,
of steady-state and is by
achieved specifyingan
incremental excesspore pressure (fixity code 2) equal to -u., Fig. 3.20(b).
The misunderstanding
canarisebecausemost engineerswould considerit a surfaceof u=O,
whereasCRISP it
considers asa surfacealongwhich excesspore water pressure is <0 (u. = 0-
u. ). SinceCRISP93, it hasbeenpossibleto specifythe absolutepore pressureon the
boundary(= 0 in this case)usingfixity code3- particularlyusefulwherethe excavatedsurface
is inclined. Thesetwo approachesare identicalandneednot be checked- but the potentialfor
makingmistakesis present,so careis needed.
Pore pressurefixities aregenerallyspecifiedin the incrementblock immediatelyfollowing the

completionof excavation.Often,this block will comprisemanyincrementsof successively
largertime step typicallyaddingup to a total time of 60-120years. The excesspore pressure
-
fixities are appliedright at the beginningof the incrementblock (unlike load changes,which
348
arespreadover all the increments),so thereis no needto specifyboundaryconditionsin one
block, followedby the time stepsfor long term equilibrationin a subsequent
block.
BoundaryF, if it exists,evolveswith excavation.In most casesit will not actuallybecomea

drainageboundary-, concreteis generallyconsideredimpermeable, in
even relationto clays.
However,contiguousboredpilesoften havegapsof up to severalcentimetresbetween
adjacentpilesandit maybe appropriateto modelthe wall with consolidationtype elementsof
suitablepermeability;F would then comprisenodesdown the rear of the wall. Alternatively,a
certaindrawdownof the groundwater levelsbehindthe wall couldbe assumed,andthis could
thenbe imposedby introducinga drainagelayerto the relevantdepth;again,down the back of
the wall.
3.11.2 Fluctuating groundwater levels
Risinggroundwaterlevelsmayoccurbecauseof increasingpore water pressuresin an

underlyinglayer,infiltration at the ground surface,or rechargepumpingvia boreholes.Falling
levelsare generallydueto the samephenomenaactingin reverse.Wherethe near-surface
layer(s)of soil is coarse-grained,
water canusuallyflow relativelyfreelyandmovementsof the
phreaticsurfacelotake placein a relativelyshorttime. In fine-grainedsoils,however,the
situationis ratherdifferent. Seepageis muchslower,andit is normallythe piezometric" level
which is saidto be rising or falling.
Take, for example, the London clay overlying the chalk aquifer in the London basin, where
underdrainagewas causedby abstraction from the aquifer in the late 19th and early 20th
centuries. By the 1930s, the phreatic surface in the London clay was close to the top of the
clay layer, but the pore water pressureprofile was considerably less than hydrostatic (culaz <
Y, ) - implying head loss (andseepageflow) downwards to the underlying chalk. Effective
stresslevels in the clay were higher, relative to the original groundwater conditions.
Following the cessationof abstraction from the chalk just after the Second World War, the
aquifer began to recharge. In some parts of London, the piezometric level in the chalk was
rising faster than I m/year in the late 20th century.
10 definedas the surfaceconnectingpointswherethe porewaterpressureis zero.

11 the level to which waterwould rise in a standpipeinsertedin the soil layer concerned.
3-49
Modellingrising (or falling) piezometriclevelsin CRISPis actuallyquite straightforward.
Referringto Fig. 3.20, onewould first establishthe currentpore water pressureprofile (u.(z))
on boundariesB and C. Typicallythis would be the "long term" profile to which the analysis
hadinitially comeinto equilibriumfollowing wall installation. Then,the new (elevatedor
depressed)
profile (u. (z)) would be determined,
andthe differencebetween
the two (Au(z))
computed.This computeddifferenceis then specifiedalongboundariesB andC as an
incrementalexcesspore water pressure,usingfL-,dty code3. The changeof pore water
pressurewill moveinto the interior of the meshEkean advancing"front" until the new steady
state(hydrostaticor non-transientseepage)is attained. In somecases,it maybe adequateto
specifychangeson the bottom boundary(C) only - an aquiferrechargebeinga casein point,
wherethe true physicalsourceof the pore water pressurechangeis from below.
In the aforementionedexampleof the London basin,piezometriclevelswererising at I m/year

in someplaces,implyingan increasein pore water pressureof 10 kPa/yearat the top of the
chalk(boundaryQ. If the drawn-downpiezometriclevel hadbeencloseto the top of the
chalkprior the it
cessationof pumping, would take of the 50 for
order of years the piezometric
levelto reachthe groundsurface(assuminga London claythicknessof about50m), at which
point aquiferpore pressureswould haveincreasedby some500 kPa.
In the writer's experience,the needto modelfuture rising groundwaterlevelsarisingfrom

is
underlyingaquiferrecharge not common, and has only happenedon one of the commercial
retainingwall analyses he has conducted. In that instance (A406 Walthamstow; AppendixA,
Case1), the former of the two strategiesabove(i.e. instantaneous recharge)was used,with a
period of 120 for
yearsallowed equilibration- which followed on from a period of similar
lengthto allow equilibrationto the present-day(underdrained)pore waterpressureprofile.
The modellingappearedsuccessfidandwas ableto showsomeimportantconsequences of this
long-termrechargeof pore water pressures.In particular,the changingporewater pressures
on the it
wall caused to rotate aboutthe singlepermanentprop at formationlevel, as the line
of action of the resultantwater thrust movedupward.
3.11.3 Time steps
The analystmust decideboth the lengthof time (Q andnumberof increments(N. ) to allow

for equalization,aswell asthe sizeof the individualtime steps(At) to be used(EAt = t. ).
3-50
Equalizationtime t. is often specifiedaspart of the analysisbrief (e.g. I-fighwaysAgency
structureswith a 120yeardesignlife), or fixed by an interestin a particulardate(e.g. back-
analysisof present-dayfield datafrom an earth-retam'ing structureconstructedsomeyears
earlier). Sometimestheremaybe a genuinedesireto achievet9s,perhapsto comparewith a
drainedanalysis.In this case,the analystwill not know how muchtime is required,andmust
guess. Thereis then the risk of using insufficient time,
or'excessive leadingto inaccurateor
inefficientresultsrespectively.
Within the overalldurationof the equilibrationblock (4), the individualtime stepscanbe

uniform or variable. This that
means for increment
'T in the block eitherAti i (t, or Ati
'ý)
f(i), wheref is a function. Uniform stepsarethe defaultin CRISPif the individualAt are
not specified(ITNE=O), but this would be an choice.
unusual It is widely known that the rate
at which excesspore water pressures dissipate decays
with time, so the At shouldbe smallto
beginwith and graduallyincreasein size- but what weightingschemeis best? Conventional
wisdom is to At
start off with small andincreasethem using a log scale(1,2,5,10,20 etc).
Time stepsmustbe largeenoughto avoid pore water pressureoscillations(both temporaland

high
spatial)yet smallenoughto preventunacceptably in
changes effective stress(bearingin
mind the possiblelack of stresscorrections; e.g. with CSSM models). The first (lower) limit
At
on arises from the fact that a "front" be
of pore water pressurechangemust givensufficient
time to movein from a drainageboundaryto the nearestline of nodeswith excesspore water
pressure degrees
of freedom (Britto,and Gunn, 1987). Take the example illustratedin Fig.
3.26 which showsa prop slaboverlyingsoil with a drainageboundaryat thejunction. The
line
nearest of nodeswith excesspore water pressure d.o.f is situateda distance Ay below the
drain. Using one-dimensional consolidationtheory andparabolic isochrones, Schofieldand
Wroth (1968) establisheda relationshipbetweenthe time At to travel a distanceAy :
At (Ay)2 / 12 C, (3.26)
=
where C, is (=
the coefficientof consolidation k/y,, m., ;t; k E' /
When manipulating boundary excesspore water pressuresto simulate rising/falling

groundwater levels, the restrictions on size of time step are no different to those which apply
when specifying drainage boundaries. In increments where drainage boundary conditions are
not being specified, a lower Emit At
on will stiff apply but will be more to do with preventing
3-51
numericalill-conditioning. In a designsituation,however,it is morelikely that therewill be a
temptationto exceedthe upperlimit on time step,asthe transitionto long term equilibrium
(Commonlyspecifiedas 120years)is normallymodelledin asfew incrementsaspossible.
Thereis no correspondinglysimplerule-of-thumbwhich canbe appliedto estimatethis upper
limit form, andin practicea form of retrospectivecheckis applied,dependingon the
constitutivemodel(discussedin the next section).
3.12 Analysis Verification
For elastic-perfectlyplasticmodels,scrutinizingthe equilibriumerrorsat the endof each

increment(seeSection3.3.6) anddecidingwhetheror not they are acceptableis the only form
of validationnormallycarriedout. The critical statemodelsoffer an additionalmeans by
which the time stepcanbe judged, in the form of the yield ratio (Y. R-). If plasticyielding is
taking placeandthe soil is strainhardening(or softening),the yield locuswill be growing (or
shrinking)and the linear scalefactor for this is
transformation definedasthe yield ratio. As all
hardening/softeningis isotropic,yield ratio canbe calculatedfrom:
Y-K = (P, )B / (P'e)A (3.27)
where the terms (p'c)Band(P"JA are definedin Fig. 3.27 (NB: YR canbe greateror lessthan
unity, depending the
on whether soil is hardeningor )
softening.
For changesin effective stressto fall within acceptablelimits, Gunn and Britto (1982,1984,
1990) recommend that the yield ratio must fall within the range 0.95: 5 Y. R. :51.05. Thus the
yield locus is allowed to change size by ± 5%, beyond which unacceptableerrors are reckoned
(by the program authors) to start entering into the analysis. This may seemrather stringent,
but in the writer's experienceit is necessary,in view of the fact that no iterative corrections
are applied to the critical state models in CRISP for stress stateswhich move outside the yield
locus. Further discussion on this can be found in Woods (1986).
The check on elastic-perfectly plastic models is rather more arbitrary. Stresspoints which
move outside the yield surface are corrected back (see Section 3.3.8), so the amount of out-
of-balance force in
generated any given increment is not so critical. The authors of CRISP
recommend (Gunn and Britto 1982,1984,1990) that the percentage equilibrium error be kept
3-52
below 10%but unlike the yield ratio (which is reportedfor everyelement)the percentage
equilibriumerror is a summaryfigure givenfor the whole mesh. This makesit very difficult
for the userto assesshow muchthe meshis sufferingfrom excessiveout-of-balanceforces
-a
very largereportederror (sayý: 100%)mayonly applyto onesinglenodeout of the several
thousandpresent.
3.13 Summary
CRISPis a widely distributedandusedgeotechnicalfmite elementpackage.Almost certainly

this is dueprincipallyto the openavailabilityof the code,which hasencouragedboth
to useit. The formergroup are ableto validateits useon
practitionersandresearchers
commercialwork becauseit is not a "black box", whereasthe latter areableto modify the
codeandaddnew models. Although CRISPhasneverenjoyedmajor commercialbacking,it
continuesto thrive andhasa committedfollowing from someof the leadingUK andoverseas
universitiesandcivil engineeringorganizations.
However,muchof the developmentandtestingseemsto haveneglectedmanyof the issues

which facethe designeron a regularbasis. The sophisticationof CRISPis increasing
continually,but manyproblemsrelevantto commercialusersgo unaddressed.This thesisaims
to makea contributionin this area;specificallywhenappliedto the analysisof embedded
retainingwalls.
This chapter has sought to

a) explain how the CRISP package has come to exist in its present form
b) summarizethe mainfeaturesandfacilitiesof CRISP,and
C) describecurrentpracticein modellingretainingwalls with CRISP- the mainissues
which arise,andhow practitionersusuallytacklethem.
Examplesfrom commercialretainingwall analysesconductedby the writer havebeenreferred

to asappropriate,alongwith other publishedexamplesin the literature.
However,it shouldbe clearto the readerthat, althoughFE (and CRISPin particular)areused

for commercialretainingwall analysis,there are still major uncertaintiesin manyaspectsof the
modellingprocess. The attentionof manyinvestigatorshasfocusedon constitutivemodelling,
3-53
but the writer is convincedthat other modellingissuesarejust asimportant. Runningthrough
the mainfeaturesof CRISPin order (Section3.2), the uncertaintieswhich aredealtwith in the
remainderof this thesisare asfollows.
The elementprovisionin CRISPis adequatefor mostgeotechnicalproblems,perhapswith the

exceptionof a stableandversatileinterfaceelement.Another areaof contentionis that of
reducedintegration,but the writer is not convincedthat a clear-cutcaseexistsfor this.
Ultimatelyit is the meshselection(grading,boundarylocation,etc.) which still causesmost
difficulty - andhasreceivedlittle attention. This will be tackledin Chapter4, whilst issues
surroundingelementaspectratios (which arenumericalin nature)areexaminedin Chapter7.
Constitutivemodelsareplentiful in the commercialreleaseof CRISP,with a commitmentto

includemore in future versions. The writer would contendthat it is the use,understanding
andparameterdeterminationfor thesemodelswhich is somehowlackingandneedsto be
addressed.For example,an appreciationof the influenceof anisotropywill assistin deciding
whetheror not it shouldbe incorporatedin an analysis.An investigationon thesefactorsis
presentedin Chapter5.
The facility for drained,undrainedandcoupled-consolidation

analysescoverseveryeventuality
for saturatedSoiIS12The real uncertaintysurroundsthe choiceof K,, in an undrainedanalysis
.
andof At in a coupledanalysis,which is consideredin Chapters6 and7.
Definition of in-situ stressesand boundary fixities is flexible enough for all caseslikely to be
encountered, based on the writer's experience(see all casesin Appendix A). Side-long
ground will always be a problem, but there is little alternative to starting with a level ground
surface and creating the profile as part of the analysis. The boundary fixity which frequently
causesconfusion is that of excesspore water pressures,but this is more a question of user
understanding than program deficiency. A range of example casesand documentation would
be helpful.
12 Theheightened thatthis is oneof

awareness soilmechanics
of unsaturated in recentyearssuggests
thenextareasto betackled,but its relevance
to embeddedretainingwallshasyetto bedemonstated.
3-54
Appliedloadsare straightforward especiallyif their usefor simulatingexcavation/fillingis
-
avoided.Therearefew reportedor experiencedproblemshere;thereforethe thesisneednot
examinethemfiinher.
Changinggeometrypresentsvariousmodellingdilemmas elementsize,incrementnumbers,
-
etc. More specialistproblemsincludeinstallationeffectswith castin-situ walls, prop insertion
andremoval,andaddingelementswhich cannothavea stresshistory. Investigationson this
arepresentedin Chapter6.
The solutionschemein CRISP,from the userviewpoint,cannotbe changed.However,

understandingthe role of incrementsizeis vital in conductinga properanalysis.This too is
exploredin Chapter6, whilst load ratio schemesare studiedin Chapter7.
Any finite element analysisproduces a large quantity of output, and a retaining wall analysisis
no exception. The wall and ground movements, stressdistributions and structural forces
require varying levels of manipulation of the "raV' output. There can be various ways of
doing this, but the user is given little guidance. Obtaining output for use in design is covered
in Chapter 8.
3-55
Table3.1 Elementtypesin CRISP
Element name Desig- No. of Type Displ. Excess Rotn Total

nation Nodes d. o.f. P-w-P- d. o.f. d. o.f.
no.
d. o.f.
Constant strain bar 14 4- - 4

-2
Constant strain beam 15 4- 2 6
-2
Linear strain bar 1 6- - 6
-3
Linear strain beam 12 6- 3 9
-3
Linear strain interface 13 12 - - 12
-6
Linear strain triangle LST 6 2 12 12

Linear strain triangle LSTp 6 3 12 3 - 15
Linear strain quadrilateral LSQ 8 6 16 - - 16
Linear strain quadrilateral LSQp 8 7 16 4 - 20
Cubic strain triangle CuST 15 4 30 - - 30
Cubic strain triangle CuSTp 22 5 30 10 - 40
Linear strain brick LSB 20 8 60 - - 60

Linear strain brick LSBp 20 9 60 8 - 68
Table3.2 Someof the constitutivemodelsimplementedin CRISPby researchers
Constitutive model University Reference
3-surface kinematic hardening (3-SKID city Stallebrass(1990)

Naylor continuous plasticity critical state Surrey Hillier (1992)
K,,-consolidated Cam-clay Cardiff Newson (1992)
Power law (non-finear elastic) Surrey Gunn (1993)

Duncan-Chang hyperbolic Surrey Woods (1993)
Partially saturated Sheffield Ali (1993)
Simpson "bricks on strings!' QMWC Chandler (1995)
Jardine (Imperial College) model Surrey Woods (1995)
Hyperelastic Sheffield Pierpoint (1996)
S-CLAYI (plastic anisotropy) Glasgow Wheeler et al. (2003)
3-56
Linearstraintriangle (LST) Linear strain triangle with excess p.w. p
6 nodes,12 d.o.f. unknowns (LTSp) :6 nodes, 15 d.o.f.
Cubic strain triangle (CuST) Cubic strain triangle with excess p.w. p
15 nodes, 30 d. o.f. unknowns (CuSTp) : 22 nodes, 40 d.o.L
Linear strain quadrilateral (LSQ) Linear strain quadrilateral with excess p.w. p.
8 nodes, 16 d. o.f. unknowns (LSQp) :8 nodes, 20 d. o.f.
Linearstrain brick (LSB) Linear strain brick with excess p.w. p

20 nodes,60 d.o.f. unknowns (LSBp) : 20 nodes, 68 d. o.f.
displacement d.o.f. excess pore water pressure d. o.f.
Fig 3.1 Continuumelementtypesavailablein CRISP
3-57
F.
"I-I
(a) basic mesh
(b) modificationfor slopingstratainterface
(C) modificationfor bermed,excavation
Fig 3.2 Simple retaining wall mesh on a rectangular grid
3-58
(d) moment-free connection
(e) dished prop slab
(f) prestressedground anchors
Fig 3.2 Simple retaining wall mesh on a rectangular grid (contd)
3-59
Fig 3.3 Simplemeshon a rectangulargrid, with more sophisticatedgrading
superimposed
INI
IN.
ZIL
Fig 3.4 Triangular elementsgeneratedwithin a rectangular grid
3-60
excavation wall retained soil
A---- --
(a) locationof Gausspointsin linear strainquadrilateralformingpart of wall
--------------
ab
----------------
(b) sectionAA through3 Gausspoints
Gya Cyyb (Tyc
t"'I
kW typical stress diagram (pure bending + compression)
iiix global axes
-t/2 0 +t/2
iiý localaxes
NB: x= (t/2) 4; dx = (t/2) dt
(d) relationsMpbetweenglobal (Cartesian)andlocal (element)axes
Fig 3.5 Transversestressdistributionin a retainingwall
3-61
(a) contiguous wall (s 2: d) I (7r/64) d 4(1/S) per m run
-------------- . ... ... ..
4
secantwall (s < d) I= (d /64)(7c-2p/3)(1/s) per m run
3
where 0= 3oc - 3sinacosa - 2sin OLCOS(X
a= cos-'(s/d)
Fig 3.6 Bored pHewaHs
excavated side
3+ 2+ 3+
wt wt (d+t/2-yc) bd bd (yc-d/2
1-2 T2-
+ bd 2/2
yc = wt(d+t/2)
bd + wt
Fig 3.7 T-section diaphragm wall
3-62
(a) fuff momenttransfer (b) partial momenttransfer
(^N
zero momenttransfer(concentric) zero momenttransfer(concentric)
(e) zero momenttransfer(eccentric) (f) zero momenttransfer(eccentric)
Fig 3.8 Wall-prop connectiondetails
3-63
(a) horizontal
(b) dished
(c) cranked (split-level)
Fig 3.9 Different permanent prop slab geometries
3-64
simpletrianciularelement
wall
Fi93.10 Rigidpropdetail
free (unbonded) length fixed (bonded) length
Fig 3.11 Components of a prestressedground anchor
3-65
single 2-noded bar
(b) single 2-node bar for free length; multiple bars for fixed length
Fig 3.12 FE representationsof a prestressedground anchor
3-66
-V-
Fig 3.13 Mesh detail for NeasclenLane Underpass(Creed, 1979)
Fig 3.14 Idealised stressdistributions on a cantilever wall
3-67
a tension
KK
rs,
8y 8x
Ks
Kn V1
compression
(a) normalstiffness (b) shearstiffness
Fig 3.15 Normal and shearload-deformationcharacteristicsof the Goodman

interfaceelement
zone1
Zi
7A .................. Z2
zone 2
0 ...............
' *"*"' " «'* .. zý
......
c
(ýý:
+ T2Z2
OvA "ý 'fIZI C7VB ý TI (Zi + Z2)
to the left of C: CT'h K%02Cy'v

to the dght of C: Cy'h KO, crl,
Fig 3.16 In-situ stress definition in the vicinity of a sloping strata boundary
//A\\V//
hi odginal phreatic surface h2
--- ........ .................
must coincide with a Ilineof nodes which

possess excess pore pressure d.o.f.
Fig 3.17 Modelling a non-horizontalphreaticsurface
3-68
pore water pressure u
depth
z
(a) comparisonof present(underdrained)andfuture (hydrostatic)pore water

pressureprofiles
---- ----------------------------------------
\_
(b) distributionsof boundarypore pressurebefore[a] andafter [a+b] a rising

phreaticsurface
Fig 3.18 Modelling a present-day underdrainedpore water pressureprofile, followed

by future recharge to full hydrostatic conditions
3-69
q
D B
C
Mesh Boundary Horizontal Movement Vertical Movement
A free free
B fixed free / fixed
c free / fixed fixed
D fixed free
E (depends on FL slab)
Fig 3.19 Typical displacementboundary conditions for an embeddedretaining wall
IE
E --------------------------------------------- --------- --
D B
C
Mesh Boundary Drainage Condition Au (incr) Au (abs)
A permeable 0
B permeable 0
c permeable 0
D impermeable -
E permeable #
F impermeable
# depends on FL slab etc - see Fig 3.20(b)
Fig 3.20(a) Typical drainage boundary conditions for a retaining wall analysis
3-70
If XXXXX is a free-draining
layer, under the slab u=0
now u=u,, + u.
and u. = y,,h,,
o= 7whw+ Ue
Ue
= -y, h,
if XXXXXis not free-draining
(or is absent), under the slab
u# 0 and nothing needs to be
specified in CRISP
Fig 3.20(b) Drainageboundaryconditionson excavatedsurface
3-71
P
(a) patch load (b) point load
(c) line load (d) strip load
14 Fy
Fx
(e) line load in 2-D (f) stressdistribution
normal stresses (h) shear stresses
Fig 3.21 Extemally applied loads in 3 dimensions
3-72
2m
wall
(a) removal in Ix 2m layer
lm
lm
wall
(b) removal in 2xIm layers
Fig 3.22 Different excavationlayer thicknesses
(a) primary mesh - prior to excavation
intermediate mesh - partial excavation
Fig 3.23 Incorporation of temporary berms into FE mesh
3-73
Scheme A Scheme B
excavatelift 1 (ii) excavatelift 1
(ii) install prop install prop and excavate

lift 2 (same inc. block)
(iii) excavate lift 2
Fig 3.24 Possibleexcavationand propping sequences
3-74
PCOSO
.................. .................. ................. ................. ................. ......
.............
PsinOf
..................
.................. L.................
..................
....................................
...... ............... ................
..................
.................. ..........................
........................................................
. ........................................................
(a) apply horizontal and vertical components of prestress load
(b) install anchor (bar element)
Fig 3.25 Installing and prestressing a ground anchor
Fig 3.26 Mesh detail near a drainage boundary
3-75
qT CSL
ýý
......... .......... hardening

(Y.R. > 1)
...... -------
fte ing
(Y.R. < 1)
((P'C)b
P'C)b Pcic P
current yield
locus
...............................
W.
(CF1
NB: q= deviatoric stress - C73) =
P, = mean effective pressure = (a', + 2a'3) /3

pIC= preconsolidation pressure
CSL = critical state line
Y. R. = yield ratio
Fig3_27 Definition of yield ratio for Cam-clay models

4ý1
3-76
CHAPTER 4
GEOMETRIC MODELLING AND DISCRETIZATION
4.1 Introduction
Thefirst decisionconfrontingthe retainingwall analystis how to convertthe actual,

continuousphysicalsystem(comprisingthe wall, supportsystem,andthe surroundingsoil)
into an assemblageof finite elements.This processis called"discretization7andinvolvesthe
following stages:
decideif a 3D representationis necessary,

or if a 2D (planestrainor axisymmetric)
modelwill suffice
judge how muchof the surroundingareaneedsto be modelled(this will becomethe
"domain" of the problem)
locateanyaxesof symmetry,in orderto reducethe sizeof the domainwhich mustbe

modelled
iv) determinewhich strataboundariesneedto be reproduced,basedon significantchanges
in engineeringproperties
V) if the existingground surfacewithin the domainis partly or wholly sloping,definea

horizontalsurfaceoverlyingthe existinggroundsurface(the surplusgroundto be
excavatedat an early stageof the analysisto createthe requifedproffle').
vi) if the structuralgeometryis complex,considerhow it canbe simplifiedwithout

introducingsignificanterror
vii) if any stageof the analysiswill involvemanipulationof groundwater levelsand/or

drainageboundaries,identify lineswithin the domainwhich will needto coincidewith
nodespossessingexcessheaddegreesof freedom.
At this point, no firm decisionhasyet beenmadeon numberof elements,elementtype, mesh

gradingetc. The decisionstakenthusfar serveto definethe frameworkwithin, aroundand
a this is a moreconvenientalternativeto attemptingto defte in-situ stressesundera sloping

groundsurface,whereprincipal planesareno longerhorizontaland vertical.
4-1
uponwhichthe finite elementmeshwill be generated.Before arriving at the final mesh,
severalfurtherstagesarerequired:
viii) decideroughly how manyelementsareto be used,within the constraintsof computer

memoryandprocessorspeed- andwithout compromisingaccuracyto unacceptable
levels
decidewhich type(s)of element(s)aregoing to be most suitablefor the various

componentsof the domain(soil, wall, supportsystem,and soil-structureinterface)
X) decideon the approximategrading(i.e. variationof elementsize)by locatingprincipal

elementboundariesrunningthroughthe domainandidentifyingthe approximatesizeof
elementthat would be appropriatein eachsub-region
xi) asit is easierto generatea meshbasedon a regulargrid of horizontalandvertical fines,

decideif it is feasibleto createsucha meshfirst and adjustit later to accommodate
the
real geometry
xii) examinethe meshto locateany elementswith unacceptablyhigh aspectratios, or to

identify areaswhereelementsarebeingusedinefficiently(the formermayleadto
numericalproblems;the latter to excessivememory/diskspacerequirements)and carry
out local adjustmentsasnecessary
xiii) for elementswhich form the wall, supportsystemandinterface,checkthe detailed

geometry- payingparticularattentionto connections,etc. (if overlayelementsare to
be used,definethemat this stage)
xiv) finally, checkfor any errorssuchasholesin the mesh,wrongly-positionednodes,

missingprops,consolidation-typeelementsbeingusedfor structuralcomponents,etc.
The key point to note with all of thesestepsis that a decisionhasto be taken,and/oran
approximationmade- and everysingleonewill havean effect on the resultsof the subsequent
analysis.The inexperienceduserwill havevery little (if any) informationon which to base
thesedecisions,andwill not be ableto distinguisha "good" meshfrom a "pooe' one. Even
the more expertuseris unlikelyto havedetailedknowledgeof the consequences of varying
theseassumptions,or of makingdifferentdecisions.In a designsituation,it is fair to saythat
4-2
the analystwill selecta "best shot" meshwhich "looks aboutright". This meshwill then be
usedfor the analysis(or analyses,if parametricstudiesare to be carriedout). Rarely,if ever,
will the meshbe adjustedanda re-analysisperformed- time andbudgetwill not permit.
Virtually all texts on FEA recommendthe userto experimentwith meshgrading,starting

with
a relativelycoarsemeshandrefiningit, in orderto find the optimumgrading. In an
unboundedphysicalproblem(e.g. mostgeotechnicalcases)thereis a further issueof boundary
location. And yet lessthan 1% of reportedfinite elementanalyses
of retainingwalls (fistedin
Table2.4) makeexplicit mentionof any suchstudieshavingbeencarriedout. The implication
is eitherthat meshgradingetc. hasbeenvariedbut not reported, that the
or meshused(and
perhapsshownin the reference)was selectedandusedwithout any sensitivitychecking,Any
errorsintroducedthrough discretizationarethereforeunquantified.It mustbe assumedthat
sucherrorsmayexist,until studiesrevealthe opposite.
MorgensternandEisenstein(1970)carriedout numericalstudieson meshbaselocation,for a

rigid wall in front of which excavationwas carriedout. The baseconditionwas eitherrough
or smooth. However,the wall hadno embedment, andthe studieswere consideringthe mesh
baseto be a real rigid boundaryratherthan an arbitrarycut-off in an unboundedmedium. The
only output quantityobservedwasthe lateralearthpressuredistributionon the wall, so the
usefulnessof this studyto embeddedretainingwall analysiswas very limited.
Ou et al (1996)conductedstudiesinto how manyelements(N) were neededin andaround

the excavatedareafor a 3D analysis,Whilstkeepingboundarylocationandconditionfixed.
They examined5:z profiles asN wasincreasedandconcludedthat only a relativelymodestN
was requiredfor the profile to convergeto (presumed)correctvalues. However,the writer
believesthat their boundariesweretoo closeto the wall andto formationlevel,andthus
convergencewas prematureandnot to the correctprofile.
So, the most fundamentalof questionsrelatingto finite elementanalysisof retainingwalls

goesunansweredin the literature. it is puzzlingthat no investigatorsappearto havevalidated
their meshes- surelya prerequisiteto demonstratingthe validity of an analysis.This chapter
will investigatethis issuein detail,alongwith manyof the decisions(i)-(xiv) listed above. In
Particular,attentionis paid to thosedecisionswhich mayappearquite arbitrary;viz., (H),(viii)
4-3
and(x). Thishasbeendoneby conductinga numberof numericalexperimentswith the
CRISPfinite elementpackage,which wasdescribedin the previouschapter.
4.2 Location of Boundaries
The first numericalstudywas conductedin orderto addressthe question:"how far away

shouldthe meshboundariesbe placed? "' The inexperienced
analystmay consulta paperor
text bookto seeexamplesof meshesusedby others,but the ground profile, stresshistory, soil
propertiesetc. maybe quite differentin his/hercase. The validity, therefore,of using
publishedmeshesmustbe consideredrathercarefully.
4.2.1 Finite element mesh
A relativelylargemeshwas created,with a reasonablyfine gradingof elementsin both

directions,Fig. 4.1. The boundarylocationshavebeencharacterizedby two geometricratios,
X/H andY/W, where
is the distancefrom the rear faceof the wall to the remotevertical boundary,
Y is the distancefrom the bottom of the excavation(i.e. formationlevel) to the bottom

horizontalboundary,
H is the wall height(or depthof excavation),
is the half-widthof the excavation.
The depthY was normalizedwith the width W on the basisthat, in foundationloading/

unloadingproblems,it is the width of the load which determinesthe depthof penetrationof
the stresschanges.The normalizingof X by H follows a similarargument,appliedin the
lateraldirection.
A wall heightH of 8m was selected,which is typical of the dimensionsencounteredin urban

underpassschemesandthe like (althoughexcavationscanobviouslygo significantlydeeper
thanthis). A penetrationdepthD equalto H was adopted- recognizingthat unpropped
cantileverwalls maywell requireD/H >I for stability,whiIst proppedcantileversmaybe quite
adequatewith D/H < 1. Onemay considerD/H =I to be a compromisewhich would allow
4-4
somesimplestudiesof boundarylocationto be madefor a numberof wall supportsystems.
The excavationhalf-widthW (symmetryis assumed)was alsofixed at 8m, giving D/H =--W/H
= 1. A wall thicknesst=0.8m was selected(whichis againfairly typical) giving t/H = 0.1-
The largemeshshownin Fig. 4.1 hasX(H = Y/W = 10,andwas thus denotedxlOyIO. The
discretizationof meshxlOyIO was arrangedto incorporatesmallermesheswith XM andY/W
ratios of 8,6,4, and2. Hencemesheswith variouscombinationsof X/H or Y/W couldbe
obtainedeffectivelyby removingrows and/orcolumnsof elements.In the first seriesof these
analyses(referredto as SeriesA), the meshsizewasvariedin sucha way that its overall
(global)aspectratio was kept constant,Fig. 4.2. Thisgeneratedfive separatemeshes,
designatedby the X/H andY/W ratios asx2y2, x4y4,x6y6, x8y8, andxlOyIO (smallestto
largestoveralldomainsize).
4.2.2 Analysis parameters
Even in an idealizedanalysissuchasthis, therearepotentiallya largenumberof parameters

which could be varied. In order to makeprogress,it was decidedto fix selectedparametersat
typical, representative
values. Examples of theseare the wall (which
thickness would be more
or lessfixed in a real designsituation),andthe in-situ lateralearthpressurecoefficient. There
were other parametersor aspectsof the problemwhichwere more likely to influencethe
presentareaof interest,andthesewerevariedover a reasonablerangeto avoid arrivingat
conclusionswith little generalapplicability. (Evenso it is recognizedthat the presentstudyis
) Tables4.1 and4.2 summarizethe problemparameters,distinguishing
far from exhaustive.
betweenthosewhichwere fixed andthosewhichwerevaried,respectively.A justificationand
discussionof the parametersselectedis givenin the following subsections.
Wall mipportlpropping
The proppingarrangements referred to in Table4.2 are illustratedin Fig. 4.3. All propswere
rigid, comprisinga single,stiff triangularelementwith onevertex nodeconnectedto the wall
andthe oppositeedgerestrainedagainsthorizontalmovement(only). Thesepropswere
installedat the earliestopportunity,which is reasonablefor the top prop (casesb, d, e) but
mayleadto unrealisticallyearlywall restraintin two of the caseswherea formationlevelprop
is used(casesc andd, but not e). However,it wasthoughtthat this would stiff produce
meaningfulresultsin the analysespresentedhere,andis discussedftifther underthe headingof
)
itwall installation!' later on in this section. k-I
4-5
Drainagerates
Excavationsin soilswith a high clay contentwill generallytake placeunderconditionsof

undrainedloading,followed by a periodof pore water pressureequalizationwhich maylast
severalyearsor evendecades.At the other extreme,excavationsin coarse-grained soilsmay
be consideredfully drained,with little if anyfurther porewater pressurechangesafter
constructionis complete. However,thereis growing evidencethat excavationsin clay soils
maybe accompanied by far more excessporewater pressuredissipationduring construction
thanis generallybelieved. This partial drainageresultsfrom fissuresopeningup understress
reliet laminationsof coarsegrainedmaterial,etc. Giventhat the stabilityof an excavationis
controlledby the long-termdrainedcondition,this partialdrainage(if not allowedfor) can
havepotentiallyseriousconsequences.
To ensurethat the conclusionsdrawnin this particularstudywere of broadapplication,both

loading/drainagelimits were considered.Undrainedanalyseswere conductedin termsof total
stress;drainedanalysesin termsof effectivestress. It was decidednot to usethe effective
stressmethodfor undrainedloading,wherebydrainageis determinedby the "switch" of
settingpore fluid bulk modulusKw eitherto zero to a valueapproximately100 x K', giving
the drainedandundrainedresponses,respectively.(This particularmodellingdecisionis
consideredfurther in Chapter7, asit hasbeenknown to give rise to numericaldifficulties.)
Soilparameters
A smallbut representative
set of constitutivemodelsandassociatedsoil parameterswere used,
so that anyemergingconclusionson meshboundarylocationcould be verified over a rangeof
soil behaviour.Theseincludednonhomogeneity, anisotropy,andyield. It could be argued
that a fourth feature,permeability,hasbeenincorporatedthroughthe decisionto carry out
both undrainedanddrainedanalyses.In a non-coupledframework,drainagerate is essentially
a Poisson'sratio effect. The undrainedconditionis characterizedby incompressibility(i.e. no
volumechange),which arisesfrom the fact that isotropicbulk modulusis definedby:
E/3 (1-2v) (4.1)
andsettingv=v,, = V2createsan infinite bulk modulus(via divisionby zero). This is a well-

known problemin finite elementanalysis,andis avoidedby usinga valuecloseto 1/2suchasv,,
ý=0.495. Of course,the exactvalueusedfor vu in a total stressanalysisis anotherdecisionto
4-6
be takenby the analyst. The drainedconditionis modelledthroughusingan effectivestress
Poisson'sratio V in the range0.1 0.3, andby scalingdown the undrainedYoung's modulus
-
throughthe relationship:
(4.2)
The "reference' soil model for these analyseswas a simple homogeneousisotropic linear
elastic MME) undrained model. This was first extendedby considering two casesof
increasingnonhomogeneity, the degree of which was characterizedby X (Fig. 4.4):
X= (EH-Eo)/EH (4.3)
whereEOandEHarethe Young's moduli at groundsurfaceandformationlevelrespectively.

This givesa non-dimensionalparametersimilar to that devisedby Butler (1975), andis more
generalthanthe rate of changeof E with depth(dE/dz)usedby Gibson(1974).
Secondly,two degreesof anisotropywere considered(independentlyto nonhomogeneity),

by the two ratios m andn where:
characterized
Gvh/ Ev
(4.4)
Eh/E,
Gibson(1974)showedthat the undrainedelasticbehaviourof cross-anisotropicsoil could be

definedbyjust threeindependentparametersF, EhandGh, togetherwith the relationships:
vhh = I-n/2
(4.5)
vh, = n/2
The maximumpermissiblevaluefor n is 4, at which point vhhis equalto I- 4/2 = -1. This

definesa materialwhich is incapableof any sheardeformationin the horizontalplanebecause:
Ghh = Eh / 2(l +Vhh) (4.6)
A valueof n=4 in the undrainedcaseimpliesthat the materialcanneitherdeformnor change

in volume.
4-7
For theundrainedanalyses,n valuesof 2 and3 were selected,whereasfor the drainedcase
ratiosof 2 and4 wereused(further discussioncanbe found in Section5.3.1).
Thefinal "real soil" featurewhich was includedin the analyseswas that of plasticyielding. In
CRISP,perfectly-plasticyieldingcanbe modelledalongwith nonhomogeneity but not with
anisotropy.For the undrained,analysesit was decidedto generatea further four setsof
throughhavingtwo diffierentratios of Edc. (thoughtto bracketthe likely rangein
parameters
practice)andtwo differentdegreesof nonhomogeneity X (the sametwo usedin the My
elasticanalyses).Note that it is generallynot possibleto havean homogenousprofile of shear
strengthwith depthunlessK, is setto unity. Tresca'syield function maybe written as:
(ah- a,) - 2q, (4.7)
To remain in the elastic range requires that f<0, where:
Kt - 1) - 2c.,,<0
*. > 1/2(Kt - 1) cr, (4.8)

.
whereK,,t is the coefficientof earthpressureat rest in termsof total stresses= 1/2(1+K.

). If
for exampleKO=2 andy= 20 kN/m3(asin the presentanalyses)thenKt = 1.5 and:
> 1/2(l. 5 - 1) 20 z=5z
The profile of undrainedshearstrengthwould haveto exceed5z' wherez is the depthbelow

groundsurface,in orderfor the initial stateof stressto be within the yield envelope.
Otherwise,someor all of the groundwould be in a stateof plasticyield beforeany excavation
commenced.Clearlyit would be unrealisticto achievethis with a uniform (homogeneous)
Profile of undrainedstrength,asit would haveto be setto exceedthe highestvalueof 5z
encounteredin the mesh(i.e. at the base). The specialcaseof K. =I (= Kt) leadsto the
simplerequirementthat cu>0 everywherein the mesh,which could be satisfiedby anynon-
zero profile. The c. profile implied by "= 1000 when E,, = 5z Wa is indeed c. = 5z kPa.
It was decided to include this for illustrative purposes being on the verge of plastic yield at
-
4-8
the in-situstageeverywherein the meshwould makethe soil-wall systemvery sensitiveto
excavationinducedload changes.The caseof E,,/c,,= 500 impliesfar lessimminentyield.
For the drainedelastic-perfectlyplasticanalyses, by a singledrained

yieldingwas characterized
angleof friction ý' (typical of a critical stateor constantvolumecondition)togetherwith zero
cohesion.No particulareffort was madeto obtainan exactcorrespondence betweenthe
drainedandthe undrainedyield surfacesin the presentanalyses,but this is relatively
straightforwardandis exploredfinther in ChapterS. A ffirther two casesin total were thus
generated,by usinga fixed ý' valuewith two degreesof nonhomogeneity.To ensurethat in-
situ yieldingwasavoidedfor the drainedelastic-perfectlyplasticcase,the coefficientof earth
pressureat resthadto be lower than the passiveearthpressurecoefficienti.e. K, < Kp =
(I+sinfl/(I-siný'). For ý'= 25", Kp =2.63 which is greaterthanthe valueof K. used(= 2).
Soil propertiesfor the linearelasticanalyses(undrainedanddrained)are summarizedin Tables

4.3 to 4.6, andfor the linearelastic-perfectlyplasticanalysesin Tables4.7 to 4.8. The effects
of soil nonhomogeneity, anisotropyand yield are exploredfurther in Chapter 5; the models
andparametersselectedherewere consideredsufficientfor the meshstudiesdescribedin this
chapter.
Structural materialproperties
The stiffhessof the wall was setto that of reinforcedconcretewith a Young's modulusof 20
GPaanda Poisson'sratio of 0.15. In practice,modulusvaluesfor reinforcedconcreterange
between15 - 25 GPa;the valueselectedherefalls in the middleof that range. No anisotropy
of wall stiffnesswas incorporated;the possibilitiesof creep,cracking,or elasticnon-linearity
were ignored. This follows standardpracticein commercialretainingwall analysis.
The other structuralmaterialin the analyseswasthat usedto form the idealizedprops.

