Geotechnical Risk and Safety

GEOTECHNICAL RISK ANDSAFETY
PROCEEDINGS OF THE 2ND INTERNATIONAL SYMPOSIUM ON GEOTECHNICAL

SAFETY & RISK, GIFU, JAPAN, 1112 JUNE, 2009
Geotechnical Riskand
Safety
Editors
Y. Honjo
Department of Civil Engineering, Gifu University, Gifu, Japan
M. Suzuki
Center for Structural Safety and Reliability, Institute of Technology,
Shimizu Corporation, Tokyo, Japan
T. Hara
F. Zhang
Department of Civil Engineering, Nagoya Institute of Technology, Nagoya, Japan
Cover photo: Traditional cormorant fishingat theNagaraRiver, GifuCity, J apan
Courtesyof GifuSightseeingAssociation
Taylor & Francis is an imprint of the Taylor & Francis Group, an informa business
2009Taylor & FrancisGroup, London, UK
Typeset byCharonTecLtd(A MacmillanCompany), Chennai, India
PrintedandboundinGreat BritainbyAntonyRowe(A CPI-groupCompany), Chippenham, Wiltshire
All rightsreserved. Nopart of thispublicationor theinformationcontainedhereinmaybereproduced,
storedinaretrieval system, or transmittedinanyformor byanymeans, electronic, mechanical, by
photocopying, recordingor otherwise, without writtenprior permissionfromthepublisher.
Althoughall careistakentoensureintegrityandthequalityof thispublicationandtheinformationherein,
noresponsibilityisassumedbythepublishersnor theauthor for anydamagetothepropertyor
personsasaresult of operationor useof thispublicationand/or theinformationcontainedherein.
Publishedby: CRC Press/Balkema
P.O. Box447, 2300AK Leiden, TheNetherlands
e-mail: [email protected]
www.crcpress.com www.taylorandfrancis.co.uk www.balkema.nl
ISBN: 978-0-415-49874-6(Hbk)
ISBN: 978-0-203-86731-0(eBook)
Geotechnical Risk and Safety Honjo et al. (eds)
2009 Taylor & Francis Group, London, ISBN 978-0-415-49874-6
Tableof Contents
Preface IX
Organization XI
Sponsors XIII
Wilson Tang lecture
Reliabilityof geotechnical predictions 3
T.H. Wu
Keynote lectures
Riskassessment andmanagement for geohazards 13
F. Nadim
Riskmanagement anditsapplicationinmountainoushighwayconstruction 27
H.W. Huang, Y.D. Xue &Y.Y. Yang
Recent revisionof J apaneseTechnical Standardfor Port andHarbor Facilitiesbased
onaperformancebaseddesignconcept 39
T. Nagao, Y. Watabe, Y. Kikuchi &Y. Honjo
Special lecture
InteractionbetweenEurocode7 Geotechnical designandEurocode8 Designfor
earthquakeresistanceof geotechnical structures 51
P.S. Sco e Pinto
Special sessions
Reliability benchmarking
Reliabilityanalysisof abenchmarkproblemfor 1-Dconsolidation 69
J.Y. Ching, K.-K. Phoon &Y.-H. Hsieh
Studyondeterminationof partial factorsfor geotechnical structuredesign 75
T.C. Kieu Le &Y. Honjo
Reliabilityanalysesof rockslopestability 83
C. Cherubini & G. Vessia
Reliabilityanalysisof abenchmarkproblemfor slopestability 89
Y. Wang, Z.J. Cao, S.K. Au & Q. Wang
Geotechnical code drafting based on limit state design
and performance based design concepts
DevelopingLRFDdesignspecificationsfor bridgeshallowfoundations 97
S.G. Paikowsky, S. Amatya, K. Lesny &A. Kisse
V
Limit StatesDesignconceptsfor reinforcedsoil wallsinNorthAmerica 103
R.J. Bathurst, B. Huang &T.M. Allen
Lossof staticequilibriumof astructure Definitionandverificationof limit stateEQU 111
B. Schuppener, B. Simpson, T.L.L. Orr, R. Frank &A.J. Bond
Geotechnical criteriafor serviceabilitylimit stateof horizontallyloadeddeepfoundations 119
M. Shirato, T. Kohno & S. Nakatani
Reliability-basedcodecalibrationof pilesbasedonincompleteproof loadtests 127
J.Y. Ching, H.-D. Lin & M.-T. Yen
Sensitivityanalysisof designvariablesfor caissontypequaywall 135
G.L. Yoon, H.Y. Kim, Y.W. Yoon & K.H. Lee
Evaluatingthereliabilityof aleveeagainst seepageflow 141
Y. Shimizu, Y. Yoshinami, M. Suzuki, T. Nakayama & H. Ichikawa
Determinationof partial factorsfor theverificationof thebearing
capacityof shallowfoundationsunder openchannels 147
A. Murakami, S. Nishimura, M. Suzuki, M. Mori, T. Kurata &T. Fujimura
Applicationof concept inGeo-code21 toearthstructures 155
M. Honda, Y. Kikuchi &Y. Honjo
Limit statedesignexample Cut slopedesign 159
W.K. Lin & L.M. Zhang
Probabilisticchartsfor shallowfoundationsettlementsongranular soil 165
Resistancefactor calibrationbasedonFORM for drivensteel pipepilesinKorea 173
J.H. Park, J.H. Lee, M. Chung, K. Kwak & J. Huh
Anevaluationof thereliabilityof verticallyloadedshallowfoundationsand
grouped-pilefoundations 177
T. Kohno, T. Nakaura, M. Shirato & S. Nakatani
Studyonrational groundparameter evaluationmethodsfor thedesignand
constructionof largeearth-retainingwall 185
Y. Yamamoto, T. Hirose, M. Hagiwara, Y. Maeda, J. Koseki, J. Fukui &T. Oishi
Systemreliabilityof slopesfor circular slipsurfaces 193
J.-Y. Ching, Y.-G. Hu & K.-K. Phoon
Correlationof horizontal subgradereactionmodelsfor estimatingresistance
of pilesperpendicular topileaxis 201
Y. Kikuchi & M. Suzuki
Risk management in geotechnical engineering
Thelongandbigtunnel fireevacuationsimulationbasedonanacceptable
level of riskandEXODUSsoftware 211
S.-Q. Hao, H.-W. Huang &Y. Yuan
Riskbaseddecisionsupport systemfor thepumpingprocessincontaminated
groundwater remediation 217
T. Hata &Y. Miyata
VI
A riskevaluationmethodof countermeasurefor slopefailureandrockfall
withaccount of initial investment 221
T. Yuasa, K. Maeda &A. Waku
Riskassessment ontheconstructionof interchangestationof Shanghai metrosystem 229
Z.W. Ning, X.Y. Xie & H.W. Huang
Challengesinmulti-hazardriskassessment andmanagement: Geohazardchain
inBeichuanTowncausedbyGreatWenchuanearthquake 237
L.M. Zhang
General sessions
Design method (1)
A studyof thenewdesignmethodof irrigationpondsusingsheet materials 247
M. Mukaitani, R. Yamamoto, Y. Okazaki & K. Tanaka
Researchonkeytechniqueof double-archtunnel passingthroughwater-erodedgroove 251
Y. Chen & X. Liu
Safetymeasuresbyutilizingtheoldridgeroadandpotential risksof land
near theoldriver alignments 257
M. Okuda, Y. Nakane, Y. Kani & K. Hayakawa
Bearingcapacityof rigidstripfootingsonfrictional soilsunder eccentricandinclinedloads 265
K. Yamamoto & M. Hira
Uncertainty
Reliabilityanalysisof slopestabilitybyadvancedsimulationwithspreadsheet 275
S.K. Au, Y. Wang & Z.J. Cao
Optimal movingwindowwidthinconjunctionwithintraclasscorrelation
coefficient for identificationof soil layer boundaries 281
J.K. Lim, S.F. Ng, M.R. Selamat & E.K.H. Goh
Soil variabilitycalculatedfromCPT data 287
T. Oka & H. Tanaka
Reducinguncertaintiesinundrainedshear strengths 293
J.Y. Ching, Y.-C. Chen & K.-K. Phoon
A casestudyonsettlement predictionbyspatial-temporal randomprocess 301
P. Rungbanaphan, Y. Honjo & I. Yoshida
Construction risk management
Reliabilityanalysisof ahydraulicfill slopewithrespect toliquefactionandbreaching 311
T. Schweckendiek, G.A. van den Ham, M.B. de Groot, J.G. de Gijt, H. Brassinga & P. Hudig
A casestudyof thegeological riskmanagement inmountaintunneling 319
T. Ikuma
Guidelinefor monitoringandqualitycontrol at deepexcavations 327
T.J. Bles, A. Verweij, J.W.M. Salemans, M. Korff, O. Oung, H.E. Brassinga &T.J.M. de Wit
A studyontheempirical determinationprocedureof groundstrengthfor
seismicperformanceevaluationof roadembankments 333
K. Ichii &Y. Hata
VII
GeoRiskScan Gettinggripsongeotechnical risks 339
T.J. Bles, M.Th. van Staveren, P.P.T. Litjens & P.M.C.B.M. Cools
Reductionof landslideriskinsubstitutingroadof Germi-Chaydam 347
H.F. Aghajani & H. Soltani-Jigheh
Risk assessment
Probabilisticriskestimationfor geohazards: A simulationapproach 355
M. Uzielli, S. Lacasse & F. Nadim
A researchproject for deterministiclandslideriskassessment inSouthernItaly:
Methodological approachandpreliminaryresults 363
F. Cotecchia, P. Lollino, F. Santaloia, C. Vitone & G. Mitaritonna
Reliability-basedperformanceevaluationfor reinforcedrailwayembankments
inthestaticloadingcondition 371
M. Ishizuka, M. Shinoda &Y. Miyata
Maximumlikelihoodanalysisof casehistoriesfor probabilityof liquefaction 379
J.Y. Ching, C. Hsein Juang &Y.-H. Hsieh
Suggestionsfor implementinggeotechnical riskmanagement 387
M.Th. van Staveren
Design method (2)
Frameworkfor evaluationof probabilityof failureof soil nail system 397
I.S.H. Harahap &W.P. Nanak
Reliabilityanalysisof embankment damsusingBayesiannetwork 405
D.Q. Li, H.H. Liu & S.B. Wu
Identificationandcharacterizationof liquefactionrisksfor high-speedrailwaysinPortugal 411
P.A.L.F. Coelho &A.L.D. Costa
Fieldcharacterizationof patternsof randomcracknetworksonvertical andhorizontal soil surfaces 419
J.H. Li & L.M. Zhang
Stochasticmethodsfor safetyassessment of aEuropeanpilot site: Scheldt 425
M. Rajabalinejad, P.H.A.J.M. van Gelder & J.K. Vrijling
A report by JGS chapter of TC23 limit state design in
geotechnical engineering practice
Codecalibrationinreliabilitybaseddesignlevel I verificationformat for geotechnical structures 435
Y. Honjo, T.C. Kieu Le, T. Hara, M. Shirato, M. Suzuki &Y. Kikuchi
Author index 453
VIII
Preface
ISGifu(2ndInternational SymposiumonGeotechnical RiskandSafety) whichisheldon11and12J une, 2009
atNagaraInternational ConventionCenter inGifu, J apan, isasymposiumheldaspartof aseriesof conferences
organizedbyagroupof peopleinterestedingeotechnical risk andsafety. TheseconferencesincludeLSD2000
(November 2000, Melbourne, Australia), IWSKamakura(April, 2002, TokyoandKamakura, J apan), LSD2003
(J une, 2003, Cambridge, USA), Georisk 2004 (November, 2004, Bangalore, India), Taipei2006 (November,
2006, Taipei), and the 1st International Symposiumon Geotechnical Risk and Safety (1st ISGSR, October,
2007, Shanghai). Besides theseevents, this grouphas organizedtechnical sessions inmany international and
regional conferencesfromtimetotime.
Themajor themesof thissymposiumare:
Evaluationandcontrol of uncertaintiesconcerninggeotechnical structures.
Performancebasedspecifications, RBDandLSDof geotechnical structures, anddesigncodedevelopments.
Riskassessment andmanagement of geo-hazards.
Riskmanagement issuesconcerninglargegeotechnical constructionprojects.
Repair andmaintenancestrategiesof geotechnical structures.
IS Gifuissponsoredby ISSMGE, J GS andGEOSNet. Twotechnical committeesnowworkinginISSMGE
are taking lead in this symposium, namely, TC 23 Limit state design in geotechnical engineering practice
(chair Y. Honjo) andTC 32Risk assessment and management in geotechnical engineering practice (chair F.
Nadim). Theorganizers greatly appreciatethesupport provided by theJ apaneseGeotechnical Society (J GS)
onthis symposium. ASCE Geo-InstituteRAM (RiskAssessment andManagement Committee) has alsobeen
involvedinpromotingthissymposium.GEOSNet(Geotechnical SafetyNetwork)isatopic-specificinternational
platformtofacilitateandpromoteactiveinteractionontopicsrelatedtogeotechnical safetyandriskamongthe
members, particularlybetweenresearchersandpractitioners. GEOSNetwasformedatTaipei 2006inviewof the
increasinginterest andmomentumtorationalizerisksinnewdesigncodesusingreliabilityandother methods.
GEOSNet isexpectedtotakeover thisactivityandbecomeapermanent bodytoorganizethisseriesof ISGSR
conferences. For thisreason, wecall ISGifualsothe2ndInternational SymposiumonGeotechnical Safetyand
Risk(2ndISGSR).
Oneof theimportant eventsrelatedtoGEOSNet inthissymposiumistheinitiationof WilsonTangLecture
series. Thelectureisnamedtorecognizeandhonour theseminal contributionsof Professor WilsonTang, who
isoneof thefoundingresearchersingeotechnical reliabilityandrisk. GEOSNet planstohost theWilsonTang
Lectureas thekey presentationinfutureISGSR events tohonour distinguishedpeers andtheir achievements.
ThefirstWilsonTanglectureisdeliveredbyProfessorT.H. Wuof OhioStateUniversity, whoisalsooneof the
foundingresearchersinthisdomain.
Finally, theorganizers aregrateful to all thosewho havehelpedandcontributedto theorganizationof this
event.A largepartof thecreditfor theproceedingsgoestotheauthorsandreviewers.Thepublicationcostof the
proceedingsissupportedbyGrantinAidforScientificResearch, ScientificResearch(A) entitledDevelopment
and promotion of performance based design and reliability based design of geotechnical structures (Grant
No. 19206051, Y. Honjoastherepresentativeresearcher). Theorganizersaredeeplyindebtedfor thisfinancial
support.
YusukeHonjo
MakotoSuzuki
Takashi Hara
FengZhang
J une2009, Gifu, J apan
IX
Organization
ISSMGETC23 Limit State Design in Geotechnical Engineering Practice
Honjo,Y. (Chair) Phoon, K.K. (CoreMembers)
Zhang, L.M. (Secretary) Schuppener, B. (CoreMembers)
Becker, D.E. (CoreMembers) Simpson, B. (CoreMembers)
Matsui, K. (CoreMembers) Steenfelt, J. (CoreMembers)
Paikowsky, S. (CoreMembers)
ISSMGETC32 Engineering Practice of RiskAssessment and Management
Nadim, F. (Chair) Leroi, E. (CoreMembers)
Fenton, G.A. (Secretary) ManfredNussbaumer, E.H. (CoreMembers)
Bekkouche, A. (CoreMembers) Pacheco, M. (CoreMembers)
Bolle, A. (CoreMembers) Phoon, K.K. (CoreMembers)
Ho, K. (CoreMembers) Roberds, B. (CoreMembers)
J aksa, M. (CoreMembers)
GEOSNet Board Members
Phoon, K.K. (Chair) Honjo,Y.
Becker, D.E. Horikoshi, K.
Chin, C.T. Huang, H.W.
Faber, M.H. Simpson, B.
Organizing Committee
Honjo,Y. (Chair) Nishimura, S.
Hara, T. Ogura, H.
Honda, M. Oishi, M.
Horikoshi, K. Okumura, F.
Kikuchi,Y. Ohtsu, H.
Kimura, T. Rito, F.
Kobayashi, K. Satake, M.
Kobayashi, S. Shirato, M.
Kusakabe, O. Suzuki, H.
Maeda,Y. Suzuki, M.
Matsui, K. Ueno, M.
Mizuno, H. Yamamoto, K.
Murakami, A. Yamamoto, S.
Nishida, H.
Scientific Committee
Suzuki, M. (Chair) KieuLe, T.C.
Hara, T. Zhang, F.
XI
Local Advisory Committee
Asaoka, A. Ohtani, T.
Daito, K. Okumura, T.
Hara, T. Rokugo, K.
Hinokio, M. Sato, T.
Honjo,Y. Sawada, K.
Itabashi, K. Shibuki, M.
Kamiya, K. Sugii, T.
Kodaka, T. Sugito, M.
Kojima, S. Tsuji, S.
Ma, G. Yamada, K.
Maeda, K. Yashima, A.
Nakai, T. Yasuda, T.
Nakano, M. Yoshimura,Y.
Narita, K. Yoshio, O.
Noda, T. Zhang, F.
Nojima, N.
International Review Panel
Akutagawa, S. Murakami, A.
Becker, D.E. Nadim, F.
Calle, E. Nagao, T.
Ching, J.Y. Nishida, H.
Coelho, P. Nishimura, S.
Cotecchia, F. Notake, H.
Fujita, M. Ohdo, K.
Furuta, H. Orr, T.
Han, J. Otani, J.
Hara, T. Paikowsky, S.
Harahap, I.S.H. Qunfang, H.
Heidari, S. Rajabalinejad, M.
Honda, M. (Makoto) Saito, T.
Honda, M. (Michinori) Scarpelli, G.
Horikoshi, K. Schuppener, B.
Huang, H. Schweckendiek, T.
Ichii, K. Shirato, M.
Karlsrud, K. Staveren, M.
Katsuki, S. Sutoh, A.
Kikuchi,Y. Suzuki, H.
Kitahara, T. Tafazzoli, N.
Kobayashi, A. Taghavi, A.
Kobayashi, S. Takada, T.
Kojima, K. Thomson, R.
Kusakabe, O. Uzielli, M.
Lee, S.R. Vessia, G.
Li, D. Wakai, A.
Li, X.Z. Wang,Y.
Lo, R. Yamaguchi,Y.
Maeda, K. Yamamoto, S.
Maruyama, O. Yoon, G.L.
Miyata,Y. Yoshida, I.
Mori,Y. Yoshinami,Y.
Moriguchi, S. Zhang, J.
Mukaitani, M. Zhang, L.
XII
Sponsors
organizedbytheJ apaneseGeotechnical Society
under theauspicesof theInternational Societyfor Soil
MechanicsandGeotechnical Engineering
withsupport of geotechnical safetynetwork(GEOSNet)
XIII
Wilson Tang lecture
Reliabilityof geotechnical predictions
TienH. Wu
The Ohio State University, Columbus, Ohio, USA
ABSTRACT: Thispaper reviewstheuseof probabilisticandstatistical methodsinreliabilityanalysis. Results
fromreliabilityestimatesarecomparedwithobservedperformancetoprovideameasureof thereliabilityof the
methodsandimproveunderstandingof theparameters. Casehistoriesof failuresinclaysareexaminedindetail.
1 RELIABILITY ANDGEOTECHNICAL
DESIGN
The concepts of risk and uncertainty are familiar
to geotechnical engineers. Since predictions are not
perfect there is always the possibility of failure or
unsatisfactory performance. Structures are designed
sothat theriskof failureisacceptablysmall. Thetwo
principlecomponents of geotechnical design arethe
calculatedrisk (Casagrande1965) andtheobserva-
tional method. Thefirst is described by Casagrande
as(1) theuseof imperfectknowledgetoestimatethe
possibleranges for all pertinent quantities that enter
intoasolution and(2) thedecisiononanappro-
priatemargin of safety, or degreeof risk taking into
considerationlossesthatwouldresultfromfailure.
Wheretheconsequenceof failureis largeandacon-
servativedesign is expensive, Terzaghi proposed the
useof theobservational method, whichisdescribed
byPeck(1969) as, "Basethedesignonwhateverinfor-
mationcanbesecured. Makeadetailedinventory of
all the possible differences between reality and the
assumptions. Then, computeon thebasis of original
assumptions, variousquantitiesthat canbemeasured
inthefield. Onthebasisof theresultsof suchmea-
surements, graduallyclosethegapsinknowledgeand,
if necessary, modify thedesignduringconstruction."
Theseconceptscanbematchedwithwell knownrela-
tions inreliability anddecisionmakingas illustrated
inFigure1and2.
InFigure1, Part(1)estimatethepossiblerange
isrepresentedbytheprobabilitydensityfunctionf (s),
wheres =theshearstrength.InPart(2),thedecision
isbasedonminimizingtheexpectedcost E(C), given
asEq.(1).InPart(3),anappropriatemarginof safety
correspondstoP
f
andlossesthat wouldresult from
failuretoC
f
inPart (4).
Figure2showstherelationshipbetweentheobser-
vational method and Bayes theorem, Eq. (2). Ele-
ment (1), Computeonthebasis of original assump-
tions, various quantities that can bemeasured in the
field, isanalogoustoP[z
i
|x
j
], wherez =performance
and x =soil property. In Element (2), the original
Figure1. Thecalculatedrisk.
assumptions is analogous to P
[x
j
], the prior prob-
ability. Element (3), on the basis of the results
of such measurements, gradually close the gaps in
knowledge, corresponds to P
[x
j
|z
i
], the posterior
probability or updated information. Element (4), if
necessary, modify thedesign, is adecision process
as describedinFig. 1. Thefollowingsections review
methods for estimating the failure probability and
updatingbyBayes theorem.
2 UNCERTAINTIESINGEOTECHNICAL
DESIGN
Uncertaintiesindesignparametersandmethodslead
to errors, which may be randomor systematic. Soil
variability is the most important source of random
error and inaccuracies or simplifications in material
and analytical models are the common sources of
systematicerrors.
Theresult is errors in predictions, represented by
the bias N, which is the ratio of the correct answer
to theprediction. N has mean N, standard deviation
[N] andcoefficient of variationO[N]. Soil variabil-
ity due to spatial variations is represented by O(x),
wherex =property.ThereisawiderangeinO(x),from
O(s
u
)=0.100.50for theundrainedshear strengthto
O(K)=2.03.0 for permeability (Lumb, 1974). The
variabilityof thepermeabilitycoefficientisthelargest
andcanrangeover morethanoneorder of magnitude
withinasoil deposit.
3
Figure2. Theobservational method.
However, in geotechnical engineering practice,
uncertaintyaboutthesubsoil stratificationisoftenthe
major issue. Terzaghis writings (e.g. Terzaghi 1929)
arefull of warnings about complications insitecon-
ditions. A well known exampleis thesoil profileat
Chicopee Dam, MA, shown in Fig. 3a. Much less
known is Fig. 3b, which shows the data Terzaghi
usedto construct Fig. 3a. Theuncertainties involved
intheextrapolationareobvious. Terzaghi recognized
the difficulty of predicting the in-situ permeability
at Chicopee Dam and, in an early example of the
observational method, provided specific instructions
onmeasurement of theseepagefromthedam. Errors
in stratification inferred from site exploration data
canbeconsideredasmappingandclassificationprob-
lems(BaecherandChristian2003).Themajorobstacle
is the difficulty of obtaining the necessary geologic
parameters.
A soil property model is often used to transform
results of laboratory or in-situ tests to the property
used for design to represent in-situ behavior of the
structure. Transformation models, or material mod-
els, are approximate and contain errors. Analytical
modelswhichincludelimit equilibriummethodsand
finite element methods (FEM) also contain errors
becauseof simplifications. Botharemodel errorsand
aresystematic.
Reliability analysis provides arational method of
evaluatingthesafetyof astructurethataccountsforall
theuncertainties.Theoriginof reliability-baseddesign
can betraced to partial safety factors. Taylor (1948)
explainedtheneedtoaccount for different uncertain-
tiesaboutthecohesional andfrictional componentsof
soil strengthandLumb(1970)usedthestandarddevia-
tiontorepresenttheuncertaintyandexpressedthepar-
tial safetyfactorsintermsof thestandarddeviations.
Inreliabilityanalysisthefailureprobabilityis
Freudenthal (1947) may be the first to provide
a comprehensive formulation of P
f
for structures.
Since the functions are complex, numerical integra-
tionisrequired.Theuseof first-order-second-moment
(FOSM) methodismoreconvenient.Animprovement
over FOSM is the use of response surface, or first-
order-reliability method (FORM). More elaborate
are simulation and stochastic methods. Simulation
has been done with stability analysis and FEM and
stochasticFEM. It isinstructivetocompareresultsof
reliabilityanalysiswithobservedperformance.Thisis
presentedinthefollowingsectionsforfailureinclays.
3 FAILURE INCLAYS
Stability of slopes on soft to medium clay under
theundrained condition is awidely studied problem
becauseit is simpleandalso becauseslopes failures
in clay are perhaps the most common among fail-
uresingeotechnical construction. Manycasehistories
areavailableandvariousreliabilityanalyseshavebeen
made.
3.1 Soil Variability
Modelsfor soil variabilityrangefromsimpleparame-
terstostochastic fields. A commonmodel represents
the mean by a trend, usually taken as a function of
depth z, and randomdepartures fromthetrend by a
standarddeviation, (s), or acoefficient of variation
(COV), O(s). Soil variability is spatially correlated
with correlation distances
x
,
z
for horizontal and
vertical directions. To account for spatial correlation
avariancereductionI
A
is usedandtheaverageof s
overaregionA hasmeanandCOV, s
A
=s, O(s
A
)=I
A
O(s), respectively(Vanmarcke1977).
For generic soil types, Rojiani et al (1991) and
PhoonandKulhawy(1999a) havesummarizedknown
data on O(s) and
x
,
z
. There is ample evidence
(Wu 2003, Wang and Chiasson 2006) that the trend
is not constant even within a small region. Exam-
ples of soil variability aregiven inTable1. Thereis
a large range in . One should note that data on
x
and
z
werevery limiteduntil the1980sandTanget
als (1976) estimatewas basedonresults fromblock
samples. If =2m(PhoonandKulhawy1999a), then
O(s
A
)0.10. Also, because of the small number of
samples, thereisanuncertaintyaboutthemeanvalue,
givenas O
o
. Thelarge for theJ ames Bay siteis to
account for thelargesegment of theslipsurfacethat
passes through a layer of lacustrine clay. This illus-
tratesthesignificanceof . Alsoshownarethevalues
4
(a) Permeability Profile at Chicopee Dam, MA. (Terzaghi and
Peck 1948) Reprinted with permission from John Wiley and
Sons.
(b) Data used by Terzaghi to construct (a)
Figure3. ChicopeeDam, MA.
fromPhoonandKulhawy(1999a) forclaysingeneral.
Uncertainty about soil stratification, or the subsoil
model, isnot includedbecauseof inadequatedata.
3.2 Model Errors
Model errorsaresystematic andincludeerrorsinthe
soil propertiesmodel andtheanalytical model. Biasin
thepropertymodel hasbeeninvestigatedbylaboratory
andin-situteststhatcomparedthepropertiesmeasured
bydifferenttests. Examplesincludetheestimateof s
u
fromresults of unconfined compression tests, vane
shear tests or conepenetration tests. Transformation
modelsandtheirreliabilitiesaresummarizedinPhoon
andKulhawy(1999b).Theengineer mustevaluatethe
resultsandexercisejudgment inestimatingthemean
andCOV of themodel error. It issubjective; thetrue
model error isof courseunknown. Table2givessome
examplesof estimatederrorsof strengthmodels. Dur-
ingearlier timestheunconfinedcompressiontestwas
thecommontest usedtomeasuretheundrainedshear
strength of soft to mediumclays. Thetwo estimates
inTable2differ inthefactors consideredandinthe
biasN
m
, andO(N
m
). Theestimatefor vaneshear test
wasbasedonthelaboratorystudiesbyLefebvreet al
(1988) onthesamesoil.
Bias in analytical models has been investigated
by comparison of results of simple models used
in practice with those of more sophisticated ones.
For limit equilibriumanalysis of slope stability, the
simple models range fromthe Fellenius method to
the Morgenstern-Price method. Sophisticated meth-
ods include FEM (Zou et al 1995) and limit anal-
ysis (Yu et al 1998). Examples of estimated model
errors for four sites are given in Table 3. The esti-
matederror for limitequilibriumanalysisisN
a
1.0,
O(N
a
)0.05. If the Fellenius method is included,
a
1.14, O(N
a
)0.05. Another factor is the dif-
ference between the plane-strain analysis and three-
dimensional analysis. The first two cases inTable 3
haveslidemassesof limitedwidth,wherethe3-deffect
is more important, while the embankment at J ames
Bayislongandaplane-strainfailureislikely.Theland-
fill atKettlementHillshasacomplicatedgeometry.A
comparison for an embankment on Bangkok Clay is
giveninCase3,Table4. Despitethedifferentsitecon-
ditions and thetimespan between theestimates, the
estimated model errors arenot very different. There
isverylittledifferencebetweenlimit equilibriumand
limit analysesfor exampleslopeswith30
--45
.
3.3 Combined Uncertainty
Thecombineduncertainty represents theerror of the
predictionmodel
Using theupper and lower limits of thevalues in
Tables13, thecombinedpredictionerrorisestimated
5
Table1. UncertaintiesduetoDataScatter.
Site Soil s kPa O(s) O
0
(m) O(s
A
) Ref.
Chicago upper clay 51.0 0.51 0.083 0.2 0.096 Tanget al.
middleclay 30.0 0.26 0.035 0.2 0.035 1976
lower clay 37.0 0.32 0.056 0.2 0.029
J amesBay lacustrineclay 31.2 0.27 0.045 40
deGroot andBaecher 1993

slipsurface
0.14 24 0.063 Christianet al. 1994

Generic clay 0.100.55 16 PhoonandKulhawy1999
x
,
For circular arcslipsurfaceandH =12m.

Table2. UncertaintiesduetoErrorsinStrengthModel.
Model Errors, N
mi
O(N
mi
) N
m
,
sample stress sample strain progress. O(N
m
)
Site, Test disturbance state anisotropy size rate failure Total Refer.
Detroit unconf. 1.15, 0.08 1.00, 0.03 0.86, 0.09 X X X 0.99, 0.12 Wu& Kraft
clay comp. 1970
Chicago
upper unconf. a) 1.38, 0.024 1.05, 0.03 1.00, 0.03 0.75, 0.09 0.80, 0.14 0.93, 0.03 Tanget al.
clay comp. b) 1.05, 0.02 1976
middle a) 1.38, 0.024 1.05, 0.03 1.00, 0.03 0.93, 0.05 0.80, 0.14 0.97, 0.03
clay b) 1.05, 0.02
lower a) 1.38, 0.024 1.05, 0.03 1.00, 0.03 0.93, 0.05 0.80, 0.14 0.97, 0.03 1.05, 0.17
clay b) 1.05, 0.02
Labrador field X 1.00, X X X X 1.00,0.15 Christian
clay vane 0.15 et al. 1994
(a) stresschange, (b) mechanical disturbance
Table3. ErrorsinAnalytical Model.
Analysis Model Error N
ai
, O(N
i
) N
a
, O(N
a
) Ref.
3-D Slipsurf. Numerical Total
Detroit, Cut =0 1.05,0.03 0.95,0.06 X 1.0, 0.067 WuandKraft 1970
Chicago, cut =0 X X X 0.98, 0.087 Tanget al. 1976
J amesBay, embankment =0 1.1,0.05 1.0,0.05 1.0,0.02 1.0,0.07 Christianet al. (1994)
Kettlement Hills, landfill c
1.1,0.16 X X 1.1,0.16 Gilbert et al. 1998

Example =0 limit eq. limit analysis 0.9, -0.01 Yuet al. 1998
tobeN 1.0, andO(N)=0.130.24forlimitequilib-
riumanalysis withcircular arc slipsurface. For total
stress analysis withs
u
, O(F
s
)=O(N). Notealsothat
soil variabilityisnot theonlysourceof uncertainty.
4 CASE HISTORIESOF FAILURES
Casehistoriesof failuresinclay areusedtocompare
theobservedperformancewithreliabilitypredictions.
This provides some measure of the reliability of
geotechnical predictions.
4.1 Reliability analysis
Table4andFig. 4showtheresultsforanumberof well
documentedcasehistories. Consider first theanalysis
byFOSM andFORM withlimit equilibriumanalysis
andwithFEM,Cases1,5,6and7.Almostall analyses,
exceptCase7,consideronlysoil variability.Thediffer-
encesinP
f
areminor.Whensimulationwasdonewith
arandomtrend, Case8, slightlyhigher failureproba-
bilitiesareobtained.Notethattheauthorscalculations
areapproximatebecauseof varioussimplificationsof
details.
Theresultsof FEM simulationarelessconclusive.
TheP
f
for St. Hilliare, calculated with thecharts of
GriffithandFenton(2004),liesbetween0.02and0.05,
depending on the spatial correlation model. This is
considerably smaller than those fromsimpler mod-
els. For the J ames Bay Dike, the FEM, which does
notincludespatial correlationbetweenelements,gives
P
f
=0.07. This is much larger than that obtained by
6
Table4. SafetyFactorsandFailureProbabilities
ReinforcedEmbankments
Site F
s