Material propertiestypical of steelwereused,with a Young's modulusof 200 GPaanda
Poisson'sratio of 0.20.
In-situ stress state
Having fixed the unit weight of soil at 20 Mmý, the coefficientof earthpressureat rest K. at
2, andthe groundwater levelat the groundsurface,the definitionof in-situ stresseswas
straightforward.For non-coupledanalyses,CRISPprovidesfor the input of eithereffective
4-9
stresseswith porewater pressures,or total stresses.The former approachwas adoptedhere
in defininga linearvariationof stressandpore pressurewith depth:
Y, h, loz
G'V yz-u= (20 - 10) z
(4.9)
a'h K, cr', = 2x 10z
'Cvh 0
For the drainedanalyses,the water tablewas droppedto the baseof the mesh. If it hadbeen
left at the top (coincidentwith ground surface),problemswould havearisenwith the pore
pressureconditions,post-excavation.Excavationsin groundwith a high water levelwill
normallybe madesuchthat the water level insideis kept drawndown to (or below) formation
level. Steadystateseepagewill be setup if this drainageis maintained(e.g. to a gravelblanket
undera road carriageway);this could be handledin a CRISPanalysisby specifyingappropriate
excessheadboundaryconditionsalongthe formationlevel,but not in an uncoupled,drained
analysis.In the undrainedcase,the positionof the water level is irrelevantaseverythingis
handledin total stressterms. It was placedat the groundsurfaceto simplifythe in-situ stress
conditions;in clay soils,water levelsarenormallyquite high, so this was considered
reasonable.
If the water tablehadbeenkept at groundlevel for the drainedanalyses,negativeeffective

stresseswould haveresultedat formationlevel. This is becausetotal vertical stresswould
drop to zero (no overburden)but pore water pressurewould still be at y,. H andhencea',, =
H.
-^t,. No problem would have arisenin the linear elastic but
analyses, in the elastic-perfectly
plasticanalyses,this tendencytoward stronglynegativea'. would haveproducedyield very
earlyin the excavationprocess. The movementof the wall andthe surroundingsoil (especially
on the passiveside)would be very great asa result. This situationcould neveractuallyexist
in practice the excavationwould eitherflood, be filled with a water-tight structure(e.g. deep
-
basement),or be drainedto formationlevel.
To try to maintainsomecompatibilitywith the undrained(total stress)analyses,the samein-

situ total horizontalpressureon the wall was applied. This was achievedthroughmatching
the coefficientof total earthpressureKt (= ah/cr,). As the primarypurposeof thesedrained
analyseswas to seeif the influenceof boundarylocationswas asimportantasfor undrained
4-10
was consideredacceptable.Trendsratherthan
this approximatecorrespondence
analyses,
absolutevalueswerebeingexaminedin thesenumericalstudies.
Initial boundaryconditions
The selectionandspecificationof displacement

constraintson the remoteboundariesof a finite
elementmeshis rarelyconsidereda problemfor the retainingwall analyst. The argumentis
normallythat, if the boundarieshavebeenlocatedsufficientlyfar awayfrom the region of
interest,thenit shouldnot matterwhetherthey arerough or smooth. But it is exactlythis sort
of assumptionwhich needsto be questionedandinvestigated.
Therewill normallybe four boundariesto the mesh,top, bottom, andthe two sides. It is only
the latter threewhich requiredisplacement
fixities to be specified,asthe top is usuallya free
surface.Many retainingwall problemsare symmetricaboutthe centrelineof the excavation,in
which caseoneof the verticalboundariescanbe fixed againsthorizontalmovement,whilst
beingfree to movevertically. However,this still leavesthe other two boundaries;at the base
andat the remoteverticalboundary. Obviouslytheremustbe fixity againstdisplacementin
directionsnormalto the two boundariesconcerned,but this leavesa choiceof full restraint
(rough) or freedomof movement(smooth)parallelto the boundary. The choicesare
illustratedin Fig. 4.5.
Very few designengineerswill checkboth possibilitiesfor thesetwo boundaries;they will

normallyadoptwhateveris their personalpreference.For the base,it could be arguedthat a
fully roughboundaryis physicallymore reasonable; the alternativeis effectivelysuggesting
that, at someparticulardepthin the ground,thereis a rigid layerwith a frictionlessupper
surfacesupportingthe soil massaboveit. A similarimplicationis madeof a ffictionless
discontinuityat the positionof the remoteverticalboundary. However,a roughvertical
boundaryhasa morenoticeableeffect on the analysis,in that it imposesa fixed point on the
ground surfacesettlementprofile. The generalflow of soil tendsto be rotational- from the
retainedside,underthe wall, andinto the excavation.The boundaryconditionswill naturally
havea major influenceon this.
The onlyjustification for smoothboundarieswhich the writer cansuggestis that they may
somehowcompensatefor the over-stiff behaviourseenin the FE method,becausea
continuumis beingmodelledwith finite degreesof freedom. In other words it is on a similar
4-11
levelto the argumentsput forward for reducedintegrationof elements.Unlessthe effectsof
usingoneor otherboundaryconditionare examinedwith referenceto a real problem,the
questionmarkremains.
For the mainbody of boundarylocationanalysespresentedin this section,fully rough

conditionshavebeenimposed(Fig. 4.5a). This choicewas madeon the basisthat it is what
mostanalystsappearto do; however,it was subsequently
examinedby re-runninga numberof
caseswith fully smoothboundaryconditions(seeSection4.4).
Wall installation,propping and excavation
Theseanalyses(of boundarylocation)were not particularlyconcernedwith wall installation

effectsper se, sothe decisionwastakento installthe wall in the simplestpossibleway. The
elementscomprisingthe wall were presentin the meshat the in-situ stage,havingthe same
unit weight (20 kN/m) asthe soil, so asnot to introduceanyunnecessary equilibriumerrors.
The unproppedfree cantileverwas straightforward;the wall was alreadyin place,andthe only

remainingtaskwas to excavatethe soil.
If a singletop level prop was used,it was alsopresentat the in-situ stage,fixed against
horizontalmovement.In practice,top level propsmaybe installedafter the first 1-2mof bulk
excavationhavebeencarriedout to exposethe top of the wall. Alternatively,narrow trenches
maybe dug betweenoppositewall panelsin the positionof the propsto allow themto be
installedbeforebulk excavationcommences.This is usuallypreferableasit minimizesboth
wall deformationsandmovementsof the surroundingground. The strategyadoptedin the
analysesrepresentsthe earliestpossibleproppingandthusputs a lower boundon wall
MOvements.
Singleproppingat bottom (formation)level is slightlymore difficult. In an undrainedanalysis,

to excavateto formationlevel andthen installthe prop would be pointless.No further
movementof the groundwould occur after final dig levelwasreached,so the prop would
makeno contributionto the wall-soil systemwhatsoever.It would only be if subsequentpore
water pressureequalizationwas permitted(in a coupledanalysis)that the formationlevel prop
would actuallytake up load. Thus,it was decidedagainto includethe bottom level prop as
4-12
beingpresentat in-situ stage. Subsequent
wall movementdueto excavationwould be resisted
by the proppingactionat formationlevelevenbeforeexcavationactuaUyreachedthat level.
It is Pertinentto enquirewhetheror not this form of proppingcanbejustified. The main area

of interesthereis how sensitivekey resultsfrom the analysesmight be to boundarylocation.
A wall whichis proppedin sucha way that it canonly rotate aboutits n-dd-pointwill behavein
quite a dilTerentfashionto onewhich is proppedat the crest. So the inclusionof this type of
Proppingarrangementcanbejustified on groundsof illustrativebehaviour.Also, it is
perfectlypossibleto constructformationlevel propsin-situ prior to excavation,throughthe
processofjet grouting (e.g. NewmanandCovil, 1993). Not withstandingthis, it is recognized
that this type of proppingis unusual,andif any startlingtrendswere to emergefrom the
analysesinvolved,they would haveto be treatedwith caution.
The doubly-proppedcasecombinesthe effects(andthe arguments)of the previoustwo

proppingarrangements.The mainpurposefor includingthis casewasto look at a rathermore
heavilyproppedarrangementwherethe wall could no longerrotate abouteithercrestor
formation level. It hasalreadybeenacceptedthat the formationlevel prop, installedat in-situ
stage,is atypical. Nonetheless,
it wasfelt that usefullessonsmight emerge.
Finally, the alternately-propped

casewas considered.The top prop was installedat in-situ
stageso that it offeredvery efficientearlyproppingto the wall, but the bottom prop was not
installeduntil after final dig levelwas reached.Following this, the top levelprop was removed
so that the wall was freeto moveforward at the crestwhilst beingrestrainedat formation
level. This is almostpreciselythe procedurefollowed in practicefor cantileverwalls singly
Proppedat formationlevel,whereoneor more levelsof temporaryproppingare installedearly
on to reducemovements.Theyareremovedlater on whenthe permanentformationlevel prop
is in place.
4.2.3 Evaluation of results
An FE analysisproducesa greatdealof information;the primaryquantitiesof nodal

displacements,
reactions,excessheads(coupledanalysesonly), etc.; the secondaryquantities
suchas stresses(Cartesianandinvariant),pore water pressures,strains,etc.; andthe tertiary
quantitiesof shearforce,bendingmomentetc. in structuralmembers.A retainingwall
designermaybe interestedin someor all of the following:
4-13
a) wall horizontaldeformations
bendingmoments
lateralpressurefrom soil
shearforces
shearstressesfrom soil
b) excavation verticalmovements(heave)of excavatedsurface
C) groundsurface verticalmovements(settlementand heave)alongcrest

horizontalmovementsalongcrest
d) SOU of
profiles movement on vertical or horizontalsections
profilesof stressor pore pressureon verticalor horizontal
sections
contoursof pore pressures,stresses,displacements, etc.
zonesof plasticyield
e) props axialforces
bendingmoments(if fixed)
0 anchors tendonloads
In order to examinethe influenceof boundarylocation,the analyseswerecharacterizedby the

following key results(thoughtto be thosemostfrequentlysoughtin practice):
wall horizontal displacements(5)

wall bending moments M
0 excavationheaves(V)
0 ground surfacemovements(S)
Each of the abovecomprisesa profile of individualdatapoints,andcould obviouslybe

handledin a numberof differentways;two were selectedhere.
The first was basedon a considerationof the whole profile, in order to evaluatethe similarities
anddifferencesbetweenthe individualprofilesgeneratedfor eachseparatemesh. The
comparisonwas necessarilyqualitativeandinvolvedthe assessment of the levelof "scatter"
betweenthe differentprofiles. All the profilesfor a givenproppingcaseandsoil parameterset
were superimposedon a singleplot for visualinspection.The valuesalonganygivenprofile
were denotedIF,with thosealongthe "true' profile denotedP. In this contextthe "true"
4-14
Profilewastakenasbeingthat obtainedfor the biggestmesh(xlOyIO; seeFig. 4.1). This
assumption was questionedandtestedseparately(see4.2.5 below). Furtherplots generated
from this datawere profile difference(r-l"*) andprofile ratio (r/T*), in
an attemptto
highlighttrendsin the data. The conceptof scatterdefinedabove,though
not rigorous,was
consideredto provide a usefulinsightinto the analysisresults.
The secondway of handlingresultsinvolvedcharacterizinga given profile by its maximum

valueIF. andto comparethis with a referencevalue. Typically this referencevaluewas the
maximumvalueof the "tfue" profile r*.. The ratio andthe mannerof its
on a value of unity wasusedasa measureof the sensitivityto (or influenceof)
convergence
meshboundarylocation on the key valueconcerned.The X/H (= Y/W = R) ratio at which
r. T*. reached0.95 (95% of unity) and0.99 (99% of unity) was identifiedfor eachplot as
x,
a meansof quantifyingthis sensitivity. The two ratioswere denotedR95andR99respectively.
Another item of informationextractedfrom the graphswas the directionfrom which the curve
of IF..,,/]F*max
was approachingunity. Mostly this was found to be from below, indicatingthat
the key resultwas underestimatedif the meshwastoo small(i.e. boundariestoo close)and
was designateda negative(-) error. Sometimes,the reversetrend was observedwith
rn=AF*. approachingunity from above(key resultoverestimated),andthis was designateda
positive(+) error. R95andR99could be regardedasindicatorsof havingreached±5% and
±1% of unity, respectively.
Onefurther point which shouldbe mentionedaboutthe profilesis the way in which r. was
identified. Wall displacement,
5 wasvirtually alwaystowardsthe excavation,andlocatingthe
maximumvalueon a given profile 5. was straightforward. (Someexceptionsto this
occurredwith the wall proppedat formationlevel,wheresometimeseitherthe upperor lower
part of the wall rotated backwardinto the retainedsoil - this is discussedlater). Similarly,
excavationheavewas alwaysupwardsso that the maximumpoint on the profile V. was easy
to locate.
However,the bendingmomentdiagramfor a cantileverwall (proppedor free) will have

positiveandnegativevalues,dependingon how the wall bendstowards(or awayfrom) the
excavation.Furthermore,it is not alwayscertainapriori which will be of greaterabsolute
magnitude;the maximum+ve moment or the maximum-ve moment(-M. ). The
designerwill normallybe concernedto know only the magnitudeof the numericallylarger
4-15
maximummoment,asthe concretewall will normallybe reinforcedsymmetricallyto give
equalmomentof resistancein either"hogging"or "sagging". In the analyseshere,the results
processingprocedurehasidentifiedboth +M.,,,,and-M. and examinedthe influenceof mesh
sizeon the largerof the two. (Normallyit was quiteobviousthat onewas significantlylarger
thanthe otherandwould tend to governthe design;only occasionallywere+M.. and-M..
foundto be of roughly equalmagnitude.
)
Profilesof groundsurfacemovementwill, in general,exhibitboth downwardsettlement(-S)

andupwardheave(+S), relativeto originalgroundlevel(+/- follows the conventionusedby
CRISPfor +ve Y upwards). As both directionsof movementmight be of interestto the
designer,the effectof boundarylocationon both and+S. hasbeensummarized.
-S.
4.2.4 Results of Analyses

.
The key resultsarepresentedin Tables4.9 to 4.12, showingR95,R99anderror direction(+/-)
for eachproppingcaseandfor eachparameterset. In Table4.13 an attempthasbeenmadeto
summarizethe mainpoints througha fairly crudestatisticalanalysis.In the following sections,
somebrief commentswill be madeon eachof the individualcases/sets.Theseconcentrateon
the sensitivity of key results to the location of the mesh boundaries, quantifying this by
convergence and scatter (as defined above). Some selectedfull-profile plots are given at the
end of this chapter, whereas examplesof typical convergenceplots are included in Appendix B
(Figs BI
-B4).
4.2.4.1 Wall horizontal displacements
Before examiningindividualcasesin more detail,it might be helpfulto take a brief overview.

From Table4.13, it canbe seenthat horizontalwall displacements have,on average,managed
to reach95% (or ±5%) of the assumedtrue valueby R (= X/H = Y/W); 4-,
51/2.In otherwords,
a modestsizedmesh(in termsof the distanceto remoteboundaries)would be ableto deliver
resultswhich havea relativeerror of 5% or less- for all likely proppingcasesandsoil stiffness
and strengthdistributions. However,in somecasesthe valuerequiredto achieve95% of the
assumedtrue valuewasR= 91/2,andthe R valuerepresentedby the mean+Ix standard
deviation(mean+Is. d.) almostreached8. A representativesampleof full-profile plots are
givenin Fig. 4.6(a-h).
4-16
Undrainedanaiyses
For the unpropped case,convergence of was very reasonable,achieving ±5% at R=

6. Convergenceimproved and scatter reduced with nonhomogeneity, indicating that the need
for boundariesto be at a distance becameless important. However, the opposite occurred
with increasingdegree of anisotropy, behaviour deteriorated noticeably, with poorer

convergenceand greater scatter. The introduction of yield similarly made convergenceand
the scatter between the 5 profiles worse, with the effects of plasticity being more obvious with
the larger meshes(yield seemedto be suppressedif the mesh was too small). This was most
unexpected,becauseyield is normally consideredto be a rather localized event - it seems
counter-intuitive that the location of distant boundaries should have a bigger effect on
retaining wall behaviour when plastic yield is permitted. Within the elastic-perfectly plastic
analyses,increasednonhomogeneity made a small improvement to convergence.
As an aside,it was quite noticeable(especiallyfor the HILE case)that the 8 profileshada

very similarcurvature,which suggeststhat wall bendingmomentsshouldbe relatively
insensitiveto R (andthis indeedprovedto be the case see4.2.4.2).
-
Unlike the free cantilever,the top-proppedwall was constrainedto havethe same(zero)

displacementat (at least)onepoint on the 5 profile, regardlessof meshsize;i.e. at the top.
This seemedto improveconvergenceoverallwith respectto the unproppedcase,ascanbe
seenfrom Table4.9. The samegeneralpointscanbe madeaboutthe influenceof soil
properties,in that convergenceof B..,,/5*. andscatterbetween5 profilesimprovedwith
nonhomogeneity, but deterioratedwith the introductionof both anisotropyandyield.
However,in contrastto the unproppedcase,convergencedeterioratedwhennonhomogeneity
was introducedat the sametime asplasticyield wasbeingpermitted.
The bottom-proppedwall producedsomethingratherdifferentwhenE, = E. +mz andEh/Ev:5

2, in that horizontaldisplacements
at the top of the wall followed the unexpectedtrend of
beingtoo largewhenR was low. However5..., was alwaysat the toe and convergencewas
basedon this, so apparentlythe directionof the error was from below. Otherwise,the scatter
of 8 profilesreducedboth with nonhomogeneity andanisotropy. (It was noted earlierthat the
methodof installingpropsadoptedhereis perhapsunrealisticfor the bottom-proppedcase,so
it is importantto be wary of anomalousbehaviour.
) The maineffect of yield was to makethe
4-17
scatterbetween5 profilesworse below formationlevel(especiallyfor Eý = mz), but this made
little differenceto convergenceas5.., was at the crest.
The heavierproppingpresentin the doubly-proppedwall appearedto imposebetteragreement

betweenthe various5 profiles (asmight be expected),with displacements
betweenthe props
virtuallyunaffectedby boundarylocation. WhenE,,= E,,,displacementbetweenprops
followed the unexpectedtrend of beingtoo largewhenR waslow. (This could, however,be
explainedby the suppressed
movementof the toe causedby smallermeshes- this wall is
similarto a simply-supportedbeamwith oneof the supportsnearmid-span,andreduced
deflectionof the cantileveredendallowsmoremovementbetweenthe supports). With
increasingnonhomogeneity, convergenceof (whichwas at the toe) improvedand
scatterof & profilesreduced. Increasinganisotropyproducedsmallimprovementsin both
convergenceandscatter,unlike earlierproppingcases.Yielding madeconvergencemuch
worse for the caseof E,,= niz, but this wasbecause5. was at the toe andmovementsbelow
formationlevelare muchmore sensitiveto yield andthereforeshow greaterscatter.
Finally, the altemately-propped wall led to similar conclusions as for the free cantilever.
Convergence improved with nonhomogeneitybut deteriorated with anisotropy; scatter of 6
profiles reduced (especially at the toe) both with nonhomogeneity and anisotropy. The
introduction of yield generally made convergenceand scatter worse.
Drainedimalyses
When the wall wasunproppedandthe soil washomogeneous isotropic linearelastic(HELE),
convergencewas ratherpoor and8,,. reducedwith increasingsizeof mesh,so that
convergencewas from above. Nonhomogeneityhada dramaticimprovementon convergence,
with ±1% reachedat aslittle asR=4. Anisotropyhadquite the oppositeeffect andmade
convergenceevenworse,removinganyasymptosy.The scatterbetween8 profileswas
reducedby nonhomogeneity,
but largelyunaffectedby anisotropy. Only oneelastic-perfectly
plasticcasewas run (E' = 40+4z) owing to stabilityproblemswith a cantileverwall of this
D/H ratio. The wall exhibitedvery little curvature,asit seemedto haveundergonerigid-body
rotation for the most part. Convergencewasrathererraticandwas from above(asin all
drainedanalyses).
4-18
19P-P-ropp-ing
producedgenerallybetterconvergence thanin the unproppedcase,which
improvedwith nonhomogeneity but was largelyunaffectedby anisotropy(similarlyfor
scatter). Whenyield waspermitted,the directionof convergencewas mostlyfrom below (i. e.
negativeerror),just rising above =I at R=8. Yielding madeconvergenceand
scattervery muchworseregardlessof the degreeof nonhomogeneity.
Scatteraboveformationlevel for the bottom-proppedcasewas much improved

with
but only slightlyimprovedby anisotropy(scatterbelow formationlevelwas
nonhomogeneity,
alreadyreasonable).Convergence,which was alwaysfrom above = 1, was generally
improvedby nonhomogeneity.Yielding madeconvergence and scattervery muchworse,
mainlybecauseit pushed5. below formationlevelwherebehaviourwasmore sensitiveto
boundaryproximity.
The behaviourof the doubly-proppedcasewas the sameasthe bottom-propped.

Nonhomogeneityproducedmore rapid convergencefor the alternately-Propped
wall, but
anisotropyhadno discernibleeffect. Scatterwasmuchreducedwith nonhomogeneityand
anisotropy(especiallythe former). Similareffectsarosefrom the introductionof soil yielding
as occurredwith bottom-propping.
4.2.4.2 Wall bending moments
Wall bendingmomentswerethe leastsensitiveto meshboundarylocation of all the key results

examined.The meanvalueof R(= XJH = Y/W) by which the analysesachieved95% (or
:1=5%)of the "true" valuewas approximately4 (Table4.13). This would indicatethat, in a
situationwherethe designerwas mainlyconcemedaboutthe structuralcapacityof the wall,
usefulinfonnation couldbe obtainedfrom evena fairly smallmesh. (However,it could be
arguedthat FE analysismight not be necessary,
asWirikler springmodelsarethought to give
adequateindicationsof wall bendingmomentin manycases. ) Somerepresentativefull-profile
plots are given in Fig. 4.7(a-h).
Undrained analyses
Convergenceand scatterfor the unproppedwall was good on the whole, but (in contrastto
observationson wall movements)deterioratedbadlyfor E,,= mz (X = 1). The scatterbetween
bendingmomentprofilesaboveformationlevelreducedwith nonhomogeneity; anisotropy
4-19
madeconvergence andscattersignificantlyworse. Whenthe soil was allowedto yield, scatter
reduceda little andtherewasno changeto convergence for Eu = E. +mz. However(againin
contrastto wall displacements),
yield caused
a greatimprovement
in convergencefor E= mz.
The introductionof top-proppin broughtabouta generalimprovementin convergenceand

reductionin scatter,anddiminishedthe effectwhich nonhomogeneity
hadon thesetwo
andscatterto worsen- asdid plastic
aspectsof the analysis.Anisotropycausedconvergence
yielding(especiallyfor E. = mz)
Changingto bottom-proppingcausedfurthergeneralimprovement.Therewas relativelylittle

scatter,andthe resultsindicatedonly a mild improvementdueto increasingnonhomogeneity.
Anisotropy,unusually,broughtaboutan improvementin convergenceandscatter. Yield
causeda deteriorationwhenEu= mz, althoughtherewas no changefor Eu= E,,+mz.
For the doubly-proppedwall. convergenceandscatterwereboth very good for all E profiles

or anisotropy.However,convergenceand
and were virtually unaffectedby nonhomogeneity
scatterbecameworsewhenyield was permitted,althoughonly significantlywhenE,,= mz.
Convergence and scatter were barely acceptablefor the altemately:propped wall, but
deteriorated badly for E,, = mz. Increasing the degree of anisotropy made convergence
significantly worse, but scatter only slightly worse. However, both aspectswere virtually
unaffected by yield for Eu = E,,+rn7, whereas convergenceactually improved for Eu = mz.
Drainedanalyses
Both convergenceandscatterfor the unproppedcantileverwere excellentfor isotropiccases,

andwere unaffectedby nonhomogeneity.However,convergenceandscatterboth
deterioratedwith increasinganisotropy. The marginalstabilityof the wall resultedin
convergencebehaviourbeingpoor andrather erratic,with scatterbetweenthe bending
momentM profilesup to asmuchas300 kN-m.
Convergencewas reasonableandalwaysfrom aboveNI., JM*,,. = unity for top-12ropped

walls, so M. was actuallytoo big for smallermeshes.Convergence was improved,and
scattereliminated,by increasingnonhomogeneity and (surprisingly)
anisotropy. The inclusion
with the M profilesactually
of plasticyield broughtabouta worseningin convergence,
4-20
deterioratingfor largermeshes.ScatterbetweenM profilesincreasedby to 200 kN-m in
up
the elasto-plasticanalyses.
For the bottom:proppedwalls, both convergence

(whichwas alwaysfrom above,andrather
poor for HME) andscatterimprovedvery significantlywith increasednonhomogeneity,
but
only mildly so with anisotropy. Convergencebecamea little erratic asyieldingwas
introduced,but generallyno worsethan for the elasticcases.Scatterwas only really
unacceptable belowformationlevel.
Doubbý:ýeqd wallsexperienced a slightimprovement in convergence

andscatterasthe
degreeof nonhomogeneityandof anisotropy increased.
However,convergencewasmuch
worseandfroma differentdirection(below)for E' = EO+mz; convergence
wasacceptable
andfrombelowwhenE'= mz. Scatterbetween M profileswasasmuchas400kN-min the
elastic-perfectly
plasticcases.
Finally, for the caseof alternate-propping, (which

convergence was generallyexcellentand
alwaysfrom below) andscatterimprovedslightlywith nonhomogeneity,
but became
significantlyworsewith anisotropy. The only noticeableeffectbrought aboutby the
introduction of yieldingwas slightlyincreasedscatter.
4.2.4.3 Excavation heaves
Excavationheaveshadthe greatestsensitivityto meshboundarylocationof all the key results

examined.It would be true to saythat conclusiveconvergenceto a valueof V. /V*. =I
was not achievedin any of the casesexamined.Examinationof Table4.13 showsthe mean
for R95to be in excessof 8, with the rangedefinedby mean±I s-d.spanning7.4 9.2. Even
-
with a meshtakenup to X/H = Y/W = 16 (seeSection4.2.5),the curveof V... N*. showed
no asymptosyto a valueof unity. In view of this, detailedcommentsaboutconvergenceand
scatterare unwarranted.A representativesampleof full-profile plots aregivenin Fig. 4.8(a-
d). Somebrief commentsare asfollows:
Undrainedanalyses
For the first caseof the unproppedfree cantilever,therewas a big scatterbetweenthe heave
Profilesfor the differentmeshes.Convergencewas not at all convincing,andwas only slightly
improvedby nonhomogeneity.Anisotropymadeno discernibledifferenceto convergence.
4-21
Whenyieldwasintroduced,therewas someworseningof convergence,but it was already
poor in thewholly elasticanalyses.
As proppingwasintroduced,the sametrendswere observed.Whetherthe wall wasLgR-,

therewasgenerallya modestimprovementto
bottom- doubly- or alternately-propped,
convergence broughtaboutby increasingnonhomogeneity.Introducinganisotropyhadlittle
effect, whereasplasticyieldinggenerallybroughtabouta deteriorationboth in scatterandin
convergence.
Drainedanalyses
Overall,the drainedanalysesappearedto havesomewhatbetterconvergencecharacteristics

than their undrained.
counterparts.
Consideringthe unproppedwall, convergencewas almostconvincingby E'= mz wherethe

scatterwasreasonablysmall. Anisotropyhasno effect. Convergenceand scatterwere both
quite good for the elastic-perfectlyplasticanalysis.
For the top-propped and bottom-propped cases,convergencewas improved to some extent by
nonhomogeneity, and the scatter was quite reasonableat E' = mz. Anisotropy had no effect.
With the introduction of yield, convergencewas unchangedfor E'= E. +mz (though more
erratic), but rather better for E' = mz.
Similar comments can be made for the doubly: propped and the alternately-propped wall,
except that with introduction of yielding, convergencewas a little better when E' = E. +mz in
the former propping case.
4.2-4.4 Ground surface movements
The profile of ground surfacemovementhasa uniquefeaturerelativeto the other threekey

being the directly
depends
profiles examined;viz., the position of the remoteendof profile
upon the location of the far verticalboundary. This hasa ratherobviouseffect on the shapeof
the settlementprofile, in additionto the changesmadeby the whole meshincreasingin size. It
alsomadethe conceptof "scattee'ratherlessmeaningful,unlesscomparisonwas confinedto a
distanceof 2H (=16m) behindthe wall.
4-22
L ike excavationheave,ground surfacemovementswere shownto havea high sensitivityto
meshboundarylocation. Conclusiveconvergenceof settlement or heave
+Snm/+S*. to a valueof unity was only achievedin a few of the casesexamined.From
Table4.13,R95for settlementis in excessof 8, with the mean±Is. d. ranging6.6 - 10. For
heave,R95for settlementis almost9, but the smallerstandarddeviationleadsto mean±Is. d.
ranging7.7 - 9.8. Onepoint to be bornein mind is that noneof the drainedanalyses
experienced groundsurfacemovementwhich was downwards in absoluteterms (relativeto
the originalundeformedground surface);i.e. therewas no settlement,-S. The unloadingof
the meshdueto the excavationis very large(yH = 160kNW), and causesthe whole areato
heavewhenvolumechangeis allowed(initially, underundrainedconditions,there is only a
distortionof the retainedmass,with someheaveandsomesettlement).Sometypical full-
profile plots aregivenin Fig. 4.9(a-h). Brief commentson the analysisresultsare asfollows.
Undrainedanalyses
For the caseof the unproppedwall, convergencefor +S improvednoticeablywith
nonhomogeneity, for
whereas -S it became much worse. Anisotropy madeno apparent
differenceto convergencefor either or +S. Yield offeredminor improvementin -S
-S
convergencefor E' = mz, andvirtually eliminatedthe scatterbetweenthe S profiles;+S
convergencewas unaffected. Samecommentsapplyto the top-proppedcase.
The bottom- ro ed anddoubly-proppedwalls were slightlydifferent. Interestingly,

convergencefor -S was quite good in the homogenouscases(for any degreeof anisotropyn),
but was from below i.e. settlementincreasedwith meshsize. Convergencefor +S improved
-
noticeably(but for -S becamemuchworse)with increasingnonhomogeneity.As before,
anisotropymadeno differenceto convergence, but plasticyield offeredsignificant
improvement;at leastin the caseof Eu = mz, E. /c. = 1000. +S convergencewas unaffected.
The convergencebehaviourfor the altematelyzproppedwall was the sameasthat for the

i. for
unproppedcantilever, e. +S it improvednoticeablywith nonhomogeneity, but for -S it
becamemuchworse. The effectsof anisotropyandyield upon convergencewere the sameas
for the bottom proppedwall.
4-23
Drained analyses
Whenthe soil was fully drainedandwholly elasticthe convergencebehaviourfor the

Pnar-o-XLed,top-propped bottom- doubly- andalternately-propped walls was virtually
identical,andwas quite similarto the
undrained.Basically,convergencefor +S improved
noticeablywith nonhomogeneity, whereasanisotropymadeno differenceto convergence.
Yield broughtabouta considerable improvementin convergencefor +S for the unProj2Pedand

19"- fo2ppedwalls, madea smallimprovementfor the doubly proppedwIal. andmadeno
L-
differenceto the bottom- andaltemately:proppedwalls. In eachproppingcase,the
scatter
betweenprofilesbecamerathergreat
whenyieldingwaspermitted.
4.2.5 Extension of boundaries
Therewasno absolutecertaintythat meshboundariesat X/H = Y/W = 10were remote

enoughto guaranteethat the key resultsof interest(wall displacements,
bendingmoments,
excavationheaves,andsurfacemovements)hadreacheda steadyvalue. Indeedthe choiceof
meshXI Oy10 asa referencemay appearsomewhatarbitrary,but it mustbe pointedout that
the distanceto remoteboundarieswas greaterthanthe vast majority of embeddedwall meshes
shownin the literature. From someof the graphsof F. AF*. againstR, it was quite obvious
that further increasesin meshsizewould haveno effect on that particularvalue;wall bending
momentsgenerallybeingthe bestexampleof this. For other key values,suchasmaximum
heave,it was equallyclearthat convergencehadnot beenreachedasthe relevantcurve
showedno signof becomingasymptoticto r. /.F*. = 1. It could well havebeenthat,
regardlessof proppingsystemandconstitutiverelationship,someof thesekey resultswere not
going to convergeto the desiredlevelwithin the rangeof meshsizesexamined.
To addressthis problem,a meshwith boundariesat X(H = Y/W = 16 (designatedxI 6y16) was

createdby addingrows andcolumnsof elementsto the baseandremotevertical sideof mesh
xlOylO- The unproppedcantileverwall.was re-runwith isotropichear elasticandelasto-
plasticsoil parameters,both undrainedanddrained,(sets1,2,3 in Table4.3, setsI d, 2d, 3d in
Table4.5, and sets 12 and 13 in Table4.7). Key results(wall displacement,
bendingmoment,
excavationheave,groundsurfacemovement)were examinedto:
verify that the resultswhich had appearedto convergesatisfactorilyby R= 10 had in

fact done so, and
4-24
b) seeif therewas any evidenceof convergencefor thoseresultswhich had obviouslynot
convergedby R= 10
The resultswere examinedby superimposing the two differentcurvesof on one

plot; in the first curve,P wasbasedon meshxI Oyl0, andin the secondit was basedon mesh
x16yl6- It would be immediatelyobviousfrom any discrepancybetweenthe two curves
whetheror not xlOyIO hadbeenan adequatechoice. The measureadoptedto quantifythe
discrepancy
wasto determinewhich valueof hadbeenachievedfor meshxI 6y16
whenR= 10 (at which point the valueof for xlOyIO was alwaysunity). A
summaryof thesevaluesis givenin Table4.14, with brief commentsasfollows:
Undrainedanalyses
As expectedfor the wall displacements,
very little extra improvementwas obtainedby
extendingthe meshboundariesto X/H = Y/W = 16. By R= 10, *,,. usingxI 6y16 as
the referencewas,on average,up to 0.978for all the parametersetsinvestigated- suggesting
that no morethan a 2% efforis involvedin assumingxlOyIO to be a reference.Bending
MQOm-- similarlyshowedno real improvementusingthe bigger mesh,with M.. /M*..
tents
averaging0.979at R= 10.
For excavationheaves,which could not be consideredto havereachedconvergenceby mesh

xIOyIO, the discrepancybetweenthe two Vni.ý*=" curveswas muchmorepronounced.
Using xl6yl6 asthe reference,V,. N*n. was only ashigh as0.89 (on average)by R= 10,
implying an error of 11%in assumingthat xl Oyl0 hadproducedthe "tru&" heave.
Surfacemovementsshowedsignificantdifferencesbetweenthe two convergencecurves,

exceptfor set 1. Usingxl6yl6 asthe reference,+S,=/+S*.. was in the range1.01-1.46with
a meanof 1.22.For -S. /-S*., the correspondingrangewas 0.68-0.83,with a meanof 0.75.
Drained analyses
Extendingthe meshboundariesto X/H = Y/W = 16 confirmedthat adequateconvergenceon
the maximumvalueof wall displacementhadprobablybeenachievedwith meshxI Oy10.
Valuesof B. /8*. comparedwith the xlOyIO referenceaveraged1.037which suggestsan
error of about 4%, althoughexaminationof Table4.14 showsthat the situationis ratherbetter
for nOnhomogeneous soil. The satisfactoryconvergenceof bendingmomentsat xlOyIO was
corroboratedby theseanalyses,with NL..,/M*. averaging1.002.
4-25
Excavationheavesonceagain,showedthat boundariesat Mf = Y/W = 10maynot be
adequate.The discrepancybetweenthe two V.. N*. curvesat R= 10 was suchthat the
xl6yl6 curvehadonly reachedan averageratio of 0.923for the threeE' profilesinvestigated.
Finally, surfacemovements, although showing rather less sensitivity to boundary location than
in the undrainedcase, still produced significant differencesbetween the two convergence
curves. By R= 10, +S. /+S*. was in the range 0.78-0.92, with a mean of 0.863. (There
was no downward movement -S in the drained analyses.)
4.2.6 Summary
It appearsthat the location of meshboundariescanhaveconsiderableeffecton the computed

profiles of excavationheaveandgroundsurfacemovement,evenwith very largemeshes.
Giventhat xI Oy10 might be consideredanunnecessarily
largemeshby manyanalysts,the
predictionof groundbehaviourawayfrom the wall maybe significantlyin error in a lot of
routine analyses.Accuratepredictionof excavationheavemay not be thoughtimportantin
practiceasimmediateheavecanbe compensated by over-dig. However,anyconcreteslab
constructedat formationlevel would be subjectto long-termswellingpressure,andthis is
Proportionalto the amountof heavewhichwould havetakenplacewithout the slabrestraint.
The computedwall displacementsand bending momentsare rather less and

sensitive, in many
instancesacceptableaccuracywould appearto havebeenachievedwith far boundariesat
modestdistance.This may,in part, explainwhy the Winkler springanalyseshaveenjoyed
considerablesuccessin retainingwall design. If full continuumbehaviouris lessimportantin
obtainingacceptableresults,a methodwhich neglectssuchbehaviourmaynot sufferunduly.
FE analysisis often promotedon the basisof its ability to predictmovementsof the adjacent
ground,but the analysesdiscussedhereinhaveshownthat this is far from straightforward.
It shouldbe emphasizedthat this sectionhasconsideredan issuewhich is quite separatefrom

that which normallyoccupiesthe attentionof geotechnicalspecialists;viz. the constitutive
model. The shapeof the heavelsettlementtrough behindthe wall is a notoriousexampleof
how misleading,for example,linearelasticitycanbe (seeSimpson,1981; alsoFig. 8.5). It is
clearthat errors canbe introducedthroughthe choiceof meshalone,beforethe constitutive
modelis evenconsidered.
4-26
However,it is probablythe casethat wherethe maindeficiencyin the meshwas that the
boundariesweretoo close,this would be mitigatedpartiallyby the use of a non-linearelastic
model,. If stifffiessis linked to strainlevel, andstrainlevelsattenuaterapidly with distance
from thewall/excavation,thenit follows that remotesoil will be significantlystiffer andthus
boundarylocationis not so important. Hence,if a meshis initially assessed via linearelastic
analyses,andfoundto be adequateunderthoseconditions,then it is almostcertainto be
adequatefor a non-linearanalysis.
4.3 Mesh Aspect Ratio
All of themeshesconsideredso far in this Chapter(denotedSeriesA) havebeen

equidimensional.This follows from the way in whichthe distancesX andY (Fig. 4.2) have
beenmadenon-dimensional by the excavationwidth andheight,which itself was equi-
dimensional.As both overall width anddepthof meshhavebeenvariedsimultaneously, it has
not beenpossibleto judge their influenceindependently.
Consequently,four additionalseriesof mesheswith differentglobal aspectratios (constantX

with variableY, andvice versa)were generated,designatedSeriesB to E. In brief, SeriesB
involved fixing X at 101Lwith Y/W taking on values2,4,6,8 and 10. SeriesCdidthe
reverse;fixing Y at I OWandvaryingX/H as2,4,6,8 and 10. Both of theseseriesterminate
at the samelargemeshof xlOylO, but via differentroutes. Two further serieswere used;
SeriesD in which Y/W = 2,4,6,8 and 10 (with X fixed at 61T),and SeriesE in which X/H
2,4,6,8 and 10 (withY fixed at 6W). Full detailsaregivenin Figs 4.10 and4.11. Theonly
caseconsideredhasbeenthe free cantilever(unpropped)wall with undrainedisotropiclinear
elasticsoil behaviour(sets 1,2 and3, Table4.3).
For eachanalysis,the horizontalwall displacements,

wall bending heave,
moments,excavation
was based
andground surfacemovementswere usedto characterizethe results. Assessment
on a Comparisonof the rates of convergence of IF.., T*.. visual comparison of the profiles,
,a
and a comparison of actual F*. values. The convergencerates are presented in Tables 4.15-
4.18, with a summary of values in Table 4.19. A few representative full-profile plots are
shown in Figs 4.124.14.
4-27
Series B (Fig. 4.12)
In this series,X was fixed at I OHwhilst Y wasincreasedfrom 2W to I OW. As might be

expected,becausethe final meshin this series(xIOy10)wasthe sameasthat in the reference
series(i.e. SeriesA), therewere no differencesin the variousP. values. However,it does
appearthat the rate of convergencehasimprovedconsistentlyfor all key values(5, KV and
S), andincreasinglyso with a greaterdegreeof nonhomogeneity.This is seenparticularlyin
the caseof wall bendingmoments,whereNI. /M*... is alreadyat 0.95 at the smallestY/W
ratio considered, being
Table4. IS. The directionof error haschangedfor wall displacements,
from aboveratherthanfrom below.
It would seemthat a more efficientuseof elementsresultsfrom havinga globalmeshaspect

ratio (XfY) greaterthanunity. Fi3dngthe far boundaryat X/H = 10 andloweringthe baseof
the meshallowsthe sensitivityto boundarylocationsto be resolvedmore quickly.
SeriesC
In this series,Y was fixed at lOW, andX variedbetween2H to 101-1.Inspectionof whole

profiles, convergence rates,andmaximumvalues indicatedunambiguouslythat SeriesC was
identicalto SeriesA (the referenceseries). This is fascinating,and suggeststhat eventhough
the baseof the meshis quite remoteat Y/W = 10,the soil-wall systemcannotexploit this if
the remoteverticalboundaryis too close. All elementsbelow Y>X are effectivelyredundant.
SeriesD (Fig. 4.13)