P
f
Method Reference
1 Muar Clay 1.22 limit eq. Chai & Bergado
stable 1.29 1.61 0.05 FOSM
c
1993
1.29 1.84 0.03 FORM Low&Tang1997
2 BangkokClay 1.00 limit eq. Bergadoet al 1994a
failure 1.00 limit eq. Low&Tang1997
Embankments
3 BangkokClay 1.03 limit eq. circular, ps Bergadoet al 1994b
failure 1.07 3d
0.98 limit eq. non-circ, ps
1.06 3d
0.60.8 PROBISH, (FOSM?)
dependson
x
4 BangkokClay, 0.96 limit eq. Bergadoet al 1994a
NongNgooHao
failure 1.00 FEM+limit eq. Zouet al 1995
0.99 limit eq. Low&Tang1997
5 J amesBay 1.53 2.60 0.002 limit eq. FOSM, O(j)=0 Christianet al 1994
stable 1.45 2.66 0.009
O(j)=0.07
2.71 0.003 limit eq. FORM, circular Xu& Low2006
1.18 0.12 non-circ.
1.49 0.07 FEM, FORM
1.46 0.005 limit equi. +simulation El-Ramlyet al 2002
6 Example 1.28 1.86 0.03 limit eq. FOSM
c
Low2008
1.96 0.025 FORM
Cut slopes
7 Chicago,
observedfailure 1.18 0.9 0.20 limit eq. FOSM, Tanget al 1976
redesignedslope 1.29 0.05
8 St. Hilliare 1.15
a
0.50
c
0.31
c
limit equi. j=0.85
e
Lafleur et al 1988
failure 1.30
b
0.78
c
0.22
c
FOSM
1.05
a
0.58 Stoch. limit eq. Wang& Chiasson2006
1.14
b
0.44
x
-10m
1.2
c
0.020.05
c
Stoch. FEM Griffith& Fenton2004
=1020m
=reliabilityindex.
a
=45
,
b
=34
,
c
=calculationsbyauthor,
d
correspondsto =20m,
e
j=correctionfor vaneshear test
FOSMandFORM. Also, theslipsurfacedepartscon-
siderably fromthe circular arc used in FOSM and
FORM, andis closer to thenon-circular slipsurface
analyzed by Spencers method, which has F
s
=1.18.
TheestimatedF
s
for theFEM solutionis1.3(Fig. 4).
Inbothcases, onewouldexpectthespatial correlation
distance, especiallyinthehorizontal directiontobean
important factor. At J amesBayalargesegment of the
slipsurfacepassesthroughthecomparativelyweaker
lacustrineclay layer and is agood illustration of the
influenceof subsoil stratification. It shouldbenoted
that theabovecases all havefairly well defined soil
strata. Thestudiesconcentratedontherandomdepar-
turesfromthetrend.Insomecases,uncertaintiesabout
thelayer thicknesswereconsidered.
Fig. 4 shows that generally, P
f
>0.10 for fail-
ures. The uncertainty about soil strength, O(s
A
), for
these sites is approximately 0.01. The largest com-
bineduncertainty O(N)=0.24. Whentheseareused
in the FOSM the calculated range in P
f
versus F
s
is shown by two curves, which includes most of the
case histories. The data in Figure 4 show that the
USArmyCorpsof Engineers (1995) designrequire-
ment that P
f
-310
5
shouldbeadequate. Another
sourceof empirical dataistheestimateof failureprob-
ability based on historic records. Meyerhof (1970)
estimated a range from P
f
=0.05 for O(F
s
)=0.3
to P
f
-10
3
for O(F
s
)=0.1. This includes a wide
rangeof soilsanddesignsbyunknownmethods, with
largely unknown safety factors. If one assumes that
therangeof thesafetyfactorisfrom1.31.5forall the
structures, theestimatedP
f
isshowninFig4.Therela-
tionshipbetweenF
s
andannual failureratehas been
estimatedfromperformancerecordandexpert judg-
ment(Silvaetal 2008).TheareashowninFig. 4isfor
CategoryIII andIV, whichincludefacilitieswith-
outsite-specificdesignandwithlittleornodesign.
Thehistoric records includeall types of failures and
7
Figure4. P
f
versusF
s
soliddotsdenotefailures.
arenot thesameas thecases inTable4, which per-
tainto failureintheundrainedstate, or immediately
or shortlyafter construction.
4.2 Prediction of failures
Therearemanywell-documentedcasehistoriesof fail-
ures, especiallyof slopesonclays.Theobservedfactor
of safetyof afailureis1.Theratioof observedtocalcu-
latedfactorsof safety providesanempirical measure
of the prediction bias N and O(N) is a measure of
uncertainty. A distinctionshouldbemadeonthetype
of prediction as definedby Lambe(1973). TypeA
predictions arethosemadebeforeobservationsType
B predictionsarethosemadeduringobservationsand
TypeC predictionsarethosemadeafter observations.
Consider first Type C predictions. Bishop and
Bjerrum (1960) summarized 27 slope and bearing
capacity failures inclays andvalues of F
s
calculated
with s
u
as measured by theunconfined compression
test. Thepredictionerror or bias, N =1/F
s
, hasmean
andCOV of N =1.01, andO(N)=0.06respectively.
Theprediction error O(N) is much smaller than the
rangegiven in Sec. 3.3. Tavenas and Lerouil (1980)
summarizedalmost 60slopefailures andtheF
s
cal-
culatedwiths
u
measuredbythevaneshear test. Their
results giveN =1.03, O(N)=0.17. Thesearecloser
totherangegiveninSec. 3.3.
Good examples of the reliability of Type A pre-
dictions are the symposia at Massachusetts Institute
of Technology(1975) andat KualaLumpur, Malaysia
(MalaysianHighwayAuthority 1989), 14years later.
Ineachcase, well respectedpredictorswereaskedto
predict theheight H
p
at failureof atest embankment
onclay. Theactual heightatfailure, H
f
, wasunknown
tothepredictors.Theobservedsafetyfactor is1.0and
themodel biasisN =H
f
,H
p
.
The values of N and O(N) for the two symposia
are given in Table 5. There are differences between
Table5. TypeA Predictions
MIT H
p
mean O(H
p
) N O(N)
H
f
=71.7ft 70.6ft 0.10 1.03 0.08
Malaysia
H
f
=5.4m 4.0m 0.14 1.12 0.07
the information given to the predictors at the two
symposia that are worth noting. At the MIT sym-
posium, observed embankment performance during
Stages 1and2wereknownandpredictors usedit to
calibrate their models before making the prediction
of the additional height required to produce failure.
However, theshear strengthat theendof stage2was
not known and had to be estimated. Table 5 shows
the observed and predicted embankment heights at
failure. Only the seven predictions that used ratio-
nal models to estimatetheshear strength areshown
in Table 5. Most predictors used limit equilibrium
analysis andafewusedFEM. Despitedifferent pre-
dictionmodels andassumptions, theresults arevery
consistent withN =1.03andO(N)=0.08. However,
althoughthesiteconditionsareconsideredtobewell
defined, Leonards(1982) suggestedtheprobablepres-
ence of a weak layer. Depending on the strength of
this layer and the failure surface, the safety factor
couldbebetween0.351.20, whichwouldimply that
N 1.3andO(N)0.25. This is onemoreexample
of theimportanceof thesubsoil stratificationandits
influenceonthefailuremode.
IntheMalaysiansymposium, theembankmentcon-
struction took 100 days and this was not known
to the predictors. All predictors used the initial
undrainedshear strengths. Thefiveinvitedpredictors
all predictedheights smaller thantheobserved, with
N =1.12. Theunderestimateof H
f
is likely because
the predictors did not account for the strength gain
duringconstruction. However, O(N)=0.07, whichis
small consideringthedifferent methodsandassump-
tionsused. Thesenumbersmayserveasameasureof
theuncertainty under themost favorabledesignsce-
nario.Thesepredictionerrorsarenotmuchlargerthan
those for Type C predictions and those estimated in
Table3.
Theaboveresults showthat, for thecasehistories
reviewed, thesimplemethodsusedindesignandreli-
ability analysis are generally satisfactory. However,
it must be emphasized that the cases represent very
small spectrumof thoseingeotechnical practiceand
that results fromsimplecases cannot always begen-
eralized. As an example, if oneextends theproblem
to includelong-termstability, thentheporepressure
mustbeestimated. Forslopesonunsaturatedsoils, the
suctionis strongly dependent onthesoil-water char-
acteristics(SWC). StudiesbyChonget al (2000) and
Zhangetal (2005) showthatuncertaintiesaboutSWC
alonecanresultinaO(F
s
) aslargeas0.2.Thisisabout
equal tothecombineduncertaintygiveninSec. 3.3.
8
5 COMPLEX PROBLEMS
Most design problems are far more complex than
thoseexamined in Sec. 3 and 4. Besides thesubsoil
stratification model, boundary conditions are usu-
ally difficult to determine. For slopes onunsaturated
soils, surface infiltration may control stability and
thechangeinporepressuredepends not only onthe
SWCbutalsoonrainfall characteristics.Anothercom-
plex boundary condition is the displacement of the
support systeminbracedexcavations. FEM analyses
canpredict thepressuredistributionwell under ideal-
izedconditions, but detailsof constructionprocesses,
which are difficult to predict, can have important
effect on displacements and stresses. Other complex
boundary conditions include wave loadings on off-
shorestructuresandgroundmotionfromearthquakes.
Evaluationof errorsinmanyof theinputdataisbeyond
thescopeof soil mechanics.
Wheretheuncertaintypredictedbyreliabilitymeth-
odsmaybetoolargetobeuseful for designdecisions,
theobservational methodprovidesanattractivealter-
native to evaluate the safety. There have been many
examples of the successful use of the observational
methodsincetheChicopeeDam.
6 BAYESIANUPDATING
The important role of the observational method in
solving complex design problems has already been
mentioned. Bayesian updating provides a valuable
model toevaluateobservedperformances. Itcancom-
bineinformationfromdifferentsourcesincludingsite
investigationdata, observations, andanalytical results.
EarlyexamplesincludeTang(1971), andMatsuoand
Asaoka (1978). Eq. (2) can be extended to solve
problems with several performance modes (z) and
properties (x) (Wu et al 2007). Two recent applica-
tionsservetoillustratethepotentialsof thisapproach
to complex design problems. Gilbert et al (1998)
usedupdatingtoevaluatethemobilizedshearstrength
of the clay-geosynthetic interface at the Kettlement
Hills landfill. The probability that the strength is x
i
giventhat theslopehas failedis givenby P[x
i
|z
j
] in
Eq(2), Fig. 2, whereP
[x
i
]=prior probabilitythatthe
strength x
i
, P[z
j
|x
i
]=probability that theslopefails
given the strength is x
i
, P[z
i
]=probability that the
slopefails. Theevaluationaccountsfor variousuncer-
taintiesinlaboratoryteststhatcontributetoP
[x
i
] and
inaccuraciesinthestabilityanalysisthat contributeto
P[z
j
|x
i
].
Regional records of slope performance, when
relatedtosomesignificant parameter, suchasrainfall
intensity, arevaluableasaninitial estimateof failure
probability. Cheung and Tang (2005) collected data
on failure probability of slopes in Hong Kong as a
function of age and rainfall. This was used as prior
probabilityinBayesianupdatingtoestimatethefailure
probability for aspecific slope. Thefailureprobabil-
ity based on age is P[x
i
] where i=0 =stable slope
andi =1=failure, P[z
j
|x
i
]=pdf of asdetermined
frominvestigationsfor failedslopes, and =reliabil-
ityindexfor thespecificslopeunder investigation. To
applyBayesianupdating, Eq. (2) isrewrittenas,
Theresults areused to formulateaslopemainte-
nancestrategyfor HongKong.
7 SUMMARY ANDCONCLUSIONS
The review of reliability methods emphasizes the
comparisonof reliability analysis withobservedper-
formance. Failure in soft to mediumclays is exam-
ined in detailed with well documented case histo-
ries. Thecalculatedfailureprobabilities of slopes on
clay with simplesubsoil profiles arein good agree-
ment with observed performance and provide some
confidenceinthemethods. Itisalsoclearthatformore
complex subsoil stratifications, modeling thefailure
mechanismis critical. Complex boundary conditions
wherestatistical dataareofteninsufficientwill require
moreinputintheformof subjectiveprobabilitybased
onjudgment. Foracomprehensivereviewof thistopic,
seeBaecher andChristian(2003).
For very complex design conditions, theobserva-
tional method is often used to achieve a success-
ful design. Bayesian updating provides an analytical
model fortheobservational method.Tworecentexam-
plesaregiventoindicatethepotential applicationsof
Bayesianupdating.
ACKNOWLEDGEMENTS
Theauthor thanks theOrganizingCommitteefor the
opportunity tomakethispresentation, Prof. Y. Honjo
foradviceandhelpinthepreparationof thispaper, and
Prof. B F. LowandProf. G. A. Fentonfor informative
discussionsabout their FEM analyses.
REFERENCES
Baecher,G.B.,andChristian,J.T.2003.Reliability and Statis-
tics in Geotechnical Engineering. Chichester: J ohnWiley
andSons.
Bergado, D. T., Long, P.V., Lee, C.H., Loke, K.H., and
Werner, G. 1994a. Performance of reinforced embank-
ment on soft Bangkok clay with high-strength geotex-
tile reinforcement. Geotextiles and Geomembranes. 13:
403420.
Bergado, D.T., Patron J r, B.C., Youyongwatana, W., Chai,
J -C., and Yudhbir, 1994b. Reliability-based analysis of
embankment on soft Bangkok clay. Structural Safety
13:247266.
Bishop, A. W., and Bjerrum, L. 1960. The relevance of
the triaxial test to the solution of stability problems. In
Proc. Research Conf. on Shear Strength of Cohesive Soils.
437501, NewYork: ASCE.
9
Casagrande, A. 1965. Roleof thecalculatedrisk inearth-
workandfoundationenmgineering. J. Soil Mechanics and
Foundation Division ASCE, 91:140.
Chai, J., and Bergado, D.T. 1993. Performance of rein-
forced embankment on Muar Clay deposiot. Soils and
Foundations, 33(4):117.
Cheung, R. W. M., andTang, W. H. 2005. Realistic assess-
ment of slope stability for effective landslide hazard
management. Geotechnique 55:8594.
Chong, P. C., Phoon, K. K., and Tan, T. S. 2000 Proba-
bilistic analysis of unsaturated residual soil slopes. In
R. E. Melchers andM. G. Stewart (eds.) Applications of
Statistics and Probability, 375382. Rotterdam: Balkema.
Christian, J. T., Ladd, C. C., andBaecher G. B. 1994. Relia-
bilityappliedtoslopestabilityanalysis. J. of Geotechnical
Engineering, 120:21802207.
DeGroot, D. J., and Baecher, G. B. 1993. Estimating auto-
covariance of in-situ soil properties. J. Geotechnical
Engineering, 119:147167.
El-Ramly, H., Morgenstern, N.R., and Cruden, D.M.
2002. Probabilisticslopestabilityfor practice. Canadian
Geotechnical J., 39:665683.
Freudenthal, A.M., 1947. The safety of structures. Trans.
ASCE, 112:125180.
Gilbert, R. B.,Wright, S. G., andLiedke, E. 1998Uncertainty
in back analysis of slopes: Kettleman Hills case his-
tory. J. Geotechnical and Geoenvironmental Engineering,
124:11671176.
Griffith, V. D., and Fenton, G. A. 2004. Probabilistic slope
stability analysisby finiteelements. J. Geotechnical and
Geoenvironmental Engineering, 130:507518.
Lafleur, J., Silvestri, V., Asselin, R., and Soulie, M. 1988.
Behaviourof atestexcavationinsoftChamplainSeaclay.
Canadian Geotechnical J. 25:705715.
Lefebvre, G, Ladd, C.C., and Pare, J -J. 1988. Comparison
of fieldvaneandlaboratoryundrainedshear strengthsin
soft sensitiveclays. InVane Shear Testing in Soils: field
and laboratory studies. STP1014, 233246. Philadelphia:
ASTM.
Lambe,T. W. 1973. Predictionsinsoil engineering. Geotech-
nique, 23:149202.
Leonards, G. A. 1982. Investigationof failures. J. Geotech-
nical Engineering, 108:222283.
Low, B. F. 2008. Practical reliabilityapproachusingspread-
sheet. In K-K. Phoon (ed.) Reliability-Based Design in
Geotechnical Engineering. London: Taylor andFrancis.
Low, B.F., andTang,W.H. 1997. Reliability analysisof rein-
forcedembankment onsoft ground. Canadian Geotech-
nical J. 34:672685.
Lumb, P. 1970. Safetyfactorsandtheprobabilitydistribution
of strength. Canadian Geotechnical J., 7:225242.
Malayasian4HighwayAuthority1989.In(eds.)R.R.Hudson,
C. T. Toh, and S. F. Chan. Proc. Intern. Symp. on
Trial Embankments on Malayasian Marine Clays. Kuala
Lumpur, Malayasia.
Massachusetts Institute of Technology 1975. Proc. Foun-
datiions Deformation Prediction Symp. Washington, DC:
Federal HighwayAdministration.
Matsuo,M.andAsaoka,A.1978.Dynamicdesignphilosophy
of soilsbasedontheBayesianreliabilityprediction. Soils
and Foundations 18(4):117
Meyerhof, G. G. 1970. Safety factors in soil mechanics.
Peck, R. B. 1969.Advantagesandlimitationsof theobserva-
tional methodinappliedsoil mechanics, NinthRankine
Lecture. Geotechnique, 19:171187.
Phoon, K. K., and Kulhawy, F. H. 1999a. Characterization
of geotechnical variability. Canadian Geotechnical J. 36:
612624.
Phoon, K. K., and Kulhawy, F. H. 1999b. Evaluation of
geotechnical propertyvariability. Canadian Geotechnical
J. 36:625639.
Rojiani, K. B., Ooi, P.S.K., andTan, C.K. (1991) Calibration
of load factor design code for highway bridge founda-
tions, Geotechnical Engineering Congress, Geotechnical
Special Publication No. 217, 2:13531364. NewYork:
ASCE.
Silva, F. Lambe, T.W., andMarr, W.A. 2008. Probabilityand
riskof slopefailure. J. Geotechnical and Geoenvironmen-
tal Engineering, 134:16911699.
Tang, W. H. 1971. A Bayesianevaluationof informationfor
foundation engineering design. In P, Lumb (ed.) Statis-
tics and Probability in Civil Engineering, 173185, Hong
Kong: HongKongUniv. Press.
Tang, W. H. 1971. A Bayesianevaluationof informationfor
foundation engineering design. In P, Lumb (ed.) Statis-
tics and Probability in Civil Engineering, 173185, Hong
Kong: HongKongUniv. Press.
Tang, W. H., Yucemen, M.S., and Ang, A. H-S. 1976.
Probability-basedshort termdesignof soil slopes. Cana-
dian Geotechnical J. 13:201215.
Tavanas, F., andLeroeil, S. 1980. Thebehavior of embank-
ments on clay foundations. Canadian Geotechnical J.
17:236260.
Taylor, D. W. 1948. Fundamentals of Soil Mechanics, New
York: J ohnWileyandSons.
Terzaghi, K. 1929 Effect of minor geologic details on the
safety of dams. Technical Pub. 215, 3144, American
Instituteof MiningandMetallurgical Engineering, New
York.
U.S. Army, Corps of Engineers 1995. Introductionto prob-
ability and reliability methods for use in geotechnical
engineering. Technical letter 1110-2-547, Washington,
DC
Vanmarcke, E. H. 1977. Probabilisic modeling of soil
profiles. J. Geotechnical Engineering Division, ASCE,
103:P1227-1246
Wang,Y-J., andChiasson, P. 2006. Stochasticstabilityanaly-
sisof atestexcavationinvolvingspatiallyvariablesubsoil.
Canadian Getechnical J. 43:10741087.
Wu, T.H. 2003. Variation in clay deposits of Chicago. In
E. VanmarckeandG. A. Fenton, (eds.) Probabilistic Site
Characterization at the National Geotechnical Test Sites.
Geotechnical Special Pub. 121. Reston: ASCE.
Wu, T.H., Kraft, L.M. 1970Safetyanalysisof slopes. J. Soil
Mechanics and Foundations Division,ASCE, 96:609630
Xu, B., andLow, B. F. 2006. andZou, J -Z., Williams, D. J.,
and Xiong, W-L. 1995. Search for critical slip surfaces
basedonfiniteelementmethod. Canadian Getechnical J.
32:233
Yu, H.S., Salgado, R., Sloan, S.W., andKim, J.M. 1998. Limit
Analysis versus limit equilibriumfor slope stability. J.
Geotechnical and Geoenvironmental Eng. 124:111.
Zhang, L. L., Zhang, L. M., and Tang, W. H. (2005)
Rainfall-induced slopefailureconsidering variability of
soil properties, Geotechnique, 55:183188.
Zou, J -Z., Williams, D.J. and Xiong, W-L. 1995. Search
for critical slipsurfacesbasedonfiniteelement method.
10
Keynote lectures
Riskassessment andmanagement for geohazards
F. Nadim
International Centre for Geohazards (ICG) / Norwegian Geotechnical Institute (NGI), Oslo, Norway
ABSTRACT: Eachyear, natural disasterscausecountlessdeathsandformidabledamagetoinfrastructureand
theenvironment. In20045, morethan300,000peoplelosttheir livesinnatural disasters. Material damagewas
estimatedat USD 300billion. Many lives couldhavebeensavedif morehadbeenknownabout therisks and
possiblerisk mitigationmeasures. Thepaper summarizes thestate-of-the-art intheassessment of hazardand
risk associatedwithlandslides, earthquakesandtsunamis. Theroleof suchassessmentsinarisk management
context is discussed and general recommendations for identification and implementation of appropriaterisk
mitigationstrategiesareprovided.
1 INTRODUCTION
Geohazards, i.e. natural hazards that aredrivenby
geological featuresandprocesses, poseseverethreats
tohumans, propertyandthenatural andbuiltenviron-
ment. During 2005, geohazards accounted for about
100,000 deaths worldwide, of which 84% were due
toOctobersPakistanearthquake. Inthat year, natural
disastersaffected161millionpeopleandcost around
US$160billionoverdoublethedecadesannual aver-
age. HurricaneKatrinaaccountedforthreequartersof
thiscost. Duringtheperiod1996to2005, natural dis-
asters causednearly onemillionlives lost, or double
thefigurefor theprevious decade, affecting 2.5 bil-
lionpeopleacross theglobe(WorldDisaster Report,
2006). Whenthetrendof fatalitiesduetonatural haz-
ardsisstudiedoverthelast100years,itappearsthatthe
increaseintheknownnumber of deathsisduetothe
increaseintheexposedpopulationinthis timescale
and the increased dissemination of the information,
andnottoanincreaseinthefrequencyand/or severity
of natural hazards.
The economic consequences of geohazards show
anevenmoredramatic increasingtrend(MunichRe,
2007). Someof thereasonsfor thisincreaseareobvi-
ous, others less so. The post-disaster effects can be
especially severe in a vast, densely-populated area
wheresewers fail and diseasespreads. Slums spring
upindisaster-proneareassuchassteepslopes, which
arepronetolandslidesor particularly severedamage
in an earthquake. Many of theworlds fastest grow-
ingcities arelocatedoncoastal landor rivers where
climatevariability andextremeweather events, from
cyclones to heat waves to droughts, pose increasing
risksof disaster.
Several well-documented studies have shown
clearly that developing countries are more severely
affectedbynatural disastersthandevelopedcountries,
especially in terms of lives lost (UNDP 2004, ISDR
Table1. Natural disastersintheperiod19912000(Source:
IFRC 2001).
Countryclassification No. of disasters No. of liveslost
Lowandmedium 1838 649,400
developedcountries
Highlydeveloped 719 16,200
countries
2004 and International Federation of Red Cross and
RedCrescent 2004). Table1showsthedatacompiled
by IFRC (2001) for the decade 19912000. Of the
total number of persons killedby natural disasters in
thisperiod, thehighlydevelopedcountriesaccounted
for only 5% of the casualties. In absolute numbers,
thematerial damageandeconomiclossduetonatural
hazards in highly developed countries by far exceed
thoseindevelopingnations. However, thisreflectsthe
grosslydisproportionatevaluesof fixedassets, rather
thanreal economicvulnerability.
Mitigationandpreventionof theriskposedbynatu-
ral hazardshavenotattractedwidespreadandeffective
public support inthepast. However, thesituationhas
changed dramatically during the past decade, and it
is now generally accepted that a proactive approach
toriskmanagementisrequiredtoreducesignificantly
lossof livesandmaterial damageassociatedwithnatu-
ral hazards.Thewidemediaattentiononmajornatural
disasters during the last decade has clearly changed
peoplesmindintermsof acknowledgingriskmanage-
ment as analternativeto emergency management. A
milestoneinrecognitionof theneedfornatural disaster
riskreductionwastheapproval of theHyogoFrame-
work for Action20052015: BuildingtheResilience
of Nations and Communities to Disasters (ISDR
2005). This document, which was approved by 164
UNcountriesduringtheWorldConferenceonDisaster
13
Reduction in Kobe, J anuary 2005, clarifies interna-
tional working modes, responsibilities and priority
actionsfor thecoming10years.
Thefirst step in any decision-making process for
disaster risk reduction is thequantitativeassessment
of the risk. This paper provides an overview of the
state-of-the-art for hazard and risk assessment for
landslides, earthquakes and tsunamis, and discusses
possible risk mitigation strategies for these geohaz-
ards.
2 RISK ASSESSMENT FRAMEWORK
Theterminology usedinthis paper is generally con-
sistent with therecommendations of ISSMGE Glos-
sary of Risk Assessment Terms (listedonTC32web
page: http://www.engmath.dal.ca/tc32/). The impor-
tant termsusedinthecontext of thispaper are:
Danger (Threat): Natural phenomenon that could
lead to damage, described by geometry, mechani-
cal and other characteristics. Description of a threat
involvesnoforecasting.
Hazard: Probabilitythataparticulardanger(threat)
occurswithinagivenperiodof time.
Risk: Measure of the probability and severity
of an adverse effect to life, health, property, or
the environment. Mathematically, risk is defined as
Risk=HazardPotential worthof loss.
Vulnerability: Thedegreeof losstoagivenelement
or setof elementswithintheareaaffectedbyahazard.
Itisexpressedonascaleof 0(noloss) to1(total loss).
InUNISDR terminology onDisaster Risk Reduc-
tion (2009), disaster is defined as a serious dis-
ruptionof thefunctioningof acommunity or asoci-
ety causing widespread human, material, economic
or environmental losses which exceed the ability of
the affected community or society to cope using its
own resources. Thetermnatural disaster is slowly
disappearing fromthedisaster risk management ter-
minology because without the presence of humans,
oneisonlydealingwithnatural processes. Theseonly
becomedisasterswhentheyimpact acommunityor a
society.
Quantitatively risk can beevaluated fromthefol-
lowingexpression:
whereR=riskassociatedwithaparticular danger
H=hazard
V=vulnerabilityof elementsat risk
E=expected cost of total loss of elements at
risk
Several riskassessmentframeworkshavebeenpro-
posed, and Dzgn and Lacasse (2005) list a large
number of these. The frameworks have the com-
mon objective of answering the following questions
(modifiedfromLee& J ones, 2004):
Whataretheprobabledangersandtheirmagnitude?
[Danger Identification]
How often do the dangers of a given magnitude
occur? [HazardAssessment]
What are the elements at risk? [Elements at Risk
Identification]
Whatisthepossibledamagetotheelementsatrisk?
[VulnerabilityAssessment]
What is theprobability of damage? [Risk Estima-
tion]
Whatisthesignificanceof theestimatedrisk?[Risk
Evaluation]
What shouldbedone? [RiskManagement]
In the following sections, the methodologies for
answering one or more of these questions for land-
slides, earthquakes and tsunamis will be discussed.
3 LANDSLIDES
3.1 Landslide threat
Landslides represent a major threat to human life,
propertyandconstructedfacilities, infrastructureand
natural environment in most mountainous and hilly
regions of the world. Statistics fromThe Centre for
Research on theEpidemiology of Disasters (CRED)
showthat landslides areresponsiblefor at least 17%
of all fatalities fromnatural hazards worldwide. The
socio-economic impact of landslides is underesti-
mated because landslides are usually not separated
fromother natural hazard triggers, such as extreme
precipitation,earthquakesorfloods.Thisunderestima-
tioncontributestoreducingtheawarenessandconcern
of bothauthoritiesandgeneral publicabout landslide
risk.
Asaconsequenceof climatechangeandincreasein
exposureinmanypartsof theworld,theriskassociated
withlandslides is growing. Inareas withhighdemo-
graphicdensity, protectionworksoftencannotbebuilt
becauseof economicorenvironmental constraints,and
is it not always possibleto evacuatepeoplebecause
of societal reasons. Oneneeds to forecast theoccur-
renceof landslideandthehazardandrisk associated
with them. Climate change, increased susceptibility
of surface soil to instability, anthropogenic activi-
ties, growingurbanization, uncontrolledland-useand
increased vulnerability of population and infrastruc-
ture as a result, contribute to the growing landslide
risk. According to the European Union Strategy for
Soil Protection (COM232/2006), landslides are one
of themaineight threatstoEuropeansoils.
Water playsamajor roleintriggeringof landslides.
Figure 1 shows the relative contribution of various
landslidetriggeringeventsfactor inItaly. Heavyrain-
fall isthemaintrigger for mudflows, thedeadliestand
most destructiveof all landslides.
Manycoastal regionshavecliffsthataresusceptible
tofailurefromseaerosion(byundercuttingatthetoe)
andtheir geometry (slopeangle), resultinginloss of
agricultural landandproperty. Thiscanhaveadevas-
tatingeffectonsmall communities. Forinstance, parts
of thenorth-eastcoastcliffsof Englandareerodingat
ratesof 1m/yr.
14
Figure1. LandslidetriggersinItaly. Source: CNR-GNDCI
AVI Databaseof areas affected by landslides and floods in
Italy.
As aconsequenceof climatic changes and poten-
tial global warming, anincreaseof landslideactivity
is expected in the future, due to increased rainfalls,
changesof hydrological cycles,moreextremeweather,
concentratedrainwithinshorterperiodsof time, mete-
orological events followed by sea storms causing
coastal erosion and melting of snow and of frozen
soils in high mountain regions liketheAlps and the
Himalayas.Thegrowinglandslidehazardandrisk, the
needtoprotect peopleandproperty, theexpectedcli-
mate change and the need to manage the risk have
contributed to set the agenda for the profession to
assessandmitigatethelandsliderisk.
3.2 Landslide hazard assessment for specific
slopes
Hazard assessment for a specific slope usually
involves a probabilistic analysis of the slope, while
hazardassessment for aregiongenerally requiresthe
computation of frequency of the landslides in the
region. For regional analyses, datatobecollectedare
intheformof maps relatedto geomorphology, geol-
ogy, land-use/cover andtriggers. For specific slopes,
the required data for hazard analysis include slope
geometry such as height, width, inclination of slope
and potential failureplane, shapeand length of fail-
ure plane etc., strength parameters data for possible
trigger suchas rainfall intensity, water level, severity
of dynamic loads e.g. earthquake magnitude, accel-
erationand/or other characteristics. Theprobabilistic
models used for aspecific slopevary depending on
thefailuremechanism(e.g. flows, fallsor slides) and
theslope-formingmaterial (e.g. soil or rock).
Analysesof specific slopesusedeterministic (fac-
tor of safety, numerical analyses) and/or probabilistic
methods, e.g. first order, second-moment (FOSM),
first order reliability method (FORM), point esti-
mate methods, and Monte Carlo Simulation (MCS)
(Ang&Tang1984). Recent trendscombinedifferent
approachesforanimprovedmodel of thehazard(s).An
uncertaintyanalysisisessential priortothecalculation
of slopefailureprobability asit allowsarational cal-
culationof total uncertaintiesassociatedwithdifferent
sourcesof uncertainty(e.g.inparametersandmodels).
Thequantificationandanalysisof uncertaintiesplaya
critical roleintheriskassessment.
The stability situation for natural and man-made
slopes is often expressed by a factor of safety. The
factor of safety is definedas theratio of thecharac-
teristicresistance(resistingforce) tothecharacteristic
load (driving force). The approach does not address
the uncertainty in load and resistance in a consis-
tent manner. The choice of characteristic values
allowstheengineertoimplicitlyaccountforuncertain-
tiesbyusingconservativevaluesof load(highvalue)
andresistanceparameters (lowvalue). Thechoiceis
somewhatarbitrary.Duncan(1992and1996)provided
an overview of deterministic slope stability analysis
method. The overview included the factor of safety
approach, equilibriummethodsof slopestabilityanal-
ysis (J anbus generalized method of slices, Bishops
method, Spencers method, Morgenstern and Prices
method among others), techniques for searching for
thecritical slipsurface, bothcircularandnon-circular,
three-dimensional analyses of slope stability, analy-
ses of thestability of reinforced slopes, drained and
undrained conditions, and total stress and effective
stress analyses. Slopes withnominally thesamefac-
tor of safety couldhavesignificantly different safety
marginsbecauseof theuncertaintiesinvolved. Duncan
(2000) pointedout that Throughregulationor tradi-
tion, thesamevalueof safety factor is oftenapplied
to conditions that involvewidely varying degrees of
uncertainty. Thisisnot logical.
To evaluatethehazard associated with thefailure
of aspecific slope, thestability assessment must be
put into aprobabilistic format usingoneof thetech-
niquesmentionedearlier(FOSM, FORM, MCS, etc.).
Anoverviewof theavailablemethodsfordoingproba-
bilisticslopestabilityassessmentforindividual slopes
isprovidedinNadimet al. (2005).
3.3 Regional landslide hazard assessment
Landslidehazardandriskassessmentisoftenrequired
on a regional or national scale and it would not be
feasible to do a stability assessment for all poten-
tiallyunstableslopesinthestudyarea.Thereforeother
techniquesbasedonGeographical InformationTech-
nology (GIT) are employed in these situations. An
exampleof thistypeof hazardassessmentisthestudy
doneby Nadimet al. (2006) in theGlobal Hotspots
study for the ProVention Consortium. That model,
whichiscurrently beingupdatedfor theGlobal Risk
Updateprojectof ISDR, assessesthelandslidehazard
by considering a combination of the triggering fac-
torsandsusceptibilityindicators.Theprinciplesof the
model aredemonstratedinFigure2.
Inthelatestversionof themodel, alandslidehazard
index was defined using six parameters: slope fac-
tor withinaselectedgridcell, lithology(or geological
conditions), soil moisturecondition, vegetationcover
index, precipitationfactor, andseismicconditions. For
each factor, an index of influence was determined
and the relative landslide hazard level H
landslide
was
obtainedbymultiplyingandsummingtheindices.The
landslidehazardindices werethencalibratedagainst
thedatabases of landslideevents in selected (mostly
European) countries to obtain the frequency of the
15
Figure2. Schematicapproachfor landslidehazardandrisk
evaluation(Nadimet al., 2006).
landslide events, i.e. the landslide hazard. Figure 3
shows the landslide hazard map for parts of Latin
AmericaobtainedbyNadimet al. (2006).
3.4 Landslide risk assessment
Themost completedescriptionof thepossiblelosses
(or risk) is quantitatively in terms of a probability
distribution, whichpresentstherelativelikelihoodof
any particular loss valueor theprobability of losses
being less than any particular value. Alternatively,
the expected value (i.e., the probability weighted
average value) of loss can be determined as a sin-
gle measure of risk. A general scenario-based risk
formulationisgivenbyNadim& Glade(2006):
where C is particular set of losses (of collectively
exhaustive and mutually exclusive set of possible
losses), Sisparticularscenario(of comprehensiveand
mutuallyexclusivediscreteset of possiblescenarios),
P[S]isprobabilityof occurrenceof scenarioS,P[C| S]
is theconditional probability of loss set C giventhat
scenarioShasoccurred, andE[Loss] istheexpected
value of loss. Loss may refer to any undesirable
consequence, such as loss of human life, economic
loss, lossof reputation, etc., intermsof itsdirect and
indirect effects (e.g. local damage of railway tracks
andrelatedinterruptionof industrial traffic), itseffects
on different social groups (e.g. individuals, commu-
nity, insurance, government) as well as its short- and
long-terminfluencesonasociety(e.g. fatalitiescould
includeall childrenof acommunity, thetouristindus-
trymightcollapse). Mostoftenthefocusisontheloss
of humanlife.
Calculationof thetermsintheaboveequationisnot
trivial.Thehazardtermintheaboveequation(i.e.P[S])
isnotconstantwithtime.Moreover,theexpectednum-
ber of fatalitiesdependsonmanyfactors, for example
onwhichweek-dayandwhattimeof thedaytheland-
slideoccurs, whether awarningsystemisinplaceand
working,etc.Thepotentiallyaffectedpopulationcould
bedividedintogroupsbasedonfor examplethetem-
poral exposuretothelandslide: peoplelivinginhouses
Figure3. Landslidehazardmapfor partsof LatinAmerica
(Nadimet al., 2006).
Figure4. Procedurefor riskassessment of slopes.
that arein thepath of thepotential landslide, locals
intheareawho happento bepasser-bys andtourists
and/or workerswhoarecoincidentallyat thelocation
duringcertainperiodsof thedayof theyear.
Figure4 summarizes ageneral procedurefor risk
assessment for slides. The key issue is the identifi-
cation of potential triggers and their probability of
occurrence, the associated failure modes and their
consequences. The triggering mechanisms could be
natural, such as earthquake, tectonic faulting, rain-
fall, temperatureincreasecaused by climatechange,
excess pore pressures or man-made. Generally, one
shouldconsiderseveral scenariosof plausibletriggers,
estimatetherun-out distanceandextent triggeredby
theseevents, andestimatetheupper andlower bounds
on the annual probability of occurrence of the sce-
narios(Roberds, 2005).Thisscenario-basedapproach
involvesthefollowingsteps:
Definescenariosfor landslidetriggering
Computetherun-off distance, volumeandextentof
landslidefor eachscenario
Estimatethelossfor thedifferent landslidescenar-
ios
Estimatetherisk andcompareit withtolerableor
acceptablerisklevels
16
4 EARTHQUAKES
4.1 Earthquake threat
Earthquakescanbeespecially devastatingwhenthey
occur inareas withhighpopulationdensity. Therisk
posedbyearthquakestolargecitiesandother densely
populated areas is by far greater than all other geo-
hazardscombined(Nadim& Lacasse, 2008). Similar
to other natural hazards, the global fatality count
fromearthquakescontinuestorise. Thishasoccurred
despitetheadoptionof earthquake-resistant building
codesinmost countries.
Inthepastfivecenturies, theglobal deathtoll from
earthquakeshasaveraged100,000per year, aratethat
is dominatedby largeinfrequent disasters, mostly in
thedevelopingnations(Bilham, 2004). J ustinthepast
few years, two of themost catastrophic earthquakes
inhistoryhaveoccurredinAsia(PakistaninOctober
2005andSichuan, ChinainMay2008). Theincrease
in earthquake-induced fatalities is mainly dueto the
steadygrowthinglobal population. At thesametime,
there is a decline in the fatality rate expressed as a
percentage of instantaneous population. It is tempt-
ing to attribute this observation to the application
of earthquake-resistant constructioncodeinnewcity
developments.A morepessimistic, andmorerealistic,
interpretationis, however, that theapparent declinein
risk isastatistical anomalyandfutureextremeearth-
quakedisastersinsomeof theworldsmegacitiesmay
arrest, or reverse, thecurrent trend.
4.2 Earthquake hazard assessment
Seismic hazard analysis methods havea widerange
of applications, frombroadly basedzonations aimed
essentiallyonlyatdescribinganddelineatingtheseis-
micity, to site-specific analyses aimed for design of
specific structures. Within both fields the analyses
rangefromrelatively cursory tohighly detailed, with
thelevel of detail fordesignpurposesbeingdependent
onthesensitivityof thestructure.
Prior to 1970, the assessment of seismic hazard
was based on a deterministic approach that consid-
ered the most likely scenarios for earthquakes that
couldaffect aparticular location. Theseminal paper
of Cornell (1968) introducedthemethodologybehind
a Probabilistic Seismic HazardAssessment (PSHA)
changedtheway most engineeringseismologists did
their hazard analyses. Traditionally, thepeak ground
acceleration (PGA) has been used to quantify the
groundmotioninPSHA. Today thepreferredparam-
eter is Response Spectral Acceleration (SA), which
gives the maximum acceleration experienced by a
damped, single-degree-of-freedomoscillator (acrude
representation of building response). The oscillator
periodischoseninaccordancewiththenatural period
of thestructureanddampingvalues aretypically set
at5%of critical. ThePSHA methodologyfor estimat-
ingtheannual probability of occurrenceof aground
motioncharacteristicisthesameforboththePGA and
theSA. Inbothsituations, PSHA involvesthreesteps:
1)specificationof theseismic-hazardsourcemodel;2)
specificationof thegroundmotionmodel (attenuation
relationship); and3) theprobabilisticcalculation.
Theseismic-hazard sourcemodel is adescription
of the magnitude, location, and timing of all earth-
quakes(usuallylimitedtothosethatposeasignificant
threat). For example, asourcemodel might becom-
posed of N total earthquake scenarios, where each
has its own magnitude, location, and annual rate of
occurrence. Thegroundmotionmodel usedinPSHA
consists of thesourcemodel andanattenuationrela-
tionship. Thelatter describeshowrapidlyaparticular
ground motion parameter decays with distancefrom
thesource. Giventhetypicallylargenumber of earth-
quakesandsitesconsideredinananalysis, attenuation
relationships must besimpleand easy to apply. The
most basic attenuation relationships givetheground
motionlevel asafunctionof magnitudeanddistance,
butmanyhaveother parameterstoallowfor afewdif-
ferentsitetypes(e.g.,rockvs.soil)orstylesof faulting.
Different relationships havealso been developed for
differenttectonicregimes.All aredevelopedbyfitting
an analytical expression to observations (or to syn-
thetic datawhereobservations arelacking). Withthe
seismic-hazardsourcemodel andattenuationrelation-
ship(s) defined, the probabilistic-hazard calculation
is conceptually simple. In practice, however, things
canget messy. Besides thenon-triviality of defining
thespatial distributionof small earthquakes onlarge
faults, thereisalsotheproblemthat different attenua-
tionrelationshipsusedifferent definitionsof distance
tothefault plane.
Thelogictreeapproachisafundamental andwell-
established tool in PSHA aimed at capturing the
epistemic uncertainties (uncertainties related to our
lackof knowledge), primarilyassociatedwithseismic
sourcesandwithgroundmotionmodelling(Kulkarni
etal., 1984; CoppersmithandYoungs, 1986).Thelogic
treeapproach,whichhasbeenstate-of-the-artinPSHA
for many years, can also be described as a means
by which one can include subjective information in
an objectiveway. Theuseof experts is afundamen-
tal component in the judgments that are needed in
order to model epistemic uncertainties. To this end
thestate-of-the-art methodology isthat developedby
SSHAC (1997), which has been summarized in a
moreeasilyavailablewayinareviewbyNRC(1997).
McGuire(2004) explainedhowseismology, geology,
strong-motion geophysics, and earthquakeengineer-
ing contribute to the evaluation of seismic risk. He
provideddetaileddescriptionof themethodsusedfor
development of consensus probabilistic seismic haz-
ardmaps, animportantprerequisitefortheassessment
of earthquakerisk.
4.3 Earthquake risk assessment
Themostcomprehensiveworktowardsearthquakerisk
calculationtodateiscondensedinHAZUS, asoftware
systempreparedforuseintheUnitedStatesbytheFed-
eral EmergencyManagementAgency(FEMA, 2003).
17
Figure5. Flowchartof earthquakelossestimationmethod-
ology(HAZUSTechnical Manual, Federal EmergencyMan-
agementAgency, 2003).
Figure6. Estimationof earthquake-induceddamageusing
thecapacity-spectrummethod.
ThemethodologyusedinHAZUSfor earthquakeloss
estimationisoutlinedinFigure5.
The HAZUS approach is based on the so-called
thecapacity-spectrummethod(seeFigures 6and7).
It combines the ground motion input in terms of a
responsespectrum(spectral accelerationversusspec-
tral displacement)withthebuildingsspecificcapacity
curve.Thephilosophybehindthisapproachisthatany
building is structurally damaged by the earthquake-
induced permanent horizontal displacement, and not
bytheaccelerationperse. Foreachbuildingandbuild-
ingtypetheinterstoreydriftisafunctionof theapplied
lateral forcethat can beanalytically determined and
transformed into building capacity curves (capacity
to withstand accelerations without permanent dis-
placements). Buildingcapacity curves naturally vary
Figure7. Probability of astructurewithspectral displace-
ment d
p
onFigure6beingindifferent statesof damageafter
theearthquake.
betweendifferentbuildingtypes,andalsobetweendif-
ferentregions, reflectingonbuildingcoderegulations
andlocal constructionpractice.
TheHAZUSmethodologyincludesstandardmeth-
odsfor:
1. Inventory data collection based on census tract
areas
2. Usingdatabasemapsof soil type, groundmotion,
groundfailure, etc.
3. Classifyingoccupancyof buildingsandfacilities
4. Classifyingbuildingstructuretype
5. Describingdamagestates
6. Developingbuildingdamagefunctions
7. Grouping, rankingandanalyzinglifelines
8. Usingtechnical terminology
9. Providingoutput.
The HAZUS approach is attractive froma scien-
tific/technical perspective. However, thefact that it is
tailoredsointimatelytotheUSsituationsandspecific
GISsoftwaremakesit difficult toapplyinother envi-
ronmentsandgeographical regions.Awareof theneed
for amoreinternationally accessibletool for seismic
risk estimation, theInternational Centrefor Geohaz-
ards (ICG), throughNORSAR andtheUniversity of
Alicante, hasdevelopedaMatlabbasedtool inorder
to computetheseismic risk inurbanareas usingthe
capacity spectrummethod, SELENA (SEimic Loss
EstimatioN using a logic tree Approach, see web
page http://www.norsar.no/pc-35-68-SELENA.aspx).
SELENA can computetheprobability of damagein
eachoneof thefour damagestates(slight, moderate,
extensive and complete) for defined building types.
SELENAisastand-alonesoftwarepackagethatcanbe
appliedanywhereintheworld. Itincludesalogictree-
based weighting of input parameters that allows for
thecomputationof confidenceintervals.Thelossesti-
mationalgorithminSELENA isbasedontheHAZUS
methodology, and144predefinedvulnerabilitycurves
detailedintheHAZUSmanual (seee.g. Figure8) can
beappliedinSELENA.
18
Figure8. Fragility curves for small refineries with unan-
chored components (HAZUS Technical Manual, FEMA
2003).
As input to SELENA, the user must supply built
area or number of buildings in the different model
building types, earthquakesources, attenuation rela-
tionships, soil mapsandcorrespondinggroundmotion
amplification factors, capacity curves and fragility
curves corresponding to each of themodel building
types and finally cost models for repair or replace-
ment. This probability is subsequently usedwith the
builtareaornumberof buildingstoexpresstheresults
interms of damagedarea(squaremeters) or number
of damagedbuildings. Simplemodels for computing
economicdamagesandcasualtiesarealsoincluded.
5 TSUNAMIS
5.1 Tsunami threat
Tsunamisaregravitywavessetinmotionbylargesud-
den changes of the sea water, having characteristics
intermediatebetweentidesandswell waves.Although
they are infrequent (ca. 510 events reported glob-
ally pr. year), tsunamis represent aserious hazard to
thecoastal populationinmanyareas, asdemonstrated
by thedevastating effects of the2004 Indian Ocean
tsunami.
Earthquakes are the most important mechanism
of tsunami generation, causing more than 75% of
all tsunamis globally. The generation mechanismis
typically dominated by the co-seismic dip-slip fault
movement, as strike-slip fault movements aregener-
ally less important for wave generation. Submarine
landslidesarealsobecomingincreasinglyrecognized
as an important trigger as well. Other sources of
tsunamis include rock slides into bodies of water,
collapsing/explodingvolcanoes, andasteroidimpacts.
Tsunamis generated by large earthquakes in sub-
duction zones along the major plate boundaries (so
called convergent plate boundaries) contribute most
totheglobal tsunami hazard. Suchimportant areasof
generationincludestheringof firealongthePacific
Rim, the Sunda Arc including Indonesia and the
Philippines, Makransouthof Pakistan, theCaribbean
Sea, the Mediterranean Sea, and the fault zones off
thePortuguesecoastline. Figure9showsthehistorical
tsunamisrecordedworldwidesince1628.
Figure 9. Historic tsunamis in the world from17th cen-
tury until today (from Tsunami Laboratory Novosibirsk,
http://tsun.sscc.ru/tgi_1.htm). Different circlesizesandcol-
orsindicatedifferent tsunami intensities(proportional tothe
averagetsunami run-up).
Tsunamisconstituteaseriousnatural hazardforthe
environmentandpopulationsinexposedareas. Future
catastrophescanbemitigatedor preventedbytsunami
hazardevaluationfromstatisticsandgeological anal-
ysis, by risk analysesfromstudiesof slidedynamics,
tsunami propagationandcoastal impact; andby risk
mitigation measures such as tsunami warning sys-
tems, seawalls anddykes, areaplanning, evacuation
routestosafeelevatedareas, andeducation, prepared-
ness and awareness campaigns. Moreover, tsunami
predictions are fundamental in engineering design
andlocationof coastal installations, dams, submerged
bridges, offshoreconstructions, aquaculture, etc.
5.2 Tsunami hazard assessment
Following the catastrophic Indian Ocean tsunami in
December 2004, several researchgroupshavestarted
work on the development of a theoretical frame-
work for Probabilistic Tsunami Hazard Assessment
(PTHA).
The PTHA methodologies are closely related to
well-establishedProbabilisticSeismicHazardAssess-
ment(PSHA) andwedefinePTHA, consistentlywith
thedefinitionof PSHA, astheprobability of exceed-
ing agiven tsunami size(either in terms of tsunami
heightorinundation) atagivenlocationandinagiven
timeinterval. Inthisrespect, thetsunami problemcan
(again in analogy with PSHA) bedivided into three
parts: thesource(for exampletheearthquakegenerat-
ingatsunami), thepath(propagationfromthesource
tosomeshortdistancefromthecoastline) andthesite
effects (inundation distanceand height based on the
local bathymetryandtopography).
In traditional PSHA, the sources are described
through azonation, which is characterized by activ-
ityratesintermsof aGutenberg-Richter relationship.
Thisisalsothemost commonapproachtofollowfor
PTHA, with the difference that also distant tsunami
sourcesmust beaccountedfor inthePTHA. Thepath
effectsintraditional PSHA aredescribedthroughsim-
ple attenuation relations giving the ground shaking
level as afunctionof earthquakemagnitudeanddis-
tancefromtherupturingfault. This approachcannot
19
beapplied for PTHA dueto thestrong influenceof
bathymetryonthetsunami propagation. Itistherefore
necessary to performfull wave propagation model-
ing to includethepath effects in PTHA. This is the
largest differencebetweenPSHA andPTHA, though
it shouldbenotedthat PSHA methodologiesarecur-
rently beingdeveloped, basedonfull groundshaking
scenariosrather thansimpleattenuationrelations.
However, because the computation time required
for the solution of the path problem may limit its
practical applicability, amoreefficient and practical
PTHA (but less accurate) approach would beto use
approximateamplificationfunctionsforthetsunami
maximuminundationor run-upheights(analogousto
attenuationfunctions for peak groundacceleration
inPSHA), whichdependontheprofileof theseafloor
fromacertainwater depthanduptothesite.
Thio et al. (2007) presented a method for Proba-
bilistic Tsunami Hazard Analysis (PTHA) based on
thetraditional Probabilistic Seismic HazardAnalysis
(PSHA). Inlieuof attenuationrelations, their method
usesthesummationof finite-differenceGreensfunc-
tions that have been pre-computed for individual
sub-faults. This enables them to rapidly construct
scenariotsunami waveformsfromanaggregateof sub-
faults that comprise a single large event. For every
fault system, it is thenpossibleto integrateover sets
of thousands of events within a certain magnitude
rangethatrepresentsafullyprobabilisticdistribution.
Becauseof theenclosednatureof ports andharbors,
effects of resonance need to be addressed as well.
Their methodthereforefocuses not only ontheanal-
ysis of exceedancelevels of maximumwaveheight,
but also of spectral amplitudes. As in PSHA, these
spectral amplitudescanbematchedwiththespectral
responseof harbors, andthusallowacomprehensive
probabilistic analysis of tsunami hazardinports and
harbors.
Asmentionedearlier, ProbabilisticSeismicHazard
Analysis(PSHA) isbasedonmethodology originally
proposed by Cornell (1968) and is well documented
inmanyreferences(e.g. SSHAC, 1997). Themajority
of tsunamis are caused by earthquake-induced dis-
placement of theseafloor. Most of theworldslargest
tsunamis, which have caused damage at locations
thousandsof milesaway, havebeencausedbymegath-
rust (subduction interface) earthquakes around the
PacificRimandIndianOcean.Theseincludethe1960
Chileearthquake, the1964Alaskaearthquakeandthe
2004Sumatra-Andamanearthquake. Onalocal scale,
smaller earthquakescancausesignificanttsunamisas
well, butusuallythehazardfromtheseeventsislower
becauseof their localizedimpact.
A crucial element inPTHA istheestimationof the
frequencyof occurrenceandmaximummagnitudesof
largetsunami-generating earthquakes in each source
region. Due to the very short historical record for
mega-thrustsandotherlargeearthquakesinrelationto
their recurrencetimes, it is not possibleto basesuch
constraints directly on theobserved seismicity. Thio
et al. (2007) therefore used models that were partly
basedonearthquakemechanics, whichcanbeassim-
pleas magnitude/arearelations but can also include
physically-based constraints in addition to empirical
data such as earthquake locations. Uncertainties in
sourceparameters,suchassliprateandmaximumpos-
sibleearthquakeonasource,wereincludedusinglogic
treeanalysis.
Tsunami hazardassessmentmethodologiesareone
of themainresearchtopicswithintheprojectTRANS-
FER (Tsunami Risk andStrategies for theEuropean
Region, http://www.transferproject.eu/). TRANSFER
aims at improving the understanding of tsunamis in
theEuro-Mediterraneanregion, includinghazardand
riskassessment andstrategiesfor riskreduction.
5.3 Tsunami risk assessment
Tsunami vulnerability and risk assessment is arela-
tively unexplored discipline, and few reliable mod-
els exist. The Tsunami Pilot Study Working Group
(2006) liststhefollowingtsunami parametersaspos-
sibleimpactsmetricsthat may enter asparametersin
tsunami vulnerability models(i.e. mortality, building
damage, forcesonstructures):
Tsunami flowdepth
Wavecurrent speed
Wavecurrent acceleration
Wavecurrent inertiacomponent (product of accel-
erationandflowdepth)
Themomentumflux(product of squaredwavecur-
rent speedandflowdepth). Inmanycircumstances
thisisthebest damageindicator.
Theabovementionedparameters areimportant in
determining themortality of thetsunami, as well as
the wave forces on structures. The selection of the
flow depth is obvious, being a direct measure of
the thickness of the flowing water; the flow depth
is also related to the current velocity. In a national
tsunami risk evaluation for New Zealand, Berryman
et al. (2005) suggestedanempiricallyderivedmortal-
itymodel solelybasedontheflowdepthof thetsunami
(Figure10), however, wenotethat suchanapproach
is most likely too simplistic (see discussion below).
The fluid force on a structure is proportional to the
momentumflux, as well as impact forces of flotsam,
andhencealsoanatural possibilityasanimpact met-
ric. Perhaps more surprising is the inclusion of the
wavecurrent acceleration. A tsunami wavethat run-
uponthebeachwill oftenacceleratewhenit hitsthe
shoreline after breaking (Synolakis, 1987), and this
effectmaybecounterintuitivefor alaypersonobserv-
ingthetsunami, leadingtoamisinterpretationof the
escapetime.
Tsunami risk evaluationis thecombinationof the
tsunami hazard, tsunami exposure, and vulnerability
as described above. A risk evaluation may focus on
different elements at risk, for instance mortality or
destructionof buildingsor installations. For aproper
evaluation, it is therefore crucial to determine the
correct damage metrics. Generally, the population,
20
Figure 10. Empirical vulnerability (mortality) model of
Berrymanet al. (2005).
buildingsetc. exposedtotsunamisarefoundbycom-
bining flood maps with population density maps,
infrastructure maps and building maps in a GIS
framework.
Regional and global hazard evaluations aim at
roughquantificationof effectsof tsunami inundation,
and simple damage metrics and measures of expo-
sure should preferably be used. The tsunami flood
maps may be found using available computational
tools. However, approximatemethods for near-shore
waveamplificationusuallyhavetobeappliedforlarge
regions. Intheory, mortalityriskmaythenbeobtained
usingrelationssimilartotheoneinFigure10. Inprac-
tice however, the regional analysis is limited to the
hazard or at most, the population exposure. This is
becausemortality models aretoo simplistic, leaving
out anumber of important factors for mortality such
aslocal evaluationof tsunami velocityandmomentum
flux, tsunami travel time, effectsof warningsystems,
timeof tsunami attack (whichseason, what timedur-
ingtheday, ) etc., andhenceaddlittlevaluetothe
analyses.
Local risk evaluations on the other hand, can be
doneindetail andprovideinsightintoappropriatelocal
risk mitigation strategies. In alocal analysis, run-up
simulations may bedonefor smaller regions, which
allowfor amoreaccuratedescriptionof boththeflow
field and theinundated area. Furthermore, mapping
of thedifferent vulnerability parameters may beper-
formedinfar moredetail thaninregional evaluations,
enabling the possibility of mapping population and
buildingvulnerabilityat ahighlevel of accuracy.
6 GEOHAZARDSRISK MANAGEMENT
6.1 Risk management framework
Riskmanagementbroadlyreferstocoordinatedactiv-
ities to assess, direct and control the risk posed by
geohazards to the society. It integrates the recogni-
tion and assessment of risk with the development
of appropriate strategies for its mitigation. The risk
Figure11. Risk estimation, analysisandevaluationaspart
of riskmanagementandcontrol (NORSOK StandardZ-013,
2001).
management process is a systematic application of
management policies, proceduresandpracticestothe
tasks of communicating, consulting, establishing the
context, identifying, analyzing, evaluating, monitor-
ingandimplementingriskmitigationmeasures(Draft
ISO / IEC 31010 Ed. 1.0: Risk Management Risk
Assessment Techniques). As depicted in Figure 11,
risk assessment is an important component of risk
management.
Inthecontext of geohazards, Fell et al. (2005) pro-
videacomprehensiveoverviewof thestate-of-the-art
in landslide risk management. A large body of lit-
erature on earthquake risk management also exists.
However, tsunami riskmanagementisarelativelynow
topicandveryfewreferencesspecificallyaddressthis
issue.
6.2 Acceptable risk
One of the most difficult tasks in risk assessment/
managementistheselectionof riskacceptancecriteria.
As guidance to what risk level a society is appar-
ently willing to accept, one can use F-N curves.
TheF-N curvesrelatetheannual probability of caus-
ingNor morefatalities(F) tothenumber of fatalities,
N. ThetermNcanbereplacedbyother quantitative
measureof consequences, such as costs. Thecurves
can be used to express societal risk and to describe
the safety levels of particular facilities. Figure 12
21
Figure12. F-Ncurves(Proske, 2004).
presentsafamilyof F-N-curves. Man-maderiskstend
to have a steeper curve than natural hazards in the
F-N diagram(Proske, 2004). F-N curves givestatis-
tical observations andnot theacceptableor tolerable
thresholds.
Who should define acceptable and tolerable risk
level? The potentially affected population, govern-
ment, or the design engineer? Societal risk to life
criteriareflect thereality that society is less tolerant
of eventsinwhichalargenumber of livesarelostina
singleevent, thanif thesamenumber of livesislostin
alargenumberof separateevents. Examplesarepublic
concerntothelossof largenumbersof livesinairline
crashes, comparedtothemuchlarger number of lives
lost inroadtraffic.
Figure13presentsaninterimrisk criterionrecom-
mendationfor natural hillsides inHongKong(GEO,
1998).Acceptableriskreferstothelevel of riskrequir-
ingnofurther reduction. It isthelevel of risk society
desirestoachieve.Tolerableriskpresentstherisklevel
whichonecompromisestoinordertogaincertainben-
efits.A constructionwithatolerablerisklevel requires
no action/expenditure for reduction, but it requires
proper control andriskreductionif possible.
Risk acceptability dependsonseveral factorssuch
as voluntary vs. involuntary situation, controllabil-
ity vs. uncontrollability, familiarity vs. unfamiliar-
ity, short/long-termeffects, existenceof alternatives,
type and nature of consequences, gained benefits,
mediacoverage, availability of information, personal
involvement, memory, andlevel of trust inregulatory
bodies. Voluntary risk levels tend to be higher than
involuntaryrisklevels. Oncetheriskisunderpersonal
Figure13. HongKongcriteria(GEO, 2001).
control (e.g. drivingacar), it ismoreacceptablethan
the risk controlled by other parties. For landslides,
natural and engineered slopes can be considered as
voluntaryandinvoluntaryrisk.
Societies experiencing frequent geohazards may
havedifferent riskacceptancelevel thanthoseexperi-
encingthemrarely. Informedsocietiescanhavebetter
preparednessfor natural hazards.
Although the total risk is defined by the sumof
specific risk, it is difficult to evaluateits sum, since
theunitsfor expressingeachspecificriskdiffer. Indi-
vidual risk has the unit of loss of life/year, while
propertylosshastheunitof lossof property/year (e.g.
USD/yr). Riskacceptanceandtolerabilityhavediffer-
entperspectives: theindividualspointof viewandthe
societyspoint of viewor societal risk.
6.3 Risk mitigation strategies
The strategies for the mitigation of risks associated
withgeohazardscanbroadlybeclassifiedinsixcate-
gories: 1) landuseplans, 2) enforcement of building
codes andgoodconstructionpractice, 3) early warn-
ing systems, 4) community preparedness and public
awarenesscampaigns, 5) measurestopool andtrans-
fer therisksand6) constructionof physical protection
barriers. The first five strategies are referred to as
non-structural measures, which aim to reduce the
consequences of geohazards; while the last strategy
comprisesactiveinterventionandengineeringworks,
whichaimtoreducethefrequencyandseverityof the
geohazards.
22
Identification of theoptimal risk mitigation strat-
egy involves: (1) hazard assessment (how often do
thegeohazardshappen?), 2) analysisof possiblecon-
sequences for thedifferent scenarios, (3) assessment
of possible measures to reduce and/or eliminate the
potential consequences, (4) recommendation of spe-
cific remedial measureandif relevant reconstruction
andrehabilitationplans, and(5) transferof knowledge
andcommunicationwithauthoritiesandsociety.
6.4 Reducing the geohazards risk in developing
countries
Onecanobserveapositivetrendinternationallywhere
preventivemeasuresareincreasinglyrecognized, both
on the government level and among international
donors. There is, however, a great need for intensi-
fied efforts, becausetherisk associated with natural
disasters clearly increases far more rapidly than the
effortsmadetoreducethisrisk.
Threekey pillars for thereduction in risk associ-
atedwithnatural hazardsindevelopingcountriesare
suggested(modifiedfromKjekstad, 2007):
Pillar 1: Identify and locate the risk areas, and
quantify the hazard and the risk
Hazardandriskassessmentarethecentral pillarinthe
management of therisk associated with natural haz-
ards. Withoutknowledgeandcharacteristicsof hazard
and risk, it would not be meaningful to plan and
implement mitigationmeasures.
Pillar 2: Implement structural and non-structural risk
mitigation measures, including early warning systems
Mitigationmeansimplementingactivitiesthatprevent
orreducetheadverseeffectsof extremenatural events.
Inabroadprospective, mitigationincludesstructural
and geo-technical measures, effective early warning
systems, andpolitical, legal andadministrativemea-
sures. Mitigationalsoincludeseffortstoinfluencethe
lifestyle and behavior of endangered populations in
order toreducetherisk. TheIndianOceantsunami of
2004, whichkilledatleast230,000people, wouldhave
beenatragedywhatever thelevel of preparedness; but
evenwhendisaster strikesonanunprecedentedscale,
therearemanyfactorswithinhumancontrol, suchasa
knowledgeablepopulation, aneffectiveearlywarning
systemandconstructionsbuiltwithdisastersinmind.
All thesemeasures canhelpminimizethenumber of
casualties.
Improvedearly warningsystemshavebeeninstru-
mental inachievingdisaster risk reductionfor floods
and tropical cyclones. Cuba has demonstrated that
suchreductionisnot necessarilyaquestionof expen-
sive means. However, the recent tropical cyclone
Nargisisasadreminder thatmuchremainstobedone
indecreasingtherisktotropical cyclones.
Meteorological forecast in region wherecyclones
generally occur is quite effective, but early warn-
ing and responseremains insufficient in unexpected
regions(e.g. Catarina2004for SouthAtlanticOcean).
AsaconsequencethefocusonEarlyWarningSystem
(EWS) developmentshouldtakeintoaccountclimatic
changesand/or exceptional situations.
Pillar 3: Strengthen national coping capacity.
Most of thedevelopingcountrieslack sufficient cop-
ing capacity to address a wide range of hazards,
especially rare events like tsunamis. International
cooperation and support are therefore highly desir-
able. A number of countrieshaveover thelast decade
beensupportivewithtechnical resourcesandfinancial
means to assist developing countries where the risk
associated with natural hazards is high. A key chal-
lengewithall projects fromthedonor countries is to
securethat they areneed-based, sustainableandwell
anchored in the countries own development plans.
Another challenge is coordination which often has
provento bedifficult becausetheagencies generally
havedifferentpoliciesandtheimplementationperiods
of variousprojectsdonot overlap. A subject whichis
gainingmoreandmoreattentionistheneedtosecure
100%ownershipof theprojectinthecountryreceiving
assistance.
The capacity building initiatives should focus on
institutions dealing with disaster risks and disaster
situationsinthefollowingfour policyfields:
Risk assessment andcommunication, i.e. theiden-
tification, evaluationandpossiblyquantificationof
thehazardsaffectingthecountryandtheir potential
consequences, and exchange of information with
andawareness-raisingamongstakeholdersandthe
general public;
Riskmitigation, i.e. laws, rulesandinterventionsto
reduceexposureandvulnerabilitytohazards;
Disaster preparedness, warning and response, i.e.
procedures to help exposed persons, communities
and organizations be prepared to the occurrence
of ahazard; when hazard occurs, alert and rescue
activitiesaimedatmitigatingitsimmediateimpact;
Recovery enhancement, i.e. support to disaster-
strickenpopulations andareas inorder to mitigate
thelong-termimpact of disasters.
In each of thesefields, institutions can operateat
local, regional, national or international levels.
7 CONCLUDINGREMARKS
Management of the risk associated with geohazards
involvesdecisionsatlocal, regional, national andeven
transnational levels.Lackof informationabouttherisk
appearstobeamajorconstrainttoprovidingimproved
mitigationinmanyareas. Theselectionof appropriate
mitigation strategies should be based on a future-
oriented quantitative risk assessment, coupled with
useful knowledgeonthetechnical feasibility,aswell as
costsandbenefits, of risk-reductionmeasures.Techni-
cal expertsactingalonecannot choosetheappropri-
atesetof mitigationandpreventionmeasuresinmany
riskcontexts.Thecomplexitiesandtechnical detailsof
23
managinggeohazardsriskcaneasilyconceal that any
strategy is embeddedinasocial/political systemand
entailsvaluejudgmentsaboutwhobearstherisksand
benefits, andwhodecides. Policymakersandaffected
parties engaged in solving environmental risk prob-
lemsarethusincreasinglyrecognizingthat traditional
expert-based decision-making processes are insuffi-
cient, especially in controversial risk contexts. Risk
communicationandstakeholderinvolvementhasbeen
widely acknowledged for supporting decisions on
uncertainandcontroversial environmental risks, with
theaddedbonusthatparticipationenablestheaddition
of local andanecdotal knowledgeof thepeoplemost
familiar with the problem. Precisely which citizens,
authorities, NGOs, industry groups, etc., should be
involvedinwhichway, however, hasbeenthesubject
of atremendousamountof experimentation.Thedeci-
sionisultimatelymadebypolitical representatives,but
stakeholder involvement, combined with good risk-
communicationstrategies,canoftenbringnewoptions
tolight anddelineatetheterrainfor agreement.
The human impact of geohazards is far greater
in developing countries than in developed countries.
Capacitybuildinginitiativesfocusingonorganizations
and institutions that deal with disaster risks and dis-
aster situationscouldgreatly reducethevulnerability
of thepopulationexposedto natural disasters. Many
of theseinitiatives canbeimplementedwithinafew
years and areaffordableeven in countries with very
limitedresources.
ACKNOWLEDGEMENT
Theauthor wishestothank hiscolleaguesat ICGfor
their direct and indirect contributions to this paper.
Special thanks are due to Prof. Hilmar Bungum
of NORSAR (earthquake), and Drs Carl Harbitz
(tsunami), Finn Lvhlt (tsunami) and Suzanne
Lacasse (landslide and risk management) of the
NorwegianGeotechnical Institute.
REFERENCES
Ang, A.H-S. & Tang, W.H. 1984. Probability Concepts in
Engineering Planning and Design I & II, J ohnWiley &
Sons, NewYork.
Berryman, K. etal. (editors) 2005. Reviewof tsunami hazard
andriskinNewZealand.Geological andNuclearSciences
(GNS) report 2005/104. 140p.
Bilham, R. 2004. Urbanearthquakefatalities: A safer world
or worsetocome? Seism. Res. Lett., December 2004.
Coppersmith, K.J., & Youngs, R.R. 1986. Capturing uncer-
tainty in probabilistic seismic hazard assessment within
intraplate tectonic environments. Proc. Third US Natl.
Conf. onEarthquakeEngineering, EarthquakeEngineer-
ingResearchInstitute, Vol. 1, pp. 301312.
Cornell, C. A. 1968. Engineeringseismicriskanalysis. Bull.
Seism. Soc. Am58, pp. 15831606.
Duncan, J.M. 1992. State-of-the-art: Static stability and
deformationanalysis. Stabilityandperformanceof slopes
andembankments-II, 1: 223266.
Duncan, J.M. 1996. Soil slopestabilityanalysis. Landslides:
investigation and mitigation. Ed. By Turner & Schuster.
Washington1996TRB Report 247.
Duncan, J.M. 2000. Factors of safety and reliability in
geotechnical engineering, J. of Geotechnical andGeoen-
vironmental Engineering126(4): 307316.
Dzgn, S. &Lacasse, S. 2005.Vulnerabilityandacceptable
risk in integrated risk assessment framework. Landslide
Risk Management, Hungr, Fell, Couture & Eberhardt
(eds), Taylor & Francis, London: 505515.
Federal EmergencyManagementAgency2003.HAZUS-MH
MR3Technical Manual Earthquake Model. Web site:
http://www.fema.gov/plan/prevent/hazus/index.shtm
Fell, R., Ho, K.K.S., Lacasse, S. &Leroi, E. 2005. A frame-
work for landslide risk assessment and management
Stateof theArt Paper 1. LandslideRisk Management,
Hungr, Fell, Couture&Eberhardt(eds),Taylor&Francis,
London: 325.
GEO (Geotechnical Engineering Office) 1998. Landslides
and Boulder Falls from Natural Terrain: Interim Risk
Guidelines. GEOReport 75, Gov. of HongKongSAR.
IFRC (International Federation of Red Cross and Red
Crescent Societies) 2001. World Disaster Report, Focus
onReducingRisk. Geneva, Switzerland, 239pp.
IFRC (International Federation of Red Cross and Red
Crescent Societies) 2004. WorldDisaster Report.
ISDR (International Strategy for Disaster Reduction) 2005.
HyogoFrameworkforAction20052015, 21pp.
Kjekstad, O. 2007. Thechallenges of landslidehazardmit-
igation in developing countries. Keynote Lecture pre-
sentedat1stNorth-AmericanLandslideConference,Vail,
Colorado38J une2007.
Kulkarni, R.B., Young, R.R. & Coppersmith, K.J. 1984.
Assessment of confidenceintervalsfor resultsof seismic
hazardanalysis. Proc. EighthWorldConf. onEarthquake
Engineering, SanFrancisco, Vol. 1, pp. 263270.
Lacasse, S. &Nadim, F. 2008. Landslideriskassessmentand
mitigationstrategy.InvitedLecture,State-of-the-Art.First
World LandslideForum, Global LandslideRisk Reduc-
tion, International Consortium of Landslides, Tokyo.
Chapter 3, pp. 3161.
Lee,E.M.&J ones,D.K.C.2004.LandslideRiskAssessment.
ThomasTelford, London.
McGuire, R. 2004. SeismicHazardandRiskAnalysis. EERI
monograph(MNO-10), ISBN: 0-943198-01-1, 221p.
MunichReGroup2007. NatCatService2007 Greatnatural
disasters19502007.
Nadim, F. 2004. Risk andvulnerability analysisfor geohaz-
ards. Glossary of Risk Assessment Terms. ICG Report
2004-2-1, NGI Report 20031091-1, OsloNorway.
Nadim, F. Einstein, H.H. &Roberts, W.J. 2005. Probabilistic
stability analysis for individual slopes insoil androck
State of theArt Paper 3. Landslide Risk Management,
Hungr, Fell, Couture&Eberhardt(eds),Taylor&Francis,
London: 6398.
Nadim, F. & Glade, T. 2006. Ontsunami riskassessment for
thewestcoastof Thailand, ECI Conference: Geohazards
Technical, Economical andSocial RiskEvaluation1821
J une2006, Lillehammer, Norway.
Nadim, F., Kjekstad, O., Peduzzi, P., Herold, C. & J aedicke,
C. 2006. Global landslideandavalanchehotspots. Land-
slides, Vol. 3, No. 2, pp159174.
Nadim, F. & Lacasse, S. 2008. Effects of global changeon
risk associated with geohazards in megacities. Keynote
Lecture, Development of UrbanAreas andGeotechnical
Engineering, St. Petersburg, Russia, 1619J une2008.
NORSOK standardZ-0132001. Risk andemergency pre-
parednessanalysis, Rev. 2, www.standard.no
24
Proske, D. 2004. Katalogder Risiken. EigenverlagDresden.
372p.
Roberds,W.J. 2005. Estimatingtemporal andspatial variabil-
ityandvulnerability Stateof theArt Paper 5. Landslide
Risk Management, Hungr, Fell, Couture & Eberhardt
(eds), Taylor & Francis, London: 129158.
SSHAC (Senior Seismic Hazard Analysis Committee)
1997.RecommendationsforProbabilisticSeismicHazard
Analysis: Guidanceon Uncertainty and Useof Experts,
US Nuclear Regulatory. Commission report CR-6372,
WashingtonDC.
Synolakis, C.E. 1987. Therun-upof solitarywaves, J. Fluid
Mech., 185, pp. 523545.
Thio, H.K., Somerville, P. &Ichinose, G. 2007. Probabilistic
Analysisof StrongGroundMotionandTsunami Hazards
in Southeast Asia. Proceedings from2007 NUS-TMSI
Workshop, National University of Singapore, Singapore,
79March.
Tsunami Pilot Study Working Group 2006. Seaside,
OregonTsunami Pilot StudyModernizationof FEMA
flood hazard maps. NOAA OAR Special Report,
NOAA/OAR/PMEL, Seattle, WA, 94pp. +7appendices.
UNDP (United Nations Development Programme) 2004.
ReducingDisaster Risk A Challengefor Development,
Bureau for Crisis Prevention and Recovery, NewYork,
146pp.
UNISDR (2009). Terminology onDisaster Risk Reduction.
http://www.unisdr.org/eng/library/UNISDR-terminology-
2009-eng.pdf
WorldDisaster Report 2006. http://www.redcross.ca
25
Riskmanagement anditsapplicationinmountainoushighway
construction
H.W. Huang&Y.D. Xue
Key Laboratory of Geotechnical & Underground Engineering, Ministry of Education
Department of Geotechnical Engineering, Tongji University, Shanghai, P.R. China
Y.Y.Yang
Chief Engineering Office, Shanghai HuShen Highway Construction Development Co., Ltd.
Shanghai, P.R. China
ABSTRACT: Moreandmorehighwayswill beconstructedinmountainousareasof westernChina.Themoun-
tainous highway projects are subject to more risk than other constructions because they entail intricate site
conditionsandinheresomany uncertainties. Risk management isaneffectiveapproachtoreduceandcontrol
therisks reasonably for achievement of project objectives. How to put therisk management techniques into
practicein thehighway projects is on thefocus in this paper. Based on thestudy of therisk mechanism, an
integratedproject risk management framework isput forward. Theorganizationof arisk management teamis
veryimportant andeachmember shouldbeselectedsuitably. Risk identificationshouldbeimposedwithmore
effortsbecauseitsresultshavesignificantinfluencetotheriskmanagementaims. A synthetical rocktunnel risk
identificationmethodbasedonFaultTreeAnalysis(FTA) isproposedandnamedasRTRI. Therisk identifica-
tionprocessisgenerally conductedinaniterativecyclewhichkeepsstepwiththeconstruction. Risk database
is important and useful to risk management which is discussed in detail in another paper. Theproposed risk
management model for mountainous highway projects is practiced in Shuifu-Maliuwan Highway inYunnan
Province, China. Therisksof constructiontimedelay, cost overrun, poor qualityandworker safetyareassessed
carefully. Theresultsareprovedrelevant andhelpthedecision-maker todeal withtheriskseffectively.
1 INTRODUCTION
In general, the highway construction projects are
always very complex and dynamic (Akintoye &
MacLeod 1997, Carr & Tah 2001). Each project
involvesavarietyof organizationsandalargernumber
of people.Itisknownthatall participantsof aconstruc-
tion project arecontinuously faced with avariety of
unexpectedor unwantedsituations. Theprojectshave
anabundanceof riskduetothenatureof construction.
It is recognized that risk is built into any actions of
people.
In a construction project, the risk is perceived as
events that influence the project objectives of cost,
quality, time, safety and environment. To reduce or
control theprojectrisk, mostpeopleagreethatproject
riskmanagement (PRM) issuitable(Wideman1986).
Lyons&Skitmore(2004) hadsurveyedthattheuseof
riskmanagement intheQueenslandengineeringcon-
struction industry is mediumto high. It has been a
critical partof integratedprojectmanagementsystem.
Even though most people think risk management
playacrucial roleinproject management andnumer-
ouspapersonsuchsubjecthavebeenprinted,theactual
use of risk management in practice is limited. Lack
of an industry accepted model of risk analysis may
be one factor that limits the implementation of risk
management (Lyons& Skitmore2004).
Along with the rapidly economic development,
moreandmorehighwayswill beconstructedinmoun-
tainous region. The characteristics of these projects
include complex geology, poor transport condition,
badweather, earthquakeandsoon. Ontheother hand,
theseprojectsalwaysincludemoreengineeringtypes:
tunnel, bridge, slope, subgradeandroadsurface. The
potential risks include tunnel collapse, water burst,
landslide, rockfall, injury, explosion, etc. Its evident
that themountainareahighway constructionisfaced
withsignificant risks.
This paper aims to supply apracticablerisk man-
agement framework, anddescribetheprocessof each
step. At last, acasestudyisconducted.
2 GENERAL RISK MANAGEMENT
Risk management is a systemwhich aims to iden-
tify and quantify all risks to which the project is
exposed so that a conscious decision can be taken
on howto managetherisks (Roger & George1996,
27
Risk Identification
Risk Classification
Risk Analysis
Risk Response
Risk Attitude
Figure 1. The risk management framework (Roger &
George1996).
Eskesen et al. 2004, MOHURD 2007). It includes a
series of processes. TheAssociationof Project Man-
agersdefinesninephasesof riskmanagement: define,
focus, identify, structure, ownership, estimate, evalu-
ate, plan, andmanage(APM Group1997, Chapman
1997). Chapman (2001) think that all theprocess of
risk analysis and management is composed of two
stages, risk analysis and risk management. Roger
defines some general stages are: risk identification,
risk classification, risk analysis, risk attitude and
risk response. The risk management framework is
illustratedinFigure1.
The definition of project risk is important before
it canbemanaged. Widemandefinesproject risk as
thechanceof certainoccurrencesadversely affecting
project objectives. Chapman(2001) definesit asan
event, which should it occur, would have a positive
or negative effect on the achievement of a projects
objectives. The main difference is that whether or
not consider therisk chance. Ingeneral, construction
project risk has two types: purerisk and speculative
risk. Inthispaper, theriskisdefinedasafunctionof
potential adverse events occurrence probability and
consequence. It is clear that only thepurerisk con-
cept is usedinthehighway constructionproject risk
management.
Risk identification is themost important stagein
risk management. Unidentifiedrisksmayhidesevere
threatstoaprojectsobjectives. Categorizationof the
sourceof risk ishelpful torisk identification. British
Standard6079considers that risks or adverseevents
generallyfall intooneof thefollowingfivecategories:
technological, political, managerial, sociological and
financial. Zayed et al. (2008) divided the highway
construction risks into two areas: company (macro)
and project (micro) levels. In the micro hierarchy,
emergingtechnologyusage, contractsandlegal issues,
resources, design, quality, weather, etc. areincluded.
Technological risk assessment is ingeneral themain
sectionof constructionriskmanagement.
3 MOUNTAINOUSHIGHWAY
CONSTRUCTION
3.1 Features
In China, with the rapidly economic development,
moreandmorehighways areconstructedor planned
inmountainous areas. Theengineeringenvironments
andconditions arefar fromtheplainarea. Themain
characteristicsof mountainareahighwaycanbesum-
marizedas:
Highmountain
inconvenient transportation
badweather
unfavorablegeology
limitedconstructionmethods
longtunnels
longbridgeswithhighpier
deepcut slope
constructionmaterial shortage
earthquakeand
soon.
Ingeneral, themainengineeringtypesincluderock
tunnel, bridge, cutting/ nature slope, subgrade and
roadsurface. Accordingtotheresults of surveys and
statisticstomostaccidentsrelatedwithmountainroad
engineeringinthelast decade, theslopefailureacci-
dentsarethemost frequent andseverest nomatter in
quantity or loss. Thenext oneislongtunnel whichis
alwayssufferedtopoorgeological condition.Thenthe
riskishighpier bridge. Newtechniquesandmaterials
arewidey usedinbridgeconstruction. This situation
hintspotential severerisks. Theaccidentsabout road-
subgradeandroadsurfacearemostfacedinoperation
phase, seldominconstructionphase.
3.2 Risk mechanism
Thereis abundant severepotential risk inherent with
highway construction under such bad conditions.
For the sake of effective risk management, the risk
mechanismshould bestudied at first. A simplerisk
mechanismisobtainedbasedonproject management
experienceandtheoretical studies. Themechanismof
riskdevelopment isillustratedinFigure2.
Figure2canhelppeopleto understandthenature
of construction risks. Activities, materials, tools and
all kinds of equipments related with human can be
considered as internal causes of risk event. Related
surrounding environment can bethought as external
causes of risk event. Through acareful screening of
theproject, themostriskfactorscanbeidentified.The
riskeffectsmainlyincludeeconomicloss, timedelay,
casualty,qualityloss,environmentloss,etc.According
to theaimof risk assessment, therisk effects canbe
categorizeddifferently.
4 MOUNTAINHIGHWAY CONSTRUCTION
RISK MANAGEMENT
It isknownthat therearenumerousproject risk man-
agementmethodsindifferentengineeringareas.A risk
28
Riskfactors
Risk
surroundings
Risk Vulnerbility Effect
Time
Quality
Safety
Environment
Cost
Construction
technology,
Machinery,
Operation,
and etc.
Casualties
Quality loss
Time delay
Economic loss
Ecological
environment
loss
Project risk
I
n
t
e
r
n
a
l

e
n
v
i
r
o
n
m
e
n
t
a
l

f
a
c
t
o
r
s
E
x
t
e
r
n
a
l

e
n
v
i
r
o
n
m
e
n
t
a
l

f
a
c
t
o
r
s
G
e
o
l
o
g
y

a
n
d

h
y
d
r
o
-
g
e
o
l
o
g
y
P
r
o
j
e
c
t

d
e
c
i
s
i
o
n
-
m
a
k
i
n
g
Figure2. Riskmechanismdiagram.
management framework for mountain highway con-
struction is proposed. It will also be accepted as a
basicriskmanagement structureinGuidelineof Risk
Assessment&Control forSafetyConstructionof Road
Tunnel inChina.
4.1 Risk management flowchart
The whole phase of a road construction project
includesprojectdevelopmentstage, constructioncon-
tract procurement stage, design stages, construction
stageand operation stage(ITIG 2006). Most people
agreethattheriskmanagementshouldbeusedthrough
thewholelifecycletoreducetherisk. Atpresent, risk
managementapplicationintheexecutionandplanning
stages of the project life cycle is higher than in the
conceptual or terminationphases(Lyons& Skitmore
2004). Inthis paper, themainproject phases include
preliminary design, construction documents design
andconstruction. Theflowchart of risk management
isillustratedinFigure3.
4.2 Risk management
4.2.1 Risk management team
To anewproject/phase, anorganizationof risk man-
agementteamshouldbeestablishedasthefirsttaskof
therisk management. A risk analysis specialist plays
acrucial rolein theteamorganization, and this will
significantlyinfluencethefollowingriskmanagement
processes. Theteammembers aredynamic, andwill
changeindifferent phasesor special riskissues.
Toariskmanagement team, riskspecialist, experi-
encedproject managersandclient representativesare
coremembers. Other membersmaybeinclude,
indesignphase:
representativesof thecoredesignteam,
designteamrepresentative,
consultant engineer and
other relatedpersonnel.
inconstructionphase:
principal designer or representative,
contractor representative,
consultant engineer,
supervisor representativeand
other relatedpersonnel.
4.2.2 Scope, objective and strategy
The scope of risk management defines the bench-
mark information such as clients objective, why
conduct the risk assessment, who will execute and
control the process, when and how to assess the
risk, anticipatedachievementandother critical issues.
Thescopeshould bedocumented as task instruction
document.
General objective of a construction project risk
management is identify and quantify all risks and
consciously manage them. For risk management is
time-person-cost-consuming, it is not reasonable to
pay much attention to all low level risks. The detail
of objectiveand clients circumstancewill influence
the depth of risk analysis. The objective of moun-
tain highway construction is dealing with the risks
as low as reasonably practicable(known as ALARP
principle).
Themountain highway construction risk manage-
ment strategyinclude:
carry out risk assessment through out the whole
project constructionprocess
clarifytheriskshareof thevariouspartiesinvolved
intheproject
aplanof dynamicriskassessment
atrainingof riskviewpoint toall personsinvolved
a standard risk document format, including risk
register andriskmeasures
29