This serieskept X fixed at 6H andallowedthe baseof the meshto drop from Y/W =2 to 10.
Convergenceshowedan evenbetterimprovementover SeriesA than was noticedfor Series
B, b-utthe maximaarrivedat were consistentlylower thanthoseobtainedfor SeriesA/B/C.
Admittedly the relativedifferenceswere not large,but 5% in oneplaceand 7% in another
contributeto a significantmodellingerror. This seriesdemonstratesthat it is possibleto arrive
at a steadyvaluewhich is in fact too low and,therefore,misleading.
Series E (Fig. 4.14)
This serieskept Y fixed at 6W andallowedthe verticalsideof the meshto moveoutward

from X/H =2 to 10. Convergenceshoweda deteriorationover SeriesA whenconsidering
4-28
wall displacements,
but was more or lessunchangedfor the other key values. The maxima
showedvaryingtrends;the V. werehigherthanthoseobtainedfor SeriesA, aswerethe
+M,7,.. However,the V*., and +S*. were an lower. Again the variations
-M*.., -S*.
werenot huge,but were noticeable.Althoughthe conclusionsfrom SeriesC suggestthat
meshdepthis of little useif the width is insufficient,obviouslyadequatedepthis still required.
4.3.2 Summary
From the foregoingobservations,it is clearthat boundarylocation cannotbe consideredin

isolationfrom overallmeshaspectratio. It would appearthat if the remoteverticalboundary
is too close,therearetwo mainconsequences.Firstly, thereis little point in loweringthe base
boundaryasthe "deeper"elementswill be wasted. Secondly,the analystcanbe misledinto
thinking that convergenceon a key resulthasbeenachieved,whenin fact the constraintof one
or otherboundarybeingtoo closehasproduceda falseminimum/maximum.
4.4 Boundary Conditions
In all the analysesreportedso far, the boundaryconditionsfor displacementon the remote

boundarieshavebeenfully rough, asdescribedin 4.2.2 andillustratedin Fig. 4.5(a). The
questionfor the analystis; with theseboundariesat a specifieddistance,doesit actuallymatter
if they arerough or smooth?
To studythis further, the unproppedwall analyses(with undrainedisotropiclinearelasticsoil

behaviour,Table4.3) were re-runwith fully smoothboundaryconditions,Fig. 4.5(b). Only
three of the mesheswereused,namelyx2y2, x6y6 andxI Oy10,to give a representative
bendingmoments,excavationheave,andground
spread. As before,wall displacements,
wasbasedon a visual
surfacemovementswere usedto characterizethe results. Assessment
comparisonof the profiles,and a comparisonof the IF*. values. A summaryof the P.
valuesandpercentagedifferencesbetweenrough andsmoothboundaryconditionsare
presentedin Table4.20. Selectedcomparisonsof key resultsfor x2y2 andxl0yl0 (full-profile
Plo,ts) are shownin Figs4.15-4.16.
4-29
Wall displacements
The introductionof smoothboundaryconditionsin meshx2y2 causedan increasein 5.. for

all degreesof nonhomogeneity,
with the inversely
increase proportionalto X (Fig. 4.15a). In
the worst case(x = 0), a differenceof nearly10%resulted,but this shouldbe takenin context
becausemeshx2y2 is muchsmallerthanthoselikely to be adoptedin a designcontext. As
biggermeshesareused(i.e. R= X/H = YAV increasing),the differencein 5.
successively
causedby changingboundaryconditionsreduced(asexpected,Fig. 4.16a),but therewas still
a 3% differencein the caseof homogeneous
isotropicsoil (x = 0). In fact, for this particular
E. profile, the signof the % differencechanged,indicatingthat S.,. was smallerwith smooth
boundaryconditions.
Wall bendingmoments
Bendingmoments,asseenelsewhere,weregenerallylessaffectedby boundaryconditionsthan
any other key analysisresult. For the smallestmesh(x2y2) removalof boundaryroughness
producedan increasein M. of 8% at worst (X = 0), reducingasX tendedto unity (Fig.
4.115b).For the two biggermeshes(x6y6 andxlOyIO), differencesbecamenegligibleand
were under 1% (Fig. 4.16b).
Excavationheave
Therewas a muchbiggerinfluenceof boundaryconditionson excavationheaves,which is
consistentwith observationsthus far concerninghow other factorsaffectthis particularresult.
The % differencecausedby the introductionof smoothboundarieswas inverselyproportional
both to degreeof nonhomogeneity(X) andsizeof mesh(R), Figs4.15(c) and4.16(c). In the
worst case(X = 0, R= 2), an increaseof over 40% in V. was caused,

reducingto only 3% at
the Otherextreme(X = 1, R= 10).
Groundsurfacemovements
Positive(downward,settlement)movement andnegative(upward,heave)movement+S
-S
havebeendistinguishedin analysingthe results,Figs4.15(d) and4.16(d). For eachmesh
considered,the differencein both -S,,,,.and+S. reducedsteadilywith increasingX, and
actuallychangedsignby X=1. Therewas not sucha clearpatternof behaviour,however,
whenthe threemesheswere compared(R = 2,6 and 10) at a givenparameterset. For sets I
4-30
and3 (X =0 and 1), the differencein-Smaxincreased
with P, whereasfor set2 it reduced.For
+S., though,somenormalityreturnedandthe percentagedifferencewas inversely
proportionalto R for all sets.
A note of cautionshouldbe soundedaboutusingthe summaryfiguresin Table4.20 in

isolation. Whencomparingthe actualsurfacemovementprofilesin their entirety,ratherthan
just consideringmaxima/minima,it becomesapparentthat the major discrepancies occurredat
the remoteverticalboundaryitself Whenfully roughboundaryconditionsare specified,the
ground surfaceprofile is effectively"hungup" on the mostremotenodeandexhibitsa dished
shape.Themajor effect of switchingto smoothboundaryconditionsis to freethis remote
node andallow the whole profile to drop. A plot of S,-S, againstdistancefrom wall (where
the subscriptsY and Y denoterough andsmoothrespectively)showsthis very clearly.
4.4.2 Summary
The purposeof this brief studywas to assess

whetheror not boundaryroughnesscould be
usedto establishsatisfactoryboundarylocation. In this type of boundaryvalueproblem,
where effectsare largelydrivenby selfweightloading,roughboundariesprovidea resistance
to the overallflow patternandthis could be expectedto havea bigger effecton ground
movementsthan on soil-wall interaction.
A fully rough boundary is overly restrictive on ground movement if it is too close to the
wall/excavation. Similarly, a smooth boundary provides too little resistance. The true
continuum does not have rough rigid interfaces at the physical position of the mesh
boundaries; neither does it have frictionless surfaces. However, it seemsreasonable
smooth
that the switching of conditions on the boundariesprovides a qualitative test on the adequacy
of their distance. Certainly it is a straightforward test. The difficulty fies in deciding what
amount of variation in, say, maximum bending moment would be indicative of boundaries
being too close. Boundary proximity affects the different key results by varying degrees,as
has been shown.
4-31
4.5 Number of Elements
After havingfixed the positionof the boundaries,the next questionfacingthe analystis: how
manyelementsof a giventype shouldbe used? A betterway of framingthe questionis not so
muchhowmanyelements,but how manynodesor degreesof freedom(d.o.f ), becausethis is
a morerelevantfigure to work with. For example,a meshof 100four-nodedquadrilaterals
will haveapproximatelythe sametotal numberof d.o.f asonecomprising50 eight-noded
quadrilaterals.All thingsbeingequal,the first meshwould not necessarilybe expectedto give
more accurateresultsthanthe second,eventhoughit hastwice the numberof elements(in
fact, the reversecouldbe true).
It is relevantto considerwhethermesheshavingthe samenumberof d-oT but comprising

differenttypes(orders)of elementsperformdifferently. It haslong beenrecognized
(Zienkiewicz,1977)that the 3-nodedconstantstraintriangle(CST) performsquite poorly in
most situations.The classicexampleoften quotedis the simplecantileverbeamwith a UDL
or point load,for which analyticalsolutionsof displacement,
moment,andsheararereadily
available.Evenwith largenumbersof elements,the solutionis deficient. A meshcomprising
higher ordertrianglesor quadrilaterals,but with the sameoverallnumberof d.o.f. Will give a
better performance.In somecases,higherorder elementsare essential.For example,Sloan
and Randolph(1982)haveshownthat undrainedcollapseanalysesof axisymmetricproblems
may requirevery high order elementsbeforea satisfactoryresult canbe achieved.
The meshesusedso far in this chapterhavebeenbasedon onemastergradingpatternand

elementtype, as seenin meshxlOylO (Fig. 4-1). Smallersub-meshes(x8y8,
x6y6, etc.) have
beencreatedby effectivelydeletingrows andcolumnsof elementsremotefrom the wall as
appropriate. Anotherway in which the sub-meshescould have been derivedwould have been
to scalethe dimensionsX andY accordingly- by 80% to get x8y8, by 60% to get x6y6, etc.
This would havekept the total numberof elements(andhenced.o.f.) constant,whilst simply
varyingtheir size. In the event,the former methodwas adopted,usingwhat wasbelievedto
be a reasonablyfine discretizationof elementswithin andaroundthe areaof the excavation.It
did mean,however,that the numberof elementswas varyingsimultaneouslywith meshsize.
It is generallyaccepted(e.g. Kikuchi, 1986)that there arethreedifferentwaysin which a finite

elementmeshcanbe refined:
4-32
a) ' h-refinement,
b) r-refinement,and
C) p-refinement.
Thesecanbestbe illustratedwith referenceto Fig. 4.17,which showsa simple4-elementmesh

which is thenrefinedin eachof thesethreedifferentways.
h-refinement simplyincreasesthe numberof elements(andhenced.o.f ) in the meshby

incorporatingadditionalgrid lines;the type (order) of elementis unchanged.In this way, the
originalmeshis containedwithin the refinedmesh;a fundamentalrequirementof classical
meshrefinementstudies.
r-rerinement adj ustssome/allof the elementsizes,whilst keepingthe total numberof

elements(andd.o.f ) constant. It is basicallya form of geometricdistortion,in which the mesh
topology is preservedbut the positionsof the internal(andperhapssomeexternal)grid fines
change.
p-refinement increasesthe order of sometall of the elements,whilst keeping the total number
of elements(and their size) constant. The number of d.o.f is increased, but the position of the
grid lines remain unaltered.
In CRISP,the scopefor p-refinementis ratherlimited asit is only the LST which hasa higher-
order relative;namelythe CuST. However,the other two forms of refinementare readily
carriedout, andin the next few sectionsh- andr- refinementwill be usedto try to establishthe
optimumnumber,sizeandtype of elementsfor a retainingwall analysis.
4.5.1 h-refinement : Series I
Before embarkingon a more structuredandfocusedprogrammeof meshrefinementstudies,

two'of the earliermeshes(x2y2 andx6y6) were usedin a very simpleform of h-refinement.
To checkon the adequacyof the original discretizationmeshx2y2bwas created,in which a
uniform Im x Im grid wasusedin eachdirection,giving a total of 650 nodesand 600
elements.The secondmeshx6y6busedaI rn xIm grid within X= 2H and Y= 2Y, anda 2m
2m,grid outsidethis area,giving 1722nodesand 1640elements.Interestingly,meshx2y2b
4-33
hasthe sametotal numberof d.o.f asmeshx8y8, andcouldhavebeenobtainedif x8y8 was
scaleddownto createa smallermeshhavingX/H = Y/W =2 (as describedabove).
The only caseexaminedwas an unproppedwall with undrainedsoil parametersetsI and3,

Table4.3. The key results5.,,, M,, V., S. are summarizedin Table4.21, from which it
is quite clearthat this additionalrefinementof the meshhasbrought no significant
improvement.Virtually all variationsin r. valueswereunder 1% evenfor excavation
-
heaveandgroundsurfacemovements,which havebeenvery sensitiveto the other mesh
variationsexaminedthusfar. This providedconfirmationthat the meshdensityadoptedin the
boundarylocationandroughnessanalyseswas probablyof adequaterefinementfor the
purposeat hand.
But would fewer elementshavesufficed?A designerwould wish to usethe leastnumberof

elementswith asfew incrementsasnecessary to obtainan answerof sufficientaccuracy.
Leavingasidethe subjectivejudgementsinherentin this statement,it is clearthat the number
of elementswill be limited by the availablememoryof the computer,whereasthe numberof
incrementswill be limited by the availabledisk space(asresultsmust be storedfor subsequent
post-processing).Both will affect the time takenfor the analysis,which is not unimportant
evenwith modemprocessorspeeds.
As an aside,one ratherusefulpieceof informationgainedfrom this investigationconcerned

the bendingmomentprofile. In manyanalysesseenby the writer, therehasbeena significant
non-zeromomentat the baseof the wall, which seemsintuitively wrong. From the results
obtainedhere,it would appearto be causedby a changeof gradingwhenthe columnof wall
elementsruns into the underlyingsoil elements,andcanbe avoidedby havingthe first element
below the wall of same(or lower) aspectratio asthe bottom elementin the wall column.
BecauseTable4.21 "homesitf' on maximumvalues,it doesnot actuallyshowthe discrepancy
at the toe of the wall, wheremuchbiggerlocal percentagedifferencesexist. This point is
returnedto in Chapter8 whendiscussingthe calculationof bendingmomentsin greaterdetail.
4.5.2 h-refinement : Series 11
The secondseriesof h-refinementanalysesusedthe samegeometryanddimensionsfor the

wall and excavationas shownin Fig. 4.2. Beginningwith boundariesat X= 8H andY= 8W
togetherwith a very coarsesubdivisionof elements,a seriesof eight finite elementmeshes
4-34
weregenerated, designatedx8y8aOto x8y8hO(with x8y8hObeingthe sameasx8y8 used
previously,andthe suffixed'0' indicatingthis wasthe original series). Eachsuccessive
mesh
was finerthanits predecessor;the maindetailsaresummarizedin Table4.22, andthe meshes
are shownin Fig. 4.18(a)-(h). The patternof refinementservesto concentrateadditionalgrid
linesabouttwo axesrunningthroughthe mesh. The first of theseaxeswasfixed on the
centrelineof the wall, andthe secondat the levelof the excavatedsurface. This choicewas
partly intuitive,andpartly basedon the ideathat thesetwo finesarein someway a focusfor
the mainareasof stressconcentrationin the problem.
All eightmesheswere usedwith a limited selectionof the sameproppingarrangements and

constitutivebehaviourdescribedearlierin this Chapter. Only the unproppedandtop-propped
caseswereconsidered,alongwith parametersets1-5 (linearelasticundrained)setsI d-5d
(linear elasticdrained),and sets12-13(linearelastic-perfectlyplastic,undrained).In all sets,
nonhomogeneity wasvaried in the 0:
range X:5 5 1; in the fully (only)
elasticanalyses
anisotropywasvariedin the range1 :!5 n:5 3.
The purposeof this particularstudywasto establishthe numberof nodes(NN) requiredto

give r., /r*,. = 0.95 and 0.99, denotedNN95andNN99respectively,wherer is a
characteristicvalueor key resultfrom the analysis.For consistencywith previousanalysesin
wall bendingmoment,excavation
this chapter,the key resultsof horizontalwall displacement,
heave,andgroundsurfacemovementswere selected.
After plotting a number of r., /r*,. v NN curves, it becameapparent that this particular
series of refined mesheswas not particularly successful. It was found that all the curves for
V. and S.. converged upon a false "plateaW' by mesh x8y8ý before .1driking9l
upwards (or downwards) to the referencevalue of r. AF*... =I at mesh x8y8 (NN 1899).
This plateau could be ±2% either side of IF., JT*. =I for wall displacementsand moments,
ý-3% side for excavation heaves,but as much as ±1 5% for surface movements. Typical
-either
convergence plots are given in Fig. 4.19(a-d).
Admittedly, most of theseerrorswould not be regardedas severe,but it wasdecidedthat it

would be unnecessary(and quite wrong) to introduceanygratuitouserrorsinto the studies.
After someconsideration,it was concludedthat concentratingadditionalgrid finesalmost
exclusivelyaroundthe excavationarea(asdescribedabove)paid inadequateattentionto the
4-35
rest of the mesh. A lessbiaseddistributionof subdivisionseemedto be necessary,anda
secondseriesof h-refinementmesheswasdevisedto redressthis.
4.5.3 h-refinement : Series IH
Thethird seriesof h-refinementmesheshadexactlythe samestartingpoint asthe second,at

meshx8y8a. Successive refinementgeneratedthe meshesx8y8b to x8y8f, maindetailsare
in Table4.23, andthe meshesshownin Fig. 4.20(a)-(e). This time the patternof
summarized
refinementwasmoreuniform throughoutthe mesh,thoughstill concentratingon obtaininga
fine discretizationin the vicinity of the wall andthe excavation.
The six meshesin this serieswere usedtogetherwith someof the proppingarrangementsand

soil parametersdescribedearlierin the Chapter. Unpropped,top-, andbottom-proppedwalls
were considered,with parametersets1-5 and 12-13for the undrainedanalyses,andsets I d-5d
for the drained.
As with seriesH, the purposeof theseanalyseswasto establishNN95andNN99,the numberof

nodes(NN) requiredto give = 0.95 and0.99 respectively.The key resultsof
horizontalwall displacement,
wall bendingmoment,excavationheave,andground surface
movementswere againusedto evaluatethe results.
Froýi the experienceof SeriesIII, it was decidedto ascertainwhetheror not the form of mesh
refinementbeingusedwas onewhich would convergetowardsthe "right" answer,before
embarkingon a detailedexaminationof the results. Severalr. /]F*.. v NN (convergence)
curveswere plotted, andit was clearthat the falseplateauseenin SeriesII hadbeen
eliminated.Following this all the convergencecurveswere plotted, togetherwith an
associatedplot in which the whole profile in questionwas shown- with superimposed curves
for the coarsest(x8y8a)andfinest(x8y8f) meshesused. This latter type of plot illustratesthe
likely error in computedresultthat will ariseif too coarsea meshis used,andrepresentative
examplesare givenin Figs 4.214.24. Tables4.24 to 4.27 summarizethe NN95andNN99
valuesfor wall displacement,
bendingmoment,excavationheave,andground surface
movements.Examplesof typical convergenceplots aregivenin AppendixB (Fig. B5a-d).
4-36
Horizontalwall displacements
For the undrainedelasticanalyses,rarelyweremorethan550 nodesrequiredto reach95% of

5.., nor morethan 1400nodesto reach99%. The introductionof plasticityincreasedthis
requirement(althoughpresumablyonly in the areasof yield), causingNN95tOrise to 650 and
NN99to 1400. It is evidentfrom Table4.24 that the drainedanalyseshaveratherlower
to reach95% 5., averagingabout400.
requirements
Consideringthe whole profile aswell asthe maximum5 valueitself, it is clearthat the

displacementsfor the coarsestmesharereducedat all pointson the profile relativeto the
finestmesh,so that the directionof convergence
for I is alwaysfrom below (Fig.
4.21).
Wall bendingmoments
In the undrainedelasticanalysesof unproppedandtop-proppedwalls,rarelywere morethan

550 nodesrequiredto reach95% of NL., but this extendedto sz-,
2000to reach99%. This was
mainlybecausethe M. /M*.. curve exhibitedvery rapid convergence
to start with, followed
by a shallowtrough, climbingup to unity at the very end.
The situationwasnot very muchdifferentwhenyieldingwas introduced,althoughtherewas

actuallya smallimprovementin reachingNN95(550 for 95%; 1400for 99%). Switchingfrom
undrainedto drainedloadinghadlittle effecton the requirementsfor numberof nodes.
Bottom propping pushed up NN9s significantly to 900, whilst bringing down NN99to 1200.
However, as noted earlier in this Chapter the form of bottom propping used in these analyses
was probably too rigid, so these anomalousresults (with respect to the other cases)must be
treated with some caution.
The bendingmomentprofile for the coarsermesheswas attemptingto be a crudebest-fit to

the profile for x8y8f, Fig. 4.22. Most of the "tru&" profileshada characteristic'S' shape,and
in somelocationsM was lessthan what it shouldhavebeen,andin others,more. The maxima
were normallyunderestimated, so againconvergence was from below. The bottom-propped
wall, with its pronounced"spike" in the bendingmomentprofile at formationlevel,proved
evenharderto fit satisfactorily,which accountsfor the observedincreasein NN95.
4-37
Excavationheave
Owin,g to the relativelystraightforwardshapeof the typicalheaveprofile, eventhe coarsest

mesheswereableto provide a quite a good fit to the "true" profile. This resultedin relatively
low requirements
for total numberof nodes,with NN95:5300 andNN95:5600 for the
undrainedanalyses,andNN95:5 100 andNN95:5 300 for the drained. Yielding broughtlittle
change,exceptfor the unproppedwall with Eu = mz, but this maybe consideredan extreme
caseowing to its marginalstability.
This relativeinsensitivityof computedexcavationheaveto meshrefinement(seeFig. 4.23)

was in sharpcontrastto the dependence on boundary location shownin previoussections.
Consideringpositive(downward)movementfirst, 500nodesbrought -Sn,, to within 95% of

its true valuein mostundrainedelasticanalyses(800 nodesto reach99%). WhenE. = 0,
theserequirementsincreasedsubstantiallyfor the unproppedwall, but this appearsto be a
specialcase.Plasticyield hadlittle affecton NN95,but madea significantdifferenceto NN99,
pushingit well over 1000.
Moving onto negative(upward)movement,NN95droppedto 100or lessin the majority of

cases,with NN99rarelyrising over 600. This was pushedup significantlyby yield however,
with 800 or morenodesneededto obtain95% +S. - exceptfor the bottom-proppedcase
which requiredsimilarnumbersof nodesto the fully elasticanalyses.
For both upward and downward ground surface movement, a switch to drained loading
brought about a very close matching of profiles regardlessof the refinement of the mesh (Fig.
4.24). As with excavation heaves,the type of mesh used seemedimmaterial, provided the
boundaries were sufficiently remote. In all casesexamined,NN95 and NN99 were both less
than 100.
4.5.4 Summary
This studyhassoughtto determinethe numberof nodesrequiredto give 0.95

and0.99, denotedNN95andNN99respectively.To summarizethe findings:
4-38
for elasticretainingwall analyses:
5.
NNgs 550 550 300 500
NN99 1400 2000 600 800
and for elasto-plastic analyses:
5. M. V. Smax
NN95 650 550 300 700

NN99 1400 1400 800 1300
On this basis,therewould be a high degreeof confidencethat, with a meshpossessing

about
700 nodes,the key resultsfrom a retainingwall analysiswould be within about±5% of the
most accurateanswerthat couldbe expected.In an arbitrarymeshof reasonablesize
comprising8-nodedLSQ elements,the numberof elementsis approximately1/3 the number
of nodes(andnot 1/8 asmight at first be thought;seeFig. 4.25). Therefore700 nodes
equatesto about233 (say250) elements,which is modestby contemporarystandards.
The total numberof nodes(or elements)in a meshis only part of the modellingissue. Of
equalimportanceis wherethe nodesareactuallylocated;i. e. in a meshcomprisinga given
numberof elements,what shouldthe relativesizesof the elementsbe? In the seriesof h-
refinementanalyses just described,it is clearthat a patternof gradinghasbeenadopted
wherebythe smallerelementsare adjacentto the wall/ excavation.However,no attemptwas
madeto carry out this gradingin anythingotherthan a simpleintervalhalving,whena new
grid line was introducedinto the mesh. In the next section,further analysesare considered
to this
which were carriedout examine questionof relative size and grading in more depth.
An overalltrend which is emergingis that profilessuchaswall displacementandbending

moment dependmoreon meshrefinementthan on boundarylocation. However, for profiles
heave
suchas excavation andground surfacemovement, the situationis reversed.
4-39
4.6 Size of Elements
Havingfixed the positionof the boundaries,anddeterminedhow manynodes(elements)are

requiredto obtaina satisfactorysolution,the last major questionrelatingto the geometric
modellingis how shoulda given numberof nodes(elements)be usedto obtainthe best
solutionpossiblefor that numberof nodes(elements).This canbe thoughtof as a questionof
efficientuseof elements.It hasalreadybeennotedfrom the SeriesII h-refinement(see4.5.2)
that it is not enoughsimplyto concentratethe detailaroundthe wall/excavationarea,at the
expenseof the rest of the mesh.
Dependingon the situationat hand,the "bestpossiblesolution" requiredby the designermay

be with respectto oneparticularresult (i.e. wall displacement),
or it maybe for the wider
behaviourof the whole soil-structuresystem. This shouldbe bornein mind when interpreting
studiesreportedhere.
the resultsof the meshsize/density/grading
Investigationsof the relativelocation of nodes(sizesof elements)comesunderthe headingof

r-refinement,Fig. 4.17, andtwo differentserieshavebeeninvestigated.
4.6.1 r-refinement : series I
The first seriesof r-refinementanalysesstartedwith a "supermesh"havingthe samewall and

excavationdimensionsasusedhitherto, andwith boundariesat X= 8H and Y= 8W, Fig.
4.26. The bareminimumof grid linesis alsoshownin Fig. 4.26, with variouskey dimensions
denotedP, Q, R, S andT. An original meshwas createdwhich hadthe sametotal numberof
nodes(elements)asmeshx8y8efrom SeriesIII h-refinement,but with the elementsbeingall
of the samesizein anygivenregion- For example,meshx8y8ehad 14 divisionsbetweenthe
rear of the wall andthe remotevertical boundary. In the startingmeshusedhere,the width of
all the elementsin this regionwould be Q/14 (which is differentto x8y8e,wherethe width of
elementswas varied).
It is necessaryto digressbriefly to describethe way in which non-uniformgradingwas defined

andgenerated.Let ai be the elementsidedimensionalongan intervalof lengthL. Thereare n
elementsidesin total andthe ratio of lengthsof the first elementto the last elementis m, i.e.
a./ai = m. So it for example,therewere five elements,andthe fifth elementwas to be three
4-40
timesaswide asthe first, then n=5 andm=3. A uniformgradingwith all elementsidesthe
samelengthwould be obtainedby settingm=1. The generalexpressionfor ai is:
ai a, ki (4.10)
where
ki m (i-1) / (n-1)
Returning to the present series of meshes,each of the five dimensionsshown in Fig. 4.26 can
be subdivided independently. Starting with x8y8el in which all the m=1, a further seven
mesheswere created, designatedx8y8e2 to x8y8e8, in which the rn were gradually increased.

The two largest dimensions Q and T were both taken up to a maximum length ratio of rn =
32.0, with the other three taken up to rn = 3.0. All the meshesare shown in Fig. 4.27(a)-(h).
In general terms, the r-refinement has been concentratedin towards a vertical axis passing
through the wall centreline, and a horizontal axis coincident with formation level. There are
clearly some similarities here with what was done in the h-refinement mesh series, and for
much the samereasons.
Using theser-refinedmeshes,a further seriesof CRISPanalyseswas conductedwith someof

the proppingarrangementsandconstitutiverelationshipsdescribedearlier. The purposeof
theseanalyseswasto establishthe degreeof gradingrequiredto extractthe bestperformance
out of a givennumberof elements.At the outset,it was recognizedthat an optimumwould
probablyexist. On the onehand,it would generallybe acceptedthat a uniform gradingwould
be highly inefficient,andthe accuracyof the resultsin areasof high stressgradientwould be
impaired. However,at the other extreme,it mayalsobe possibleto havetoo largea ratio of
sizebetweenthe smallestelements(presumablynearthe wall) andthe largestelementsfurther
away,for two reasons.
Firstly, individualelementsof excessiveaspectratio C'IengtW'/"width" in broadterms)may

result, andthis could affectthe accuracyof the analysis(dependingon wherethey are
located). The generalrule is to try to ensurethat elementsareasequidimensional
aspossible,
and severalauthorshavegoneasfar asto suggestEmitsfor aspectratios,internalangles,area
ratio of smallestto largestelementin the mesh,etc. (for example,seeFagan,1992).
4-41
Secondly,aswas seenin the h-refinementstudies,certainaspectsof the analysismaybeginto
deteriorateif insufficientattentionis paidto the meshin areasremotefrom the excavation.
This maybe lessof an issueif the analystis not concernedwith attemptingto predict what will
happenin the groundsomedistanceawayfrom the retainingwall, but it is possiblethat the
qualityof resultsin the areaof interestwill alsobe affected.
Giventhe relativelymuchlargervolumeof the meshinfluencedby the gradingof dimensions

Q andT, it seemedreasonableto interpretthe resultsof the analysesin termsof a singlelength
ratio LR = mQ= mT. To be consistentwith the earlierstudiesdescribedso far in this chapter,
the presentanalyseshavebeenusedto establishLR95andLR99,the lengthratios requiredto
give r. AF*,.,,= 0.95 and0.99 respectively.The parameterr canbe the maximumvalueof
wall displacement,bendingmoment,excavationheave,or ground surfacemovement,as
before. In additionto plotting curvesof againstLP, completeprofiles havealso
beenexamined,superimposing curvesobtainedfor the uniform (LR=I) andmost nonuniform
(LR=32) meshes,to ensurethat misleadingconclusionswere not drawn.
The LR95 and LR99values obtained are summarizedin Tables 4.28 to 4.3 1, and representative
fuU-profile plots are shown in Figs 4.28-4.3 1. Examples of typical convergenceplots are
given in Appendix B (Fig. B6a-d.) Before going on to discussthe different key values in turn,
some general commentswill be made. The results were somewhat surprising in that there
appearedto be very little sensitivity to LR on the whole. In other words, having used a mesh
with a considerablenumber of nodes Oust over 1400) a modest concentration of nodes into
the area of interest was all that was required to optimize the accuracy of the analysis. Upon
reflection, the SeriesIII h-refinement studies indicated that mesh x8y8e would be expected to
have more than sufficient nodes to give answersto within ±1% of the "true" FE results in the
majority of propping casesand soil parameter sets (see Tables 4.24-4.27). Mesh x8y8el (r-
refinement, Fig. 4.27a) was graded a little differently to mesh x8y8e (h-refinement, Fig. 4.20e),
but these differences seemnot to have had a significant effect on the calculated quantities.
Theseobservationswould suggestthat, for a generallysatisfactorymesh,further r-refinement

will do little (if anything)to improveresults. Onepossiblestrategyfor testingmeshadequacy
would, therefore,be to carry out r-refinementandobservethe effects. This hasthe attraction
of convenience;r-refinementonly needsnodal co-ordinatechanges,whereash-refinement
requiresalterationsto nodeandelementnumbering.
442
Horizontal wall displacements
Virtually all of theundrainedanalyseshadachieved±5% of 5*. with the initial uniform

grading(i.e. at LR = 1), the exceptionbeingan unproppedwall with parameterset2
(requiringLR = 2) andunpropped/set 13 (requiringLR = 16). Onecandismissthe latter
owing to its marginalstability. To achieve±1% of 5*,,. neededratherhigherlengthratios,
,Aith LR99averagingabout5 for the unproppedandtop-proppedcases(elasticandelasto-
plasticalike). The drainedanalysesshowedno changebeyondLR = 1.
Consideringthe whole profile, the displacements for the uniformlygradedmeshwere reduced

at all pointson the profile relativeto the mosthighly gradedmesh,so that the directionof
convergencefor =I was alwaysfrom below,Fig. 4.28.
Wall bendingmoments
In the undrainedanalysesof unproppedandtop-proppedwalls,rarelywas it necessaryto

exceedLR =I to reach±5% of M., althoughthe lengthratio requiredto get to within ± I%
variedbetweenI- 18. Therewassomediscernibletrend for nonhomogeneity andanisotropy
both to reduceLR99,andfor yield to increaseit, but beyondthat it was hardto draw any
generalconclusions.As with displacements, drainedloading(i.e. the removalof the virtually
incompressiblecondition)brought both LR95and LR99 down to below unity. Full profiles
were generallyvery similar,exceptwhen significantyieldingwaspermitted,Fig. 4.29(a).
Excavationheave
The simple shapeof the typical heave profile (Fig. 4.30) led to a very good agreementbetween
the profiles at LR =I and LR = 32, showing that length ratio made little difference once an
adequatenumber of nodes had been selected. As can be seenin Table 4.30, all LR95were <1,
and the majority of LR99were also <1. Yielding had the effect of increasing LR99up to 5 and
beyond.
Changinglengthratio hadratherEferent effectson groundsurfacemovements.To achieve

±5% of -S*.,,. or +S*. generallyrequiredno morethanLR9s= 1, exceptwhenE. =0 (with
or without yield) at which point LR95wasintherange2-22. To convergeupon±S*.
substantiallyincreasedthe lengthratio, but therewas still a wide range. In general,sometrend
4-43
couldbe seenwherebynonhomogeneity, anisotropyandyielding all increased
LR95,but little
elsecouldbe discerned.Profile detailnearthe wall wasmost affectedby LR, Fig. 4.31.
4.6.2 r-refinement : series 11
The first seriesof r-refinement mesheswas topologically equivalent to x8y8e (441 elements,
1408 nodes), with the mesh grid fines adjusted to give a range of element side length ratios.
For the secondseriesof r-refined meshes,it was decided to select a coarser mesh from the
third h-refinement series,namely x8y8b (90 elements,299 nodes) and create meshes
topologically equivalent to it, Fig. 4.32. A total of six mesheswas created, designatedx8y8bi
to x8y8b6, along very similar lines to SeriesI, Fig. 4.33. There were some n-dnordifferences;
no r-refinement was attempted along lengths R and S (i. e. both mRand ms = 1.0), as it did not
seemfrom SeriesI that any benefit accrued from having the excavation layers of non-uniform
thickness. Also, there may be an argument in favour of maintaining a uniform length of
elementsmaking up the retaining wall itself (certainly abrupt changesin size should be
avoided). Another difference is that the principal length ratio LR (= MQ= mT) was taken from
I to 32 in slightly fewer steps (LR =12 and 24 were omitted), as it was clear from SeriesI that
as many as eight mesheswas probably not required.
The LR95and LR99values obtained for SeriesII are summarizedin Tables 4.32 to 4.35, and
representative full-profile results plotted in Figs 4.34-4.37; some typical convergence plots are
given in Appendix B (Fig. B7a-d). As a general comment, the length ratios required are
significantly higher than for Series1, by a factor of between 2-3, but solutions that are within
±5% of IF*. can still be obtained with only a modest concentration of nodes into the area of
interest. A much bigger gap now exists between the LR95and LR99values for any given
case/setcombination, indicating that considerable extra refinement has to be performed on the

mesh in order to achieve± I% of the "true" value.
An importantpoint to makehereis that the valueswhich these mesheswill be

convergingupon canno longerbe regardedascloseto the "true' values(i.e. thosewhich
would be obtainedwith a very fine, well-gradedmeshpossessing distantboundaries).The
reasonis that the SeriesHI h-refinementanalysesdemonstratedthat a minimumof 500-600
nodeswere requiredin order to be surethat thosekey resultsexaminedwere within 1:5% of
the bestanswerswhich couldbe obtainedwith boundariesat X/H =Y/W =8. Only 300 nodes
arepresentin eachof the SeriesH r-refinementmeshes.
4-44
However,the valueof theseanalysesis to test the hypothesisthat a meshwith inadequate
degreesof freedomwill be significantlymore sensitiveto grading. It doesnot matterif the
bestsolutionwhich canbe obtainedis, itself, in error if sensitivityto gradingcanbe
-
demonstrated,
thenthis would be sufficientgroundson which to question(andperhapsreject)
the solutionanyway.
Horizontalwall displacements
On average,the undrainedanalysesrequiredthe lengthratio LR to be at least2-3 in order to
reach±5% of V., andabout 10-12to reach±1%. Unpropped,set 13 was againrather
anomalous.If the loadingwas drained,a uniform mesh(LR=I) was sufficientto provide 95%
of 8*., with 99% beingachievedat LR = 3.
Consideringthe whole displacement

profile, therewas a morepronounceddifferencebetween
the uniform (x8y8bI) andmost highly graded(x8y8b6)meshes,thanwas observedfor Series
1,Fig. 4.34. This is confirmationthat thereis greatersensitivityto gradingwhenthe total
numberof degreesof freedomis on the low side.
Wall bendingmoments
The lengthratio requiredto achieve±5% of M*. in the undrainedelasticanalysesvaried
between1-4; which is both a greatervariationanda highermagnitudethanwas requiredfrom
Series1. The LR99valuesaveragedabout7 for the unproppedand 15 for the top-propped
walls. Yield appearedto haveonly a smalleffect on LR95,but almostdoubledLR99.
The drained casewas of particular interest; the requirementsfor ±5% were met without any
refinement (as per Series1), but to achieve±1% M*. required length ratios as high as 15
(unpropped) or 6 (top-propped). Full profile plots, for these relatively coarse meshes,were
rather discontinuous though not too dissimilar (Fig. 4.35).
Excavationheave
The good agreementobservedbetweenthe profilesat LR =I andLR = 32 for SeriesI
disappearedfor SeriesH, Fig. 4.36. It was still possibleto get within ±5% V*.. without
refiningthe meshbeyondthe initial uniform elementsizes(evenwith yield), but to achieve
±1% requiredasmuchasLR= 10 (unpropped)orLR= 5 (top-propped).Foralldrained
analysesLR95< LR99<1, indicatingthat meshrefinementmadeno difference.
4-45
As might havebeenexpected,SeriesII r-refinementplacedmuchmore onerousdemandson

lengthratio thanwas observedwith Series1. The smoothness
of the computedground surface
movementprofile is heavilydependenton the gradingof the meshin the horizontaldirection.
Thereare simplyfewer nodepointsto fit the profile to, andthis accountsfor much of the
discrepancybetweenmesheswhich are in the sameseries,but which possessdifferentdegrees
of refinement(e.g. comparingx8y8bI andx8y8b6)- seeFig. 4.37(a) and(b).
From Table4.35 it canbe seenthat, for the undrainedanalyses,thereis not one casewhich
givesLR95below 12, or an LR99below 25 for settlement-S*n.. The situationimproves
slightlyfor heave+S*,with most of the analysesachieving±5% without requiringany
grading;to reach±1% maystill requirea lengthratio between4 to 25. For all drained
analyses,however, LR9s< LR99<1, thusmeshrefinementmakesno difference,Fig. 4.37(c).
4.6.3 Summary
Basedon the initial seriesof r-refinedmeshes,whenenoughnodesarepresentto ensurethat

resultsarewithin:1:1% of IF%., very little relocationof nodalpositionsis necessaryto
optimizethe solution. The samewould not be expectedof a meshwhich was deficientto
beginwith; if therewere insufficientnodes,it would be more critical wherethey were located.
This was confirmedby the secondseriesof r-refinedmesheswhich showedthat, when

insufficientnodesare presentto ensurethat resultsarewithin ±1% of "true" the
solutionbecomesmuchmore sensitiveto the positionsof the nodes. Therewas a bigger
differencebetweenthe profilesobtainedwith a uniform mesh(LR=I) anda highly graded
mesh(LR=32), thanwas seenwith the first seriesof r-refinedmeshes.
4.7 Type of Elements
Every meshconsideredso far in this chapterhasemployedthe 8-nodedlinear strain

quadrilateral(LSQ) with full 3x3 integrationfor both soil andwall, with triangularelements
usedonly for props(whenrequired). The majority of analystsseemto favour quadrilaterals
for Soil-structureinteraction(SSI) applications,andsomeof the reasonsfor tl-ýshavealready
beendiscussedin Chapter3.
446
It is difficult to saycategoricallythat the LSQ performsbetterthanits triangularcounterpart
(or vice versa)on a "per node" basis,andthe writer is unawareof anystudieswhich have
examinedthis in detail, especiallyin an SSI context.
This sectionwill not attemptto comparethe relativemeritsof trianglesandquadrilaterals,but

ratherwill considerthe possibilityof usingtrianglesasa meansof selectivelyrefininga mesh
originallycomprisingquadrilaterals,to checkits adequacy.Fig. 4.38 showsan LSQ divided
first into two LSTs (linear straintriangles)andtheninto four LSTs. In one sensethis is a form
of h-refinement;the numberof elements(andhencedegreesof freedom)is beingincreased,
the originalnetwork of grid linesis still visible,andthe elementshavethe sameorder (i.e.
quadraticin displacement,
linearin strain).
Alternativelyeachquartetof trianglescouldbe consideredto be forming a new higherorder