Project phases
Preliminary design
phase
Construction documents
design phase
Construction phase
Client
Risk management
specialist
Scope
Objectives
Strategy
Risk management team
Engineering
type or area
Risk identification
Risk evaluation
Risk response and
monitor
Risk document and
database
Risk acceptance criteria
Contractor and others
Figure3. Riskmanagement flowchart.
4.2.3 Engineering types or areas
A typical mountain highway construction project is
alwayssocomplexthatriskmanagementisveryhard.
Separation is an effectivesolution to complex prob-
lem. Soit isusedheretoseparateaproject intoaset
of baseelementsfor structuringthemanagement. Itis
natural and reasonableto separateproject according
toitsengineeringtypes. Ingeneral, amountainhigh-
wayconstructionprojectcanbeseparatedintotunnel,
bridge, slope, subgrade, road surface, etc. A typical
kind engineering is usually complex, too. It can be
separatedintosub-engineeringaccordingtoengineer-
ingareasfutures.Atunnel projectcanbeseparatedinto
tunnel portal section, poor geology sectionandother
sectiondependsonthenatureof thetunnel engineer-
ing. A bridge will be separated into superstructure,
substructureand connection section. A slopecan be
categorizedas natureslope, cuttingslope, rock slope
andsoil slope.
4.2.4 Risk identification
Risk identification is themost important step in the
overall process of risk management. Unidentified
andthereforeunmanagedrisksareclearlyunchecked
threats to a projects objectives, which may lead to
significant overruns. Floricel & Miller (2001) got
that regardless of a thorough and careful identifica-
tion phase, something unexpected occurred in every
project. A survey result shows that risk identifica-
tionandrisk assessment arethemost oftenusedrisk
management elements aheadof risk control andrisk
documentation. Thequalityof identificationresultsis
greatly dependent on theteams professional experi-
enceandknowledge. Atthesametime, theidentifica-
tiontechniqueplaysanimportant role.
Risk identification methods generally include
brainstorming, risk checklists, expert analysis/
interviews, modellingandanalyzingdifferent scenar-
ios and analyzing project plans. Brainstorming and
expert questionnaire(Ahmad& Minkarah 1988) are
themost commonrisk identificationtechniquesused
inroadconstructionproject inChina.
Based on the Fault TreeAnalysis method, a syn-
thetical identification method(namedRTRI) is used
effectively in many tunnel construction projects. Its
operationstructureis illustratedinFigure4. Thekey
of this method is that severeand general risk events
aredistinguished. It iseasy tocontrol therisk identi-
ficationdepth, andcanhelptounderstandtheinternal
logicrelationof differentrisks. Itisnecessarytobuild
upariskdatabaseandregister all therisksidentified.
Thedatabaseisveryuseful andhelpful toidentifyrisk
of anewanalogousproject.
4.2.5 Risk evaluation
After theriskshavebeenidentified, theymustbeeval-
uated in terms of the probability of occurrence and
30
Figure4. Rocktunnel constructionriskidentificationmethod(RTRI).
Table1. Riskoccurrenceprobabilityranking.
Ranking 1 2 3 4 5
Impossible Seldom Occasional Possible Frequent
Occurrence P-0.01% 0.01%p-0.1% 0.1%p-1% 1%p-10% P10%
probability
Table2. Riskimpact ranking.
Ranking 1 2 3 4 5
Impact Insignificant Considerable Serious Severe Disastrous
impact. In practice, the risk probability and impact
can be analyzed based on the historic statistic data.
But for mountain highway construction project, the
data are always very scarce. In this paper, probabil-
ityandimpact rankingislistedinTable1andTable2
(MOHURD2007) .Theriskimpactrankingisvariable
withdifferent risksandclient riskattitude.
When the risk occurrence probability and impact
is defined, therisk canberatedaccordingtotherisk
evaluation matrix (Table 3). For easy use, a colored
standing for risk ranking is made as list inTable 4.
The mountain highway construction risk acceptance
criterionis describedinTable5. Thecriteriacanaid
thedecisionmaker todeal withtherisk.
4.2.6 Risk response and monitor
Risk responseisastrategytakentomanagetheiden-
tifiedrisks. Ingeneral, therearefour basic forms of
31
Table3. Riskevaluationmatrix(riskranking).
Impact
Risk Insignificant Considerable Serious Severe Disastrous
Probability 1 2 3 4 5
Impossible 1 I I II II III
Seldom 2 I II II III III
Occasional 3 II II III III IV
Possible 4 II III III IV IV
Frequent 5 III III IV IV IV
Table4. Coloredstandingfor riskrank.
Riskranks Logo Colour
Table5. Riskacceptancecriteria.
response or control strategies which can be used in
risk management. Thefour typesof risk responseare
acceptance, mitigation, transfer and avoidance. The
decision of risk response should consider the risk
acceptance criteria (Fig. 5). When a risk is identi-
fiedandevaluated, theundertakershouldtakespecific
actions to control it. Any actions or measures should
beanalyzed carefully in order to achievetheproject
objectives.
Asanimportant part of risk response, project con-
tingencyarrangementshouldbeestablishedforcritical
risk. It includestheriskactionschedules:
Actionsrequired(what istobedone);
Resources(what andwho);
Responsibilities(who) and
Timing(when).
Once the risk responses have been defined, the
project risk source should be monitored in time
throughout the construction. The states of risk are
logged into therisk database. Therisk status can be
definedasTable6.
4.2.7 Risk document and database
Theriskmanagementisasystematicprocess.All mate-
rials relatedwiththeprocess shouldbedocumented.
Thematerials includeall kinds of memos, statistical
data, photos, design and construction specifications,
etc. A comprehensiveriskregister formlogstherisks
statusandinformation.Formattedriskinformationcan
Table6. Proposal for riskstatusdefinitions.
Riskstatus
Identified
Assessed
Responsesimplemented
Occurred
Avoided
Closedout
becommunicatedeffectively amongall parties. Risk
communicationisakeypoint tosuccessful risk man-
agement. A softwarebasedondatabasetechniquehas
been developed for standard implementation of risk
management in mountainous highway construction
project.
5 CASE STUDY
5.1 Project introduction
Shuifu-MaliuwanHighway(shortedasShui-MaHigh-
way) locatesintheadjacentareaof theYungui Plateau
and Liangshan mountain, which is in northeast of
YunnanProvinceinChina.Thetotal lengthof thehigh-
way is135.55kmwiththreeyears constructiontime
and the total general estimate of project investment
is92billionRMB. OwingtoYanshanandHimalayan
tectonicmovement, thetectonicdeformationisheavy.
Therearehighmountains,steepgorgeswithheavyero-
sion, rapidriversandsaw-cutseverywhere. Thereare
39tunnelswithatotal lengthof 27.21km, 365bridges
withatotal lengthof 91.4kmandmanycutslopesand
talus slopes alongthis highway inthewholeproject.
Thewholeproject was formally commencedwith28
biddingcontractsinMarch2005. Somecharacteristics
of thisproject areillustratedinFigures6to8.
5.2 Shui-Ma Highway construction risk
management
According to the aforementioned method, the risk
management iscarriedout for theShui-MaHighway
constructionproject, includingriskidentification, risk
evaluation, riskresponseandriskdocumentation. The
risksarecategorizedintotime,cost,qualityandhuman
32
Risk response
Risk acceptance criteria
Negligible risk Unwantedrisk Acceptable risk Unacceptable risk
Risk acceptance Risk transfer Risk mitigation Risk avoidance
Figure5. Theapproximaterelationbetweenriskresponseandacceptancecriteria.
Figure6. Photoof constructionof Guanhebridge.
Figure7. Thepanoramaof theancient landslide.
safety. Becausethisproject isverylongandcomplex,
theproject is separatedinto 28sections accordingto
thebiddingcontracts. Asanexampleof theriskman-
agement, therisksof Contract 11wereevaluated. The
occurrenceprobabilityrankingof riskeventisdecided
according to Table1 based on historic events statis-
tic or experienceof highway constructionspecialists.
Theimpact ranking is decided according toTable2.
TheRisk evaluationmatrix (Table3) is usedfor risk
ranking. TheresultsarelistedinTable7Table10(P
Figure8. Hazardof rockfall.
Table7. Riskevaluationresultsof timedelay.
No. Riskevents P C R
R10 Inconvenient trafficconditions 3 2 II
R11 Poor weather conditionresult in 5 3 IV
short effectiveconstructiontime
R12 Poor social environment including 2 2 II
thegovernment mismanagement and
thebackwardnessof peopleidea
R13 Delayof thedesignalteration, design 3 2 II
changeandthedesignapproval
R14 Supplement of theunifiedsupply 3 2 II
materialsarenot intime
R15 Narrowconstructionsiteandpoor 4 2 III
constructionconditionsresult
indifficult transportation
R16 Power supplyisnot intime 2 3 II
andthesupplyisunstable
R17 Thetreatment andinfluence 3 3 III
of unfavorablegeology
R18 Unreasonableresourcepreparing 3 3 III
result inrun-idle
33
Table8. Riskevaluationresultsof cost overrun.
R20 Insufficient credit level of the 3 2 II
insurancecompany
R21 Thepricerisingof rawmaterials, 4 4 IV
fuel andlabor cost
R22 Natural disastersincludingflood 4 4 IV
anddebrisflow
R23 Owner improvesthequalitystandard 2 3 II
andcontractor underestimatesthe
exist risks
R24 Unfavorablegeologyincluding 3 5 IV
landslide, rockheapandrockfall
R25 Inconvenient trafficconditionsresult 3 3 III
intheincreasingof thematerials
transportationcost
R26 Constructioncontract risks 3 3 III
R27 Poor humanenvironment 1 2 I
R28 Designrisks(designalterationresults 2 4 III
intheincreasingof investment)
R29 Narrowconstructionsiteresults 2 4 III
intheincreasingof thespoil cost
Table9. Riskevaluationresultsof poor quality.
R30 Constructioncontract risks(short 5 3 IV
constructionperiodandlowbidprice)
R31 Poor qualityof equipmentsandmaterials 1 2 I
R32 Protectionrisks(incorrect constructionof 2 3 II
anti-slidepileandshot Creteandanchor)
R33 Sitepersonnel quality 3 3 III
R34 Technologyrisksof tunnel construction 2 4 III
R35 Technologyrisksof subgradeand 3 3 III
pavement construction
R36 Technologyrisksof bridgeconstruction 4 1 II
R37 Unfavorablegeology 2 4 III
R38 Poor weather condition 4 2 III
inthetables means probability, C means impact and
R meansriskrate).
The risks of time delay (Table 7) show that the
poor weather condition is a critical risk factor to
the schedule. The adverse factor is mainly raining
in the project area. There are almost 2/3 days rain-
inginayear, andtheeffectiveworkdays arelimited.
Becausetheweatherconditionisuncontrolled, thebet-
ter riskresponseshouldbedrawingagoodplanwhich
arranges the project processes thinking of weather
influence. Table 8 shows that the major cost risks
includepricerisingof materials, natural disastersand
geo-hazards. The materials cost risk should be con-
sidered in the tendering. Different natural disasters
risks should beassessed in detail. Then thecontrac-
tor has ameetingto proposepracticablerisk control
measures and emergency plans. To reduce the geo-
hazardsrisk, thedesignof rock reinforcement should
be certificated base on the new geological survey
Table10. Riskevaluationresultsof humansafety.
R40 Unfavorablegeology 3 3 III
R41 Insufficient safetyconsciousness 3 3 III
of constructors
R42 Improper constructionprotection 3 2 II
R43 Improper equipmentsmanipulate 2 1 1
R44 Highaltitudeconstruction 3 2 II
R45 Contact of injurant 1 2 I
R46 Roadtrafficaccident 1 2 I
R47 Natural disasters 3 3 III
R48 Custodyof initiatingexplosive 2 3 II
deviceandblastingconstruction
R49 Power failurewhentheconstruction 1 5 III
of thediggingpileandthe
overturnof thebridgemachine
data, and thesupporting parameters can bechanged
if necessary. Themajor quality risk iscontract which
is quoted with unreasonable low-price bid. Another
majorfactorisunreasonableprojectscheduleandtime.
Theworker safety risk is acceptable. Risk education
andreminder areeffectivemanagementapproachesto
improvesafety.
Shi rock tunnel project is the key sub-project of
Contract11, andtherearesomanyuncertaintiesinits
constructionthat thepotential risksarevery big. The
risk management systemmust beapplied to achieve
thefinal objectives.
5.3 Shi rock tunnel construction risk management
Shi tunnel isaseparatedtwo-single-tunnel situatedin
Contract 11. Its left tunnel is approximately 4,752m
in length, and the cover depth varies from1.5mto
352m. Theright tunnel is approximately 4,755min
lengthanddepthvariesfrom0mto357m. Thetunnel
cross-sections are 11.7mwide and 7mhigh. Tradi-
tional Drill andBlastmethodisemployedinthetunnel
excavation. The conditions of the portal section are
evenworsewhichcontainsfaultsandfragmentzones.
5.3.1 Risk identification and assessment
Based on in-situ investigation, tunnel construction
designandother relatedmaterials, themainrisksand
general risksareidentifiedwithRTRI.Thetunnel risk
breakdown structure is listed in Figure 9. When all
therisksareidentified, Delphi methodisemployedto
evaluatetherisk rating. Therisk evaluatedresultsare
giveninTable11.
FromTable10, themanager can easily grasp and
understandtheriskprofileof thetunnel project. Then
thekey points of risk management transportedto all
project parties. Because the Shi tunnel is very long
anditsgeological conditionschangegreatlyalongthe
tunnel alignment, thetunnel is separatedinto several
sections according to thegeological conditions. The
risks of eachsectionareevaluatedanddiagramedin
Figure10.
34
Risks of Shi Tunnel during construction
Main risks General risks
T
u
n
n
e
l

p
o
r
t
a
l

s
l
o
p

f
a
i
l
u
r
e
W
a
t
e
r

o
r

m
u
d

b
u
r
s
t
i
n
g

T
u
n
n
e
l

c
o
l
l
a
p
s
e
S
u
r
r
o
u
n
d
i
n
g

r
o
c
k

l
a
r
g
e

d
e
f
o
r
m
a
t
i
o
n
R
i
p

s
p
a
l
l
i
n
g

a
n
d

f
l
o
o
r

h
e
a
v
e
B
l
a
s
t
i
n
g

a
c
c
i
d
e
n
t
s
E
l
e
c
t
r
i
c

s
h
o
c
k

a
c
c
i
d
e
n
t
s
M
a
c
h
i
n
e

a
c
c
i
d
e
n
t
s
H
i
g
h

f
a
l
l
i
n
g

a
c
c
i
d
e
n
t
s
H
a
r
m
f
u
l

g
a
s
e
s

p
o
i
s
o
n
i
n
g
F
i
r
e

a
c
c
i
d
e
n
t
s

G
a
s

a
c
c
i
d
e
n
t
s

W
a
t
e
r
f
l
o
o
d
i
n
g

a
c
c
i
d
e
n
t
s
C
o
n
s
t
r
u
c
t
i
o
n

t
i
m
e

d
e
l
a
y
E
n
v
i
r
o
n
m
e
n
t
a
l

i
m
p
a
c
t
s
Figure9. Possiblerisksof Shi tunnel duringconstruction.
Table11. Riskevaluationresults.
5.3.2 Sub-risk factors analysis
Forriskcontrol,thesub-riskfactorsof mainriskevents
are analyzed with specialist questionnaire method.
FromFigure 11 we can find that supporting qual-
ity, overburden, excavationmethod, etc. arethemain
factors that result in the occurrence of collapse, so
during the construction, special attention should be
paid to these factors in order to reduce the proba-
bility of collapse occurrence. Figure 12 shows that
advanced geological forcast, construction organiza-
tiondesign,undergroundwater disposal, etc. affect the
water bursting and mud surging significantly. Spe-
cial attention should be paid to these factors during
the construction. Figure 13 shows that the excava-
tionmethod,blastingmethodandadvancedsupporting
influencethetunnel portal stabilitysignificantly. Asa
systemrisk control measure, therock slopreinforce-
mentqualityandportal sunrroundingrocksupporting
paramaters must satisfy thedesigndocuments (Yang
et al. 2006, Pine& Roberds 2005). FromFigure14,
wecanfindthat excavationmethod, blastingmethod,
cyclical footageandadvancedsupport, etc. aremuch
more important factors which will result in large
deformation.
Theriskcontrol methodsof mainrisksandgeneral
risksareproposedaccodingtotheresultsof riskiden-
tification, riskassessment, analysisof riskfactors, and
theactual conditionsof Shi tunnel. Inthemeantime,
emergency plans against the possible risks are also
put forward. Therisk management should beimple-
mentedthroughout thelifeof theproject andkeepall
risksunder control.
6 CONCLUSIONS
The mountainous highway construction projects are
generally very complex, costly andtime-consuming.
Inevitably, therearealot of potential riskswhichmay
hindertheprojectdevelopmentandoftenresultinpoor
performance with increasing costs and time delays.
Most peopleagreethat therisk management is very
useful inprojectmanagementof complexsystems, but
fewpeopleanalyzetherisksinhighway construction
practiceother thanbyusingintuitionandexperience.
The major factors that limit the implementation of
risk management maybe the lack of risk awareness
and thelack of accepted risk assessment method. In
thispaper, asystematicriskmanagementframeworkis
proposedandpracticedinShui-mahighwayconstruc-
tion project. Thepracticeproved that therisk evalu-
ation results can help thedecision maker with more
confidence. Intherisk management process, therisk
identification,riskevaluation,riskmonitoringandrisk
databasearekey steps as well as project experience.
The risk identification is the most important phase.
This paper proposes aFTA based synthetical identi-
fication method which has been used in more than
10roadtunnels. Quantitativeor semi-quantitativerisk
analysis method for complex mountainous highway
constructionproject shouldbedevelopedinfuture.
35

I
I

M
i
d
d
l
e

r
i
s
k
,

a
c
c
e
p
t
a
b
l
e
I
I
I
H
i
g
h

r
i
s
k
,

u
n
w
a
n
t
e
d

I
V

E
x
t
r
e
m
e
l
y

h
i
g
h

r
i
s
k
,

u
n
a
c
c
e
p
t
a
b
l
e

S
h
u
i
k
o
u

F
a
u
l
t

P
o
r
t
a
l

C
o
l
l
a
p
s
e

W
a
t
e
r

b
u
r
s
t
i
n
g

D
e
f
o
r
m
a
t
i
o
n

F
a
u
l
t

e
f
f
e
c
t

a
r
e
a
,
f
r
a
g
m
e
n
t
i
z
e
d
D
e
p
t
h
:
1
0
1
~
1
7
6
m
,
,
b
l
a
s
t
o
p
s
a
m
m
i
t
e
,
,
[
B
Q
]
=
3
6
0

D
e
p
t
h
:
1
7
0
m
,
,
[
B
Q
]
=
3
4
0

D
e
p
t
h
:
1
5
4
~
3
5
0
m
,

r
o
c
k
:

b
l
a
s
t
o
p
s
a
m
m
i
t
e
,

l
e
s
s

i
n
t
e
g
r
i
t
y
,
[
B
Q
]
=
3
7
0

[
B
Q
]
=
2
5
5

[
B
Q
]
=
2
3
0

F
a
u
l
t

e
f
f
e
c
t

a
r
e
a
F
i
g
u
r
e
1
0
.
D
i
a
g
r
a
m
o
f
c
r
i
t
i
c
a
l
r
i
s
k
s
r
a
t
i
n
g
a
l
o
n
g
t
h
e
t
u
n
n
e
l
a
l
i
g
n
m
e
n
t
.
36
Figure11. Diagramof influencefactorstotunnel collapse
risk.
Figure12. Diagramof influencefactors to water bursting
andmudsurging.
Figure 13. Diagramof influence factors to tunnel portal
slopefailure.
Figure 14. Diagram of influence factors to tunnel large
deformation.
ACKNOWLEDGEMENTS
The authors wish to acknowledge the support
of National Natural Science Foundation of China
(No.40772179) and Western Science & Technol-
ogy Project of Ministry of Transport of China
(No.2006318799107).
REFERENCES
Ahmad, I. & Minkarah, I. 1988. Questionnaire survey on
biddinginconstruction. Journal of Management in Engi-
neering 3(4): 229243.
Akintoye, A. & MacLeod, M. 1997. Risk analysisandman-
agementinconstruction. International Journal of Project
Management 15(1): 3138.
APM Group. 1997. Project risk analysis and manage-
ment. http://www.eurolog.co.uk/apmrisksig/publications/
minipram.pdf.
Bao, H.L. & Huang, H.W. 2008. Risk assessment for the
safegradeof deepexcavation. InNgC.W.W. et al. (eds),
Geotechnical Aspects of Underground Construction in
Soft Ground: 507512. London: Taylaor & Francis.
Carr, V. & Tah, J. 2001. A fuzzy approach to construction
project risk assessment andanalysisconstructionproject
riskmanagement system. Advances in Engineering Soft-
ware 32: 847857.
Chapman, C. 1997. Project risk analysis and management-
PRAM the generic process. International Journal of
Project Management 15(5): 273281.
Chapman, R.J. 2001. The controlling influences on effec-
tive risk identification and assessment for construction
design management. International Journal of Project
Management 19: 147160.
Eskesen, S.D. et al. 2004. Guidelines for tunnelling risk
management:International TunnellingAssociation,Work-
ing Group No. 2. Tunnelling and Underground Space
Technology 19(3): 217237.
Floricel, S. & Miller, R. 2001. Strategizing for anticipated
risks and turbulencein largescaleengineering projects.
International Journal of Project Management 19:
445455.
Huang, H.W. et al. 2006. Risk analysisof buildingstructure
dueto shieldtunnelinginurbanarea. InZhuH.H. et al.
(eds), Underground construction and ground movement;
Proc. of sessions of Geoshanghai, Shanghai, 24 June
2006. NewYork:ASCE.
Lyons, T. & Skitmore, M. 2004. Project risk management
in the Queensland engineering construction industry: a
survey. International Journal of Project Management 22:
5161.
MOHURD. 2007. Guideline of Risk Management for Con-
struction of Subway and Underground Works. Beijing:
ChinaArchitecture& Building.
MTPRC. 2004. Code for design of road tunnel. Chinese
Standard: JTG D70-2004: 6265.
Pine, R.J. & Roberds, W.J. 2005. A risk-based approach
for thedesign of rock slopes subject to multiplefailure
modesillustratedbyacasestudyinHongKong.Interna-
tional Journal of Rock Mechanics & Mining Sciences 42:
261275.
Roger, F. & George, N. 1993. Risk Management and Con-
struction. London: Blackwell scienceLtd.
TheInternational TunnellingInsuranceGroup(ITIG). 2006.
A codeof practicefor risk management of tunnel works.
37
http://www.munichre.com/publications/tunnel_code_of_
practice_en.pdf
Wideman, R.M. 1986. Risk management. Project Manage-
ment Journal 17(4): 2026.
Yang, Z.F. et al. 2006. Research on the Geo-hazards of
Sichuan-Tibet Road and its Prevent and Control. Beijng:
SciencePressof China.
Yao, C.P. & Huang, H.W. 2008. Riskassessment onenviron-
mental impactinXizangRoadtunnel. InNgC.W.W. etal.
(eds), Geotechnical Aspects of Underground Construction
in Soft Ground:601606. London: Taylaor & Francis.
ZayedT. et al. 2008. Assessing risk and uncertainty inher-
entinChinesehighwayprojectsusingAHP. International
Journal of Project Management 26: 408419.
38
Recent revisionof J apaneseTechnical Standardfor Port andHarbor
Facilitiesbasedonaperformancebaseddesignconcept
T. Nagao
National Institute for Land and Infrastructure Management, Yokosuka, Japan
Y. Watabe&Y. Kikuchi
Port &Airport Research Institute, Yokosuka, Japan
Y. Honjo
Gifu University, Gifu, Japan
ABSTRACT: Thepurposeof this paper is to introducetherevisionof theTechnical Standards for Port and
Harbor Facilities (TSPHF) which was recently revised inApril 2007. It is thought that theTSPHF is oneof
thefirst casesof arevisionof adesigncodebasedonaperformancebaseddesign/specificationconcept. First,
the reason why a performance based design concept was introduced to theTSPHF is explained. Then, the
philosophyof providingaperformanceconcept isexplained. ThestandardverificationprocedureintheTSPHF
guidelinesisexplainedusinganexample. Thepolicyfor determiningthegeotechnical parametersusedfor the
performancebased design concept is introduced. Finally, theadequateness surveillancesystemintroduced is
explained. Thiskindof organizationisinevitablyrequiredfor thenewdesignsysteminorder toachievehigher
levelsof transparencyandfairness.
1 INTRODUCTION
The J apanese government published the first guide-
lines for port and harbor facilities in 1930. These
weresomething likeacollection of design casehis-
tories. Engineers at that timedesignedport facilities
by themselves withtheaidof suchdesignexamples.
In 1967, the Port and Harbor Bureau published the
StandardsforPortandHarbourFacilityDesign.These
standardswerethebasisof thestandardsfor approxi-
mately40years. However, atthattime, suchstandards
had no concretelegal background. In 1973, thePort
andHarbor Lawwasrevisedtoprovidealegal back-
ground for these standards. In 1979, standards and
commentaries wererevised to suit thelaw. ThePort
and Harbour Bureau then revised them twice dur-
ing the period up until 1999. In these revisions, the
concept of the standards was the same as those in
1967.In2007,newtechnical standardswerepresented.
Theconcept of thesestandardswasdifferent fromthe
former standards. These standards were formulated
in order to coincidewith theWTO/TBT agreement.
This paper presents thefeatures of thenewJ apanese
Technical Standard for Port and Harbor Facilities
(TSPHF).
2 JAPANESE GOVERNMENT POLICY ON
TECHNICAL STANDARDSAND
ACCREDITATIONS
Since1995(theyear inwhichtheWTO/TBT agree-
ment cameintoeffect.), theJ apanesegovernment has
adoptedapolicyof deregulationwithregardtoavari-
ety of laws and rules related to economic activities
andtrade. InMarch1998, theThree-Year Programfor
Promoting Deregulation was determined as a result
of a cabinet decision, and the following tasks were
delineated:
1 All economic regulations should beeliminated in
principle, and social regulations should be mini-
mized.
2 Rationalization of regulation methods. For exam-
ple, testingcanbeoutsourcedtotheprivatesector.
3 Simplificationandclarificationof thecontents of
theregulations.
4 International harmonizationof theregulations.
5 Speedingupof regulationrelatedprocedures.
6 Transparencyinregulationrelatedprocedures.
Followingtheaboveplan, theThree-Year Planfor
thePromotionof RegulatoryReformwasdetermined
39
Table1. Revisionof Port andHarbor Law.
Article56Item2-2
(Beforerevision)
Thoseport andharbor facilities, suchasnavigationchannelsandbasins, protectivefacilitiesfor harbors,
andmooringfacilities, shouldcomplywiththelawthat specifiessuchmattersif suchalawexists. Inaddition, their
construction, improvement andmaintenanceshouldcomplywiththeTechnical Standardsfor Port andHarbor Facilities
that havebeenspecifiedasaministerial ordinancebytheMinistryof Land, InfrastructureandTransportation.
(After revision)
Thoseport andharbor facilities, suchasnavigationchannelsandbasins, protectivefacilitiesfor harbors, andmooring
facilities(termed facilities covered by the TSPHF), shouldcomplywiththelawthat specifiessuchmatters
if suchalawexists. Inaddition, construction, improvement andmaintenanceconcerningtheperformance of facilities
covered by the TSPHF should comply with theTechnical Standardsfor Port andHarbor Facilitiesthat havebeen
specifiedasaministerial ordinancebytheMinistryof Land, InfrastructureandTransportation.
asaresult of acabinet decisioninMarch, 2001. This
planconsistedof theobjectivesshownbelow:
1 Realizationof sustainableeconomic development
throughthepromotionof economicactivities.
2 Realizationof atransparent, fair andreliableeco-
nomicsociety.
3 Securediversifiedalternativelifestyles.
4 Realizationof aneconomic society that isopento
theworld.
Inordertorealizesuchobjectives, thepromotionof
essential andactivederegulationsinvariousadminis-
trativeserviceswasplanned. Inthefieldof standards
andaccreditations, thefollowingbasic policies were
implemented.
Essential reviewsof standardsandaccreditationsin
order tocheck thenecessity of theinvolvement of
thegovernment.
In cases where administrative involvement was
still required, administrativeroles should bemin-
imized, andself-accreditationor self-maintenance
of standardsandaccreditationsbytheprivatesector
shouldbepromoted.
Theinternational harmonizationof standards, per-
formancebasedspecificationsandtheelimination
of multiple examination procedures in accredita-
tionprocessesshouldbepromoted.
The third itemhad a very strong impact on the
revision of design standards and codes for civil
structures. The Ministry of Land, Infrastructure and
Transportation(MLIT) startedaprogramentitledthe
Restructuringof PublicWorksCostsinMarch, 2003,
andwhichincludesthetasksshownbelow:
Revisionof commonspecificationsforcivil works.
Reviewof theHighwayBridgeSpecifications.
Revision of theTechnical Standards for Port and
Harbor Facilities (TSPHF) to performancebased.
The revision of the TSPHF was started around
this time with the goal of achieving harmoniza-
tion between the standards and the international
agreement.
3 REVISIONOF THETSPHF
Based on the background explained in the previous
section, the Port and Harbor Law was revised in
parliament, whichmadeaproclamationinSeptember
2006, andwasimplementedon1April 2007.Theitem
that influenced therevision of thestandards, Article
56Item2-2, isshowninTable1. Intherevision, rather
than prescribing thespecifications of design details,
theperformancesof facilitiesareregulated.
Based on the revised Port and Harbor Law, the
TSPHF wasfullyrevised. Themainpointsof therevi-
sionsummarizedherearefromtwo aspects, namely,
the system for the performance based specifica-
tions, and the performance verification. Previously
establishedcomprehensivedesigncodes(MLT, 2002;
J SCE, 2003; J GS, 2004) providedfoundationsfor the
revisionof thesetechnical standards.
3.1 Performance based specifications system
Thebasic systemfor theTSPHF is that therequired
performances of the structures are given as manda-
toryitemsinthreelevels, i.e. objectives, performance
requirements and performance criteria, whereas the
performance verification methods are not manda-
tory but are given in the annex or in the reference
documentsassomeof thepossiblemethods(Figure1).
Theobjectivesstatethenecessityof thestructures,
whereastheperformancerequirementsstatethefunc-
tionsof thestructuresthat needtobeimplementedin
thestructuresinorder tosatisfytheobjectivesinplain
languagefromtheview point of accountability. The
performancecriteriarestatetheperformancerequire-
ments fromtechnical viewpoints, thus making each
performancerequirement verifiable.
The performance requirements are classified into
basic requirements andother requirements (Table2),
where the former specifies structural performances
against various actions and their combinations, and
the latter specifies structural dimensional require-
mentsarisingfromusageandconveniences.Thebasic
requirements are further classified into serviceabil-
ity, reparabilityandsafetyrequirementsasdefinedin
Table2.
Thebasic requirements should becombined with
theactions consideredinthedesign, whicharesum-
marized in Table 3. The combinations of perfor-
mance requirements and actions are termed design
situations, where performance verification of the
structure should be carry out for each design situ-
ation. The actions are classified into accidental and
40
Figure1. Performancebasedspecificationsfor theTSPHF.
Table2. PerformancerequirementsintheTSPHF.
Classification Definition
Basic Performanceof structural response
requirement (deformation, stress, etc.) against
actions.
Serviceability Thefunctionof thefacilitywouldbe
recoveredwithminor repairs.
Reparability Thefunctionof thefacilitywouldbe
recoveredinarelativelyshort period
of timeafter somerepairs.
Safety Significant damagewouldtakeplace.
However, thedamagewouldnot
causeanylossof lifeor serious
economicdamagetothehinterland.
Other Performancerequirementsfor
requirements structural dimensionsconcerning
usageandconvenienceof thefacilities.
Table3. Summaryof thebasicrequirements.
Design Performance
situation Definition Requirements
Persistent Permanent actions(self Serviceability
Situation weight, earthpressures)
aremajor actions.
Transient Variableactions(waves, Serviceability
Situation level 1earthquakes) are
major actions.
Accidental Accidental actions Serviceability
Situation (tsunamis, level 2 Reparability
earthquakes) aremajor Safety
actions.
permanent/variable actions employing an annual
occurrence rate of approximately 0.01 (i.e. a return
periodof 100years) asathresholdvalue. Forbothper-
manent andtransient designsituations, serviceability
needstobesatisfied, whereasinaccidental situations,
either of thethreeperformancerequirementsneedsto
be satisfied depending on the importance and func-
tions of the structure under design. This concept is
furtherillustratedinFigure2. Itshouldbenoticedthat
theperformanceof astructuremay not beverifiable
inaccidental situationsinsomecases.
Figure 2. Classifications of performances, actions and
frequency.
Theobjectives and theperformancerequirements
areprescribedintheMLIT ministerial ordinancepart
of theTSPHF, whereas the performance criteria are
specifiedintheMLIT declarationpart of theTSPHF
that defines the details of the TSPHF. In this way,
the hierarchy of the performance specifications is
maintained.
Table4shows anexampleof theprovisions inthe
newTSPHF. Thisexampleisfor aprotectivefacility.
A breakwater is a representative protective facility.
Figure 3 shows the cross section of a caisson-type
breakwater.Table5showstheprovisionsfor breakwa-
ters intheformer TSPHF. InthenewTSPHF, objec-
tives, performance requirements, and performance
criteriaarewrittenclearlyinaccordancewiththehier-
archy shown in Figure 1. However, these were not
clearlydescribedintheformerTSPHF. Withregardto
verification, thiswasmandatoryintheformerTSPHF,
but is not mandatory in the new TSPHF. As per-
formance verification in accordance withTSPHF is
Approach B verification shown in Figure1, therec-
ommended verification method is presented in the
guidelines, but isnot mandatory.
3.2 Performance verifications
In order to harmonizetheTSPHF with international
standardssuchasISO2394andintroducenewlydevel-
oped verification methods using more sophisticated
designcalculationmethods suchas seismic response
41
Table4. Exampleof provisionsinthenewTSPHF.
Mandatory
Level Definition situation Examplefor breakwaters
Objectives Thereasonwhythe Mandatory Thecalmnessof navigationchannelsandbasinsshouldbe
facilityisneeded (Port and maintainedinorder tosafelynavigateandmoor ships, to
Harbor Law) handlecargosmoothly, andtosafelymaintainbuildings
andother facilitieslocatedinports. LawArticle14
Performance Levelsof performance Mandatory Damagesduetotheactionsof self weights, waves,
requirements facilitiesarerequired (Port and andlevel 1earthquakesshouldnot affect theobjectives
topossess Harbor Law) of thebreakwater anditscontinuoususage.
LawArticle 14 -Serviceabilityrequirements-
Performance Concretecriteriawhich Mandatory -NotificationArticle35-
criteria represent performance (Notification) 1st Danger of thepossibilityof slidingfailuresof theground
requirements under persistent situationsinwhichthemainactionisself
weight shouldbelower thanthelimit level.
2ndDanger of thepossibilityof slidingandrotation
failuresof gravitystructuresandof failuresof theground
duetoinsufficient bearingcapacityunder variablesituations
inwhichthemainactionsarewavesor level 1earthquakes
shouldbelower thanthelimit level.
Performance Performancesshould Not Mandatory (Guidelinespresent thestandardprocedurefor performance
verification beverifiedusing (Guidelinesare verificationsfor referencepurposes)
engineeringprocedures presentedas
references)
Figure3. Typical caissontypebreakwater.
analyses, thefollowingverificationmethodsareintro-
ducedintherevisedTSPHF (Table6):
Reliability based design (RBD) methods, mainly
level I partial factor approach.
Numerical methods (NM) capable of evaluating
structural responseproperties.
Model tests.
Methodbasedonpast experiences.
Theperformanceverificationimpliesdesignproce-
durestoverifyfor thestructuresinorder tosatisfythe
specifiedperformancerequirementsand/ortheperfor-
mancecriteria. Inprincipal, therevisedTSHPF does
not specifyanyconcreteallowablevaluesfor strength
nordisplacement.Inordertoperformthetolerablefail-
ureprobability, safetyindicesandcharacteristicvalues
for basicvariablesinthedesignareintroduced. These
areall decidedbydesigners.
However, itisconsiderednecessarytoprovidestan-
dardverificationmethodstogether withtheminimum
tolerablelimit values for thedesign in order for the
users of therevisedTSPHF to understand theinten-
tionsof thecodewriters. For thispurpose, itisjudged
appropriatetoprovidetheinformationintheformof
anannexandsupportingdocuments.
Figure4shows theverificationprocedures for the
slidingsafetyof acaissontypebreakwatershowninthe
former andnewTSPHF guidelines. For thedesignof
abreakwater inapersistent situationwherethemajor
action is waves, it is recommended to employ RBD
basedonforceequilibrium. Level I RBD is adopted,
andrecommendedpartial factorsareprovidedintables
in theannex. In thecaseof acaisson-typebreakwa-
ter onarubbleembankment, serviceabilityisrequired
for awaveforcewitha50year returnperiod(avari-
able action), and partial factors are provided which
aredeterminedbasedontheannual failureprobability
of 0.01or belowfor sliding, overturningandbearing
capacityfailure.
3.3 Geotechnical parameters
The soil parameters of the ground and the qual-
ity parameters of industrial products are completely
different in terms of their treatments. Statistical
treatments suitable for geotechnical parameters are
strongly required in consideration of non-uniform
sedimentary structures, investigation errors, testing
errors, and limited numbers of data entries, etc. In
thenewTSPHF, asimplifiedandreasonablemethod,
which pursues practical usability by simplifyingsta-
tistical treatments, to determine soil characteristic
values was introduced. Details have been published
byWatabeetal. (2009a; 2009b).Themethodisbriefly
introducedinthefollowingsections.
42
Table5. Exampleof provisionsintheformerTSPHF.
Contents ProvisionsinformerTSPHF
Objectivesandperformance Function Protectivefacilitiesfor harborsshouldmaintaintheir functionsunder
requirements all natural situationssuchasgeographical, meteorological, andmarine
phenomena, etc. (LawArticle7)
Safety Protectivefacilitiesshouldbesafeagainst self weight, water pressure,
waveforces, earthpressure, anearthquakeforces, etc. (LawArticle7)
Performanceverifications Calculationof Thewaveforceactingonastructureshall bedeterminedusing
(alsodescribedin forces appropriatehydraulicmodel experimentsor designmethodsin
notifications) thefollowingprocedure. (NotificationArticle5)
Safetyverification Examinationsof thesafetyof themembersof reinforcedconcrete
of members structuresshall beconductedasstandardusingthelimit state
designmethod. (NotificationArticle34)
Stabilitycheck Examinationsof thestabilityof upright sectionsof gravitytype
breakwatersshall bebasedonthedesignproceduresusingthesafety
factorsagainst failures. (NotificationArticle48)
Table 6. Summary of basic performance verification
methods.
Recommended
Design performance
situation Major actions verificationprocedures
Persistent and Self weight, RBD
transient earthandwater
situations pressures, live
loads, waves,
wind, ships, etc.
Level 1 Non-linear response
earthquakes analysestakinginto
considerationsoil
structureinteractions
RBD
Pseudo-staticprocedures
(e.g. seismiccoefficient
method)
Accidental Level 1 Numerical procedureto
situations earthquakes, evaluatedisplacements
tsunamis, ship anddamageextents
collisions, etc.
For example, it iswell knownthat undrainedshear
strengths obtained through unconfined compression
testsaremorevariablethanthoseobtainedasaresult
of triaxial tests, indicatingthatthereliabilityof thefor-
mer ismuchsmaller thanthat of thelatter (Tsuchida,
2000; Watabe and Tsuchida, 2001). In each design
procedure, however, it is difficult to takeaccount of
datavariationswhicharedependent ontestingmeth-
ods. Consequently, themethodinthenewTSPHF has
adopted a concept in which the characteristic value
is correctedincorrespondencewiththereliability of
thetestingmethod. Thisconcept aimstousethepar-
tial safety factor, whichis independent of thetesting
method. Therefore, theconcept inthecaseof alarge
number of dataentries is slightly different fromthat
of thegeneral designcode, inwhichthecharacteristic
value is generally the expected value of the derived
values.
3.3.1 Principle of soil parameter determination
J GS4001: Principles for foundationdesigngrounded
in the performance-based design concept were pub-
lishedin2004by theJ apaneseGeotechnical Society.
Theseareguidelines for determiningsoil parameters
forperformance-basedreliabilitydesigninJ apan. Fig-
ure5showstheflowchartintheTSPHF for determin-
ingsoil parametersusedforperformanceverifications.
Thisflowchart wasmodifiedfor thenewTSPHF, but
reflectsthepurposeof J GS4001.
Themeasuredvalueis thevaluedirectly recorded
inafieldor laboratory test. Thederivedvalueis the
valueobtainedby usingtherelationshipbetweenthe
measuredvalueandthesoil parameter.
Thecharacteristic valueistherepresentativevalue
obtainedbymodelingthedepthprofileof thedataby
takingintoaccount thevariationof theestimatedval-
ues. The value must correspond to the critical state
for the performance considered in the design. Tak-
ing account of the application range of either the
verification equation or the prediction equation, the
characteristicvalueisconvertedintothedesignvalue
by multiplyingwithanappropriatepartial safety fac-
tor.Thepartial safetyfactorsforeachfacilityarelisted
in the design code corresponding to both the varia-
tionandsensitivityof thesoil parameter inthedesign
verification.
Thecharacteristicvalueof thegeotechnical param-
etersinEurocode7isalsodefinedfollowingthesame
concept asJ GS4001. Inthecaseof industryproducts,
thecharacteristicvalueisgenerallydefinedas5%frac-
tile corresponding to Equation 1, e.g. in Eurocode0
(EN1990, 2002).
wherej(x) istheaverageof x, and(x) isthestandard
deviation of parameter x. This kind of characteristic
43
Figure4. Exampleof theverificationof persistent andtransient situationsfor caisson-typebreakwaters.
Figure5. Flowchart for thedeterminationof soil parame-
tersinTSPHF.
valueisapplicabletostructural materials; however, it
isnot applicabletosoil parametersbecausetheyvary
significantly.
If weconsidergroundfailure, forexample, wehave
totreatthewholegroundfailure, notthefailureineach
element. Fromthis background, Eurocode7 adopted
thevaluecorrespondingtothe95%confidencelevel
insteadof the5%fractile(Orr, 2006). InJ GS4001, the
characteristic valueis described in thesamemanner
as Eurocode7, but the confidence level is not fixed
at 95%.
Ovesen (1995) proposed a simple equation
expressedasEquation(2) inorder toobtainthelower
limit of the95%confidencelevel.
wheren isthenumberof dataentries.Schneider(1997)
proposedamoresimplifiedequationfor n =11.
Whenn islarger than12, Schneidersequationgivesa
moreconservativevaluethanOvesensequation.
ThenewTSPHF adoptedamorepractical method,
which uses the outline of both Schneiders and
Ovesens equations and is partly consistent with
J GS4001, in order to determine the characteristic
value.
Because an engineer who performs geotechnical
investigations andanengineer who performs facility
design are usually different engineers, the engineer
who performs facility design cannot determine an
appropriate partial safety factor taking into account
datavariationsfor soil parametersinassociationwith
theinvestigation and testing methods. In addition, it
isvirtuallyimpossibletodetermineeachpartial safety
factortakingintoaccountdatavariationsderivedfrom
theheterogeneityof thegrounditself reflectingthesoil
locality. Therefore, it isideal that thereliabilityof the
soil parametersisalwaysguaranteedtoremainat the
samelevel whenthegeotechnical informationistrans-
mittedfromthegeotechnical engineer to thefacility
44
designengineer. Consequently, thepartial safety fac-
tor asageneral valuelistedinthedesigncodecanbe
used.
3.3.2 Characteristic value determination with
correction factors
InthenewTSPHF guidelines, thepartial safetyfactor
isdeterminedfromtheempirical calibrationtaking
intoconsiderationthedatavariation. Becausethedata
variationisagivenconditioninmostcases,anyefforts,
e.g. examinationsdesignedtoobtainthemost appro-
priate depth profile, adopting a reliable laboratory
testing method, brushing up skills for siteinvestiga-
tionsandlaboratory testing, designedtodecreasethe
datavariationarenotrewarded. Consequently, wepre-
fertouseconventional investigationmethods.Thenew
TSPHF guidelinesaimtosolvethesekindsof issues.
InJ GS4001, becausetheconfidenceinterval range
narrowswithincreasesinthenumbersof dataentries,
the characteristic value is coincident with the mean
valuewhenthenumber of dataentriesbecomeslarge.
InthenewTSPHFguidelines, becausethecharacteris-
ticvalueisdeterminedaccordingtothedatavariation,
effortstoreducethedatavariationarerewardedinthe
design.
Because the derived value is influenced by the
sampling method, laboratory testing method, sound-
ing method, and empirical/theoretical equation, etc.,
the design values must reflect these influences. For
example, it has been well known that the reliability
of the undrained shear strength obtained using the
unconfinedcompressiontest is muchlower thanthat
obtained using therecompression triaxial test. How-
ever, it isvery difficult totakeaccount of thisfact in
design.
ThemethodinthenewTSPHF guidelinesadopted
the concept in which the characteristic value is cor-
rectedaccordingto thereliability level of thetesting
method. Thecoefficient of variation (COV) is intro-
duced to represent the data variation. To reflect the
datareliabilityinthecharacteristicvalue,theestimated
valueiscorrectedaccordingtotheCOV.Consequently,
wecanestablishadesigncodewithacommonpartial
safety factor by using thecharacteristic valuedeter-
mined using this method, even though the derived
value of the soil parameter has been obtained from
adifferent soil test.
A larger number of dataentries is moredesirable
inorder toreducetheCOV. However, 10dataentries
aresufficient inpractice, becausethenumber of data
entriesisgenerallyverylimited. Infact, inmostcases
theCOVconvergesonacertainvaluewhenthenumber
of dataentriesismorethan10.
ItisknownthattheCOVforderivedvaluesobtained
by highly skilledtechniciansislessthan0.1(Watabe
et al., 2007). Inother words, thevariationat thislevel
is inevitable due to ground heterogeneity and labo-
ratory testing errors. Ground heterogeneity, sample
disturbances, inappropriatesoil tests, andinappropri-
atemodelingof depthprofiles,etc.,resultinlargeCOV
values. Insuchcases, itisreasonabletoconservatively
Table7. Valuesfor correctionfactor b
1
.
Correctionfactor b
1
Coefficient of Parameter for Parameter for
variationCOV safeside unsafeside
COV-0.1 1.00 1.00
0.1-COV-0.15 0.95 1.05
0.15-COV-0.25 0.90 1.10
0.25-COV-0.4 0.85 1.15
0.4-COV-0.6 0.75 1.25
0.6-COV Reexaminationof thedata/
Reexaminationof thesoil test
determinecharacteristicvaluesbytakinguncertainties
intoaccount.
Inordertocalculatethecharacteristicvaluea
k
from
theestimatedvaluea
, thecorrectionfactorb
1
isintro-
ducedasafunctionof theCOV, thena
k
isdefinedas
Equation(4).
Whenthesoil parameter a contributestoeither the
resistancemoment inthesafety verification(e.g. the
shear strength in thestability analysis) or thesafety
marginintheprediction(e.g. theconsolidationyield
stressp
c
; thecoefficientof consolidationc
v
inthecon-
solidationcalculation), thecorrectionfactorisdefined
asEquation(5).
Ontheother hand, whenit contributestoeither the
slidingmomentinthesafetyverification(e.g. theunit
weight of the earth fill in the stability analysis) or
the unsafe factor in the prediction (the compression
indexC
c
; coefficientof volumetriccompressionm
v
in
theconsolidationcalculation), thecorrectionfactor is
definedasEquation(6).
In thesedefinitions, thecharacteristic values cor-
respond to either 30% or 70% of the fractile value.
Becausetheaimof thismethodissimplification, the
valueslistedinTable7aretobeusedinsteadof thecor-
rectionfactorswithdetailedfractions. WhentheCOV
islarger than0.6, itisjudgedthatthereliabilityof the
soil parameter is too lowfor thedesign. Inthis case,
thetest results arereexamined; i.e. thedepthprofile
is remodeledif necessary. Insomecases, theground
investigationmaybeperformedagain.
InJ GS4001or Eurocode7, thecharacteristicvalue
is defined as the upper/lower boundary of a certain
confidence level (95% in most cases) as mentioned
above. ThenewTSPHF guidelinesusingasimplified
45
method without real statistical treatment are partly
consistent with J GS4001. Thecharacteristic valueis
definedas30%or 70%of fractilevalueswhichcorre-
spondtoa95%confidencelevel whenthenumber of
dataentriesnis10andthedatavariationCOV is0.1.
If thenumber of data entries is not sufficient for
statistical treatment, another correction factor b
2
is
introducedtocorrectb
1
.Then, thecharacteristicvalue
isexpressedasEquation(7).
Approximately 10 or more data entries in the depth
profilecanbethought tobesufficient toreliablycal-
culate COV. In cases with less than 10 data entries,
whenthesoil parametercontributestoeitherresistance
moment inthestability verificationor safety margin
in theprediction, thecorrection factor is defined as
Equation(8).
On the other hand, when it contributes to either
theslidingmoment inthestability verificationor the
unsafefactor intheprediction, thecorrectionfactor is
definedasEquation(9).
Here, b
2
for cases with only onedataentry is set
at 0.5or 1.5, andthereliability isassumedtorapidly
increasewiththenumberof dataentries. Inthisregard,
however, thecorrectionfactor b
1
cannot beobtained
inthecaseof n =1, becausetheCOV cannot becal-
culated. Thisindicatesthatmorethantwodataentries
arerequiredinthismethod. InthenewTSPHF guide-
lines, thecorrectionfactor b
2
is introducedwhenthe
number of dataentries is less than10. However, this
number canbevariedbyeachdesignguideline.
Noteherethat b
1
=1andb
2
=1areusedfor soil
parametersthat contributeequivalentlytobothaction
andcounteraction.
3.4 Institution for surveillance of adequateness to
the TSPHF
ThenewTSPHF admits designs which arenot com-
pletelyverifiedbytheguidelines.Therearetwoproce-
duresfor designverification. Oneprocedureistouse
ApproachA inFig. 1, whichisaverificationapproach
withdesignersoriginal considerationcertificatingthe
satisfaction of the requirement of performance cri-
teria provided in TSPHF. The other procedure is to
use Approach B, which is a verification approach
in accordance with recommended procedures in the
guidelines. EveninApproachB, designers canmake
their owndecisionsintheverificationprocedures.
For verificationresultsusingApproachA, itisnec-
essary tocertify that theverificationresultsconform
Figure6. Designverificationsurveillancesystem.
to the TSPHF. On the other hand, for verification
results obtainedusingApproachB, it is necessary to
certify that theverificationresultsareevaluatedade-
quately.Technical standardsorverificationprocedures
inpublic constructionworks inJ apanareauthorized
by operatingbodiessuchasrailways, roads, or ports.
In the existing system, evaluations of design results
areperformedby suchoperatingbodies. Thissystem
seemstobelesstransparent or fairnessviewingfrom
outsideof concernedgroup.Inordertoavoidthisprob-
lem, a third party for certifying design verification
resultsisrequired.
This surveillance organization is required to sat-
isfytheitemsrequiredfor anadequatenessevaluation
organization provided by ISO/IEC guide#65. When
adesignverificationsurveillanceorganizationissues
acertification for theadequateness of adesign, it is
responsiblefor conductingasurveydesignedtoshow
that thedesign results arefreefromfaults. It is also
required to prepare a discharge of liability for any
damagesarisingfromamisevaluation.
To perform this certification, the TSPHF has a
rulethat designresultsfor important facilitieswill be
checkedbyagovernmental institutionor athirdparty
institution which is authorized by the government
(Figure6).
4 CONCLUSION
Thepurposeof thispaperistointroducetherevisionof
theTechnical Standardsfor Port andHarbor Facilities
(TSPHF) whichwasrecentlyrevisedinApril 2007. It
isthought that theTSPHF isoneof thefirst casesof
arevision of adesign codebased on aperformance
baseddesign/specificationconcept.
First, thereasonwhy aperformancebaseddesign
concept wasintroducedtotheTSPHF isexplainedin
thispaper. Then, thephilosophyof providingaperfor-
manceconceptisexplained. Thestandardverification
procedureintheTSPHF guidelinesisexplainedusing
anexampleconsistingof theslidingfailureof acaisson
typebreakwater.
Thepolicyfordeterminingthegeotechnical param-
eters usedfor theperformancebaseddesignconcept
46
isintroduced. TheTSPHF guidelinesintroduceasim-
plified determination method allowing for ease of
use for the practitioner, and with this determination
procedureinnovativegeotechnical investigationmeth-
ods and laboratory testing methods can be easily
introduced.
Finally, theadequatenesssurveillancesystemintro-
duced is explained. This kind of organization is
inevitablyrequiredfor thenewdesignsysteminorder
toachievehigher levelsof transparencyandfairness.
Engineers in the field are experiencing a certain
degreeof confusionwithregardtoapplicationof the
newTSPHF. Misunderstandingsof theconcept isone
of thereasons for this confusion, withtheresult that
theintroductionof abrandnewconcept isinevitable.
Ascodewriterswealsointendtoworkhardtoimprove
thedesigncode.
REFERENCES
EN1990: 2002: Eurocode0, Basisof structural design.
EN1997-1: 2004: Eurocode7, Geotechnical designPart 1:
General rules.
J GS (2004), J GS-4001-2004: Principles for Foundation
DesignsGroundedonaPerformance-basedDesignCon-
cept (nicknameGeocode21).
J SCE (2003), Principles, Guidelines andTerminologies for
drafting design codes founded on performance based
design concept (nickname code PLATFORM ver.1),
J apanSocietyof Civil Engineers.
MLIT (2002); Basis for design of civil and building struc-
tures, Ministryof Land, InfrastructureandTransportation.
Orr, T.L.L. (2006): Development and implementation of
Eurocode 7, Proceedings of the International Sympo-
sium on New Generation Design Codes for Geotechnical
Engineering Practice Taipei 2006, CDROM, 118.
Ovesen, N.K. (1995): Eurocode 7 for geotechnical design,
Proceedings Bengt B. Broms Symposiumon Geotechnical
Engineering, Singapore, 333360.
Schneider, H.R. (1997): Definition and determination of
characteristic soil properties, Proceedings 12th Interna-
tional Conference on Soil Mechanics and Geotechnical
Engineering, Hamburg, Vol. 4, 22712274.
The J apan Port and Harbour Association (2007): Techni-
cal standards, and commentaries for port and harbour
facilities. (inJ apanese)
Tsuchida, T. (2000): Evaluationof undrainedshear strength
of softclaywithconsiderationof samplequality, Soilsand
Foundations, 40(3), 2942.
Watabe, Y. andTsuchida, T. (2001): Comparativestudy on
undrained shear strength of OsakaBay Pleistoceneclay
determinedbyseveral kindsof laboratorytests, Soils and
Watabe, Y., Shiraishi, Y., Murakami, T. and Tanaka, M.
(2007): Variability of physical and consolidation test
resultsfor relativelyuniformclaysamplesretrievedfrom
OsakaBay, Soils and Foundations, 47(4), 701716.
Watabe,Y., Tanaka, M. andKikuchi,Y. (2009a): Soil param-
eters inthenewdesigncodefor port facilities inJ apan,
Proceedings of the International Foundation Congress &
Equipment Expo09: IFCEE09. (inprint)
Watabe, Y., Tanaka, M. and Kikuchi, Y. (2009b): Practical
determinationmethodfor soil parametersadoptedinthe
newperformancebaseddesigncodefor port facilitiesin
J apan, Soils and Foundations. (inprint)
47
Special lecture
InteractionbetweenEurocode7 Geotechnical designandEurocode8
Designfor earthquakeresistanceof geotechnical structures
P.S. ScoePinto
Faculty of Engineering, University of Coimbra, National Laboratory of Civil Engineering (LNEC),
Lisbon, Portugal
1 INTRODUCTION
TheCommissionof theEuropeanCommunities(CEC)
initiated awork in 1975of establishing aset of har-
monisedtechnical rulesforthestructural andgeotech-
nical design of buildings and civil engineers works
basedonarticle95of theTreaty. Inafirststagewould
serveasalternativetothenational rulesappliedinthe
variousMemberStatesandinafinal stagewill replace
them.
From1975to1989theCommissionwiththehelp
of aSteeringCommitteewiththeRepresentatives of
Member StatesdevelopedtheEurocodesprogramme.
TheCommission, theMember statesof theEUand
EFTA decidedin1989basedonanagreementbetween
theCommissionandCEN totransfer thepreparation
andthepublicationof theEurocodestoCEN.
TheStructural Eurocodeprogrammecomprisesthe
followingstandards:
EN1990Eurocode Basisof design
EN1991Eurocode1 Actionsonstructures
EN 1992 Eurocode 2 Design of concrete
structures
EN1993Eurocode3 Designof steel structures
EN 1994Eurocode4 Designof compositesteel
andconcretestructures
EN1995Eurocode5 Designof timber structures
EN 1996 Eurocode 6 Design of masonry
structures
EN1997Eurocode7 Geotechnical design
EN 1998 Eurocode 8 Design of structures for
earthquakeresistance
EN 1999Eurocode9 Designof aluminiumalloy
structures
The work performed by the Commission of the
European Communities (CEC) in preparing the
Structural Eurocodes in order to establish a set
of harmonised technical rules is impressive. Never-
theless, due the preparation of these documents by
several experts, someprovisionsof EC8withthespe-
cial requirementsfor seismicgeotechnical designthat
deservemoreconsiderationwill bepresentedinorder
to clarify several questions that still remain without
answer.
Theactual tendencyistoprepareunifiedcodesfor
different regions but keeping the freedomfor each
country to choose the safety level defined in each
National Document of Application. Theglobal safety
of factor was substitutedby thepartial safety factors
appliedtoactionsandtothestrengthof materials.
This invited lecture summarises the main top-
ics covered by Eurocode 7 and the interplay with
Eurocode 8 and also identify some topics that need
further implementation.
In dealing with thesetopics weshould never for-
get the memorable lines of Lao- Tsze, Maxin 64
(550B.C.):
The journey of a thousand miles begins with
one step.
2 EUROCODE 7 GEOTECHNICAL DESIGN
2.1 Introduction
TheEurocode7 (EC7) Geotechnical Design gives
a general basis for the geotechnical aspects of the
designof buildings andcivil engineeringworks. The
link between the design requirements in Part 1 and
theresultsof laboratory testsandfieldinvestigations
runaccordingtostandards, codesandother accepted
documentsiscoveredbyPart 2
EN 1997 is concerned with the requirements for
strength, stability, serviceability and durability of
structures. Other requirements, e.g. concerning ther-
mal or soundinsulation, arenot considered.
2.2 EUROCODE 7 Geotechnical Design Part 1
Thefollowingsubjectsaredealt withinEN 1997-1-
Geotechnical design:
Section1: General
Section2: Basisof Geotechnical Design
Section3: Geotechnical Data
Section4: Supervisionof Construction, Monitoring
andMaintenance
Section5: Fill, Dewatering, GroundImprovement
andReinforcement
Section6: SpreadFoundations
Section7: PileFoundations
Section8: Anchorages
51
Section9: RetainingStructures
Section10: Hydraulicfailure
Section11: Overall stability
Section12: Embankments
2.2.1 Design requirements
Thefollowingfactorsshall beconsideredwhendeter-
miningthegeotechnical designrequirements:
Siteconditionswithrespect tooverall stabilityand
groundmovements;
Natureand sizeof thestructureand its elements,
including any special requirements such as the
designlife;
Conditionswithregardtoitssurroundings(neigh-
bouring structures, traffic, utilities, vegetation,
hazardouschemicals, etc.);
Groundconditions;
Groundwater conditions;
Regional seismicity;
Influence of the environment (hydrology, surface
water, subsidence, seasonal changesof temperature
andmoisture).
Eachgeotechnical designsituationshall beverified
that norelevant limit stateisexceeded.
Limitstatescanoccur either inthegroundor inthe
structureor by combinedfailureinthestructureand
theground.
Limitstatesshouldbeverifiedbyoneoracombina-
tionof thefollowingmethods: designby calculation,
designbyprescriptivemeasures, designbyloadstests
andexperimental modelsandobservational method.
To establish geotechnical design requirements,
three Geotechnical Categories, 1, 2 and 3 are intro-
duced.
Geotechnical Category 1 includes small and rela-
tivelysimplestructures.
Geotechnical Category 2 includes conventional
typesof structureandfoundationwithnoexceptional
riskor difficult soil or loadingconditions.
Geotechnical Category 3 includes: (i) very large
or unusual structures; (ii) structuresinvolvingabnor-
mal risks, or unusual or exceptionallydifficultground
or loading conditions; and (iii) structures in highly
seismicareas.
2.2.2 Geotechnical Design by calculation
Designbycalculationinvolves:
Actions, which may be either imposed loads or
imposeddisplacements, for examplefromground
movements;
Propertiesof soils, rocksandother materials;
Geometrical data;
Limiting values of deformations, crack widths,
vibrationsetc.
Calculationmodels.
Thecalculation model may consist of: (i) an ana-
lytical model; (ii) a semi-empirical model; (iii) or a
numerical model.
Where relevant, it shall be verified that the
followinglimit statesarenot exceeded:
Lossof equilibriumof thestructureor theground,
consideredasarigidbody, inwhichthestrengthsof
structural materialsandthegroundareinsignificant
inprovidingresistance(EQU);
Internal failure or excessive deformation of the
structureorstructural elements, includingfootings,
piles, basement walls, etc., in which the strength
of structural materials is significant in providing
resistance(STR);
Failureor excessivedeformationof theground, in
whichthestrengthof soil or rock is significant in
providingresistance(GEO);
Lossof equilibriumof thestructureor theground
duetoupliftbywater pressure(buoyancy) or other
vertical actions(UPL);
Hydraulic heave, internal erosion and piping
in the ground caused by hydraulic gradients
(HYD).
Theselectionof characteristicvaluesforgeotechni-
cal parametersshall bebasedonderivedvaluesresult-
ingfromlaboratoryandfieldtests, complementedby
well-establishedexperience.
Thecharacteristicvalueof ageotechnical parameter
shall beselected as acautious estimateof thevalue
affectingtheoccurrenceof thelimit state.
For limit state types STR and GEO in persistent
andtransient situations, threeDesignApproachesare
outlined. Theydiffer inthewaytheydistributepartial
factorsbetweenactions, theeffectsof actions, material
propertiesandresistances. Inpart, thisisduetodiffer-
ingapproachestothewayinwhichallowanceismade
foruncertaintiesinmodelingtheeffectsof actionsand
resistances.
InDesign Approach 1 partial factorsareappliedto
actions, ratherthantotheeffectsof actionsandground
parameters,
InDesign Approach 2 approach, partial factorsare
applied to actions or to theeffects of actions and to
groundresistances.
InDesign Approach 3 partial factorsareappliedto
actionsor theeffectsof actionsfromthestructureand
togroundstrengthparameters.
It shall beverified that alimit stateof ruptureor
excessivedeformationwill not occur.
It shall be verified serviceability limit states in
the ground or in a structural section, element or
connection.
2.2.3 Design by prescriptive measures
Indesignsituationswherecalculationmodelsarenot
available or not necessary, the exceedance of limit
states may be avoided by the use of prescriptive
measures. These involve conventional and generally
conservative rules in the design, and attention to
specificationandcontrol of materials, workmanship,
protectionandmaintenanceprocedures.
52
2.2.4 Design by load tests and experimental
models
Whentheresultsof loadtestsor testsonlargeor small
scalemodelsareusedtojustifyadesign, thefollowing
featuresshall beconsideredandallowedfor:
Differences inthegroundconditions betweenthe
test andtheactual construction;
Timeeffects, especiallyif thedurationof thetestis
muchlessthanthedurationof loadingof theactual
construction;
Scaleeffects, especially if small models areused.
The effect of stress levels shall be considered,
together withtheeffectsof particlesize.
Testsmay becarriedout onasampleof theactual
constructionor onfull scaleor smaller scalemodels.
2.2.5 Observational method
When prediction of geotechnical behaviour is diffi-
cult, itcanbeappropriatetoapplytheapproachknown
astheobservational method, inwhichthedesignis
reviewedduringconstruction.
The following requirements shall be met before
constructionisstarted:
Thelimitsof behaviour whichareacceptableshall
beestablished;
Therangeof possiblebehaviour shall beassessed
and it shall be shown that there is an acceptable
probabilitythattheactual behaviour will bewithin
theacceptablelimits;
A plan of monitoring shall bedevised which will
reveal whether theactual behaviour lieswithinthe
acceptablelimits. Themonitoringshall makethis
clear at a sufficiently early stage and with suffi-
cientlyshortintervalstoallowcontingencyactions
tobeundertakensuccessfully;
Theresponsetimeof theinstrumentsandtheproce-
duresfor analysingtheresultsshall besufficiently
rapid in relation to the possible evolution of the
system;
A plan of contingency actions shall be devised
which may be adopted if the monitoring reveals
behaviour outsideacceptablelimits.
2.3 EUROCODE 7 Part 2
EN1997-2isintendedtobeusedinconjunctionwith
EN 1997-1 and provides rules supplementary to EN
1997-1relatedtothe:
Planningandreportingof groundinvestigations,
General requirements for anumber of commonly
usedlaboratoryandfieldtests,
Interpretationandevaluationof test results,
Derivation of values of geotechnical parameters
andcoefficients.
Thefieldinvestigationprogrammeshall contain:
A planwiththelocationsof theinvestigationpoints
includingthetypesof investigations,
Thedepthof theinvestigations,
The type of samples (category, etc) to be taken
includingspecifications onthenumber anddepth
at whichtheyaretobetaken,
Specificationsonthegroundwater measurement,
Thetypesof equipment tobeused,
Thestandardsthat aretobeapplied.
Thelaboratory test programmedependsinpart on
whether comparableexperienceexists.
The extent and quality of comparable experience
for thespecificsoil or rockshouldbeestablished.
Theresults of field observations on neighbouring
structures, whenavailable, shouldalsobeused.
Thetests shall berunonspecimens representative
of therelevantstrata. Classificationtestsshall beused
to check whether thesamples andtest specimens are
representative.
This canbecheckedinaniterativeway. Inafirst
stepclassificationtestsandstrengthindextestsareper-
formedonasmany samplesaspossibletodetermine
the variability of the index properties of a stratum.
In a second step the representativeness of strength
andcompressibility tests canbecheckedby compar-
ingtheresultsof theclassificationandstrengthindex
tests of the tested sample with entire results of the
classificationandstrengthindextestsof thestratum.
Theflowchart shownbelowdemonstratesthelink
between design and field and laboratory tests. The
design part is covered by EN 1997-1; theparameter
valuespart iscoveredbyEN1997-2.
3 EUROCODE 8 DESIGNOF STRUCTURES
FOR EARTHQUAKE RESISTANCE
3.1 Introduction
The Eurocode 8 (EC8) Design of Structures for
EarthquakeResistant dealswiththedesignandcon-
structionof buildings andcivil engineeringworks in
seismicregionsisdividedinsixParts.
ThePart 1isdividedin10sections:
Section 1 containsgeneral information;
Section 2 contains the basis requirements and
compliancecriteriaapplicabletobuildingsandcivil
engineeringworksinseismicregions;
Section 3 gives the rules for the representation
of seismicactionsandtheir combinationwithother
actions;
Section 4 contains general design rules relevant
specificallytobuildings;
Section 5 presents specific rules for concrete
buildings;
Section 6 givesspecificrulesfor steel buildings;
Section 7 containsspecificrulesforsteel-concrete
compositebuildings;
Section 8 presents specific rules for timber
buildings;
Section 9 gives specific rules for masonry
buildings;
Section 10 contains fundamental requirements
andother relevant aspectsfor thedesignandsafety
relatedtobaseisolation.
53
Further Partsincludethefollowing:
Part 2 containsrelevant provisionstobridges.
Part 3 presentsprovisionsfor theseismicstrength-
eningandrepair of existingbuildings.
Part 4 gives specific provisions relevant to tanks,
silosandpipelines.
Part 5 contains specific provisions relevant to
foundations, retaining structures and geotechnical
aspects.
Part 6 presentsspecificprovisionsrelevant totow-
ers, mastsandchimneys.
In particular the Part 5 of EC8 establishes the
requirements, criteria, and rules for siting and foun-
dationsoil andcomplementstherulesof Eurocode7,
whichdonotcoverthespecial requirementsof seismic
design.
Thetopics covered by Part 1 Section 1namely:
seismic action, groundconditions andsoil investiga-
tions, importancecategories, importancefactors and
geotechnical categories andalso thetopics treatedin
Part5slopestability,potentiallyliquefiablesoils,earth
retaining structures, foundation system, topographic
aspectsarediscussed.
3.2 Seismic action
The definition of the actions (with the exception of
seismic actions) and their combinations is treated in
Eurocode1ActiononStructures.
Nevertheless the definition of some terms in EN
1998-1further clarificationof terminology is impor-
tant to avoid common misunderstandings and short-
comings in seismic hazard analysis as stressed by
Abrahamson(2000).
Ingeneral thenational territoriesaredividedbythe
National Authoritiesintoseismiczones, dependingon
thelocal hazard.
Figure1. Elasticresponsespectrum(after EC8).
InEC8, ingeneral, thehazardisdescribedinterms
of asingleparameter, i.e. thevalueagof theeffective
peak ground acceleration in rock or firmsoil called
designgroundacceleration (Figure1) expressedin
terms of: a) the reference seismic action associated
with a probability of exceeding (P
NCR
) of 10% in
50years; orb) areferencereturnperiod(T
NCR
)=475.
Theserecommendedvaluesmaybechangedbythe
National Annex of eachcountry (e.g. inUBC (1997)
theannual probabilityof exceedanceis2%in50years,
or anannual probabilityof 1/2475).
where:
Se(T) elasticresponsespectrum,
T vibration period of a linear single-degree-of-
freedomsystem,
g
designgroundacceleration,
T
B
, T
C
limits of theconstant spectral acceleration
branch,
T
D
value defining the beginning of the constant
displacement responserangeof thespectrum,
54
Ssoil parameterwithreferencevalue1.0forsubsoil
classA,
damping correction factor with referencevalue
1.0for 5%viscousdamping.
TheearthquakemotioninEC8isrepresentedbythe
elasticresponsespectrumdefinedby3components.
It isrecommendedtheuseof twotypesof spectra:
type1if theearthquakehasasurfacewavemagnitude
Msgreater than5.5andtype2inother cases.
The seismic motion may also be represented by
ground acceleration time-histories and related quan-
tities(velocityanddisplacement). Artificial accelero-
grams shall match the elastic response spectrum.
The number of the accelerograms to be used shall
give a stable statistical measure (mean and vari-
ance) and aminimumof 3 accelerograms should be
used and also some others requirements should be
satisfied.
For the computation of permanent ground defor-
mations the use of accelerograms recorded on soil
sitesinreal earthquakesorsimulatedaccelerogramsis
allowedprovidedthatthesamplesusedareadequately
qualified with regard to the seismogenic features of
thesources.
For structures with special characteristics spatial
models of theseismic action shall beused based on
theprinciplesof theelasticresponsespectra.
3.3 Ground conditions and soil investigations
For thegroundconditionsfivesubsoil classesA, B, C,
DandE areconsidered:
Subsoil classA rockorothergeological formation,
includingatmost5mof weakermaterial atthesurface
characterisedby ashear wavevelocity V
s
of at least
800m/s;
Subsoil class B deposits of very dense sand,
gravel or very stiff clay, at least several tens of
min thickness, characterised by a gradual increase
of mechanics properties with depth shear wave
velocity between360800m/s, N
SPT
>50blows and
c
u
>250kPa.
Subsoil class C deepdepositsof denseormedium
densesand, gravel or stiff clays with thickness from
several tenstomanyhundredsof meterscharacterised
byashearwavevelocityfrom160m/sto360m/s,N
SPT
from1550blowsandc
u
from70to250kPa.
Subsoil class D deposits to loose to medium
cohesionless soil (with or without some soft cohe-
sivelayers), or of predominantlysofttofirmcohesive
soil characterised by ashear wavevelocity less than
180m/s, N
SPT
lessthan15andc
u
lessthan70kPa.
Subsoil class E asoil profileconsistingof asur-
facealluviumlayer withVs, 30 values of typeC or
Dandthicknessvaryingbetweenabout5mand20m,
underlainbystiffer material withVs, 30>800m/s.
Subsoil S
1
deposits consisting or containinga
layer at least 10mthick of soft clays/siltswithhigh
plasticityindex(PI >40) andhighwatercontentchar-
acterised by ashear wavevelocity less than 100m/s
andc
u
between1020kPa.
Table 1. Values of the parameters describing the Type 1
elasticresponsespectrum*.
Groundtype S T
B
(s) T
C
(s) T
D
(s)
A 1.0 0.15 0.4 2.0
B 1.2 0.15 0.5 2.0
C 1.15 0.20 0.6 2.0
D 1.35 0.20 0.8 2.0
E 1.4 0.15 0.5 2.0
Table 2. Values of the parameters describing the Type 2
elasticresponsespectrum*.
Groundtype S T
B
(s) T
C
(s) T
D
(s)
A 1.0 0.05 0.25 1.2
B 1.35 0.05 0.25 1.2
C 1.5 0.10 0.25 1.2
D 1.8 0.10 0.30 1.2
E 1.6 0.05 0.25 1.2
Subsoil S
2
deposits of liquefiablesoils, of sen-
sitiveclays, or any other soil profilenot included in
typesAE or S
1
.
For thefivegroundtypestherecommendedvalues
fortheparametersS,T
B
,T
C
,T
D
, forType1andType2
aregiveninTables1and2.
The recommended Type 1 and Type 2 elastic
responsespectrafor ground typesA to E areshown
inFigures2and3.
Therecommendedvaluesof theparametersfor the
fiveground types A, B, C, D and E for thevertical
spectra are shown in Table 3. These values are not
appliedfor groundtypesS1andS2.
Theinfluenceof local conditions on siteamplifi-
cation proposed by Seed and Idriss (1982) is shown
in Figure4. Theinitial responsespectraproposed in
the pre-standard EC8 based in Seed and Idriss pro-
posal was underestimating the design levels of soft
soil sitesincontradictionwiththeobservationsof the
last recordedearthquakes.
Basedonrecordsof earthquakesIdriss(1990) has
shownthat peak accelerationsonsoft soilshavebeen
observed to be larger than on rock sites (Figure 5).
The high quality records from very recent earth-
quakesNorthridge(1994),Hyogo-ken-Nambu(1995),
Kocaeli (1999), Chi-Chi (1999) andTottoriken(2000)
haveconfirmedtheIdriss(1990) proposal.
Based on strong motions records obtained dur-
ing Hyogoken-Nanbu earthquake in four vertical
arrays sites and using and inverseanalysis Kokusho
and Matsumuto (1997) have plotted in Figure 6 the
maximumhorizontal acceleration ratio against max-
imumbaseacceleration andproposedtheregression
equation:
ThistrendwithabaseinaPleistocenesoil issimilar
to the Idriss (1990) proposal where the base was in
rock.
55
Figure2. RecommendedType1elasticresponsespectrum
(after EC8).
Figure3. RecommendedType2elasticresponsespectrum
(after EC8).
Table 3. Recommended values of the parameters for the
fivegroundtypesA, B, C, DandE.
Spectrum
vg
/
g
T
B
(s) T
C
(s) T
D
(s)
Type1 0.9 0.05 0.15 1.0
Type2 0.45 0.05 0.15 1.0
Thedownholearrays areuseful: (i) to understand
theseismic groundresponse; and(ii) tocalibrateour
experimental andmathematical models.
Following the comments proposed by South-
ern Member States the actual recommended elastic
response spectrumof EC8 incorporates the lessons
learnt byrecent earthquakes.
Thesoil investigationsshall followthesamecriteria
adoptedinnon-seismicareas, asdefinedinEC7(Parts
1, 2and3).
Thesoil classificationproposedinthepre-standard
of EC8, basedonlyonthe3groundmaterialsandclas-
sifiedby thewavevelocities was simpler. Theactual
groundclassificationof EC8follows aclassification
basedonshear wavevelocity, onSPT values andon
undrainedshear strength, similar toUBC (1997) that
isshowninTable4.
Based on the available strong-motion database
on equivalent linear and fully nonlinear analyses of
responsetovaryinglevelsandcharacteristicsof exci-
tationSeedetal. (1997)haveproposedforsitedepend-
ingseismicresponsetheFigures7and8, whereA
0
, A
andAB arehardtosoft rocks, B aredeepor medium
depth cohesionless or cohesivesoils, C, D soft soils
andE soft soils, highplasticitysoils.
Comments: Thefollowingcommentsarepointed:
(i) as seismic conetests haveshowngoodpotentiali-
ties they shouldalso berecommended; (ii) theEC 8
(Part 5) stresstheneedfor thedefinitionof thevaria-
tionof shear modulusanddampingwithstrainlevel,
but doesnt refer totheuseof laboratorytestssuchas
cyclicsimpleshear test, cyclictriaxial test andcyclic
torsional test. It is important to stress that adetailed
descriptionof laboratorytestsfor thestaticcharacter-
isationof soils is giveninEC 7Part 2andthesame
criteriaisnot adoptedinEC 8 Part 5.
3.4 Importance categories, importance factors
and geotechnical categories
The structures following EC 8 (Part 1.2) are classi-
fied in 4 importancecategories related with its size,
valueandimportancefor thepublicsafetyandonthe
possibilityof humanlossesincaseof acollapse.
Toeachimportancecategoryanimportantfactor
I
isassigned.Theimportantfactor
I
=1,0isassociated
withadesignseismicevent havingareferencereturn
periodof [475] years.Theimportancecategoriesvary-
ing I to IV (with the decreasing of the importance
andcomplexity of thestructures) arerelatedwiththe
importancefactor
I
assumingthevalues[1,4], [1,2],
[1,0] and[0,8], respectively.
Toestablishgeotechnical designrequirementsthree
Geotechnical Categories1, 2and3wereintroducedin
EC 7with thehighest category related with unusual
structures involving abnormal risks, or unusual or
exceptionally difficult ground or loading conditions
andstructuresinhighlyseismicareas.
Alsoitisimportanttorefer thatbuildingsof impor-
tancecategories[I,II,III]shall generallynotbeerected
intheimmediatevicinityof tectonicfaultsrecognised
as seismically activeinofficial documents issuedby
competent national authorities.
Absenceof movementinlateQuaternaryshouldbe
usedtoidentifynonactivefaultsfor most structures.
It seems that this restriction is not only very dif-
ficult tofollowfor structuressuchasbridges, tunnels
andembankmentsbutconservativeduethedifficultto
identifywithreabilitythesurfaceoutbreakof afault.
AnastapoulosandGazetas(2006) haveproposeda
methodology to designstructures against major fault
rupturesvalidatedthroughsuccessful ClassA predic-
tionsof centrifugemodel testsandhaverecommended
somechangestoEC8 Part 5.
56
Figure4. Influenceof local soil conditionsonsiteresponse(after SeedandIdriss, 1982).
Figure5. Influenceof local soil conditionsonsiteresponse(after Idriss, 1990).
Comments: The following comments are pre-
sented: (i) no referenceis madefor theinfluenceof
strong motion data with the near fault factor (con-
fined to distances of less than 10kmfromthe fault
rupture surface) with the increases of the seismic
designrequirementstobeincludedinbuildingcodes;
(ii)alsonoreferenceisestablishedbetweentheground
motionandthetypeof thefault suchasreversefault-
ing, strikeslipfaultingandnormal faulting; (iii) EC8
referstothespatial variationof groundmotionbutdoes
not present any guidance; (iv) basin edge and other
2Dand3Deffectswerenot incorporatedinEC8. The
importanceof shapesof theboundariesof sedimentary
valleys as well as of deeper geologic structures in
determiningsiteresponsewasshownfromtheanalysis
of recordsinNorthridgeandKobeearthquakes.
3.5 Potentially liquefiable soils
Following 4.1.3.(2) Part5 EC8 An evaluation
of theliquefactionsusceptibility shall bemadewhen
thefoundationssoilsincludeextendedlayersor thick
lenses of loose sand, with or without silt/clay fines,
beneaththewater level, andwhensuchlevel is close
tothegroundsurface.
57
Figure 6. Maximumhorizontal acceleration ratio plotted
against maximum base acceleration (after Kokusho and
Matsumoto, 1997).
Table4. Groundprofiletypes(after UBC, 1997).
Shear Undrained
Ground wave shear
profile Ground velocity SPT strength
type description Vs(m/s) test (kPa)
S
A
Hardrock 1500
S
B
Rock 7601500
S
C
Verydense 360760 >50 >100
soil and
soft rock
S
D
Stiff soil 180360 1550 50100
S
E
Soft soil -180 -15 -50
S
F
Special
soils
Figure7. Proposedsite-dependent relationship(after Seed
et al., 1997).
Soil investigations should include SPT or CPT
testsandgrainsizedistribution(IdrissandBoulanger,
2004). Normalisation of the overburden effects can
be performed by multiplying SPT or CPT value by
Figure 8. Proposed site-dependent response spectra, with
5%damping(after Seedet al., 1997).
Table5. Magnitudescalingfactors.
Magnitude Seed& Idriss IdrissNCEER Ambraseys
M (1982) (1997) (1988)
5.5 1.43 2.20 2.86
6.0 1.32 1.76 2.20
6.5 1.19 1.44 1.69
7.0 1.08 1.19 1.30
7.5 1.00 1.00 1.00
8.0 0.94 0.84 0.67
8.5 0.89 0.72 0.44
thefactor (100/
vo
)
1,2
where
vo
(kPa) istheeffective
overburdenpressure.Thisnormalisationfactorshould
betakennot smaller than0.5andnot greater than2.
Theseismic shear stress
e
canbeestimatedfrom
thesimplifiedexpression:
where
gr
isthedesigngroundaccelerationratio,
f
is
theimportancefactor, Sisthesoil parameterand
vo
is
thetotal overburdenpressure. Thisexpressionshould
not beappliedfor depthslarger than20m. Theshear
level shouldbemultipliedbyasafetyfactor of [1.25].
ThemagnitudecorrectionfactorsinEC8followthe
proposal of Ambraseys(1988) andaredifferent from
theNCEER(1997) factors.A comparisonbetweenthe
different proposalsisshowninTable5.
Anewproposal withasummaryof differentauthors
presentedbySeedet al. (2001) isshowninFigure9.
Empirical liquefaction charts aregiven with seis-
micshear wavevelocitiesversusSPT valuestoassess
liquefaction. A comparison between NCEER (1997)
and EC8 proposal for pre-standard is shown in Fig-
ure 10. It is important to refer that the proposal for
EC8isbasedontheresultsof Roberstonet al.(1992)
58
Figure9. Recommendations for correlations with magni-
tude(after Seedet. al., 2001).
Figure 10. Liquefaction potential assessment by NCEER
(1997) andEC8(pre-standard).
andtheproposal of NCEER(1997) incorporatesvery
recent results.
However theEC8 standard version considers that
thesecorrelationsarestill underdevelopmentandneed
theassistanceof aspecialist.
Theimportanceof thistopichasincreasedandthe
assessmentof liquefactionresistancefromshear wave
crosshole tomography was proposed by Furuta and
Yamamoto(2000).
A newproposal presentedby Cetinet al. (2001) is
shown in Figure11 considered advanced in relation
withthepreviousones, asintegrates: (i) dataof recent
earthquakes;(ii)correctionsduetheexistenceof fines;
(iii) experiencerelatedabetter interpretationof SPT
test; (iv) local effects; (v) caseshistoriesrelatedmore
than200earthquakes; (vi) Baysianatheory.
Figure11. Probabilistic approachfor liquefactionanalysis
(after Cetinet al., 2001).
Figure 12. Relationship between (N1) 60 and undrained
residual strength(after SeedandHarder, 1990).
Bray et al. (2004) have shown that the chinese
criteriaproposed by Seed and Idriss (1982) was not
reliablefor theanalysisof siltysandsliquefactionand
haveproposedtheplasticityindex.
Thetopicrelatedwiththeassessmentof postlique-
factionstrengthisnottreatedinEC8, butitseemsthat
thefollowingvariablesareimportant: fabricortypeof
compaction, directionof loading, voidratioandinitial
effectiveconfiningstress(ByrneandBeaty, 1999).
A relationship between SPT N valueand residual
strengthwasproposedbySeedandHarder(1990)from
direct testingandfieldexperience(Figure12).
Ishihara et al.(1990) have proposed a relation of
normalized residual strength and SPT tests, based
on laboratory tests compared with data fromback-
analysis of actual failure cases (Figure 13). Also
59
Figure 13. Relation of normalized residual strength and
SPT tests(after Ishiharaet al., 1990).
Figure14. Normalizedresidual strengthplottedversusCPT
values(after Ishiharaet al., 1990).
Ishiharaet al. (1990) by assemblingrecordsof earth-
quakecausedfailuresinembankments, tailingsdams,
andriverdykeshaveproposedtherelationof Figure14,
in terms of the normalized residual strength plotted
versusCPT value.
Alba(2004) has proposed Binghammodel, based
intriaxial tests of largesamples, tosimulateresidual
strengthof liquefiedsands.
Thesusceptibilityof foundationssoilstodensifica-
tion and to excessivesettlements is referred in EC8,
buttheassessmentof expectedliquefaction induced
deformationdeservesmoreconsideration.
By combination of cyclic shear stress ratio
and normalized SPT N-values Tokimatsu and Seed
(1987) haveproposedrelationships with shear strain
(Figure15).
To assess thesettlement of theground dueto the
liquefactionof sanddepositsbasedontheknowledge
of thesafetyfactoragainstliquefactionandtherelative
densityconvertedtothevalueof N1achart(Figure16)
wasproposedbyIshihara(1993).
Following EC8 ground improvement against liq-
uefaction should compact the soil or use drainage
to reduce the pore water pressure. The use of pile
foundationsshouldbeconsideredwithcautionduethe
Figure15. Correlationbetweenvolumetric strainandSPT
(afterTokimatsuandSeed, 1987).
Figure16. Postcyclicliquefactionvolumetricstraincurves
usingCPT andSPT results(after Ishihara, 1993).
largeforcesinducedinthepilesbytheliquefiablelay-
ers andthedifficulties to determinethelocationand
thicknessof theselayers.
Twocategoriesof remedial measuresagainstlique-
factionwereproposed:
(i) Solutions aiming at withstanding liquefaction
confinement wall: stiff walls anchored in a
60
nonliquefiedlayer (or abedrock) toavoidlateral
spreading in caseof liquefaction; soil reinforce-
ment transfer of loadstoanon-liquefiablelayer.
(ii) Solutions to avoidliquefaction: soil densifica-
tion: compactiongroutingtominimisethelique-
factionpotential; dewatering: tolower thewater
table in order to minimise the risk of liquefac-
tion; drainage: to facilitate the dissipation of
porepressure; Finegrouting: toincreasethesoil
cohesion.
Theliquefactionprediction/determinationmethods
arecoveredbythefollowingmainjapanesestandards:
(i) Designstandardsfor Port andHarbour Structures,
(ii) Design Standards of Building Foundations, (iii)
DesignStandardsfor RailwayStructures; (iv) Design
Specificationsfor Roads.
Following EC8 ground improvement against liq-
uefactionshouldcompact thesoil or usedrainageto
reducetheporewaterpressure.Theuseof pilefounda-
tionsshouldbeconsideredwithcautionduethelarge
forcesinducedinthepilesbytheliquefiablelayersand
thedifficultiestodeterminethelocationandthickness
of theselayers.
Theremedial measuresagainst liquefactioncanbe
classified in two categories (TC4 ISSMGE, 2001;
INA, 2001): (i) the prevention of liquefaction; and
(ii) the reduction of damage to facilities due to
liquefaction.
The measures to prevent of occurrence of lique-
factionincludetheimprovement of soil propertiesor
improvementof conditionsforstress, deformationand
porewaterpressure. Inpracticeacombinationof these
twomethodsisadopted.
Themeasurestoreduceliquefactioninduceddam-
age to facilities include (1) to maintain stability by
reinforcing structure: reinforcement of pile founda-
tionandreinforcement of soil deformationwithsheet
pileandundergroundwall; (2) torelieveexternal force
bysofteningor modifyingstructure: adjustingof bulk
unit weight, anchorageof buriedstructures, flattering
embankments.
In NEMISREF Project the following criteria for
selection was used (Evers, 2005): (i) Potential effi-
ciency; (ii) Technical feasibility; (iii) Impact on
structureand environmental; (iv) Cost-effectiveness;
(v) Innovation.
Twomethodswereselected: (i) Soil groutingusing
calcifyingbacteria; (ii) confinement wall.
Relatedwithcalcifyingbacteriatheobjectiveof soil
consolidation is to createacementation between the
grainsof soil skeletonincreasingthecohesion.
Withconfinement wall evenif partial liquefaction
couldoccur thefinal deformationswill becontrolled.
The improvement of soil properties, to prevent
soil liquefaction, by soil cementation and solidifica-
tionisperformedby deepmix method(Port Harbour
ResearchInstitute,1997),sowithinthisframeworkthe
useof bacteriatechniqueisinnovative.
The structural strengthening is performed by pile
foundation and sheet pile (INA, 2001) and so the
confiningwall canbeconsideredinnovative.
Theproposedmethodsof remediationhaveanaddi-
tional advantage minimizing the effects on existing
structuresduringsoil improvement.
Comments: Fromthe analyses of this section it
seems that thefollowingitems deservemoreclarifi-
cation(ScoePinto, 1999):
i) It isimportant toquantify thevaluesof extended
layersor thicklensesof loosesand;
ii) What is themeaningof whensuchlevel is
closetothegroundsurface? What depth? What
isthemaximumdepthliquefactioncanoccur?
iii) Norecommendationispresentedtocomputeseis-
micshear stress
e
for depthslarger than20m;
iv) Theuseof Beckerhammerandgeophysical teststo
assesstheliquefactionof gravelymaterialsshould
bestressed;
v) TherecommendedmultipliedfactorCMforearth-
quake magnitudes different from 7.5 deserves
moreexplanation. It isimportant torefer that the
well known correlation proposed by Seed et al
(1984) for cyclic stress ratio versus N
1
(60) to
assess liquefaction and adopted in Annex B of
EC8 Part 5 use different correction factor for
earthquakemagnitudesdifferent from7.5;
vi) Noreferenceisgivenfor theresidual strengthof
soil.
3.6 Foundation system
Ingeneral for theSoil-StructureInteraction(SSI) the
design engineers ignore the kinematic component,
consideringafixedbaseanalysisof thestructure, due
thefollowingreasons: (i) insomecasesthekinematic
interaction may be neglected;(ii) aseismic building
codes, withafewexceptions e.g. Eurocode8do not
refer it; (iii) kinematic interaction effects are more
difficult to assess than inertial forces (Sco ePinto,
2003).
Thereisstrongevidencethatslender tall structures,
structuresfoundedinverysoftsoilsandstructureswith
deepfoundationstheSSI playsandimportant role.
TheEurocode8states: Bendingmoments devel-
opingduetokinematicinteractionshall becomputed
only when two or more of the following conditions
occur simultaneously: (i) thesubsoil profileisof class
D, S
1
or S
2
, and contains consecutive layers with
sharply differing stiffness; (ii) the zone is of mod-
erateor highseismicity, >0.10; (iii) thesupported
structureisof important categoryI or II.
Thestabilityof footingsfor theultimatestatelimit
design criteria shall be analysed against failure by
slidingandagainst bearingcapacitylure.
Forshallowfoundationsunderseismicloadsfailure
can not bedefined for situations when safety factor
becomes less than 1, but is related with permanent
irrecoverabledisplacements.
Theseismiccodesrecommendtocheckthefollow-
inginequality:
61
Figure 17. Bounding surface for cohesive soils (after
Pecker, 1997).
whereS
d
istheseismicdesignactionandR
d
thesystem
designresistance.
In the inequality (3) partial safety factors shall
be included following the recommendations of
Eurocode8.
Theoretical and experimental studies to provide
bearingcapacitysolutionstoincludetheeffect of soil
inertiaforcesledtotheinequality(Pecker, 1997):
where =0 defines the equation of the bounding
surface(Figure17).
The combination of the loading lying the outside
thesurfacecorresponds to an unstablesituation and
the combination lying inside the bounding surface
correspondstoapotentiallystablesituation.
Piles and piers shall be designed to resist the
following action effects: (i) inertia forces fromthe
superstructure; and (ii) kinematic forces resulting
fromthedeformationof thesurroundingsoil duethe
propagationof seismicwaves.
Thecompletesolution is a3D analysis very time
demandinganditisnotadequatefor designpurposes.
Thedecompositionof theprobleminstepsisshownin
Figure18andimplies(GazetasandMylonakis, 1998):
(i) thekinematicinteractioninvolvingtheresponseof
the base acceleration of the systemconsidering the
mass of superstructure equal to zero; (ii) the iner-
tial interaction that involves the computation of the
dynamic impedances at thefoundation level and the
dynamicresponseof thesuperstructure.
For the computation of internal forces along the
pile, as well as thedeflectionandrotationat thepile
head, both discrete(basedinWinkler Springmodel)
or continuummodels can be used (Finn and Fujita,
2004).
Thelateral resistanceof soil layers susceptibleto
liquefactionshall beneglected.
Ingeneral thelinear behaviour is assumedfor the
soil. Thenonlinear systems aremoregeneral andthe
termnonlinearitiesincludethegeometricandmaterial
nonlinearities(Pecker andPender, 2000).
The engineering approach considers two sub-
domains(Figure19):
i) A far field domain where the non linearities are
negligible;
ii) A near field domain in the neighbouring of the
foundationwheretheeffectsof thegeometrical and
material linearitiesareconcentrated.
Thefollowing effects shall beincluded: (i) flexu-
ral stiffness of the pile; (ii) soil reactions along the
pile; (iii) pilegroup effects; and (iv) theconnection
betweenpileandstructure.
The use of inclined piles is not recommended to
absorbthelateral loads of thesoils. If inclinedpiles
areusedtheymustbedesignedtosupportaxial aswell
bendingloads.
Pilesshall bedesignedtoremainelastic,if thisisnot
possiblepotential plastic hingingshall beconsidered
for: (i) aregionof depth2d(d-diameter of thepile)
fromthepilecap; (ii) aregionof 2dfromanyinter-
facebetweentwolayerswithdifferent shear stiffness
(ratioof shear moduli >6).
Evidencehasshownthatsoil confinementincreases
pile ductibility capacity and increases pile plastic
hingelength. Pileshaveshownthecapabilitytoretain
much of their axial and lateral capacity even after
crackingandexperiencedductibility levels upto 2.5
(GerolymosandGazetas, 2006).
Theinvestigationmethodsforpilefoundationdam-
ageare: direct visual inspection, theuseof borehole
camerainspectionandpileintegrity test. Theground
deformationcanbeinvestigatedbyvisual surveyand
GPSsurvey(Matsui et al. 1997).
Comments: The following topics deserve more
consideration:
i) Theinfluenceof pilecap;
ii) Themoment rotationcapacityof pilefooting;
iii) Theincorporationof thenonlinear behaviour of
thematerialsinthemethodsof analysis;
iv) The instrumentation of the piles for design pur-
poses;
v) Some guidelines about group effects, as there
are significant different opinions on the influ-
enceof groupeffects relatedwiththenumber of
piles, spacing, direction of loads, soil types and
constructionmethodsof piles.
For theevaluationof mitigationmethodsaprelimi-
naryanalysisof thefollowingsolutionswasperformed
(Evers, 2005): (I) Stiffening solutions hard layer,
reinforcedconcretewalls, soil stiffeningatfoundation
level and inclined piles; (ii) Soft material barriers
softlayer,expandedpolystyrene(EPS)walls,air-water
balloonsandsoft caisson; (iii) oscillators.
For the criteria of selection the following factors
were used: Potential efficiency, technical feasibil-
ity, impact on structure and environment, cost-
effectivenessandinnovation.
From this analysis the following two mitigation
methods:i)soil stiffening(inclinedmicro-piles)andii)
deformablesoft barriers(soft caisson) wereselected.
62
Figure18. Soil-structureinteractionproblem(after GazetasandMylonakis, 1998).
4 INTERACTIONWITHOTHER SEISMIC
CODES
The continuous process of elaborating codes and
standards that incorporates the lessons learned by
earthquakes has a significant effect for the design
andconstructioninseismicareas. Asanexampledue
theconsequencesof Kobeearthquakesnewimprove-
ments havebeen implemented for theassessment of
liquefactioninJ apanesecodes(Yasuda, 1999).
63
Figure19. Conceptual subdomainsfor dynamicsoil struc-
tureanalyses(after Pecker andPender, 2000).
Theactual tendency inalmost every regionof the
worldistoelaborateunifiedcodes. Thiswasthemain
purposeof Eurocodes.
InUnitedStates, neverthelessthelongtraditionof
existingdifferent codesinthestates, duethedifferent
geographic regions of the U. S., where the western
region in comparison with eastern region faces the
largest levels of probabilistic seismic risk, efforts to
elaborate an uniform code incorporating the UBC
(Uniform Building Code) with NEHRP (National
SeismicHazardReductionProgram) inIBC(Interna-
tional BuildingCode), notaninternational code, buta
unifiednational code, areundertaken(SeedandMoss,
1999).
In general all geotechnical codes use factored
parametersforloads,resistanceandstrengthtoharmo-
nizewithstructural practice. But theway of doingis
different, for instancetheEurocodeusespartial safety
factors for loads and strength (Cuellar, 1999) and in
theNewZealandcodetheLRFD(loadandresistance
factoreddesign) isused(Pender, 1999).
It is important to stress that Eurocodes represent
asignificant stepforwardandallowsvariousNation-
ally Determined Parameters (NDP) which should be
confirmedor modifiedintheNational Annexes.
5 FINAL REMARKS
Thework performedbytheCommissionof theEuro-
peanCommunities(CEC) inpreparingtheStructural
Eurocodes inorder to establishaset of harmonised
technical rules is impressive. However we feel that
sometopicsdeservemoreconsideration.
Earthquakes arevery complex anddangerous nat-
ural phenomena, which occurs primary in known
seismiczones, althoughsevereearthquakeshavealso
occurred outside these zones in areas considered
beinggeologicallystable.Asaresult, regulatoryagen-
cies becamemorestringent intheir requirements for
demonstrationof adequateseismicstabilityanddesign
engineers responded by developing new and more
convincing design approaches than had previously
used. Thus thepast years haveseen amajor change
ininterest andattitudetowardsthisaspect of design.
Thelessons learnedfromrecent earthquakes such
as: Northridge(1994), Kobe(1995), Umbria-Marche
(1997),Kocaeli (1999),Athens(1999),Chi-Chi (1999)
and Bhuj (2001) have provided important obser-
vational data related with the seismic behavior of
geotechnical structures.
The need of cost effective methods to upgrade
buildings by developing new specific foundations
techniques is a major problem. So the objective of
reducing the earthquake motion transferred to the
structurethroughthefoundationby developinginno-
vative constructive techniques for soil improvement
andsoil reinforcement isgettingincreaseattention.
One very important question to be discussed is:
(i) how detailed a seismic code must be; (ii) what
is the time consuming to establish a set of har-
monisedtechnical rules for thedesignandconstruc-
tionworks?(iii) Howtoimprovetherelationsbetween
theusers: relevant authorities, clients anddesigners?
and(iv) howtoimplement inpracticethat codesmay
not cover indetail everypossibledesignsituationand
it mayrequirespecialisedengineeringjudgement and
experience?Itishopedthatthecontributionstobepre-
sentedby CEN members, inthenext years, will help
to clarify several questions that still remain without
answer.
It isimportant tonoticethat trueinnovatorshavea
mantra:Theyareconstantlydaringtomakethingsbet-
ter. Theychallengethecommonlyaccepted. Theysee
nolimits. Weshouldnot forget that growth, evolution
andreinventionsustainlife.
Soweneedtokeepchallengingourselvestothink
better, do better and be better. Confront our limita-
tions. Failureisagift anyway. It takesuscloser toour
dreams, equipsuswithmoreknowledgeandtoughus
up. Successandfailuregohandtohand.
IndealingwithEurocodes weshouldnot forget 4
lessons:
Improve: Alwaysbegettingbetter;
Observe: We need to keep our eyes open to
absorbthechanges;
Connect: Weneedtoreceivedifferent inputs;
Adapt: Theconditionsaredifferent, soweneed
tokeepmonitoringtheprocess.
Indealingwiththissubject weshouldalwayshave
inmind:
All for Love
Errors, like straws, upon the surface
flow;
He who would search for pearls must
dive below.
(J ohnDryden)
REFERENCES
Abrahamson, N.A. (2000) Stateof thepracticeof seismic
hazardevaluation. Vol. 1, pp659-685. GEOENG 2000,
Melbourne.
64
Alba, P. (2004) Residual strength.after liquefaction: arheo-
logical approach. Proc. of 3rdInternational Conference
onEarthquakeGeotechnical Engineering, Berkeley, Edi-
tors D.Doolin, A. Kammerer, T. Nogami, R.B. Seed e
I.Towhata, Vol. 1pp. 740746.
Ambraseys, N. N. (1998) Engineeringseismology. Earth-
quakeEngineeringandStructural Dynamics, Vol. 17, pp.
1105.
Anastassopoulos, I. andGazetas, G. (2006) Designof foun-
dations and structures against fault displacement. ETC
12- Workshop, Athens.
Bray, J., Sancio, R.B., Riemer, M. and Durgunoglu, H.T.
(2004) Liquefaction susceptibility of finegrained soils.
Proc. of 3rd International Conference on Earthquake
Geotechnical Engineering, Berkeley, Editors D. Doolin,
A. Kammerer,T. Nogami, R.B. SeedandI.Towhata,Vol. 1
pp. 665662.
Cetin, K.O e Seed, R.B, and Kiureghion. (2001) Reliabil-
ity based assessment of seismic soil liquefaction initia-
tion, XV ICSMGE TC4SatelliteConferenceonLessons
Learned fromRecent Strong Earthquakes, pp 327332.
EditedbyAtillaAnsal,.
Cuellar, V. (1999) Codes and standard for Europe. Proc.
of the Second International Conference on Earthquake
geotechnical Engineering.EditedbyPedroS.ScoePinto.
PublishedbyBalkema.Vol. 3, pp. 11291133.
EN1990Eurocode0 Basisof structural design.
EN1991Eurocode1 Actionsonstructures.
EN1997Eurocode7 Geotechnical design.
EN 1998Eurocode8 Designof structuresfor earthquake
resistance.
Evers, G. (2005) Horizontal shakingmitigationimplemen-
tation. NEMISREF Seminar, CD-Rom, Athenes.
Evers, G. (2005) Liquefactionmitigationimplementation.
NEMISREF Seminar, CD-Rom, Athenes.
Finn, W. D. and Fujita, N.(2004) Behavior of piles in liq-
uefiable soils during esrthquakes: Analysis and design
issues, Proc. Of theFifthCaseHistoriesinGeotechnical
EngineeringConference, NewYork, NY, SOAP, pp.16.
FurutaandYamamoto (2000). Liquefactionassessment by
shear wave crosshole tomography tests. Paper n
o
831.
12thWCEE, Auckland, NewZealand.
Gazetas, G. andMylonakis, G(1998) Seismicsoil structure
interaction: newevidenceandemergingissues. Geotech-
nical Earthquake Engineering and Soil Dynamics,
ASCE II, pp11191174.
Gerolymos, G. andGazetas, G. (2006) Capacity designof
pile foundations: Evidence in support of allowing pile
yielding. ETC 12- Workshop, Athens.
Idriss, I. M. (1990) Responseof soft soil duringsoil earth-
quakes. Proc. H. Bolton Seed Memorial Symposium,
pp. 273290.
Idriss, I.M. andBoulanger, R.W.(2004) Semi-empirical pro-
ceduresforevaluationliquefactionpotential duringearth-
quakesProc. Of theFifthCaseHistoriesinGeotechnical
EngineeringConference, NewYork, Ny, SOAP, pp.16.
Ishihara, K. 1985. Stabilityof natural depositsduringearth-
quake. Proc. 11 th ICSMGE, S. Francisco, Vol.2, pp.
321376.
Ishihara, K., Yasudfa, S. and Yoshida, Y. (1990) Liq-
uefaction induced flow failure of embankments and
residual strength of silty sands. SF, Vol. 30, n
o
3,
pp. 6980.
Ishihara, K.(1993). Liquefaction and flow failure during
earthquakes, 33rdRankineLecture. Geotechnique43(3),
pp351415.
Kokusho, T. and Matsumoto, M (1997) Nonlinear
site response during the Hyogoken-Nanbu earthquake
recorded by vertical arrays in view of seismic zonation
methodology. Proc of theDiscussionSpecial Technical
SessiononEarthquakeGeotechnical Engineeringduring
14thICSMFE, Hamburg, EditedbyPedroS.ScoePinto.
PublishedbyBalkema., pp. 6169.
Matsui, Y., Kitazawa, M., Nanjo, A andYasuda, F. (1977)
Investigationof damagedfoundationsintheGreat Han-
shinearthquake.Procof theDiscussionSpecial Technical
SessiononEarthquakeGeotechnical Engineeringduring
14thICSMFE, Hamburg, EditedbyPedroS.ScoePinto.
PublishedbyBalkema., pp. 235242.
NCEER (1997) Proc. NCCER Workshop on Evaluation
of Liquefaction Resistance of Soils, Summary Report,
Edited by T. LeslieYoud and I.M. Idriss, National Cen-
ter for Earthquake Engineering Research, University of
Buffalo, Technical Report NCEER-97-0022.
Paolucci, R. (2006) Numerical Investigation of 3D Seis-
mic Amplification by real Steep Topographic Profiles
andCheck of theEC8TopographicAmplificationCoef-
ficients. ETC 12- Workshop, Athens.
Pecker, A. and Pender, M.J. (2000) Earthquake resistant
design of foundation. New construction. Vol. 1, pp
313332. GEOENG2000, Melbourne.
Pecker, A. (1997) Analytical formulaefor theseismicbear-
ing capacity of shallow strip foundations. Proc of
theDiscussionSpecial Technical SessiononEarthquake
Geotechnical Engineering during 14th ICSMFE, Ham-
burg, Edited by Pedro S.Sco e Pinto. Published by
Balkema., pp. 262268.
Pender, M.J. (1999) Geotechnical earthquake engineering
design practice in New Zealand. Proc. of the Sec-
ondInternational ConferenceonEarthquakegeotechnical
Engineering. Editedby Pedro S.Sco ePinto. Published
byBalkema.Vol. 3, pp. 11291133.
Roberston, P.K.,Woeller, D.J. andFinn,W.D.L. (1992). Seis-
micconepenetrationtestforevaluatingliquefactionunder
cyclicloading, CanadianGeotechnical J ournal, Vol, 29,
pp. 686695.
Sco ePinto, P. S. (1999) Therelation between Eurocode
8andEurocode7. Proceedingsof theTwelfthEuropean
Conference on Soil Mechanics and Geotechnical Engi-
neering, Vol. 3, pp. 22232228, Amsterdam, Edited by
F.B.J. Barends, J. Lindenberg, H.J. Luger, L. deQuelerit
andA. Verruit. PublisherA. A. Balkema.
Sco e Pinto, P. (2003) Seismic behaviour of geotechni-
cal structures. Inaugural lecture, Proc. 13 th Regional
African Conference of Soil Mechanics and Geotechni-
cal Engineeringpp324., Marrakech, EditedbyM.Sahli,
L.Bahi & R.Khalid.
Seed, H. B. and I. M. Idriss (1982). Ground motions
and soil liquefaction during earthquakes. Earthquake
EngineeringResearchInstitute, Oakland, California.
Seed, H. B., Tokimatsu, K.,Harder, L.& Chung, R. (1984)
The influence of SPT procedures in soil liquefac-
tion resistance evaluations. Earthquake Engineering
ResearchCentreReport n
o
84/15U.C.
Seed, R.B., Cetin, K.O and Moss, R.E.S (2001) Recent
advances in soil liquefaction hazard assessment, XV
ICSMGE TC4SatelliteConferenceonLessons Learned
fromRecent StrongEarthquakes, Istanbul, pp. 319326.
EditedbyAtillaAnsal,
Seed, R.B., Chang, S. W., Dickenson, S.E. and Bray, J. B.
(1997) Site dependent seismic response including
recent strong motion data. Proc of the Discussion
Special Technical Session on Earthquake Geotechni-
cal Engineeringduring14thICSMFE, Hamburg, Edited
by Pedro S.Sco e Pinto. Published by Balkema.,
pp. 125134.
65
Seed, H.B. and Harder, L.F. (1990) SPT-based analysis of
cyclic pore pressure generation and undrained residual
strength. Proc. of Memorial Symposiumof H. B. Seed,
Vol. 2, pp. 351376.
Seed, R.B. andMoss, R.E.S. (1999) RecentadvancesinU.S.
codes and policy with regard to seismic geotechnics.
Proc. of theSecond International Conferenceon Earth-
quakegeotechnical Engineering. EditedbyPedroS.Sco
ePinto. PublishedbyBalkema.Vol. 3, pp. 11111116.
TC4 (ISSMGE) (1999) Manual for Zonation on Seismic
Geotechnical Hazards. (RevisedVersion).
TC4(ISSMGE). (2001) CaseHistoriesof Post-Liquefaction
Remediation. Committee on Earthquake Geotechnical
Engineering.
Tokimatsu, K andSeed, H. B. (1987). Evaluationof settle-
mentsinsandsduetoearthquakeshaking, J GE, ASCE,
113, pp. 861878.
UBC (1997) UniformBuidingCode, International Confer-
enceof BuildingOfficials, Whittier, California, Vol. II.
Yasuda, S. (1999) Seismic designcodesfor liquefactionin
Asia. Proc. of theSecond International Conferenceon
Earthquakegeotechnical Engineering. EditedbyPedroS.
Sco e Pinto. Published by Balkema. Vol. 3, pp. 1117
1122.
66
Special sessions
Reliability benchmarking
Reliabilityanalysisof abenchmarkproblemfor 1-Dconsolidation
J.Y. Ching
National Taiwan University, Taipei, Taiwan
K.-K. Phoon
National University of Singapore, Singapore
Y.-H. Hsieh
National Taiwan University of Science and Technology, Taipei, Taiwan
ABSTRACT: Thispaper presentstheanalysisresultsof areliabilitybenchmarkproblemfor one-dimensional
consolidation.Thepeculiarityof thisexampleisthattherearetwolocal solutionsforthedesignpoints, depending
ontheinitial pointof thesearchalgorithm. Oneof thelocal designpointsistheglobal solutionof thefirst-order
reliability method, whiletheother is afakeone. Unfortunately, it turns out to berelatively easy to find the
fakesolution. Thisphenomenonof multipledesignpointsisstudiedanddocumentedindetail inthispaper. The
analysisresultsbyusingother reliabilitymethodsarealsopresentedfor comparison. Finally, recommendations
aregivenfor thesuitablereliabilitymethods.
1 INTRODUCTION
Thepurposeof thisbenchmarkexampleistoexamine
therobustnessof variousreliabilitymethodsfor prob-
lems with non-differentiable performance functions,
whichis clearly thecasefor aone-dimensional con-
solidationproblemwhereswitchingbetweenNC and
OC regimesispossible.
1.1 Description of the problem
Considerasaturatedclaylayerof thicknessH between
twosandlayers, asshowninFigure1. Thesurcharge
pressureat thegroundsurfaceisq. Theclayislightly
overconsolidated with an uncertain OCR. The com-
pressionC
c
andrecompressionC
r
indicesof theclay
arealso uncertain. It is assumedthat therecompres-
sionindexisanuncertainfractionof thecompression
index:
The failure is defined as the consolidation settle-
mentexceedsaprescribedallowablesettlementS
allow
.
In other words, the performance function g can be
writtenas(g -0definesfailure):
Figure1. Theconsolidationproblem.
wheree
clay
istheinitial voidratioof theclay;
0
and
aretheeffectiveconsolidationstressesbeforeandafter
the surcharge is applied;
p
is the preconsolidation
stress;
and
where and
sat
are the moist and saturated units
weightsof thesoils;
w
=9.8kN/m
2
istheunitweight
of water. The moist and saturated soil unit weights
69
Table1. Thevaluesanddistributionsof theinputvariables.
Variable Distribution Statistics
S
allow
Deterministic 0.05m
q Lognormal mean=20kN/m
2
; cov
=20%
H Gaussian mean=4m; cov=10%
e
clay
Lognormal mean=1.2; cov=15%
e
sand
G
clay
s
Uniform [2.5, 2.7]
G
sand
s
Uniform [2.5, 2.7]
OCR Uniform [1.5, 2.5]
C
c
Uniform [0.1, 0.2]
* cov standsfor coefficient of variation.
are not independent, because they are related to the
specific gravity of the soil solids G
s
and the void
ratioe:
where a degree of saturation=20% is assumed for
moist.Therearenineindependentrandomvariables
in this problem, including q, H, e
clay
, e
sand
, G
clay
s
,
G
sand
s
, OCR, C
c
and.
1.2 Input variables
Thevaluesanddistributionsof thebasicinputvariables
aresummarizedinTable1.
1.3 Solution methods
Thereliabilitymethodsemployedinthisstudyinclude
the First-Order Second Moment (FOSM), First-
Order Reliability Method (FORM) (Ang and Tang
1984), Second-Order Reliability Method (SORM)
(Der Kiureghian and Stefano 1991), direct Monte
Carlosimulation(MCS) andSubset simulation(Sub-
sim)(AuandBeck2001).TheFOSMmethodisimple-
mented with the assumption that the performance
function g is Gaussian. The FORM is implemented
with thegradient projection algorithm(Liu and Der
Kiureghian1991)forthesearchof thedesignpoint;the
SORM isimplementedwiththealgorithmdeveloped
byDer KiureghianandStefano(1991).
1.4 Results of reliability analyses
Table2summarizestheanalysisresultsof theadopted
reliability methods. Thefactor of safety of thisprob-
lem, definedasS
allow
dividedbythesettlement when
all uncertainvariablesarefixedat their meanvalues,
isaround2.1. Thereliabilityindex for FORM isthe
distanceof theobtaineddesignpointtotheorigin. For
Table2. Primaryanalysisresults.
Solution
method FOSM FORM SORM Subsim MCS
2.94 1.43 1.55 1.53 1.52
2.15
2.29
P
F
estimate 0.0017 0.076 0.061 0.063 0.064
0.016 0.011
%error in 97.4 19.3 4.7 0.8
P
F
estimate 75.3 82.6
#of g 19 152 261 1900 10
6
evaluations
Estimator n/a n/a n/a 10.2% 0.4%
cov
multipledesignpoint solutions.
theother methods, itissimplytakentobe+
1
(P
F
),
whereP
F
denotesthefailureprobabilityestimate, and
+isthecumulativedensityfunction(CDF)of thestan-
dardGaussiandistribution. For Subsim, 1000samples
aretakenineachstage.
TheFOSM, FORMandSORMareanalytical meth-
ods, sotheP
F
estimatorsaredeterministic. However,
MCS and Subsimare simulation methods: their P
F
estimators are random. The coefficient of variation
(cov) of their estimates are listed in the table. The
cov for MCS is calculated based on the formula
cov=[(1P
F
),P
F
,n]
0.5
, where n =10
6
is the total
number of MCSsamples. Thecovfor Subsimisesti-
matedbasedontheP
F
estimatesfrom100independent
Subsimruns.
The%error inP
F
estimateisthepercentageerror
comparedtotheMCSsolution, whichshouldbevery
closetotheactual failureprobabilityjudgingfromthe
0.4%cov. Inspiteof thehighefficiency, FOSMsignif-
icantly underestimates P
F
. Thepoor performanceof
FOSMmaybeduetoitssimplifiedassumptiononthe
distributionof theg functionandduetotheinaccuracy
of thelinearizationfor theg function.
For FORM, theadopted gradient projection (GP)
algorithmyields two local solutions in the standard
Gaussianspace: thefirst onegivesareliability index
of 1.43, and thesecond onegives areliability index
of 2.15. It is clear that thefirst solutionis theglobal
solutionbecauseit iscloser totheorigin. Thisglobal
solutioncorrespondstoafailureprobabilityof 0.076,
whichisreasonablyclosetotheMCSsolution0.064.
Whether the GP algorithm would converge to the
global solutiondepends onthelocationof theinitial
trial point in the standard Gaussian space. Unfortu-
nately, whentheinitial trial point is randomly gener-
atedfromthestandardGaussiandistribution, thereis
only about 20% chance of converging to the global
solution: 80%of thetimethealgorithmwill converge
tothefakesolution. Inparticular, whentheinitial trial
pointistakentobetheoriginof thestandardGaussian
space,theGPalgorithmconvergestothefakesolution.
TheSORMusesthesolutionfromFORM,therefore
SORM alsoleads totwosolutions. Thefirst solution
of P
F
=0.061 follows from the global solution of
70
Table3. Coordinatesof thetwolocal designpoints.
Coordinatesof thedesignpoints
inthestandardGaussianspace
Global solution Fakesolution
=1.43 =2.15
Component P
F
=0.076 P
F
=0.016
Q 0.82 0.97
H 0.05 0.39
e
clay
0.04 0.32
e
sand
0.14 0.16
G
clay
s
0.10 0.12
G
sand
s
0.09 0.10
OCR 1.09 0.00
C
c
0.35 1.59
0.18 0.92
FORM, while the second solution of P
F
=0.011
follows fromthe fake solution of FORM. Note that
the first SORM solution is fairly close to the MCS
solution 0.064, indicating SORM indeed improves
FORMforthiscasestudyintheglobal solutioncanbe
found. However, SORM suffers fromthesameissue
of multiplelocal design points becauseSORM uses
theresultsfromFORM.
TheP
F
estimatemadebySubsimisquiteaccurate
althoughtherequiredcomputationismorethanthose
for FOSM, FORM andSORM.
2 ISSUE OF MULTIPLE DESIGNPOINTS
This section discusses in detail theissueof multiple
local solutions(designpoints). Thisissueonlyaffects
theanalysis results of FORM andSORM; MCS and
Subsimarerobust against thisissue.
2.1 Multiple solutions
Table3liststhecoordinatesof thetwolocal solutions
obtained in the GP algorithm. After verification, it
isfoundthat bothsolutionssatisfy thefollowingtwo
necessaryconditionsfor designpoint:
a. Thesolutionshouldresideright onthelimit-state
lineg =0.
b. InthestandardGaussianspace, thegradientvector
of theperformancefunctionevaluatedat thesolu-
tionshouldbeparallel tothesolutionvector itself,
i.e. thesolutionisthepointontheg =0linethatis
locallyclosest totheorigin.
Moreover, fromthe numerical values of the coordi-
nates of thetwo solutions, it is not trivial to identify
which solution is global or fake. As a consequence,
for theinstancethat asinglerunof theGP algorithm
converges to thefakesolution, thereseems to beno
viablewayat handtodetect thefalsity.
Figure 2. The contour lines of the performance function
aroundthetwosolutions(upper: global solution; lower: fake
solution).
2.2 Non-differentiability of performance function
Figure 2 is employed to demonstrate the geometry
around these two local design points. The first plot
showsthecontour linesof theperformancefunctiong
inthestandardGaussianspaceof q andOCR, where
all other sevenuncertainparametersarefixedat their
solutioncoordinatesof theglobal solution.Thesecond
plot isfor thefakesolution. Themarker x indicates
thelocationsof thetwolocal designpoints. Thedark
dashedlinesrepresentthenon-differentiableboundary
for
p
.
It is clear that the existence of two local
design points has something to do with the non-
differentiability of the performance function. More-
over, thefailureregioninthestandardGaussianspace
seemstobetheunionof failureregionsdefinedbythe
followingtwoperformancefunctions:
71
Figure3. Thefailureregionsdefinedbyg
1
andg
2
function
aroundthesecondlocal designpoint.
and
Let ustaketheregionaroundthesecondlocal design
point (thelower plot inFigure2) asanexample. The
region in thelower plot in Figure2 satisfying g -0
(failure) resides at the right hand side of the g =0
contour line, which is clearly the union of the two
failure regions defined by g
1
-0 and g
2
-0 shown
in Figure 3. In other words, the consolidation prob-
lemis similar to anin-series systemwhosefailureis
definedbythedisjunctionof thetwofailureevents, i.e.
failureoccurs either thefailureof g
1
or failureof g
2
occurs. Itisinterestingtoseethattheoriginal problem
in(2) is not in-series, but it behaves likeanin-series
systemduetonon-differentiabilityof theperformance
function.
2.3 Other design point algorithms
Apart fromtheGP algorithm, another twoalgorithms
for FORM of finding the design point are exam-
ined, includingthefmincon.mfunctioninmatlaband
thesimplealgorithminSection6.2.4, p. 361inAng
andTang (1984). Table 4 summarizes the results of
convergence.
All algorithms may converge to the fake solution,
depending on the location of the initial trial point.
Whentheinitial trial pointisrandomlygeneratedfrom
the standard Gaussian distribution, there is always
larger chancetoconvergetothefakesolution. When
the initial trial point is taken to be the origin, only
thematlabalgorithmconvergestotheglobal solution:
it is likely that this only happens by luck. Further-
more, thereis somechancefor non-convergence, i.e.
thealgorithmnever reachesastationarypoint.
Table4. Convergencefor thethreealgorithms.
Chanceof convergence
Convergence
Fake Global No if startingfrom
Algorithm solution solution converge theorigin
GP 78% 15% 7% Fakesolution
Matlab 70% 28% 2% Global solution
Ang and 83% 17% 0% Fakesolution
Tang
Table5. Analysisresultsfor variouschoicesof j
q
.
(P
F
estimatefromFORM)
MCS P
F
P
F
j
q
Fake Global No P
F
2-order point
kN/m
2
solut. solute. converge (n =10
6
) bound est.
5 0% 23% 77% 0 [6.2e-9 6.2e-9
(n/a) (6.2e-9) 6.2e-9]
10 0% 66% 34% 5.6e-5 [6.1e-5 6.1e-5
(n/a) (6.1e-5) 6.1e-5]
20 78% 15% 7% 0.064 [0.083 0.085
(0.016) (0.076) 0.087]
30 0% 91% 9% 0.37 [0.36 0.39
(n/a) (0.35) 0.40]
50 0% 100% 0% 0.88 [0.91 0.95
(n/a) (0.85) 1.00]
3 SENSITIVITY ANALYSIS
It is instructive to understand how the existence of
multiplelocal designpointsfor FORM/SORM would
beaffectedbythechangeof theinput parametersand
distributions.Again, thisissueof multiplelocal design
pointsonly affectsFORM/SORM; MCS andSubsim
arerobust against thisissue.
3.1 Sensitivity over mean value of q
Table 5 shows the analysis results by using the GP
algorithmfor variouschoicesof themeanvalueof q,
denotedbyj
q
, whiletheinputvaluesanddistributions
of other variablesareidentical tothoseinTable1. The
chanceof non-convergenceisbasedonthecasewhere
theinitial trial pointisrandomlygeneratedbythestan-
dardGaussiandistribution. Itisclearthattheexistence
of multiple solutions only happens for j
q
near 20.
Moreover, the chance of non-convergence increases
asthefailureprobabilitybecomessmaller. Compared
withtheMCSresults, theP
F
estimatedfromtheglobal
solutionof FORM isreasonably accurate, althougha
biasisnoticeable.
3.2 Sensitivity over range of OCR
algorithmfor various choices of the range of OCR,
72
Table6. Analysisresultsfor variouschoicesof OCR range.
(P
F
estimatefromFORM)
MCS P
F
P
F
OCR Fake Global No P
F
2-order point
range solut. solute. converge (n =10
6
) bound est.
010 76% 16% 8% 0.16 [0.16 0.16
(0.016) (0.15) 0.16]
01.5 0% 100% 0% 0.95 [0.92 0.92
(n/a) (0.94) 0.92]
1.52.5 78% 15% 7% 0.064 [0.083 0.085
(0.016) (0.076) 0.087]
2.55 0% 98% 2% 0.011 [0.016 0.016
(n/a) (0.016) 0.016]
510 0% 93% 7% 0.011 [0.016 0.016
(n/a) (0.016) 0.016]
Table7. Analysisresultsfor variouschoicesof S
allow
.
(P
F
estimatefromFORM)
MCS P
F
P
F
S
allow
Fake Global No P
F
2-order point
mm solut. solute. converge (n =10
6
) bound est.
1 0% 100% 0% 0.99 [0.99 0.99
(n/a) (0.98) 1.00]
3 10% 89% 1% 0.28 [0.29 0.33
(0.15) (0.24) 0.34]
5 78% 15% 7% 0.064 [0.083 0.085
(0.016) (0.076) 0.087]
10 19% 39% 42% 4.9e-3 [8.4e-3 8.4e-3
(8.2e-6) (8.4e-3) 8.4e-3]
15 0% 29% 71% 3.7e-4 [7.2e-4 7.2e-4
(n/a) (7.2e-4) 7.2e-4]
while the input values and distributions of other
variables areidentical to thoseinTable1. It is clear
that the existence of multiple solutions happens for
the range of 010 and 1.52.5. The chance of non-
convergenceis reasonably small regardless theOCR
range. Compared with theMCS results, theP
F
esti-
matedfromtheglobal solutionof FORMisreasonably
accuratedespiteacertainamount of bias.
3.3 Sensitivity over S
allow
algorithmfor variouschoicesof S
allow
, whiletheinput
valuesanddistributionsof othervariablesareidentical
to thoseinTable1. It is clear that multiplesolutions
occur for most cases. Thechanceof non-convergence
increases as thefailureprobability gets small. Com-
pared with the MCS results, the P
F
estimated from
theglobal solution of FORM is reasonably accurate
despiteacertainamount of bias.
4 POSSIBLE REMEDY FOR FORM/SORM
Asmentionedearlier, thenon-in-seriessystemdefined
in(2) behavessimilarlyasthein-seriessystem.There-
fore, apossiblesolutiontoresolvetheissueof multiple
local solutions of FORM/SORM is to replace the
original systemby the in-series system. For an in-
series systemwith two failure modes of g
1
and g
2
described in (7) and (8), two methods can betaken:
(a) thesecond-order bound(AngandTang1984) and
(b) point estimate(Mendell and Elston 1974; Phoon
2008).Thesetwomethodsarebrieflyreviewedherein.
Outside the FORM/SORM framework, the issue of
multiplelocal designpointscanbeeasily handledby
adoptingeither MCSor Subsim.
4.1 Second-order bound
Thesecond-order boundsimplysaysthat
where P
F
is the system failure probability;
P
i
=+(||z
i
||) and z
i
is the design point for the
i-th failuremode: e.g. z
1
is thedesign point if g
1
in
(7) is theonly performancefunction(similar for z
2
),
sotheprocessof findingz
1
or z
2
will not suffer from
theissueof multiplesolutions; P
+
21
=P(B
1
)+P(B
2
);
P
21
= max[P(B
1
), P(B
2
)]; and
and
where is the angle between the two design point
vectorsz
1
andz
2
.
Theresultingupper andlower boundsfor thecases
studiedintheprevioussectionarelistedinTables57.
Although the issue of multiple solutions is resolved
(because the process of finding z
1
or z
2
does not
involvemultiplesolutions), thebiasof theboundscan
beobserved, i.e. someMCS results fall outsides the
bounds.
4.2 Point estimate
Thesecond-orderboundsdonotofferasingleestimate
of thesystemfailureprobability. Thefollowingpoint
estimatecanmitigatethisissue:
whereP
F
isthesystemfailureprobability;
73
isanestimateof theprobabilitythat g
1
andg
2
failure
eventshappensimultaneously, and
The resulting point estimates are also listed in
Tables 57. Again, although the issue of multiple
solutions is resolved, thebias of thepoint estimates
isevident.
5 RECOMMENDEDRELIABILITY
METHODS
Fortheconsolidationproblem,FOSMdoesnotprovide
aconsistentreliabilityestimate. Moreover, thefollow-
ing two issues exist in FORM/SORM: (a) thepossi-
bility of findinganon-global (fake) designpoint and
(b) the possibility of non-convergence. The second-
order bound and point estimatemethods seemto be
able to resolve the first issue although they cannot
removethebias in thereliability estimate. However,
they cannot resolvethesecond issue. Therefore, the
useof thelatter two methods is recommendedat the
conditionthat thesecondissuedoesnot exist andthe
amount of biasisacceptable.
MonteCarlosimulationandSubset simulationare
completelyrobustagainsttheexistenceof local design
points. Therefore, their use is recommended at the
expenseof morecomputation.
6 CONCLUSION
The reliability analysis of a one-dimensional con-
solidation is presented. This simple geotechni-
cal problemis challenging becausetheperformance
function is not differentiable. It is found that this
non-differentiability induces abehavior similar to an
in-series systemalthough theconsolidation problem
isclearlynot anin-seriessystem.
Several reliability methods are implemented to
investigatetheir feasibilityandconsistency. TheFirst-
Order Second-Moment (FOSM) method provides
inconsistent reliability estimates. For First-Order and
Second-Order Reliability Methods (FORM/SORM),
three algorithms of finding design points are
examined, and all of themshow possibility of con-
verging to alocal design point that is different from
the global one, hence giving inconsistent reliability
estimate. Sometimesthey donot convergeat all. The
second-order bound and point estimatemethods can
resolve the issue of multiple local design points for
FORM/SORM. Monte Carlo simulation (MCS) and
Subset simulation (Subsim) are robust and always
provideconsistent reliabilityestimates.
Itisrecommendedthatthesecond-orderboundand
point estimate methods can be used if convergence
is not an issue and the amount of bias in reliability
estimate is acceptable. MCS and Subsimare highly
recommended because they seemcompletely robust
against theexistenceof local designpoints.
REFERENCES
Ang, A. H.-S., & Tang, W. H. 1984. Probability Concepts
in Engineering Planning and Design, Vol. II, Wiley, New
York.
Der Kiureghian, A., & Stefano, M. D. 1991. Efficient algo-
rithmfor second-order reliabilityanalysis. ASCE Journal
of Engineering Mechanics, 117(12), 29042923.
Liu P. L., & Der Kiureghian, A. 1991. Optimization algo-
rithms for structural reliability. Structural Safety, 191,
9(3), 161177.
Mendell N. R. &ElstonR. C. 1974. Multifactorial qualitative
traits: geneticanalysisandpredictionof recurrencerisks.
Biometrics, 30, 4157.
Phoon, K. K. 2008. Reliability-Based Design in Geotechnical
Engineering: Computations and Applications, Taylor &
Francis.
74
Studyondeterminationof partial factorsfor geotechnical
structuredesign
T.C. KieuLe&Y. Honjo
ABSTRACT: Someof theaspects concerning design verification equations in geotechnical design aredis-
cussed,suchastheformatof verificationformula,andtheeffectivenessof theBaselineApproachindetermination
of characteristicvalues.
TheDesignValueMethod(DVM) combinedwithFirst-Order ReliabilityMethod(FORM) analysishasbeen
usedasastandardproceduretodeterminepartial factorsinthepast. However, themethodencountersdifficulties
whenappliedtodeterminepartial factorsof someof thegeotechnical structures: (i) Someof thebasicvariables
areincludedbothinresistanceandload; (ii) Highnon-linear natureof theperformancefunctions. Inorder to
overcomethedifficultiesincodecalibrationby thetraditional method, aprocedure, namedasMcDeva, which
keeps philosophy of theDVM, anduses effectiveness of Markov ChainMonteCarlo Simulation(MCMC), is
proposedtodetermineloadandresistancefactors.
Themethod is applied to determinateload and resistancefactors in retaining wall design to illustratethe
feature.
1 INTRODUCTION
Theprimary purposeof structural design codes is a
simple, safeandeconomically efficient basis for the
design of ordinary structures under loading, opera-
tional andenvironmental conditions. Therearemany
sourcesof uncertaintiesincludingphysical, statistical
and modeling uncertainties that may impact on the
performanceof thestructure.
In traditional design practice, for exampleAllow-
ableStressDesign(ASD), all uncertaintiesareembed-
ded within a factor of safety. This single factor of
safety depends onthepast experiences anddoes not
giverational basisfor thetreatmentof uncertaintiesin
design.
As opposedtoASD, theLevel 1 Reliability-Based
Design(RBD) ismoreflexibleandrational becauseit
providesconsistent levelsof safetyover varioustypes
of structural components by applying many factors
tocharacteristic values intheverificationformulato
treattheuncertaintiesinvolvedinthestructural design.
InLevel 1 RBD, thedesignverificationformulafor-
mats are classified into two categories: (i) Material
Factor Approach (MFA), (ii) Load and Resistance
Factor Design(LRFD).
Thebasicphilosophyof MFA istotreat theuncer-
taintiesat their origin. However, it isvery difficult to
evaluatethereliabilityof structurebyreliabilityanal-
ysis basedonsuperpositionof all uncertainsources.
InLRFD, theresistanceandtheload(external action)
arefirst calculatedbasedonthecharacteristic values
of thebasicvariables.Thenloadandresistancefactors
are applied to resulting resistance and load compo-
nents to assure appropriate safety margin. LRFD is
consideredtobebetter format thanMFA ingeotech-
nical engineering design codeat least in thepresent
situation, duetothesomeof thereasonsasfollows:
Especially for the resistance side, the design cal-
culation based on LRFD predicts the most likely
behaviorof thestructuretoitslaststageof design. It
isincoincidencewiththephilosophythata designer
should keep track of the most likely behavior of the
structure to the last stage of the design as much as
possible.
In geotechnical design, where the interaction
between a structure and ground is so high, it is
not possible to know whether the discount of the
material parameter valuemay result safedesignof
astructure. It is especially truefor asophisticated
designcalculationmethodlikeFEM.
Another important issue in Level 1 RBD is the
choiceof characteristic value. It isvery conventional
tochoosecharacteristicvaluesof load, andresistance
as very high/low fractile values, such as 95%/5%
based on so called Baseline Approach (e.g. Phoon
et al., 2003). In order to check the effectiveness of
thisapproach, theauthorshavecarriedout somecal-
culations and realize that the advantage of Baseline
Approach is much more obvious in determining the
loadfactor comparingtothat of resistancefactor. It is
recommendedtotakehighfractilevalueforcharacter-
isticvalueof load, however thelowfractilevalueneed
not tobechosenascharacteristicvalueof resistance.
75
Theprocedureof determiningloadandresistance
factorsistermedcodecalibrationprocess.TheDesign
ValueMethodcombinedwithFirst-Order Reliability
Method(FORM) analysisisusedasastandardproce-
dure to determine partial factors of basic variables,
mainly due to the fact that they are computation-
ally inexpensive and the results are of acceptable
accuracies.
Themethod, however, encountersdifficultieswhen
applied to determine partial factors of some of the
geotechnical structures: (i) Some of the basic vari-
ablesareincludedbothinresistanceandload; (ii)High
non-linear natureof theperformancefunctions.
To overcome the above-mentioned difficulties, in
this paper, amethod, which keeps philosophy of the
DesignValueMethod, andusingMonteCarloSimula-
tion(MCS), isproposedtodetermineloadandresis-
tancefactors. Designof agravityretainingwall under
sliding failure mode is introduced as an illustrative
exampleof themethod.
2 DETERMINATIONOF LOADAND
RESISTANCE FACTORS
2.1 Design Value Method
The simplest formof a limit state function is RS,
where the load, S, and resistance, R, are statisti-
cally normally distributedindependent variables, i.e.
RN(j
R
,
2
R
) andS N(j
s
,
2
S
).
whereM alsofollowsnormal distributionN(j
M
,
2
M
),
wherej
M
=j
R
j
S
and
M
=
2
R
+
2
S
. Therelia-
bilityindex isgivenas:
Theprobabilityof failurep
f
maybedefinedby:
where+(.) is thestandard normal probability distri-
butionfunction.
Onthefailuresurface, thedesignpointD(Z
R
, Z
S
)
isthepoint whichisat theshortest distancefromthe
origininthestandardnormal spaceinFigure1. The
design point is thepoint, which maximizes thejoint
probabilitydensityfunctiononthefailuresurfacefora
givenproblem.Thisconceptwill beusedtodetermine
thedesignpointutilizingthesamplesgeneratedduring
theSubset MonteCarlosimulation.
For a performance function given in Eq. (1), it
is required in design that
T
should always be
satisfied, where
T
isatarget reliabilityindex. Thus,
Figure1. Definitionof designpoint.
Ontheother hand,
where
R
and
S
arecalledthesensitivityfactorsof R
andS, whichisdefinedbythefollowingequations.
Then, thefollowingequationcanbeobtained:
Let characteristic values of R and S be R and S,
respectively, the load and resistance factors which
can fulfill the target reliability index,
T
, can be
determinedasfollows:
whereV
R
andV
S
arethecoefficient of variationof R
andS, respectively.
Theresults areexact only whenall thebasic vari-
ables arenormal, andtheperformancefunctioncon-
sists of linear combination of thesevariables. When
basicvariablesfollowlognormal distributions, thefol-
lowing approximation is used (e.g., Ditlevsen and
Madsen, 1996):
where =ln(1+V
2
) isthestandarddeviationof log-
normal distribution, sensitivityfactorsaredefinedas:
76
2.2 Subset Markov Chain Monte Carlo simulation
SubsetMarkovChainMonteCarlo(MCMC) method,
proposedbyAuandBeck (2003), is oneof themost
effective and stable methods to carry out reliability
analyses by MCS. The basic idea of this method is
that asmall failureprobability can beindicated as a
product of conditional probabilitiesof someinterme-
diateevents.Thus, asimulationproblemof arareevent
is convertedinto theproblemof asequenceof more
frequent events.
Letthetotal regionandthefailureregionbedenoted
by F
0
and F, respectively. The subsets of F
0
are
F
i
(i =1,m) that satisfiedtherelationship:
ThefailureprobabilityP(F)canbecalculatedbased
on the conditional probability of these intermediate
subsetsasfollows:
2.2.1 Samples generation by Metropolis-Hastings
algorithm
In order to calculate P(F
i
), the samples are gener-
atedbyMCSfollowingaspecifiedprobabilitydensity
function (PDF). For the calculation of the failure
probability of the intermediate subsets, a procedure
of samples generation will be carried out for any
PDF defined on intermediate subset F
i
by using
an MCMC algorithmknown as Metropolis-Hastings
(M-H) algorithm.
Let thecurrent Markov chainstate, or theseedfor
thenextstate, isx
(k)
.TheM-Halgorithmgeneratesthe
following series of samples x
(k+1)
that follows given
PDF (.) bytheprocedurebelow:
Step 1: Usingthecurrent statex
(k)
, generateacan-
didatestatex
fromaproposal distribution q(x
|x
(k)
).
Also, generateasampleu fromuniformdistribution
U(0,1).
Step 2: Calculateanacceptanceprobability as
Step 3: x
(k+1)
isgeneratedbasedon as:
Step 4: Repeat steps(1) to(3) for necessarycycles
toobtaintherequirednumber of samples.
In M-H algorithm, thegenerated samples may be
affectedby: (i) thechoiceof theproposal densityand
(ii) theburn-in period or thenumber of runsrequired
for thechaintoapproachitsstationarydensitysothat
candidatestatesfollowthetarget distribution.
2.2.2 Subset MCMC algorithm
Suppose the limit state function g(X) composed of
n basic variables X
i
, i =1,,n. Basedonthesubsets
andMCMC, thefailureprobability canbecalculated
bythefollowingprocedure:
Step 1: k =1, N
t
samples for each basic variables
X
j
aregeneratedbasedongivenPDF of X
j
. Thenthe
performance function g(x) is also calculated for N
t
number of generatedx (x ={x
1
, . . . x
n
}).
Step 2: Theintermediatefailureevent F
k
is deter-
mined. A number of samplesN
s
that arecloser tothe
limitstatewill bechosenasseedsfor usinginthenext
step.
Step 3: k =k +1. For each of N
s
seeds of each
basicvariable, generateN
t
,N
s
samplesbyusingM-H
algorithm. Thus, thesubset F
k+1
isdefined.
whereg
Ns
istheN
s
th lowestvalue(N
s
-N
t
)of theper-
formancefunctionobtainedusingx
(k)
whenarranged
intheascendingorder asg
1
-g
2
--g
Nt
.
Step 4: Evaluate limit state function g(x
(k+1)
) at
eachconsecutivestateof thebasicvariables. If x
(k+1)
lie in F
k
, accept the state, while if they lie in F
k1
,
reject thestateandtakex
(k+1)
=x
(k)
.
Step 5: Repeat steps2to4for appropriatenumber
of times.
Step 6: Thefailureprobability of thetarget failure
event P
f
isestimatedbythefollowingequation:
whereN
fail
is thenumber of samples fallinginto the
failureregion.
A numberof variationsintheimplementationof the
above mentioned algorithmfor Subset simulation is
possible. For example, thechoiceof thetotal number
of samples N
t
the definition of intermediate failure
eventsF
k
(k =1,,m1) andthestoppingcriteria.
2.3 Procedure to determine load and resistance
factors
In reliability analysis of structural design, the resis-
tancecomponent, R, andtheloadcomponent, S, may
containseveral basicvariablesX
i
(i =1n), i.e. R(X)
and S(X). Some of the basic variables may appear
in both resistance and load components of the limit
state function. This leads to a troublesome situation
thatthedesignvaluemethodusingFORMmaynotbe
applicable.
Asanalternativeapproach, aprocedure, referredto
asMcDeva (MonteCarlosimulationbasedonDesign
ValueMethod) hereafter, todirectlydetermine
R
and
S
for R and S, in place of basic variables X
i
, is
proposed.
Step 1: Definebasic variables X
i
(i =1,,m) and
their PDFs.
77
Figure2. Concept of thesubset MCMC method.
Step 2: Carry out subset MCMC in a number of
runs to obtain failure probability and many sets of
randomsamplesof basic variablesX
i
(i =1,,n). In
addition, by using thesamples generated in thefirst
step of MCMC procedure, it is possible to estimate
thejoint densityfunctionf
R,S
(R, S).
Step 3: At the final step, it is expected that most
of the points (or samples) are located closely to the
limit stateline(seeFigure2). A point whichgivesthe
maximumlikelihoodfor thegivenPDFsisconsidered
toapproximatethedesignpoint.
Step 4:Withthespecifieddesignpoint(R
, S
), one
canevaluatethesensitivityfactors.
Thesensitivityfactorscanbeobtainedasfollows:
where, (Z
R
, Z
S
) is the estimated design point
in normalized space, i.e. Z
R
=(R
j
R
),
R
and
Z
S
=(S
j
S
),
S
.
Another simple but less accurate way to obtain
sensitivity factors is shown in Eq. (19), using the
probabilisticparametersof R andS obtainedinStep 2.
The sensitivity factors obtained fromEq. (18) may
bemoreaccuratethan thoseobtained fromEq. (19)
becauseEq. (18) isbasedonthemaximumlikelihood
point, whereas Eq. (19) is just simply based on the
samples generatedinthefirst stepof subset MCMC
procedure.
Althoughthestructurethathasbeenevaluatedmay
not haveexactly thesametarget reliability level, one
canobtainloadandresistancefactorsthat correspond
to the target reliability index by using the equation
developed for thedesign valuemethod, i.e. Eqs. (7)
and(8).
If it is judged more appropriate to fit lognormal
distribution, the sensitivity factors can be estimated
bytwowaysshowninEqs. (20) and(21) asfollows:
where, (Z
R
, Z
S
) is the estimated design point in
normalized space, i.e. Z
R
=(lnR
j
lnR
),
lnR
and
Z
S
=(lnS
lnj
S
),ln
S
.
And Eqs. (8) and (9) should beused to estimatethe
loadandresistancefactors.
Therefore, the method proposed in this study is
combination of MCS and the design value method.
Theformer isaveryflexiblemethodtoevaluaterelia-
bilityof thestructureunderconsideration, whereasthe
advantageof thelater isthatoneneedsnottoredesign
the structure during code calibration as long as the
safetylevel of theinitiallyset structureisnot veryfar
fromthetarget reliabilitylevel.
Inorder tocheck theresultsobtainedby McDeva,
another procedure, basedonacombinationof OMC
(OrdinaryMonteCarlosimulation) andDVM, isalso
attemptedasfollows:
Step 1: Definebasicvariablesandtheir PDFs.
Step 2: Carry out OMC. TheOMC is carried out
based on the probability distributions of basic ran-
domvariablesX
i
(i =1,,n). A number of N
t
samples
aregenerated. p
f
may beestimatedastheratioof N
f
(numberof samplesfall intothefailureregion) andN
t
.
Step 3: Definetheapproximatedesignpoint X
Step 4: Calculate
R
,
S
and
R
,
S
.
3 EXAMPLES
3.1 Linear limit state function Z =RS
R and S areindependently normally distributed ran-
domvariables, i.e. R N(7.0,1.0)andS N(3.0,1.0),
where true reliability index
True
=2.83 and true
designpoint is(5.0,5.0). Anexampleof thegenerated
samplesisshowninFigure3.
The results obtained by Subset MCMC are as
follows:
Probabilityof failure: p
f
=0.2610
2
Reliabilityindex: =2.79
Thesamplesof thelaststepareusedtoestimatethe
designpoint andnecessaryfactorsasshownbelow.
Designpoint: D(R
,S
) =(5.1,5.0)
Sensitivityfactors:
R
=0.69
S
=0.73
Partial safetyfactors:
R
=0.73
S
=1.08
The true design point in this problem is
(R
, S
)=(5.0,5.0), which is very closeto thevalue