element,possessing13 nodesinsteadof the normal8. However,this is not the sameas
replacingthe quadrilateralwith another(single)quadrilateralwhich employsa higherorder
displacementapproximation,so it is not strictly p-refinementasdefinedearlier(seeFig. 4.17).
If eachLSQ were replacedby a 9-nodedquadrilateral(Lagrangian),that would truly be p-
refinement,but this is not possiblein CRISPwith the currentlyavailableelementtypes. A
meshcomprisingLSTs couldbe p-refinedin CRISP,if eachLST was replacedby a 15-noded
CuST. The subdivisionproposedin Fig. 4.38 is thus a hybrid of h- andp-refinement.
Two of the meshesencounteredpreviouslywhererefinedin this manner,namelyx8y8e6(from

r-refinementSeriesI) andx8y8b4(from r-refinementSeriesII). Both of thesemesheswould
be consideredto havereacheda reasonablyadvancedstateof refinementin their respective
series,andfurtherrefinementwould not be expectedto improveresults. In eachcase,every
quadrilateralsoil element(only) wasreplacedby 4 LSTs, Fig. 4.38(c);the wall was maintained
asa columnof LSQs so that its bendingstiffnesswasunaffected,andbecausecalculating
bendingmomentsfrom Gausspoint stressesis mucheasierwith quadrilaterals.Meshes
x8y8e6/tandx8y8b4/t- with the 'V denoting triangulated- were thus formed,Fig. 4.39.
Only the unproppedundrainedcasewas considered,asit hadbecomeapparentfrom earlier
it
studiesthat was a particularlydemanding case. A rangeof soil parameterswas used (sets I-
5, and 12-13).
4-47
4.7.1 Results of analyses
Tables 4.36 and 4.37 summarizethe key results for x8y8e6/t and x8y8b4/t, quoting r*. and
percentagedifferencesw. r. t. the original (untriangulated) mesh. It is clear that only very small
changeshave occurred in the relatively fine mesh through this type of hybrid refinement;
confirmation of its adequaterefinement. By a small margin it appearsthat wall displacements
are influenced the most. As might have been anticipated, it was those analyseswhich permit
plastic yielding of the soil which show the biggest variation.
For the rathercoarsermesh(x8y8b4/t),the percentagevariationsare muchmore significant,

with manyexceeding±5% and somegreaterthan±10%. In general,wall bendingmoments
are the mostaffectedof the key results,with excavationheavesbeingthe least.
4.7.2 Summary
This type of mesh refinement is probablyjust as effective as standard h-refinement, for the
purposes of improving results and/or testing for adequacyof existing mesh. In common with
h-refinement, it need only be carried out in selectedparts of the mesh, where it is felt that
benefit to the solution may accrue. It is particularly useful where refinement is only required in
one or two elements,as the alterations to adjacent elementsare minimized (if not eliminated).
One examplewhere this could be very useful in a retaining wall analysisis in the element
immediately in front of the wall and just below final dig (formation) level. Many investigators
have observedunusually high horizontal and vertical stressesin this element,when compared
with the full profiles which may be plotted down the wall on the passive side, and along the
underside of the slab (if present) - seeChapter 7. There are genuinely high stressgradients
and constraints placed on this element, and so its anomalousbehaviour is not altogether
surprising. Nonetheless,to be able to improve accuracy and avoid spurious results in this area
through some localized refinement would be beneficial.
Somedesigners/analysts to the
mayprefer avoid useof triangular so
elements, it is worth
noting that refinementwithin a singlequadrilateralelementcanbe carriedout without the need
for anytriangles,asillustratedin Fig. 4.40(b)&(d). The potentialfor further refinementalong
theselinesis evident. Anotherpossibilityinvolvesa combinationof quadrilateralsand
triangles,Fig. 4.40(c),whereagainthe potentialfor further refinementis clear.The only
limitation would be to ensurethat unacceptable
aspectratioswere not generated.
4-48
4.8 Discussion and Summary
In this chapter,someof the geometricmodellingissuesfacingthe retainingwall analysthave

beenconsidered.To commencewith, all of the decisionswhich hadto be takenwere listed,
togetherwith the assumptionswhich hadto be madeen route to producingthe final FE mesh.
Then,a numberof thesewere selectedfor further studyusingan idealizedexcavationandwall
geometry. Severaldifferentproppingconditions(againidealized)were selected,alongwith a
representativerangeof constitutivemodelsand soil parameters,andthe two limiting
conditionsof drainage.
The following havebeeninvestigatedto varyinglevelsof thoroughness:
a) boundarylocations
b) meshaspectratio
C) boundaryconditions
d) numberof elements
e) sizeof elements
f) type of elements
Their influencehasbeenquantifiedby studyinga numberof "key" resultswhich the designeris

most likely to be interestedin, namely:
i) horizontalwall displacements
ii) wall bendingmoments
iii) excavationheave
iv) ground surfacemovements
Neitherof the abovelists (a)-(f) or (i)-(iv) are exhaustive,but were consideredto be sufficient
to makea genuineattemptat addressingthe perceivedproblems.
For horizontalwall displacementsirrespectiveof proppingarrangement,an increasingdegree

of nonhomogeneity in stifffiessmadethe resultslessvulnerableto choiceof boundarylocation,
relativeto the baseline(homogeneous isotropiclinearelastic)case. However,increasingthe
degreeof anisotropyhadlittle effect or madethe resultsmoresensitiveto boundarylocation.
Neitherof thesewas particularlyunexpected.What was a surprisewas that permittingthe soil
to yield plasticallymadethe resultsmore sensitiveto boundarylocation. Intuitively, this
4-49
seemswrong,asplasticyield occursin a small,well-definedpart of the mesh(evenat
collapse).Overall,analysesconvergedfasteron the "true" maximumdisplacementif
conditionsweredrained. The implicationof this is that, if the adequacyof a meshis assessed
usinganundrainedanalysis,then it will be adequatefor a drainedanalysis.Unfortunately,it
cannotbe saidthat if a meshis shownto be adequatefor an elasticanalysis,that it win be
adequatefor a plasticanalysis.
For wall bending moments, the degree of nonhomogeneityseemednot to make much

difference for undrained analysis,but was still important for the drained condition. Increasing
the degree of anisotropy made the results more sensitiveto boundary location, whereasplastic
yield caused little change (undrained) or brought about a marginal improvement (drained).
For excavationheavesincreasingthe degreeof nonhomogeneity generallymadethe results

lessvulnerableto boundarylocation,whereasincreasingthe degreeof anisotropymadelittle
difference. Yield madethingsmarginallyworse(undrained)or muchimproved(drained).
For downwardgroundsurfacemovements(-S), increasingdegreesof both nonhomogeneity

and anisotropymadethe resultsmorevulnerableto boundarylocationunderundrained
conditions,whereasincreasingthe amountof plasticyield broughtabouta slight improvement.
For upward groundsurfacemovements(+S), increasingthe degreeof nonhornogeneity made
the resultslessvulnerableto boundarylocation,whereasincreasingthe degreeof anisotropy
andthe amountof plasticyield hadvirtually no affect(drainedconditions).
The overall aspect ratio (AR) of the mesh was shown to be important, and to have an
interaction with the location of remote mesh boundaries. The convergenceof a key result (5,
KV or S) on its "true" value depends on whether the mesh is essentially square (AR 1)
sts or
rectangular (AR # 1) to begin ikith, and the relative rate at which the remote boundary
locations are extendedaway from the excavation. Convergenceon a false maximum/n-dnimum
can occur if a4ppropriate AR is used. The location of both remote boundaries must be
tested; it is not sufficient to test only one or the other in isolation.
Boundaryconditionis anotheraspectwhich is normallyglossedover, with rough/smooth

boundariesimposedaccordingto the preferenceof the analyst. If the boundariesare
sufficientlyremote,their conditionshouldnot matter;henceswitchingfrom rough to smooth
4-50
(or viceversa)providesa convenienttechniqueof showingwhetheror not sufficientdistance
hasbeenachieved.
Concerningthe numberof elements,thesestudieshaveindicatedthat a relativelymodestsized

meshcanproduceresultsto within ±5% of their expectedvalue- but at leasttwice the number
of nodesareneededto bring this to within±1%.
The locationof the nodeswithin the meshis alsoimportantand,for a givennumberof d.o.f,

therewill be an optimumgradingwhich is situationdependent.A balancemustbe struck
betweenthe needto havea high densityof nodeswherestressgradientsare high, whilst
maintaininga reasonabletransitionto largerelementsizestowardsthe remoteboundaries
(avoidingabruptchangesin size). The brief studiesreportedin this chapterindicatethat
gradingprovidesanotherway of examiningthe adequacyof a mesh. If too few nodesare
used,the analysisis particularlysensitiveto their location;if the total numberof d.o.f. is
adequate,far lesssensitivityis evident.
The actualjyj2eof elementsused(trianglesor quadrilateralsin the contextof this study)would

not appearto be an issue. It is difficult to do a straightcomparison,becauseif eachelementin
an initially quadrilateralmeshis subdividedinto two triangles(the simplestconversion)this
will generatearound33% extra nodes(d.o.f ). An increaseof 167%happensif quadrilaterals
are split into four triangles. This offers a potentiallyconvenientway of checkingthe adequacy
of a quadrilateral-based mesh- the increase in d. f.
o. shouldonly change the key resultsif the
originalmeshwastoo coarse. It seemsthat the mainpotentialfor usingdifferentelementsis
in very localizedrefinement- whereparticularanomaliescanbe eliminatedor minimized.
Brief mentionshouldbe madeof adaptivemeshing(e.g. Babuskael al., 1986). Mesh

adaptivityconcernsiterativerefinementof an initially coarsemeshduring an analysis,using
error estimates"in flight" to dictatewhat refinementis carriedout (andwhere). The physical
geometryis only onefactor in the adaptivity- the natureandlocationof appliedloadsis
equallyimportant. This is an importantandongoingareaof researchin FEM methods,though
stressanalysis.Commercial
the presentthrust is aimedmuchmore at structural/mechanical
FE packageswith adaptivemeshingare currentlyavailablein CAD/CAK but not yet in
geornechanics.Simpson(1973) incorporateda form of adaptivemeshingin his work on
retainingwall modelling,but an intervalof morethan20 yearspassedbeforeinterestresumed
4-51
(e.g. Mar, 1993),thoughnot specificallyfor retainingwalls. Much of the work on refinement
in this chapterwould be superseded
if adaptivemeshingwere matureenoughfor inclusionin
CRISP.
Furthermorethe entireboundarylocation/conditionissuecouldbecomeirrelevantif "infinite

elements"werewidely available(Zienkiewiczet al., 1983). Theseelementsincludeautomatic
of the far field on one or two boundaries,andarebeingemployedincreasingly
representation
in rock mechanicsFEA (e.g. Pandeet al., 1990). However,their usein soil problemshas
beenlimitedthus far, andthereareno plansfor infinite elementsto be incorporatedin CRISP
at the time of writing.
In summary,manygeometricmodellingdecisionshaveto be takenin settingup an embedded

retainingwall mesh. It hasbeendemonstrated
that incorrect(or poorly informed)decisions
canleadto potentialerrorsin the computedresultsmostfrequentlysoughtbe designers.
Engineeringmodellingis concernedwith obtaininganswersof adequateaccuracyand
precision,giventhe natureof the task at hand. It is the writer's contentionthat most
geotechnicalengineershaveconcentratedalmostexclusivelyon materialcharacterization, and
havelargelyignoredthe needfor carefulmeshdesign. Whilst someof the errorsquantifiedin
this chaptermaycanceleachother out in somecircumstances,
they could well reinforceeach
other in a cumulativeeffect.
4-52
Table4.1 Boundarylocationanalyses- fixed parameters
Parameter Value(s) adopted
Wall: type reinforced concrete diaphragm

thickness 0.8m
retained height 8m
penetration depth 8m
Excavation: half-width 8m.

(symmetric ahout centreline)
Elements: soil and wall 8-node quadrilaterals (LSQ)

props 6-node triangles (LST)
Concrete: linear elastic (E = 20 GPa, v=0.15)
Groundwater: hydrostatic from ground surface downwards
& (ah/C7,): 2.0
Wall installation: wished in place
Analysis stages: 0) in-situ conditions (wall in place)

1) install prop(s) *
2) excavate soil to formation level
3) remove any temporary props
Notes:
1. K, in termsof total stresses= Kt = 1/2(l+ Kj = 1.5 for theseconditions
2. * if applicable
3. props formed from singleelement:point contactwith wall, and oppositeedgefree to
moveverticallyon a roller boundary
4-53
Table4.2 Boundarylocation analyses- variableparameters
Parameter Value(s) adopted
Wall support: a) none (free cantilever)

b) singly propped at top
C) singly propped at bottom (i. e. formation level)
d) doubly propped top and bottom
e) alternately propped; first at top and then at bottom
Analysis type: a) undrained (total stress)

b) drained (effective stress)
Soil: a) linear elastic

i) homogeneousisotropic
H) nonhomogeneousisotropic
iii) homogeneousanisotropic
b) linear elastic-perfectly plastic

i) nonhomogeneous isotropic
Boundary a) rough on base and on remote vertical boundaries

conditions: b) smooth on base and on remote vertical boundaries
Notes:
1. Boundary conditions on centreline of mesh always smooth (axis of symmetry)
2. Tresca or Mohr-Coulomb yield criteria used for elastic-perfectly plastic analyses
3. All props rigid and installed 'perfectly' (see Section 4.2.2for discussion)
Table4.3 Soil parameters- linearelastic,isotropic,variablenonhomogeneity(undrained)
Set x Eu (NDa) vu
1 0 50 0.497
2 0.444 50 + 5z 0.497
3 1 5z 0.497
4-54
Table4.4 variableanisotropy(undrained)
Soil parameters- linear elastic,homogeneous,
Set n Eu,, (NTa) Euh (NTa) Vuvh VWih
1 1 50 50 0.497 0.497
4 2 50 100 0.497 0.000
5 3 50 150 0.497 -0.500
Table4.5 Soil parameters- linearelastic,isotropic,variablenonhomogeneity(drained)
Set x E'(NDa) v?
Id 0 40 0.200
2d 0.444 40 + 4z 0.200
3d I 4z 0.200
Table 4.6 Soil Parameters- linear elastic, homogeneous,variable anisotropy (drained)
Set nt E'V(Wa) Eh (NVa) vFVh vF hh
Id 1 40 40 0.200 0.200
4d 2 40 80 0.200 -0.250
5d 4 40 160 0.200 -0.825
NB: Gh in all cases=16671vlPa m(=GhlE, ) =5112 =0.417
.-
Table 4.7 Soil parameters- linear elastic-perfectly plastic, isotropic (Tresca)
Set x E. (,%Va) V, E. / c. c, (kPa)

10 0.444 50 + 5z 0.497 500 100 + loz
11 1 5z 0.497 500 100 + loz
12 0.444 50 + 5z 0.497 1000 50 + 5z
13 1 5z 0.497 1000 50 + 5z
NB: higherE. 1c. meanslowerc. (for afixedE, ) and hence)deldingis more likely
Table4.8 Soil parameters- linearelastic-perfectlyplastic,isotropic (Mohr-Coulomb)
Set x E' (NTa) V, c' (kPa) V (0)
12d 0.444 40+4z 0.200 0 25

13d I 4z 0.200 0 25
4-55
Table4.9 Influenceof meshboundarylocationon horizontalwall displacement
undrained drained
Case Set Rgs R99 Error Set Rgs R99 Error
1 6 8 Id 8 9 V2 +
unpropped -
2 6 9 - 2d 3 7
3 5 7 - 3d 4
4 7 9 - Q 8Y2 9V2 +
5 8Y2 9 Y2
- 5d 9Y2 10 +
10 6 9 - 12d 5 9
12 7Y2 9 - 13d 4Y2 9
11 5Y2 8 -
13 7 9 -
top- 1 5 7 - Id 7 9 +
propped 2 5 8 - 2d 8 +
3 4 7 - 3d 7 +
4 5 V2 8 - 4d 7 9 +
5 6 9 - 5d 8 9V2
10 5 8 - 12d 4 6
12 6Y2 9 - 13d 4 6
11 5 8
13 8 9Y2 -
bottom- 1 6 9 - Id 6 9 +
propped 2 3Y2 8 - 2d 6 9 +
3 4 - 3d 3 6 +
4 6 9 - 4d 4 9 +
5 4 9 - 5d 5Y2 7 +/-
3Y2 2Y2
10 8 - 12d 7 +
12 3Y2 8 - 13d 5 +
11 3
13 3 8 +
doubly- 1 6 9 - Id 7 9
propped 2 5 8 - 2d 4 8 +
3 4 7Y2 - 3d 4 +
4 5 9 - 4d 3 9 -/+
5 5 9 - 5d 9 9V2 +
8Y2 +
10 5 8 12d 5
5V2
12 5 8 13d 8 +
11 4 8
13 8 9Y2 -
alternately 1 5 7 - Id 8 9Y2 +
propped 2 5Y2 9 - 2d 8 +
3 4 7 - 3d 6 +
4 5Y2 8 - 4d 8 9Y2 +
5 7 9 - 5d 9 9Y2 +
10 5 Y2 9 - 12d 3 4
12 7 9 - 13d 3 5
11 5 8
13 6 9
* Rgsachievedwith smaHest
mesh(x2y2)
4-56
Table4.10 Influenceof meshboundarylocation on wall bendingmoment
undrained drained
Case Set Rgs R" Error Set Rgs R" Error
1 3 6 Id 2 4
unpropped -
2 3 4 2d 4
-
3 5 7 3d 3
- 4V2
4 5 9 4d 3
-
5 7 9 V2 - 5d 4 6
10 3 5 12d 9
-
12 3 6 13d 9
3V2
-
11 4 -
13 4 -
top- 1 3Y2
5 Id 6 9 +
-
2 4 8 2d 7 +
propped -
3 4 6 3d 6 +
-
4 5 8 4d 7 9 +
-
5 6 9 5d 7 9
-
10 4 8 12d 3Y2 5
12 5 81/2 13d 3 5V2
11 4 7
f 13 6V2 9
8V2 9 V2 +
bottom- 1 4 9 Id
2 4 8 2d 5 8 +
propped 3V2
3 2Y2
6 3d 7 +
4 2 8 4d 9 10 +
+
5 3V2
5 5d 81/2 10
10
4 7Y2
12d 2V2 7 +
2Y2
12 4 7 13d 6 +
11 2 5
13 5 81/2
doubly- 1 3 5 Id 7 9 +
2 3 6 2d 5 8 +
propped
3 5 3d 4Y2 8 +
4 3 5 4d 7 9 +
5 3Y2
6 5d 8 9 +
10 3 6 12d 4 8 +
+
12 3 6 13d 3 6Y2
11
5
13 6 9
1 3 8
alternately 1 2 Y2 4 Id
propped 2 3 2d
3 5 8 3d 3
-
4 4 8 4d 3Y2 7
-
5 6 9 5d 4 V2 7
-
10 3 12d 7Y2
- 2Y2
12 4 13d 4
11 5 8
13 4 8
uuuvuisrýu%oF.
mesh(x2y2)
* Rgsachievedwith smaHest /x/\ mconsismin
4-57
Table4.11 Influenceof meshboundarylocation heave
on excavation
drained
undrained
Case Set Rgs R99 Error Set Rgs R99 Error
1 9 9 V2 Id 9 9V2 -
unpropped -
2 8V2 9 V2 2d 7 9-
- 6V2
3 8 9Y2 3d 9-
-
4 9 10 4d 9 9Y2 -
-
5 9 10 5d 9 9V2
-
10 8 9 V2 12d 5 7
- 4V2
12 9 9 Y2 13d 9
-
11 8 9 V2 -
13 8Y2 9 Y2
8Y2
1 9 10 Id 9V2
top- -
2 8 9 Y2 2d 7Y2 9
propped -
3 8 9Y2 3d 61/2 9
-
4 9 10 4d 81/2 9Y2
- Y2
5 9 10 5d 8 9Y2
- 6
10 8 9 Y2 12d 4
-
12 8Y2 9V2 13d 4 6-
-
11 8 9 Y2 -
13 9 9Y2
bottom- 1 9 10 Id 8Y2 9V2 -
-
2 8Y2 9Y2 2d 7 9-
propped -
8 9Y2 3d 6Y2 9-
3 -
4 9 10 4d 8V2 9V2 -
-
5 9 10 5d 9 10 -
-
10 8Y2
9Y2 12d 5 8
-
12 8Y2 9 V2 13d 4 6
-
11 8 9Y2 -
13 9 10 -
doubly- 1 9 10 ld 8Y2 9Y2
-
2 8 Y2 9V2 2d 7 9
propped -
3 8 9 V2 3d 6Y2 9
-
4 9 10 4d 8Y2 9V2
-
5 9 10 5d 9 10
-
8Y2 9 V2 12d 4Y2 8
10 - Y2
8Y2 9Y2 13d 3 81/2 -
12 -
11 8 9 V2 -
13 9 10
9 10 Id 9 10 -
alternately 1 -
8Y2 9Y2 2d 7Y2 9-
propped 2 -
8 9Y2 3d 6Y2 9-
3 -
10 4d 9 10 -
4 9 -
10 5d. 9 10 -
5 9 -
8Y2 9Y2 12d 4Y2 8-
10 -
8Y2
9 Y2 13d 3Y2 6
12 -
11 8 9V2 -
1 13 81/2 9 Y2
/\/\- inconsistentconvergence
4-58
Table4.12 Muence of meshboundarylocationon surfacemovement
undrained drained
heave settlement heave
settlement
Case Set Rgs R99 Err Rgs R99 Err Set Rgs R99 Err Rgs R99 Err
.'
6 9 + 9Y2 10 Id 9 10 -
unpropped 1 -
2 9 10 + 9 10 2d 8V2 9V2 -
-
3 9 10 + 9 10 3d 8 9Y2 -
-
4 6 9 + 9 10 4d 9 9V2 -
-
5 8 9V2 + 9 10 5d 9 10 -
-
10 9 10 + 9 10 12d 69V, 1 8Y. 9Y2 -
-
12 9Y2 10 + 9 Y2 10 13d 61/2 8Y2 7Y7 9
-
II 7Y2 9Y2 + 9 10 -
13 31/2 6 9Y2 10 -
top- 1 4 8 + 9 10 ld 9 9V2
-
2 9V2 10 + 1:
1 9 Y2 10 2d 81/2 9V2 -
propped - V2
3 9Y2 10 + 9 10 3d 8Y2 9 -
-
4 5 8Y2 + 9 10 4d 9 10 -
-
5 5 8Y2 + 9 10 5d 9 10 -
-
10 9 10 + 9 10 12d 8 9Y2 -
-
12 9 10 + 9 10 13d 8Y2 9Y2 -
-
11 9Y2 10 + 9Y2 10 -
13 9 10 + 9 Y2 10 -
bottom- 1 7 9 9Y2 10 ld 9 9Y2 -
2 9 10 + 9 10 2d 8Y2 9/2 -
propped -
3 9Y2 10 + 9 10 3d 8 9Y2 -
-
4 7 9 1 9 Y2 10 4d 9 9Y.
- - 1, 9
5 6 9 9Y, 10 5d 9Y2
- -
10 9 10 + 9 10 12d 8Y2 9Y2 -
-
12 9 10 + 9 10 13d 8 9yi -
-
11 9Y2 10 + 9 10 -
1 13 9 10 + 9 10 -
doubly- 1 7 9 9 10 Id 9Y2 9Y2 -
-
2 9 10 + 9 10 2d 7Y2 9-
propped -
3 9Y2 10 + 9 10 3d 8 9Y, -
i -
4 7 9 9 Y2 10 4d 9 9Y2 -
-
5 7 9 9Y2 10 5d 9 9Y2 -
-
9 10 9 10 12d 8Y2 9Y2 -
10 + -
12 9 10 9 10 13d 8 9Y.
+ -
11 9V2 10 + 9 10 -
13 9V2 10 9 10 -
5 8Y2 + 9Y2 10 Id 9 9Y2
altemately 1 -
2 9Y2 10 + 9Y2 10 2d 8Y2 9Y2
propped -
3 9 10 + 9 10 3d 8 9Y2
-
5 Y2
9 9 Y2 10 4d 9 10
4 +
6 9 + 9Y2 10 5d 9 10
5 - Y2
10 9Y2 10 + 9 Y2 10 12d 8Y2 9 -
-
12 9V2 10 + 9 Y2 10 13d 8 9Y, -
-
11 8Y2 9Y2 + 9 10 -
13 8Y2 9/2 + 9 Y2 10
Note: ". " becauseno settlementin drainedelasticcases
4-59
Table4.13 Statisticalanalysisof boundarylocationanalyses
R9,s R99 R95range frequency
wall displacement
mean 5.39 8.32 <2 9
std.dev(s.d.) 2.45 1.40 2-3 1
mean-s.d. 2.94 6.92 3-4 9
mean+s. d. 7.84 9.71 4-5 10
min <2 3 5-6 21
max 9.5 10 6-7 11
7-8 8
8-9 8
9-10 3
wall bendingmoment
3.96 6.95 <2 15
mean
std.dev 2.50 2.15 2-3 8
mean-s. d. 1.46 4.80 34 20
d. 6.46 9.10 4-5 14
mean+s.
<2 <2 5-6 9
min
9 10 6-7 5
max
7-8 5
8-9 3
9-10 1
excavationheave
8.30 9.56 <2 0
mean
0.37 2-3 0
std.dev 0.92
d. 7.38 9.20 4 2
mean-s.
d. 9.21 9.93 4-5 6
mean+s.
3.5 8 5-6 2
min
9.5 10 6-7 5
max
7-8 5
8-9 33
9-10 27
ground surfacemovement
a) b)
a) settlement
8.30 9.58 <2 0 0
mean
0.72 2-3 0 0
std.dev 1.71
d. 6.58 8.86 34 1 0
mean-s.
10.31 4-5 1 1
mean+s.d. 10.01
3.5 6 5-6 4 2
min
10 10 6-7 4 2
max
7-8 6 2
b) heave 8-9 3 11
8.75 9.78 9-10 35 62
mean
std.dev 1.01 0.35
mean-s. d. 7.74 9.43
mean+s. d. 9.76 10.14
4 8
max 9.5 10
4-60
Table4.14 Valueof reachedat R=10 usingmeshxl6yI6 asthe reference
(unproppedcantilever)
Characteristicvalue
MMSI / M* VM31 / Vk SM., S*Mal
Case Set /8 *Max
8mal max owl
settlement heave
undrained 1 0.995 0.995 0.820 0.680 1.010

2 0.980 0.908 0.910 0.780 1.400
3 0.990 0.992 0.940 0.830 1.200
12 0.960 0.997 0.900 0.740 1.460
13 0.966 1.003 0.890 0.730 1.030
drained Id 1.100 1.003 0.850 0.780
2d 1.010 1.001 0.950 0.890
3d 1.001 1.001 0.970 0.920
not relevant- all ground movementupward (i.e. heave,+S) in this case
4-61
Table4.15 Influenceof meshboundarylocationon wall displacement: differentmesh
aspectratios
Series Set Rq!; R99 Error Comments
A 1 6 8
2 6 9 Referencecase
3 5 7
B 1 7 9 + convergence improvedwhen
2 4Y2 7Y2 + X>O; convergence from
3 <2 6 + differentdirection
6Y2
C 1 9
6Y2
2 9 sameas SeriesA
3 5 7
D 1 5 7Y2 + convergence alwaysimproved

3V2
2 6 + andfrom differentdirection,
3 2 5 + but 8, valuestoo small
n.
E 1 8 9Y2 convergenceworse,diough
2 7 9 from samedirection
3 5V2 8Yz
Table4.16 Muence of meshboundarylocationon wall bendingmoment: differentmesh

aspectratios
Series Set Rgs R99 Error Comments
Y2
A 1 3 6
2 3 4 Referencecase
3 5 7
Y2
B 1 <2 5 convergenceimproved for all X;
2 <2 2 different direction when X=I
3 <2 6
Y2 Y2
C 1 3 5
2 3 4 sameas Series A
3 5 7
D 1 <2 5 convergenceimproved, and

2 <2 2 from different direction, but
3 <2 5 + ±M. values too smaU
V2
E 1 3 5
2 3 4 roughly same as Series A
Y2
3 5 8
4-62
Table4.17 Influenceof meshboundarylocation on excavationheave: differentmesh
aspectratios
Series Set R95 R99 Error Comments
A 1 9 9 Y2 -
2 8 Y2 9V2 - Referencecase
3 8 9 Y2 -
V2
B 1 7 9 convergenceimproved, especially
-
2 6 8 - with increasing X
3 5 8
C 1 9 9 Y2
2 8 9 Y2 same as SeriesA
3 8 9 V2
D 1 5 8 convergencemuch improved,
2 4 6 but V. values too small
3 4 6
Y2 9 V2
E 1 8 marginal improvement over
Y2
2 7 9 SeriesA
3 7 9
Table 4.18 Influence of mesh boundary location on ground surface movement : different
mesh aspect ratios
settlement heave
Series Set Rgs R99 Error Rgs R99 Error Comments
A 1 6 9 Y2 10 Reference
9 + -
2 9 10 + 9 10 - case
3 9 10 + 9 10 -
B 1 8 9 Y2 8 Y2 9 V2
- convergence
2 3Y2 8 7 V2 9 Y2 improvedfor
3 6Y2 8 Y2 + 7 9 - X>O
C 1 6 9 + 9 Y2 10 - same as
2 9 10 + 9 10 + SeriesA
3 9 10 + 9 10
D I 4Y2 8 7 9 convergence
2 <2 5 6 8Y2 muchimproved
3 4 6 5 Y2 7 V2 for all X
+
E 1 9 10 + 9 10 similarto
2 9 V2 10 9 9 Y2 SeriesA
+
3 8Y2 9 Y2 + 9 9V2
4-63
Table4.19 Comparisonof maximumvaluesof key resultsfor differentmeshaspectratios
(SeriesA to E)
wall displacement (mm)

Set 1 Set 2 Set 3
anny SFnaz
Series Smax
A -18.9 -12.9 -45.1

B -18.9 -12.9 45.1
C -18.9 -12.9 45.1
D -17.9 -12.1 43.8
E -20.5 -13.2 45.6
wall bending moment (kN-m)

Set 1 Set 2 Set 3
Series -M.. +MM&X +MMAX -MBWI +MMMX
-MMRI
A -440.6 94.1 -314.3 156.4 -161.0 399.1
B 440.6 94.1 -314.3 156.4 -161.0 399.1
C -440.6 94.1 -314.3 156.4 -161.0 399.1
D -438.1 82.0 -313.2 146.4 -162.4 388.6
E 437.5 113.9 -314.2 161.3 -160.7 403.4
excavation heave (mm)

Set I Set 2 Set3
VMAIE VMXI VMSX
Series
A 46.0 14.9 23.8
B 46.0 14.9 23.8
C 46.0 14.9 23.8
D 36.3 12.9 21.3
E 41.7 14.3 23.1
ground surface movement (nun)

Set 1 Set 2 Set 3
+SMMI +SMAI
Series -SMUZ
+Sntax
-SM21 -SRM31
A -11.6 21.1 -3.3 4.8 -9.6 7.5

B -11.6 21.1 -3.3 4.8 -9.6 7.5
C 21.1 -3.3 4.8 -9,6 7.5
-11.6
D 12.4 4.6 3.1 -12.1 5.4
-12.0
E -10.1 16.8 -3.2 4.2 -10.2 6.8
Notes
1. SeriesA, B and C (asexpected)all endup with the samemaximumvalues,as they all endat
meshxlOylO.
2. SeriesD (which hasX/H fixed at 6) givesmaximaconsistentlylower than SeriesA/B/C,
exceptfor downwardsurfacemovement(-S), so adequatemeshwidth is required.
3. SeriesE (whichhasY/W fixed at 6) giveshigherS.., alternatelyhigher/lower±M.,,,,,,lower
V., andlower ±S. Um SeriesA/B/C.
4-64
Table4.20 Influenceof meshboundaryconditionson unproppedcantileverwall
Set I Set 2 Set 3

Mesh rough smooth rough smooth rough smooth
("lodiffi difj)
(016 r1od1j))
wall disPlacementB.,,, (mm)

x2y2 -12.16 -13.25 -8.19 -8.74 -33.69 -34.76
(9-0) (6.7) (3.2)
x6y6 -18.26 -17.63 -12.21 -12.21 43.93 43.96

(-3.5) (0-0) (0-1)
XIOYIO -18.92 -18.30 -12.88 -12.86 45.09 45.13

(0-1)
wall bending moment ±M,.., (kN-m)
x2y2 -386.5 -418.3 -291.9 -304.3 307.5 317.8

(8.2) (4.2) (3.3)
x6y6 -436.8 -440.0 -313.2 -313.7 389.5 390.2

(0.7) (0.2) (0.2)
XlOylo 440.6 441.6 -314.3 -314.0 399.1 400.3

(0.2) (-0-1) (0.3)
excavationheave V... (mm)

x2y2 14.96 21.19 6.55 8.55 12.18 14.72
(41.6) (30.5) (20.9)
x6y6 35.24 40.93 12.78 13.75 21.13 22.18

(161) (7.6) (5.0)
XlOylo 45.96 51.75 14.92 15.57 23.78 24.47

(12.6) (4.4) (2.9)
ground surfacemovement -S... and +S... (mm)

x2y2 -13.57 -17.92 -7.67 -8.63 -20.14 -19.40
(32.1) (12.5) (-3.7)
0.00 0.94 0.00 0.01 0.00 0.91
(00) (00) (00)
x6y6 -12.16 -16.36 4.66 4.86 -12.17 -11.23

(34.5) (4.3) (-77)
11.37 16.63 2.96 3.84 5.22 6.19
(46.3) (29.7) (18.6)
XIOYIO -11.65 -16.18 -3.34 -3.40 -9.63 -8.98

(38.9) (1.8) (-6.7)
21.15 26.85 4.82 5.46 7.51 8.18
(270) (13.3) (8.9)
4-65
Table4.21 Influenceof meshrefinement: h-refinementSeriesI
mesh
Value Set x2y2 x2y2b x6y6 x6y6b
r1odifO rlo diffi
wall 1 -12.16 -12.31 -18.26 -18.36

displacement (1.23) (0.55)
8. (mm) 3 -33.69 -33.85 -43.93 -4199
(0.47) (0.14)
wall bending 1 -386.50 -387.10 -436.80 -436.30

moment (0.16) (-0.11)
±K.. (kN-m) 3 307.50 307.90 389.50 389.30
(0.13) (-0.05)
excavation 1 14.96 15.06 35.24 35.04

heave (0.67) (-0.57)
V.. (mm) 3 12.18 12.25 21.13 21.09
(0.57) (-0.19)
ground 1 -13.57 -13.59 -12.16 -12.02

surface (0.15) (-1.15)
movement 0.00 0.00 11.37 11.26
-S. and 00 (-0.97)
+S, (mm) 3 -20.14 -20.01 -12.17 -12.16
ý,,
(-0.65) (-0.08)
0.00 0.00 5.22 5.23
Go (0-19)
Table4.22 Meshesusedin h-refinementSeriesII
fNVTX CNEL
mesh NVTX NEL NN
x8y8aO 42 30 113 065 050

. .
x8y8bO 72 56 199 111 093
. .
x8y8cO 110 90 309 170 150
. .
x8y8dO 168 143 478 258 238
. .
x8y8eO 255 224 733 392 340
. .
x8y8fD 380 342 1101 585 620
. .
x8y8gO 528 483 1538 812 805
. .
x8y8hO 650 600 1899 1.00 1.00
4-66
Table4.23 Meshesusedin h-refinementSeriesIII
fNM CNEL
mesh NVTX NEL NN
x8y8a 42 30 113 065 050

. .
x8y8b 110 90 299 169 150
. .
x8y8c 210 182 601 323 303
. .
x8y8d 342 306 989 526 510
. .
x8ySe 484 441 1408 745 735
. .
x8y8f 650 600 1899 1.00 1.00
Table4.24 Influenceof meshh-refinement(SeriesHI) on wall displacement
Case Set NN9!; NN99 Error Comments

unpropped 1 550 1000
undrained 2 550 1000
3 500 900
4 550 1000
5 550 1000 -
12 600 1400 -
13 850 >1400 -
top-propped 1 500 1300 -
undrained 2 600 1400 -
3 500 1200
4 500 1400 -
5 500 1400 -
12 700 1400 -
13 550 1400 -
bottorn- 1 300 900 -
propped 2 600 900 -
undrained, 3 600 900 -
4 300 600 -
5 300 1000 -
12 600 1000 -
13 700 1000 -
unpropped Id 300 1000 -
drained 2d 200 600 -
3d 200 600 -
Q 400 1200 -
5d 500 1600 -
top-P ped Id 300 1000 -
drained 2d 400 1800 -
3d 300 1000 -
4d 300 1400 -
5d 450 1400 -
4-67
Table4.25 Influenceof meshh-refinement(SeriesIII) on wall bendingmoment
Case Set NNqj NN99 Error Comments
unpropped 1 550 1600

3 570 1200 - (M).. +ve
4 550 1700 -
5 550 1800 -
12 550 600 -
13 850 1400 -
top-propped 1 550 1600
3 470 1600 -
4 500 600 -
5 500 600 -
12 550 1700 -
13 500 1000 -
bottom- 1 900 1100 - (M),. +ve
propped 2 900 1200 -
4 900 1100 -
5 900 1100 -
12 900 1200 -
13 900 1100 -
unpropped Id 550 1800 -
drained 2d 550 1900 -
3d 500 800 - (MI. +ve
4d 500 1900 -
5d 480 1900 -
top-propped Id 500 1900 -
drained 2d 450 1000 -
3d 450 600 -
4d 500 1900 -
5d 500 600 -
Table4.26 Influenceof meshh-refinement(SeriesIII) on excavationheave
Case Set NNgs NN99 Error Comments
unpropped 1 300 600

undrained 2 300 600
3 250 600
4 200 600 -
5 200 600 -
12 250 600 -
13 300 1400 -
/contd
4-68
Table4.26 Influenceof meshh-refinement(SeriesIII) on excavationheave(con1d)
Case Set NNgs NN99 Error Comments

top-propped I <100 600 -
3 250 600 -
4 200 800 -
5 200 600 -
12 250 600 -
13 300 600 -
bottom- 1-5 <100 500 -
propped 12 <100 500
unpropped 1-5 <100 300 -
drained
top-propped 1-5 <100 300 -
drained
Table 4.27 Influence of mesh h-refinement (Series III) on ground surface movement
settlement(-S) heave(+S)
Case Set NNgs NN99 Error NN9!i NN99 Error
unpropped 1 500 600 + 350 500 -

undrained 2 500 600 + <100 <100 -
3 900 1000 + <100 500 -
4 500 600 + <100 500 -
5 500 600 + 300 550 -
12 600 1000 +/- 800 1000 +
13 900 1600 + 800 1400 +
top-propped 1 400 1000 + <100 500
undrained 2 500 800 + <100 600
3 500 800 + <100 600
4 500 800 + <100 500
5 500 600 + <100 500
12 500 1200 800 900
13 <100 900 950 1700
bottom- 1 400 500 + <100 500
propped 2 500 600 + <100 300
undrained 3 500 1200 + <100 400
4 500 600 + <100 500
5 500 600 + <100 500
12 500 600 <100 300
13 400 1300 200 1500
unpropped 1-5 (nodownward
movement) <100 <100 +
drained
top-propped 1-5 (nodownward
movement) <100 <100 +
drained
4-69
Table4.28 Influenceof meshr-refinement(SeriesI) on wall displacement
Case Set LR95 LR99 Error Comments
unpropped I <1 4
undrained 2 <1 4
3 2 8
4 <1 2
5 <1 1
12 <1 6
13 16 27
top-propped I <1 4
undrained 2 <1 6
3 <1 3
4 <1 4
5 <1 I-
12 1 8-
13 <1 7-
unpropped 1-5 <1 <1 -
drained
top-propped 1-5 <1 <1 -
drained
NB: <1 denotes5,. alreadywithin ±5% (or :L]Yo)of&,,., at a lengthratio of unity
Table 4.29 Influence of mesh r-refinement (SeriesI) on wall bending moment
Case Set LRgs LR99 Error Comments
unpropped I <1 10
undrained 2 <1 <1
3 2 6 (M). +ve
4 <1 2 +
5 <1 I +
12 <1 18
13 15 25
top-propped I <1 10
undrained 2 1 6
3 <1 5
4 <1 10
5 <1 7
12 1 10
13 1 17
runpropped
l 1-5 <1 <1 varies
a ined
t op propped
z 1-5 <1 <1 varies
drained
NB: <1 denotesM.. alreadywithin =LJ%(or ±1Yq)ofM*mavat a lengthratio of unity
4-70
Table4.30 Influenceof meshr-refinement(SeriesI) on excavationheave
unpropped I <1 <1 +

undrained 2 <1 <1 +
3 <1 <1 +
4 <1 <1 +
5 <1 <1 +
12 <1 2 +
13 <1 10
top-propped I <1 <1 +
undrained. 2 <1 <1 +
3 <1 <1 +
4 <1 <1 +
5 <1 5 +
12 <1 I +
13 <1 5 +
unpropped. 1-5 <1 <1 no discernible variation
drained
top-propped 1-5 <1 <1 no discernible variation
drained
NB: <1 denotesV.,. already within +5% (or::L1Vq)of V*., at a lengthratio of unity
Table 4.31 Influence of mesh r-refinement (Series I) on ground surface movement
settlement heave
Case Set LRgs LR99 Error LRgs LR" Error
unpropped 1 <1 <1 + <1 <1 +
undrained 2 <1 5 <1 5
3 7 10 + 2 5
4 <1 10 <1 12 +
5 <1 15 <1 13 +
12 <1 11 <1 21 +
13 20 28 20 30 -
top-propped I <1 <1 <1 <1 -
undrained 2 <1 5 <1 <1 -
3 <1 14 22 30 -
4 <1 <1 <1 <1 +
5 <1 <1 + <1 10 +
12 <1 8 <1 13
13 <1 15 14 20
unpropped 1-5 (nodownward
movement) <1 <1
drained
movement) <1 <1
drained
NB: <1 denotesS. already within :L5%(or Yq)

of S*nm, at a length ratio of unity
-+I
4-71
Table4.32 Influenceof meshr-refinement(SeriesH) on wall displacement
unpropped. 1 2 10 -
undrained 2 2 10 -
3 5 15 -
4 1 10 -
5 <1 5
12 2 11 -
13 22 30 -
undrained 2 3 14 -
3 2 11
4 2 12 -
5 2 12 -
12 4 14 -
13 2 7-
unpropped 1-5 <1 2-
drained
top-propped 1-5 <1 3-
drained
Table4.33 Influenceof meshr-refinement(SeriesH) on wall bendingmoment
Case Set LRgs LR99 Error Comments

unpropped 1 2 7
undrained. 2 2 4
3 3 7
4 <1 7
5 <1 7
12 3 12 -
13 24 30 -
undrained 2 4 15 -
3 2 11
4 3 15 -
5 2 13 -
12 6 20 -
13 4 15 -
unpropped Id <1 8-
drained 2d <1 10 -
3d 1 15 -
4d <1 4-
5d <1 I-
top-propped Id <1 6-
drained 2d <1 5-
3d <1 4-
4d <1 3-
5d <1 2-
4-72
Table4.34 Influenceof meshr-refinement(SeriesII) on excavationheave
unpropped I <1 <1

undrained 2 <1 5
3 <1 10
4 <1 <1
5 <1 5
12 <1 6-
13 <1 16 -
top-propped I <1 <1 -
undrained 2 <1 <1 -
3 <1 4
4 <1 <1
5 <1 5
12 <1 2-
13 <1 18 -
unpropped 1-5 <1 <1 -
drained
top-propped 1-5 <1 <1 -
drained
Table4.35 Influenceof meshr-refinement(SeriesH) on groundsurfacemovement
settlement heave
Case Set LR95 LR99 Error LR9!; LR99 Error
unpropped 1 12 25 - <1 18
undrained 2 15 28 - 2 4
3 15 25 . 3 8
4 13 27 - <1 18
5 13 27 - <1 18
12 15 28 - <1 25
13 27 31 - 18 30
top-propped 1 13 27 <1 15
undrained 2 15 28 - <1 2
3 15 28 - 3 8-
4 14 27 - <1 18 -
5 14 27 - <1 20 -
12 15 28 - <1 22 -
13 16 28 61 28 -
-
unpropped 1-5 (nodownwardmovement) <1 <1 -
drained
movement) <1 <1 -
drained
I
4-73
Table
4.36 Influenceof pseudop-refinement(LSQsto LSTs) on relativelyfine mesh
(x8y8e6)- unpropped.wall / undrainedloading
r Set LSQ LST (016

difl)
et
displacement8.,, (mm)
displacen
waH
1 -18.74 -19.01 (1.4)
2 -12.68 -12.84 (1.2)
3 45.04 -45.34 (0.7)
4 -18.47 -18.67 (1.1)
5 -12.50 -12.61 (0-9)
12 -16.44 -16.90 (2.8)
13 -141.50 -153.30 (8.3)
waUbendingmoment±N4.. (kN-m)
1 437.4 -441.8 (1.0)
2 (0.5)
-315.3 -316.8
3 397.9 394.7 (-0.8)
4 446.4 (0.4)
-448.1
5 (0.4)
-379.9 -381.6
12 (2.4)
-340.9 -348.8
13 619.9 664.9 (7.3)
excavationheave V.,.,, (mm)

1 41.21 41.31 (0.2)
2 14.03 14.07 (0.3)
3 22.74 22.74 (0.0)
4 41.00 41.33 (0.8)
5 26.90 27.15 (0.9)
12 16.91 17.26 (2.0)
13 81.80 83.59 (2.1)
. ..........
........
ground surface movement S,. (mm)
settlement (-S) heave (+S)

1 (0.2) 16.73 16.76 (0.2)
-11.86 -11.88
2 -3.88 (0-0) 4.04 4.06 (0.5)
-3.88
3 (0.2) 6.61 6.58 (-0..5)
-10.68 -10.70
4 (0.5) 16.19 16.43 (1.5)
-11.42 -11.48
5 -6.73 (0.7) 10.13 10.30 (1.7)
-6.78
12 -4.71 4.76 (1.1) 3.00 2.90 (-5-2)
13 (13-3) 9.00 9.35 (3.8)
-106.00 -120.10
4-74
Table4.37 Influenceof pseudop-refinement(LSQsto LSTs) on relativelycoarsemesh
(x8y8b4)- unpropped. loading
wall / undrained.
Set LSQ LST % diff
wall displacement 5. (mm)
1 -17.20 -18.30 (6.4)

2 -11.64 -12.39 (6.5)
3 -42.16 -44.66 (5.9)
4 -17.19 -18.12 (5.4)
5 -11.69 -12.22 (4.5)
12 -14.77 -16.24 (10-0)
13 -110.00 -139.60 (269)
..........
wall bending moment ±A4.,,. (kN-m)
1 -361.9 -397.2 (9.8)
2 -238.8 -274.0 (14-7)
3 310.8 359.0 (15-5)
4 -371.2 -404.5 (8-9)
5 -316.3 -343.5 (8.6)
12 -281.5 -323.1 (14-8)
13 360.9 574.5 (59-2)
.... . .......
........ ................ ........................
. .......
...................
excavation heave V. (mm)
1 40.46 41.21 (1-8)

2 13.76 14.03 (1-9)
3 22.18 22.69 (2-2)
4 40.48 41.15 (1-6)
5 26.52 26.97 (1-7)
12 16.58 16.91 (2-0)
13 77.89 79.78 (2-5)
...............
. .............. ............
ground surface movement S. (mm)
settlement (-S) heave (+S)

1 (-1.3) 16.80 16.81 (0-1)
-12.02 -11.86
2 4.05 (-4-5) 4.14 4.08 (-1.5)
-3.87
3 (-13.8) 6.70 6.65 (-0.8)
-12.19 -10.51
4 (4.6) 16.23 16.47 (1-5)
-12.02 -11.47
5 (-S.O) 10.05 10.29 (2-4)
-7.14 -6.78
12 (-0.2) 3.59 3.13 (-12.8)
-4.67 -4.66
13 (51.7) 10.22 9.56 (-65)
-65.91 -100.00
4-75
Notes:
1. wall elements highlighted by grey shading
2. formation level (FL) indicated by bold line
Fig 4.1 ParentmeshxI Oy10 usedin boundarylocationanalyses
4-76
0 2H 4H 6H 8H IOH
2W
4W
6W
8w
low
mesh XY NN NEL
x2y2 2H 2W 380 342

x4y4 4H 4W 506 462
x6y6 6H 6W 600 552
x8y8 8H 8w 650 600
xl Oyl 0 10H low 702 650
X= distance from wall to remote vertical boundary

Y= distance from base of excavation to bottom boundary
H= retained height of wall
W= half-width of excavation
NN= number of vertex nodes
NEL = number of elements
Fig 4.2 Different mesh sizesused in boundary location analyses(constant global

aspect ratio)
4-77
(a) Unproppedcantilever
prop placed prior

to excavation
(b) Top-proppedcantilever
prop placed prior

to excavation
(c) Bottom-propped cantilever
both props placed

prior to excavation
(d) Doubly-proppedcantilever
bottom prop installed after excavation -
prior to removal of top prop
cantilever
Fig 4.3 Wall propping arrangementsused in boundary location analyses
4-78
z E,, (MPa)
(M) Set 1 Set 2 Set 3
0 50 50 0
H 50 90 40
IIIL 50 130 80
z132 <-- set
degree of nonhomgeneity X= (EH- Eo) / EH
NB: same values of X apply for drained elastic parameters (sets 1d, 2d and 3d)
Fig 4.4 Definition of degree of nonhomogeneity for an embeddedretaining wall
a) rough remote boundaries b) smooth remote boundaries
Fig 4.5 Boundary conditions used in boundary location analyses
4-79
0.0
-10.(
-20.C
-30.0
-40.0
P4
P-4
-50.0
0.04.0 8.0 12.0 16.0

Depth below top of wall Cm)
X292 * A94 + X696 + X899 X10910
(a) cantileverwall/ undrained

/ elastic-perfectly
plasticsoil(set11)
0.1
E -101(
E
%00
-20.C
-30.0
P-4
-40.0
-50.0,
0.0 4.0 8.0 12.0 16.0
Depth below top of wall (n)
X292 * A94 + X696 + x8ye -qL X10910
(b) top-proppedwall / undrained/ elastic-perfectlyplasticsoil (set 13)
Fig 4.6 Influenceof remoteboundarylocation on wall displacement

(SeriesA)
selectedprofiles
4-80
I
O.c
-5.0
(V
-10.0
. pf
-15.0
13
P.4
1-4
-20.0
0.0 408.0 12.0 16.0

DZth below top of wall (n)
-EF X292 * AY4 + X696 + X898 A.- X10910
(c) bottom-proppedwall / undrained/ elastic-perfectlyplasticsoil (set 13)
£ -5.0
)A
. F(
la
pq -15.0
P.i
0.0 4.0 8.0 12.0 16.0

x2Y2 * A94 + x6Y6 + x8Y8 XIOYIO
(d) altemately-propped wall / undrained / elastic-perfectly plastic soil (set 12)
Fig 4.6 Influenceof remoteboundarylocationon wall displacement

(SeriesA)
selectedprofiles(contd)
4-81
-2.1
-4.0
u -6.0
m
-8.0
P-1
-10.0
0.0 4.0 8.0 12.0 16.0

Depth below top of wall Cn)
-a X292 * A94 + A96 + Aye X10YI0
cantileverwall / drained/ elasticsoil (set4d)
51C
%OF
0.0
4J
-2.5
m
IA
.1.4 -5.0
v
P-4
P4
m
-7.5
0.0 4.0 8.0 12.0 16.0

-8- x2Y2 * x4Y4 + X696 -4- X8Y8 X10Y10
(f) bottom-proppedwall / drained/ elasticsoil (set I d)
Fig 4.6 Influenceof remoteboundarylocation on wall displacement(SeriesA)

4-82
0.0 ......... . .......
41
r.
. .... . ...... . ... .........
-5.0 ..
. ............ . .........
-10.0 t ---------------------
.......... .....
P.i -15.0
-20.0
0.0 4.0 8.0 12.0 16.0
X292 * A94 -#- X696 + x8ge X10910
(9) bottom-proppedwall / drained/ elastic-perfectlyplasticsoil (set 12d)
0.0
III1EE
II1
u -2.0
m
P-1 -4.0
P-4
m
3
-A-n
0.0 4.0 8.0 12.0 16.0
4 x292 * AY4 -0- AY6 + x8ge X10910
(h) doubly-propped wall/ drained/ elastic soil (set Id)
Fig 4.6 hffluenceof remoteboundarylocationon wall displacement(SeriesA)

4-83
201
.x
0.0 ...... . .... ........ ..... .................... ............
............... .................. ..... . ... ..................... . ....... ...
. ........ - -- -----
. ........ .. --- ------
.C
P4
-400.0
n
-5nn.
0.0 4.0 8.0 12.0 16.0
X292 * AA + x696 + x8ge X10YI0
(a) cantileverwall / undrained/ elasticsoil (set 1)
501
.X
200.0
100.0
c 0.0
-100.0 - ----- .......
-20010
0.0 8.0 12.0
4.0 16.0
Depth below top of wall (M)
-0- X292 * AY4 -#- AY6 + X898 XJOY10
(b) cantileverwall undrained/ elasticsoil (set3)
Fig 4.7 Influenceof remoteboundarylocationon wall bendingmoment(SeriesA)

selectedprofiles
4-84
20C
O.C
z
X
-200.0
0
-400.0 ...............
..
. pq
-600.0 . ....
.....
41
a
P14 -800.0
P4
-1000.0
0.0 4.0 8.0 12.0 16.0
x2y2 * AY4 -+- x6y6 + X8Y8 xID910
(c) top-proppedwall / undrained/ elasticsoil (set3)
200
-200.0
-400.0
E
0
z -600.0
.pq -800.0
-1000.0
P-4
-1400.0' 111
0.0 4.0 8.0 1210 16.0
Depth below top of wall (m)
4ý x2Y2 * A94 + AY6 + X898 XINIC
(d) top-proppedwall / undrained/ elastic-perfectlyplasticsoil (set 13)

4-85
?UU.U
600.0 .......... ..........

. ........
. ...............................
..... .......
.X 500.0
400.0 ..... ........... . .... ......... .........................................

...................
...............
300.0 . .......
200.0
100.0
-100.0.0 12.0
408.0 16.0
_EFX292 * AY4 -#- AY6 + X8Y8 -,v- XIOYIO
(e) altemately-propped wall / undrained / elastic-perfectly plastic soil (set 13)
Mulu
600.0
500.0
400.0
300.0
200.0
100.0
p4 0.0
-inn-n 12.0
0.0 4.0 8.0 16.0
-& x2Y2 * AA + xGYfi + x8ye -V- X10Y10
(f) cantileverwall / drained/ elastic-perfectlyplasticsoil (set 12d)

4-86
14UU.L
1200A
r
Z
low. v
.X
%ýo
-W 800.0
c
0 600.0
400.0
13
c
0) 200.0
J2
P-1
P4 0.0
m
2
-200.0,
0.0 4.0 8.0 12.0 16.0
-[ý x292 * A94 -0- X696 + x8ge X10YI0
(g) bottom-proppedwall / drained/ elasticsoil (set3d)
6D
400A
%dp
200.0
0.0
-200.0
-400.0- 11
0 8.0 12.0 16.0
.04.0
x2Y2 * x4Y4 +"+ X898 XIDYID
(h) doubly-proppedwall / drained/ elasticsoil (set 4d)

4-87
100
801C
m
60.0
40.0
m
D,
m
u
x 20.0
LU
L-
0.0
-8.0 -6.0 -4.0 -2.0 0.0
Distance from wall Cm)
x2y2 A94 + xfiy6 + x8ge X10910
(a) cantilever wall / undrained / elastic-perfectly plastic soil (set 13)
50A
40.C
E
a, 30.0
a,
C
0 20.0
ci 10.0
x
UI
CIO
Distance fron wall Cn)
4 x2Y2 A94 + x6y6 + x8ge xlDulO
(b) aftemately-propped wall / undrained / elastic soil (set 1)
Fig 4.8 Influenceof remoteboundarylocationon excavationheave(SeriesA)

selectedprofiles
4-88
V
80.1 --- .........

--------------- ............
60A .......... ............. .............. .......... ...................... .....
c
0 40.0
.pq - --------- ------
----------------------
- ---- ------ ............
.1i
u 20.0
x
w
0.0
-8.0 -6.0 -4.0 -2.0 0.0
Distance from wall Cn)
-a- x2y2 AY4 + x6y6 + X898 JgL X10YI0
(c) top-proppedwall / drained/ elasticsoil (set 5d)
80.C
6010
40.0
20.0 ..... . ...........................

. ........ . .....................
. ...
uj
n.n
-8.0 -6.0 -4.0 -2.0 0.0
Distance from wall (n)
X2Y2 A94 + X6Y6 + Aye XICY10
(d) doubly-propped wall / drained / elastic-perfectly plastic soil (set 12d)
Fig 4.8 Influenceof remoteboundarylocationon excavationheave(SeriesA)

4-89
lu. L
P-4
41
.40
10
91
w
-5.0
0.0 20.0 40.0 60.0 80.0

-8- X292 * A94 + X696 + x8y8 -7- X10YI0
(a) cantileverwall / undrained/ elasticsoil (set2)
10.1
CA
-10.0
z -20.0
al
-30.0
IA
w
-40.0
X rangetd.showmre detail nearWID
: reduced
-50.0
0.0 4.0 8.0 12.0 16.0

Distance from wall (m)
x2Y2 x4Y4 + *6 + XBY8 XIMO
(b) altemately-propped wall / undrained / elastic-perfectly plastic soil (set 13)
Fig 4.9 Influence of remote boundary location on ground surface settlement (Series
A): selectedproffles
4-90
2
(MB:reducedX rangeto showdetail Nýarwall)
20.U
15.0 .......... .............
pq 10.0 .... ........... ... ... . .... .... ....................... .........

. ................................................................
3.0 ......................
w
u
19
4.
0.0
-s.n0.0 4.0 8.0 12.0 16.0

Distance fron wall (n)
-0- X292 x494 + X696 + X898 XIMO
(c) top-propped wall / drained / elastic-perfectly plastic soil (set 12d)
6U.U
50.0
40.0 . ....... ....... .........
30.0 ................... ..... ......
V
20.0 ------ .................. ....... ........
U 1010 . ..... .... ... .......
L
0.0 ---------------
-- --- ................... ---------- ......
:zero settlement at bmidarv
RHS
-10.0,
0.0 20.0 40.0 60.0 80.0
x292 A94 + x6Y6 + x8ve -7- X10Y10
(d) bottom-proppedwall / drained/ elasticsoil (set 5d)
Fig 4.9 Influenceof remoteboundarylocationon ground surfacesettlement(Series

A) : selectedprofiles(contd)
4-91
0 IOH
2W
4W
Series B
(variable depth
6W of mesh)
8W
low
0 2H 4H 6H 8H 10H
0
Series C
(vadable width
of mesh)
low
Series B Series C
mesh NN NEL mesh NN NEL
xlOy2 520 475 x2ylO 513 468

xlOy4 598 550 x4ylO 594 546
xlOy6 650 600 x6ylO 648 598
XloY8 676 625 x8ylO 675 624
xloylo 702 650 xloylo 702 650
Fig4.10 Different meshsizesusedin boundarylocationanalyses-variable global

aspectratio
4-92
0 6H
0 -di
2W
4W
Series D
(variable depth
of mesh)
6W
8w
low
0 2H 4H 6H 8H 11014
0
Series E
(variable width
of mesh)
6W
Series B Series C
mesh NN NEL mesh NN NEL
x6y2 480 437 x2y6 475 432

x6y4 552 506 x4y6 550 504
x6y6 600 552 x6y6 600 552
x6y8 624 575 x8y6 625 576
x6ylO 648 598 XIOY6 650 600
Fig4.11 Different meshsizesusedin boundarylocationanalyses-variable global

aspectratio
4-93
0.0
-5.0
-10.0
-15.0
-20.0
0-4
-25.0
-30.0 1
0 8.0 12.0
.04.0 top (n)
Depth below of wall
X1092 * 4094 + X1096 + X1098 -q- X10YI0
(a) wall displacement: cantileverwall / undrained/ elasticsoil (set 1)
500.0
400.0
.X 300.0
%0
200.0
(U
100.0
0.0
-100.0
-200.0
-300.0 -- 8.0 12.0 16.0

0.0 4.0
XIOY2 * xIOY4 + X1096 + XIOY8 X10YI0
(b) wall bending I /

moment:cantileverwall undrained elasticsoil (set 3)
Fig 4.12 Influenceof remoteboundarylocationon key output value(SeriesB)

selectedprofiles
4-94
r4l
I
20.0
t
£
w 15.0
CA
0 10.0
4,
ID
a
C 310
w
CIO,
-8.0 -6.0 -4.0 -2.0 0.0
-EFXIOY2 x1m + XIOY6 + AM XIDY10
(c) excavation heave : cantilever wall / undrained / elastic soil (set 3)
2.5
(U 0.0
E
w
P-4
-W
41
IV -2.3
IA
41
u
m
14.
L -510
-7.5 'I
0.0 10.0 20.0 30.0 40.0
-EFXIOY2 4094 + XIOY6 + XIOY8 XIOYIO
(d) ground surface settlement : cantilever wall / undrained / elastic soil (set 2)
Fig 4.12 Influenceof remoteboundarylocationon key output value(SeriesB)

4-95
-5.C
-10.0 ............... .7
u
m
-13.0 ....... ........ . .........
. ......... ...
.13
P4
P4
-20.0
-25.0
0.0 4.0 8.0 12.0 1610
-a- X692 * A94 + x6Y6 + X698 x6ylo
DUUA
400.G
-W 200.0
c
100.0
c 0.0
.pq
v
c
-100.0
pq
-200.0
-300.0111
CIO 4.0 8.0 12.0 16.0
AY2 * x694 + X696 + X698 -v- X6910
wall bending moment: cantilever wall / undrained,/ elastic soil (set 3)
Fig 4.13 Influenceof remoteboundarylocationon key output value(SeriesD)

selectedprofiles
4-96
25
Pi 20.0
w 15.0 .... . .....
m
C
C
0
"M
10.0 . ..........
U 5.0
x
LU
n.n
8.0 -6.0 -4.0 -2.0 0.0
ý3- AY2 * A94 -0- x6y6 + x6ya -v- X6Y10
(c) excavationheave: cantileverwall / undrained/ elasticsoil (set 3)
5.0
£ 2.5
4,
C
ci 0.0
ci
4
4'
ci
U.
' -2.5
ci
ci
'4-
I. -3.0
-7.5 L-
0.0 10.0 20.0 30.0 49.0
x6yz * X6Y4 + x696 + X6Y8 -v- X6Y10
(d) ground surfacesettlement: cantileverwall / undrained/ elasticsoil (set 2)
Fig 4.13 Influenceof remoteboundarylocation on key output value(SeriesD)

4-97
I
-5.C
£
-1010
P4
P4 -20.0
0.0 4.0 8.0 12.0 16.0

x296 * A96 + x696 + x896 X1096
3uu.t
400.C
300.0
.X
%0
200.0
al
E
0 100.0
0.0
-100.0
P-4
P-1 -200.0
m
3
-300A
0.0 4.0 8.0 12.0 16.0
-EF X296 * MY6 + AY6 + X896 -VL X1096
(b) wall bendingmoment:cantileverwall / undrained/ elasticsoil (set 3)
Fig 4.14 Influenceof remoteboundarylocationon key output value(SeriesE)

selectedprofiles
4-98
2
20A ................ ... ............ ....

E
C,
15.0 .......... . ...... ..
C,
0 10.0 ....... ..... ............
C.' 5.0 ........................ .
n.n
8.0 -6.0 -4.0 -210 0.0
-& X296 A96 + X696 + X896 -IgLY1096
(c) excavationheave: cantileverwall / undrained/ elasticsoil (set3)
2.5
z at far RHSboundary
.W0.0
c
-2.5
IA
41 -5.0
u
-7.3
0.0 10.0 20.0 30.0 40.0

x2Y6 AY6 + x6Y6 + XBY6 XIM
(d) ground surfacesettlement: cantileverwall / undrained/ elasticsoil (set 2)
Fig 4.14 Influenceof remoteboundarylocationon key output value(SeriesE)

4-99
Ll
-1010
-20.0
-30.0
Ic
m4
-40.0
-50.0L-
0.0 4.0 8.0 12.0 16.0
I-Smooth -0- 2-Rough + 2-Smooth -qL Hough 3-Smooth
toN
wall displacement: cantileverwall / undrained/ elasticsoil (sets 1-3)
500.0
.X 250.0 ......... .........

W
0.0
-250.0
-rirkA n
0.0 4.0 810 12.0 16.0
I-Rough I-Smooth 2-Rough + 2-Smooth -y- Hough 3-Smooth
-0-
(b) wall bending moment: cantilever wall / undrained / elastic soil (sets 1-3)
Fig 4.15 Influenceof remoteboundaryconditionon key output value,usingmesh

x2y2 : selectedprofiles
4-100
25.0
20.0 ............
.............. . ........
1510 .................... ........... ........

............... .
Z 10.0 ...... .......... .......... ...... . .... . ............ ... . .............
C..............
.............
"I'll,
. ...............
...............
.pq
----..........
----- ..... ........
1U
WX 0.0 .......
-5.0 1 ;
8..0 -6.0 -4.0 -2.00.0
-& I-Rough I-Smooth -0- 2-Rough + 2-Smooth -v- 3-Rough 3-Smooth
(c) excavation heave : cantilever wall / undrained / elastic soil (sets 1-3)
5.0
0.0
-W
c -5.0
-10.0
vi
(u -15.0
u
to
4.-
-20.0
0.0 4.0 8.0 12.0 16.0

I-Rough 1-1mooth 2-Raigh + Mmooth -v- 3-Raugh -a, 3-INGth
-B. -4-
(d) :
ground surface settlement cantilever wall / undrained / elastic SOU(sets 1-3)
Fig 4.15 Influence of remote boundary condition on key output value : mesh x2y2
selectedprofiles (contd)
4-101
oý
-10.1
-20.C
u
m
-30.0
Ul
.la
pq
P-1 -40.0
P-4
m
3
L-
-50.0
0.0 4.0 8.0 1210 16.0
-EF I-RMh I-Smooth 2-Rough + 2-1mooth Hough 3-Smooth
-0- -v-
ON
wall displacement: cantileverwall / undrained/ elasticsoil (sets 1-3)
3uu. t
.X 250.0
%WO
0.0
-250.0
L-
-500.0
0.0 4.0 8.0 12.0 16.0
I-Pb* i-shooth
-EF -#- 2-NK* + 2-Smooth -vý 3-Rough 3-Imooth
(b) wall bending / elasticsoil (sets 1-3)

moment:cantileverwall / undrained.
Fig 4.16 Influenceof remoteboundaryconditionon key output value,usingmesh

xlOyIO : selectedprofiles
4-102
61
50.C
40.0
30.0
.pq
20.0
u
x 10.0
w
-4.0 -2.0 0.0

I-Rough I-Smooth + Hough + 2-Smooth Hough 3-Smooth
(c) excavationheave: cantileverwall / undrained/ elasticsoil (sets1-3)
31
2010 ..............
4.'
C
w 10.0
4.'
w 0.0 ----------
m
w
ci
. ......... ......... ... .... . .........
-10.0
On n
0.0 4.0 8.0 12.0 16.0
-EF I-RwO I-Smoth 2-PA* + 2-Smooth Haigh 3-Smooth
-0-
(d) groundsurfacesettlement: cantileverwall / undrained/ elasticsoil (sets1-3)
Fig 4.16 Muence of remoteboundaryconditionon key output value,usingmesh

xI Oy10 : selectedprofiles(contd)
4-103
(a) originalmesh
(b) h-refinernerit (c) r-refinement (d) p-refinement
Notes:
h-refinement extra elements incorporated by subdivision; original mesh is

contained within the refined mesh
2. r-refinement element size adjusted, whilst keeping total number of elements

constant - topological equivalence maintained
3. P-refinement order of elements (and hence number of degrees of freedom)

increased,whilst keeping total number of elements (and their size) constant
Fig 4.17 Various types of meshrefinement
4-104
(a) mesh x8y8aO (b) meshx8y8bO
LLL-1
(C) meshx8y8cO (d) meshx8y8dO
Note: dashed lines indicate most recent mesh sub-divisions
Fig 4.18 SeriesII h-refinement meshesx8y8aO- x8y8hO
4-105
meshx8y8eO (0 mesh x8y8fU
I
.mI,
(g) mesh x8y8gO (h) meshx8y8hO
Note: dashed lines indicate most recent mesh sub-divisions
Fig 4.18 SeriesH h-refinementmeshesx8y8aO- x8y8hO(contd)
4-106
1.3
. po
a
P-4
pq
0.8
aP4
0.7
4J
m
m
016L
0 200 4ý 600 goo 1000 1200 1400 1600 1800 2000
Total Number of Modes
(a) wall displacement cantilever wall undrained elastic soil (set 4)
1.0
E 0.8
P4
P4
0.6
E
0.4 . ........ .............
0.2
cc
0.0
0 200 400 600 800 1000 1200 iiý 16M 18,00 2000
Total Number of Modes
-Ve Nowt * +yehommt
(b) wall bendingmoment:cantileverwall / undrained/ elasticsoil (set 2)
Fig 4.19 Influenceof meshgrading(SeriesH h-refinement)on key output values

selectedconvergenceplots
4-107
IA
Ul
. .......... . ..
u
x
LLI
1.0 ......... ....... . ......... . .............. .
ifo
E
0.9
ic
n.20
200 400 600 800 1000 1200 1400 1600 1800 2000
Total Number of Nodes
(c) excavationheave: cantileversoil / undrained/ elasticsoil (set 1)
1.2
E
dl
4.
0 0.9 . .....
. ...... ...... ....
....
.Pq
0: 0.8 1--L
0 200 400 6M Boo 1000 1200 1400 1600 1000 2000
Total Number of Hodes
fue(Settlement) * -ve (Heave)
(d) ground surface settlement : cantilever soil / undrained / elastic soil (set 2)
Fig 4.19 Influenceof meshgrading(SeriesII h-refinement)on key output values

selectedconvergenceplots
4-108
(a) meshx8y8a (b) mesh x8y8b
AIIII
AN
, Jill
(C) mesh x8y8c meshx8y8d
Millill IIIIII
(e) mesh x8y8e (1) mesh x8y8f
Fig 4.20 SeriesHI h-refinement meshesx8y8a - x8y8f
4-109
0.
-10.0
0
-1-4
13
P-4
-20.0
L-
-23.0
0.0 4.0 8.0 12.0 16.0
-a- neshx8v8a * neshx8g8f
0.
-5A
41
E
10
u -10.0
m
.pq
V
-15.0
P-4
9-4
m
2
-20.0 L-
0.0 4.0 0.0 12.0 16.0
-FF resh x8g8a * meshx8y8f
(b) top-proppedwall / undrained,

/ elasticsoil (set3)
Fig4.21 Influenceof meshgrading(SeriesIII h-refinement)on wall displacement:

selectedprofiles
4-110
5.0
p..
.ij
Qg
-1e.o
. -15.0
P-1
-20.0
-23.0 L-
0.0 4.0 0.0 12.0 16.0
neshx898a * meshx8y8f
(c) bottom-proppedwall / undrained/ elastic-perfectly

plasticsoil (set 13)
0.0
-W
c
41 -4. C
E
m
-6.0
IA
WI
13
pq
L-
-10.0
0.0 4.0 8.0 12.0 16.0
neshx$sjga * mh x8yor
(d) unproppedwall / drained/ elasticsoil (set4d)
Fig4.21 Influenceof meshgrading(SeriesIII h-refinement)on wall displacement:

4-111
200.
100.1
.X
0.1
-100.(
-200A -- ---------
(9
12
-300.0
P-1
pq
-4nfl. n
0.0 4.0 8.0 12.0 16.0
reshx8yea WA x8yof
(a) cantilever wall / undrained / elastic soil (set 5)
200.(
100A
z 0.0 .........
Jv
-100.0
-200.0
C
_300.0 ------
.0
13
C
-400.0
P-1 -300.0
m
-600.0
0.0 4.0 8.0 12.0 16.0
msh x8y8a rash Xguef
(b) / / (set 12)

top-proppedwall undrained elastic-perfectlyplasticsoil
Fig 4.22 Influenceof meshgrading(SeriesIII h-refinement)on wall bending

moment: selectedprofiles
4-112
1000.0
800.0
X 600.0 .... . .........
400.0
200.0
010 - -------
-
00.0
_200.0
M
-400.0
0.0 4.0 8.0 12.0 16.0
-0- reshxgg$a neshx898f
(c) bottom-propped wall / undrained / elastic soil (set 2)
700.1
.X
%.f
400.t
15
c 100.0 . ..... . ......
gi
12
Pf 0.0
4.0 8.0 12.0 16.1

Depth below top of wall Cnj
msh x8v8a meshx8q8f
unproppedwall / drained/ elasticsoil (set3d)
Fig 4.22 Influenceof meshgrading(SeriesIII h-refinement)on wall bending

moment: selectedprofiles (contd)
4-113
100.
80.1 .........
-------
..........
(9
60A
0)
40.1 -------- ---
20.0
0.0 L-
-8.0 -6.0 -4.0 -2.0 0.0
-8- mesh
x898a * x8y8f
mesh
100A
BOA
60.0
w
C
0 40.0
.1
U 20.0
C
L-
0.0
-8.0 -6.0 -4.0 -2.0 0.0
roeshx8y8a * meshx8y8f
(b) top-proppedwall / drained/ elasticsoil (set I d)
Fig4.23 Influenceof meshgrading(SeriesIII h-refinement)on excavationheave:

selectedprofiles
4-114
41
Z
ai M
. ........ . ..
. .........
-20.0
0.0 16.0 32.0 48.0 64.0
-9- neshx8g8a meshx8y8f
(a) cantileverwall I undrained/ elasticsoil (set4)
20.0
15.0
10.0
5.0
0.0
-3.0
u
-1010
L
-15.0
-20.0
0.0 16.0 32.0 48.0 64.0
-& mshx8y8a * mesh
x8y8f
(b) top-propped wall / undrained / elastic-perfectly plastic soil (set 13)
Fig 4.24 Influenceof meshgrading(SeriesIII h-refinement)on ground surface

movement:selectedprofiles
4-115
20.0
0.0
L-
-3.0
0.0 16.0 32.0 48.0 64.0
-EFmh * x8g8f
mesh
(C) cantileverwall / drained/ elasticsoil (set3d)
Fig 4.24 Influenceof meshgrading(SeriesIII h-refinement)on groundsurface

movement: selectedprofiles(contd)
each new element contributes 3 new nodes
Fig 4.25 Relationshipbetweennumberof nodesandnumberof elementsin a large

meshof LSQ elements
4-116
PQ
R
S
Segment Length (m) No. of divs

- ------ ----
p 8.0 6
Q 64.0 14
R 8.0 6
S 8.0 7
T 56.0 8
mesh Mp MQ MR MS MT
----- ----- ----- -----

x8y8el 1.0 1.0 1.0 1.0 1.0
x8y8e2 1.2 2.0 1.2 1.2 2.0
x8y8e3 1.4 4.0 1.4 1.4 4.0
x8y8e4 1.6 8.0 1.6 1.6 8.0
x8y8e5 1.8 12.0 1.8 1.8 12.0
x8y8e6 2.0 16.0 2.0 2.0 16.0
x8y8e7 3.0 24.0 3.0 3.0 24.0
x8y8e8 3.0 32.0 3.0 3.0 32.0
Fig 4.26 Supermeshfor Series I r-refinement for embeddedwall analysis : key

gridlines, dimensions and divisions
4-117
(a) meshx8y8el (C) mesh x8y8e3
(b) mýshx8y8e2 (d) meshx8y8e4
Fig 4.27 SeriesI r-refinementmeshesx8y8el - x8y8e8
4-118
ta*l mesh x8y8e5 (g) mesh x8y8e7
kv)
(f) meshx8y8e6 (h) meshx8y8e8
Fig 4.27 SeriesI r-refinementmeshesx8y8el - x8y8e8 (contd)
4-119
U. 1
£
£
-10.u
. pf
13
P-4
-20.0
............
-25.0
0.0 4.0 8.0 12.0 16.0
-8- wuh x8g8el meshxBgBeB
-10.1
-20.0
(v
-30.0
P4
-40.0
11
-50.0.1
0
.04.0 8.0 12.0 16.0

meshx8y8el * meshxBySeB
Fig 4.28 Influenceof meshgrading(SeriesI r-refinement)on wall displacement

selectedprofiles
4-120
0
41
., 4
13
P-4
-10.0
0.0 4.0 8.0 12.0 16.0

-a- heshx8g8el neshx8q8e8
(c) cantileverwall / drained/ elasticsoil (set4d)

Fig 4.28 Influence of mesh grading (Series I r-refinement) on wall displacement
selectedprofiles (contd)
700.0
600.0
z 500.0
400.0
300.0
. pe
rq 0.0
-inn.n0.0 8.0 12.0 16.0

4.0
neshx8yBel * rmh xByBeB
(a) cantileverwall / undrained/ elastic-perfectlyplasticsoil (set 13)

Fi84.29 Influenceof meshgrading(SeriesI r-refinement)on wall bendingmoment:
selectedprofiles
4-121
20
1 0.( ........... .......
-400.0
13
c
-600.0
m
-800.0
0.0 4.0 810 12.0 16.0
meshxOgOel meshxSggeO
top-propped wall / undrained/ elasticsod(set4)
20
.XD. C
%0
E
-100.0
-200.0
V . 30040 - -------- ------ -- - -------- -
P-i -400.0 . .........

P4
-500.01
0.0 4.0 8.0 12.0 16.0
heshx8y8el meshx8goes
(C) cantileverwall / drained/ elasticsoil (set I d)
Fig 4.29 Influenceof meshgrading(SeriesI r-refinement)on wall bendingmoment

4-122
51
40.C
%OP
30.0
c
0
. P4 20.0
x 10.0
w
0.01 11
8.0 -6.0 -4.0 -210 0.0
meshx8y8el neshx8y8e8
(a) unproppedwall / undrained/ elasticsoil (set 1)
100.0
80A
E
60.C
0 40.0
.pq
41
la
:0
19
u 20.0
x
w
0.01 1
-8.0 -6.0 -4.0 -2.0 0.0
Distance from wall Cn)
meshx8g8el meshx898e8
(b) / elastic-perfectlyplasticsoil (set 13)

top-proppedwall / undrained.
Fig4.30 Influenceof meshgrading(SeriesI r-refinement)on excavationheave:

selectedprofiles
4-123
6.0
4.c
I 2.0
C
OJ
ci 0.0
0
cl -2.0
ci
'C
L
-4.0
0.0 16.0 32.0 48.0 64.0

meshx8y8el meshx8gBeO
(a) unproppedwall / undrained/ elasticsoil (set2)
15.1
IM
010
-5.0
(U
u
-10.0
-15.0
-20.0,1 11
0.0 16.0 32.0 U. 0 64.0
-0- iiesh x8y8el * neshx8y8e8
Fig 4.31 Influenceof meshgrading(SeriesI r-refinement)on ground surface

movement: selectedprofiles
4-124
F
p
R
..
Ss.
S
Segment Length (m) No. of divs
p 8.0 2
Q 64.0 7
R 8.0 2
S 8.0 2
T 56.0 5
mesh Mp MQ MR MS MT
x8y8bl 1.0 1.0 1.0 1.0 1.0

x8y8b2 1.2 2.0 1.0 1.0 2.0
x8y8b3 1.5 4.0 1.0 1.0 4.0
x8y8b4 1.9 8.0 1.0 1.0 8.0
x8y8b5 2.4 16.0 1.0 1.0 16.0
x8y8b6 3.0 32.0 1.0 1.0 32.0
Fig 4.32 Supermeshfor SeriesH r-refinement for embeddedwall analysis : key

gridlines, dimensionsand divisions
4-125
(a) meshx8y8bI (b) meshx8y8b2
(C) mesh x8y8b3 (d) mesh x8y8b4
(e) mesh x8y8b5 (f) mesh x8y8b6
Fig 4.3 3 SeriesH r-refinement meshesx8y8b I- x8y8b6
4-126
01
%0
P.-I
a
IA
.pq
13
P.4
-50.0L-
0.0 4.0 8.0 12.0 16.0
mesh
x8y8bI * mh xBvBI)6
0.
%.0
-3.0
u
m
0-4
k2 -7.3
LQ
.pq
13
P14
-10.0
-12.5L- 12.0
0.0 4.0 8.0 16.0
-0- RmhxSuBU * meshx8981A
Fig 4.34 Influenceof meshgrading(SeriesII r-refinement)on wall displacement

selectedprofiles
4-127
0.
1%
£
£
-10.0
u
m
IA -15.0
.rj
13
P4
P4 -20.0
m
3
0.0 4.0 8.0 12.0 16.0

-8- meshx8y8U meshx8y8b6
(c) cantileverwall/ drained/ elasticsoil (set 1d)

Fig4-34 Influenceof meshgrading(Series1E1 r-refinement)on wall displacement:
200.1
E 100.1
I
z
%Ilp
0.1
41
c
0 -100.t
--- -------
-
-200.C
13
c
IN
in
-300.0
P-4
-AM n
0.0 4.0 8.0 12.0 16.0
-0- msh x8y8b1 meshx8y8b6
Fig4.35 Influenceof meshgrading(SeriesH r-refinement)on wall bendingmoment
: selectedprofiles
4-128
200.
33.
.X
-13L
-300A
-96.1
-633.3 ...........
-800.0
4.0 8.0 12.0 16.1
ueprn i3eiLow top or waii tni
Mh X89OU * neshx8y8b6
(b) top-proppedwall / undrained.