78
Figure3. Generatedsamplesfor simplelinear performance
function.
Figure4. Descriptionof retainingwall.
obtained.Thetruesensitivityfactors, i.e.
R
=0.707
and
S
=0.707arealsoveryclose.Themethodseems
toworkwell inthissimpleexample.
3.2 A tetaining wall example
3.2.1 Determination of load and resistance factors
McDeva is now applied for reliability analysis of a
gravity retainingwall (Figure4) under slidingfailure
mode (Orr, 2005). The necessary parameters of the
retainingwall, soil belowandfill abovetheretaining
wall aredescribedbelow.
Propertiesof sandbeneathwall: c
s
=0,
s
,
s
Propertiesof fill behindwall: c
f
=0,
f
,
f
Groundwater level isat depthbelowthebaseof the
wall
Friction angle between base wall and underlain
sand:
bs
Thicknessof retainingwall: w=0.4m
Thelimit statefunctionof theproblemisgivenas:
whereR: total horizontal forcethat resists horizontal
movement of the wall, S: total horizontal force that
movesthewall.
Table1. Probabilisticparametersof basicvariablesinlimit
statefunctionof retainingwall under slidingfailuremode.
Variables Distribution j COV
X
1
=
f
Lognormal 20 1.000 0.05 2.994 0.050
X
2
=
s
Lognormal 19 0.950 0.05 2.943 0.050
X
3
=
c
Lognormal 25 1.250 0.05 3.218 0.050
X
4
=tan
f
Lognormal 0.781 0.107 0.14 0.256 0.136
X
5
=tan
s
Lognormal 0.675 0.086 0.13 0.402 0.127
X
6
=tan
bs
Lognormal 0.577 0.070 0.12 0.557 0.120
X
7
=q Lognormal 15 1.500 0.10 2.703 0.100
Theactual formof R andS aregivenasfollows:
where, tan(45
f
,2)=
1+(tan
f
)
2
tan
f
tan(45
s
,2)=
1+(tan
s
)
2
+tan
s
,
f
,
c
, tan
s
, tan
f
, tan
bs
andq areconsidered
asbasicvariablesof thelimit statefunctionandsome
of themareincludedbothinR andS.Theprobabilistic
parametersof basicvariablesareshowninTable1.
Subset MCMC simulationisthencarriedout 1000
runs to obtain p
f
of thegravity retaining wall under
slidingfailuremode.Thenumberof samplesgenerated
foreachvariableinastepisN
t
=100.Inaddition,from
thegeneratedsamplesof basicvariablesduringsubset
MCMC simulation, seriesof numbersof R andS are
also calculated. Thereafter, theseR and S values are
utilizedfor determinationof partial safetyfactors.
The estimated p
f
by 1000 simulation resulted
the mean value of 0.2710
2
(i.e. =2.70), and
COV(p
f
)=1.40.
Designpoint:
f
=20.2
s
=19.1
c
=24.1
tan
f
=0.588 tan
s
=0.624
tan
bs
=0.472
q =14.9
Thisapproximateddesignpoint isequivalent topoint
(R=169.1,S=166.8).
N
t
points (R, S) calculated from N
t
points
X
i
(i =1,,n), which aregenerated in thesubset F
0
,
areutilized to estimatemean and standard deviation
of R andS.
j
R
=201.1
R
=21.8
j
S
=120.8
S
=20.2
Inaddition, basedonfittingtest onthegenerated
points, the density function f
R,S
(R, S) seems to be
79
Table2. Results obtained for
T
=3.1, sensitivity factors
areobtainedbasedontheestimateddesignpoint.
ByMcDeva ByOMC
Normal Lognormal Normal Lognormal
R S R S R S R S
0.54 0.84 0.61 0.79 0.57 0.82 0.64 0.77
0.62 1.76 0.61 1.83 0.61 1.79 0.61 1.85
Table3. Results obtained for
T
=3.1, sensitivity factors
areobtainedbasedonf
R,S
(R, S).
ByMcDeva ByOMC
R S R S R S R S
0.73 0.68 0.54 0.84 0.75 0.67 0.56 0.83
0.57 1.66 0.63 1.87 0.57 1.69 0.63 1.91
best-fittedtojointNormal andLognormal distribution
functions.
Finally, sensitivityfactors, loadandresistancefac-
torsareobtainedfor atargetreliabilityindexassumed
tobe
T
=3.1.Thejointdensityfunctionf
R,S
(R, S) are
assumedtobejointnormal andlognormal distribution.
TheobtainedresultsareshowninTables2and3.
Notethatthecharacteristicvalueof RandS adopted
inthisexampleiscalculatedfromcharacteristicvalues
of basicvariableswhicharechosenasthemeanvalue
for X
1
toX
6
, while95%of fractilevaluefor X
7
.
Therearesomeremarksontheresultsasfollows:
LoadandResistanceFactorsobtainedall thecases,
i.e. different joint density functions, different way
tocalculatesensitivityfactors, areverysimilar.
Theresultsobtainedbasedontheestimateddesign
point (Table2) is better thanthoseobtainedbased
on only thesamples generated in thefirst step of
subset MCMC procedure(Table3), asexpected.
OMC simulationhasbeencarriedout for 100,000
samples and the results obtained by OMC are also
showninTables2and3, showingthatresultsobtained
bythetwoproceduresareverysimilar.
3.2.2 Some reflections of Subset MCMC and
verification of the robustness of McDeva
In this example, the true failure probability p
fTrue
is
not available. Therefore, OMC simulation has been
carriedoutwiththetotal generatedrandomsamplesof
N
t
=1,000,000toobtainthefailureprobabilitywhich
is 0.28910
2
. This valuewas assumed as thetrue
valuehere.
McDeva has also been carried out for different
values of N
t
, N
s
, and N
f
, where N
t
is total samples
generatedfor eachsubset, N
s
isnumber of seedsused
Table4. Mean(p
f
) for N
t
=100.
N
f
,N
t
N
s
,N
t
0.02 0.05 0.1 0.2 0.3 0.4 0.5
0.02 0.0030 0.0030 0.0027 0.0020 0.0017 0.0012 0.0009
0.04 0.0029 0.0028 0.0025 0.0017 0.0014 0.0012 0.0011
0.10 0.0033 0.0029 0.0027 0.0026 0.0023 0.0020 0.0018
0.20 0.0035 0.0031 0.0026 0.0026 0.0025 0.0023 0.0022
0.50 0.0040 0.0033 0.0028 0.0026 0.0024 0.0025 0.0024
Table5. COV(p
f
) for N
t
=100.
N
f
,N
t
N
s
,N
t
0.02 0.05 0.1 0.2 0.3 0.4 0.5
0.02 1.17 1.08 1.21 1.68 2.24 2.71 3.06
0.04 1.21 0.96 1.14 1.32 1.19 0.88 0.56
0.10 1.22 1.04 1.07 1.05 1.12 1.28 1.35
0.20 1.20 1.15 0.95 0.89 0.92 0.92 0.97
0.50 1.07 1.01 0.86 0.77 0.77 0.73 0.73
Table6. log(p
f
)/log(p
fTrue
) for N
t
=100.
N
f
,N
t
N
s
,N
t
0.02 0.05 0.1 0.2 0.3 0.4 0.5
0.02 0.96 0.96 0.98 1.02 1.06 1.12 1.16
0.04 0.96 0.97 0.99 1.05 1.09 1.11 1.12
0.10 0.94 0.97 0.98 0.99 1.00 1.02 1.04
0.20 0.93 0.96 0.98 0.98 0.99 1.00 1.01
0.50 0.91 0.95 0.97 0.98 0.99 0.99 0.99
for generatingasubset, andN
f
istheminimumnum-
ber of samplesfallinginthefailureregionsothat the
SubsetMCMCalgorithmwill besatisfiedandstopped
or cut-off criteria.
N
t
, N
s
, andN
f
areset asfollows:
N
t
=50, 100, 150
N
s
,N
t
=0.02, 0.04, 0.10, 0.20, 0.50
N
f
,N
t
=0.02, 0.05, 0.10, 0.20, 0.30, 0.40, 0.50
For eachcombinationof theaboveN
t
, N
s
andN
f
,
McDevahasbeencarriedout 1000runs. For thecase
N
t
=100, tables 4to 6showmean(p
f
), COV(p
f
) and
thebiasof failureprobabilitycomparingtotruefailure
probabilityinlogscale, i.e. Log(meanp
f
)/Log(p
fTrue
).
The mean value of p
f
for all N
t
cases (i.e. 50,
100, 150) rangedin(0.100.47)10
2
, whereasits
COV rangedbetween0.903.36. Most of thecases,
thebiaslog(p
f
)/log(p
fTrue
)=1.00.05, whichmaybe
consideredacceptable.
To define the optimum combinations of input
parameters,thefollowingfactorshavebeenconsidered:
The number of times that the performance func-
tion is called, Zcall. Smaller Zcall means less
computational time.
80
Table7. Zcall for N
t
=100.
N
f
,N
t
N
s
,N
t
0.02 0.05 0.1 0.2 0.3 0.4 0.5
0.02 158.5 166.0 175.5 188.4 194.2 197.9 199.3
0.04 168.6 179.4 192.1 202.3 207.5 211.8 213.7
0.10 188.2 206.5 220.3 236.7 244.3 251.1 254.4
0.20 214.8 244.2 264.6 286.8 299.4 307.2 313.0
0.50 308.8 393.4 451.7 503.1 535.6 552.4 571.5
Table8. TotP for N
t
=100.
N
f
,N
t
N
s
,N
t
0.02 0.05 0.1 0.2 0.3 0.4 0.5
0.02 111.8 127.6 148.2 176.0 189.5 197.6 200.5
0.04 127.7 150.3 177.2 199.6 211.1 219.5 223.8
0.10 158.1 193.6 221.9 256.8 272.6 287.9 295.9
0.20 199.5 255.3 295.2 341.6 367.5 384.8 396.7
0.50 345.5 496.3 607.2 708.1 774.4 808.7 847.6
Table9. RejP for N
t
=100.
N
f
,N
t
N
s
,N
t
0.02 0.05 0.1 0.2 0.3 0.4 0.5
0.02 83.5 96.4 114.4 137.9 149.1 155.9 158.4
0.04 93.4 111.6 134.0 152.9 163.1 170.1 173.7
0.10 110.7 138.2 160.8 189.5 202.9 215.9 222.8
0.20 133.5 175.1 206.0 242.7 263.8 278.5 288.3
0.50 206.9 310.1 392.6 470.6 523.3 550.1 581.3
The number of total points generated in Subset
MCMC, TotP. Smaller TotP means less computa-
tional time.
The number of points that are rejected by Sub-
set MCMC fromTotP points, RejP. Theless RejP
meansthemoreeffectiveMcDevaprocedure.
Theresultsshowthat Zcall andTotP increasewith
N
t
, N
s
and N
f
. When N
s
,N
t
=0.5, ToP and RejP
become very large. COVs of TotP, and RejP seem
todecreasewhenN
t
increases. Meanvaluesof Zcall,
TotP, andRejP after Nrun =1000runs for N
t
=100
areshowninTables7to9, respectively.
Byconsideringthebiasof failureprobabilitycom-
paringtotruefailureprobabilityvariesinlog-scale,i.e.
(log(P
f
)=log(P
fTrue
)),anditsCOV,andthenchecking
other parameters indicating the computational time,
for N
t
variesfrom50to150, McDevaseemstowork
well whenN
s
,N
t
=0.10.2, andN
f
,N
t
=0.10.2.
4 CONCLUSIONS
McDeva, basedonthecombinationof subset MCMC
simulation and DVM, is proposed to determineload
andresistancefactorsfor thelimit statefunctionthat
contains basic variables in both load and resistance
functions, and show high non-linearity. This kind of
performance function is frequently encountered in
geotechnical design in which a retaining wall is an
example. Theresultsshowthat:
Withacceptablebiasandcoefficient of variationof
failureprobability, McDevacan beused to obtain
theloadandresistancefactorseffectively.
McDevaseems to work well intheexamplewhen
N
s
,N
t
=0.10.2, andN
f
,N
t
=0.10.2.
Forfurtherdevelopments,theissuethatisnecessary
tobestudied: inthepresentexample, resultedcorrela-
tionbetweenRandS wereveryweak, bychance.Thus,
R andS havebeentreatedasindependentvariables. If
thereissomenonnegligiblecorrelationbetweenR and
S, someadditional considerationisrequiredindeter-
miningloadandresistancefactorstosecuresufficient
safetymargin.
REFERENCES
Au, S.K. & Beck, J.L. 2003. Subset simulationanditsappli-
cationtoseismicriskbasedondynamicanalysis. Journal
of Engineering Mechanics, ASCE, 901917.
Ditlevsen, O. & Madsen, H.O. 1996. Structural reliability
methods, J ohnWiley& SonsLtd, England.
Gilks, W.R., Richardson, S. & Spiegelhalter D.J. (1998).
Introducing Markov chain Monte Carlo. Markov chain
Monte Carlo in practice (W.R. Gilks et al. Eds.), 119,
ChapmanandHall/CRC.
Honjo, Y., Kikuchi, Y., Suzuki, M., Tani, K. & Shirato, M.
2005.J GSComprehensiveFoundationDesignCode:Geo-
code21, Proc. 16thICSMGE, pp. 28132816, Osaka.
Orr, L.L.T. 2005. Proceedingsof theInternational Workshop
on the Evaluation of Eurocode 7, p. 71, pp. 127149,
Dublin.
Phoon,K.K.,Becker,D.E.,Kulhawy,F.H.,Honjo,Y.,Ovesen,
N.K., andLo, S.R. 2003. Whyconsider ReliabilityAnaly-
sisfor Geotechnical Limit StateDesign? LSD2003 Inter-
national Workshop on Limit State Design in Geotechnical
Engineering Practice, USA
Phoon, K.K. & Honjo, Y. 2005. Reliability analyses of
geotechnical structures: towards development of some
user-friendlytools, Proc. 16thICSMGE, pp. 28452848,
Osaka.
81
Reliabilityanalysesof rockslopestability
Politecnico di Bari, Bari, Italy
ABSTRACT: Thebenchmark proposedisrelatedtothetopicof instabilityanalysesinanchoredrock slopes.
As it can be observed the definition of a safety factor can be addressed in two different ways so that even
differentvaluescanbeprovidedassafetylevel indeterministicapproach. Herereliabilityanalysesareperformed
by means of FORM, SORM and MCS in order to investigate the effects of uncertainties and variability of
rock design resistances on safety level. Moreover acomparison between thepartial factor approach and the
reliability based design has been performed pointing out the rule of characteristic values in modern design
provisions.
1 INTRODUCTION
1.1 Stability assessment for rock slopes
The main feature of mechanical behaviour in rock
slopeinstability isthat shear takesplacealongeither
adiscreteslidingsurfaceor withinazonebehindthe
face: if theactingshear forceisgreater thantheshear
strength of the rock on this surface, then the slope
becomes unstable. The definition of instability will
differ dependingonthemethodadopted.
Commonly, thestabilityof aslopecanbeexpressed
inoneormoreof thefollowingterms(WyllieandMah,
2004):
1) Factor of safety, FS. Stabilityisquantifiedbylimit
equilibriumof theslope:FS>1meansstableslope.
2) Strain. Failureisdefinedby onset of strainsgreat
enoughtopreventsafeoperationof theslope,orthat
therateof movement exceedstherateof miningin
anopenpit.
3) Probability of failure. Stability is quantified
by probability distribution of difference between
resisting and displacing forces (Safety Margin),
which areeach expressed as probability distribu-
tions.
4) LRFD (load and resistance factor design). Sta-
bility is defined by the factored resistance being
greater than or equal to the sumof the factored
loads.
Inorder toinvestigatethestability of arock slope
wedge involved in sliding mechanism, the equilib-
riummethod is usually employed and the factor of
safety is computed. In the case studied the pres-
ence of an anchorage force has been considered
(seeFig. 1).
Here, accordingtoMohr-Coulombfailurecriterion
for therockmassbehaviour, twodifferentexpressions
Figure1. Loadsandgeometryof arockslopewedge.
for the factor of safety (FS) have been investigated
(DeMello1988):
whereP is thewedgeweight; T is theanchor pull;
istheanchor andtheslopeinclination; isthesliding
surfaceinclinationand istheinternal frictionangle.
Suchfactor of safetyvaluescanbecomparedwith
some target values proposed by Terzaghi and Peck
(1967) andbyCanadianGeotechnical Society(1992)
that for earthworks vary between 1.3 and 1.5. The
upper valueappliestousual loadsandservicecondi-
tionswhilethelower valueappliestomaximumloads
andtheworst expectedgeological conditions.
Therefore, apart from the ambiguity in comput-
ing procedure, the factor of safety is affected by
uncertaintiesandvariabilitywhichcannotbeavoided.
83
The study performed below focuses on reliabil-
ity approach for computing the safety of the slope
by accounting for randomvariables to describerock
resistanceparameters.
2 THE CASE STUDIED
2.1 Uncertainties in rock mass characterization
Strength of arock mass is characterized by strength
of intact rock and of discontinuities. Depending on
number, orientation and condition of joints therock
mass behaviour is affected by anisotropy and weak-
ness which can lead to the failure when slopes are
concerned. Moreover humanactivity andweathering
processes may contribute to increase failure prone
conditions by means of increasing acting forces or
reducingrockmassresistance.
Besides, thefirststepinrockslopestabilitystudyis
themechanical characterizationof rockmassbymeans
of classificationsystems. Then, after agoodanalyses
of discontinuitiesandtheirshearresistancethetypeof
slidingmovement canbeforecast.
Intheproposedcasestudyarockslopeslidingalong
slidingpronediscontinuitiesisconsidered(seeFig. 1)
wheretheslopeistherockfaceoneof anopenquarry
reinforcedbyananchorage.
Thefrictionangleof ajoint inarock mass, canbe
computedasBartonandChoubey(1977):
where
n
isthevertical effectivestressactingondis-
continuitywall;J RC
n
isthejointroughnesscoefficient
for joint of actual length; J CS
n
isthejoint wall com-
pression strength for joint of actual length;
r
is the
residual frictionanglethat canbedrawnexperimen-
tallyandi istheroughnessof discontinuityat large
scale.
J RC
n
and J CS
n
are calculated by Bandis et al.s
formulation(1981), that is:
where J CS
0
and J RC
0
are joint wall compression
strength and joint roughness coefficient respectively
computedfor referencejoint lengthL
0
of 10cm; L
n
is
theactual joint length. J CS
0
canbeprovidedbyrefer-
encetableswhereasJ RC
0
canbedrawnfromSchmidt
hammer test.
Asexplainedbefore,theevaluationof forajointin
arock massisaffectedbyvariabilityanduncertainty.
Here,thefrictionangleprobabilitydistributionismod-
elled as lognormal with the coefficient of variation
equal to30, 40and50%.
Nonetheless, the sliding plane inclination, named
, canbeconsideredalso as arandomvariable. It is
Table1. Variablevaluesfor benchmark.
Standard
Mean deviation Characteristic
Variable value value value
Slidingsurface 40 4 38
slope [
]
Rockface 7.5 7.5
slope [
]
Frictionangle 35 10.5 29.8
[
]
Unit weight 23 0.5 22.8
[kN/m
3
]
Slopeheight 51015
H[m]
determinedby meansof structural investigationsand
consequentlyaffectedbyhumanerrors.Therefore, for
anormal probabilitydistributionwithameanvalue
of 40
is considered acoefficient of variation equal

to10%.
2.2 Factor of safety method
Thecommonest approachtostabilitystudiesisbased
on the computation of the factor of safety. Accord-
ingtotheequilibriummethodandMohr-Coulombsoil
behaviour, thesafety factor iscomputedby meansof
different expressions, namely Eq. (1) and (2): such
two cases arise fromdifferent interpretation of the
pull component along the sliding surface direction.
As a matter of fact, it can be taken as a stabilizing
forceorasanegativecontributiontotheslidingweight
component.
The choice between the two approaches can be
rationally undertakenconsideringaparametric deter-
ministicandprobabilisticstudywhereanchoragepull
andtheheight of theslopearevaried.
Table 1 shows values of variables used in such a
benchmark.
Three values for slope height have been inves-
tigated: 5, 10 and 15m. According to anchor pull
increasing the two factors of safety increase with a
different trendascanbeseeninFig. 2.
TheexpressionEq. (1), indicatedasFS
+
increases
linearly although slope decreases when height
increases. So that, the factor of safety is stronger
affected by anchoragepull for smaller height of the
slope.
On the contrary, expression Eq. (2), reported as
FS
, shows a parabolic trend with a rapid increase

asanchor pull increases.
Suchatrendprovidesobviouslynegativevaluesas
thecontributionof anchoragepull got higher thanthe
weight component ontheweakplane.
Accordingly, FS
wont bediscussedfurther on.

2.3 Reliability method
AstheFS
fails,onlyexpressionEq.(1)canbeconsid-
eredforthesafetyfactor.Then, itisworthytocompare
84
Figure2. Variationof safetyfactor (Eq. 1) withtheanchor
pull.
Table2. Randomvariabledistributiontype.
Distribution Variation Min Max
Variable type coefficient value value
Slidingsurface Normal 10%
slope [
]
Rockface Uniform 5 10
slope [
]
Frictionangle Lognormal 30, 40, 50%
[
]
Unit weight Normal 2%
[kN/m
3
]
Anchoragepull Constant
T [kN]
Slopeheight Constant
H[m]
thesafetyfactorwiththereliabilityindexcomputedby
meansof arandomvariableapproach. Suchaparame-
terisstrictlyrelatedtotheprobabilityof failure. Many
studies havebeenperformedontheallowablevalues
of theprobabilityof failureof theengineeringproject
acceptedbypeople. Inthecaseof slopestabilityvalues
of reliabilityindexabout2-4aresuggestedbyBaecher
(1982).
Accordingly, in this analysis such interval will be
investigated.
Table1 summarizes mean and standard deviation
values for randomvariables whileTable2shows the
assumption about probability distribution for such
providedvariables.
As the unit weight is concerned, it is can be
assumedasarandomvariablewithasmall variability,
provedbyliterature(Cherubini 1997). Inthiscaseitis
considerednormally distributedwithacoefficient of
variationequal to2%.
InthisbenchmarkFORM, SORMandMonteCarlo
methodsareperformedinorder toevaluatetherelia-
bility index by means of COMREL code (1997). In
thiscase, theperformancefunctionconsideredis:
Figure 3. Reliability index versus anchorage pull for
CV
=30%.
CV
=40%.
Results fromFORM andSORM techniques coin-
cide, thusjust FORM arereportedinthefollowing.
At first, reliability index is reported versus the
anchoragepull, asalreadydonefortheFactorof safety
(Fig. 2). Then, reliability index is comparedwiththe
safetyfactor for thesameanchoragepull values.
Fig. 3showscurvedtrendsof reliabilityindexver-
susanchoragepull. For thecaseof slopeheight equal
to5m, as anchoragepull varies from100to150kPa
thereliabilityindexincreasesfrom2to4whenthefric-
tionanglecoefficient of variationCV
is 30%. Such
trendcanbeseenfor other twoheight valuesbut with
aslopereductionas theheight increases: this means
that when height increases an higher anchoragepull
increaseis needed for stabilizing theslopeas in the
caseof thefactor of safety.
Such evidence is true also when CV
increases
(Figs. 45). So that, when the variability of friction
angleincreasesthesafetylevel of theslopereducesand
muchhigheranchoragepull incrementsareneededfor
improvingstabilitycondition.
Accordingly, Figure6 shows, for theslopeheight
equal to5mthesafetyfactor andthereliabilityindex
values correspondingto thesameanchoragepull for
thethreeCV
values.Hence,whenthereliabilityindex
varies between2and4thesafety factor increases its
values according to theCV
. As amatter of fact, in
order togetareliabilityindexequalsto2.5, thefactor
of safetyshouldbeincreasedfrom1.8to2.2.
85
CV
=50%.
Figure6. Factorof safety(Eq. 1) versusreliabilityindexfor
CV
=30%, 40%and50%andH=5m.
Figure 7. Reliability index computed by MCS versus
anchoragepull forCV
=30, 40e50%: H=5m(continuous

lines); H=10m(dottedlines); H=15m(dashedlines).
Therefore, thereferencevalues 1.31.5 suggested
for the factor of safety are definitively inadequate
when variabilityistakenintoaccount.
Astheother twoslopeheight cases(H=10mand
15m) are concerned they give the same results as
coincident curvesareprovided.
Results from MCS with an adaptive sampling
scheme have been investigated: 30000 samples are
neededforthecasestudiedandreliabilityindexvalues
drawnarethesameasFORM. Fig. 7showsthesame
evolutionof reliabilityindexcurvesfromMCS.
2.4 LRFD method
Finally,aninterestingcomparisonisproposedbetween
reliability index andthepartial factor designmethod
suggestedby Eurocode7(1997) andby ItalianTech-
nical BuildingCode(TU 14/01/2008).
Thecombinationof factorssuggestedfortheglobal
stability analysis is named as Combination 2, where
load factors (A2), resistancefactor (M2) and global
factor (R2) areconsideredasfollows:
InEq. (7), designvariablesvaluesemployedarechar-
acteristic values: such values are introduced by the
Eurocodefor limitstatedesignbutnosuggestionsare
providedabout thewaytocomputethem.
Characteristic valuesfor reinforcedconcretecom-
pressionstrengthis commonly assumedas thevalue
correspondingtothe95%probabilitytobeexceeded.
Assumingit isnormallydistributedwehave:
wherex
m
isthemeanvalueandsisthestandarddevi-
ation of theconcretecompression strength. Such an
assumptioncouldbetoomuchconservativeinthecase
of geotechnical randomvariables. Asamatter of fact,
letsconsiderthefrictionangle, withmeanvalueequal
to35
andstandarddeviationequal to10.5
;itscharac-
teristicvalueis14.4
, whichistoomuchconservative
value.
HeretheexpressionfromSchneider(1997)reported
alsobyCherubini andOrr(1999) isused:
where CV is the coefficient of variation; x
m
is the
randomvariablemeanvalue.Threecasesarethencon-
sidered according to theCV
variation. In this case,

thefrictionanglecharacteristicvalueis30
.
Moreover, as Eq. (7) is concerned, building
code requests that such difference is higher than
zero although no minimumvalue suggests. Hence,
Figs. 810showthepositivedifferencebetweenresis-
tanceand action according to theexpression Eq. (7)
versus theanchoragepull values. Thetrend, for each
slopeheight, is not linear andit shows adecreaseof
slopeasheight increases.
The three values of CV
do not affect the slopes

of thetrendwhileaffect themagnitudeof anchorage
pull needed: whenCV
increases theanchoragepull
must beincreasedinorder to havethesamepositive
differencebetweenresistanceandaction.
A meaningful correlationcanbeobservedinFig. 11
where reliability index is related to the positive dif-
ference between resistance and action for the same
anchorage pull value. Each graph corresponds to
different CV
valuesfor thecaseof 5mslopeheight.

86
Figure 8. Combination for sliding condition in LRFD
versusanchoragepull for CV
=30%.
=40%
=50%.
FromFig. 11 the LRFD differences between 30
and60kPacorrespondtothereliabilityindexvarying
between2and4for thecaseof CV
=30%.
Suchdifferenceshall bethesamefor higher vari-
ability although the lower boundary of the range
versus reliability index for slope height H=5m and
CV =30%, 40%and50%.
considered increases as friction angle variability
increases. This means that, in such a case, simply
considering:
1) resistancejust higher thanactions;
2) characteristicvaluesandpartial safetyfactors;
it cannot give a constant reliability level. These
changes are related both to anchorage pull values,
slopeheight andvariability.
Hence, when variability of the friction angle is
enlarged higher differences between resistance and
loads areneededbut thesevalues vary accordingnot
onlytoCV
but alsotoslopeheight variation.