/ elasticsoil (set3)
auu.I
£ 600.(
4.
C 400.C
ci
£
0
£
200.0
C
C
ci
0.0
-200.a L-
0.0 4.0 1.0 12.0 16.0
-0- rieshx8y8bI * msh x8y8b6
(c) cantileverwall / drained/ elasticsoil (set 3d)
Fig4.35 Influenceof meshgrading(SeriesH r-refinement)on wall bendingmoment

: selectedprofiles(contd)
4-129
50,
40A
30.0
20.0
10.0
L-
0.0
-8.0 -6.0 -4.0 -2.0 0.0
41- rieshx8q8bI * msh x898b6
100.
2
60.0
. P4
40.0
-W
20.0
x
w
L-
0.0
-8.0 -6.0 -4.0 -2.0 0.0
Distance fron wall Cm)
Msh XBYBbl * mh x8y8b6
Fig4.36 Influenceof meshgrading(SeriesH r-refinement)on excavationheave:

selectedprofiles
4-130
100.
BOA
6O.
U
Cs
40.0
L-
0.0
-8.0 -6.0 -4.0 -2.0 0.0
-9- msh x8q8bI * heshx898b6
(c) cantileverwall / drained/ elasticsoil (set I d)

Fig4.36 Influenceof meshgrading(SeriesH r-refinement)on excavationheave:
13.
10A
3.0
0.0
E
w
u -5.0
0.0 16.0 32.0 48.0 64.0

weshx8y8b1 nuh x8y9b6
(a) unproppedwall / undrained/ elasticsoil (set 5)
Fig4-37 Influenceof meshgrading(SeriesII r-refinement)on ground surface
movement:selectedprofiles
4-131
13.0
10.
-I
E
C
Ci
Ci
(0
-20.0L- t
__________
0.0 16.0 32.0 48.0 64.0

?Wshx8gobI * mh x898b6
(b) top-proppedwall / undrained/ elastic-perfectly

plasticsoil (set 13)
20.
13.
%dp
10.1
3.(
c --- ------ ----
L-
-3.0
0.0 16.0 32.0 U. 0 64.0
msh x8q8bI * mh xIy8b6
(c) top-proppedwall /drained I elasticsoil (set3d)
Fig 4.37 Influenceof meshgrading(SeriesII r-refinement)on ground surface

movement: selectedprofiles(contd)
4-132
(a) 8 nodes(I LSQ) (b) 9 nodes(2 LSTs)
(c) 13 nodes(4 LSTs) (d) 29 nodes(12 LSTs)

Wi
NB:
no refinement of elements adjacent to the original quadfilateral is required
nodes shown only in (a) for dafity
Fig4.38 for a quadrilateralelement

Possibletriangulationschemes
4-133
Fig4.39 Pseudop-refinementmeshesfor an embeddedwall analysis(detail nearwall
shownonly)
4-134
(a) 8 nodes(I LSQ) (b) 20 nodes(5 LSQs)
(c) 25 nodes(4 LSQs and4 LSTs) (d) 32 nodes(9 LSQS)
NB.
no lefinement of elements adjacent to the original quadWateral is required
nodes shown only in (a) for dafity
Fig 4.40 Further possibilitiesfor refmernentwithin a quadrilateralelement(mixture

Of linear strain quadrilateralsand triangles)
4-135
CHAPTER 5
CONSTITUTIVE MODELLING AND PARAMETER SELECTION
5.1 General Introduction
CRISPoffers the usera numberof differentconstitutivemodels,rangingfrom simpleelasticity

to sophisticatedelasto-plasticstrainhardening/softening
models(seeSections3.3 and3.6).
The choiceof modelis closelylinked to the selectionof appropriatesoil parameters,and the
principalissuefacingthe designeris how muchcomplexityis requiredin order to ensurea
realisticresult.
In the LJKconstructionindustryit is not uncommonfor the designerto be presentedwith data

from a site investigationafter it hasbeenconducted,ratherthanbeingableto indicatewhat
informationwould actuallyhavebeenrequired. The soil datawhich is availablemaybe of
suchsparsityandlow qualitythat it is quite unreasonable
to suggestthat a powerful
constitutivemodelcould be built upon it. The choiceof constitutivemodelmaythereforebe
dictatedby what datais available,unlessthe project warrantsfurther expenditureon high
quality samplingandtestingfor site characterization.
isotropic linear elastic

The simplestform of soil modelwhich canbe usedis the homogeneous
OME) model,basedon Hooke's Law. On top of this, severallayersof sophisticationmaybe
in to
added order reflect the known behaviour
of soil, including:
a) nonhomogeneity,
b) anisotropy,
C) small-strain behaviour,
d) non-linearity,
e) yielding, and
f) non-associated flow.
Each of thesefeatureswiff now be discussedin more detA with referenceto previouswork.
5.1.1 Nonhomogeneity
For most soils,meaneffectivestress(andhencestiffness)increaseswith depthbelow ground

Nonhomogeneous
surface. is in the form E=E. + mz, Fig. 5.1(a)
stiffness normallyexpressed
but this may causeproblemswhenE,,,= 0. For thoseelementsat the soil surface,integration
5-1
pointsjust belowthe top boundaryhavea very low stiffnesscomparedwith integrationpoints
at thebottom boundary,Fig. 5.1(b). Within a singleelementit is thuspossibleto havea large
stifffiessratio, with implicationsfor ill-conditioning(seeChapter7). This may be a more
importantissuefor foundationloadingproblemssuchasthe platetest (Iffier, 1992),where
the structurewould be everywherein contactwith soil of very low stiffiaess.In a retaining
Wall,only the upperportion is likely to be affected. Althoughtherearenaturally-occurring
soilswhich truly exhibitE, =0 (e.g. uncementedcoarse-grainedsoils),it is possiblethat some
analystshaveuseda low E. to minimizethe effectof the retainedsoil holdingback the wall
(perhapsinsteadof using"no-tensioif' interfaceelements).For example,Potts andBurland
(1983)usedE. =0 for their Bell Commontunnelanalyses,despiteshowingthat cý. = 40 kPa
and statingthat a constantE, /c, was applicableto the stiff clayspresenton site.
Another form of nonhomogeneity in

which canoccur real soils is that of discretelayering
ratherthan continuousvariation. Different soil stratacould havecontrastingstiffnesses,
relativeto one another,whilst beinguniform in anygivenlayer. Suchlayeringmayhaveas
much (if not more)influenceon wall behaviour,but owing to the largenumberof possibilities
which exist,no analyseshavebeenconductedwith layeredE profilesin the presentstudy.
Although plasticyield (an issueof shearstrengthratherthan stiffness)is consideredin more

detail later in this chapter(Section5.6), it shouldbe pointedout that a nonhomogeneous
stiffnessprofile is usuallyaccompanied
by oneof nonhomogeneous shearstrength. In clay
soils,it is commonpracticeto definea profile of undrainedshearstrengthc. with depth,and
then to correlatestiffnessto it througha fixed E. /c. ratio. Typicalvaluesof Eu/cumayrange
from 250-1000,dependingon soil type, directionof loading', etc.
5.1.2 Anisotropy
Many naturalsoilsexhibitanisotropy,asa result of depositionandstresshistory. For

retainingwalls embeddedin suchsoilsit would seemlogical to useanisotropicelasticityasa
constitutivemodel(e.g. Simpsonel al., 1979;Creed,1979). However,stressinduced
anisotropydependson overconsolidationratio (OCR) which is not itself constantbut will vary
I Problemsof foundationsettlement,wherevertical stiffnessis moreimportant,may begiven

ratherdifferent " ratiosthan excavationsupportproblem - even if both are in the samesoil and at
the samelocation.
5-2
(usuallydecreasing)with depth. To modelstressinducedanisotropyrigorouslywould
thereforerequirea fairly complexvariationof Ehand/orE, with depth.
Anisotropyof shearstrengthis ascommonasanisotropyof stiffhess,yet the assumedyield

criterion (definingthe onsetof yield, usuallyin termsof soil strength)is virtually always
isotropic. CRISPgivesno opportunityto incorporateanisotropyof strength;all yield surfaces
are assumedto havesymmetryaboutthe hydrostaticaxis. (This is true for both elastic-
perfectlyplasticand Cam-claybasedmodels.) Someattempthasbeenmadeto incorporate
anisotropicyieldingin CSSMmodels,suchthat the yield surfaceis symmetricaboutthe K.
compressionline ratherthanthe meanstressaxis(e.g. Newson,1992). Othershaveattempted
to recognizethe anisotropicconsolidationhistory of naturalclays,usingstresspath testingto
determinethe true shapeof the yield locus(e.g. PicklesandWoods, 1989). This will be more
importantwithin the contextof plasticyield, andis relevantto Section5.5.
The characterization of anisotropy has tended to focus on the ratio n (= Eh/F-,). However, the
ratio m (= Gh/E, ) has also been shown to be important; in particular for predicting settlement
troughs over tunnels. Lee and Rowe (1989) were the first to demonstrate how manipulation
of m could give a good match of numerical prediction (using linear elastic-perfectly plastic
analysis) with centrifuge and field data. Adenbrooke et al. (1997) used more sophisticated
non-finear elastic models, and again found that the value assumedfor m had a big effect on
predicted settlements. To the writer's knowledge, no similar investigations have been carried
out for embeddedretaining walls.
5.1.3 Small-strain behaviour
The extentand magnitudeof groundmovementsaroundretainingwalls areoften over-

predictedusing simpleelasticmodelstogetherwith stiffnessparametersmeasuredin
conventionallaboratorytests. This is not only becauseconventionallaboratorytestsover-
estimatestrainsby includingbeddingeffectsin their measuredvalues,but alsobecausethe
strain levelsaroundwalls aretypicallyvery small(seeJardineet al., 1986). Two approaches
arepurrentlyin use. In the first, backanalysisof comparablestructuresin similarground
conditionsis usedto obtainparameters(typicallylinearelastic,perhapscross-anisotropic),and
theseare then usedfor forward prediction. In the secondapproach,smallstraintriaxial testing
is carriedout on samplesobtainedfrom the actualsite,anda singlestifffiessis selectedfrom
the test results. In eachcaseit is commonto usea simpleconstitutivemodeldefinedby a few
5-3
parametersin order to performthe FE computation(non-linearanalysisper se is discussedin
Section5.1.4).
Both approaches havepotentialproblemsassociated

with them. In the first, it is implicit that
both geometryandgroundconditionsare sufficientlyidenticalto allow satisfactoryforward
simplifiedsoil behaviour,aswell asa

prediction,whilst the secondapproachassumes
knowledgeof a suitableaverageoperationalstrainlevel. Thesetwo approacheswin now be
consideredin further detail.
5.1.3.1 Back analysis
Back-analysismustbeginwith the selectionof a constitutivemodel,in order to decidewhat

parametersare actuallygoingto be inferred. If Hooke's Law is usedwithout anisotropy(n
1) or nonhomogeneity(X = 0), only E andv (or G andK) needto be deduced.Drainage
conditionsanda knowledgeof whethertotal or effectivestressesare appropriatewill normally
fix V (vu or V) in a narrow band,leavingjust E (F, or E'). An FE analysisis madewith trial
valuesfor the soil parameter(s)concerned,andthe resultscomparedwith observations.
Adjustmentsaremadeandseveralreanalyses are conducteduntil acceptable agreementis
2.
obtained Hooke's Law is attractivedueto its simplicityandthe fact that the parametersare
physicallymeaningful.However,it will not generallybe possibleto matchall observations-
for example,agreementwith the ground surfacesettlementtrough maybe obtainedwith a
combinationof E andv which givesa ratherpoor matchwith the wall deflectionprofile. All
of the intrinsic aspectsof behaviour(anisotropy,nonhomogeneity,
non-linearity,yield, etc.)
which go to producingthe true stress-strain
responsewill only be by
equivalenced a single
linearE valueat onepoint on the curve. This hasimplicationsfor, andexplainswhy thereare
difficulties with
a) matching more than one computation/observation, and
b) from
usingparameters the back analysisof one structurefor the forward predictionof
-
another.
If a constitutivemodelmore sophisticatedthan simpleHooke's Law is used,different

problemsarise. The principalproblemis that of ambiguity,wherebya measuredaspectof
behaviourcanbe reproducedby morethan one combinationof soil parameters.NUier(1992)
2 Barkhordari(1998)hasshownthat this parametersearchprocesscanbe automatedvery

efficiently by couplinga geneticalgorithmprogramto the FE code.
5-4
hasinvestigatedthis thoroughlyin the contextof interpretingplateload tests,wherethe
inferenceof soil parameters(evenwhen simpleconstitutivebehaviouris assumed)is subjectto
ambiguitybecauselimited measurements areavailable(usuallyonly appliedload andresulting
settlement).Oneexampleconcernsnon-linearmodelsof the Gunn(1993) power-lawform,
Fig. 5.2, which canproducean inferredE. profile underneatha loadedplatewhich is similarto
a nonhomogeneous distributionF, ;.-, mz (Woodset al., 1992). The inducedstrainprofile,
togetherwith the power-lawmodelresultsin the operationalEu beingvery smallnearthe
ground surface,increasingwith depth.
In order to restrict the numberof analysesusedfor the presentwork, it wasdecidednot to

carry out any specificinvestigationsof back-analysis.Of course,all of the pointsmadein
Chapters4 to 8 of this thesisare relevant any decisionwhich could affectthe quality or
-
reliability of computedresultsin a designpredictionexercisewill do exactlythe sameto a
back-analysis3.
5.1.3.2 Triaxial testing
The basicapproachis to conducttriaxial compression testswith small-strainmeasurements

enablingE (E' or EJ to be determinedat differentstrainlevels. The appropriateE valuesare
then enteredinto the FE analysis- basedon the engineer'sjudgementof what strainlevels
pertainin the field.
This procedurecanbe taken one stepfiirther if the triaxial equipmenthasthe ability to follow
any desiredstresspath, suchasthe hydraulictriaxial cell (BishopandWesley,1975). Noting
that soil behaviouris dependenton stresspathaswell aslevel, representative elementsof soil
can be subjectto the appropriateloadingdirection. In broadterms,soil elementsbehindthe
wall will experiencehorizontalunloadingwith constantvertical stress,whilst soil elementsin
front of the wall Oustbelow firial dig level)undergoverticalunloadingwith (approximately)
Back-analyses by
are often supplemented published observationson similar structuresin
-j
comparablesoils. Theseare givenconsiderable by because they represent
weight some engineers
reality. In the QueensberryHouseanalysisdescribedin AppendixA, the consultantssteadfastly
refusedto acceptthe resultsof non-linearelastic-perfectlyplastic FE analysis,with parameters
Providedby pressuremeter and small-straintriaxial tests- choosinginsteadto accepttheir own much
cruderanalysissimply becauseit agreedbetterwith datafrom the Houseof Commonscar park
(Burlandand Hancock,1977). The consultantseventuallyconcededthe point, but not beforesite
measurements made during (as
construction part of the ObservationalMethod) supportedthe writers
analyses.
5-5
constanthorizontalstress,Fig. 5.3. Triaxial testsfollowing thesestresspathswill showthe
engineerif the stiffhess,pore pressureresponse,andstrengthare path dependentto any
significantdegree.The appropriateparameterscanthenbe assignedto differentregionsof the
mesh.Thewriter hasbeeninvolvedin at leastoneprojectwherethis was done(A406
Walthamstow;Case1, AppendixA).
5.1.4 Non-linearity
It haslong beenrecognizedthat the elasticbehaviourof particulatematerialssuchas soil is

only reallylinearover smallincrementsof strain. The overallpatternof displacements
predictedusinga simplelinearmodelmaybe incorrect. A well-known exampleof this is the
predictionof how muchthe Big Ben clock tower would tilt asa result of the constructionof
the Houseof Commonsundergroundcar park (BurlandandHancock, 1977). Initial linear
elasticanalysescalculatedthat tilt would be in the oppositedirectionto that subsequently
measured;a mistakerectified only by the incorporationof non-finearityin later computations.
Nonetheless,linearelasticanalysesremainvery populardueto their simplicity small
-a
numberof well-understoodparametersandfast computation(no iterationsrequired).
The hyperbolicmodel(Duncanand Chang,1970)wasoneof the first non-linearelasticmodels

to gain acceptancein geotechnics,basedon the approximationof triaxial stress-straincurves
by a hyperbola.Its successes(andlimitations)havebeenwidely documented.Piecewise
linear and splinemodelshavealsobeenusedto representreal stress-straincurvesin a
convenientmathematicalform for numericalimplementation(e.g. DesaL1971). More recent
proposalsincludea two-parameterpower law modelby Gunn(1993), specificallydevisedfor
improvedpredictionsof tunnel-inducedsettlementtroughs. The higheststateof refinementof
this type of modelis probablyrepresentedby the ImperialCollege(1C)periodic-logarithmic
model of Jardineet at (1986),Fig. 5.4. This modelis definedby five parameters,obtainedby
curve-fitýingtriaxial test data.
The Cam-claymodel(SchofieldandWroth, 1968)reproducednon-linearelasticbehaviour

below the yield surfaceasa by-productof its formulation,with both KI and G depended
directly on meaneffectivepressurep'. However,this aspectof Cam-clayand its variantshas
not beenconsideredasimportantasits ability to reproduceplasticyield andhardening.
5-6
A differentapproachwas followed by Simpson
el al. (1979)with the so-calledLC (London
clay)model. This incorporatedthe conceptsof thresholdstrain,andkinematicsurfacesin
strainspace.It wasusedsuccessffillyin a ClassC predictionof the aforementioned Big Ben
clocktower andhelpedto establishthe importanceof usingnon-linearelasticityin such
situations.Over a decadelater, Simpson'swork producedthe "bricks-on-strings"(orjust
BRICK) model,basedon the unusualanalogyof a mandraggingbehindhim severalbricks
of
differentweight, via stringsof differentlength(Simpson,1992). This model
simulatesboth
non-linearityandrecentstresshistory,reproducingthe so-called"S-shaped"curve describing
the variationof E, with the logarithmof strain.
A relateddevelopmentwithin the frameworkof critical statesoil mechanicsbeganwith the

"bubble"modelof Al-Tabbaa(1987), a single kinematic yield surfaceinside
whichuses a
modifiedCam-clayboundingsurface. Stallebrass (1990)developedthis further, culminatingin
a 3-surfacekinematichardening(3-SYJI) modelcapableof reproducingthe effectsof current
state,recentstresshistory,andchangesin stresspath direction(StallebrassandTaylor, 1997).
3-SKH requireseight parameters,but hasshownconsiderable promisein predictingground
movementin over-consolidatedclay soils. It is perhapsthe way of the future, but its usein
designat the presenttime is impracticable.Furthermore,subsequent work by Claytonand
Heymann(2001) hascastdoubt on the needto modelrecentstresshistory, suggestingthat
excessivecreepstrainrateshadled othersto infer (incorrectly)that recentstresshistory was a
genuinesoil feature,ratherthan an artefactof the experimentalprocess.
Since CRISP was releasedpublicly in 1982, many constitutive models have been implemented
by university-based investigators, including most of those mentioned above (see Table 2.3).
However, commercially available CRISP did not possessnon-linear elastic models
versions of
until 1999. Consequently, at the time of writing, there has been little published experience in
their use within CRISP.
5.1.5 Yielding
All soils exhibit irrecoverable(i.e. plastic)behaviourabovecertainlevelsof stressor strain.

Classically,a yield surfacein the appropriatestressspacehasbeenusedto differentiateregions
of purely elasticandplastic(or elasto-plastic)behaviour.Theyield surfacemaybe fixed in
sizeand position (perfectlyplastic),maygrow or shrinkaboutsomereferenceaxis (isotropic
hardening),or maytranslatein (kinematichardening).Familiarexamples
stressor strainspace
5-7
of perfectlyplasticcriteria are TrescaandMohr-Coulomb(Fig. 5.5), whereasthe Cam-clay
familyexemplifyisotropic hardeningyield criteria(Fig. 3.27). Kinematichardening,pioneered
in geotechnics
by Mroz et al. (1979), playsa fundamentalpart in the "bubble" and3-SKH
models.For soft clays,the shapeof the yield surfaceandthe mannerin which it strainhardens
areof majorimportancein the accuratepredictionof groundmovement. For stiff clays,it is
the behaviourwithin the yield surfacewhich is usuallymoreimportant,and thus the focushas
beenprimarilyon describingelasticmodulusandits variation(e.g. non-linearity).
The currentconsensus (e.g. Smithel al, 1992)is that yield actuallycommencesafter small
stressincrements,well within what hastraditionallybeenconsideredan elasticregion. An
alternativeapproachwhich recognizesthis is the "boundingsurface"-a limit on possiblestates
within which both elasticandplasticstrainingmaytake place. Oneexampleis the continuous
plasticitycritical statemodelof Naylor (1985),in which plasticstrainsare proportionalto the
proximity of the stresspoint to a modifiedCam-claysurface.The 3-SKH modelalsoemploys
a boundingsurface,within which are (nested)yield andhistory surfaces.From a practical
plasticyield surfacemay seemto
viewpoint, variableelasticitywithin an elastic-perfectly
producethe sameresult asevolvingplasticitywithin a boundingsurface. In manyinstancesof
monotonicloading,this maywell prove to be the case. However,wherethereare multiple
reversalsin loadingdirection(e.g. loading-unloadingcyclesin a platebearingtest), differences
betweenthe two typesof constitutivemodelwould be expectedto emerge.
Notwithstandingthe above,in practicalretainingwall analyses,elastic-perfectlyplasticmodels

are muchmorelikely to be usedthan anyothertype of plasticitymodel(evidencedby the
frequencyof its occurrencein Table2.3). Consequently,the investigationsdescribedin this
sectionhavebeenconfinedto the former type of constitutivemodel,to attemptto provide
insightswhich maybe helpfulto the practitioner.
When performingan undrainedelastic-perfectlyplasticanalysis,the user can electto work in

1/2)and describingyield as
terms of either(a) total stress,usingundrainedstffffiesses(E., v,,%ts
a function of the undrainedshearstrengthcý, or (b) using
effectivestress, E' and V (together
with K,, >>0 to producean incompressible response), and definingthe yield criterion with c'
and ý'. In the former, Tresca'syield function (Fig. 5.5a)would be used,written as:
(al
2. 2.t - 2.c. (9,0) (5.1)
-a3)
- c =
5-8
For 2D stresssystems,CRISPusesthe slightlymore generalexpression:
2.a.. cos.0 - 2.cü (:9 0) (5.2)

where
GS [ V2(S)? + SY2 + SZ2) +T 2 ]V2
xy
Sx a'X - a'mt (similarly for syand sz)
at Od (cyx + Wy + WJ/3
0 sin-V-3/2)(43) J3/
a, 3)/3
J3 SX.SY.S'. SX.
T 2
Z- xy
If working in termsof effectivestresses,the Mohr-Coulombyield function (Fig. 5.5b) canbe

written as:
c' - cr'taný' (5 0) (5.3a)
or
(c'cotý' + S')siný, (:g 0) (5.3b)
where t' = '/2(a'l - CF'3)and s' =1/2(CFtl + C6)-
The more general expressionused in CRISP is:
-a', Asiný' + cy.,

cosO+ a,,sjnýlsjno/43- clcosýl (5.4)
Closely allied to the selection of the yield criterion is the specification of in-situ stresses. As
observed by Burland (1978), there is little point in carrying out a sophisticated non-linear
elasto-plastic analysisif the initial stressesare incorrect. In this respect the at-rest earth
pressure coefficient K, can be particularly difficult to measureor estimate over the full depth
of interest. CRISP provides considerableflexibility in spec4ing variable Ko profiles; the real
limitation is in knowing what they are for a particular field situation.
5.1.6 Non-associatedflow
Whenthe stressstateof an elementof soil reaches(andremainson) the yield surface,plastic

by continuedelasticstrainingwhich is typically of a much
strainingwill occur;accompanied
lower magnitude.The relativecomponentsof plasticvolumetricandplasticshearstrainare
5-9
determined by the flow rule, with plastic strainvectorsbeingnormalto the plasticpotential,
whichis alsoa surfacein stressspace. If the flow rule is associatedandnormalityapplies,
thenthe yield surfaceandplasticpotentialarecoincident,andthe plastic strainvector is
normalto the yield surface.This assumptionwill leadto excessivedilation in a drained
sis;if the loadingis undrained,stronglynegativeexcesspore pressuresat yield/failurewill
be generated.
Associatedflow is considerablyeasierto implementin a finite elementpackagethannon-

associatedflow, andis thereforemore common. Oneattractiveaspectis that it generatesa
global systemof stiffnessequationswhich aresymmetric,with all the advantageswhich this
impliesfor the storageandsolutionof the equations.But dilation of this magnitude(i.e. angle
of dilation W is unrealistic,andcanleadto over-enhanced
strengthand stiffnessaround
the wall (Ponnampalarn,
1990;Powrie andLi, 1990a). Thereis little consensuson the most
suitableangleof dilation (1/2ý'is usedby some;e.g. Day, 2001) or on which aspectsof
retainingwall analysisare most affected. Potts andBurland(1983)havestudieddilation and
cameto the conclusionthat wall movementswerelargelyunaffectedby the value of W
adopted. Earth pressuredistributions,however,maybe more sensitive,and it is preciselythis
sort of informationwhich the practitionerneeds- if dilationis not a key issue,then neitherthe
FE codewhich permitsits variation,nor the laboratorydatadescribingitare required.
Most of the elasto-plasticmodelsprovidedin CRISPassumenormalityat yield (i.e. flow is

associatedandthe yield function and plasticpotentialare coincident),includingthe elastic-
perfectly plasticandCam-clayfamily of modeW. The exceptionto this is a non-associated
Mohr-Coulombmodelwhich was introducedin 2000,but with which experienceis still fairly
limited. Normality requiresthe vector of plasticstrainincrementto be normalto the yield
surface. In the caseof Tresca,the yield surfaceis parallelto the meanstressaxis,and so the
volumetric componentof plasticstrain UP, is (i.
zero e. no dilation), Fig. 5.6(a). In contrast,
the yield surfaceis inclinedat an angle(a functionof fl to the meanstressaxis,andthe
is
plastic volumetricstrain negative,implying an angleof dilation %V
= ý', Fig. 5.6(b). If this
dilation is suppressed
becauseloadingis undrained,negativeexcesspore water pressurewill
4 An unfortunateerror was introducedin earlierversions(CRISP 82 and 84), wherea sign was

reversedon a term in the Mohr-Coulombsubroutine.1he error arosefrom usingthe codepublishedby
OwenandHinton (1980),who adopteda tension+ve conventionfor stressinsteadof the compression
+ve usedin CRISP. The net cffect of this error wasto suppressdilation almostentirely-a desirable
effect in many cases,but achievedin a mannerlacIdngany soundtheoreticalbasis.
5-10
begenerated.This in turn will causethe effectivestresses in the soil to be elevated,with an
attendantincreasein failure stress(which is, of course,proportionalto meaneffectivestress).
Diagrammatically,this correspondsto the stresspoint migratingup the failure fine, andwhilst
this is seento someextentin the behaviourof real soils,dilation is rarely as strongasV= ý'.
5.1.7 Overview of numerical studies
Despitefairly widespreadknowledgethat soil is not homogeneous,isotropic, or linearelastic

CHILE),it is frequentlyassumedto be so. Beforedesignersincludemore realisticsoil models
in their FE analyses,
they will needto understandhow the resultsmaybe affected,andif the
extracomplexityis warranted. To this end,numericalstudiesinto constitutivebehaviourhave
beenconductedandare presentedin the following sections.In line with Chapter4, horizontal
Walldisplacement,
wall bendingmoment,excavation heave,andground surfacemovement
havebeenhighlightedaskey resultsfrom eachanalysis.The evaluationof resultshasbeen
along similarlines,in that maximumvalues aswell asentire profiles (17*)havebeen
exan-dned.The principalaim hasbeento highlightwherethe introductionof somefeatureof
real soil behaviourcauseda significantdifferenceto the profile or maximumvalueconcerned.
Someof the resultshavebeenobtainedby reinterpretationof the analysesreportedin Chapter

4, focusingon constitutivebehaviourratherthangeometricmodelling(seeSection4.1 for
analysisdescription).Only meshxI0yl0 hasbeenusedsothat the resultscanbe deemedto be
largely,unaffectedby boundaryproximity, meshgrading,etc. The samefive proppingcases
havebeenused(Fig. 4.5). Additional analyseshavebeenconductedon constitutivefeatures
flow.
not includedin Chapter4; namely,non-linearityandnon-associated
Before consideringeachconstitutivefeaturein turn, somecommentsWill be madeaboutthe

mannerin which the resultshavebeenpresented.For nonhomogeneity,anisotropy,andyield
Ithe key (5,,,, V,,. )
±S.,,,,, initially extracted. Then, the
output values ±N", and were
percentage differenceswere calculatedbetween key its
each residtand corresponding
"baseline7'value,asan indexof the influenceof the constitutivefeatureconcerned.The
baselineresult wastakento be that for the homogeneous isotropiclinearelastic(HILE)
in
parameterset eachproppingcase,exceptwhenyieldwas introduced,
when the baseline
result was takento be that for the fully elasticanalysishavingthe samedegreeof
nonhOmogeneity in stiffness.To summarize,resultsfor parameterSets2-5 havebeen
5-11
to Set 1; Sets12 & 22 havebeenreferencedto Set2; Sets 13 & 23 havebeen
referenced
to Set3- andsimilarlyfor drainedanalyses(seeTables4.3 to 4.8).
referenced
Thepercentagedifferencesare summarizedin Tables5.1 (undrainedanalyses)and5.2

(drainedanalyses).Selectedplots showingthe influenceof different constitutivefeaturesare
includedat the endof this chapter
and arereferencedfrom the appropriatesections.
5.2 Nonhomogeneity
5.2.1 Description of analyses
Briefly recappingChapter4, nonhomogeneity by

was characterized a dimensionless
parameter
x, definedin termsof E at ground surfaceandat formationlevel (seeEqn 4.3 andFig. 4.4),
andwhich rangesfrom 0 (homogeneous) to I (highlynonhomogeneous).This is one of
severalWaysof characterizingnonhomogeneity(seealsoButler, 1975and Gibson,1974).
Analyseshavebeenconductedfor threevaluesof X- the limits of 0 and 1, togetherwith an
intermediatevalueý--1/2(seeTables4.3 and4.5). The effectsof nonhomogeneitycanbe seen
by comparingresultsfor Sets2 and 3 with Set I (the baseline)in Tables5.1 and 5.2.
Horizontal wall &splacement
Nonhomogeneityhada pronouncedeffecton both the magnitudeand shapeof the deflected

profile of an mnpjppýedwall when E. =0 (X = 1), Fig. 5.7(a). The apparentrestraintseenat
the top of the wall (wheresoil with E, >0 was apparentlyholdingthe wall back) was
eliminated,andthe profile was muchmoreEkewhat would be expectedfor a cantileverwall.
(This changeof deflectedshapewill haveimplicationsfor the bendingmomentprofile too.)
The maximumdisplacement more than doubled, it
so would seem importantto establish
whetheror not sucha nonhomogeneous
profile might exist at a site before conductinganFE
analysis.The magnitudeof V. is more likely to be governedby the averagestiffnessE.,.
over the retainedheightof the wall = 1/2(E,+ EH),thanby the degreeof nonhomogeneityX.
)
For Sets 1-3,E.,. = 50,70 and20 Wa respectively,andit canbe seenfrom Table 5.1 that the
orderingand relativemagnitudeof 5..,, correspondsto this in an inversemanner.
5-12
Whenthe soil was allowedto drain fully duringloading,broadlythe
samedifferencesbetween
theunpropped5 profileswere apparent.If anythingthe trendswere accentuatedand,as
the overallmagnitudeof displacement
expected, was increasedby drainage.
Theintroductionof a top-prop suppressed the influenceof nonhomogeneity on displacement

profileseenin the unproppedcase,Fig. 5.7(b),anda simplescalingof the 5 proportionalto
I/E.,. was apparent.Essentiallythe samepatternwasobservedwhenthe loading
wasdrained
although,ratherunexpectedly,the magnitudeof displacements was consistentlylessthanfor
the corresponding
undrainedE profiles. This is at variancewith the generalbeliefthat drained
movementswill be greaterthanthoseoccurringundershortterm undrainedloading(aswas
observedfor the unproppedwall).
For,the bottom-proppedwall, a markeddependence on the degree of nonhomogeneitywas

againin evidence,with the deflectionof the upperhalf of the wall beingmost affected. It has
alreadybeennotedthat this form of propping,installedprior to excavation,is unusualand
unrealisticallystiff However,it is clearthat a knowledgeof the actualdistributionof stiffness
with depthis ratherimportantin this case. With a uniformstiffhessthereis roughly equal
resistanceto wall movementaboveandbelow formationlevel,andthe higherlateralpressures
below the hingeat formationlevel are sufficientto causethe toe to moveout further thanthe
crest. As X increases,thereis rathermoreresistanceto wall movementbelow formationlevel,
and consequentlythe wall toe doesnot deflectquite so much. WhenE.. =0 (X = 1) the
balanceis tippedthe other way andthe wall crestdeflectionis morethantwenty timesgreater
than that for E. = constant.Full drainageservedonly to increasethe magnitudeof ma)dmum
displacements for the bottom-proppedwall, whilst leavingthe effectsof increasing
nonhomogeneityessentiallythe same,Fig. 53(c).
The profile andmagnitudeof wall movementsin the doubly-proppedcasewere only slightly

affectedaboveformationlevel,Fig. 5.7(d),with mosteffectnoticedat the wall toe (possibly
of lessinterestto the designer).Drainedconditionsmadeno differenceto the observations
about deflectedshape,andhadonly a smalleffect on the magnitudeof movementrelativeto
undrainedloading.
Finally, for the alternately-propped the

case, influence of nonhomogeneitywas almost identical
to that on the unpropped. A
wall. comparisonof the relevantcolumnsin Table5.1 revealsthat
5-13
thechangesin 8., relativeto the homogeneous E profile werevirtually the same. Similarly,
for fiffly drainedconditions,the alternatelyproppedandunproppedcasesexhibitvirtually
identicaltrendsregardingthe influenceof nonhomogeneity.
Wall bendingmoment
Nonhomogeneityhada striking effect on the bendingmomentprofile for the unproppedwall

in that, asX increasedfrom 0 to 1, the wholeprofile shiftedfforn beingpredominantlynegative
(tensionfacetowardsthe excavation)to predominantlypositive(tensionfacetowardsthe
retainedsoil). This broughtan attendantshift in the positionof NL. from aboveformation
levelto just below. for x=0 was aboutthe samemagnitude as +NL. for X=1;
-M.
intermediatex valuesproducedintermediate±NL.. In a designsituationit could be argued
that the sign of the maximumbendingmomentis not crucial,becausemost concreteretaining
walls will be reinforcedsymmetrically.However,shouldthis be important,thenit would seem
vital to establishthe actualprofile of E,
This sametrend was seenwith fiffly drainedloading,Fig. 5.8(a). The differencebetween
-N" and+N" on anygivenprofile was slightly largerfor the drainedcase,andwhenX=1,

virtually all the negativemomenthad disappeared from the drainedprofile.
Top-proppingbroughtabouta uniformity of shapeof the bendingmomentprofiles,Fig.

5.8(b), with the maindifferencecausedby the varyingdegreesof nonhomogeneity
beingin the
magnitudeof M..., which wasalwayspositiveandwasinverselyproportionalto E.,,.. There
was very little bendingmomentbelow formationlevel,beingat most about 10%of -N".
The trend for drainedloadingwas identical,andin fact the magnitudeof the bendingmoment
'
profiles was almostthe same,indicatingthat drainagehadlittle effect. Therewas a small
increasein +K. below formationlevel,but this was still relativelysmallcomparedwith
momentsabovefonnation level.
Despitethe big changesin displacement by in

profile caused nonhomogeneity the bottom-
PLM-ed case,the shapeof the bendingmomentprofilesremainedlargelyunchanged
regardlessof degreeof nonhomogeneity.This impliedthat the curvaturepatternof the wall
was about the sameirrespectiveof whetherit wasthe crestor the toe which had displacedthe
most. Therewas almosta "see-saw"effect aboutthe formationlevelprop. In fact the profiles
for X=0 andx =. 444 were more or lessidentical,with changeonly apparentwhenX=I- in
5-14
theform of a 50% increasein maximum(positive)momentat the formationlevel prop (due to
Wallcurvatureincreasing)andthe disappearance
of virtually all negativemoment,Fig. 5.8(c).
Drainedloadingpresenteda similarpicture. Nl, for x=I was unaffectedby drainage,

=
whereasfor other X valuestherewas a 25% decreaseof maximummomentover the prop.
Thismeantthat, comparingthe X=0 andX=I drainedcases,M.,, over the prop increasedby
100%. For both typesof drainage,onemayconcludethat it is importantto establishwhether
or not E, -> 0 asthis hasthe mostprofoundeffect on bendingmomentswhenthe wall is
bottom-propped.
All bendingmomentprofilesfor the doubly-proppedwall hadthe sameshapeandmore or less

similarmagnitudesfor the degreesof nonhomogeneity
considered.The maximummoment
was alwayspositiveandlocatedat the formationlevel prop, andhadthe samemagnitudefor X
-0 and=1, beingslightlylower for intermediateX. Full drainagepreservedthe shapeof the
profiles,and simplycauseda smallshift in the bendingmomentdiagramtowardsthe negative;
i. e. +M.. reducedand increased by the sameamount. The degreeof
-9. approximately
nOnhomogeneity was,therefore,unimportantin this proppingcase.
The shapeof the alternately-propped wall bending moment profiles was affected by
nonhomogeneity in much the sameway as for the unpropped wall (i. e. N" gradually pushed
below formationlevel), althoughnot quite so markedly. Therewas a slightlymore
pronounced"spike' in the bendingmomentprofile over the formationlevel prop whenX=1.
Thesecommentsare equallyapplicableto the wall underdrainedconditions,althoughthe peak
on the profile whenX was evenmorepronounced, Fig. 5.8(d). The overall magnitudesof
bendingmomentswere aboutthe samefor the two drainageconditionswhen0 :9X but
--5.444,
full drainagebroughtabouta doublingof +K., whenx=1.
Excavationheave
The shapeof the excavationheaveprofile was the samefor eachdegreeof nonhomogeneity

considered,regardlessof proppinganddrainageconditions. On reflectionthis is reasonable,
becauseexcavationheaveis basicallya responseto a 2D unloadingsituation,and the upward
movementof the soil at formationlevel is influencedby the vertical stiffnessof the soil. No
lateral variation of E. acrossthe areaof the excavationwas permitted.
5-15
Themagnitudeof heaveswere certainlyaffectedby the profile of E,,, aswould be anticipated.
In general,the largestheaveswere associatedwith uniform stiffness(Eu= Ej andthe smallest
with nonuniformstiffiless,nonzeroat the surface(Eu= E. + mz). Switchingto drained
conditionspreservedthe generalpicture and,comparinglike for like in termsof X,
approximatelydoubledthe heavevalue,Table5.2.
A typical pair of heaveprofilesis given in Fig. 5.9 for the top-proppedcase,both undrained
(a) anddrained(b). The profilesfor otherproppingcaseswerevirtually identical- which was
somewhatsurprisingasa lesswell proppedwall might be expectedto move in moreat
formationlevel and (via undraineddistortion)causemoreuplift of the excavatedsurface. One
point to draw attentionto (whichthe figuresshow,but the tablesdo not), is that higherV..
on the centrelineis accompaniedby higherV at the wall. This meansthat the whole area
aroundthe wall is heavingmore in responseto excavationunloading. The differentialheave
acrossthe excavation,AV, is more relevantbut visualinspectionsuggeststhat the percentage
differencesquotedfor V. in Table5.2 areaboutright for AV too.
Groundsurfacemovement
As a generalpoint, it shouldbe notedthat the analysesbeingdiscussedherewere conducted

with fully rough boundaryconditionsandit was shownearlier(Section4.4) that the surface
movementprofile is quite stronglyaffectedby the choiceof boundaryconditionon the remote
vertical boundary.
Furthermore,wheneverthe loadingconditionswere switchedto fiffly drained,all settlement

(relative to the datumof the originalgroundsurface)was lost, andthe profile was alwaysone
of upward movement- from zero at the far boundaryto a maximumvalueat the wall. The
Proppingmade little (if any) differenceto the drainedS profileswhich was (again)surprising.
Vertical surfacemovementshaveonly beenplotted backto a distanceof 40m (= 5M,

although the back
surfaceextends a fiffl 80Mto the remoteverticalboundary- The cut-off of
5H was basedon the extentof movementobservedfor walls in the field (e.g. Carder,1995).
Furthermore,truncatingat 40m allowedthe detailnearthe wall to be shownmore clearly.
For the uUropped wA the basiceffect of increasingnonhomogeneity wasto alter the shape
of the surfacemovementprofile quite considerably,Fig. 5.10(a). The position of maximum
5-16
settlementwas pushedmuchcloserto the wall asX changedfrom 0 -ý 1. In fact X=I yielded
-
a "classicar'settlementtrough immediatelybehindthe wall, which someclaim is only observed
whennon-linearstiffnessis used. Upward movement+S reacheda maximumimmediately
behindthe wall, andwas a function of globalheavecausedby excavationunloading(in the
absenceof slip elementsto allow relativesoil-wallmovement).
Top-proppin enforcedmoreuniformity betweenthe groundsurfacemovementprofiles, so

that the shapeswere not very muchdifferentover the rangeof nonhomogeneityconsidered,
Fig. 5.10(b). The magnitudeof settlementvariedsomewhat,beinggreatestwhenx=0.
Upward movement+S reacheda maximumimmediatelybehindthe wall, andwas a function of
global heavecausedby excavationunloading.
Nonhomogeneityaffectedthe bottom-proppedwall in a similarmannerto the Unproppedcase,

in so far asthe positionof rna-,
drnurnsettlementwas pushedmuchcloserto the wall, Fig.
5.1O(c). The trough was not quite so pronouncedthough,andwas muchmore localized. The
samecommentsapplyto +S asfor top-propping.
The shapeof the surfacemovementprofilesfor the doubly-proppedcaseseemedlargely

unaffectedby nonhomogeneity-,
only the and
magnitudesof -S,,,,. +S. were influenced. The
samecommentsapplyto +S asfor top-propping.
Surface movement profiles for the alternately::proj2peggi

casewere virtually identical to those
for the unpropped wall; therefore the samecomments apply. The drained case is shown in Fig.
5.1 0(d), which is typical of aUthe propping cases.
5.2.3 Summary
Nonhomogeneity(of elasticstiffness)with depthcanbe expectedin virtually any depositof

naturally-occurringsoil, owing to the depth-dependencyof confiningpressure.The limited
studiesreportedherehavedemonstrated that a knowledgeof the actualdegreeof
nonhomogeneitycanbe very important- especiallyover the depthof the wall itself
Wall displacements to
areparticularlysensitive the degreeof nonhomogeneityX whenthe wall
is cantileveredor alternately-propped.Thesetwo proppingcasesarealsothosewhich show
the biggestsensitivityof bendingmomentto X, in termsof distributionand/ormagnitude.
5-17
Excavationheavealsodependson nonhomogeneity it to
and would appear be importantto
knowhow E varieswith depthif heavepredictionsare important.
Finally,ground surfacemovementpredictionsdependheavilyon the correctidentificationof x

(for mostproppingcases) particularlythe depthandlocationof any"trougW'behindthe
-
of damageto adjacent
wall. Thislast point hasobviousimplicationsfor the assessment
structures.
5.3 Anisotropy
by the ratio n (= Eh/E,) rangingfrom 1 to 3 in

As in Chapter4, anisotropywas characterized
the undrainedanalysesandfrom 1 to 4 in the drainedanalyses(seeTables4.4 and4.6). The
ratio m waskept constantat 0.417 (= 1/2(1+v,,
h)). As the selectionof anisotropicelastic
parameterswas ratherglossedover in Chapter4, further be
explanationwill now given.
Initially, parametersin termsof effectivestresswere selectedfor the drainedanalysesand

then, through suitablemanipulation,convertedto equivalenttotal stressparameterswhich
be in
could used the undrainedanalyses. This is a simpleenoughprocess for isotropic
materialsvia the identity G' = G., togetherwith Eqn 4.2. Valuesquotedby Atldnson(1973)
and by Creed (1979) for the London Clay were initially and
selected, then the processdetailed
by Bishop andI-light (1975)was followed to obtaintotal stressparameters.The degreeof
anisotropywas defined in (i.
termsof effectivestresses e. n' = )
E'&/E', as this was how the data
was presentedin the originalreferences.
However, it was found that the effectsof usingn'= 2 andn'= 4 were minimal,basedon a
comparisonof key output values from the analyseswith thosefrom the isotropic case. A
useful investigation hasbeen presented by Lee and

of this problemof parameterselection
Rowe (1989), who usedGibson's(1974)observationthat the undrainedelasticbehaviourof
cross-anisotropicsoil canbe definedthroughjust 3 independent E,
parameters Eh and GA.
The Procedureadoptedultimatelywasto selecttotal stressanisotropicelasticparameters,and
subsequentlyto infer effective
equivalent stress from
parameters these.
5-18
Horizontalwall displacement
Thegeneralshapeof the 5 profilesfor the unproppedwall was unaffectedby anisotropy,with

only a simplescalingdown of the magnitudeof displacements
as n increased(and eventhis
wasnot really significantuntil approachingn= 3). In the drainedcases,therewas a stronger
tendencyfor the wall to be pulledbackinto the retainedsoil asthe degreeof anisotropy
increased,Fig. 5.11(a), accompanied by muchstrongerreductionsin overall magnitude
(compareTables5.1 and 5.2).
For the W: qroppedwall, the shapesof the proffleswerevirtually identical,showingonly a

gradualreductionin magnitudeof 5 asn increased,Fig. S.11(b). Drainagecauseddeflections
to be ratherlessthan their undrainedcounterparts(at the samen), andalsothe reductionin 8
asn' increasedwasmore pronounced- wall displacements at n' =4 were very low indeed.
Increasedanisotropycausedsomeof the same"see-sawing"of the displacementprofile for the

bottgap-r-oo-edcaseaswas noticedwith increasingnonhomogeneity.Apart from that, the
overall shapeandcurvatureof the wall seemedlargelyunaffectedby n. Introduction of
drainagecausedthe wall to be pulledbackinto the retainedsoil asthe degreeof anisotropy
increased,Fig. 5.11(c).
For the doubly-proppýdwall, againanisotropyhadvery little effect on the shapeof the

displacementprofile - or indeedon the magnitudeaboveformation level,Fig. 5.11(d). Below
formation level therewas a gradualreductionin 5 asn increased,but this is of lessinterestto
the designer. The sameobservationsandcommentsmadefor earlierproppingcasesapply,
after changingfrom undrained,
to drainedloading.
Finally, the alternately-propped

casefollowed the
very closely patternof displacements
for
exhibited the unproppedcase,with a slightly smallermagnitude over all. Therefore,
the
samecommentsare applicable(for both undrainedanddrainedbehaviour).
Wall bending moment
Anisotropy hadlittle effect on the shapeof the bendingmomentprofile for the unpropped
wall; the maximummomentremainednegativeandwas alwaysabove formation level, Fig.
5-19
5-12(a).Therewas a smaHshift in the bendingmomentdiagramtowardsthe negative,with
+Nf.. reducingand-M. increasingby aboutthe sameamount. The drainedprofileswere
skfflarto the correspondingundrainedprofiles(at samen). Consideringthe drainedseries
increasinganisotropy(n') reducedboth +NL. and-N4,,. by about30% at most;
separately,
whichwasnot a greatdealconsideringEIh/E, hadto be increasedto 4 to achieveit.
Thebendingmomentprofile shapefor the top-proppedwall was similarlyunaffectedby

.sotropy,with the maximummomentalwaysnegativeandaboveformationlevel. The
magnitudeof -N" decreasedvery slightly(about 10%)at high n, but no more. The shapesof
the profileswhenloadingconditionsbecomefiffly drainedwere somewhatmore affectedby
increasinganisotropy,Fig. 5.12(b). AlthoughNL. wasalwaysnegativeandaboveformation
level,a more pronouncedpeakfor +XL. developedat high njust below formationlevel even
thoughthe magnitudeof +N1. did not changemuch. At high n', -N" was effectively
halved. Thus, establishingthe degreeof anisotropyis clearlymore importantfor long term
drainedloading,thanfor the shortterm.
Anisotropyappearedto makeno differencewhatsoeverto the bottom-proppedwall bending

momentprofiles. The profile shape,+NL.,, and_N'LMvalueswere virtually identical. For
fWly drainedconditions, was unaffectedby anisotropy and was equal to the undrained
-NI.
value. However, +NL. startedoff 25% lower than the undrainedvalue and then decreased by
a fitrther 50% asn' approached4, Fig. 5.12(c). Hencethe magnitudeof maximummomentto

be carriedby the wall would appearto be greatlyaffectedby the degreeof anisotropyunder
drainedconditions,but not for undrained.
The doub proppedwall behavedmuchthe sameasthe bottom-proppedwall, with the

in
exceptionof a smallreduction +NL. (at the formationlevel prop) whenn=3, Fig. 5.12(d).
The drainedprofilesfor n' =I andn' =2 areequivalent;+M. reducedby only 25% when
E'h/E'v = 4. Anisotropyappearednot to be importantin this proppingcase.
Finally, the alternately-propped only a modest to

response increased
anisotropy;
wall exhibited
+Mn= increasedby up to 40%, but -M.. (whichwas the largerof the two) stayed
approximatelythe same.The drainedandundrained, bendingmomentprofilescoincidedwhen
conditionswere isotropic;thereafter,both +N" and-N" steadily (by
decreased and
1--20%
60% respectively)with increasingn.
5-20
Excavationheave
As with nonhornogeneity,
the shapeof the excavationheaveprofile was the samefor each
degreeof anisotropyconsidered,regardlessof proppinganddrainageconditions. Thiswas
probablypredictable,asthe way in which anisotropywasincreasedin theseanalyseswasby
keepingF, constantandincreasingEh. It would be reasonableto expectthat verticalheavein
an aPProximatelyID unloadingsituationwould be governedby F, andnot by Eh.
In a supportedexcavationpart of the mechanism causingheaveis one of extrusion,asthe wall

is forcedto moveinwardsandthus squeezethe soil at the baseof the excavationupwards. It
night be expected,therefore,that if horizontalwall movementis restricted(becauseof
increasedlateralstiffness),that this mayleadto somereductionin heave.
Examinationof Table5.1 showsa high degreeof consistencybetweenthe differentundrained

proppingcases,in termsof the effectsof anisotropy.Therewas no changein V. at EVE,
2; it was not until n was increasedto 3 that anychangewas apparent,by which point a 33%
reductionhadtakenplace;e.g. for the top-proppedcase,Fig. 5.13(a).
Drained heavesalso showed insensitivity to propping arrangement(see Table 5.2). A small
reduction in V. was evident when n' = 2, and the reduction at n' =4 was only just over 10%;
e.g. Fig. 5.13(b) for the alternately-propped case.
Ground surfacemovement
The points madeearlierconcerningrough boundaryconditions,andthe loss of any downward
surfacemovementwhen conditionswere drained,are alsoapplicableto this section. The basic
shapeof the surfacemovementprofile andthe magnitudeof the movementswas the samefor
eachdegreeof anisotropyconsidered,regardlessof proppingcondition.Consistently,-S.
and +S. - werevirtually unaffectedby anisotropyuntil n=3, whentherewere reductionsof
50% and 40% respectively,e.g. Fig. 5.14(a)for the bottom-proppedcase.
Consideringdrainedsurfacemovements,againtherewas very little influencefrom propping

conditionson eitherthe profile shapeor magnitudeof movement;e.g. Fig. 5.14(b)for the
unproppedcase. Upward movement(+S.. ) experiencedonly a 10%reductionif Eh/F' was
increasedto 3.
5-21
Theinsensitivityof surfacemovementto anisotropywaslessexpectedthan for excavation
heave.Near to the excavation,surfacemovementsWill be influencedmainlyby the behaviour
of thewall; further awayit is dictatedmoreby the overallpattemof soil movement.It would
appearthat the samemechanisms which causeexcavation heavesto reducewhen a high
degreeof anisotropyis reached,alsoput somerestrictionon ground surfacemovement.
5.3.3 Summary
Anisotropygenerallyseemedto havethe effectof scalingthe wall displacementprofiles

without alteringtheir shape.The degreeof anisotropy(n) was constantwith depthin all the
analyses,so it is perhapsnot surprisingthat this happened.Genuineanisotropyvarieswith
depthasit is often relatedto OCR,but CRISPcannotmodelthis unlessthe groundis zoned
into horizontallayers,eachwith a differentEh/F,. If the deflectedwall shapeis largely
unaffected,then wall bendingmomentis lesslikely to be affectedthan displacement.
Paradoxically,bendingmomentwas affectedby the degreeof anisotropy,especiallyfor the
top-proppedwall where50% reductionswereobserved.
Effect on heaveseemsminimal,unlessvery largevaluesof n (or n) wereused. Thereare two

mechanismsat work - verticalrebounddueto removalof overburden(dominantin wide
excavationswhereW >> M anda Poisson'sratio effectasthe appliedlateralstresson the
block of sofl below dig level is increased.IncreasingEhwith F, constant(asdonehere)would
seemmore likely to influencethe latter mechanism, it
as would resistthis lateral"squeezing".
Somesignificantreductionsin groundsurfacemovementwere observed,but only for

undrained loading andonly whenextreme Eh/Evratioswere employed. There has been
insufficient time to examinehorizontalsurfacemovements,yet in somecasesthis is neededfor
the damageprediction- if gradientsaretoo steep,tensilecracIdngof adjacentstructuresmay
occur at foundationlevel.
Anisotropy, as definedhere,maynot be particularlyimportantfor embeddedretainingwalls -

unless it is of a very high degree(say 2). Anisotropy of strengthmay be importantif
n>
is
yielding significant. Anisotropy has
of permeability greater implicationsfor drainagerates
during and after construction.
5-22
5.4 Non-linearity
TheJardineet al. periodic-logarithmicmodelhasbeenusedin a short seriesof analyses.This

modelwasconsideredto be a compromisebetweenthe more elaborateBRICK modelandthe
ratherbasichyperbolicmodel. Furthermoreit hasbeenusedextensivelyby Potts andhis co-
workersat ImperialCollegeon a numberof retainingwall / deepbasementprojects,with
apparentsuccess(Jardineet al. 1991). TheJardineet al. (1986) non-finearityfunctionis
,
shownin Fig. 5.4 andis definedby:
Eu/cu =A+B cos [cc [logio (s./C)]') (5.5)
wheree,,is the axial strain,andA, B, C, a andy aredimensionless If

parameters. Eqn 5.5 is
normalizedby the valueof E" at somereferenceaxial strain,e*, andif A=B (often
approximatelythe casewith real soil test data),the following is obtained:
A* (I + cos fa [logio (e./C)ly) ) (5.6)
This formulationwasusedfor the Queensberry HouseFE analyses(AppendixA, Case8). It

allows a convenientdefinitionof Young's modulusat anypoint in the ground whereboth the
current strainlevel andE. at the referencestrainareknown. The latter is obtainedfrom:
F,,,(*) =+mz (5.7)
Then, knowing the currentstrainlevel at that point C2(andhenceEj is evaluatedfrom Eqn

5.6. The soil parametersrequiredare A*, C, a andy - thoughit hasalsobeenfound necessary
to imposelower andupperstrainlimits, ej. ande..., at which the stiffinessis at its greatest
and leastvaluesrespectively.Theseparametersarenormallyobtainedfrom laboratorytest
data; hypotheticalbut realisticvalueswere adoptedfor the presentinvestigation.Fixing A*
C
and at representative initial of 0.01%was adopted(thought to
values,an referencestraine*
be typical of the level at which commerciallaboratoriesquote stfffnessvalues). Then,ccandy
E,, / E,,
were varied so that the ratio gooj) (o.ol) e. (i. the 0
valueof when e=0.001 %) took on
successivevaluesof I (linear),2,3 and 4 (most non-linear). This is illustrated in Fig. 5.15,
and is denotedSeries EO
NLI. The valuesof and m for use in Eqn 5.7 were from parameter
5-23
sets1,2 and3 (Table4.3). A secondseries(NL2) was devisedin which a referencestraine*
of 0.1%was adopted,with (x andy variedto give flo.oo,= 1,2,3,4 asbefore. All parameters
usedaresummarizedin Table5.3.
In summary,seriesNLI had all Dx. curves coinciding at 0.01% strain (where 0= 1), with
individual 0 values having dropped from 1,2,3, or 4 at o.001% strain, and continuing to
diminish as far as 0.1% strain, after which D was constant. SeriesNL2 had all f2x. curves
coinciding at 0.1% strdA dropping from their maximum values (I - 4) at 0.00 1% strain,
diminishing as far as 1 strain, and constant thereafter. In both series, there was one log
.0%
cycle of strain between the reference and maximum strains (e* and F-ý), but different gaps
between the minimum and reference strains and e*) - one log cycle for series NLI, but
two cycles for seriesNL2. With so many parametersto select, numerous permutations were
possible. The values adopted in the parametric study were partly arbitrazyýthough were
influenced quite heavily by values derived for a number of projects on London clay sites where
the writer had performed commercial and/or researchanalyses. The intention was to carry out
illustrative analysesof the influence of non-linearity, rather than an exhaustive study of the
Jardine et al. model. Only unpropped and top-propped wall caseshave been considered.
Horizontal wall &splacement
For SeriesNLI, the effectof increasingsoil non-linearityon anonp-ropped wan was to causea
huge increasein deflection,for all distributionsof Eupj). The most dramaticincreasewas for
Set 3, wherethe maximumdeflection&. wasnearly200%greaterat D=4, Fig. S.16(a).
However, SeriesNL2 actuallycauseda reductionin displacement n
as was increased(exactly
the oppositetrend to SeriesNLI) andthe changeswerein somecasesvery modest- e.g. Set
3, where 8.. droppedby lessthan 10%.
The addition of tonropping amplifiedthe effect of increasingnon-linearity,with evengreater

increases in 5. (as muchas almost700% in Set 3), occurringat the wall toe. This suggests
that there was a dramaticreductionin stiffnessin the soil aroundthe embeddedportion of the
wall, allowing the toe to movemuch fixther in
than the linearcase. Once again, the in
changes
deflection profile were in the oppositedirection(i.e. a reduction),Fig. 5.16(b)andwere of
much smallermagnitude,when SeriesNL2 non-linearitywasused (greatest change-30%).
5-24
A generalobservationfi-omtheseanalysesis that the deflectionprofiles appearedto be scaled
geometrically(ratherthan changingshape)as 0 was varied. This impliesthat, whilst there
would be changesin the magnitudeof bendingmoment(due to increasingwall curvature),the
overall shapeof the bendingmomentdiagramshouldbe unchanged.
Wall bendingmoment
For SeriesNLI (e* = 0.01%), M.. increasedby up to 60% in the unproppedwall asthe
degreeof non-linearityincreasedto f2 = 4. (For setsI and2, M.. was negative,whereasfor
Set 3 it waspositive,but in all casesM. increasedin magnitude).As observedfor wall
deflection, SeriesNL2 non-linearityhad a more modesteffect on bendingmomentprofiles,
and changeswerein the oppositesense- i. e. causingM. to reducein all cases(regardlessof
sign), e.g. Fig. 5.17(a).
Top-propping(again)amplifiedthe effect of increasingnon-linearity,with 9. increasingby

more than 100% in somecases,Fig. 5.17(b). The changes in bending momentprofile were
smallerand in the oppositesensewhen the degree of non-hearity followed SeriesNU For
both proppingarrangements, the predictionsconcerningdeflectedshapewere correct- non-
linearity indeedchangedthe magnitudeof bendingmomentsbut not the shapeof the bending
momentdiagram.
Fxcavation heave
In all casesexanýned,increasingnon-linearityscaledup (or down) the heaveprofile, with very
few changesin its actualshape.With e* = 0.01%(SeriesNLI), the scalingwas >1, with V.
increasingby asmuchas 100%(unpropped)or over 500% (top-propped)for Set I (Eý(o.
oj)=
50 Wa). With e* = 0.1% (SeriesNL2), scalingwas <1 andlesspronounced,with V. as
much as halving (both proppingcases)when Ewa) -= 50 + 5z Wa. Occasionally,some
flatteningof the heaveprofile was noted, suggestingthat therewould be a more uniform
in
swellingpressure caseswherea formation level to
slabwas present restrainheave. Figs
5.18(a)& (b) compareSeriesNLI andNL2 for an unproppedwall, at Set 3 (E,,(.) - 5z MPa).
Ground surfacesettlement
The effectsof non-Enearityon surfacesettlementprofileshaveprobablybeenthe most widely
documentedaspectsof this featureof constitutivemodelling. The trendsdescribedin the
5-25
literature were all notedin the analysesperformedhere. In SeriesA, non-linearityaffected
both the shapeandmagnitudeof the settlementtrough for unproppedwalls; increasing
-S.
with! Q (as muchas400% in Set 3) and the point at which it occurredmoving closerto the
wall, e.g. Fig. 5.19(a). Even in SeriesNL2 thereis a clearchangeand, although-S. reduced
in most cases,the point of maximumsettlementhasmovedmuchcloserto the wall andso the
profile appearsmorerealistic(comparedwith linear elasticity).
TM-propped walls, SeriesNLI, showeda similarlymajor effect,though with a muchlarger

"jump" betweenKI =3 to 4, than from A=I to 2, or from 2 to 3. This ties in with the
pronouncedIdck-outof the wall toe observedfor 0=4. SeriesNL2 showedbasicallysimilar
trends to the unproppedcase,exceptat Eup.i) = Sz,whererestraintfrom the prop preventsa
major trough developingjust behindthe wall, Fig. 5.19(b).
Whereasfor wall deflection,wall bendingmoment,andexcavationheaveit was generallya

scaling of the profile which hadtakenplace,for groundsurfacesettlementtherewere
in In
significant changes shape. general (andfor Set 2 cases in particular),therewas definite
evidenceof reversedslopeof ground surfaceover a considerable distance behind the wan -
comparing the linear This
andnon-linearanalyses. is, of course,the very featurewhich
in the calculatedtilt of Big Ben referredto earlier.
accountedfor the discrepancies
5.4.3 Summary
All designquantitiesexaminedwere affectedby the incorporationof non-linearelasticity-

sometimesvery dramatically.Not surprisingly,the definitionof non-finearitywas a key
component,anda majorfactor uncoveredherewas the significanceof the referencestraine*,
where all stiffness:straincurvespassedthrougha commonpoint.
The resultsobtainedin Section5.6 suggestthat operationalstrainsaroundthe rctainingwalls

analysedweregenerallygreaterthan 0.01%but probablylessthan 0.1%. WhenusingSeries
NLI (e* = 0.01%),greaternon-linearity(higher91)led to the overall soil responsebeingless
stiff and hencegroundmovementswere larger. This would only happenif operationalstrains
were pushingsignificantlyinto the region beyond 0.01%. Conversely, with SeriesNL2 (e*
0.1%), increasingnon-linearityled to more soil beingstiffer thanat the referencestrain,and
henceground movementsare smaller. This would only happenif operationalstrainswere
generallylessthan0.1%.
5-26
In trying to estimatethe impact of non-linearityit is not just the absolutereductionin E. with
e, which is important,but alsothe rapidity with which this reductiontakesplace. Both Series
NL I andNL2 hadthe sameinitial stiffnessesat e=0.00 1%, but thereafterE,, decayedat
different rates. It is abundantlyclear from the analysisresultsthat the rate of declinehasa
major impact on how muchchangetakesplacein quantitiessuchas8, KV or S, andalso
what directionthe changeis (increaseor decrease).Previousinvestigatorshaveemphasized
only the effect of non-linearityon the shapeof the surfacesettlementtrough - these
investigationshaveshownthat other quantitiesof interestto the designerare also influenced.
Furthermorethe natureof the influenceis more subtlethan at first thought.
5.5 Yielding
Plastic yieldinghasbeencharacterizedeitherby E,,/c. for undrainedanalyses(Trescayield

function) or by ý' for drainedanalyses(Mohr-Coulombyield function). In the former case,
Eu/cuvariedfrom 0 (elastic,cu oo)to 1000(elasto-plastic,c. -+ 0), via an intermediate
-).
value of 500. In the latter, ý' was eithernot applicable(elastic)or set to a typical critical state
value of 25* (elasto-plastic).
Horizontal wall displacement
Yield preservedthe generalshapeof the displacementprofile for the unproppedwall under

undrainedloading,simplycausingit to moveout further. In the caseof Eu - mz with "-
1000 (wherethe initial stressstatewas on the vergeof yield beforeany excavationtook
place), the upperportion of the wall straightenedout andany contraflexurewas lost, Fig.
5.20(a). The additionaloutwardmovementin this latter casewasgreat (over 200%), but this
was due to widespreadyieldingbecausethe in-situ stagewas closeto limiting equilibrium.
The drainedcasedid not merit seriousconsideration;the increasesin 5. of 500% and200%
(for E' = E. +mz andE' = mz respectively)arehardly credibleandarisefrom the fact that the
wall is unstablein the long term at D/H =I (seeSection4.2.2).
5-27
For the Wzpmroppedd
wall, the generalshapeof the undrainedprofile was unalteredby yield;
there was a simplescalingof movementrelativeto the fully elasticcase. WhenE,, = mz (and
Eu/cu= 1000)the smallamountof contraflexurebelow FL was lost aswell. Drainedloading
brought about a somewhatstrongercontrastin profile shapebetweenthe elasticand elasto-
plastic analyses,Fig. 5.20(b),aswell asa massiveincreasein movement(nearly200%). The
contraflexurebelow FL was lost (for both E': z variations),which would obviouslyaffect the
bending momentprofilesin the lower half of the wall. In other words, the embeddedlength of
the wall is not beingclampedby the surroundingsoil so effectively,asyield is reducingoverall
soil stiffhess.
Bottom-propPing,in the way it hasbeenimplementedhere(andpossiblywith the wall

geometry beingsuchthat the prop is at the midpoint)was so efficientthat it virtually
preventedanyyieldingof the retainedsoil. Consequently, as can be seen in Table S.1,
(which was at the top of the wall) was completelyunaffectedby yield. -Examinationof the
5
whole profile for Eu = mz revealed that the toe of the wall had movedout much more due to
yielding on the passiveside,which increased curvature (and hencebendingmoment)over the
prop, Fig. 5.20(c). It would appear,therefore, that bending moment would be affectedmore
than displacement.The picturewas broadly similarunderdrainedconditions. WhenE=
Eo+mz,yield causedthe curvatureof the wall aboveFL to disappearand to be replacedby a
slightly increasedcurvaturebelow; 5,,. increasedby over 100%. WhenE' - mz, lesschange
in displacementprofile was observed,with 5n.. virtually unaltered.
The doubly-propped wall imj)osed even greater restraint on the formation of plastic zones in
the retained soil, so it was not surprising that the introdu cti on of Yield had li ttle cff ect on w all
movement above FL for this propping case. However when Eu = mz there was passive yield
in the excavation area, with a substantial kick out at the toe, Fig. 5.20(d), and it was this
which has been picked up in the summary of 5r,. in Table S.1. Under drained conditions, the
wall below FL tended to move out more than above FL, such that the position of 5.
switched from being midway between the props, to being at the toe (for both E' distributions).
Otherwise, profile shapeswere largely unchanged. The value of 8. increased by about 50%.
Similar commentsapplyto the altemately--12rol2ped

wall aswere madefor the unproppedwall.
The additionaloutwardmovementwhenE. = mz was not quite asgreat (approximately80%).
but was still significant. Again, asthe soil parametersimply the possibilityof extensive
5-28
yielding from an earlystage,this figure maybe excessivein termsof what could be realistically
expected- it couldbe consideredto be an upperbound. The final deflectedshapeof the wall
under drainedyieldingwas fairly similarto the drainedelasticanalysis,but with significantly
reducedcurvature(especiallyfor E' = E. +mz, wherecontraflexurejust aboveFL disappeared).
This would be expectedto reducebendingmoments(aswas indeedthe case- seelater). The
position of 8. remainedat the crest,but with largeincreases(up to 200%) in value.
Wall hendingmoment
Bending momentprofilesfor the unproppedwall were affectedonly whenEu= mz, when

+NL.,, increasedby 50% andall negativemomentdisappeared completely,Fig. 5.21(a). The
range between+M.,., and-M. was the same(approx.600 kN-M); yield simplyservedto
displacethe bendingmomentprofile sideways.It was notedearlierthat the displacement
profile for E. = mz hadalteredin sucha way that momentswould be greatlydiminishedover
the upper portion of the wall. Another interestingfeatureseeminglyremovedby yield was the
slightly erratic natureof the bendingmomentprofile below FL, seenin all previousanalyses.
However, therewas a largenon-zeromomentat the toe whenEu= mz (equalto one third of
+M. ) which wasintuitively incorrect. (Non-zeromomentsat the wall toe are studiedfurther
in Chapter8.) The drainedcasewas not really admissiblebecauseof its inherentinstability,
but essentiallythe profile shapewas maintainedwith -Nt. reducedby ;ýs100%and+NL.
more than doubled (when El = E. +mz) or by
increased 3
Fu 0% (when E' = mz).
With top-propping.vield madevery little differenceto the shapeof the bendingmoment

profiles, but did causesomelateraltranslationof the bendingmoment diagram (BMD). When
Eu ý mz, -K= was increasedby 50% andthe positivemomentwas eradicated(NB: exactly
the oppositeof the unproppedcase). The profile shapesdid change,however,when drainage
by
was permitted;principally the virtual eliminationof positivemoment below FL, because
contraflexurewas lost, Fig. 5.21(b). Furthermore,substantialincreasesof 50-60%in
took place,Table5.2.
Yield was largely suppressed

by bottom-proppLng,
andboth the shapeandmagnitudeof the
bendingmomentprofileswere unaffected,exceptfor a small(25%) increasein +Ntu over the
prop whenE,,= mz. Full drainagebrought aboutmore changewhenE'= E. +mz, with +N"
increasedby 60% - but therewere few changeswhenE' = mz. Slight changesin profile were
observed,wherebythe positionof negativemomentmovedfrom aboveFL, to a point below.
5-29
Double-proppingbroughtevenmore suppressionof plasticyield, but only reallyfor undrained
loading. BendingmomentprofileswereidenticalwhenE,,= E. +mz, but still underwenta
significantshift to the positivewhenE,,= mz, Fig. 5.21(c) andTable5.1. Whenloading
conditionsswitchedto drained,positivemomentsover the FL prop increasedsignificantly(by
about 50%) for both E' distributions. Therewas a slight changein bendingmomentprofile,
with more negativemomentbelow FL whenthe soil was permittedto yield.
The shapeof the bendingmomentprofile for the altemately:proppedwall wasunaffectedby

E,,
yield when = E. +mz, althoughtherewas a 40% drop in +AL. over the FL prop. WhenE,,
= mz bending moment in the upper half of the wall disappeared
andtherewas a more
pronounced"spike" in the profile over the FL prop, eventhoughthe actualvalueof +M...
increasedonly modestly. Under drainedloading,the ability to yield broughtabout
significantlydifferentprofiles,e.g. Fig. 5.21(d)for E'= E,,+mz. Above FL, negativemoment
reduced,and +M. over the prop also reduced,by up to 25%.
Excavationheave
When investigatingthe influenceof nonhomogeneity (Section5.2) and anisotropy(Section

5.3), it was observedthat the proppingconditionmadeno differenceeitherto the shapeof the
heaveprofile nor its maximumvalueV.. Changingthe degreeof nonhomogeneity or
did the
anisotropy not affect shapeof the profile either,though it did influencethe magnitude
of uPwardmovement.
An examination of Table 5.1 shows that unpropped, top- and altemately=propnedwalls

experienced an increasein V,,. of about 250% when yield occurs under undrained conditions
for E,, = mz, but an increaseof less than 25% when E,, = E, +mz. For the bottom- and doubly_-
proppedwalls, V. increasedby approximately160%whenE,,= mz, andwas unchanged
when =Eu E,,

+mz. Comparing whole profiles, it is clearthat yield causeda steepergradientof
the excavatedsurfacenearto the wall, with the profile then flatteningoff towardsthe
centreline, e.g. Fig. 5.22(a) for E,, = E. +mz.
Table 5.2 showsthat the increasesin heavecausedby drainedyield for the unpropped Lop-
than in the bottom- and doubly=
: proppedwalls were, agdin, rathergreater
and alternately
proppedcases. The actualpercentage increases were very considerable, rangingfrom 150-
500%. In onecasea maximurnheaveof 250mmwas predictedwhenE'= mz, althoughit is
5-30
acceptedthat manyexcavationsof this sort will, in the long tern4 generallybe coveredby a
slab(e.g. for a carriagewayor basementfloor), and so heavesof this magnitudewould not be
observed. Instead,the tendencyto heavewould give rise to swellingpressureson the
undersideof the slab. Comparingwhole profilesfor correspondingelasticandelastic-perfectly
plastic analysesunderdrainedloadingrevealedthe samesteepergradientnextto the wall; e.g.
Fig. 5.22(b) for the doubly-proppedwall.
Ground surfacemovement
For the unproppedwall, whenE,,= E,+mz insufficientyield took place for thereto be
any significantchangesto the ground surfacemovementprofile. However,whenE. = mz' the
trough immediatelybehindthe wall becamemassivelyaccentuated,droppingvery sharplyto a
value of -S. 10timesgreaterthan for the elasticanalysis.Therewas a 50% increasein +S..
too, but the relativemagnitudewas muchless. Furtherbackfrom the wall, the average
settlementwas approximatelydoubled.
The drainedsurfacemovementprofilesfor both distributionsof E' were similarlyaffectedin a

major way, Fig. 5.23(a). In the there
elasticanalyses, was no downward movementof the
to
surface-S relative originalground level. When yield was allowed, a sharplydefinedtrough
appeared behind the the
wall, although remainderof the profile beyond 10mbackfrom the
wall was largelyunchanged.(It shouldbe recalledthat this wall was on the vergeof yield in
the in-situ condition,so the resultsmaybe consideredsomewhatextreme.)
Predictably,yield hadlesseffect on undrainedsurfacemovementswhenthe wall wasLop

propped. Profile shapeswere largelyunchangedcomparedwith the elasticcase,regardlessof
E,, distribution,Fig. 5.23(b). The magnitudeof -S... and+S. both trebledwhenE,,= mz; no
real changewas evidentwhen = E. E,,
+mz. When conditionsbecamedrained,
the profile
shapechangedjust behindthe wall, showinga sharpupward"kinle' (especiallyfor E'
E. +mz), and a3 00% increasein upwardmovement.
Whenthe wall wasbottom-propMd,the surfacemovementprofile wasnot influencedat all by

plasticyield whenEu= E. +mz, andonly showedsomechangein shapewhenF, = mz. In this
latter case,both -S.. and+S. increasedby over 100%,andthe very localizedtroughjust
behindthe wall becamea little more accentuated,
Fig. 5.23(c). Underdrainedloading,yield
5-31
causeda pronouncedbut localized"dip" in the surfacemovementprofiles,thoughvirtually no
changeat all in +Sm.
-,.
Similar commentsapplyto the doubly-proppedwall; changeswere still confinedto the

stiffnessprofile E,,= mz, whereboth -Sn.. and+S. more than doubled,but therewas no real
evidenceof a troughjust behindthe wall. Yield underdrainedconditions,on the other hand,
had no effect on the profile shapeor magnitude.
The altemately-propped
wall producedtrendsthat were very similarto thoseseenin the
unproppedcasewhenyieldingwas permitted. Very smallchangestook placewhenE,,=
E, +mz, but a very distincttrough formedright behindthe wall whenEu = mz' Fig. 5.23(d); -
S,. increasedby nearly350%, and+S,,, increasedby nearly30%. Ground surface
movements for drainedyield had profilesof similarshapeto the elastic but
analyses, with a
smalltrough in evidencejust behindthe wall. Changesin +S. valuesfluctuatedbetween
+60% (E'= E. +mz) and-20% (E'= mz).
5.5.3 Summary
The introductionof plasticyieldingcanproducemuchlargerwall displacements, but generally

only if the proppingarrangementallowsthe developmentof zonesof plasticyield (which is
not surprising). Hence the bottom and doubly-propped
walls were virtually identicalto their
fully elasticcounterparts.Drainedyieldingwas always"worse' in the sensethat long-term
be
conditionswould more dependent on this aspectof soil behaviour. Another important
point is that the wall cannotbe forced into taking up the same degreeof curvatureif it is
in
embedded an elasto-plasticmaterial. Elastic materials can sustainvery high stresses(infinite
if required),whereaselastic-plasticmaterialswill yield and allow stressredistribution.
Computedbendingmomentprofileswere smoothedrelativeto the fully elasticcase,

sometimesaccompanied by lateral shiftingof the diagram. Yielding hasa more pronounced
effect on the distributionandmagnitudeof bending moments when loadingis drained. In the
context of diaphragmwalls, this is relevantto the long term conditionrather than during
construction. However,a drainedanalysismaybe consideredexcessivelyonerousfor an
embeddedwall in clay, asit is unlikely that full drainagewill occur during the construction
stage. In practice,constructiontakesplaceunderapproximatelyundrainedconditions,
followed by long-termpore pressureequalization(which could give rise to swellingand/or
5-32
consolidation).A key questionfor the designeris whetheror not a drainedanalysisgivesthe
samedeflections,bendingmoments,etc. asanundrainedanalysisfollowed by swelling/
consolidation.Different degreesof partial drainagecould alsobe investigatedvia coupled
loading andconsolidationanalyses.Theseissuesbelongmoreproperlyunderthe headingof
constructionandlong-termmodelling,andare investigatedmore fully in the next chapter.
Introducing yield hada major effect on the magnitudeof heavefor high degreesof elastic
nonhomogeneity,but was rather modestotherwise. The overalldisplacedprofile showed
more uniform heaveacrossthe excavation,asgreaterdistortionof the soil took place
immediatelyin front of the wa.U.
Yielding servesto accentuatesettlementbehindthe wall, althoughthis is only really

pronouncedwhen -+ E. 0 (X = 1). The degreeof propping is important
especially when
consideringbehaviourimmediatelybehindthe wall (saywithin the distanceX= 11).
5.6 Non-associatedFlow
A full implementationof non-associated Mohr-Coulombwas not availablein CRISPuntil after

the relevantanalysesfor this thesishadbeencompleted.Instead,an approximatealternative
was used,which attemptedto enforceequivalenceof the (total stress)Trescaand (effective
stress)Mohr-Coulombyield surfacesin order to studypost-yieldbehaviour. This requireda
priori knowledgeof the Lode 0
angle andan assumption that it remainedconstantthroughout
the analysis.Unfortunatelythis wasdifficult because,although0= 30* at the in-situ stagein
planestrain(due to 'X 'ý #
c; = c; cy'y), thereafter this cannotbe Nonetheless,
guaranteed. it was
decidedto pursuethis approach,andthe first stepwas to establisha procedurewherebythe
two yield curvesmight be equivalenced.This will be illustratedusingparameterSet23 (Table
4.7).
With c,,= 5z, the Trescayield function(Eqn 5.1) becomesf=2. t -I Oz. Equatingthis with
the Mohr-Coulombyield function at yield/failure(Eqn 5.3b) andeliminatingt (= V) gives5z
1/2(crv + a'h)
s'.sin(ý'). Let s= =
(l + K. ). If the unit weight of the soil y= 20 kN/m3,
1/2cr'.,
5-33
the groundwaterlevel is at ground surface,andK. = 2, then s' = (20-1O)z.(I + 2)/2 = 15z.
Hence 5z = 15z.sin(fl, which impliessin(fl = 1/3 andthus ý' = 19.47*.
Using the moregeneralexpressions(Eqns5.2 and 5.5) producesthe sameresult,provided0=

30* whenyield occurs. If cohesionis non-zeroat the surface(i.e. c.. > 0), a simHarprocedure
is followed by keeping fixed and equating cý. + mz = ccosý' + s'siný' At the surface,
. z
0 and this leadsto c' cu,.se4 $I
.
Before performingfull retainingwall analyses,a preliminarycheckwas carriedout on the

procedurefor equivalencingthe TrescaandMohr-Coulombyield criteria presentedabove.
Using a singlequadrilateralelementto simulatea strain-controlledbiaxial compressiontest,
four runs were carriedout all startingfrom the initial stressstate(a Pxo
= app = cr'zo= 100kPa,
uo = 0) andhavingthe sameE and v. Otherdetailsare asfollows:
run yield criterion drainage stresses soil strength

BYI Tresca undrained total cu= 50 kPa
BY2 Mohr-Coulomb undrained effective c'=O,ý'=3011
BY3 Mohr-Coulomb undrained effective c'=50 kPa, V=0
BY4 Mohr-Coulomb drained effective c?=O,ý'=30*
Runs BYI and BY3 were expected to produce identical results, as they describe the same
yield surface (Tresca.being a special caseof Mohr-Coulomb when ý= 0). Graphs of t: e.,
t: &,, and e.:e, for both runs showed first yield at t= 50 kPa.,after which there was no increase
in either t or e,. Runs BYI and BY2 experiencedfirst yield at identical stages(t = 50kPa =
s.siný' = c,) but thereafterBY2 showeda tendencyto dilate (5e, < 0) - resultingin a
significantdrop in pore water pressure,andan increasein the deviatoricstresst carriedby the
sample,Fig. 5.24. Run BY4 showedfirst yield at t= 100kPa, followed by very strong
dilation (8e, < 0) but no changein t, indicatingthat associatedflow did not leadto increased
strength(thoughin someboundaryvalueproblemsthis can occur). On the basisof theseruns
(in particular,BYI andBY2) it was concludedthat TrescaandMohr-Coulombcould be
constrainedto yield at the samepoint but thereafterdisplaythe divergenceexpectedbetween
full (V = fl andzero (W= 0) dilatancy.
5-34
Four pairs of 0 undrained, ,,,wall analyseswere
retainin( then carriedout on both the
unpropped,andthetop-proppedcantileverwall (usingmeshxi Oyl0, Fig. 4.1) to compare
Tresca(total stress)andMohr-Coulomb(effectivestress). In the former E,,= 5z MPa, v,,
0.497, andK. = 0; in the latterE' = 4z, V=0.2, andK,, >>O. The rest of the parametersare
given in Table5.4,in whichY1 indicatesa comparisonof YT I with YE I, etc. In order to
comparethe resultswith thoseobtainedwith a smallerK,,, all analyseswere re-run with K. =I
(but otherwisewith exactlythesameparameters).
Selectedfull profilesaregivenat the endof the chapter,includingwall displacement(Fig