3 CONCLUDINGREMARKS
Abenchmarkhasbeenproposedtoinvestigatestability
of ananchoredrockslopeaccordingtothreedifferent
methods: thefactor of safety, thereliabilityindexand
theloadandresistancepartial factor.
Resultsshowthat, whenvariabilityof thoserandom
variableswhichgovernthelimitequilibriumof arock
slopehastakenintoaccount:
1 it causes a safety reduction which depends on
variabilitymagnitude;
2 target values for safety factor shall be increased
accordingtovariabilitymagnitude;
3 the difference between factorized resistance and
loadmust beincreasedaccordingtothevariability
magnitudeandtheheight of theslope.
REFERENCES
Bandis, S.C., Lumdsden, A.C. & Barton, N. 1981. Exper-
imental studies of scale effects on the shear behaviour
of rock joints. Int. Journal of Rock Mech. Min. Sci. and
Geoch. Absts. 18: 121.
Baecher G.B. 1982. Simplified geotechnical data analysis.
Reliability theory and its application in structural and soil
engineering. Thoft-ChristensenP. (ed.), Dordrecht: Reidal
Publishing.
Barton, N. & Choubey, V. 1977. Theshear strengthof rock
jointsintheoryandpractice. Rock mechanics 10(1): 154.
87
Canadian Geotechnical Society 1992. Canadian Founda-
tion Engineering Manual. Vancouver, Canada: BiTech
PublishersLtd.
Cherubini, C. 1997. Dataandconsiderationonthevariabil-
ity of geotechnical properties of soils. In Proceedings
of the ESREL97 International Conference on Safety and
Reliability, Lisbon: 15831591.
Cherubini, C. & Orr, T.L.L. 1999. Considerations on the
applicability of semiprobabilistic Bayesian methods to
geotechnical design. InAtti del XX Convegno Nazionale
di Geotecnica, Parma:421426.
COMREL 1997. Reliability Consulting Programs RCP
GmbH, Munchen, Germany.
DeMello, Victor F.B. 1988. Risksingeotechnical engineer-
ing: conceptual and practical suggestions. Geotechnical
Engineering 19(2): 171208.
Eurocodice 7, Progettazione geotecnica Parte 1: Regole
generali, UNI ENV 1997 1- Aprile1997.
Schneider, H.R. 1997. Definitionanddeterminationof char-
acteristicsoil propertiescontributiontodiscussion. InXIV
ICSMFE, Hamburg: Balkema.
Terzaghi, K. &Peck, R. 1967. Soil mechanics in Engineering
Practice. NewYork: J ohnWileyandSons.
Testo Unitario 2008. D.M. Infrastrutture 14 gennaio 2008.
NuoveNormeTecnicheperleCostruzioni.Ministerodelle
Infrastrutture, Ministero delInterno, Dipartimento della
ProtezioneCivile.
Wyllie, DuncanC. & Mah, Christopher W. 2004. Rock slope
engineering. LondonandNewYork: SponPress.
88
Reliabilityanalysisof abenchmarkproblemfor slopestability
Y. Wang, Z.J. Cao, S.K. Au& Q. Wang
Department of Building and Construction, City University of Hong Kong, Hong Kong, China
ABSTRACT: This paper presents a benchmark study for slope stability analysis using the Swedish Circle
methodinconjunctionwithgeneral procedureof slices. Thepurposeis toillustrateimplementationof spatial
variability of soil properties(i.e., undrainedshear strengthS
u
) intheanalysis. Thestudy revealsthat different
reliability analysis methods usedifferent critical slip surfaces. Thepitfalls of using different slip surfaces in
different methodsarediscussed.
1 INTRODUCTION
Limited published studies/implementation examples
arebelievedasonereasonfor geotechnical practition-
ers reluctance to apply reliability methods to slope
stability analysis. In view of this aspect, this paper
providesabenchmark examplethat illustratesimple-
mentation of reliability methods in slope stability
analysis. The example focuses on spatial variability
of soil properties and different critical slip surfaces
usedindifferent reliabilityanalysismethods.
After this short introduction, a simple slope sta-
bilityexampleisdescribed, together withtherandom
variables andsolutionmethods usedintheexample.
Then, theanalysisresultsarepresented, anddifference
among the results from different solution methods
are highlighted. Finally, the pitfalls of using differ-
ent critical slip surfaces in different methods are
discussed.
2 BENCHMARK EXAMPLE
2.1 Description of the example
Figure1 shows aclay slopewith aheight and slope
angleof 5mand45
,respectively.Stabilityof theslope
is assessedusingtheSwedishCirclemethodincon-
junctionwithgeneral procedureof slices(Duncanand
Wright2005).Thefactorof safetyFSisdefinedasthe
critical (minimum) ratio of resisting moment to the
overturningmoment, andtheslipsurfaceis assumed
to beacircular arc centered at coordinate(x, y) and
withradiusr. Theoverturningandresistingmoments
are summed about the center of the circle to calcu-
late the factor of safety, as shown in Figure 1. For
moment calculationsthesoil massabovetheslipsur-
face is subdivided into a number of vertical slices,
eachof whichhas aweight W
i
, circular slipsegment
lengthLl
i
, undrainedshear strengthS
ui
alongtheslip
Figure1. Slopestabilityexample.
segment, andanangle
i
betweenthebaseof theslice
andthehorizontal.Thefactorof safetyisthengivenby
wheretheminimumistakenover all possiblechoices
of slip circles, i.e., all possiblechoices of (x, y) and
r. Note that Ll
i
, W
i
, and
i
change as (x, y) and/or
r change(i.e., geometry of theithslicechanges). In
addition, the W
i
is a function of the clay total unit
weight. Therefore, for agivenchoiceof (x, y) andr,
FSonlydependsonS
ui
and.
TheperformancefunctionP of this slopestability
problemcanbeexpressedas
Note that the mathematical operation of min in
Equations(1) and(2) makestheperformancefunction
implicit andnon-differentiable.
2.2 Input variables
This exampleconsiders thespatial variability of soil
properties, and the soil between the upper ground
89
Table1. Thevaluesanddistributionsof theinputvariables.
Variable Distribution Statistics
S
u
Log-Normal (30entries Mean=20kPa
inthevector areiid.) Cov
=20%
Deterministic 18kN/m
3
Cov standsfor coefficient of variation.

surface and 15m below is divided into 30 equal
layers with a layer thickness of 0.5m. A vector
S
u
=[S
u
(1), S
u
(2), . . . , S
u
(30)]
T
isdefinedfor theval-
uesof S
u
inthese30layers. Then, S
ui
for theithslice
is averageof theentries in S
u
that correspondto the
depths of slip segment of theith slice. All entries in
S
u
aretakenasindependentandidentically-distributed
(i.e., iid) log-normal randomvariablewithameanand
coefficientof variation(i.e., COV) of 20kPaand20%,
respectively. Thetotal unit weight of clay is taken
as deterministic with a value of 18kN/m
3
. The val-
uesanddistributionsof thesebasicinputvariablesare
summarizedinTable1.
2.3 Solution methods
For a given choice of (x, y) and r, FS is calculated
by implementingEquation1inanExcel spreadsheet
whereeachrowrepresentsaparticular sliceandeach
column represents the variables and terms in Equa-
tion1. A VBA codehasbeenwrittentocalculatethe
ratio of resistant to overturning moment for differ-
ent valuesof (x, y) andr andthenpick theminimum
value as the factor of safety. As a reference, the
nominal value of FS that corresponds to the case
where all S
u
values equal to their mean values of
20kPais equal to 1.25. For thecritical slip surface,
r=15mand (x, y)=(2.7m, 8.8m). Thecalculation
resultsarefoundtobeconsistentwithresultsfromthe
commercial slopestabilityanalysissoftwareSlope/W
(GEO-SLOPEInternational Ltd.2008).Consequently,
theExcel spreadsheet model is validatedandusedin
thereliabilityanalysis.
The reliability methods employed in this study
includetheFirst-OrderSecondMoment(FOSM) with
a given critical slip surface (Ang and Tang 1984,
Tang et al. 1976, Wu 2008), First-Order Reliabil-
ity Method (FORM) using the object-oriented con-
strained optimization tool in the Excel spreadsheet
(LowandTang1997, Low2003, LowandTang2007),
direct MonteCarlosimulation(MCS) usingcommer-
cial softwareSlope/WandtheExcel spreadsheet, and
Subset simulationusingtheExcel spreadsheet (Sub-
sim) (AuandBeck 2001). TheMCSinSlope/Wuses
theSwedishCirclemethodandgeneral procedureof
slices.Therefore, theperformancefunctiondefinedby
Equation(2) isalsoapplicable.
To facilitate direct Monte Carlo simulation and
SubsetsimulationusingtheExcel spreadsheet, apack-
age of spreadsheets and VBA functions/Add-In are
developedinExcel.Anuncertaintymodel spreadsheet
Table2. Primaryanalysisresults.
Relative
Solutionmethod P
f
(%) error inP
f
FOSM 4.74 1.110
4
99.9%
FORM 3.89 5.110
3
96.8%
MCSwithSlope/W 4.34 7.010
4
99.6%
MCSwithExcel 2.95 1.610
1
N/A
SubsimwithExcel 3.04 1.210
1
25.0%
isdevelopedfor generatingarandomsample(realiza-
tion) of therandomvariablesS
u
. Startingwithuniform
random numbers provided by the built-in function
Rand() in Excel, transformation is performed to
producetherandomsamples of desired distribution.
AvailableVBA subroutinesinExcel areusedtofacil-
itatetheuncertainty modeling. Fromaninput-output
perspective, theuncertaintymodelingworksheettakes
noinput but returnsarandomsampleof S
u
asitsout-
put whenever a re-calculation is commanded. Then,
theuncertaintymodel worksheetsarelinked together
withthedeterministic slopestability analysisspread-
sheet through their input/output cells to produce a
probabilistic analysis model of the slope stability
problem. The value of S
u
shown in the determinis-
ticanalysisworksheetisequal tothatgeneratedinthe
uncertainty modelingworksheet, andsotheFS value
calculated in thedeterministic analysis worksheet is
random. Inother wordsat thisstageonecanperform
a direct Monte Carlo simulation of the problemby
repeatedly executingthebuilt-infunctionRand() in
Excel. In addition, aVBA code for Subset Simula-
tionisdevelopedthat functionsasanAdd-IninExcel
and can be called by selecting fromthe main menu
Tools followed by SubSim. A user formappears
upon invokingof thefunction, andtheSubset simu-
lationcanbeperformedaccordingly. Moredetailson
theExcel spreadsheetsandVBA functions/Add-Inare
referredtoAuet al. (2009).
2.4 Analysis results
Table2 summarizes theanalysis results fromdiffer-
ent reliabilitymethods. Totally5,000,000samplesare
taken in MCS with Slope/W, as the P
f
obtained is
extremely small. For MCS with Excel, the number
of samples is 10,000. For Subsimwith Excel, three
levels of simulationareperformedwith500samples
takenineachlevel. Thesecondandthirdcolumnsof
Table2showtheequivalent reliabilityindex =+
1
(1P
f
) and its corresponding probability of failure
P
f
fromFOSM, FORM, MCS with Slope/W, MCS
withExcel andSubsimwithExcel, respectively. The
P
f
fromFOSMandMCSwithSlope/Wisontheorder
of magnitudeof 10
4
%. Incontrast, theP
f
fromMCS
andSubsimwithExcel isontheorderof magnitudeof
10
1
%, andtheP
f
fromFORM fallsbetween10
4
%
and10
1
%(i.e., ontheorderof magnitudeof 10
3
%).
If theP
f
fromMCSwithExcel isusedasthereference
90
for comparison, therelativeerror inP
f
isabout 100%
for FOSM, FORM, and MCS with Slope/W. TheP
f
fromSubsimwithExcel isreasonablyaccuratewitha
significant increaseof computational efficiency (i.e.,
decreasingnumber of samplesneeded).
The substantial difference between the P
f
from
FOSMandMCSwithSlope/WandthatfromMCSand
SubsimwithExcel might seemsurprisingat thefirst
glance.Detailedexaminationsshowthatthedifference
canbeattributedtodifferentcritical slipsurfacesused
indifferentreliabilitymethods, whicharediscussedin
thenext section.
3 PITFALL OF CRITICAL SLIP SURFACE
3.1 FOSM with a given critical slip surface
FOSM is basedonthefirst order approximationand
uses the first terms of aTaylor series expansion of
performance function P. Therefore, the mean of FS
is estimated by setting all S
u
values equal to their
meanvalues of 20kPaandsearchingfor critical slip
surface. The resulting mean of FS is 1.25, and the
correspondingcritical slipsurfacehasar=15mand
(x, y)=(2.7m, 8.8m). For thisgivencritical slipsur-
face,thestandarddeviationof FSisestimatedas0.053.
Notethat, for thegivencritical slipsurface, Equation
1 only involves linear operation (i.e., summation) of
randomvariablesS
u
, andhigh-order partial derivative
of theEquation is zero. Thesolution fromFOSM is
thereforereasonablyaccurateif onlyonegivencritical
slipsurfaceisconcerned.
3.2 Form
A practical object-oriented constrained optimization
approachproposedbyLow(2003) isusedinthiswork
to calculate the Hasofer-Lind reliability index. The
approachisimplementedinanExcel spreadsheet and
uses thebuilt-inoptimizationtool Solver to obtain
theminimumdistancebetweentheperformancefunc-
tionandcenter of anexpandingequivalent dispersion
ellipsoidintheoriginal spaceof therandomvariables
asthereliability index. Thesearchingfor critical slip
surface is accounted for in the approach by includ-
ing (x, y) and r as additional optimization variables
and adding a constraint for FS (i.e., FS=1) in the
optimization for theminimumdistance. Variation of
potential critical slip surfaces is implicitly factored
in the analysis. Consequently, as shown in Table 2,
theP
f
fromFORM ismorethanoneorder of magni-
tudelargerthanthatfromFOSMwhichonlyonegiven
critical slipsurfaceisused.
Ontheotherhand, whencomparedwiththeP
f
from
MCSwithExcel whichincludesthevariationof criti-
cal slipsurfaceexplicitly, FORM significantlyunder-
estimates the P
f
. The poor performance of FORM
might be attributed to the inadequate linear approx-
imation of the failure criterion (i.e., Equation 1),
particularlywhenthevariationof potential critical slip
surfacesisaccountedfor.
Figure2. Critical slipsurfaceinSlope/W.
Figure3. Examples of critical slipsurfaces obtainedfrom
MSC withExcel.
3.3 MCS with slope/W
Table2showsthat theP
f
fromMCSwithSlope/Wis
onthesameorder of magnitudeas that fromFOSM.
InSlope/W, theMCSonlytakesintoconsiderationthe
variabilityof soil properties, andit usesacritical slip
surfacethatisfirstdeterminedbasedonthemeanval-
uesof therandomvariables. Figure2showsthecritical
slipsurfaceobtainedinSlope/Wusingthemeanval-
uesof S
u
. Thecritical slipsurfaceinSlope/Wisquite
consistentwiththeoneusedinFOSM.Asaresult, itis
notsurprisingtofindthattheP
f
fromFOSMandMCS
withSlope/W agrees well witheachother. However,
asthevariationof potential critical slipsurfacesisnot
properly accountedfor ineither method, their results
arebiased.
3.4 MCS and Subsim with Excel
MCS inExcel startswithgenerationof randomsam-
ples (realizations) for therandomvariables S
u
. Then,
for eachrandomsampleof S
u
, critical slipsurfaceis
searched,andtheminimumFSisobtainedaccordingly.
Figure3showexamples of different critical slipsur-
faces obtainedfromdifferent randomsamples of S
u
.
It isobviousthat thecritical slipsurfacechangessig-
nificantlyasthespatial distributionof S
u
changesfor
different randomsamples. Asareference, thecritical
91
Table3. Rangesof centercoordinatesandradiusforcritical
slipsurfacesobtainedfromMCSwithExcel.
Parameter Minimum Maximum Range
Coordinatex(m) 1.0 3.0 2.0
Coordinatey(m) 6.0 9.6 3.6
Radiusr (m) 9.0 16.0 7.0
Table 4. Comparison of simulation results with different
critical slipsurfaces.
Relativeerror
Solutionmethod P
f
(%) inP
f
(%)
MCSwithSlope/Wand 4.34 7.010
4
99.6%
fixedcritical slip
MCSwithExcel and 4.38 6.010
4
99.6%
fixedcritical slip
MCSwithExcel and 2.95 1.610
1
N/A
changingcritical slip
SubsimwithExcel and 4.53 3.010
4
99.8%
fixedcritical slip
SubsimwithExcel and 3.04 1.210
1
25.0%
changingcritical slip
slip surface#4 in Figure3 is theoneobtained from
deterministic analysis withmeanS
u
values andused
intheFOSM andMCS inSlope/W. Table3summa-
rizes ranges of (x, y) and r for critical slip surfaces
obtained fromMCS with Excel. The r varies from
9.0mto 16.0mand has arangeof 7.0m. When the
variation of potential critical slip surfaces is consid-
ered explicitly in the simulation, the P
f
fromMCS
withExcel isthreeorder of magnitudelarger thanthat
fromFOSMandMCSwithSlope/Wwhichuseagiven
critical slipsurface.
Tofurtherillustratetheeffectof variationof critical
slipsurfacesontheP
f
,MCSisalsoperformedinExcel
usingthecritical slipsurface#4inFigure3, whichis
obtainedfromdeterministicanalysiswithmeanS
u
val-
ues andusedintheFOSM andMCS inSlope/W. As
shown inTable4, theresulting P
f
is on theorder of
10
4
%and agrees well with those fromFOSM and
Slope/Wwhichusesthesame/similar critical slipsur-
face.ThecomparisonsummarizedinTable4confirms
thatthesubstantial differenceamongP
f
fromdifferent
methodsismainlyattributedtothevariationof critical
slipsurfaces.
Subsimis carriedout inExcel witheither afixed
critical slipsurface(i.e., #4inFigure3) or searching
for critical slip surfaces for different randomsam-
ples of S
u
. As shown in Table 4, the results agree
well with thosefromMCS with Excel. TheP
f
from
Subsimisreasonablyaccuratewithincreasingcompu-
tational efficiency(i.e., decreasingnumberof samples
needed).
4 RECOMMENDEDRELIABILITY
METHODS
Whenspatial variabilityof soil propertiesistakeninto
considerationinreliability analysis of slopestability
problem, variation of potential critical slip surfaces
has a profound impact on the calculated P
f
or .
Similar to FOSM with a given critical slip surface,
MCS in Slope/W relies on a given critical slip sur-
faceobtainedfromdeterministic analysis withmean
valuesof soil properties, andhence, theimpactof vari-
ation of critical slip surfaces is not accounted for in
the analysis. Their results are therefore biased. The
FORM approachconsidersimplicitlythevariationof
potential critical slip surfaces. However, it signifi-
cantly underestimates the P
f
due to the inadequate
linearapproximationof thefailurecriterion(i.e.,Equa-
tion 1), particularly when the variation of potential
critical slip surfaces is accounted for. Therefore, the
useof MCS or Subsimwithexplicit considerationof
critical slipvariationisrecommended. Suchconsider-
ationcanbeimplementedintheanalysiswithrelative
ease, asillustratedinthispaper.
5 CONCLUDINGREMARKS
This paper presents a benchmark example for slope
stability analysis using the Swedish Circle method
in conjunction with general procedure of slices.
Spatial variability of soil properties are taken into
consideration in the reliability analysis, and the
effect of variationof potential critical slipsurfacesis
highlighted.
Several reliability methods are implemented to
investigatetheir feasibility andefficiency. Similar to
FOSM with a given critical slip surface, MCS in
Slope/Wdoesnotaccountforthevariationof potential
critical slip surfaces and only uses a given criti-
cal slip surfaceobtained fromdeterministic analysis
withmeanvalues of soil properties. Their results are
therefore biased. The variation of potential critical
slip surfaces is implicitly considered in the FORM
approach proposed by Low and his collaborators.
However, it significantly underestimates the P
f
due
to theinadequatelinear approximationof thefailure
criterion. MCS or Subsim with explicit considera-
tion of variation of critical slip surfaces are imple-
mentedinanExcel spreadsheetenvironment.Theyare
showntoprovidereasonableresults, andhence, their
use is recommended at the expense of computation
time/efforts.
ACKNOWLEDGEMENTS
The work described in this paper was supported by
General ResearchFund[Project No. 9041327(CityU
110108)] andCompetitiveEarmarkedResearchGrant
[Project No. 9041260 (CityU 121307)] from the
Research Grants Council of theHong Kong Special
92
AdministrativeRegion, China. Thefinancial supports
aregratefullyacknowledged.
REFERENCES
Ang, A. H.-S. andTang, W. H. 1984. Probability Concepts
in Engineering Planning and Design, Vol. II, Wiley, New
York.
Au, S. K., Wang,Y., andCao, Z. J. 2009. Reliabilityanalysis
of slope stability by advanced simulation with spread-
sheet. Proceeding of the 2nd International Symposium on
Geotechnical Safety and Risk (IS-Gifu2009), J une2009,
Gifu, J apan(submitted).
Au, S. K., and Beck, J. L. 2001. Estimation of small fail-
ureprobabilitiesinhighdimensionsbysubsetsimulation.
Probabilistic Engineering Mechanics, 16(4): 263277.
Duncan,J.M.andWright,S.G.2005.Soil Strength and Slope
Stability, J ohnWiley& Sons. Inc. NewJ ersey, 2005.
GEO-SLOPE International Ltd. 2008. Stability Modeling
with Slope/W 2007 Version, GEO-SLOPE International
Ltd, Calgary, Alberta, Canada.
Low, B. K. 2003. Practical probabilisticslopestabilityanal-
ysis. Proceedings of Soil and Rock America, MIT, Cam-
bridge, MA, J une2003, Verlag Gluckauf GmbH Essen,
Germany, Vol. 2, 27772784.
Low, B. K. and Tang, W. H. 1997. Efficient reliability
evaluation using spreadsheet. Journal of Engineering
Mechanics, 127(7): 149152.
Low, B. K. and Tang, W. H. 2007. Efficient spreadsheet
algorithmfor first-order reliability method. Journal of
Engineering Mechanics, 133(2): 13781387.
Tang, W. H., Yucemen, M. S., and Ang, A. H.-S. 1976.
Probability-basedshort termdesignof soil slopes. Cana-
dian Geotechnical Journal, 13: 201215.
Wu, T. H. 2008. Reliability analysis of slopes. Reliability-
Based Design in Geotechnical Engineering: Computa-
tions and Applications, Chapter 11: 413447, Edited by
Phoon, Taylor & Francis.
93
Geotechnical code drafting based on limit state design
and performance based design concepts
DevelopingLRFDdesignspecificationsfor bridgeshallowfoundations
S.G. Paikowsky
Geotechnical Engg Research Lab, University of Massachusetts, Lowell, Massachusetts, USA &
Geosciences Testing and Research Inc., N. Chelmsford, Massachusetts, USA
S. Amatya
Geotechnical Engg Research Lab, University of Massachusetts, Lowell, Massachusetts, USA
K. Lesny&A. Kisse
University of Duisburg-Essen, Essen, Germany
ABSTRACT: Anongoingproject,supportedbytheNational CooperativeHighwayResearchProgram,NCHRP
Project 2431isaimedtodevelopLRFDproceduresandtomodifythecurrentAASHTOdesignspecifications
forUltimateLimitState(ULS) designof bridgeshallowfoundations.Thecurrentstudyutilizesacomprehensive
databaseof 549cases of shallowfoundationloadtests under various loadingconditions (i.e. vertical-centric,
vertical-eccentric, inclined-centric andinclined-eccentric). Inthis paper, theprocedureto establishtheLRFD
designadoptedintheresearchstudyisintroduced. Thedesignmethodsusedfor ULSdesignof bridgeshallow
foundationsarepresentedandtheuncertaintyintheestimationof theultimatebearingcapacityhasbeenexpressed
intermsof abias, definedasmeasuredover calculatedcapacities. Thebiasesintheestimationof theultimate
bearing capacity have been appraised based on the database. Typical bridge foundation loadings and their
uncertaintiesaredefinedandutilizedalongwiththeresistanceuncertaintiestoestablishresistancefactors. The
investigationsleadtotheconclusionthatonesingleresistancefactor for thebearingcapacityisnotsufficient, as
differentloadingconditionsresultindifferentlevelsof uncertainties.Hence,differentresistancefactorshavebeen
establishedbasedontheFirstOrder SecondMoment(FOSM) method, andtheMonteCarlosimulations(MCS),
eachfor thevertical-centric, vertical-eccentric, inclined-centric andinclined-eccentric loadingconditions. The
recommendedpreliminaryresistancefactorsthusobtainedinthestudyarepresented.
1 INTRODUCTION
An ongoing project, NCHRP Project 2431: LRFD
designspecificationsforshallowfoundations,isaimed
at developing LRFD procedures and modifying the
current AASHTO design specifications for theUlti-
mate Limit State (ULS) design of bridge shallow
foundations. It issupportedbytheNational Coopera-
tiveHighway ResearchProgram(NCHRP) under the
TransportationResearchBoard(TRB) of theNational
Academyof Science(NAS). TheAASHTOspecifica-
tionsaretraditionallyobservedasaNational Codeof
theUShighwaypracticeonall federallyaidedprojects,
hence, they influence the construction of highway
bridgeandotherstructurefoundationsacrosstheUSA.
The current AASHTO specifications as well as
other existing codes based on Load and Resistance
Factor Design (LRFD) principles were calibrated
using a combination of reliability theory, fitting to
ASD (allowablestress design) andengineeringjudg-
ment. The main objectives of this project therefore
are the compilation of a database of load tests on
shallowfoundations andthecalibrationof resistance
factorsbasedonthereliability analysisof thedatato
obtainmorerational designswithconsistent levelsof
reliability. Thechallenges for therequirement of the
secondobjectiveincludeovercominggenericdifficul-
tiesapplyingtheLRFDmethodologytogeotechnical
applications, i.e. theevaluation of uncertainty in the
geotechnical model incorporating e.g. indirect vari-
ability (site or soil parameters interpretation), load
dependencyof thegeotechnical resistance(especially
in the case of shallow foundations, where a strict
separation between load and resistance is not possi-
ble), judgment (e.g. apreviousexperience), andother
similar factors.
2 EVALUATIONOF BEARINGCAPACITY
UNCERTAINTY
2.1 Database
Thisresearchstudyutilizesacomprehensivedatabase
of load tests on shallow foundations, UML-GTR
ShalFound07, for the evaluation of uncertainties in
bearingcapacity(BC)estimation.Itcontains549cases
of load tests, mostly performed in Germany and the
USA. It has been compiled from various publica-
tionsnoticeablyusingfour major sources: (a) ShalDB
Ver5.1 (Briaud & Gibbens 1997), (b) Lutenegger
97
Table1. Summaryof UML-GTR ShalFound07database.
Predominant soil type
Foundation
type Sand Gravel Mix Others Total
Plateload 346 46 2 72 466
testsB1m
Small footings 26 2 4 1 33
1-B3m
Largefootings 30 1 31
3-B6m
Rafts& 13 5 1 19
MatsB>6m
Total 415 48 12 74 549
Notes: Mix: alternatinglayers of sandor gravel andclay
or silt Others: either unknownsoil typesor other granular
materialslikeloamyScoria
& DeGroot (1995), (c) German test database in a
set of volumes (e.g. Muhs & Weiss 1972) compiled
by DEGEBO (Deutsche Forschungsgesellschaft fr
Bodenmechanik) and(d) testscarriedoutor compiled
bytheUniversityof DuisburgEssen, Germany(some
of whicharepresentedinPerau1995). Thedatabase
summaryispresentedinTable1.
Most cases relate to foundations subjected to
vertical-centricloadinginorongranularsoils.Testsof
foundationsubjectedtocombinedloadings (vertical-
eccentric, inclined-centric and inclined-eccentric)
weremainlysmall scalemodel testsperformedincon-
trolled soil conditions (in laboratories using soils of
knownparticlesizeandcontrolledcompaction).
2.2 Failure load criterion and measured bearing
capacity
Vesi c (1975) suggested thefailure(ultimate) load to
betheloadwhichcorrespondstothepoint wherethe
slopeof theload-settlement curvefirst reaches zero
or asteady, minimumvalue. Theinterpretedultimate
loadsfordifferentloadtestsareshowninblackdotsin
Figure1(Vesi c1963).Insoilswithhigherrelativeden-
sities, thereisahigher possibilityof failureingeneral
shear modeandthefailureloadcanbeclearlyidenti-
fied, e.g. fortestnumber61.Therearealsocaseswhen
the identification of the minimumvalue becomes
subjective, accordingtothesoil relativedensitiesD
r
,
e.g. for test number 64inthefigure. Hence, theinter-
pretationof failureloadof ashallowfoundationfrom
aloadtest is complex as thefailuremodes (general,
local or punching) not onlydependonthetypeof soil
(categorized according to the relative density or the
relativestiffness), but alsoonthefootingembedment
andloadingtype. Except for thecaseof general shear
failure, inwhichthefailureloadisclearlydefinedby
apeakintheload-settlement curve, judgment isoften
requiredtointerpretauniquefailureload. Theexami-
nationof theloadtestresultsinthedatabasereveal that
thefailuresforwhichtheload-settlementcurvesdonot
showaclear peakprevail, hencetheinterpretationsof
the failure loads for footings in most cases become
Figure1. Failurecriterionbasedontheminimumslopeof
the load-settlement curve (Vesi c 1963; modified to show
settlement of 0.1B); 1psi 6.9kPa, 1pcf =0.157kN/m
3
,
1in=25.4mm
difficult. Inaddition, many loadtests inthedatabase
havebeenfoundtobeunsuitablefor useinthepresent
studyastheyhavenotbeencarriedouttofailure. This
is especially thecasefor larger sizedfoundations for
whichfailurewouldbeassociatedwithverylargeloads
andexcessivedisplacements well beyondtheservice
limits.
It was found from the comparison of the fail-
ure loads for 195 cases of database UML-GTR
ShalFound07 that the MinimumSlope failure crite-
rionprovidedthemost consistent interpretationwhen
comparedto interpretedloads usingtwo other meth-
ods: log-log load-settlement curvemethod (DeBeer
1967) andtwo-slopecriteriondescribedinNAVFAC
(1986). Hencethemeasuredbearingcapacityhasbeen
establishedusingtheMinimumSlopefailurecriterion.
A total of 267 load test cases in/on granular soils
couldbeusedfor theevaluationof uncertaintiesinthe
bearingcapacityemployingtheMinimumSlopefail-
urecriterion. Outof the267cases, 172foundationsare
under vertical-centric, 42under vertical-eccentric, 39
underinclined-centricand14underinclined-eccentric
loadings.
Fourteenoutof the172footingcasesundervertical-
centric loadings are in natural soil conditions (for
which SPT blow counts are available) from8 sites.
The average friction angles and unit weights are in
theranges29.8
to39
and14.5kN/m
3
to19.7kN/m
3
,
respectively. There are 158 cases in soils with con-
trolled soil conditions in7sites. Theaveragefriction
angles of thesesoils rangefrom34.6
to 46.0
, with
mostcaseslyingbetween42
and46
.Theaverageunit
weights range from10.2kN/m
3
to 18.4kN/m
3
. The
widthof footingsundervertical-centricloadingranges
from5cmto 3m, withnearly half of thefootings of
size9cm9cm; 104 of thefootings aresquare, 63
arerectangular and5arecircular.
Theloadtestsunder all other loadings, i.e. vertical-
eccentric, inclined-centric and inclined-eccentric,
have been carried out in controlled soil condi-
tions. The averages of the soil friction angles for
98
Figure 2. Positive moment (upper) and negative moment
(lower) for inclined-eccentricloadings.
vertical-eccentric, inclined-centric and inclined-
eccentricloadtestsare42.0
, 43.4
and44.9
, respec-
tively, while the averages of the soil unit weights
are16.8kN/m
3
, 17.2kN/m
3
and17.4kN/m
3
, respec-
tively. The footing widths range from5cmto 1.0m
for vertical-eccentric and inclined-centric load tests
whiletheinclined-eccentric loadtestsarecarriedout
onfootings9cmwide.
The loadings in the inclined-eccentric load tests
havebeenappliedintwoways: (a) inradial loadpath
and(b)instep-likeloadpath.Intheradial loadpath,the
loadcomponentsaremutuallyincreasedsuchthatboth
theloadeccentricityandtheloadinclinationangleare
kept constant until failure. Inthestep-likeloadpath,
thevertical loadingiskeptconstantwhilethehorizon-
tal loading is gradually increased till failure. Hence,
for thestep-likepathloading, thefailureloadcanbe
interpretedfromtheload-displacement curvefor the
horizontal directiononly. Inaddition, for thecaseof
inclined-eccentric loading, dependingontherelative
directionsof loadeccentricityandinclinationasshown
inFigure2, thebearingcapacityof thefoundationmay
increaseor decrease. Theupper combination in Fig-
ure2isthecaseof positiveor reversiblemoment, and
thelower isthecaseof negativemoment. Thefounda-
tionBC is greater inthecaseof positivemoment as
comparedtothatwithanegativemomentfor thesame
magnitudeof loadeccentricitiesandinclinations.
2.3 Calculated bearing capacity
Theequation specified inAASHTO (2007) given in
Equation1, whichisbasedonVesi c(1975), hasbeen
usedtocalculatethebearingcapacity of afootingof
lengthL andwidthB supportedbyasoil withcohesion
c, averagefrictionangle
f
andaverageunitweight.
where q
u
is the calculated bearing capacity,
N
cm
=N
c
s
c
i
c
, N
qm
=N
q
s
q
d
q
i
q
, N
m
=N
andC
wq
andC
w
arethereductioncoefficientsforthepresence
of groundwater. For the depth of groundwater table
from the ground surface D
w
=0.0, C
wq
=0.5 and
C
wq
=1.0whenD
w
=footingembedment depth(D
f
)
or below. C
w
=0.5for D
w
D
f
, and1.0whenD
w
is
greater than1.5B+D
f
, withthevalues for interme-
diategroundwater locationdepthsinterpolated.
For granular soils, c =0andhenceonly theterms
withN
qm
andN
m
inEquation1comeintothepicture.
Theequationsfor thebearingcapacityfactorsN
q
and
N
, basedonproposalsby Reissner (1924) andVesi c

(1973, 1975) arerespectively.
Theshapefactorss
i
usedinthepresent calculation
arethoseproposedbyDeBeer (1961, 1970) andVesi c
(1973), whicharealsousedinAASHTO(2007).
For depthfactorsd
i
, it islogical touseaconsistent
set of equations givenby thesameauthor, at present
givenbyVesi c (1973, 1975). Hence, theproposal by
BrinchHansen(1970) andVesi c(1973) for depthfac-
tor d
q
areusedinsteadof thediscretevaluesprovided
inAASHTO(2007).
Forinclinedloadingcases, thefollowingloadincli-
nationfactors i
i
proposedbyVesi c (1975) havebeen
used.
whereH andV arethehorizontal andvertical compo-
nentsof theappliedinclinedloadP, c issoil cohesion;
and
where istheprojecteddirectionof loadintheplane
of the footing, measured fromthe side of length L
indegrees. Incaseof eccentric loading, theeffective
footingdimensionsL
=L 2e
L
andB
=B2e
B
are
aretobeusedinEquations1through6insteadof the
full footingdimensionBL.
For the bearing capacity calculations, the soil
strengthparametersaretakenastheweightedaverages
99
of thestrengthparameterstoadepthof twicethefoot-
ing width, belowthefooting base. For footing cases
withmissingreportedsoil parameters, correlationsof
the parameters with SPT values have been used for
estimation. A correlationgivenby Peck, Hansenand
ThornburnasmodifiedbyKulhawy& Mayne(1990)
hasbeenusedtoestimatethesoil frictionangles, while
for the soil unit weight, a correlation proposed by
Paikowsky et al. (2005) has been used as given by
Equations7and8, respectively.
where the corrected values of N
60
for overburden,
(N
1
)
60
, havebeenobtainedbasedontheproposal by
Liao&Whitman(1986):
where p
a
is the atmospheric pressure (100kPa or
1tsf) and
v
istheeffectiveoverburdenpressureinthe
sameunit asthat of theatmosphericpressure.
2.4 Summary of mean bias values
The uncertainties in the aforementioned design
method areexpressed as abias, defined as theratio
of themeasuredoverthecalculatedbearingcapacities.
Thislumpedvalueincludesall sourcesof uncertainties
intheBC predictionsuchas themodel uncertainties
(e.g. BC factors, foundationscaleeffects etc), varia-
tionof soil propertiesandtheirinterpretation, capacity
interpretationetc. Thebiases inthis present research
have been studied according to the loading types,
namely vertical-centric, vertical-eccentric, inclined-
centric andinclined-eccentric loadings as well as the
natureof thesoil, differentiatingbetweennatural and
controlled soil conditions.
Vertical-Centric Loading:
Themeanandcoefficientof variations(COV)of the
biasesarecalculatedfor the172total vertical centric
loadingcases. Figure3summarizesthemeanof bias
grouped by test soil conditions and footing widths.
The mean bias for the footings in natural soil con-
ditions is found to bearound 1.0 irrespectiveof the
footingsizes(thelargestfootingtestedbeingof 3.0m
width). Incontrast, for thefootingsincontrolled soil
conditions the mean bias value is about 1.7 to 2.2.
For footingsincontrolledsoil conditions, thereisless
variationinthebiasesforsmall footings(B0.1m)as
comparedtothelarger footings, eventhoughthetests
are froma larger number of sites. Compared to the
testsincontrolledsoil conditions, thebiasesfor those
in natural soil conditions havehigher variation even
when thenumber of sites is comparable. Thehigher
mean bias reflect conservatism(under-prediction) in
the theoretical prediction of the BC factor N
by
Vesi c(1973), whichwasfoundtorepresent thelower
boundof theback-calculatedvaluesfromtheloadtests
Figure 3. Summary statistics of the bias (measured over
calculatedBC) for footingsunder vertical-centricloadings.
especiallyintherangeof soil frictionanglesbetween
42
and 46
. The bias of N
(back-calculated from
testsoverVesi c) isfoundtoincreasefrom0.95to2.18
for anincreaseinthefrictionanglefrom42
to 46
.
In both thenatural and controlled soil conditions, it
wasfoundthat thereisatrendof increaseinthebias
withanincreaseinthefootingsize. It may behence
logicallystatedthattheevaluationof uncertaintiesfor
small scalemodel tests to failurearevalidfor proto-
typelargefootingstoo.Thisinferenceisof importance
especiallyfor thecasesunder combinedloadings, for
whichthetestingonlargescalemodelsarelimitedby
economical andpractical reasons.
Other Loadings:
For the42casesof vertical-eccentric loadings, the
biasisfoundtohaveameanof 1.81andCOV of 0.349.
For the39casesof inclined-centricloadings, thebias
is found to haveamean of 1.43 and COV of 0.295.
There are 8 cases of positive or reversible moment
forinclined-eccentricloadingsand6casesof negative
moment; themeanandCOV fortheformerisfoundto
be1.41and0.278, respectively, andfor thelatter 2.03
and0.094, respectively.
3 RELIABILITY ANALYSES
3.1 Typical bridge foundation loading and load
factors
Theloadingconditionhasbeentakenasthat usedby
Paikowsky et al. (2004) inestablishingtheLRFDfor
deep foundations. The load combination defined as
Strength I in AASHTO is applied as follows in its
primaryform.
100
Table2. Statistical detailsof thebiasesof thebearingcapacitiesof shallowfoundationsin/ongranular soilsandresistance
factorsunder different loadings.
Resistancefactor
Underlying No.of No.of Mean
Loadingtype soil conditions cases sites bias COV
MCS FOSM Recommended

Vertical-centric Controlled 158 7 1.73 0.271 0.937 0.793 0.90
Natural 14 8 1.00 0.329 0.457 0.396 0.45
Vertical-eccentric Controlled 42 4 1.81 0.349 0.779 0.680 0.75
Inclined-centric Controlled 39 3 1.43 0.295 0.722 0.617 0.70
Inclined-eccentric
Positiveor Controlled 8 1 1.41 0.278 0.748 0.635 0.70
reversiblemoment
Negativemoment Controlled 6 1 2.03 0.094 1.773 1.318 1.00
Note: =bias=measuredover predictedCOV
=coefficient of variationof thebiasMCS=MonteCarloSimulationusing

100,000iterationsFOSM=First Order SecondMoment
whereRistheresistanceorbearingcapacityof shallow
foundation, D is the dead load and LL is vehicular
liveloads.Thestatistical characteristicsof therandom
variablesD andLL areassumedtobeasthoseusedin
NCHRPReport368(Nowak, 1999). Theloadfactors,
L
for liveloadand
D
for deadloadfromAASHTO
(2007) (Tables3.4.1-1and3.4.1-2), andthestatistical
characteristicsusedareasgivenbelow.
Further,Paikowskyetal.(2004)examinedtheinflu-
enceof thedeadloadtoliveloadratiosdemonstrating
very littlesensitivity of theresistancefactors to that
ratio, with overall decrease of the resistance factors
with theincreasein thedead load to liveload ratio.
Thelargedead-to-liveloadratiosrepresentconditions
of bridgeconstruction, typicallyassociatedwithvery
long bridge spans. The relatively small influence of
thedead-to-live-loadratioontheresistancefactorlead
Paikowskyet al. (2004), thereby, touseatypical ratio
of 2.0 knowing that the obtained factors are by and
large applicable for long span bridges being on the
conservative side. This ratio was adopted, therefore,
for thepresent studycalibrationsaswell.
Equation10abovedoesnotincludetheeffectsof the
horizontal earthpressure, hence, theresistancefactors
developedandpresentedinthis paper donot include
theuncertaintiesduetothehorizontal earthpressure.
Except for the footings under vertical-centric load-
ing, theresistancefactorswouldlikelytobedifferent
especially for the cases under the inclined-eccentric
loadings when the lateral loading due to horizontal
earthpressureisalsoconsidered.
3.2 Resistance factor calibration
The preliminary resistance factors based on the
evaluationof biasesintheultimatelimitestimationof
shallowfoundationsin/ongranular soils, presentedin
theprevioussection, havebeencalibratedfor atarget
reliability
T
of 3.0or target exceedanceprobability
of 0.135%, assuminglognormal distributionsforloads
andresistance.
The resistance factors obtained from the FOSM
(original AASHTOcalibrationprocedure, Barkeretal.
1991) and from the Monte Carlo simulation using
500,000 simulations, along with the recommended
factors, for shallow foundations under the different
loadingtypesarepresentedinTable2.
4 SUMMARY ANDCONCLUSION
TheNCHRP 24-31research project aims to develop
LRFDproceduresandtomodifythecurrentAASHTO
design specifications for the Ultimate Limit State
(ULS) design of bridge shallow foundations. The
research study utilizes a comprehensive database of
loadteststoestablishtheuncertaintyinbearingcapac-
ity (BC) calculation. The failure loads in the model
testshavebeendeterminedbydifferentfailurecriteria,
amongwhichtheminimumslopecriterionproposed
by Vesi c (1963) and was employed for the database
loadtestsinterpretation.Thiscriterionwasestablished
asthemostappropriateonetoidentifythefailureload.
Theuncertaintiesinthedesignmethodwereexpressed
by thebiasdefinedastheratioof measuredover cal-
culatedBCs. This lumpedvalueincludes all sources
of uncertaintiesintheBCpredictionoriginatingfrom
themodel (e.g. BC factors, scaleeffects), variationin
soil properties, etc.
The bearing capacity analysis of shallow founda-
tions oncontrolledsoil conditions was foundto sys-
tematicallyunder-predictthemeasuredcapacityforall
examinedloadcombinations,namely;vertical-centric,
vertical-eccentric,inclined-centricandinclinedeccen-
tric.Thisunder-predictionresultswiththebiasvarying
between1.4and2.0withCOV valuesof approximately
0.3(excludingthelimitedcasesof inclined-eccentric
loadingunder negativemoment).
Investigation of the bearing capacity factor N
suggeststhatasimilar biasexistsinthisfactor assum-

ing all other factors are known and measured. The
bias in N
was found to be related to the soil inter-

nal friction angle, where the bias increases (i.e. BC
under-prediction increases) with the increase in the
101
internal frictionangle. Thecontrolledsoil conditions
of theexaminedcasessuggest that thebiasintheBC
factor N
may explaininalargepart theuncertainty

intheBC calculations. Thisstatement anditsvalidity
iscurrentlyfurther evaluated.
Thefindingsrelatedtothefoundationstestedin/on
natural soils is more difficult at it introduces larger
variability in the soil type and strength parameters
interpretation. Itisnoticeablehoweverthattheinternal
frictionangles associatedwithall of thetestedcases
onnatural soils resultedwithinternal frictionangles
lower thanthosefor controlledsoil andhencematch
theareaforwhichtheBCparameterN
didnotexhibit
abias. Thoughqualitativelytheseobservationsmatch
thedataanalyzed, theimplementationof theindicated
findings for natural soil conditionis thus incomplete
at thisstage.
Inboththenatural andcontrolledsoil conditions,
however, therewasatrendof increaseinthebiaswith
anincreaseinthefootingsize.Itmaybehencelogically
stated that the evaluation of uncertainties for small
scalemodel teststofailurearevalidforprototypelarge
footingstoo.Thisinferenceisof importanceespecially
for thecasesunder combinedloadings, for whichthe
testingonlargescalemodelsarelimitedbyeconomical
andpractical reasons.
Theinitial investigationleadstotheconclusionthat
one single resistance factor for the bearing capac-
ity is not sufficient as different soil and/or loading
conditions result in different level of uncertainties.
Hence, different resistance factors were established
based on probabilistic analyses (FOSM and Monte
Carlosimulations), eachfor vertical-centric, vertical-
eccentric, inclined-centric and inclined-eccentric
loadingconditions.
DISCLAIMERANDACKNOWLEDGEMENTS
Thepresented results arepreliminary findings of an
ongoing research; the final results may be changed
due to further ongoing investigation. The presented
researchwassponsoredby theAmericanAssociation
of StateHighwayTransportationOfficials(AASHTO)
under project NCHRP 24-31awardedtoGeosciences
Testing and Research, Inc. (GTR) of North Chelms-
ford, Massachusetts. The presented parameters are
preliminary and do not reflect the recommended or
approved parameters. The opinions and conclusions
expressed or implied in this paper are entirely of
those of the research agency and not necessarily
those of theTransportation Research Board (TRB),
the National Research Council (NRC), the Federal
HighwayAdministration(FHWA) orAASHTO.
Thecollectionandanalysisof databaseUML-GTR
ShalFound07 was conducted under a contract with
theGeotechnical Research Laboratory of University
of Massachusetts Lowell by the graduate students
Ms. Yu Fu, Mr. J enia Nemirovsky and postdoctoral
fellowDr. ShailendraAmatya. Thehelp and dedica-
tionof Ms. Mary Canniff inconductingtheresearch
isappreciated.
REFERENCES
AASHTO. 2007. Section 10: Foundations, LRFD Bridge
Design Specifications,AmericanAssoc. of StateHighway
andTransportationOfficials, Washington, D.C.
Barker, R.M., Duncan, J.M., Rojiani, K.B., Ooi, P.S.K., Tan,
C.K. & Kim, S.G. 1991. Manuals for the Design of
Bridge Foundations, NCHRP Report 343, Transportation
ResearchBoard, National ResearchCouncil,Washington,
D.C., 308p
Briaud, J.L. & Gibbens, R. 1997. Large scale load tests and
database of spread footings on sand, USDoT, Federal
Highway Association, Publication No. FHWA-RD-97-
068, Washington, D.C., 228p.
BrinchHansen, J. 1970. A RevisedandExtendFormulafor
BearingCapacity.Akademiet for deTekniskeVidenskaber,
GeotekniskInstitut, BulletinNo.28, Copenhagen: 511.
De Beer, E.E. 1967. Proefondervindelijke bijdrage tot
de studie van het gransdragvermogen van zand onder
funderingen op staal; Bepaling von der vormfactor
sb. Annales des Travaux Plublics de Belgique, 68(6):
481506.
De Beer, E.E. 1970. Experimental Determination of the
ShapeFactorsandtheBearingCapacityFactorsof Sand.
Gotechnique, 20(4): 387-411.
Kulhawy, F. &Mayne, P. 1990. Manual on Estimation of Soil
Properties for FoundationDesign,ReportEPRI-EL-6800,
ElectricPower ResearchInstitute, PaloAlto, Calif.
Lutenegger,A.J. &DeGroot, D.J. 1995. Settlement of shallow
foundations on granular soils, ReportUMTC-94-4, Univ.
of MassachusettsTransportationCenter, Amherst, Mass.,
USA (for Massachusetts Highway Department, Trans-
portationresearchproject Contract #6332Task#4).
Muhs, H. & Weiss, K. 1972. Der Einfluss von Nei-
gung und Ausmittigkeit der Last auf die Grenztrag-
faehigkeit flachgegruendeter Einzelfundamente, Degebo-
Mittelungen, Heft 28(inGerman).
NAVFAC. 1986. Naval Facilities Engineering Command
Desing Manual DM7.01.
Nowak, A. 1999. Calibration of LRFD Bridge Design Code,
NCHRP Report 368. Transportation Research Board of
theNational Academies, Washington, D.C.
Paikowsky, S.G., Birgisson, B., McVa, M., Nguyen, T., Kuo,
C., Baecher, G., Ayyub, B., Stenersen, K., OMalley,
K., Chernauskas, L. & ONeill, M. 2004. Load and
Resistance Factor Design (LRFD) for Deep Foundations,
NCHRP Report 507. Transportation Research Board of
theNational Academies, Washington, D.C., 126pp.
Paikowsky, S.G., Fu, Y. & Lu, Y. 2005. LRFD foundation
design implementation and specification development,
NCHRP Project 20-07, Task 183, National Coopera-
tiveHighwayResearchProgram,TransportationResearch
Board, National ResearchCouncil, Washington, D.C.
Perau, E.W. 1995. Ein systematischer Ansatz zur Berechnung
des Grundbruchwiderstands von Fundamenten,
Mitteilungenaus demFachgebiet GrundbauundBoden-
mechanik, Heft19der UniversitaetEssen, editedbyProf.
Dr-Ing. W. Richwien(inGerman).
Vesi c, A. 1963. Bearing capacity of deep foundations in
sand, Highway Research Record 39, National Academy
of Sciences, National ResearchCouncil: 112153.
Vesi c, A. 1973. Analysis of ultimate loads of shallow
foundations, J. of the Soil Mechanics and Foundations
Div., SM1, ASCE, 99(1): 4573.
Vesi c, A. 1975. Bearing capacity of shallow foundations.
In H.F. Winterkorn & H.Y. Fang (eds.) Foundation
Engineering Handbook: 121147. NewYork: Van Nos-
trandReinhold.
102
Limit statesdesignconceptsfor reinforcedsoil wallsinNorthAmerica
R.J. Bathurst & B. Huang
GeoEngineering Centre at Queens-RMC, Royal Military College of Canada, Kingston, Canada
T.M. Allen
Washington State Department of Transportation, Olympia, Washington, USA
ABSTRACT: Limit statesdesign(LSD) calibrationfor reinforcedsoil retainingwall design(calledMechani-
callyStabilizedEarthWallsinUSA terminology) hasuntil recentlybeenrestrictedtocomparisontoallowable
stressdesign(ASD) practice(oftencalledcalibrationbyfitting). Thispaper presentsLSDcalibrationprinci-
plesandtracesthestepsrequiredtoarriveat loadandresistancefactorsusingclosed-formsolutionsor Monte
Carlosimulationfor onetypical limit state pullout of steel reinforcement elementsintheanchoragezoneof a
reinforcedsoil wall.A uniquefeatureof thispaperisthatmeasuredloadandresistancevaluesfromadatabaseof
casehistoriesareusedtodevelopthestatistical parametersintheexample. Furthermore, model biasisconsidered
explicitlyinthecalculations. Thepaper alsoaddressesissuesrelatedtotheinfluenceof outliersinthedatasets
andpossibledependencesbetweenvariablesthat canhaveanimportant influenceontheresultsof calibration.
1 INTRODUCTION
The United States and Canada are committed to a
limit states design(LSD) approachfor thedesignof
geotechnical structures. IntheUnitedSates theterm
loadandresistancefactordesign(LRFD)isused.Most
engineers are familiar with LSD practice based on
experiencewithcivil engineeringdesigncodes. How-
ever, the methodologies to carry out calibration are
less well understood and this has proven to be an
impediment to acceptanceof LSD for thedesign of
conventional retainingwall structures. Thechallenge
to adopt LSD for modern reinforced soil wall struc-
turesisfurthercompoundedbycomplexityof thelimit
statemechanisms,poorpredictionaccuracyof someof
theunderlyingdeterministic models andlack of data
toperformrigorouscalibration.
In this paper we present LSD calibration princi-
plesandtracethestepsrequiredtoarriveat loadand
resistancefactorsusingclosed-formsolutionsfor one
typical limitstate pulloutof steel reinforcementele-
mentsintheanchoragezoneof areinforcedsoil wall.
A uniquefeatureof this paper is that measured load
andresistancevaluesfromadatabaseof casehistories
are used to develop the statistical parameters in the
example. Furthermore, thestatisticsarecomputedfor
model bias which allows theaccuracy of theunder-
lyingdeterministic models for loadandresistanceto
beconsideredquantitatively inLSD calibration. The
paper also addresses issues related to the influence
of outliers inthedatasets andpossibledependences
betweenvariablesthatcanhaveanimportantinfluence
ontheresultsof calibration.Anexpandedtreatmentof
thesamegeneral topiccanbefoundinthereferences
(Bathurst et al. 2008, Allenet al. 2005).
2 BACKGROUND
InNorthAmerica, LSD (or, loadandresistancefac-
tor design LRFD) is basedonafactoredresistance
approach. Thegeneral approachcanbeexpressedas:
where: Q
ni
=nominal (specified) load, R
n
=nominal
(characteristic) resistance,
i
=load factor, and =
resistance factor. In the North American approach,
uncertainty in the calculation of the resistance side
of theequationiscapturedbyasingleresistancefac-
tor while load contributions are assigned (typically)
different loadfactors. Loadfactor terms havevalues
i
1, whiletheresistancetermshouldhaveavalue
1.
It is important to note that there is a fundamen-
tal differencebetweenNorthAmericanandEuropean
approaches to LSD. A factored strength approach is
usedinEurope(Eurocode7) whereresistancefactors
areappliedtoindividual parametersinresistance-side
expressions. Examples of NorthAmericancodes are
the AASHTO LRFD Bridge Design Specifications
(AASHTO2007), National BuildingCodeof Canada
(NRC2005)andtheCanadianHighwayBridgeDesign
Code(CHBDC 2006). Thegeneral approachfor LSD
of earthstructuresdescribedintheUK standard(fac-
toredstrengthapproach) hasbeenadoptedbyagencies
103
inAustralia(RTA 2003) and Hong Kong (Geoguide
6 2002). An excellent discussion of the difference
in approaches between Europe and North America
and thedevelopment of LSD methods in geotechni-
cal engineeringcanbefoundinthepaper by Becker
(1996).
3 EXAMPLE LIMIT STATE
Inthispaperweusetheexampleof pulloutof steel grid
reinforcement for theinternal stability of reinforced
soil walls. TheresistancetermR isexpressedas:
where:T
po
=pulloutcapacity(kN/m),L
e
=anchorage
length(m),
v
=
s
z=vertical stress(kPa),
s
=soil
unit weight (kN/m
3
), andF
=dimensionless pullout
resistancefactor (inthis case, F
is afunctionof the
thicknessandhorizontal spacingof thereinforcement
transverse bars, and depth z of the reinforcement)
(AASHTO2007).
TheloadtermQiscomputedas:
where: T
max
=maximum tensile load in the rein-
forcement (kN/m), S
v
=the tributary spacing of the
reinforcement layer (m), and K
r
= dimensionless
lateral earth pressure coefficient acting at the rein-
forcement layer depth. For steel grid reinforced soil
walls, K
r
varies from 2.5K
a
to 1.2K
a
at the top
of the wall to a depth of 6m below the wall top,
respectively, andremains at 1.2K
a
below6m. Here,
K
a
=f()=dimensionlesscoefficientof activelateral
earthpressure. Notethatthesoil frictionangleiscon-
strainedto 40
accordingtoAASHTO(2007).This
restrictionisappliedincalculationsfor predictedload
valuesthat aredescribedlater.
UsingEquations1-3, thedesignlimitstateequation
for reinforcement pullout withasingleloadtermcan
beexpressedas:
wheretheloadfactor notation
Q
is nowadoptedfor
thecaseof asingleloadterminthelimitstatefunction.
FromEquation4theminimumpullout capacityis:
Inotherwords,thenominal resistancevalue(R
n
=T
po
)
must always be greater than the nominal load value
(Q
n
=T
max
) byafactor of
Q
/.
4 CALIBRATION
4.1 Probability of failure
An important first step in limit states calibration is
theselection of probability of failure. Thechoiceof
acceptableprobabilityof failurewill berelatedtothe
level of redundancyinthestructure. For example, for
pilegroups, theloss of onepiledoes not mean that
the group will fail since load shedding to the other
piles is possible. A similar argument can beapplied
totheinternal stabilitydesignof reinforcedsoil walls.
Loadsheddingfromasinglereinforcingelement that
hasfailedtotheremaininglayersispossible. For this
reasonaprobabilityof failureof p
f
=1in100recom-
mendedbyDAppolonia(1999) andPaikowskyet al.
(2004) for pilegroups is judgedto beapplicablefor
internal stability design of reinforced soil walls (i.e.
forpulloutandtensilerupturelimitstates) (Allenetal.
2005).
The probability of failure for a prescribed unfac-
toredlimitstateequationisillustratedinFigure1.The
reliabilityindex isusedtorelatethenormal statistics
for thedistributionof limit statevalues(computedas
g=R Q) to probability of failure. Load and resis-
tancefactorsareselectedthroughcalibrationtoensure
that factored values of R and Q result in an accept-
ableprobability of failure. A probability of failureof
p
f
=1/100correspondstoareliability index valueof
=2.33. A useful meanstoequatethesetwoparame-
tersistousethestandardnormal cumulativefunction
(NORMSDIST) inMicrosoft Excel asfollows:
Figure1. Probabilityof failureinreliability-baseddesign.
104
4.2 Model bias
Model biasisameasureof theaccuracy of predicted
values to measuredvalues usingalimit statemodel.
It isreasonabletoassumethat selectionof magnitude
of resistanceandloadfactorstomeet atarget value
(i.e. probability of failure) will beinfluenced by the
accuracyof theunderlyingdeterministicmodel. Inthis
paperwehaveintroducedtwosuchdeterministicmod-
els (onefor theload sideand onefor theresistance
side).Toquantifymodel accuracyandtocarryoutcal-
ibrationit isuseful tointroducetheratioof measured
to predicted values as a model bias, or bias for
brevity. For averygoodmodel andalargenumber of
datapoints, themeanbiasvalueisclosetooneandthe
coefficient of variation(COV) issmall. However, the
choiceof acceptableCOV valueissubjective. Theuse
of biasstatisticstoquantifytheaccuracyof loadpre-
dictionsinreinforcedsoil wallshasbeendemonstrated
byBathurst et al. (2008a,b).
Adoptingnomenclatureintroducedearliertheresis-
tancebiasvalueX
R
for apullout resistancedatapoint
canbeexpressedas:
andtheloadbiasvalueX
Q
as:
Meanandstandarddeviationcanbecomputedfor any
set of biasdata. ThemeanandCOV of theresistance
biasvaluesaredenotedhereasj
R
andCOV
R
, andfor
theloadbiasvaluesasj
Q
andCOV
Q
.
If thelimit statefunction is linear (thecasehere)
andtheloadandresistancebiasdataarenormallydis-
tributed, then canbecomputedusingthefollowing
formula:
If thebiasvaluesarelognormallydistributedthen:
A necessaryrequirement for thesecalculationsisthat
there are no dependencies between bias values and
correspondingloadandresistance(predicted) values.
Thisisdiscussedlater inthepaper.
Theequationsdevelopedabovedemonstratethatfor
our LSDcalibrationpurposes, statistical characteriza-
tion of thedistribution of actual load and resistance
values arenot used directly. Theadvantageof using
bias values is that variability in predicted load and
resistancevaluesresultingfromthemodel selectedfor
designisincludedexplicitlyinthesubsequentcalcula-
tionof loadandresistancefactors(Withiametal.1998,
Allenetal. 2005). Furthermore, if thedatabaseof mea-
suredloadandresistancevaluesaretakenfromactual
field measurements (for load) and laboratory mea-
surements(for resistance) then, inherent variabilityin
computedloadandresistancetermswill becaptured,
provided that the data represent typical quality and
typeof constructioninthefield, representativemate-
rial components, consistentlaboratorytechniquesand
thesiteconditionsfor thestructurebeingdesigned.
4.3 Statistical treatment of resistance and
load data
Predicted and measured pullout data are used here
to demonstrate how resistance bias statistics can be
computed and interpreted. The data are taken from
Christopher et al. (1989) and represent n=45 tests
withfivedifferent reinforcement products incombi-
nationwith15different granular soils. It may appear
temptingtousetheentiredatasettocomputethemean
andCOV bias values. However, this canleadtopoor
selectionof biasstatistics.Thebestapproachistosort
thebiasdatainrankorder fromsmallesttolargestand
thenplotthedataasacumulativedistributionfunction
(CDF) plotusingzX
R
axeswherezisthatstandard
normal variable, i.e. z=+
1
(p
f
). It canbecomputed
usingtheMicrosoft Excel NORMSINV function:
Here, i is the rank order of the data point and n is
the number of data. Resistance bias data plot close
to astraight lineona(CDF) plot withsemi-logaxes
(Figure 2) indicating that the data are sensibly log-
normally distributed. Note that lognormal statistics
canbecomputeddirectlyusingdatapointsconverted
to natural logarithms. If thedatadivergefromaper-
fectlognormal distributiontheremaybediscrepancies
Resistance bias, X
R
0.4 0.6 0.7 0.8 2 3 4 1.0
S
t
a
n
d
a
r
d

n
o
r
m
a
l

v
a
r
i
a
b
l
e
,

z
22
21
0
1
2
filtered data points
Resistance bias values (n = 45)
Filtered resistance bias values (n = 41)
Predicted lognormal distribution to filtered data set
using normal statistics (n = 41)
Predicted lognormal distribution - best fit-to-tail
of lognormal data (n = 45)
Lognormal prediction from normal statistics
(n =45)
Figure2. Cumulativedistributionfunction(CDF) plots of
resistancebias(X
R
) valuesfor pulloutcapacityfor steel grid
reinforcement andfittedapproximations.
105
between statistics computed in these two different
ways(Bathurst et al. 2008). However, thedifferences
herearesmall andit is convenient tocharacterizeall
distributionsusingnormal meanandCOV values.
Two data sets are shown on this plot. All n=45
bias values areplottedas theopencircles. Thereare
a number of data points that are treated as outliers
andthereasonsforeliminatingthemfromanalysisare
discussed later. If the filtered data set is considered
only, thentheremainingn=41datapointsplot asthe
open squares in thefigure. Approximations to these
curvesaresuperimposedonthetwodatasetsandhave
beencomputedas:
withX usedheregenerically to calculatebias values
for z in a lognormal CDF plot described by normal
statisticsj andCOV.
Figure1ademonstrates that it is thelower tail of
theresistanceCDF plotthatisimportantsinceitisthe
overlapbetweenthelower tail of theresistancedistri-
butionandtheupper tail of theloaddistributionthat
governs the probability of failure. Simply using the
normal statisticsfor theentiren=45biasdatapoints
provides avisually poor fit to this important tail. By
adjustingtheselectionof meanandCOV of thebias
statistics it is possibleto performabest fit-to-tail as
shown in Figure2 and thus abetter match. Therea-
son for the poor fit using all of the data is that the
approximationisinfluencedbyfour datapointsat the
oppositetail. If thesedataareremoved(filteredout)
thentheresultingmeanandCOV of thebiasstatistics
(n=41) give a visually better match with the lower
tail. Examination of thetest conditions that resulted
inthesefour datapointsrevealedthat theycamefrom
thesametest series carried out under lowconfining
pressure(
v
-40kPa) withdensecompactedgranu-
lar soils(compactedunit weight of 14.9kN/m
3
) anda
frictionangleof 45degreesor more. Loadsgenerated
insteel reinforcement systemsaredifficult topredict
accurately for these conditions using current design
equationsduetocomplicatedsteel reinforcement-soil
interactioneffects(Bathurstetal. 2008, 2009). Hence,
the data for these four data points is at the limit of
therangeof valuesthatcanbeusefullyextractedfrom
conventional pullout testing.
A necessaryconditiontocarryoutcalibrationusing
model bias statistics is that there must not be hid-
den dependences between bias statistics (X=X
R
or
X
Q
) and the magnitude of nominal prediction (i.e.
R
npredicted
or Q
npredicted
inEquations 7and8) (Phoon
andKulhawy2003). Possibledependencybetweenthe
bias values and the magnitude of the nominal pre-
dictions can be examined using the Spearman rank
correlationcoefficient(Bathurstetal.2008),Kendalls
tau or Pearsons correlation coefficient. Using all
n=45 pullout resistance data points, the Spearman
rank correlation coefficient between X
R
and R
n
filtered data points
n = 13 n = 13 n = 19
n = 45 (total number of data points)
n = 41 (total less filtered data points)
predicted resistance, R
n
(kN/m)
0 20 40 60 80 100 120 140 160
r
e
s
i
s
t
a
n
c
e