5.25), bendingmoment(Fig 5.26),excavationheave(Fig 5.27) andgroundsurfacemovement
(Fig 5.28). Thevariouspercentagedfferences(Trescarelativeto Mohr-Coulomb- showing
dilation)havebeensummarizedin Table5.5.
the effect of totally suppressing
It is quite apparentthat majordifferencesexistbetweenthe computedresultsobtainedby the

two alternativeapproaches, especiallywhen shearstrengthat the ground surfacewas zero.
Agreementbetweencorresponding pairsof analysesgradually improved as cý. (andhencec')
increased,so that YT4 andYE4 gavesimilarresultsin most cases.This was almostcertainly
becauseyield becamelessdominantandthe resultssimplyreflectedan equivalenceof the total
to undrainedelasticanalysis.In line with earlierobservations,
and effectivestressapproaches
maximumwall displacements andbendingmoments both convergedto a reasonableagreement
fairly quickly, asci.. (andc) wasincreased.Excavationheavesalso showsvery good
agreementat YT4 / YE4, but surfacemovementsstill showedsignificantdifferences.
Becauseyield commenced
very earlyin the analysis(as a consequence
of havingt just below
sj'siný'at all pointsbelow groundlevel),thesedifferencescanbe consideredto arisewholly
out of way in which undrainedplasticyieldingwas represented.However,it is far from clear
if the discrepancies were a consequence principallyof the disparatedilation rates,or of the
different stresspathstakenby elementsof soil in reachingthe yield surface(variationsare
possiblein both the t: s' and 7cplanes).
Effective stresspathsin the Mohr-Coulombanalyseswill be at constants', but someelements

of soil will reachthe yield surfaceafter havingfollowed extensionstresspaths,whilst others
5-35
will havefollowed compressionpaths. An approximateway in which this could be addressed
would be to zonethe mesharoundthe excavation,on the basisof expectedstresspaths.
Elementsof soil immediatelybehindthe wall will undergohorizontal stressreductions,andif
Ko is relativelyhighthesestresschangeswill be elasticasthe elementsgo from passiveto
active states. On the other hand,elementsof soiljust underneathfinal dig level will mainly
undergo vertical stressreductionswith somelateralstressincrease,and if K,, is relativelyhigh
then early passiveyieldingis almostinevitable'. With this zoning,agreementcould be
enforcedbetweenTrescaandMohr-Coulombcompressionpathsbehindthe wall, and
extensionpathsin front of the wall.
Notwithstandingthe above,it seemsmorelikely that the differencesseenbetweenthe Tresca

and Mohr-Coulombanalysesdo indeedstemfrom the amountof dilation which hasbeen
permitted. Whenstressstatesstart so closeto the yield surface,they are unlikely to take
significantly different in
routes reachingit. However, this issue is unlikely to be resolved
without further detailedstudy,which was not possiblein the time available.
ReducingK,, from 2 to I brought about a significantreductionin the differences(apparently)

causedby plasticvolumetricstrain (seeTable5.5). This canbe attributedto the smaller
influencewhich yield hason the results,asmore of the soil canbe expectedto be operatingin
the elasticrange.
5.6.3 Summary
The starting conditions in runs YT1 and YE I were selected so close to yield to be able to
isolate the plastic response, From these limited analyses,it would appear that the
incorporation of non-associatedflow could, in extreme cases,almost double predicted wall
movements and bending major increasesin predicted ground movements

moments and cause
in front of and behind the wall. However, it is unlikely that the potential for this much yield to
occur early on in the excavation process would be present in most real cases. The simplified
in-situ stress conditions in the analysespresentedhere have a constant K. of fairly high
magnitude, whereas in reality K. would reduce with depth until it eventually reachedK.
3 Passiveresistanceis often factoredby 2 on the basisthat passiveconditionsare only mobilized

at largestrains. However,this is only true if initial conditionsare normally consolidated(Y, stsI -sinfl;
whenY%. > 1, the soil may alreadybe closeto passivefailure and so doesnot needmuchadditional
strain to reachit (e.g. Padfieldand Mair, 1984).
5-36
(when OCR = 1). The implicationsfor practicalanalysesmay possiblynot be particularly
serious.
A separateissueraisedby the aboveis how bestto achieveequivalencebetweenMohr-

Coulomb andTrescayield criteria. It is quite commonfor a retainingwall analysisusing
CRISP to be carriedout as a coupledconsolidationevent;to allow actualconstructiontimes
to be followed, andto give the opportunityto examinelong term equilibrationwithin one
in terms of effectivestress-
analysis. This requiresthe input of soil strengthsandstiffinesses
even for the shortterm phase. Onecheckwhich a designermight reasonablewant to carry out
on a coupledanalysisis to run an "equivalent"total stressanalysison the undrained
construction/excavation stage. And yet this may give differentresultsfrom an undrained
analysisbased in yield
on effectivestresses, caseswheresignificant is expectedto occur.
5.7 Discussion and Summary
Returning to the openingquestion- how muchcomplexityis requiredto ensurea realistic

result? Therearea numberof issuesthat canbe investigatedquantitativelythroughnumerical
experimentsof the sort describedhere.
Before choosinga model,the engineermusthavean ideaof the effects(andpenalties)of

adding increasingrealism. Better predictionmust be offset the
against needfor more (and/or
'
higher quality)parameters,more caretakenin analysis,more time (datapreparation,
for of
execution,andpost-processing) analysis,moreaspects output to be checked,etc.
In this sectiononly thosefeaturesavailablein the standardcommercialreleasesof CRISPup

until 1999 have been used (with the exception of elasticnon-linearity). These are the versions
which a designeris more likely to haveat his disposal,but it hasmeant that featuressuchas
kinematichardeninghavebeenomittedfrom this study,eventhoughthey are known to exist
in real soils. However,the findingsreportedin this sectionare no lessusefulbecauseof this.
Constitutive modelling is probably the single largest area of researcheffort in geotechnical

finite elements,and the writer acknowledges that it has an important place in the quest for
better prediction of retaining wall behaviour. But designerswill not and cannot concern
themselves with this crusade for a more complete description of constitutive behaviour - they
5-37
aremore interestedin how bestto applythe eNistingtechnology. Nonhomogeneous linear
elastic-perfectlyplasticmaywell be the most sophisticatedlevelto which practisingengineers
will rise for manyyearsto come.
The usefulnessof really complexmodelsto the designeris questionable.An importantpart of

the designprocessis beingableto developa "feel" for the relativeimportanceof the different
parameters.For example,most engineershavesomeunderstandingof undrainedstrengthasit
can be relatedto consistency,and the useof a singlestrengthvaluein a yield criterion(e.g.
Tresca)is very attractive. Thosemodelswhich needonly a few parameterswith recognizable
physicalsignificancehavea clear advantageover thosewhich requirespecializedtestsand/or
curve fitting procedures(suchasthe Jardineet al. periodiclogarithmicfunctions). It cannot
be disputedthat suchmodels,in the handsof their inventors,canbe very powerfulandthe
literaturecontainsmanyexamplesof good agreementbetweenpredictionandobservation
madeby theseexperts. It is right andproperthat researchanddevelopmentof this nature
shouldbe taking place,becausetherewill alwaysbe a needto makehigh quality,absolute
predictionsin the caseof certainprestigiousor particularlysensitivestructures.
But would a geotechnicalengineerwith somebasicFE literacybe ableto makegood

predictionswith oneof the more complexmodelsavailabletoday? For example,the BRICK
model (Simpson,1992)and3-SKH model(StallebrassandTaylor, 1997)both offer superior
representationof soil behaviour,but are well beyondthe comprehension
of most engineers.
The hyperbolicmodel(Duncanand Chang,1970)enjoysgreatpopularityin North America,
and hasbeenusedwith apparentsuccessto representa wide variety of soilsin a rangeof
engineeringproblems.However,this successreliesupon the calibrationof the modelagainst
observedbehaviour. The debateon the requiredcomplexityof constitutivemodelswill
continuefor manyyearsto come.
The intendedcontributionof this thesisis not to attempta detailedexaminationof available

constitutivemodels,but to the
assist retainingwall in
designer decidingwhich aspectsof soil
behaviourmaybe importantin a given situation. Also, wheretheremaybe one or moreways
in which a given aspectof behaviourcould be modelled,to explorethe implicationsof the
decisionwhich maybe taken. Having consideredgeometricandconstitutivemodelling,the
thesiswill now moveon to considerthe way in which the wall is constructed,bulk excavation
performed, temporarysupportis removed,and long-term equalisationachieved
5-38
Table 5.1 Influenceof constitutivebehaviouron key output values(undrainedloading)
percentagedifferenceswith respectto baselinecase
propping unpropped top bottom double alternate
Wall deflection
Set
00000
2 -32 -38 42 -61 -28
3 138 14 134 -37 143
4 -1 -6 -16 -16 0
5 -32 -35 -52 -52 -29
10 10000
12 32 30 00 22
11 27 18 20 15
13 214 161 1 228 77
Wall bending moment

-tMn=
ýC-t
: --; 7-
M +" +M
-M TM -M -M . --+M--- -M
1 00 00 00 00 00
2 66 59 -17 -3 21 -17 -27 43
-29 -16
3 324 41 141 48 56 43 0 -71 262
-63
4 3 66 6 59 25 16 -2 2 24
5 102 151 2 -5 24 -15 -10 39
-12 -11
10 13 01 00 00 00
12 9 10 00 00 5 40
-17 -7
11 4 10 34 01 00 -34 .9
-20
13 52 45 9 26 -33 45 -78 17
-98 -50
Ercavallon heave V.
set
1 0 0 0 0 0
2 -68 -68 -71 -71 -68
3 48 -50 -59 -57 -51
4 0 -1 .1 0 0
5 -33 -34 -34 -33 -33
10 0 0 0 0 0
12 23 15 0 0 16
11 13 0 0 0 .1
13 272 232 161 159 246
Ground surface settlement

-tS.
Set -S +S -S +S -S +S -S +S -S +S
2 -71 -77 -74 -78 -76 -77 -77 -77 -72 -77
3 -17 -64 -62 -72 -68 -65 -70 -68 -34 -64
4 -4 -2 -6 -3 -7 -5 .7 -4 -4 -2
5 -43 -37 -45 -39 -46 -41 -47 -40 -43 -38
10 00 00 00 00 00
12 19 -21 13 -12 00 00 13 -14
11 121 -13 95 51 00 86 -6
13 960 48 203 198 132 109 147 123 347 27
5-39
Table 5.2 Influenceof constitutivebehaviouron key output values(drainedloading)
percentagedifferenceswith respectto baselinecase
propping =lop bottom double alternate

unpropped
Wall deflection
Set
0 0 0 0 0
2 -5 48 -47 4
-26
3 286 31 561 -29 287
4 -47 -32 -27 -28 -38
5 -87 -74 -79 -75 -81
12 512 187 136 46 203
13 186 184 -8 58 72
Wall bending moment (-, +) M,.,.

Set -M +M -M +M -M +M -M -M +M
00 00 00 00 00
2 44 55 25 19 -8 -29 42
-28 -6 -25
3 263 51 75 . 94 138 59 5 -90 272
-83
4 24 3 -18 -11 -7 -1 -28
-7 -20 -20
5 -27 -39 -52 63 -6 -45 -29 -26 -20 -56
12 123 53 -38 59 -17 46 -67 -15
-90 -53
13 32 63 560 15 -23 50 -93 -23
-98 -48
Excavation heave V..,

Set
0 0 0 0 0
2 -70 -70 -69 -69 -70
3 -53 -50 -50 -49 -51
4 -2 -4 -4 -5 .3
5 -9 -13 -13 -15 -11
12 408 301 153 154 341
13 508 350 178 194 378
Ground surface settlement (-, +)S.

Set
1 n/a 0 n/a 0 n/a 0 n/a 0 n/a 0
2 -78 -77 -77 -77 -78
3 -68 -69 -65 -66 -66
4 -2 -5 -4 -5 -3
5 -8 -13 -13 -15 -10
12 68 7 5 61
-12
13 1 2 -21
-21 -21
Note: n1aIndicatestherewas only heave(+S) and no settlement(-S) in drained analyses
540
Table 5.3 Parameters in to the
used analyses estimate influenceof non-linearity(Jardineet
al model)
Series A* c Enün 6* Enm a
(0/.) (0/.) (0/.) (lY.) n= 1 n=2 fl =3 fl =4

1=0.0 y=0.07 1=0.14 1=0.21
NLI 4.00 0.0007 0.001 0.01 0.1 2.4188 2.3946 2.3706 2.3467
NL2 4.00 0.0007 0.001 0.1 1.0 2.4188 2.2823 2.1536 2.0321
Table 5.4 usedin analysesto estimatethe influenceof non-associated

Parameters flow
Run yield criterion c, (kPa) c' (kPa)
YTI Tresca 5z -
YT2 99 10 + 5z -
YT3 cc 20 + 5z -
YT4 44 50 + 5z -
--------- -------------- ---------- --------- ---------
YEI Mohr-Coulomb - 0 19.5
YE2 (.9 - 10.6 19.5
YE3 99 - 21.2 19.5
YE4 cc - 53.0 19.5
5-41
Table 5.5 Uluence of non-associated flow on key output values: percentagedifferences
betweenTresca(zero dilation) andMohr-Coulomb(fiA dilation)yield criteria
Propping top-propped
unpropped
K. =2 Y%.
=2
-7 K.=I
Wall dej7ection
Run
Yi 91 10 45 -13
Y2 30 34
-11 .8
rY3 19 26
-2 -4
Y4 6 0 8 -11
Wall benchngmoment
-t-M,,.
Run -M +M +M +M -M +M
-M -M T3
Yi 86 40 20
-96 -94 -13 -4
Y2 17 5 14 14 -3
-8 -6 -2
Y3 7 20 27 10 -2
-16 -2 -22
Y4 9 0 14 6 -1
-3 -4
I
Excavafion heave V.
Run
Yi 61 48 -18
-5
Y2 48 35 -6
-12
Y3 37 25 -3
-4
Y4 9 1 -3
-3
Ground surface settlement S,.

-+
Run +S +S +S -S +S
-S -S -S
Yi 288 68 - 63 33 2 -14
-8 -2
Y2 78 52 8 -3
-47 -31 -7 -10
Y3 42 2 45 -13 11 -5
-48 -4
Y4 17 8 22 -12 12 -8
-19 -7
Arote: Case Yj refers 10 he co,,nparisonofrun yTj with yv, in which thepercentage

. difference would be calculatedas 100x (TyTI Fad lFrE, etc. 7hetabulated
-
valuesShawthe effect of suppressingdilation at yeld - in the Trescacriterion
V=0, whereasin theMohr-Coulomb ;v= ý'
542
ground surface
E
++/+ //A\V
very low stiffness

at shallow depth
element bounclades
z-
(a) Soffstiffhessprofile (b) UppermostGausspoints
Fig 5.1 Low stiffness near the ground surface for profile E= mz
I
deviator stress
deviator strain es
Tangent stiffness Eng., = dq/des = n. a(es)'I-l

i
Secant stiffness = Aq/Aes = a(es)
Fig 5.2 Power-law model for non-linear elastic response(Gunn, 1993)
5-43
CYV av A
element B
ACYhI'll 0
Aav<0
TSP elementA
Aah4 0 TSP
0
Aa, ste
Ch CTh
Fig 5.3 Typical stresspathsfor elementsof soil adjacentto a retainedexcavation
or projected, maximum
JObserved.
EulCu A+8 cos Ja (log, 0 #/C)I
28
Peak deviator (and of stillnoss data)

J.
Axial st(sin
010-3 - 0 ..............
10-2 IL
0.1 Jou - 10
aos.
, -. o.
Projected minimum E Projected
function
Fig 5.4 Pcriodic-logarithnýcfunctionmodelfor non-lincarclasticresponse(aflcr

Jardineel at 1985)
,
544
(c, cotV + s) sin(V
St
(a) (effectivestress)
Mohr-CoUlomb
Cu
f= 2A - 2.cu
S
(b) Tresca(total stress)
Fig 5.5 The TrescaandMohr-Coulombyield loci in t: s invariant space
t t Incrementalplastic
UP,S strain vector
UPv=
S
(a) Tresca (zero dilation V" =
Incrementalplastic
r
strainvector
tp I 8CPS
8CPV
st
Mohr-Coulomb(full dilation y'- ý')
Fig 5.6 Plasticstrainvectorsfor the TrcscaandNfohr-Coulombyicld loci in vs

invariantspace(assurningan associatedflow rule)
545
0.(
-10.c
c
-20.0
u
m
-30.0
IA
.pq
13
P-4
P-4 -40.0
m
3
-50.0 , 8.0 12.0

0.0 4.0 16.0
Depth beloW top of wall (n)
4 Eu: Eo * Eu : Eo 4 liz Li -. KZ
(a) cantileverwall / undrainedloading
0.1
-3.( . ... . ..... .......
-10A
-13.0
-20.0
0.0 4.0 1.0 12.0 11.0

Depth below top Of wall (n)
4 Eu: ED * Eu: Eo 4 NZ + lu - KZ
(b) top-proppedwall / undrainedloading
Fig 5.7 Influenceof nonhomogeneityon horizontalwall displacement: sclectcd

profiles
546
10.0
0.(
%0
-10.C
-20.0
-30.0
-40.0- 11.0
0.0 4.0 8.0 16.0
E' : Eo * El : Eo + mz +r-. KZ
-EF
bottom-proppedwall / drainedloading
-7.3
M
P.4
0.0 4.0 1.0 12.0

Depth
nk--nfh below
hnlnu top
too o'
of Uall Cn )
EU: EO * Eu: Eofnz + cu-- KZ
(d) doubly-propped wall / undrained loading
Fig 5.7 Influenceof nonhomogencityon horizontalwall disPlaccmcnt: sclected

profiles(contd)
5-47
750
-500.0, 8.0 12.0 16.0

4.0
v -- Eo * E, : EO# III EI - nz
(a) cantilever
wall / drainedloading
200,
%W
-200A
.1i
c
AMC
-600.0
P4 -800.0
P-4
m
-1000.0L- III
0.0 4.0 1.0 12.0 11.0
Depth below top of wall (11)
-13- Eu : E0 * Eu : Eo f mz Eu-. mi
(b) tOP-Proppedwall / undrainedloading
Fig S.8 Influence of nonhomogencity on wall bending moment : sclcctcd prorilcs
548
1400.0
1200,0
1000.0
.X
800.0 .......
600.0 ........
400.0
200.0
0.0
-20010
-400.0 8.0 13.0 11.0

CIO 4.0
-EF Eu Eo * Eu: ED# NZ -#- Eu: mz
(c) bottom-proppedwall / undrainedloading
1000.1
.X 600.1
%0
400.1
200.t
CA
-200.0
pq
P-4 -400.0
0 1
-6W.0.0 4.0 1.0 12.0 11.0
F Eo *V- Eo # nz +V-. KZ
-& --
(d) wall / drainedloading

altemately-propped
Fig 5.8 Influenceof nonhornogeneityon wa bendingmomcnt: selectcdproriles

(contd)
549
IUU.(
80A
a' 60.0
a'
0 40.0
U 20.0
x
Lii
0.01
8.0 -4.0 -2.0 0.0
-6.0 (n)
Distance from wall
-a- Eu-- Eo * Eu-- Eo4 nz -+. Eu- tiz
(a) top-proppedwall / undrainedloading
100.
80.1
E
60.1
C
C
0 40.(
I
tu
tu ...............
ci 20A
x
UI
0.0
4V: Ea * El -_Eo# mz -f- V -. KZ
(b) top-proppedwall / drainedloading
Fig 5.9 Influenceof nonhomogeneityon excavationheave: selectedprofiles
5-50
30.(
20.1
E
10.0
0
u 0.0
L
3
IA
------- ..... ..
-10.0
-20.0.
00 10.0 20.0 30.0 40.0
4k Eu: Ea * Eu.-Eo #Piz + Eu-. KZ
(a) cantileverwall / undrained,loading
30A
20.(
C
Ui
Ui 10A
0
£
Ui
S
I
U CIO . ___
In
_ ___
-10.0
I
I.
UI
-20.0, 20.0 30.0 40.0

0.0 10.0
-& Eu: Ea * Eu- Ec4 nz Eu-. mi
(b) top-proppedwall / undraincdloading
Fig5.10 Influenceof nonhomogeneityon ground surfacemovement: selected

profiles
5-51
30.1
20.( ... ...........
10A .............
0.0
.....
........
..........
..............
IA
-10.0 ...........
-20.0, 20.0 30.0 40.1

10.0
Distance Tron waai %no
Eu-- Ea * Eu: EO*KZ -f- Eu: Kg
(c) bottom-proppedwall / undrainedloading
60.1
50.(
40.C
30.0
19
-- -----
------------------
20.0
10.0
0.0 ,I
0.0 10.0 20.0 30.0 40.
.0 Distance fron (n)
wall
-& V -- EO *V -- Eo# mz -+- r: "I
(d) altematcly-proppcdwall / draincdloading
Fig5.10 Influenceof nonhomogcncityon ground surfacemovement: selected

profiles(contd)
5-52
10.
S.
0.
V
-3.1I-
#A
.0.4 -10.(
13
mw
-13.0
-20.0,
0.0 4.0 1.0 11.0 16.0
4 FWPV -- I + EIWVY-IV NT 4
(a) cantileverwall / drainedloading
3.1
0.(
-3.0
u
m
P-4
12 -10.0 - -------- -------
PA
.pj
13
P14
-15.0
-20.0,
0.0 4.0 1.0 11.0 11.0
-G- DAV + DAV: 2 DAv :3
(b) top-proppedwall / undrainedloading
Fig 5.11 Influenceof anisotropyon horizontalwall displaccmcnt: sclcctedprofiles
5-53
3.1
2.5
0.0
u
m
P-4
-2.3
13
P14
P-4 -3.0
-7.5 ,
0.0 4.0 8.0 12.0 11.0
-fk EI
h/p v 1 + E'Wry -- I --r- El
NT v:
--
(c) bottom-proppedwall / drainedloading
2.1
0.1
-2.5 .............
-3.0
-7.5
*0
p4
p4
-10.0
-12.5,
0.0 4.0 1.0 12.0
Depth below top of' wall Cn)
-a- + Mu -. 2 ql. DAv. - 3
(d) doubly-proppedwall / undrainedloading
Fig 5.11 Influcnceof anisotropyon horizontalwall displaccment: sclectcdprorilcs

(contd)
5-54
200.
.X#
%.
0.1
. ...........
-200.[
-300.C
P.
-I -400.0
P-1
0.0 4.0 1.0 12.0 11.0

EMU 1+ DAY -- 2 Mv -3
-13- --
(a) cantileverwall / undrainedloading
200.
.Z0.1
-300.C
-400.0
P-4
-600.0
0.0 4.0 1.0 U. 0 16.0
-EFEl
WE'
v :I+ EIWY :I r tvT v*4
(b) top-propped wall / drained loading
Fig5.12 Influenceof anisotropyon wall bendingmoment: selectedprofiles
5-55
600.0
400A
.X
200.0
0.0
-200.0
-400.0 '
0.0 4.0 8.0 12.0 11.0
+ FWVV: 2 Q6 VWVV: 4
-&
(c) bottom-proppedwall / drainedloading
1000.1
.X
%OF WO.(
. ..... ..........
12
P4 -200.0
-400.0 1.0 12.0

0.0 4.0 11.0
Depth below top of wall (1j)
Mv :1+ DAV: I .96
(d) doubly-proppedwall / undrainedloading
FigS. 12 Influenceof anisotropyon wall bendingmoment: selectedprofiles (contd)
5-56
60.0
50.0
%dp
40.0 .......... ................... ..

. ..........................
30.0 . .... ...... .......... ................

.
20.0
0.0 : 1
8..0 -6.0 -4.0 -2.0 0.0
DAV -- I+ Eh/Ev-- I Mv -3
(a) top-proppedwall / undrainedloading
100.
El
80.1
%.0
60A
40.C
----------
20.0
.0 -6.0 -4.0 -2.0 0.0

oistance rron waxi L"j
EI
WV 1 + VWU --I rWrV. - 4
-B- v: -q6
(b) altemately-proppedwall / drainedloading
Fig 5.13 Influenceof anisotropyon excavationheave: selectedprofiles
5-57
JU. 1
%0 20.t
10.C
al
0.0
-10.0
ci
-20.01 20.0 30.0
0.0 10.0 40.0
+ DAY: 2 .96 DAV
-0- _-
(a) bottom-proppedwall / undrainedloading
60.
50.1
.4-'
U
U 40A
0
U
ci 30.C
(0
In
20.0 . .... ......... . ..
.4-'
I.
U
10.0,
0.0 10.0 20.0 30.0 40.1
Distance rrom Wall Fni
-& VtVE'u :I + EIWVv -2 4- rwrv. - 4
(b) cantileverwall / drainedloading
Fig S.14 Influenceof anisotropyon ground surfacemovement: sclectedprofilcs
5-58
Eu/ Eu(-)
3
a and y selectedto
force all E, / Er)
EO
curves through
2 commonpoint
£min r* emax log (ca
Emin = minimum strain, below which Eu Is constant at a maximum value

C* = reference strain, at which C2= EutEu(-)=I for all curves
emax = maximum strain, above which Eu Is constant at a minimum value
Fig 5.15 Normalizedstiffness: straincurvesusedin non-fincaranalyseswith the

Jardineel al. (1986) model
5-59
0.0
-23.0
-50.0
E
a)
u
to
P4
-75.0
P4
-IW. o
-123.0- 4.0 1.0 12.0 16.0

0.0
01) -I
Eu(.001)/Eu(. * :2 -#- -3+
(a) unproppedwaH/undrained /Series NLI /E. - = mz
0.(
%00
-5.c
-10.0
P.4
a
41
44
10
p4 -15.0
P-1
-20.0, 4.0 1.0 12.0 16.0

0.0
1) :I
Eu(.001)/Eu(. *: I+-3+4
(b) top-proppedwaH/ undrained/ SeriesNL2 / E,,. -=
Fig. 5.16 Influenceof non-linearityon horizontalwall displacement: selectedprortles
5-60
700.
600.1
500.1 ....................
%0
400.0
300.0
0
E
200.0 ----------
10 100.0
J2 0.0
P-4
------------------
-100.0
4ý A
-iGuu. u
0.0 4.0 8.0 12.0 16.0
-0- Eu(. I)
001)/Eu(. :I* -- 2 4- -- 3+: 4
(a) cantileverwall/ undrained/Series NL2 / Eu- - Eo+mz
251
x
%0 -25(
4.'
-501
............ .. .........
-1000
P4 -1250 -II -- - ------------

P4
I
0.0 4.0
4. n 1.0 11.0
12.0 16.1
Depth below top Of wall (n)
Eu(.DM)/Eu(.M) :1*-2+: 3
(b) top-proppedwall/ undrained/Series NLI IF,. - 13ý
Fig. 5.17 Influenceof non-linearityon wall bendingmoment: sclectedprofiles
5-61
50.
........
...........
40.
z
. ..............
30.1 ... ....... .
a;
C
a 20.(
tu
tu
ci 10.C
x
UI
0.01
8.0 -6.0 -4.0 -2.0 0.0
-& Eu(.
001)/Eu(.01) :1 * :2 -#- -- 3+4
(a) cantfleverwaff /undrained /Series NLI /Eu- = mz
25.1
I
I
20A xx -*CY--- ------
_IX
..
10.0
DA'
0.0 2.0 4.0 6.0M
a 1)
-9- Eu(.001)/Eut. 1
Distance
*: 2+3+
from wall (n)
--
(b) cantileverwall / undrained/ SeriesNL2 / rj,. - mz
Fig. 5.18 Influenceof non-linearityon excavationheave: selectedprofiles
5-62
2.1
0.1
E
(U
-3.0
IA
w
u
-10.0
-15.01
0.0 10.0 20.0 30.0 40.0
Distance fron wall (m)
Eu(.001)/Eu(.M) :1 * :2 -#- :3 4
(a) cantileverwall/ undrained/Series NLI /E, * = Eo+mz
10.
5.
4,
U
p4 0.1
4,
4,
U
U
U
-5.(
-10.0, 30.0
0.0 ln. n 20.0 40.0
.0 fron (n)
Distance wall
-fl- Eu(.
001)/Eu(.
I) --I *-2+ :3+ :4
top-proppedwall / undrained/ SeriesNL2 / E,,- - mz
Fig. 5.19 Influenceof non-linearityon ground surfaccmovement: selectcdproriles
5-63
%.0
-60.1
Ul
"q
13
p"
-120.0
-150.01
0.0 4.0 8.0 12.0 16.0
-8- EmTu
=0 * Eu/Cu: 500 1000
(ON
M cantileverwall / E,, = mz / undrainedloading
-$.I
%OR
-10.1
-15.C
. p4
13
P-1
p4
-20.0
-25.0L- 8.0 12.0

0.0 4.0 16.0
e= im, o, -- 25-1 +e: c, w
(b) top-proppedwall /E= E. +mz /drained loading
Fig. 5.20 Influenceof yielding on horizontalwall displacement: selectedprofiles
5-64
10.0
0.0
E
W
u -1010
to
PA
.P4
13
P4 -20.0
P4
-30.01 4.0 8.0 12.0 16.0

0.0
Euleu=0 * EuMu: 500 + EAU : 1000
-G-
(c) bottom-proppedwall / E, = mz / undrainedloading

ý
5.0
0.0
-510
cu
-10.0
-15.0
0.0 4.0 8.0 12.0 16.0

-I]- EUIN :0* EWICU-- 500 ELM -- 1000
(d) doubly-proppedwall / Eu = rm / undrainedloading
Fig. 5.20 Influenceofyielding on horizontalwall displacement: selectedprofiles

(coutd)
5-65
000
600.0
.X
400.0
200.0
0.0
-200.0
-400.0 '
0.0 4.0 8.0 12.0 16.0
-Eý ELM =0 * Eu/Cu: 500 + EuMu-- 1000
(a) cantileverwall / E. = mz / undrainedloading
200.0
0.0
(U -200.0
E
-400.0
-600.0
P4
-800.01 ,"I
0.0 4.0 8.0 12.0 16.0
e= moc, f= 25# +e=c, f= 2s,
(b) top-proppedwall /E'= E,,+mz /drained loading
Fig. 5.21 Influenceof yieldingon wall bendingmoment: selectedprofiles
5-66
125(
1000.0
75010
500.0
0
E
250.0
13 0.0
c
V
A
P-4 250.0
P-4
m
3
0.0 4.0 8.0 12.0 16.0
-a- EU/0j
:a* : 500
EU/CU + EU/CU
- 1000
(c) doubly-propped wall / E,, = mz / undrained loading
400.0
300.0
.X
200.0
100.0
0.0
m
-100.0
12
P4
-200.0--------- --
---- ----------- . --J(L
-300.0
0.0 4.0 8.0 12.0
-o- e= im, jr = 25* +e: o, f: 251
(d) alternately-propped.wall /E'= E. +niz /drained loading
Fig. 5.21 Influence of yielding on wall bending moment : selectedprofiles (contd)
5-67
25.0
20.0
E
E
%0
13.0
m
w
Ja
c
0 10.0
5.0
x
w
CIO.
-4.0 -2.0 0.0
-8.0 -6.0
-EF EuMu:0 * Eu/Cu= 500 -- 1000
+ Eu/Cu
(a) cantileverwall / E. E, / loading

= )+mz undrained
200.0
150.0
%.0
100.0
50.0
L-
0.0 0.0
8.0 -6.0 -4.0 -2.0
e= 1000,IV = 25- -#- e=o, o, = 25*
(b) doubly-propped wall /E'= mz/ drained loading
Fig. 5.22 Influenceof yieldingon excavationheave: selectedprofiles
5-68
13.0
E 10.0
%OF
41
c ... ...... ......
3.0
0.0
--- -----
------
--- ---------
------------------
-----------------
L -5.0
D
IA
-10.0
IV
D
-is.n
-0.0 L- 10.0 20.0 3u.u 40.0
e= 1000, IV = 250 +e: c, jv = 25#
(a) cantileverwall /E'= E,,

+mz /drained loading
7.5
p..
£
5.0
w
U 2.3
0
U
U 0.0
-2.3
I.
Cl
0.0 10.0 20.0 30.0 40.0

-fl- ELVCu
=0* Eu/W = 500 Etvcu: 1000
(b) top-propped wall / E. = E,,+mz / undrained loading
Fig. 5.23 Influenceof yieldingon groundsurfacemovement: selectedprofiles
5-69
20.0
C IM
I
a) 10A
C
0
C 3.0
a)
U
0.0
4J -5.0
0.0 10.0 20.0 30.0 40.0

4 EwTu=0* EtVCu= 500 Eu/Cu
: 1000
(c) bottom-proppedwall / E. = mz / undrainedloading
20.1
10.0
L -20.0
2
IA
4; -30.0
L
-40.0 .I
0 10.0 20.0 30.0 40.0
.0
4y EwTu :0* EwTu = 500 EwTu: 1000
(d) alternately-propped wall / E,, = mz / drained loading
Fig. 5.23 Influence of yielding on ground surface movement : selectedprofiles

(contd)
5-70
H 1
31
CL
.Y
v
41 4C
,A
IA
3C
41
IA
u
-pq
L
20
10
1.
0 ir- ________ ________ _________________ ________
0.00 0.10 0.20 0.30 0.40 0.50

deviatoric strain fs Cv. )
run BYI * run BY2
(a) deviatoricstress: deviatoricstrain
0.50
UI
x 0.40
w
0.30
.1.1
In
C., 0.20
I.
0
0.10
w
L--
0.00
0.moo 0.0005 D.OMO 0.0013 0.0020 0.0025
volumetric strain Ev CV.)
nin BYI * run M
(b) deviatoricstrain: volumetric strain
Fig. 5.24 Stress and strain responsesin biaxial test simulation analyses
5-71
0.1
-10.(
-20.1
m
P14
im
-30.0
Iq
. P4
13
P4
N-4 -40.0
m
-30.01
0.0 4.0 8.0 12.0 16.0
run Wl * run VE2 -0- run W3 + run VE4
(a) effectivestressanalysis(Mohr-Coulombyield criterion: full dilation)
-10.0
-20.0
P-4
-30.0
P4
P-4 -40.0
L-
-50.a 4.0 8.0 12.0 16.0
0.0
-& run Yrl * runYFZ + run Yr3 + nin YM
(b) total stressanalysis(Trescayield criterion : zero dilation)
Fig. 5.25 Influence of flow rule on horizontal wall displacement : top-propped wall
undrained loading / K, =2
5-72
VLRJ.
.X
%0
400.0 ......
200.0 .... ............. ...........

..
--- -------------------
---------
-----
P-4
-200.0
0.0 4.0 8.0 12.0 16.0
run M run M+
* run VE3 + run NE4
(a) effectivestressanalysis(Mohr-Coulombyield criterion: full dilation)
80
600.0 -
Z
X
(U 400.0
C 200.0 .............
,Pq
(U
J2 0.0
-200.0--
0.0 4.0 8.0 12.0 16.0
-B- run YU * run Yr2 + run Yr3 + run Yr4
(b) total stress analysis (Tresca yield criterion: zero dilation)
Fig. 5.26 Influence of flow rule on wall bending moment : cantilever wall undrained
loading / K. =I
5-73
100.
Pl
80.
ci 60.1
tj
cl In
C
C 40.( .... .......

.p
U 20.C
x
Iii
0.0 .
-8.0 -6.0 -4.0 -2.0 0.0
run VEL * run VE2 -0- run NO + run W4
(a) effectivestressanalysis(Mohr-Coulombyield criterion : full dilation)
100
80.(
(0
60.C
40.0
20.0
0.01
8.0 -6.0 -4.0 0.0
-2.0
-& run Yll * run Yr2 run VT3 run Yr4
-0-
total stressanalysis(Tresca yield criterion: zero dilation)
Fig. 5.27 Influenceof flow rule on excavationheave: top-proppedwall / undrained

loading/ K. =2
5-74
10.1
0.(
- ----------
--
'.
-20.(
-----------
-30A
V
-40.0
-50.0
0.0 5.0 10.0

Ila 15.0 20.0
Distance behind wall (n)
run VU * run M+ run VE3 + run VE4
(a) (Mohr-Coulomb
effectivestressanalysis yield criterion: full dilation)
10.c
0.0
£
"
41
c
-40.0
-50.0
10.0 15.0 20.0

Distance behind wall (m)
-0- run YrI * run W2 + run W3 nin Y14
(b) total stress analysis (Tresca yield criterion: zero dilation)
Fig. 5.28 Influenceof flow rule on ground surfacemovement: cantileverwall

undrainedloading / K. =I
5-75

Application of FEM On Retaining Walls

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Application of FEM On Retaining Walls

Încărcat de

Drepturi de autor:

Formate disponibile

THE APPLICATION OF FINITE ELEMENT

ANALYSIS TO THE DESIGN OF

RICHARD IAN WOODS

A Thesissubmittedfor the degreeof

Q Richard Ian Woods, 2003

insightinto the behaviourof embeddedretainingwalls canbe providedby

andto identify critical designparameters (the Uhat if' approach),andto obtainrealisticand

motivationfor using finite elementsin retainingwall designhasarisenfrom increased

constructionactivity in urban areas,and the needto demonstrate

Finite elementanalysisof embeddedwalls presentsspecialchallenges.For example,the

Systematicnumericalinvestigationshavebeenconductedinto eachof theseareas.Basedon

I would like to thankProfessorChrisClaytonfor encouragingme to do this research

A numberof organizationshaveprovidedmewith opportunitiesto be involvedwith

Last but by no meansleast,I mustrecordmy deepestgratitudeto my wife Lesleyand

CHAPTER I GENERAL FVMODUCTION

CHAPTER 2 GENERAL BACKGROUND

CHAPTER 4 GEOMETRIC MODELLING AND DISCRETIZATION

CHAPTER 6 MODELLING OF CONSTRUCITONAND LONG TERM EFFECTS

CHAPTER 7 COMPUTATIONAL DIFFICULTIES

CHAPTER8 OBTAINING REQUIRED DESIGN OUTPUT

CHAPTER 9 CONCLUSIONS AND RECOMMENDATIONS

A. percentage error in computedstress

2x2 (reduced)numericalintegrationrule for 2-D elements

AGS Associationof GeotechnicalSpecialists

CRISP CRItical State Program

1.1 Historical Background

company, and not just a single-stepelastic analysiswith a few tens of elements.

Exponentialprogressin the computerindustryhasbroughtthe 486 andPentiumprocessors

"The new marketfor finite elementswill be dominatedby the ordinarydesigner.

strong intuitive senseof what is happening physically."

Writing nearly 10 years later, Zienkiewicz asserted:

"The revolutionbroughtaboutby the combinationof numericalmethodsandthe

This statementmaybe contrastedwith a morerecentarticleby Clark (1994),which describes

...... finite elementanalysisis no longerthe preserveof stressengineersand

Someauthoritieshaveexpressedgraveconcernover the releaseof sophisticatedsoftwareinto

1.2 Finite Elements in Geotechnical Design

The earlyfocusof finite elementmethodsin the late 1950swaswithin structuralengineering

Most (if not all) reportedapplicationsof geotechnicalfinite elementsin the earlyyears

Someof thesevendors(notablyCambridgeandDelft) recognizedthe needfor educatingusers

- andnot just sellingthem powerful software andorganized training

Of course,it is not merelythe availabilityof the hardwareandsoftwarewhich hasprovided

1.3 Background to Present Work

Typicalrequirementshavebeento estimatewall deformationsandassociatedground

Oneotherpublicationwhich appearedin the sameyear asthe originalNAFEMS guidelines

"... SophisticatedFE programsincorporatingelasto-plasticmodels are

A subsequent publicationof relevancewas the Associationof GeotechnicalSpecialistsGuide

At the time of writing, a revisionof CHUA 104(CP96)hasjust beenreleasedto a restricted

(a) the ývhatif'approack wherethe sensitivityof a particularcalculatedvalueis

(b) the 'absolute'approach,wherevaluesarerequiredto be realisticand accurate,

There is an increasingtrend towards using FE analysesin the secondway. Programmers are

experience of the problem being solved.

* geometry is usually complex, and changesincrementally

o both short and long term behaviour is required

e therearehigh contrastsin materialstiffness

accuratepredictionsof groundandwall movements,structuralforces,andprop/anchor

In the courseof his consultancywork and otherindustrialcontacts,the writer hasencountered

Whenusingfinite elementsfor retainingwall analysistherearemanyfactorswhich are

a knowledgeof the relevantfactors,andhow they caninfluencethe results

0 of whereandwhen (andwhy) problemsare known to haveoccurred

0 strategiesfor checkingfinite elementoutput,to help ensurethat the resultsare reasonable

Another strandto the backgroundof the presentwork camefrom variouscommunications

a) bugswithin the codeitself (programnot worldng asstated)

Obviously,(a) is beyondthe control of the user(thoughhe/sheis not without responsibility),