b
i
a
s
,

X
R
0
1
2
3
4
Figure 3. Resistance bias values (X
R
) versus predicted
(nominal) resistance(R
n
) values.
(i.e. T
po
predicted) is =0.374, whichcorresponds
to aprobability p=0.0065that thetwo distributions
areindependent. Hence, thenull hypothesis(thedistri-
butions areindependent) is rejectedandthebias and
predicted load values are considered correlated at a
level of significance of 0.05. Removing these four
points and recalculating theSpearman rank correla-
tioncoefficient withn=41gives =0.242which
corresponds to a probability p=0.063 that the two
populationsareindependent at alevel of significance
of 0.05 (i.e. the null hypothesis that the two distri-
butions are independent cannot be rejected). Visual
evidenceof theeffect of thefour outliers onstatisti-
cal dependencycanbeappreciatedinFigure3. When
all data(n=45) areconsidered thereis adetectable
visual dependency between X
R
and R
n
. If the four
outliersareremovedthedependency disappearscon-
sistent with hypothesis testing described earlier. An
alternativestrategyistogroupthedataintoresistance
ranges as shown on the figure and compute normal
bias statistics for each range (Bathurst et al. 2008).
Thedisadvantageisthat thenumber of datapointsin
eachrangeismuchlessandahigher COV valuemay
result for anyindividual subset.
Allen et al. (2001, 2004) collected a data set of
20well-instrumentedreinforcedsoil wallsconstructed
with bar mat and welded wire steel reinforcement
(atotal of 34datapoints fromsix different wall sec-
tions constructedwithcompactedgranular backfill).
Thefriction angleused to calculatethelateral earth
pressure value K
r
in Equation 3 was taken as the
reported valuefromtriaxial or direct shear tests and
cappedat 40degreesasrequiredinAASHTO(2002,
2007) designcodes.
Usingtheentiredataset,thedistributionof loadbias
datapointsisreasonablywell approximatedbylognor-
mal distributions inFigure4. However, it is thedata
values intheupper tail of thedistributionthat areof
interest sinceit isthisregionwhichwill contributeto
thecalculation of probability of failurefor thesame
reasondescribedfor thedistributionof resistancebias
values (seeFigure1a). Anapproximationto theload
distributionusingalognormal fit to theupper tail is
106
load bias, X
Q
0.3 0.4 0.5 0.6 0.7 0.8 2 3 1
s
t
a
n
d
a
r
d

n
o
r
m
a
l

v
a
r
i
a
b
l
e
,

z
2
1
0
1
2
load bias values (n = 34)
lognormal prediction from normal statistics
predicted lognormal distribution
best fit-to-tail of lognormal data
Figure4. Cumulativedistributionfunction(CDF) plots of
loadbias (X
Q
) values for reinforcement loads for steel grid
reinforcedsoil wallsandfittedapproximations.
plotted in Figure 4. A test of dependency similar to
thatdescribedforresistancebiasdatarevealsthatload
biasvaluesandpredicted(nominal)loadvaluesarenot
correlatedat alevel of significanceof 0.05.
Basedonbestfit-to-tail for resistanceandloadbias
datapresentedinFigures2and4thefollowingnormal
statistics are proposed: j
R
=1.30, COV
R
=0.400,
j
Q
=0.973andCOV
Q
=0.462.
4.4 Estimating the load factor
As noted earlier in the paper it is desirable to have
aload factor
Q
>1. However, this may not bepos-
sible if the underlying deterministic model is poor.
Thishasbeenthecasefor current designmethodsfor
geosyntheticreinforcedsoil wallsthatusethetie-back
wedgemethod (Miyata and Bathurst 2007, Bathurst
et al. 2008). Fortunately, the prediction of load for
steel-reinforced soil walls is reasonably accurate as
demonstratedbyBathurst et al. (2008, 2009).
The following equation can be used as a starting
point toestimatetheloadfactor, if loadbiasstatistics
areavailable:
wheren
is aconstant. For agiven valueof n
, the
probabilityof exceedinganyfactoredloadisaboutthe
same. Thegreater thevalueof n
, thelower theprob-
ability the measured load will exceed the predicted
nominal load. A valueof n
=2for thestrengthlimit
state was used in the development of the Canadian
Highway BridgeDesign CodeandAASHTO LRFD
Bridge Design specifications (Nowak 1999, Nowak
andCollins 2000). This valueis usedintheexample
computationstofollow.
Using load bias statistics reported in theprevious
section gives 1.87 as astarting point. Bathurst et al.
(2008) reportedvaluesfrom1.73to1.87dependingon
variousminor adjustmentstotheselectionof loadand
resistancenormal statistics.A valueof 1.75isselected
here. A visual check on the reasonableness of the
Measured load (kN/m)
0 20 40 60 80 100 120 140
P
r
e
d
i
c
t
e
d

l
o
a
d

(
k
N
/
m
)
0
20
40
60
80
100
120
140
unfactored loads
factored loads (
Q
= 1.75)
1
1
Figure5. Predictedloadsversusmeasuredvaluesfor steel
grid reinforced soil walls using the AASHTO Simplified
Method.
selectedloadfactor canbecarriedoutbyplottingpre-
dictedloadvaluesagainstmeasuredvalues. Figure5
showsunfactoredandfactored(predicted) loadvalues
forsteel gridreinforcedsoil walls, usingtheAASHTO
SimplifiedMethod, plottedagainst measuredvalues.
Thefigureshowsthatmanyof theoriginal datapoints
arebelowthe1:1correspondenceline.Applyingaload
factor
Q
=1.75 moves almost all of thedatapoints
abovethe1:1line, whichisadesirableendresult.
4.5 Estimating the resistance factor using
a closed-form solution
Once the load factor is selected, the resistance fac-
tor canbeestimatedthroughiterationto producethe
desired magnitude for , using Equations 9 and 10
(as applicable), a design point method based on the
Rackwitz-Fiessler procedure (Rackwitz and Fiessler
1978), or the more adaptable and rigorous Monte
Carlo method demonstrated later. Here, the exam-
pledescribed earlier in thepaper is continued using
Equation10rewrittenasfollows:
Using the normal statistics for best fit-to-tail, load
factor
Q
=1.75 and a target reliability index value
of =2.33(probabilityof failurep
f
1/100) givesa
resistancefactor of =0.612. Fromapractical point
of view, =0.60isconvenient andsufficientlyaccu-
rate. If theresistancedataisparsedandrestrictedtothe
middlerangeof datapointsinFigure3, thenthecom-
putedresistancefactor is 0.81(Bathurst et al. 2008).
Clearly judgment isrequiredinany calibrationeffort
of thetypedescribedhere.
4.6 Trial and error approach using Monte Carlo
simulation
An alternative approach to select the resistance fac-
tor is to carry out a Monte Carlo simulation. It is
107
computationally convenient to assume a nominal
(mean) value of T
max
=1kN/m, then according to
Equation 5, thenominal (mean) valuefor resistance
must be equal to (
Q
/)1kN/m. The variation of
T
max
valuesislognormal andcanbequantifiedusing
the normal statistics for load bias (j
Q
, COV
Q
) and
for thefactoredresistancevaluesusing(
Q
/)j
R
and
COV
R
computed fromresistancebias values. Recall
that normal statisticscanbeusedtocomputelognor-
mal distributionswithsuitableaccuracy. Thefactored
limit stateequationisnowexpressedas:
Random values of R
i
and Q
j
can be computed
using:
and
where, z
i
and z
j
are randomvalues of the standard
normal variable. Randompairs of R
i
andQ
j
arethen
usedtocomputeasetof gvalues.Thesevaluesarethen
sortedinincreasingorder andaCDF plot constructed
asshowninFigure6.
Thecalculationscanbeeasilycarriedout usingan
Excel spreadsheet. Differentvaluesof
Q
and canbe
trieduntil thevalueof theCDF plotintersectsg=0at
thetargetvalueof althoughthegeneral approachisto
fixtheloadfactor(whichisoftenprescribed) andthen
adjust. Intheexamplehere, thenumerical approach
gives =2.35whichisvery closetothetarget value
of 2.33 used in theclosed-formsolution to compute
=0.60assumingaresistancefactor of 1.75.
g (kN/m)
3 2 1 0 1 4 6 8 10 11 12 13 14 15
s
t
a
n
d
a
r
d

n
o
r
m
a
l

v
a
r
i
a
b
l
e
,

z
5
4
3
2
1
0
1
2
3
4
5

2 3 5 7 9
Figure6. MonteCarlosimulationforthepulloutlimitstate,
usingtheAASHTOSimplifiedMethodof design(steel grid
wallsonly lognormal distributionassumedforT
max
andT
po
,
Q
=1.75, and
R
=0.60, and10,000valuesof ggenerated).
Anadvantageof MonteCarlosimulationisthatnor-
mal andlognormal distributionscanbeusedtogether.
Infact, anyfitteddistributionfunctioncanbeusedto
calculaterandomvalues of R
i
andQ
i
intheexample
usedhere.
5 CONCLUSIONS
This paper has demonstratedsomefundamental con-
cepts for LSD calibrationof geotechnical structures.
Anattempt has beenmadehereto break throughthe
obscurelanguagethat has traditionally been used in
LSD practice. Thepaper has highlightedtheneedto
carry out LSD calibration using model bias statis-
ticsinorder tocapturetheaccuracyof theunderlying
deterministicmodelsonLSDloadandresistancefac-
tors. Finally, the paper illustrates how simple Excel
spreadsheetscanbeusedtocarryoutcalibration, treat
dataoutliersandavoidhiddendependenciesbetween
variables.
REFERENCES
AASHTO.2007.LRFDBridgeDesignSpecifications,Amer-
ican Association of State Highway and Transportation
Officials, FourthEdition, Washington, D.C., USA.
AASHTO. 2002. Standard Specifications for Highway
Bridges, American Association of State Highway and
TransportationOfficials, 17thEdition, Washington, D.C.,
USA, 686p.
Allen, T.M., Bathurst, R.J., Holtz, R.D., Lee, W.F. &
Walters, D.L. 2004.A newworkingstressmethodfor pre-
dictionof loadsinsteel reinforcedsoil walls, ASCE J our-
nal of Geotechnical andGeoenvironmental Engineering,
Vol. 130, No. 1, 11091120.
Allen, T.M., Christopher, B.R., Elias, V. & DiMaggio, J.
2001. Development of thesimplifiedmethodfor internal
stability of Mechanically Stabilized Earth (MSE) walls,
Washington StateDept of Trans, Report WA-RD 513.1,
108p.
Allen, T.M., Nowak, A.S. & Bathurst, R.J. 2005. Calibration
to Determine load and resistance factors for geotechni-
cal andstructural design, TransportationResearchBoard
Circular E-C079, Washington, DC.
Bathurst, R.J., Miyata,Y., Nernheim, A. &Allen, T.M. 2008.
Refinement of K-Stiffnessmethodfor geosyntheticrein-
forced soil walls, Geosynthetics International, Vol. 15,
No. 4, 269295.
Bathurst, R.J., Allen, T.M. & Nowak, A.S. 2008. Calibration
concepts for load and resistance factor design (LRFD)
of reinforcedsoil walls, CanadianGeotechnical J ournal,
Vol. 45, 13771392.
Bathurst, R.J., Nernheim, A. &Allen, T.M. 2008. Compari-
sonof measuredandpredictedloads usingtheCoherent
GravityMethodforsteel soil walls, GroundImprovement,
Vol. 161, No. 3, 113120.
Bathurst, R.J., Nernheim, A., Miyata,Y. &Allen, T.M. 2009.
Predicted loads in steel reinforced soil walls using the
AASHTOSimplifiedMethod,ASCEJ ournal of Geotech-
nical andGeoenvironmental Engineering,Vol. 135, No. 2,
177184.
Bathurst, R.J., Miyata, Y., Nernheim, A. & Allen, T.M.
2008. Refinementof K-Stiffnessmethodforgeosynthetic
108
reinforcedsoil walls, GeosyntheticsInternational,Vol. 15,
No. 4, 269295.
Becker, D.E. 1996. EighteenthCanadianGeotechnical Col-
loquium: Limit states designfor foundations. Part I. An
overview of the foundation design process, Canadian
Geotechnical J ournal, Vol. 33, 956983.
CFEM.2006.CanadianFoundationEngineeringManual (4th
Ed). Richmond, BC, Canada.
CSA. 2006. Canadian Highway Bridge Design Code
(CHBDC), CSA Standard S6-06, Canadian Standards
Association, Toronto, Ontario, Canada.
Christopher, B.R., Gill, S.A., Giroud, J.-P., J uran, I.,
Mitchell, J.K., Schlosser, F. & Dunnicliff, J. 1989. Rein-
forced soil structures, Vol. II Summary of research and
systemsinformation, FHWA Report FHWA-RD-89-043,
158pp.
DAppolonia. 1999. Developing new AASHTO LRFD
specifications for retaining walls, Report for NCHRP
Project 20-7, Task 88, Transportation Research Board,
Washington, DC., USA 63p.
Eurocode 7, 1995, ENV 1997-1 Eurocode 7, Geotechni-
cal design, Part 1: General rules (with theUK National
Application Document), British Standards Institution,
London.
Geoguide6, 2002, Guidetoreinforcedfill structureandslope
design, Geotechnical Engineering Office, Hong Kong,
China.
Miyata,Y. &Bathurst, R.J. 2007. Developmentof K-stiffness
methodfor geosyntheticreinforcedsoil wallsconstructed
with c- soils, Canadian Geotechnical J ournal, Vol. 44,
No. 12, 13911416.
Nowak,A.S.1999.Calibrationof LRFDBridgeDesignCode,
NCHRP Report 368, Transportation Research Board,
Washington, DC, USA.
Nowak, A.S. &Collins, K.R. 2000, Reliabilityof Structures,
McGrawHill, NewYork, NY.
NRC, 2005. National Building Code. NRC of Canada,
Ottawa, Ontario, Canada.
Paikowsky, S.G., Birgisson, B., McVay, M., Nguyen, T.,
Kuo, C., Baecher, G., Ayyub, B., Stenersen, K.,
OMalley, K., Chernauskas, L. &ONeill, M. 2004, Load
andresistancefactordesign(LRFD)fordeepfoundations,
NCHRP Report 507, Transportation Research Board of
theNational Academies, Washington, D.C., 126p.
Phoon, K-K. & Kulhawy, F.H. 2003. Evaluation of model
uncertainties for reliability-based foundation design.
Applications of Statistics and Probability in Civil Engi-
neering (Der Kiureghian, Madanat and Pestana (eds).
Millpress, Rotterdam, Netherlands.
Rackwitz, R. & Fiessler, B. 1978. Structural reliability
under combinedrandomloadsequences, Computersand
Structures, Vol. 9, 484494.
RTA 2003, Design of Reinforced Soil Walls, QA Specifi-
cation R57, Roads andTraffic Authority of New South
Wales, Australia.
Withiam, J.L., Voytko, E.P., Barker, R.M., Duncan, J.M.,
Kelly, B.C., Musser, S.C. & Elias, V. 1998, Load
and Resistance Factor Design (LRFD) for Highway
BridgeSubstructures, FHWA HI-98-032, Federal High-
wayAdministration, Washington, DC, USA.
109
Lossof staticequilibriumof astructure Definitionand
verificationof limit stateEQU
B. Schuppener
Federal Waterways Engineering and Research Institute, Karlsruhe, Germany
B. Simpson
Arup Geotechnics, London, UK
T.L.L. Orr
Trinity College, Dublin University, Ireland
R. Frank
Universit Paris-Est, Ecole nationale des ponts et chausses, Navier-CERMES, Paris, France
A.J. Bond
Geocentrix, Banstead, Surrey, UK
ABSTRACT: In order to satisfy the essential requirements for construction works, the Eurocodes require
that structures fulfil thedesign criteriafor both serviceability and ultimatelimit states. Threeultimatelimit
statesareof particular importanceingeotechnical design: lossof staticequilibrium(EQU), failureor excessive
deformationof thestructure(STR), andfailureor excessivedeformationof theground(GEO). This paper is
concernedwithEQU.Theproblemsrelatingtotheuseof GEOingeotechnical designaredescribedinthispaper
andalternativeviewsarepresented. Theauthorshavesought, bymeansof anumber of illustrativeexamples, to
examinetheseproblemsandclarifyissuesrelatingtotheapplicationof EQU.
1 INTRODUCTION
In order to satisfy the essential requirements for
constructionworks, theEurocodesrequirethat struc-
tures fulfil the design criteria for both serviceabil-
ity and ultimate limit states (SLSs and ULSs). It
is a principle of the Eurocodes that ultimate limit
states shall be verified as relevant. Three ultimate
limit states are of particular importance in geotech-
nical design: loss of static equilibrium(EQU), fail-
ureor excessivedeformationof thestructure(STR),
and failure or excessive deformation of the ground
(GEO).
This paper is concerned with EQU, which the
authorshavedebatedatlengthhowtoapplyingeotech-
nical design, encountering problems which are also
relevant to structural design. Particular difficulties
arise when the stability required by EQU has to be
augmented by structural or ground resistance. The
problems relatingto theuseof GEO ingeotechnical
designaredescribedinthispaperandalternativeviews
arepresented. Theauthors havesought, by means of
a number of illustrative examples, to examine these
problemsandclarifyissuesrelatingtotheapplication
of EQU.
2 DEFINITIONS
Ultimatelimit states STR andGEO aredescribedin
EurocodeBasisof design(EN 1990) andEurocode7
Geotechnical design Part1General rules (EN1997-1)
bythefollowingdefinitions:
STR: Internal failure or excessive deforma-
tion of the structure or structural members,
includingfootings, piles, basement walls, etc.,
wherethestrengthof constructionmaterialsof
thestructuregoverns [EN 1990] (or) where
thestrengthof structural materialsissignificant
inprovidingresistance[EN1997-1].
GEO: Failureorexcessivedeformationof the
ground, inwhichthestrengthof soil or rock is
significant inprovidingresistance.
Intheverificationof bothSTR andGEO, it must be
shownthat ineverysection, member andconnection,
thedesign valueE
d
of theeffect of actions (such as
internal force, moment or avector representing sev-
eral internal forces or moments) does not exceedthe
designvalueR
d
of thecorrespondingresistanceof the
structureor theground:
111
Insertingpartial factors, thismaybeexpandedto:
where:
F
rep
istherepresentativevalueof anaction, derived
fromindividual characteristic actions takingaccount
of combination of variable actions (in the case of
variable actions Q, F
rep
=Q
rep
=Q
k
where is a
factor for converting the characteristic value to the
representativevalue; inthecaseof permanentactions,
F
rep
=G
rep
=G
k
)
X
k
isthecharacteristicvalueof agroundproperty
a
nom
isthenominal valueof geometrical data(i.e.
dimension)
E
and
F
arepartial factors for effects of actions
andactions, respectively
M
and
R
arepartial factorsfor groundproperties
andgroundresistance, respectively
Laisasafetymarginor tolerance
(Note: all valuesareusually1andLaistypically
zero.)
Limit stateEQU is describedinEN 1990andEN
1997-1bytwoslightlydifferent definitions:
EN1997-1, 2.4.7.1(1)Pgives: Lossof equilibrium
of thestructureor theground, consideredasarigid
body, in which thestrengths of structural materi-
als and the ground are insignificant in providing
resistance
EN 1990, 6.4.1(1)P gives: Loss of static equilib-
riumof thestructureor anypartof itconsideredas
arigidbody, where:
minor variations in thevalueor thespatial dis-
tribution of actions from a single source are
significant, and
thestrengthsof constructionmaterialsor ground
aregenerallynot governing.
Theconceptof asinglesource isimportant: EN1990
notes (seeNote3 inTableA.1.2 (B) of Annex A.1):
For example, all actions originating fromthe self
weight of thestructuremay beconsideredascoming
fromonesource; thisalsoappliesif differentmaterials
areinvolved.
As the definitions state that the strength of the
materialsor thegroundeither playsnopartinthever-
ificationor isnot governing, theexpressionfor EQU
isdifferentfromEquation(1). Forequilibrium, itmust
beverifiedthat thedesignvalueE
d,dst
of theeffect of
destabilisingactionsdoesnotexceedthedesignvalue
E
d,stb
of theeffect of stabilisingactions:
Insertingpartial factors, thismaybeexpandedto:
where:
F
rep,dst
andF
rep,stb
arerepresentativedestabilisingand
stabilisingactions, respectively
F,dst
and
F,stb
arepartial factorsfor destabilisingand
stabilisingactions, respectively
andtheothersymbolsareasdefinedforEquation(1a)
above
Alternatively, partial factors may be applied to the
effectsof actionsinstead:
where
E,dst
and
E,stb
arepartial factors for destabil-
isingandstabilisingeffectsof actions, respectively.
InAnnex A.1 of EN 1990 for buildings, the fol-
lowing partial factors are recommended in NOTE 1
of TableA1.2 (A):
G,dst
=
G,sup
=1.1 for destabil-
ising permanent actions G
dst
,
G,stb
=
G,inf
=0.9for
stabilising permanent actions, and
Q,dst
=1.5 for
destabilisingvariableactionsQ
dst
.
Insituationswheretheexpressionfor EQU cannot
besatisfied, EN 1990alsoallows theintroductionof
additional stabilisingterms inEquation(2) resulting
fromfor example, acoefficient of friction between
rigidbodies (EN 1990, 6.4.2(2)). Theseadditional
terms are an important issue for this paper. For
suchsituations, NOTE 2of TableA1.2(A) inAnnex
A.1 of EN 1990 also allows an alternative proce-
dure, subject to national acceptance, in which the
twoseparateverificationsof STR/GEOandEQU are
replaced by a combined EQU+STR/GEO verifica-
tion with recommended partial action factor values
of
G,dst
=1.35,
G,stb
=1.15, and
Q,dst
=1.50 (pro-
videdthat applying
G,dst
=
G,stb
=1.0doesnot give
amoreunfavourableeffect) combinedwith therele-
vant STR/GEOpartial material andresistancefactors
fromDesignApproachesDA 1(Combination1), DA 2
andDA 3.
EN 1997-1requires verificationof alimit stateof
static equilibriumor of overall displacements of the
structureor ground(EQU) by:
andanoteisadded: StaticequilibriumEQUismainly
relevant in structural design. In geotechnical design,
EQUverificationwill belimitedtorarecases, suchas
arigidfoundationbearingonrock, andis, inprinciple,
distinct fromoverall stability or buoyancy problems.
If any shearingresistanceT
d
isincluded, it shouldbe
of minor importance.
Indiscussionsbetweentheauthors, twoconcepts
for theinterpretationandapplicationof ENs1990and
1997-1havebeendeveloped.
Concept 1 proposes verifying only EQU in those
caseswherelossof static equilibriumisphysically
possiblefor thestructureor partof it, consideredas
arigidbody. Similarly Concept 1proposes verify-
ingonlySTR/GEOinsituationswherethestrength
112
of material or ground is significant in providing
resistance.
Concept 2 proposes verifying EQU in all cases;
it is interpreted as a load case. Where minor
strengthof material or groundisinvolved, thecom-
binedEQU/STR/GEOverificationmay beused, if
allowedbythenational annex.
Further discussioninvolves thetermT
d
inexpres-
sion EN 1997-1 2.4, particularly with the use of
Concept 1: it might beregardedeither asaresistance
(Concept 1-R) or asanaction(Concept 1-A).
Using Concept 1-R, the design resistance of the
anchor is:
and a similar approach is generally used with
Concept 2.
If load factors can be applied directly to action
effects, R
d
isgivendirectlyas:
whereE
Q,rep,dst
isE{Q
rep,dst
}.
UsingConcept 1-A andsubstitutingthecharacter-
isticvalueof astabilisingactionA
k,stb
(assumedtobe
permanent, from, for example, ananchor), expression
EN1997-12.4becomes:
Hence:
andA
k
isthenusedinaSTR/GEOverificationtoshow
that thedesignstabilisingactionA
d
canbeprovided
bytheresistanceR
d
of, for example, ananchor:
Whenvariableactionsaresignificantlylarger thanthe
permanent actions, thissimplifiesto:
whichismoreonerousthannormal GEOdesignfor
whichR
d
E
Q,rep,dst
Q,dst
.
ThedifferencebetweenConcepts1-R and1-A lies
intheratio
G
/
G,stb
(compareequations 3aand4b),
where
G
istheloadfactorinaSTR/GEOverification.
Example4, given below in section 6, illustrates this
difference.
EN 1997-1allowsnational standardsbodiestoset
values of partial factors for soil strength in EQU
Figure1. Balancedstructureonpiledfoundation.
that are different fromthose of STR/GEO. In con-
trast, no values aregivenfor EQU for partial factors
for geotechnical resistances, and none are offered
for structural materials or resistances in EN 1990; it
might beinferredthat theseunspecifiedmaterial and
resistance factors will be the same for EQU as for
STR/GEO.
Thefollowingexamples showhowthesediffering
concepts lead to different results in practical design
situations.
3 EXAMPLE 1: BALANCEDSTRUCTURE ON
PILEDFOUNDATION
3.1 General
Figure1showsabalancedstructuresittingonapiled
foundation. Intheverificationsit isassumedthat:
the representative values of the two forces W are
equal (=W
r
).
thecolumnandfootingareassumedtobeweightless
all structural componentsareof reinforcedconcrete
thestructureconsistsof asinglebeam, onecolumn
andtwopilesand
thereisnotransfer of bendingmoment tothepiles.
The details and the results of some verifications
proposedbytheauthorsareinpresentedinTable1.
Concept 1onlyrequireslimit statesSTR andGEO
to be considered, since failure can only occur from
113
a lack of strength in the structure or in the ground
surrounding thepiles. Thereis no possibility of loss
of equilibrium, providedthestructureandfoundations
arestrongenough.Itmightbeuseful insuchasituation
tointroducetheloadcasegivenbythefactorsof EQU
asanadditional STR verification. Alternatively, other
provisionsof ENs1992and1993mightdominatethis
problem(e.g. geometricallowancesetc).
Concept2requiresthespatial distributionof actions
fromtheself weightof thehorizontal element(asingle
source) tobeconsidered, soadditionallyEQUmustbe
verified.
3.2 Verification of limit states STR and GEO
If thereareno wind or snowloads to beconsidered
thecolumnwill onlycarrythevertical loadof theself
weight 2W
r
of thehorizontal beam. Thepartial factor
for permanent actions of
G
=1.35 must be applied
toW
r
to determinethedesignvalueof theeffects of
actionsfor thepiledesign.
Clause 5.2(1)P of EN 1992-1-1 states that the
unfavourable effects of possible deviations in the
geometry of the structure and the position of loads
shall be taken into account in the analysis of mem-
bers and structures. For isolated members, such as
that shown in Figure 1, a vertical lean of approxi-
mately1/200shouldbeincludedinthecalculationof
moments.
3.3 Verification of limit state EQU via Concept 2
According to thenotes of Annex A of EN 1990 two
alternativeprocedureswithtwosetsof partial factors
Table1. Resultsof theverificationsof example1 balancedstructureonpiledfoundation.
Concept 1, 2 2 1
(1)
Sourceof factor values EN1990TableA1.2(B) EN1990TableA1.2(A) EN1990TableA1.2(B)
Note1=UKNA
(2)
for
buildings
Valuesof factors
G
=1.35
Gj,sup
=1,10
Gj,sup
=1,35
Gj,inf
=0.9
Gj,inf
=1,00
Designvalue(s) of thebending 0 0.2W
r
a 0.35W
r
a
moment M for structural designof the
bottomof thecolumn
Designvaluesof theforcesF1andF2
G
W
r
=1.35W
r
W
r
(1+0.2a/b) W
r
(1.175+0.35a/b)
(compression>0) for geotechnical
designof thepile
Designvaluesof theforceF1andF2 Notension W
r
(10.2a/b) W
r
(1.1750.35a/b)
(tension-0) for geotechnical
designof thepile
Designvaluesof theforceF1andF2
G
W
r
=1.35W
r
W
r
(1+0.2a/b) W
r
(1.175+0.35a/b)
(compression>0) for structural
designof thepile
Designvaluesof theforceF1andF2 Notension W
r
(10.2a/b) W
r
(1.1750.35a/b)
(tension-0) for structural designof
thepile
(1) Without theuseof thesinglesourceconcept; thisisasolutiononemight think of whenonewantstodesignproperly all
thestructural connections, aswell asthefoundations, inorder totakeintoaccount thegeometrical andloaduncertainitesof
thebalancedstructure
(2) UK NA=UnitedKingdomNational Annex
tobeappliedtotheself weightof thehorizontal beam
maybeusedtodesignthestructure:
Note1:
G,dst
=1.1and
G,stb
=0.9
Note2:
G,dst
=1.35 and
G,stb
=1.15 (not shown in
Table1 theresultsareasfor Note1).
3.4 Conclusions
Fromtheresultsof theverificationsinTable1, it can
beseenthatthedesignvaluesof thedifferentverifica-
tionsdependtoalargeextentontheratioa/b(widtha
betweentheforcesrelativetothewidthbbetweenthe
piles). Comparing thedifferent EQU-verifications it
canbeseenthatinthisexampletheEQU-verifications
do not represent aseparateultimatelimit statebut a
differentsetof partial factorsonactionstoaccountfor
aspecial designsituation herethepossibilityof the
varianceinthespatial distributionof theself weight
of thehorizon-tal concretebeam. Ineffect, thedesign
valuesof theactionstakenfromEQUareappliedwhen
verifyinglimit stateSTR for theconcreteof thecol-
umnandpilesandwhenverifyinglimitstateGEOfor
thepiles groundbearingresistance.
4 EXAMPLE 2: TOWER SUBJ ECT TOA
VARIABLEACTION
4.1 General
ThesecondexampleshowninFigure2isatower sub-
jectedtoavariableaction. Herenotonlythestructural
designandEurocode7sthreeDesignApproachesfor
theverificationof thegroundbearingcapacityneedto
114
beconsideredbut alsothequestionastowhether and
howoverturning of thetower should beinvestigated
as aseparateEQU limit state. For this examplealso
twodifferent conceptsmaybeconsidered. Thedetails
andtheresultsof theverificationsareinpresentedin
Table2.
Thedesignmust demonstratethat:
the base of the column has sufficient strength
to resist combined axial compression, shear, and
bending(limit stateSTR)
the ground beneath the column has sufficient
strengthtoresist theappliedbearingpressuresand
slidingforces(limit stateGEO)
thefoundationis wideenoughto prevent toppling
(limit stateEQU).
4.2 Verification of limit states STR and GEO
Thebottomof thecolumnmust bedesignedfor two
separateconditions,oneinwhichtheself-weightof the
column is treated as an unfavourableaction and one
inwhichit istreatedasfavourable. Themost onerous
combination of design axial forceN
Ed
, design shear
Figure2. Tower subject toavariableaction.
Table2. Partial factorsusedinverificationof EQU for Example2.
Concept 1 2
Actionfactors
G,dst
1.1 1.1
G,stb
0.9 0.9
Q
1.5 1.5
Material factors
,
c
Not used 1.1
Resistancefactors
R,v
,
R,h
Not used 1.0
Source EN1990TableA1.2(A) EN1990TableA1.2(A)
UK NA toEN1997-1TableA.NA.2
R,v
: partial factor for bearingresistance
R,h
: partial factor for slidingresistance
forceV
Ed
, anddesignbendingmoment M
Ed
is given
byeither:
where the partial factors
G
=1.35,
G,fav
=1.0 and
Q
=1.5arespecifiedinTableA1.2(B) of EN1990.
The ground beneath the column must resist the
resultant (inclined) design force coming from the
structure, givenby
F
d
=
[(
G
W
k
)
2
+(
Q
Q
k
)
2
]. This force acts at
a design angle of inclination to the vertical of
d
= tan
1
Q
Q
k
G
W
k
and, for the calculation of effec-

tive area for determining bearing resistance, the
design eccentricity of this action is given by:
e
d
=h tan
d
=

Q
Q
k
h
G
W
k
. (Note: inDesignApproachDA
2
the calculation is performed with characteris-

tic (representative) values. Thecharacteristic ground
resistanceR
k
isdeterminedonthebasisof thecharac-
teristicvaluesof angleof inclination
k
tothevertical
and theeccentricity e
k
. In theend of thecalculation
thepartial factors areapplied and theultimatelimit
stateisverifiedbycheckingtheexpression:
where E
v.
is the vertical component of the force
comingfromthestructure.
Thevaluesof thepartial actionfactorstobeusedin
theseequationsdependonthemagnitudeof theloads
eccentricity. For large eccentricities,
G
=1.0 and
Q
=1.5aremoreonerous; whilefor small eccentrici-
ties,
G
=1.35and
Q
=1.5aremoreonerous. (Note:
in DesignApproach DA 1, Combination 2,
G
=1.0
and
Q
=1.3areused, regardlessof eccentricity.)
The design resistance is determined according to
the Design Approach adopted. In Design Approach
DA 1, Combination 1:
=
c
=
Rv
=
Rh
=1.0; in
DA 1, Combination 2, and DA 3:
=
c
=1.25
and
Rv
=
Rh
=1.0; while in DA 2 and DA 2
,
115
=
c
=1.0,
Rv
=1.4, and
Rh
=1.1. See Frank
et al., (2004) or Bond and Harris (2008) for further
detailsof theDesignApproaches.
(assuming rigid ground)
In Concept 1, only if the ground is considered as
infinitelystrong, anEQUcheckwill becarriedout; for
thisthe
G,dst
and
G,stb
factorsareused.Apartfromthe
EQUcheckaseparateSTR/GEOcheckiscarriedout,
using
G
and
Q
factorsfor theactionsandfactorson
either thefinitegroundstrengthor groundresistance,
inaccordancewiththeselectedDesignApproach.
Informer Germanstandards, whichemployedthe
global safety concept, sufficient safety against foun-
dationoverturningwasdemonstratedby ensuringthe
eccentricityeof theresultant loaddidnot exceedone
thirdof thefoundationswidth.Thisensuredthatthere
wasalwayscompressionover at least half of thetotal
base area (assuming a triangular distribution of this
basepressure). If thegroundisassumedtoberigidand
thestabilisingmoment iscausedbyaforcecentral to
thefoundation, considerationof overturningaboutthe
edgeof thefoundationbaseleadstoaglobal factor of
safety of 1.5(equal totheratioof thestabilisingand
destabilisingmomentsabout thefoundationedge).
With the implementation of Eurocode 7 in Ger-
many, overturning of a shallow foundation is now
coveredbyapplyinglimit stateEQU withpartial fac-
tors(summarizedinTable2)takenfromEN1990.This
verificationof EQUisaccompaniedbyverificationof
limit stateGEOfor thebearingcapacity of ashallow
foundation, asdiscussedintheprevioussection.
(accounting for ground strength)
InConcept 2, thegroundis treatedas beingof finite
strengthfor EQU, againwithaseparatecheck for the
loadfactorsof STR/GEO.
Concept 2requires EQU to bechecked simply as
afurther set of factorsappliedinthecompletestruc-
tural andgeotechnical design. Here, twosetsof partial
factors from EN 1990 have to be investigated, as
summarizedinTable2.
The weight W of the tower may prove to be rel-
evant for design as a favourable action, and in this
casethefactors of EQU aremoreseverethan those
of GEO, leadingtolarger inclinationandeccentricity
of theloadonthefoundation, althoughitsmagnitude
is slightly smaller. Theseactions arethen applied in
averificationof thebearingresistanceof thespread
foundation applying a partial factor on the angle of
shearingresistanceof
. If
istakenequal tothat
forGEO(e.g.
=1.25), thenEQUbecomesthegov-
erningcase. For thisreason, theNational Annexof the
UK hasusedthepossibility, allowedbyEN1997-1, to
specifyadifferent valuefor
, adopting1.1.
If, however, national annexes allow the use of
the factors
G,sup
=1.35 and
G,inf
=1.15 (or 1.0
Figure3. Overturningduetoearthpressurefromdrysoil.
for both factors), then the critical case will be
G,sup
=
G,inf
=1.0, andtheloadingfor EQU will be
thesameasfor GEOfor thisproblem.
4.5 Conclusions
Intheverificationof limit stateEQU assumingrigid
ground (Concept 1), Equation 2 is used in its basic
form. In somecases this means that thesamestruc-
tureismodelledasimposingmoresevereloading(i.e.
greater inclinationandeccentricity) onthegroundif
it isassumedtoberigid thanif it isassumedtobe
soil.
Concept 2 requires that the structure is demon-
strated to besafein all respects for EQU, STR, and
GEO loading, so avoiding difficult decisions about
whether thegroundshouldbeconsideredrigid.
5 EXAMPLE 3: OVERTURNINGOF A
RETAININGSTRUCTURE DUETOEARTH
PRESSURE
5.1 General
The third example shown in Figure 3 is a retain-
ing structure built on infinitely strong rock. Here
the requirements for rotational stability and sliding
in terms of the earth pressure F and self weight
of the retaining structure W need to be considered.
As the rock is infinitely strong, overturning of the
retainingwall about itsedgecanbeconsideredasan
EQU limit state. Somecolleagues consider that both
overturning and sliding should also be checked for
the requirements of STR/GEO, while others would
check overturning for EQU only, disregarding over-
turninginstabilityintheSTR/GEOcasewhensliding
ischecked.
Inthecalculationsit isassumedthat
earthpressures fromthesandareactivepressures,
withtheresultant characteristic valueof F
k
acting
at depth2/3h,
thewall frictionof theearthpressureisneglected,
theangleof frictionfor interfacebetweenwall and
rockis, and
thewall weight W acts at thecentreof therectan-
gular wall, itscharacteristicvalueisW
k
.
116
Table3. Resultsof theverificationsof example3 overturningof aretainingstructure.
Concept 2 1Overturning 1Sliding
Sourceof factor EN1990 EN1990 EN1990 EN1990 EN1990
values TableA1.2(B) TableA1.2(C) TableA1.2(A) TableA1.2(A) TableA.1.2(B)
andEN1997-1 andEN1997-1 andUK NA EN1997-1 EN1997-1
TableA4 TableA4 TableA2 TableA.2 TableA.5
Valuesof factors
Gj,sup
=1,35
G
=1.00
Gj,sup
not used
G,sup
=1, 10
G,sup
=1,35
Gj,inf
=1.0
=1.25
Gj,inf
=0.9
G,inf
=0,90
G,inf
=1,00
=1.00
=1.1
R;h
=1.1
Designvalueof 0.333 0.41 0.37 0.333 0.333
K
a
(K
a,d
)
Ratio: K
a,d
/K
a.k
1 1.23 1.11 1 1
(if relevant)
Requirement 1.35F
k
(h/3)/ 1.23F
k
(h/3)/ 1.11F
k
(h/3)/ 1,1F
k
(h/3)/
(inequality) for 1.0W
k
1.0W
k
0.9W
k
=1.23 0.9W
k
=1.22
rotational b/20.1m b/20.1m F
k
(h/3)/W
k
F
k
(h/3)/W
k
-
stability interms b/20.1m (b/20.1m(?)) [1a]
of F
k
andW
k
Requirement 1.35F
k
/ 1.23F
k
/1.0W
k
1.11F
k
/0.9W
k
1.35F
k
-
(inequality) for 1.0W
k
tan
k
tan
k
/1.25 tan
k
/1.1So 1,0W
k
tan
k
/1.1
sliding in So1.54F
k
/W
k
1.37F
k
/W
k
tan
k
So1.49F
k
-
termsof F
k
tan
k
1,0W
k
tan
k
andW
k
: structure-groundinterfacefrictionangle
Thesolutionsdiffer inthewaythedesignvalueof the
actionof theearthpressureisdetermined. Itcaneither
beregardedasanactiontobefactored(DA2andDA1
Combination 1) or as derived fromfactored ground
strength(DA1Combination2andDA3). Thedetails
and the results of the verifications are presented in
Table3.
5.2 Overturning
ThereisagreementforConcepts1and2that, forover-
turning,itmustbeshownthatthedesignvalueof desta-
bilisingmoment of theearthpressurewithrespect to
theedgeof thefoundationisnotgreaterthanthedesign
valueof thestabilisingmomentof theself weightof the
structure. Inparagraph6.5.4(2) of EN1997-1itisrec-
ommendedthattolerancesupto0.10mbeconsidered
whereloadswithlargeeccentricitiesoccur.
Opinionsdifferabouttheneedtocheckoverturning
for STR/GEO. If therequirements of STR/GEO are
appliedforoverturning, asinConcept2, theyaremore
severethanthoseof EQU, so theEQU requirements
havenoeffect tothisproblem.
5.3 Verification of sliding for limit state EQU
In principle, Concept 2 also requires a verification
of sliding for limit state EQU, whereas Concept 1
does not. However, if therequirements of STR/GEO
areapplied for sliding (as all agreethey should be),
they aremoreseverethanthoseof EQU, sotheEQU
requirementshavenoeffect tothisissue.
Figure4. Beamstructure staticequilibrium.
6 EXAMPLE 4: BEAM STRUCTURE
ThefourthexampleshowninFigure4isacontinuous
horizontal beamrestingontwosupports. Whenba,
theleftsupporthastobedesignedasatension-support
to provide sufficient safety of the horizontal beam
against lossof stability. Potentially thisdesignhasto
becheckedfor bothEQU andSTR/GEO. Thedetails
andtheresultsof theverificationsareinpresentedin
Table4.
Bothconceptsagreethattoachieveequilibriumfor
limit stateEQU, thedesignvaluesM
dst,d
andM
stb,d
of
destabilisingandstabilisingmomentsaboutthecentral
support arecompared:
In Concept 1-A using EQU Equation 2.4, an addi-
tional term is introduced as an action fromtheleft
support:
where G
a,k
and G
b,k
are the characteristic values of
thetwo components of theself weight andA
k
is the
117
Table4. Resultsof theverificationsof example4- beamstructure.
Concept 1-A 1-R 2
Sourceof factor values EN1990: TableA1.2(A), EN1990: TableA1.2(A), EN1990: TableA1.2(a)
Note1andTableA1.2(B) Note2 Note1
Partial factors
G,stb
=0.90
G,stb
=1.15
G,stb
=0.90
G,dst
=1.10
G
=1.35
G,dst
=1.10
G,dst
=1.35
EQU-ULSexpression M
dst,d
M
stb,d
M
dst,d
M
stb,d
M
dst,d
M
stb,d
G
b,k
b
G,dst
G
b,k
a
G,stb
+ G
b,k
b
G,dst
G
b,k
b
G,dst
A
k
2a
G,stb
G
b,k
a
G,stb
+A
d
2a G
b,k
a
G,stb
+A
d
2a
A
d
=(G
b,k
b
G,dst
G
b,k
a
G,stb
)/2a
Characteristicvalueof A
k
(G
k
G,dst
G
k
G,stb
)/ A
k
not required
thetensionforceA
k
tobe 2
G,stb
takenbythesupport or anchor
if a=bandG
a
=G
b
=G
Characteristicvalueof the A
k
0.2G
k
/1.80 A
k
=0.1G
k
1.35 A
k
not required
tensionforceA
k
if a=band A
k
0.11G
k
A
k
=0.1.35G
k
G
a
=G
b
=G
DesignvalueA
d
of theaction A
d
=A
k

G
=0.11G
k
G
A
d
(G
k
G,dst
G
k
G,stb
)/2 A
d
=(G
b,k
G,dst
onananchor if G
a
=G
b
=G A
d
=0.148G A
d
=0.1G
k
G
b,k
G,stb
)/2=0.1G
k
DesignresistanceR
d
of R
d
A
d
=0.148G R
d
A
d
=0.1G
k
R
d
=A
d
=0.1G
k
theanchor if G
a
=G
b
=G
characteristic valueof thenecessary tensionforceof
theleftsupporttosatisfylimitstateEQU. Thetension
forceisthecharacteristicvalueof theaction(tension)
to design the support or an anchorage. If a=b and
G
a
=G
b
=Gthetensionforceis:
In the next step the anchor can be designed fulfill-
ing the expression for limit state STR for anchors
(Equation8.1inEN1997-1):
whereR
d
is thedesignvalueof theresistanceof the
anchor whichis:
InConcept 2, EQU equation2.4isusedtodetermine
the necessary design value of the anchor resistance
directlyintheform:
7 CONCLUSION
Situations in which actions balance each other are
important andmust beproperlyconsideredindesign.
EQU is defined in the Eurocodes to accommo-
date this, but it is clear that its definition and use
requires further development, both for geotechnical
andstructural design.Thispaperhassummarisedsome
alternativeviewsbymeansof examplesinvolvingboth
actions and resistances. The authors hope that these
will contributetoclarificationof EQU.
REFERENCES
Bond, A.J., and Harris, A.J. (2008), Decoding Eurocode7,
Taylor andFrancis, London, 598pp.
CEN(2002) EurocodeBasisof design. (EN1990) European
Committeefor Standardization: Brussels.
CEN (2004) Eurocode 7 Geotechnical design Part 1:
General rules. (EN 1997-1) European Committee for
Standardization: Brussels.
Frank, R. C. Bauduin, R. Driscoll, M. Kavvadas, N. Krebs
Ovesen, T. Orr and B. Schuppener (2004): Designers
Guide to EN 1997-1, Eurocode 7: Geotechnical design
Part 1: General rules, London: ThomasTelford.
118
Geotechnical criteriafor serviceabilitylimit stateof horizontally
loadeddeepfoundations
MasahiroShirato, TetsuyaKohno, & Shoichi Nakatani
Bridges and Structures Research Group, Center for Advanced Engineering Structural Assessment and Research
(CAESAR), Public Works Research Institute (PWRI), Tsukuba, Japan
ABSTRACT: Thispaper presentsafirst approximationof theelastic limit displacement of soil resistanceto
horizontallyloadedpiles. A total of 37fieldtestdatasetswereconsistentlyretrievedfromavastdatabase. Both
mathematical andgraphical approaches wereadoptedto interpret theelastic limit of soil resistancefromthe
measuredload-displacement curve. Thedatasetsindicatethat themeanvalueof theelasticlimit displacement
is 56%of thepilediameter, and its coefficient of variation is approximately 4060%. A design horizontal
thresholddisplacement canalsobeproposedas24%of thepilediameter, consideringthevariationinelastic
limit displacement.
1 INTRODUCTION
The concept of serviceability limit state is widely
acceptedinfoundationdesignandusually takes into
consideration the influence of foundation displace-
mentonsupportingstructures.Tolerabledisplacement
is affectedby many aspects includingstructuretype,
materials, and design philosophy. While a tolerable
displacementhasbeenproposedbymanyresearchers,
especially for buildings, earlier study is limited for
highway bridge foundations. For example, Moulton
(1985) proposed a tolerable displacement based on
practical experiencein theU.S. Recently, reliability-
basedstudyontheallowabledisplacement of founda-
tionsappearstohaveprogressed(PhoonandKulhawy
2008). Asfor highwaybridges, ZhangandNg(2005)
have given reliability-based criteria on the tolera-
bledisplacement (i.e., settlement) of highway bridge
foundationsbasedonexperiencesinHongKongand
they have also commented that abutment displace-
ment duetothemovement of approachembankments
is a primary concern rather than the settlement of
foundations.
Since1990, theJ apaneseSpecifications for High-
way Bridges (J RA 2002) has requiredverificationof
the tolerable displacement of foundations in terms
of the following two aspects: 1) tolerable displace-
ment based on the required performance or design
of the superstructure; and 2) tolerable displacement
basedontherequiredperformanceof thefoundation.
The first tolerable displacement is equivalent to the
serviceability limit state mentioned above. The sec-
ond tolerable displacement, traditionally considered
critical infoundationdesignduetoJ apansearthquake-
prone nature, is the elastic limit displacement of
foundations. While structures frequently undergo
small-scale earthquakes, residual displacement of
foundations fromeach frequent-scale earthquake is
not allowed.
Thisstudydealswiththesecondtolerabledisplace-
ment.Itisworthnotingthatthefirsttolerabledisplace-
ment is not critical in design except for the lateral
movement of abutment foundations on soft ground,
and preventivemeasures areusually applied in such
cases. In addition, dueto their long length, highway
bridgesuperstructuresarenotsensitivetotheunequal
vertical displacement of neighboringsubstructures.
The second tolerable displacement must confirm
that the foundation provides the superstructure with
steady-statereversiblerestoringresistanceagainstfre-
quentloadsduringitsservicelife.Therequiredbridge
performanceintheJ apaneseSpecificationsfor High-
wayBridgesunder normal, wind, andfrequent earth-
quake designconditionsissuchthatthebridgemustbe
fullyfunctional without changingthewayit responds
to loads. Accordingly, the corresponding limit state
for thefoundationistheelasticlimit. Sectional stress
in shafts must not go beyond the allowable stress
froma structural viewpoint, and the load effect at
the foundation top must not exceed the allowable
elastic bearingcapacity. Horizontal soil resistanceis,
in reality, nonlinear even at a very small displace-
ment level. However, if thefoundation displacement
remainswithinacertainlevel andnonotableresidual
displacementappears, itisexpectedthatthehorizontal
soil resistancetothefoundationwill maintainasteady
state, avoidingpossibleadverseeffects.
In addition, theelastic limit stategives an advan-
tage in structural design, which can be modeled so
that the soil resistance is reversible (elastic) and is
not a function of load history within the tolerable
foundationdisplacement. If foundationresistanceand
119
displacement wereassumed to beafunction of load
history, thedesignerswouldtheoreticallyhavetotrace
theentirehistory of foundationdisplacement against
the entire load history for the expected service life,
whichisunacceptableinpractice.
Traditionally, the allowable horizontal bearing
capacity isdefinedwithadisplacement at thedesign
ground level. TheJ apaneseSpecifications for High-
way Bridges describe allowable horizontal displace-
ment as the larger value of 1%of the pile diameter
or 15mmunder normal, wind, and frequent earth-
quake designconditions.Theengineeringbackground
of these empirical values is given by Okahara et al.
(1991a, 1991b) whocollectedin-situtestdataonpiles
subjectedtohorizontal loads, andexaminedtheelastic
limitdisplacementontheobservedload-displacement
curves. Also, 1% of the pile diameter is reasonable
fromapractical viewpoint becausethetypical design
calculation model is considered to bevalid only for
smaller displacement.
However, wedo not knowwhether soil resistance
or shaft resistance is attributed to the elastic limit
points on the load-displacement curves derived in
earlier studies. Structural verification is separately
implementedindesignandthethresholddisplacement
shouldbeafunctionof soil resistance, not shaft resis-
tance. Inaddition, larger-diameter shaftsandstronger
materials arenowavailable. A newnonlinear design
calculationmodel wasrecentlyintroducedalongwith
the advent of a new seismic design of foundations
under rare earthquake conditions. Technically, the
foundationresponseat larger displacement levelscan
becalculatedtotakeadvantageof shaftstrength, while
traditional values havebeenseenas abarrier against
newer technologies.
Thisstudyre-examinestheelasticlimitof horizon-
tal soil resistanceusingaPWRI databaseforfieldtests
on piles subjected to horizontal loads. First, wepro-
posean interpretation for extracting theelastic limit
of soil resistance fromthe load-displacement curve
measured in testing. Then, field test data sets are
consistently analyzedtoillustratethestatistical char-
acteristics of the elastic limit displacement in terms
of soil resistance, sothat wecanfurnishmodel statis-
tics for reliability calibrations. Finally, weproposea
revised design tolerable horizontal displacement of
pilesunder thereliabilitydesignconcept.
2 BEHAVIOR OF HORIZONTALLY
LOADEDPILES
Typical nonlinearbehaviorof horizontallyloadedpiles
is showninFigure1. Nonlinearity intotal pileresis-
tanceisattributedtothenonlinearityinsoil andshaft
resistance as indicated by A and X in Figure 1.
Accordingly, it is necessary that the shafts do not
becomeplastic, so that theelastic limit in soil resis-
tance marked with A in Figure 1 can be identified
fromatotal load-displacementcurve. Finally, theload
testdatashouldagreewiththefollowingrequirements
Figure1. Typical nonlinear behavior of horizontallyloaded
piles.
for identifying the elastic limit point of soil
resistance:
A. Theload-displacement curvehas clear nonlinear-
ity.
B. It canbeassumedthat theshaft will not become
plasticuptoalargedisplacement level.
3 INTERPRETATIONOF ELASTIC
LIMIT POINT
There is no specific definition for identifying the
elastic limit point of soil resistance from a load-
displacementcurve,andbothmathematical andgraph-
ical approaches are available for determining the
point at which the transition occurs fromthe initial
linear portionto thesecondlinear portionof aload-
displacementcurve(e.g., HiranyandKulhawy, 1989).
Basically, amathematical model is employed in this
study, because it is considered more robust than a
graphical approach. This study employs exponential
(orWeibull) curvefittingreferringtoUtoetal. (1985)
andOkaharaet al. (1991a, 1991b).
where R
uw
=ultimate total pile resistance estimated
viaexponential fitting, d =displacement, d
0
=elastic
limit displacement estimated in exponential fit-
ting, B=pile diameter, and m=a constant that
defines the shape of the curve. Typical curves
are shown in Figure 2. The load level corre-
sponding to elastic limit displacement is always
R
0
=R
uw
(1e
1
)=0.63R
uw
for any m. For the
sake of simplicity, this study assumes m=1 except
in the case of drilled shafts where this assumption
results in errors based on a trial and error anal-
ysis. Also regarding drilled shafts, the data points
that exist under a displacement level of 0.01B are
omitted in the exponential (Weibull) fitting analysis
becauseapseudoelasticlimitcouldappearontheload-
displacementcurvesduetothedevelopmentof cracks
inthecover concreteprior totheyieldof longitudinal
reinforcement.
120
Figure2. Weibull curve.
4 GRAPHICAL METHODTODETERMINE
ELASTIC LIMIT
Sometimes, the use of a graphical method is con-
sidered more useful for interpretation compared to
amathematical methodfromtheviewpoint of physi-
cal meaning, andaccordingly, analternativegraphical
methodproposedbyOkaharaet al. (1991a, 1991b) is
alsoappliedtoconfirmthemathematical method. Fig-
ure3(a) shows theresults of acyclic loadtest. From
theresults,setsof residual displacementatanunloaded
point of R=0andtheprecedingpeak point displace-
mentareextractedas[1], [2], [3], [4], and[5] asshown
inFigure3(b) andplottedinFig. 3(c). Itcanbeseenin
Figure3(c) thatasuddenchangeinincrementof resid-
ual displacementoccursatapeakdisplacementlevel of
d
1
. Therefore, thisdisplacementof d
1
canberegarded
as theelastic limit displacement, because, if thepile
didnot go beyondthis thresholddisplacement, there
wouldbenoresidual displacementremaining, andthe
initial soil resistancepropertycouldstill bemobilized.
5 SELECTIONOF SINGLE PILE LOAD
TEST DATA
For singlepiles, only in-situtestsaredealt withhere.
The data adopted here agrees with the following
conditions:
1) Thepileisasslenderaspossible,isassumedtohave
asemi-infinitelength, andisafree-head, straight-
sidepile.
2) Theheightof thepointof loadingfromtheground
level is lower than the pile diameter, B.
3) Themaximumdisplacementlevel islargerthan5%
of thepilediameter, B.
4) The observed maximumload in the load test is
larger than 1.2 times the elastic limit load, R
0
,
obtainedthroughexponential fitting.
5) Theshaftwill notbecomeplasticuptoaloadlevel
of 1.2timesthecalculatedelasticlimit load, R
0
.
Items3),4),and5)areassociatedwithrequirements
A andB mentionedearlier. A very simplemethodis
employedtoverifyItem5) above. Usingadesigncal-
culationmodel basedonthetheorybeamonWinkler
foundation, the maximumbending moment in the
shaft at aloadlevel of 1.2R
0
isestimated.
Eventually, 37 data sets were available. The
observed and fitted load-displacement curves are
showninFigure4. Thedisplacement at groundlevel
wasadoptedif available, andif not, thedisplacement
at theclosest point to theground level was taken. If
neitheronewasavailable, thedisplacementattheload-
ingpoint wasadoptedinthisstudy. Whenit comesto
cyclicloadtests, onlythebackbonecurvesareshown
inFigure4. A summary of thedatasets usedhereis
showninTable1andFigure5interms of construc-
tion type and pile geometry. Approximately 81%of
the employed data sets are associated with the cat-
egory of non-composite piles (or typical piles) and
therest areassociated with thecategory of compos-
itepiles. Drivenpiles, large-diameter screwpiles, and
inner-auguredcompressivelyinstalledpilesareclassi-
fiedasnon-compositepiles. Pre-boredpilesandsteel
pipe/soil cementpilesaretobeclassifiedascomposite.
Inthedataselectionprocess, all drilledshafttestswere
excludedduetononconformancetoItem5) above.
Inner-augured compressively installed piles are
constructedasfollows: (1) A pileisset, andanaugur
is used to drill through thepile, making aholewith
adiameter somewhat smaller than thepilediameter.
(2) As the auger drills lower, the pile is simultane-
ouslypushedintotheground. (3) Steps(1) and(2) are
repeateduntil thepilereachesapredefineddepth.
Pre-boredpilesareconstructedasfollows: (1) Soil
isauguredwithadiameter of B andapredefinedpile
length. (2) Runny cement is mixedwithresidual soil
intheauguredhole. (3) A factory-madehigh-strength
prestressed concrete pile with a diameter somewhat
smaller than B is placed into the augured hole. The
soil cementmixturemadeinStep(2) risesup, andthe
spacebetweenthesoil andpileshouldbefilledsothat
acompositepileismade.
Steel pipe/soil cement pilesareconstructedasfol-
lows: (1) A soil cement column with a diameter of
B is constructed. (2) A pile with a diameter some-
what smaller than B is pushed into the soil cement
column, sothat it becomesacompositepilewiththe
soil cement column part. Steps (1) and (2) can be
conductedsimultaneouslyor separately.
An augured diameter B is somewhat larger than
theinsertedfactory-madepilediameter inpre-bored
pilesandsteel pipe/soil cementpiles. Thisstudydeals
withthediameter of B astherepresentativediameter
of these piles, because the soil cement part should
121
Figure3. Graphical analysis for estimatingthresholddisplacement interms of rapidevolutionof residual displacement in
cyclicloadtest.
Figure4. Observed, normalizedobserved, andfittedload-displacement curves.
Table1. Numberof loadtestresultsintermsof construction
methodsfor singlepiles.
#of Meanof COV of
Constructionmethods data d
0
/B
d
0
/B
Non- Drivenpile 21 0.0599 0.39

composite
(Typical) Large-diameter 7 0.0693 0.22
screwpile
Inner-augured 2 0.0515 0.23
compressively
installedpile
Composite Steel pipe/soil 6 0.0356 0.28
cement pile
Pre-boredpile 1 0.0329
* d
0
=ElasticlimitdisplacementobtainedbyWeibull fitting
analysis
respond together with the factory-made pile and it
shouldbedirectlysubjectedtothesoil resistance. The
bendingrigidity of theshaft is madeequal to that of
theinsertedfactory-madepile.
6 STATISTICAL CHARACTERISTICSOF
ELASTIC LIMIT OF HORIZONTAL SOIL
RESISTANCE
First, thecyclicloadtest dataischosenfromtheload
test datasets employed above. Although thenumber
of cyclicloadtest dataislimited, theelasticlimit dis-
placementlevel identifiedthroughgraphical analysis,
d
1
/B, is comparedwiththat estimatedthroughexpo-
nential fitting, d
0
/B. For example, oneof theresults
is shown in Figure3, in which an exponential curve
fitted for thebackbonecurveis shown in thegraph
ontheleft-handside. Theelasticlimitpointismarked
onthefittedcurve. Figure6comparestheelasticlimit
displacement level derived using both graphical and
mathematical methods. Thenumber of employedload
tests, n, eventually totals twelve. Theresults derived
usingbothmethodsagreewell witheachother, which
confirmsthatbothmethodsarecapableof accounting
for theelasticlimit of soil resistance. Thebiasfactor,
=d
1
/d
0
, hasameanvalueof 1.01andacoefficient
of variation(COV) of 0.45obtainedusingthesquare
root of (
2
/(n 1)), inwhich =standarddeviation
andn =numberof loadtests.Inaddition,themeanand
COV of d
1
/B thereof wereinterpretedviaagraphical
methodas0.048and0.39, respectively.
Then, usingall thedata, theelastic limit displace-
ment estimated through exponential fitting, d
0
, is
plotted in the left panel of Figure 7 versus the pile
diameter,B.Adisplacementlevel of 20mmsafelycov-
ersalmostall thedata. However, thereisatendencyfor
thevalueof d
0
toincreasewithanincreaseinthepile
diameter, B. Therefore, thenormalizeddisplacement
level, d/B, which more or less represents the strain
inthesoil infront of thepile, is usedto characterize
theelastic limit. Therelationshipbetweentheelastic
122
Figure5. Frequenciesinpilegeometryandtypesof construction.
Figure6. Comparisonof elasticlimit displacements, d
1
andd
0
, obtainedthroughgraphical analysisandexponential fitting
(N.C. =non-compositepilesandC. =compositepiles).
Figure 7. Relationship between estimated elastic displacement level and pile diameter (N.C. =non-composite piles and
C. =compositepiles).
limit displacement level, d
0
/B, and thepilediameter
is shownintheright panel of Figure7. Thedatadis-
tributesaroundthemeanvalueof 0.565B withaCOV
of 0.39. Thecalculatednormalizedexponential curve
with amean valueof d
0
/B, 0.565, and m=1 is also
shownintheright panel of Figure4.
Theempirical distributionof elasticlimit displace-
mentlevel, d
0
/B, isshowninFigures8and9. Itshould
be noted that there are small variations in the num-
ber of testscitedherebecausesometest resultsinthe
database do not have sufficient log data on boring.
Based on Figure 8, the value of d
0
/B of composite
piles tends to besomewhat smaller thanthat for typ-
ical non-compositepiles. Compositepiles hadasoil
cement layer around the factory-made shaft. There-
fore, thedevelopmentof cracksinthesoil cementpart
during the loading could be considered to result in
nonlinearity in the load-displacement curve even at
smaller displacement levels. For example, all driven
piles intheadopteddatasets aresteel pipepiles. As
123
Figure8. Variationinelasticlimit displacement level, d
0
/B, of horizontallyloadedpilesfor eachpiletype
Figure9. Relationshipof elasticlimit displacement level, d
0
/B, withsoil typeandSPT-N value
showninTable1, whilethemeanvalueof d
0
/B for the
drivenpilesis0.599, themeanvalueforsteel pipe/soil
cement pilesis0.356.
Therelationshipbetweenestimatedelasticlimitdis-
placement level, d
0
/B, soil type, andsoil stiffness or
strength, i.e., SPT-N value, is showninFigure9. For
eachpileloadtest, thepredominant soil typedepends
on thesubsoil layers within thecharacteristic length
of pileas defined in theSpecifications for Highway
Bridges, inwhich1/ =characteristicpilelength, and
=[(kB)/(4EI )]
1,4
, E =Youngsmodulusof theshaft,
I =second sectional moment of inertia of the shaft,
andk =subgradereactioncoefficient. Thetotal depth
of sandy subsoil layers and clayey subsoil layers is
calculated, andthepredominant soil typeisbasedon
thelarger total depthof either thesandy subsoil lay-
ersor theclayeysubsoil layers. Thesubgradereaction
factor for calculatingthecharacteristic pilelengthis
obtainedusingan empirical formulain theJ apanese
SpecificationsforHighwayBridges.TheSPT-N value
is themean valueof themeasured values within the
characteristic pile length. The difference in the esti-
matedelasticdisplacement level, d
0
/B, resultingfrom
thedifferenceinsoil typeor SPT-N valueissmall.
7 DESIGNTHRESHOLDDISPLACEMENT
ANDITSRELIABILITY
A general designequationcanbeexpressedas
Aslongastheresistancecapacityislarger thanthe
load effect, there is a margin of safety for the limit
stateunder consideration. However, asboththeresis-
tance capacity and load effect are considered to be
randomvariables, factored resistance and loads are
usually usedindesigntoachieveapredefinedsafety
margin. Intermsof thesoil resistancetohorizontally
loadedpiles, Eq. (2) canberewrittenas:
in the Load and Resistance Factor Design (LRFD)
format, in which =resistance factor, d
L
=elastic
limit displacement, =load factor, Q=load, and
d =calculated piledisplacement in design with fac-
toredloadQ.
A crudeestimation of thethreshold valueconsid-
ering uncertainty in theelastic limit displacement is
still better than that obtained using a deterministic
approach. Accordingly, this study presents a condi-
tional reliability analysis, disregarding thestatistical
issuesintheloadeffect,d,andapplyingtheloadsinthe
current J apaneseSpecificationsfor HighwayBridges
as the deterministic factored values. Eventually, Eq.
(3) canberewrittenas
whered
d
=threshold displacement. Theelastic limit
displacement level, d
L
/B, is assumed to follow a
log-normal distribution.Hence,thefirst-ordersecond-
moment (FOSM) methodgives arelationshipfor the
124
design threshold displacement level, d
d
/B=d
L
/B,
andreliabilityindex, , asfollows:
where COV
R
=coefficient of variation of the resis-
tancecapacity, d
L
/B.
The mean and COV for the resistance capacity,
d
L
/B, can be estimated based on the results of the
exponential fitting analysis and graphical analysis.
Although it appeared in Figure 8 that there may be
a difference in uncertainty due to the difference in
construction type, this study does not take this dif-
ference into account, assuming that the elastic limit
in the horizontal soil resistance is not a function
of the behavior of the soil cement part in compos-
itepiles. If thegraphical interpretationis considered
to be more reasonable, the transformation error in
elastic limit displacement between the exponential
fitting analysis and the graphical analysis should
be incorporated into the mean and COV of d
L
/B.
Thus, the mean and coefficient of variation of elas-
ticlimitdisplacement,d
L
/B,isd
L
= 0.565=0.057
and COV
R
=(0.39
2
+0.45
2
)
0.5
=0.60. Accordingly,
basedonall theresults obtainedabove, thevalues of
the mean and COV of d
L
/B are likely to fall some-
whereintherangeof 0.05to 0.06and40%to 60%,
respectively. Eventually, this study sets thevalues of
themeanandCOV of d
L
/B at 0.055and50%.
Estimation of the target reliability index,
T
, is
attempted here. As mentioned above, the allowable
horizontal displacementinthecurrentspecificationsis
basedonOkaharaetal. (1991a, 1991b). Theyadopted
anequivalenttheorytoEq. (5) andconsideredarelia-
bilitylevel of approximately1usingtheloadtest data
that they compiled. Accordingly, thetarget reliability
level should be set around that value. In addition, a
typical safety marginassociatedwithdesigningshal-
lowfoundations andpiles subjectedto vertical loads
canassist inthecalculation. Inthecurrent Specifica-
tions for Highway Bridges, safety factors j=3 and
2areappliedtotheultimatebearingcapacity toesti-
matetheallowablebearingcapacityundernormal and
frequent earthquake design conditions, respectively,
and the allowable bearing capacity is considered to
provide foundations with a sufficient safety margin
for both serviceability (i.e., elastic limit state) and
safety(i.e., ultimatebearingcapacity) (Nakatani etal.
2007). Theideahereis that equivalent safety factors
shouldhavebeenappliedtothehorizontal soil resis-
tance. A generalizedload-displacement curvecanbe
modeledas
whereR
L
=soil resistanceattheelasticlimitloadcor-
responding to the elastic displacement capacity, d
L
.
Therefore, safety factors j=3 and 2 applied to the
ultimatehorizontal soil resistancearetranslated into
safety factors j
=1.89 and 1.2 to the elastic limit

soil resistance, R
L
, respectively. Inaddition, because
the resistance factor is equal to 1/j
, safety fac-
torsj
=1.89and1.2correspondtoresistancefactors
=1/1.89(=0.53) and1/1.2(=0.89). Therefore, Eq.
(5) leadstothedesignthresholdvaluesandreliability
levels. Thedesignthresholddisplacement level, d
d
/B,
atresistancefactor =0.53isestimatedas0.022with
reliability level =1.67andthat at resistancefactor
=0.89 is estimated as 0.038 with reliability level
=0.54. Based on these results, the target reliabil-
ity indices caneventually beroundedoff to
T
=1.5
and0.5under normal andfrequent earthquake design
conditions, respectively. The corresponding values
of design threshold displacement level, d
d
/B, are
obtainedas0.024and0.039, respectively.
Asit turnsout, wecanexpect at least 0.02B under
normal conditions and0.035B under frequent earth-
quake conditions as therounded-off designtolerable
displacement for theserviceabilitylimit state.
8 HORIZONTALLY LOADEDPILE GROUPS
Whiletheresultsshownaboveareassociatedwithsin-
gle piles, pile foundations usually comprise several
piles with close spacing between neighboring piles
equal to2.53.5timesthepilediameter. Accordingly,
aroughanalysis for theelastic limit displacement of
pile groups is conducted for reference using expo-
nential fitting. Eventually, a drilled shaft group and
ninesteel-pipepilegroupsareadopted. Toensurethe
largestpossibleamountof data, laboratorytestdatais
alsousedaswell asin-situtest data. Theadopteddata
setssatisfythefollowingconditions:
1) Thepilegroupcomprisesseveral rowsof pilesboth
parallel andperpendicular totheloadingdirection,
2) The observed maximum displacement level is
larger than5%of thepilediameter,
3) The observed maximumload in the load test is
larger than1.2timesthecalculatedyieldload, R
0
,
usingexponential fitting.
Therelationshipbetweentheelasticlimit displace-
ment level, d
0
/B, estimatedusingexponential fitting,
andthepilediameter, B, is showninFigure10. The
solidanddashedlinesindicatethetolerabledisplace-
ment under normal and frequent earthquake design
conditionsasproposedabove, andtheproposedtoler-
abledisplacementscover mostdataonthesafetyside.
This result supports the application of the proposed
tolerabledisplacementstothegeneral designfor pile
groupshavingclosepile-to-pilespacing.
9 CONCLUDINGREMARKS
When thesoil resistanceto piles behaves within the
elastic limit, thepiles haveareversiblerestoringsoil
125
Figure 10. Elastic limit displacement obtained using the
exponential fitting analysis for horizontally loaded pile
groups.
resistance force, and are expected to be firmly sup-
portedovertheyearsagainstserviceloads, windloads
and frequent earthquakes. Accordingly, this study
examinedthestatistical issuesfor theelasticlimitdis-
placement. Thefollowingresultswereobtainedfrom
thisstudy.
1) Clarification of the philosophy of serviceability
limit states in highway bridge design in J apan
fromthe viewpoint of superstructure design and
foundation design. The latter is the elastic limit
of foundation restoring resistance as a structural
component of highwaybridges.
2) Verificationof asimplemathematical methodfor
identifyingtheelasticlimit of soil resistancefrom
theload-displacement curveinthehorizontal load
test of apile.
3) When the elastic limit displacement was inter-
preted consistently froma number of field load
test results, weobtainedameanvalueof 46%of
thepilediameter withacoefficient of variationof
4055%.
4) Wefoundthatthedesignthresholddisplacementat
thegroundlevel canbesettoapproximately24%
of thepilediameterfortheserviceabilitylimitstate
under normal andfrequent earthquake conditions.
These values are generally larger than the previ-
ously proposed values of 0.01B and 15mmfor a
typical rangeof pilediameters inhighway bridge
foundations.
5) Theproposedthresholddisplacementscanalsobe
consideredrelevant tothedesignof pilegroups.
Asaremainingissue,thebehaviorof thesoil cement
portionof steel pipe/soil cement piles andpre-bored
pilesrequiresinvestigationinorder toestablishmore
detailedserviceabilitylimit criteria.
REFERENCES
Hirany, A. and Kulhawy, F. H. 1989. Interpretation of load
testsondrilledshafts, Part 3: Lateral andmoment. Foun-
dation Engineering: Current Principles andPractices, 2,
ASCE, NewYork, pp. 11601172.
J apan Road Association (J RA). 2002. Specifications for
HighwayBridges. Tokyo.
Moulton, L. K. 1985. Tolerablemovement criteriafor high-
waybridges.ReportFHWA/RD-85/107,Federal Highway
Administration, Washington, DC, USA.
Nakatani, S., Shirato, M., Iochi, H., and Nomura, T. 2007.
Stabilitycheckof highwaybridgepilefoundationsundera
performance-baseddesignconcept. PWRI Technical Note
(4036), PublicWorksResearchInstitute. InJ apanese.
Okahara, M., Nakatani, S., and Matsui, K. 1991a. A study
onvertical andhorizontal bearingcharacteristicsof piles,
J SCEJ. of Struct. Engrg. 37, pp. 14531466. InJ apanese.
Okahara, M.,Takagi, S., Nakatani, S., andKimura,Y. 1991b.
A studyonthebearingcapacityof singlepilesanddesign
methodof columnshapedfoundations. PWRI Technical
Memorandum(2919), PublicWorksResearchInstitute. In
J apanese.
Phoon K. K. and Kulhawy, F. H. 2008. Serviceability limit
statereliability-baseddesign,Chapter9,Reliability-based
design in geotechnical engineering: Computations and
applications, pp. 344384, Taylor & Francis.
Uto, K., Fuyuki, M., andSakurai, M. 1985. Anexponential
mathematical model togeotechnical curves. International
SymposiumonPenetrabilityandDrivabilityof Piles, San
Francisco, USA, pp. 16.
Zhang,L.M.andNg,A.M.Y.2005.Probabilisticlimitingtol-
erabledisplacements for serviceability limit statedesign
of foundations. Gotechnique, 55(2), pp. 151161.
126
Reliability-basedcodecalibrationof pilesbasedonincomplete
proof loadtests
J ianyeChing
National Taiwan University, Taipei, Taiwan.
Horn-DaLin& Ming-TsoYen
National Taiwan University of Science and Technology, Taipei, Taiwan.
ABSTRACT: Thispaperaddressesanissueoftenencounteredincalibratingresistancefactorsforpileultimate
capacitiesbasedonloadtest database. Inpractice, manypileloadtestsarenot conductedtofailuresbut onlyto
amultiple(e.g.: 2) of thedesignload. Thisleadsadifficult situationof incompleteinformation: for thesetest
results, theultimatebearingcapacitiesof thetest pilesareunknown. Howcanthesetest resultsstill beusedto
calibrateresistancefactorsof piles?A full probabilistic framework isproposedinthisresearchtoresolvethis
issue. A local piletestdatabaseof Taipei (Taiwan) ispresentedfor demonstration. Theanalysisresultsshowthat
theinclusionof theincompletepileloadtestdatahelpsincalibratingtheresistancefactors. Moreover, itisfound
that thecalibratedresistancefactors areconsistent tothesafety factors that areadoptedinthecurrent Taiwan
designcode.
1 INTRODUCTION
Recently, reliability-based design approaches have
emerged as anewdesign paradigmfor pileultimate
capacities becausereliability is aconsistent measure
of uncertainties. For reliability-based designs, one
designs a pile so that its failure probability is in an
acceptablerange. Moreover, asinglefactor, calledthe
resistancefactor, is oftenusedto quantify resistance
uncertaintiestofacilitatereliability-baseddesigns.
Inthecasethat ultimatecapacitiesof thetest piles
in a region areknown, theresistancefactors of that
region can be calibrated conveniently based on the
loadtestdatabase, e.g.: Barker etal. (1991), McVayet
al. (2000), andPaikowsky et al. (2004). However, in
practicemanyloadtestsarenotconductedtoultimate
capacity failures but only to two times of thedesign
loadsdependingonthelocal codeprovisions. Taking
theboredpilesof theTaipei CityinTaiwanasanexam-
ple, thereareabout57loadtestsinthelocal database,
butamongthem,only8of themwereconductedtofail-
ures. Figure1illustratestheload-deformationcurves
of twopilesinthedatabase, onewasloadedtofailure
whiletheotherwasnot. ByDavissonscriteria(Davis-
son 1972), theultimatecapacity of thefirst pilecan
beidentified, whilethatof thesecondpileisnotiden-
tifiable because of the incomplete load-deformation
curve.
Although the load-deformation curve is incom-
plete, itdoesprovideacertainamountof information.
Accordingtotheload-deformationcurveof thesecond
pilein Figure1, it is clear that theDavisson capac-
ity of thepileis greater thanthemaximumtest load
Figure1. Complete(left:(a)) versus incomplete(right:(b))
load-deformationcurves.
(2436tons).Also, theactual Davissoncapacityshould
beless than theload of point A in thefigure(3300
tons), theDavisson capacity if theload-deformation
curveisindefinitelylinearlyextended(i.e.: thedashed
line). Therefore, it isevident that theactual Davisson
capacity of the pile should be among [2436, 3300]
tons.
In the context of interpreting load test results of
proof piles, Zhang (2004) addressed the issue of
incomplete information and proposed a systematic
method of updating reliabilities of within-site pile
designs. By assuming the coefficient of variation
(c.o.v.) of within-sitepilecapacitiestobearound20%,
heproducedtablesandchartsthat cananswer impor-
tantquestions. ThefocusinZhang(2004) istoupdate
thereliabilities of within-site piles. For calibrating
resistance factors for a region, a different approach
shouldbetaken.
127
The focus of this paper is to propose a rigorous
framework of calibrating resistance factors of piles
basedonincompleteloadtestdatabase. Imaginemost
informationavailablefromaregional databaseis the
bounds on the Davisson capacities of the test piles.
Howcantheregional resistancefactorsbecalibrated
based on the incomplete information? A test pile
databaseof Taipei will bestudied as demonstration,
andafull probabilisticframeworkwill beproposedto
resolvetheissueof incompleteinformation. Inprinci-
ple,theproposedframeworkisnotlimitedtotheTaipei
database: the same framework can be implemented
to other regional test pile database with incomplete
informationtocalibrateregional resistancefactors.
2 DATABASE OF TAPEI
This section describes a database of the load test
resultsof 57boredpilesintheTaipei region.Thebasic
information of thetest piles is listed inTable1. The
diametersof theboredpilesinthedatabaserangefrom
0.6mto2m, andthelengthsrangefrom13mto81m.
Most piles have lengths from30mto 60m. In the
table, the column of ultimate capacity reports the
Davisson capacity if the load test is complete; oth-
erwise, the [L,U] of the Davisson capacity will be
reported, wherethelower boundL andupper boundU
areobtainedfromtheload-deformationcurvebyusing
theinterpretingapproachshowninFigure1(b). Only
theDavissoncriterionis usedinthis paper for illus-
trativepurposes. For any particular purpose, suitable
failurecriteriashouldbeselected. It isclear that there
areonly8completeloadtests.
Theend-bearinglayersfor all pilesareeither sand-
stones or gravels, depending on their geographical
locations. For eachtest pile, thesoil profilefromthe
nearest boreholefromthetest pileis taken. Thesoil
profileincludes thefollowinginformation: (a) thick-
nessof eachlayer; (b) classificationof eachlayer; (c)
SPT-N blow counts of each layer, manifesting itself
either as arangeor as anaveragevalue; (d) location
of thewater table; (e) unit weight of eachlayer.
3 QUATIFICATIONOF UNCERTAINTIES
Twotypesof uncertaintiesareconsideredinthispaper:
(a) uncertainties in the basic soil parameters and
indices(includingwater tablelocations) and(b) mod-
elinguncertainties. Theformer is duetotheinherent
variabilities of soils and rocks, measurement errors
of thetest methods (e.g.: thereported SPT-N values
may deviatefromtheir actual valuesduetomeasure-
ment errors in SPT) and transformation errors (e.g.:
the errors induced in the process of estimating fric-
tion angles fromSPT-N values). Thelatter is dueto
theinaccuracy of thepredictive(design) models for
theultimatepilecapacities. Whatfollowssummarizes
our waysof determiningtheprobabilitydensityfunc-
tions (PDF) of theseuncertainties basedonthefield
dataandprior information.
3.1 Uncertainties in soil parameters and indices
Thehereinconsidereduncertaintiesinclude: (a) SPT-
N valueof each layer; (b) unit weight of each layer;
(c) undrainedshear strengthof eachclayey layer; (d)
frictionangleof eachsandyorgravellylayer; (e) water
tablelocationateachpile; (d) endbearingcapacityfor
thesandstonelayer.Theseparameters/indicesarenec-
essaryforpredictingtheultimatebearingcapacitiesof
thetest piles.
Theaverageortherangeof SPT-Nvalue(N)isavail-
able for each soil layer. Note that the SPT-N values
reportedaremeasuredvalues of SPT-N rather than
theiractual values: evenif theSPT-Nvalueof alayeris
reported, its actual SPT-N valueis still uncertaindue
to the measurement errors of SPT. It is desirable to
model theseuncertaintiesasprobabilitydistributions,
describedasfollows.
Inthecasewheretherange[N
low
, N
up
] of SPT-N
valueisreportedforalayer, theuncertainSPT-Nvalue
ismodeledtobeuniformlydistributedover the[N
low
,
N
up
] interval. Inthecasethattheaveragevalueof SPT-
N is reported, theuncertainSPT-N valueis modeled
to belognormally distributedwithmeanvalueequal
thereportedaveragevalue, whilethec.o.v. equal 50%.
The50%c.o.v.isbasedonthe15%45%measurement
error c.o.v. estimated by Phoon (1995) but is made
slightlylargertoaccommodatetheinherentvariability
of SPT-Nvalues.
The c.o.v. of the measurement errors for unit
weights () is negligible, while that of the inherent
variabilityisaround9%.Therefore, inthecasethatthe
unitweightof alayer isreported, thisvalueistakento
bethemeanvalueandthec.o.v. is takento be10%.
Thedistributionistakentobelognormal becausethe
unit weight cannot benegative.
For Taipei clays, the ratio of undrained shear
strengthdividedby vertical effectivestress (s
u
/
v
) is
roughlyconstant0.21for depthdeeper than12mwith
c.o.v. =30%andisroughly0.36for depthwithin12m
withc.o.v. =30%. Thesenumbersaretakentobethe
mean valueand c.o.v. for thes
u
/
v
ratio of aclayey
layer.Thedistributionistakentobelognormal because
thes
u
/
v
ratiocannot benegative.
The v.s. (N
1
)
60
relation summarized in Chen
(2004) istakentoestimatetheeffectivefrictionangle
of sandsandgravels. Thestandarddeviationof this
relation is roughly 3
o
, quantifying themagnitudeof
the transformation uncertainties of the v.s. (N
1
)
60
relation.
If thelocationof watertableisknownataborehole,
itswater tablelocationistreatedasknown. However,
thereareboreholeswherethelocationsof water table
arenotdocumented. For thesecases, thedepthsof the
water table locations are taken to be uniformly dis-
tributedover[0m, 4.5m].Thisrangeisconcludedfrom
our Taipei borehole database, which shows in most
areasof Taipei, thewater tableisbetween0mto4.5m
deep.
Many of the test piles in the database rest on
sandstone layer. The end bearing capacities on the
sandstone (q
r
) are very uncertain. According to the
128
T
a
b
l
e
1
.
B
a
s
i
c
i
n
f
o
r
m
a
t
i
o
n
o
f
t
h
e
5
7
t
e
s
t
p
i
l
e
s
.
F
o
r
t
h
e
c
a
s
e
s
w
i
t
h
i
n
c
o
m
p
l
e
t
e
l
o
a
d
-
d
e
f
o
r
m
a
t
i
o
n
c
u
r
v
e
s
,
[
L
,
U
]
a
r
e
t
h
e
l
o
w
e
r
a
n
d
u
p
p
e
r
b
o
u
n
d
s
o
f
t
h
e
D
a
v
i
s
s
o
n
c
a
p
a
c
i
t
i
e
s
.
F
o
r
t
h
e
c
a
s
e
s
w
h
e
r
e
t
h
e
D
a
v
i
s
s
o
n
c
a
p
a
c
i
t
i
e
s
a
r
e
k
n
o
w
n
,
t
h
e
i
r
c
a
p
a
c
i
t
i
e
s
w
i
l
l
b
e
s
h
o
w
n
.
N
O
.
P
1
P
2
P
3
P
4
P
5
P
6
P
7
P
8
P
9
P
1
0
D
a
i
m
e
t
e
r
(
m
)
0
.
9
0
.
8
0
.
7
0
.
6
0
.
9
1
.
5
1
.
5
1
.
2
0
.
9
1
.
2
L
e
n
g
t
h
(
m
)
4
2
4
2
4
2
4
2
4
4
.
0
2
4
6
4
7
.
2
5
4
6
.
3
5
4
7
.
6
4
7
.
4
C
a
p
a
c
i
t
y
(
t
o
n
s
)
[
6
0
0
,
1
6
5
0
]
[
5
0
0
,
8
0
0
]
[
4
0
0
,
7
0
0
]
[
3
2
0
,
6
0
0
]
[
4
8
2
,
1
1
0
0
]
[
1
1
2
0
,
7
0
0
0
]
[
1
1
2
0
,
2
1
0
0
]
[
8
1
4
,
1
9
0
0
]
[
4
8
2
,
1
9
0
0
]
[
1
1
0
0
,
2
4
0
0
]
E
n
d
-
b
e
a
r
i
n
g
l
a
y
e
r
G
r
a
v
e
l
G
r
a
v
e
l
q
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
o
u
n
d
w
a
t
e
r
d
e
p
t
h
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
N
O
.
P
1
1
P
1
2
P
1
3
P
1
4
P
1
5
P
1
6
P
1
7
P
1
8
P
1
9
P
2
0
D
a
i
m
e
t
e
r
(
m
)
1
.
2
1
.
3
1
.
2
1
.
2
1
.
1
1
.
1
0
.
9
0
.
8
0
.
8
0
.
9
L
e
n
g
t
h
(
m
)
4
6
.
3
5
3
3
.
6
2
8
.
4
3
4
.
2
3
3
.
3
3
3
.
8
3
2
.
4
2
9
.
1
3
7
3
5
C
a
p
a
c
i
t
y
(
t
o
n
s
)
[
1
1
0
0
,
1
6
5
0
]
[
7
3
0
,
2
2
5
0
]
[
6
0
0
,
1
6
0
0
]
[
7
1
4
,
8
0
0
]
[
6
1
0
,
2
2
0
0
]
[
5
1
0
,
1
5
0
0
]
[
4
1
0
,
1
2
0
0
]
[
4
0
5
,
2
0
0
0
]
[
4
3
4
,
1
0
0
0
]
[
3
2
2
,
8
0
0
]
E
n
d
-
b
e
a
r
i
n
g
l
a
y
e
r
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
o
u
n
d
w
a
t
e
r
d
e
p
t
h
U
n
k
n
o
w
n
2
.
5
2
.
5
2
.
5
2
.
5
2
.
5
2
.
5
2
.
5
2
.
5
2
.
5
N
O
.
P
2
1
P
2
2
P
2
3
P
2
4
P
2
5
P
2
6
P
2
7
P
2
8
P
2
9
P
3
0
D
a
i
m
e
t
e
r
(
m
)
0
.
9
0
.
9
1
1
1
.
4
1
.
1
1
.
4
1
.
4
1
.
3
1
L
e
n
g
t
h
(
m
)
3
4
.
1
3
4
.
5
3
4
.
7
3
2
.
1
3
5
3
3
5
6
.
3
5
6
.
1
5
4
.
9
4
5
.
4
6
C
a
p
a
c
i
t
y
(
t
o
n
s
)
[
3
2
2
,
8
0
0
]
[
3
2
2
,
9
0
0
]
[
3
7
6
,
1
0
0
0
]
[
3
6
6
,
1
2
5
0
]
[
5
9
8
,
2
9
0
0
]
[
4
5
6
,
5
2
0
0
]
[
1
0
1
6
,
1
0
0
0
0
]
[
1
0
1
6
,
1
0
0
0
0
]
[
9
1
8
,
1
0
0
0
0
]
[
8
1
7
,
1
5
0
0
]
E
n
d
-
b
e
a
r
i
n
g
l
a
y
e
r
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
G
r
o
u
n
d
w
a
t
e
r
d
e
p
t
h
2
.
5
2
.
5
2
.
5
2
.
5
2
.
5
2
.
5
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
0
.
0
N
O
.
P
3
1
P
3
2
P
3
3
P
3
4
P
3
5
P
3
6
P
3
7
P
3
8
P
3
9
P
4
0
D
a
i
m
e
t
e
r
(
m
)
1
1
1
1
.
2
0
.
8
1
.
2
1
.
2
1
.
5
1
.
2
1
.
2
L
e
n
g
t
h
(
m
)
2
9
3
0
.
8
5
4
4
.
5
5
0
5
2
5
5
4
1
.
3
4
5
.
2
5
0
5
4
.
5
C
a
p
a
c
i
t
y
(
t
o
n
s
)
[
6
5
0
,
9
0
0
]
6
6
4
[
5
9
6
,
8
5
0
]
[
2
0
0
0
,
1
0
0
0
0
]
7
0
0
1
5
4
0
1
0
4
0
1
9
0
0
1
3
0
0
3
6
0
0
E
n
d
-
b
e
a
r
i
n
g
l
a
y
e
r
G
r
a
v
e
l
S
a
n
d
s
t
o
n
e
S
a
n
d
s
t
o
n
e
G
r
a
v
e
l
S
a
n
d
s
t
o
n
e
S
a
n
d
s
t
o
n
e
S
a
n
d
s
t
o
n
e
S
a
n
d
s
t
o
n
e
G
r
a
v
e
l
G
r
a
v
e
l
G
r
o
u
n
d
w
a
t
e
r
d
e
p
t
h
4
.
5
0
.
0
3
.
0
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
N
O
.
P
4
1
P
4
2
P
4
3
P
4
4
P
4
5
P
4
6
P
4
7
P
4
8
P
4
9
P
5
0
D
a
i
m
e
t
e
r
(
m
)
1
.
2
1
.
5
1
.
5
1
.
5
1
.
5
1
.
2
1
.
2
1
1
.
5
1
.
5
L
e
n
g
t
h
(
m
)
5
0
6
6
.
1
7
2
.
5
7
4
8
1
.
1
5
3
.
8
5
5
7
.
4
5
5
.
5
7
2
.
8
7
9
.
8
C
a
p
a
c
i
t
y
(
t
o
n
s
)
[
1
4
5
0
,
1
0
0
0
0
]
[
2
1
9
4
,
1
0
0
0
0
]
[
2
0
6
1
,
1
0
0
0
0
]
[
2
0
6
1
,
1
0
0
0
0
]
[
2
0
6
1
,
1
0
0
0
0
]
[
9
5
0
,
1
0
0
0
0
]
[
1
9
7
0
,
2
2
5
0
0
]
[
1
5
0
0
,
1
6
5
0
]
3
0
0
0
[
3
5
5
0
,
4
4
0
0
]
E
n
d
-
b
e
a
r
i
n
g
l
a
y
e
r
G
r
a
v
e
l
S
a
n
d
s
t
o
n
e
S
a
n
d
s
t
o
n
e
S
a
n
d
s
t
o
n
e
S
a
n
d
s
t
o
n
e
G
r
a
v
e
l
G
r
a
v
e
l
G
r
a
v
e
l
S
a
n
d
s
t
o
n
e
S
a
n
d
s
t
o
n
e
G
r
o
u
n
d
w
a
t
e
r
d
e
p
t
h
[
0
,
4
.
5
]
[
0
,
4
.
5
]
[
0
,
4
.
5
]
[
0
,
4
.
5
]
[
0
,
4
.
5
]
[
0
,
4
.
5
]
[
0
,
4
.
5
]
[
0
,
4
.
5
]
[
0
,
4
.
5
]
[
0
,
4
.
5
]
N
O
.
P
5
1
P
5
2
P
5
3
P
5
4
P
5
5
P
5
6
P
5
7
D
a
i
m
e
t
e
r
(
m
)
1
.
5
2
1
.
2
1
1
.
5
1
1
L
e
n
g
t
h
(
m
)
6
8
.
3
3
7
.
5
5
0
1
3
2
3
2
9
4
4
.
5
C
a
p
a
c
i
t
y
(
t
o
n
s
)
[
2
4
3
6
,
3
3
0
0
]
[
2
0
0
0
,
1
0
0
0
0
]
[
1
3
8
0
,
4
5
0
0
]
[
5
5
0
,
1
0
8
0
]
[
8
0
0
,
1
2
4
0
]
[
6
5
0
,
1
0
0
0
]
[
6
0
0
,
8
0
0
]
E
n
d
-
b
e
a
r
i
n
g
l
a
y
e
r
S
a
n
d
s
t
o
n
e
G
r
a
v
e
l
G
r
a
v
e
l
S
a
n
d
s
t
o
n
e
S
a
n
d
s
t
o
n
e
G
r
a
v
e
l
S
a
n
d
s
t
o
n
e
G
r
o
u
n
d
w
a
t
e
r
d
e
p
t
h
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
U
n
k
n
o
w
n
129
Table2. Thetwo deterministic predictivemodels adopted
for theanalysis.
Model Skinfriction Endbearing
SPT-N Clay: f
s
=S
(1)
u
Clay: q
b
=9S
u
Sandandgravel: Sandandgravel:
f
s
=N/315T/m
2
q
b
=30N 1500T/m
2
Sandstone: q
b
=q
r
Static Clay: f
s
=S
(1)
u
Clay: q
b
=9S
u
Sandandgravel: Sandandgravel:
f
s
=0.67
v
tan f
(2)
max
q
b
=
v
N
q
q
(3)
max
Sandstone: q
b
= q
r
Remarks: N istheaverageSPT-N valueof thesoilsnear the
piletip; (1) suggestedbyTomlinson(1994);
(2) f
max
suggested by DM7-2 (1982); (3) N
and q
max
suggestedbyDM7-2(1982)
codeinTaiwan, thepilesocket length is required to
begreater than 2 to 3 meters (most test piles in our
databasesatisfy this requirement). Under this condi-
tion, theend bearing capacities on sandstonecan be
predictedby itsuniaxial compressionstrength. How-
ever, theactual endbearingcapacitiescanstill bevery
different fromthe predicted ones depending on the
constructionqualityof piletips.
Inthecasethat theconstructiontechniquesarerel-
atively uniformandthat thebedrock quality doesnot
vary significantly, it may bereasonableto assumeq
r
of all test piles areidentically distributed. Moreover,
variabilitiesinq
r
duetoconstructionqualityvariabil-
ities can beassumed independent. Therefore, in this
studyq
r
of all testpilesareassumedtobeindependent
but identically distributed. The distribution is taken
to belognormal becauseq
r
cannot benegative. The
commonmeanvalue(j
r
) andc.o.v. (
r
) areunknown.
AccordingtoChen(2000), theendbearingcapacities
inTaipei areusually intherangeof [50tons/m
2
, 300
tons/m
2
]. Therefore, inthisstudyj
r
isconstrainedin
thisrange.
3.2 Modelinguncertainties
Themost important uncertaintyintheproblemisper-
haps themodelinguncertainty inpredictingultimate
bearing capacity of abored pile. Predictions on pile
bearingcapacitiescanbemadebasedonthefollowing
equation:
where R=ultimate bearing capacity of a pile,
Q
s
=skinresistanceof thepileshaft,Q
b
=endbearing
resistance,f
s
=skinresistancestress, q
b
=endbearing
stress, A
s
=surface area of pile shaft, A
b
=cross-
section area of pile tip. In Taiwan, there are two
commonmethodsof calculatingf
s
andq
b
: theSPT-N
andstaticmethods, listedinTable2. Thesetwometh-
ods are developed primarily based on the Taiwans
Foundation Design Code for Building (TGS 2001).
Thepredictedcapacity R is clearly afunction of the
soil parametersandindicesdescribedinthepreceding
section.
However, theactual capacity of apile, denotedby
C, will generally deviatefromthepredictedvalueR
duetotheinaccuracy of thepredictivemodel. Let C
j
bethemeasuredDavissoncapacity of thej-thpilein
thedatabase, andletR
j
bethecapacitypredictedeither
by theSPT-N model or by thestatic model. Theratio
betweenC
j
andR
j
iscalledthemodel factor
j
, i.e.:
j
=C
j
/R
j
. Thisfactor characterizesthemodel uncer-
tainty. In most literature (e.g.: Whitman (1984) and
Baker etal. (1991)), themodel factor
j
isassumedto
belognormally distributed. Furthermore, in thecase
that all pile tests are fromthe same region, it may
bereasonableto assumethemodel factors of differ-
ent pilesareidentically distributed. Therefore, inthis
studyitisassumedthatthemodel factorsof the57test
pilesareindependent identically distributedasalog-
normal distributionwhosemeanvalue(j
) andc.o.v.
(
) areunknown.
Among all the uncertain parameters and indices,
there are four hyperparameters including j
,
j
r
and
r
. Thesehyperparametersarespecial because
the model factor and the end bearing capacity q
r
are probably the most influential parameters in the
entiremodel. Other parameters suchas unit weights,
undrainedshearstrengths,etc.,arerelativelylessinflu-
ential. Moreover, these four parameters are special
becauseall pilessharethesamehyperparameters.The
consequenceisthatthecapacitiesof differentpilesare
dependent althoughtheuncertainsoil parametersand
indicesof differentpilesareindependent.Thesehyper-
parametersplaycrucial rolesincalibratingresistance
factors.
Intheperspectiveof Bayesiananalysis,itispossible
to updatethePDFs of thehyperparameters by using
thetest resultsof the57piles. After theupdating, the
designof anewpilecanbemadebasedontheupdated
PDFsof thehyperparameters, whichhaveabsorbed
theinformationof the57piles. By doingso, calibra-
tionof resistancefactors will bepossible. Therefore,
thehyperparametersplaytheroleof passingoldinfor-
mation(thetest results of the57piles) into newpile
designs becausethe57piles andnewpiles sharethe
samehyperparameters.
3.3 Results of preliminary deterministic analysis
Apreliminarystudyisperformedtocomparetheactual
Davissoncapacitiesof the57pileswiththeirpredicted
values. Inorder to computethepredictedcapacities,
all uncertainsoil parameters andindices requiredby
theSPT-N or staticmodel suchasunit weights, shear
strengths, water tablelocations, etc. arefixedat their
averagevalues.Theendbearingcapacityq
r
isfixedat
200tons/m
2
.Thepredictedbearingcapacitiesof the57
pilesandtheiractual Davissoncapacitiesarecompared
inFigure2forboththeSPT-Nandstaticmodels. Inthe
cases wheretheload-deformation curves areincom-
plete, the lower and upper bounds of the Davisson
capacitiesaredrawninthefigure.
130
Figure2. ThecomparisonbetweentheDavissoncapacities
and thepredicted ones: left is for theSPT-N model, while
right is for thestatic model. Thex datapoints correspond
to the piles with complete load-deformation curves, while
thebars showthelower andupper bounds for thepilewith
incompletecurves.
Atthefirstsight, itisevidentthattheSPT-Nmodel
ismoreconservative, but itscorrelationwiththedata
points with complete load-deformation curves (i.e.:
thex datapoints inthefigure) seems stronger than
thatof thestaticmodel. Moreover, fromthescattering
of thedatapoints, it seems reasonableto assumethe
57model factorstobeidenticallydistributedbecause
thedeviationsof thepredictedcapacitiestotheactual
Davissoncapacities or to theDavissonbounds seem
relativelyhomogeneous.
4 FULL PROBABILISTICANALYSIS
4.1 Calibration of resistance factors
Giventheloadtestdataof the57piles, itispossibleto
calibratethereliability-basedresistancefactor for pile
designsinTaipei. Thereliability-baseddesigncanbe
achievedbyrestrictingthefailureprobabilityof anew
designconditioningonthepast data, i.e. consider the
followingequation:
whereC
new
istheuncertainbearingcapacityof anew
pile, andc
isthedesignloadof thepile; theC

new
c
is thefailureevent; C
1:57
denotes our load test data;
P(C
new
c
| C
1:57
) is the failure probability condi-
tioningonthedatabasedata, which, inprinciple, can
becomputedwithBayesiananalysis; P
F
isthetarget
failureprobability. Equation(2) canberewrittenas
where
new
isthemodel factor of thenewpile, which
is lognormally distributed with mean value equal to
j
andc.o.v. equal to
; q
r,new
is thesandstoneend
bearingcapacityof thenewpile, whichislognormally
distributedwithmeanvalueequal toj
r
andc.o.v.equal
to
r
; Y
new
contains all uncertain soil parameters of
thenewpilesiteother thanthesandstoneendbearing
capacity. NotethatR(Y
new
, q
r,new
) isuncertainbecause
bothY
new
andq
r,new
areuncertain.
Themeanvaluesof thesoil propertiesY
new
canbe
obtained fromproper site investigations of the new
pile site through in-situ and laboratory tests, while
theirvariabilitiescanbeestimatedbasedonarguments
similar to those in the section of Quantification of
Uncertainties to obtain their prior PDFs. The rela-
tionshipbetweentheresistancefactorandtargetfailure
probability for thenewpiledesigncanbederivedas
follows:
whereY
n
new
andq
n
r,new
arethenominal values of Y
new
and q
r,new
, usually taken to be their mean values;
=c
/R(Y
n
new
, q
n
r,new
) is exactly theresistancefactor,
whichisthereciprocal of thesafetyfactor. Onecansee
thattheresistancefactorcanbesimplyestimatedasthe
100 P
F
percentileof G(
new
,Y
new
,q
r,new
)conditioning
onthedataC
1:57
. If onecanobtainstochasticsimula-
tionsamplesof G(
new
,Y
new
,q
r,new
)conditioningonthe
dataC
1:57
, denotedby {G
1
,, G
N
}, therelationship
betweenresistancefactor andtarget failureprobabil-
itycanbeestimatedasthefollowingaccordingtothe
Lawof LargeNumber:
where1() is theindicator function: it is unity if the
insidestatement is trueand is zero otherwise. Once
thisrelationshipisobtained,thecalibrationof theresis-
tancefactor isachievedsinceonecannowdetermine
therequiredresistancefactorof anewpiledesignfrom
thetarget failureprobability.
Regarding the question of how to draw samples
of {G
1
,, G
N
} conditioning the data C
1:57
, since
G(
new
, Y
new
,q
r,new
) is a function of
new
, q
r,new
and
Y
new
, it suffices to draw samples of
new
, q
r,new
and
Y
new
conditioning on the data C
1:57
, i.e. draw sam-
plesfromf (
new
,Y
new
,q
r,new
|C
1:57
). Firstof all, thesoil
parameters at the new site Y
new
are independent of
the past data C
1:57
, so drawing Y
new
samples condi-
tioningonC
1:57
isthesameasdrawingY
new
samples
fromits prior PDF. However, drawing {
new
,q
r,new
}
samplesfromf (
new
, q
r,new
|C
1:57
) ismorechalleng-
ing since the past data C
1:57
contains information
about {
new
,q
r,new
}, i.e. they are dependent through
thehyperparameters {j
}and {j
}. In fact,
thesesamplesof {
new
,q
r,new
}playtheroleof convey-
inginformationfromthetest piledatabasetothenew
piledesign. Inthenext section, stochastic simulation
techniquesthatareusedtodraw{
new
,q
r,new
}samples
conditioningonC
1:57
will bediscussed.
131
4.2 Stochastic simulation of {
new
,q
r,new
} samples
conditioning on past data C
1:57
Inthecasethat thesoil parametersandindicesof the
57testpilesY
1:57
andendbearingcapacitiesq
r,1:57
are
fixedat their nominal values, i.e.: thetest valuesY
n
1:57
andsomechosenvaluesof q
n
r
, f (j
, j
|C
1:57
)
hasaverysimpleexpression:
wheref (C
1:57
|j
, j
) is called thelikelihood
function. In this study, theprior PDFs f (j
), f (j
),
f (
) andf (
) aretakentobeflatPDFs, i.e.: noprior

informationonthem. Therefore
where j =1,, 8 indicates the piles with complete
load-deformation curves (therefore their Davisson
capacities are known), while j =9,, 57 indicates
the piles with incomplete load-deformation curves
(therefore only the lower and upper bounds {L
j
,U
j
}
of their Davisson capacities are known). Since the
model factorsareindependentidenticallylognormally
distributed,
where+()isthecumulativedensityfunction(CDF)of
thestandardGaussianPDF. TheMarkovchainMonte
Carlo method can then beapplied to obtain samples
fromf (j
,

, j
|C
1:57
). Oncethe{j
, j
}samplesareobtained, the
new
samplecanbeeas-
ily obtainedfromalognormal PDF whosemeanand
c.o.v. are at the sampled value {j
}, while the
q
r,new
sample can be easily obtained froma lognor-
mal PDF whose mean and c.o.v. are at the sampled
value{j
}. Thesesampleswill beusedin(4) and

(5) to obtain the -P
F
relation, i.e.: to calibrate the
resistancefactors.
Figure3. Thetimehistoriesandhistogramsof thesamples
for thehyperparameters.
5 ANALYSISRESULTSANDVERIFICATIONS
5.1 Samples of stochastic simulations
Ten thousand samples are drawn fromf (
new,
q
r,new
|C
1:57
) forbothdesignmodels, andtheburn-inperiods
are shown in Figure 3. The samples after the burn-
in periods contain interesting information about the
behaviors of the57test piles. For instance, thesam-
plesof theq
r,new
parameter, asshowninFigure3, are
distributedasf (
new
|C
1:57
).Thehistogramof thesam-
ple values, also shown in the figure, basically gives
therelativedegreeof plausibilityof thesandstoneend
bearingcapacity learnedfromthedata. If theSPT-N
model isadoptedfor designs, thesandstoneendbear-
ingcapacity q
r
is inaverage246tons/m
2
withc.o.v.
equals55%.
Thesamplesfromf (
new
|C
1:57
), alsoshowninFig-
ure2, gives thepossiblerangeof themodel factor
learned fromthe data of the 57 test piles. Fromthe
histogramsinFigure3, itisfoundthatthemeanvalue
and c.o.v. of the model factor are 1.247 and 34.9%
for theSPT-Nmodel andare1.102and23.6%for the
static model. Therefore, theactual Davissoncapacity
of apileinTaipei isinaverage1.247timeslarger than
thecapacitypredictedbytheSPT-Nmodel withc.o.v.
equals34.9%. Forthestaticmodel, theactual capacity
of apileinTaipei isinaverage1.102timeslarger with
c.o.v. equals23.6%.
Figure4showsthecomparisonbetweentheactual
capacitiesandthenominal onespredictedbytheSPT-
N model multiplied by the average values of the
samples, wherethenominal capacities arecomputed
withall uncertainsoil parametersheldattheirnominal
valuesandthesandstoneendbearingcapacityheldat
the average value of q
r
samples. It is clear that the
132
Figure4. ThecomparisonbetweentheDavissoncapacities
and thepredicted ones multiplied by theaveragevalues of
the samples.
predictedcapacitiesunbiasedlyreflecttheactual ones,
indicating that the analysis results are consistent. A
similar conclusionisfoundfor thestaticmodel.
5.2 Relationship between resistance factor and
target failure probability
The procedure introduced in a previous section is
employedto estimatetherelationshipbetween resis-
tancefactorandtargetfailureprobability(orthetarget
reliabilityindex
=+
1
(P
F
). Ingeneral, theesti-
mated relationship is not unique: it depends on the
configurationof thenewpiledesign, thesoil profiles,
and the chosen design model. This non-uniqueness
imposesadifficulty, butitcanberesolvedbyselecting
arepresentativeset of piledesignscenarios inTaipei
andestimatingthe
relationshipsfor all scenar-

iostoobtainthepossiblerangeof therelationship. One
canfirstdeterminethetargetfailureprobabilityP
F
for
piledesigns of Taipei, thenconvert ininto thetarget
reliability index
=+
1
(P
F
), so therangeof the
requiredresistancefactor canbefound.
Inthispaper, thesitesof the57pilesinthedatabase
aretaken as therepresentativeset of soil profiles in
Taipei. Numerous scenarios of piledesigns arecon-
sidered and the corresponding
relationships
are computed. It is found that the
relation-
ships strongly depend on the chosen design model,
i.e.: SPT-Nor staticmodel.
FortheSPT-Nmode,thecalibratedrelationbetween
thetarget reliabilityindexandtherequiredresistance
factor is showninFigure5. Inthecasethat only the
eighttestswithcompleteload-deformationcurvesare
used for the calibration, the required resistance fac-
tor (see the left plot in Figure 5) is in the range of
[0.2, 0.3] (required safety factor between 3.3 and 5)
for thetarget reliability index
=3andis less than

0.2(requiredsafety factor greater than5) for
=4.
In thecurrent Taiwan codefor piledesigns, asafety
factor of 3(equivalenttoaresistancefactor of 0.33) is
requiredfor thebearingcapacityconsideration. From
theleft plot inFigure5, sucharesistancefactor cor-
responds to reliability index betweenintherangeof
[2.3, 2.8], or failureprobabilityintherangeof [0.003,
0.01].
In thecasethat all information (all test results of
the57piles) areusedfor thecalibration, therequired
resistancefactor (seetheright plot inFigure4) is in
Figure5. Thecalibratedrelationbetweentarget reliability
indexandresistancefactor for theSPT-Ndesignmodel. The
left figure is based on the eight test results with complete
load-deformationcurves, whiletheright figureis basedon
all availableinformation.
therangeof [0.3, 0.45] (requiredsafetyfactorbetween
2.2and3.3) for thetarget reliabilityindex
=3and
is intherangeof [0.22, 0.35] (requiredsafety factor
between 2.8 and 4.5) for
=4. For thecoderegu-

lation of resistance factor=0.33, the corresponding
reliability index betweenis intherangeof [3, 4], or
failure probability in the range of [0.00003, 0.001].
It is obvious that when the test results with incom-
plete load-deformation curves are incorporated into
theanalysis, thecalibratedresistancefactor increases
significantly, i.e.: therequiredsafetyfactor decreases
significantly.
Notice that there is one curve in the left plot in
Figure5wheretherequiredresistancefactor issignif-
icantlylessthantheothers. That curvecorrespondsto
avery short pilerestingonthesandstonelayer. This
phenomenonwill bediscussedindetail later.
When thestatic design model is chosen, thecali-
brated
relationsareshowninFigure6. Unlike
the SPT-N model, the calibrated resistances with or
without all availableinformationarenot very differ-
ent. Therequired resistancefactor is in therangeof
[0.3, 0.4] (requiredsafetyfactor between2.5and3.3)
for the target reliability index
=3 and is in the
range of [0.2, 0.28] (required safety factor between
3.6 and 5) for
=4. For the code regulation of

resistancefactor=0.33, thecorrespondingreliability
index betweenis intherangeof [2.8, 3.4], or failure
probabilityintherangeof [0.0003, 0.003].
Again, thereareseveral curves in Figure6 where
therequiredresistancefactor issignificantlylessthan
theothers.Thesecurvescorrespondtoveryshortpiles
restingonthesandstonelayer.
6 CONCLUSION
(a) Fortheloadtestswithincompleteload-deformation
curves, itispossibletoidentifythelowerandupper
boundsof theDavissoncapacities. Moreover, it is
evidentthattheseincompleteloadtestdatamaybe
valuable in calibrating resistance factors. For the
133
Figure6. Thecalibratedrelationbetweentarget reliability
index andresistancefactor for thestatic designmodel. The
left figure is based on the eight test results with complete
load-deformationcurves, whiletheright figureis basedon
all availableinformation.
SPT-Ndesignmodel, thecalibratedresistancefac-
tors changesignificantly after theincompletetest
resultsaretakenintoconsideration.
(b) Basedonall loadtestresults(includingtheincom-
pleteones) intheTaipei region, themeanvalueand
c.o.v. of themodel factorareidentifiedtobearound
1.247and34.9%for theSPT-N designmodel and
around1.102and23.6%forthestaticdesignmodel.
Thisindicatesthat bothdesignmodelsareconser-
vative. Moreover, the SPT-N design model is in
average more conservative than the static model.
However, thevariabilityof theSPT-Nmodel seems
larger, judgingfromthe34.9%c.o.v., thanthestatic
model.
(c) Basedontheanalysisresultswithall availabledata,
a safety factor of 3 required by theTaiwan code
correspondstoareliability index between3and4
(or failureprobabilitybetween0.00003and0.001)
fortheSPT-Nmodel andcorrespondstoareliability
index between 2.8 and 3.4 (or failure probability
between 0.0003 and 0.003) for the static model.
For designs of pileultimatecapacity, therequired
resistancefactorisusuallyaround3.Therefore, the
safetyfactorof 3requiredintheTaiwancodeseems
reasonable.
(d) Thecalibratedresistancefactorsforshortpilesrest-
ing on thesandstonelayer areparticularly small.
This is becausetheir endbearingcapacities dom-
inatetheir ultimatecapacitiesandtheendbearing
resistanceq
r
of thesandstonelayer is very uncer-
tain. Therefore, the ultimate capacities for these
short piles are also very uncertain, hence small
resistancefactors may berequired to compensate
the high uncertainties. It is found that for piles
longerthan30m, theendbearingcapacityceasesto
dominate,andtheir
relationswill fall intothe

ordinaryrangesshowninFigures5and6.Asacon-
sequence, when implementing the recommended
resistance factors mentioned in the previous sec-
tion, oneshouldbecareful whenthepilesareshort
(shorterthan30m) andarerestingonthesandstone
layer. A muchsmaller resistancefactor shouldbe
takenfor suchpiles.
REFERENCES
Barker,R.M.,Duncan,J.M.,Rojiani,K.B.,Ooi,P.S.K.,Tan,
C. K. andKim. S. G. (1991). Manuals for the Design of
Bridge Foundations. NCHRP Report 343, TRB, National
ResearchCouncil, Washington, DC.
Chen, D.S. (2000). Piletypes, bearingcapacity andmecha-
nisms considerationsinconstruction. Publication D19,
TaiwanConstructionResearchInstitute, 133.
Chen, J.R. (2004). Axial Behavior of Drilled Shafts in
Gravelly Soils, PhDthesis, Cornell University.
Davisson, M.T. (1972). High capacity piles. Proceedings,
Lecture Series, Innovations in Foundation Construction,
ASCE, IllinoisSection.
DM 7-2(1982). Foundations and Earth Structures Design
Manual 7.2, Department of the Navy Naval Facilities
EngineeringCommand, Alexandria, VA.
McVay, M. C., Birgisson, B., Zhang, L. M., Perez, A., and
Putcha, S. (2000). Load and resistance factor design
(LRFD) for driven piles using dynamic methodsA
Floridaperspective. Geotech. Test. J., 23(1), 5566.
Paikowsky, S. G., Birgisson, B., McVay, M., Nguyen, T.,
Kuo, C., Baecher, G.,Ayyab, B., Stenersen, K., OMalley,
K., Chernauskas, L., and ONeill, M. (2004). Load and
Resistance Factor Design (LRFD) for Deep Founda-
tions. NCHRPFinal Report507,TransportationResearch
Board, Washington, DC.
Phoon, K.K. (1995). Reliability-based Design of Founda-
tions forTransmission Line Structures, Ph.D. Dissertation,
Cornell University.
Tomlinson, M.J. (1994). Pile Design and Construction Prac-
tice. FourthEd., E & FNSpon, London.
TGS, Taiwan Geotechnical Society (2001). Foundation
Design Code for Building. TaiwanGeotechnical Society
(TGS), Taipei.
Whitman, R.V. (1984). Evaluatingcalculatedriskingeotech-
nical engineering. ASCE Journal of Geotechnical Engi-
neering Division, 110(2), 145-188.
Zhang, L.M. (2004). Reliabilityverificationusingproof pile
load tests. ASCE Journal of Geotechnical and Environ-
mental Engineering, 130(11), 1203-1213.
134
Sensitivityanalysisof designvariablesfor caissontypequaywall
G.L.Yoon& H.Y. Kim
Korea Ocean Research & Development Institute, Ansan, Korea
Y.W.Yoon
Inha University, Incheon, Korea
K.H. Lee
Konyang University, Nonsan, Korea
ABSTRACT: Sensitivityanalysiswascarriedout toquantifytheeffectsof designvariablesonthereliability
indicesof caissontypequaywall usingHasofer-Lindapproach.A targetreliabilityindexfromreliabilityanalysis
wasestimatedwiththreefailuremodesof overturning,slidingandbearingcapacityforstabilitiesof thestructures.
Theresultsshowedthat thereliability indicesfor bothbearingcapacity andslidingfailuremodesweremostly
lowerthanthoseforoverturningfailuremode.Consequently,themajordesignvariablesaffectingonthereliability
indices were found to be coefficient of friction, residual water pressure and resistance moment for sliding,
overturningandbearingcapacity failure, respectively. Especially, coefficientsof sensitivity of boththeinertia
forceanddynamic water pressurewithhighvariationinseismic cases werenot largeandthis corresponds to
theprevious researchobtainedby correlationstudy betweenvariations of designvariables andcoefficients of
sensitivity.
1 INTRODUCTION
Newdesigncodesbasedonlimit statedesign, perfor-
mancebased design and so on havebeen developed
accordingtothechangesof global codeinportandhar-
bordesign.Also, duringseveral years, therehavebeen
efforts to apply reliability theory to existing safety
factor method.
Nagao et al. carried out the research about reli-
ability based design (RBD) of quay wall structures
and investigated correlations between lifecyclecost
and probability of failureusing FORM (2001). And
Yoneyama et al. estimatedprobabilityof failuretoget
exacttargetreliabilityindexfrommaterial of thefailed
gravitytypequaywall andloadfactorsfor earthquake
load (2000). In 2004, as a result of an international
joint study, Chineseresearchers investigated thetar-
get reliability index of quay walls and models for
designvariables andpartial safety factors for sliding
andoverturningresistanceof existingtengravityquay
walls.
Thepurposes of this study is to consider reliabil-
ity indices against each failure mode using reliabil-
ity analysis of quay walls, to investigate the effects
of each design variable on reliability indices using
sensitivity analysis, and to estimate partial safety
factors.
2 RELIABILITY ANALYSIS
2.1 Limit state functions
Failure modes of a caisson type quay wall are
extensively classified into sliding, overturning,
bearingcapacity andso on. Limit statefunctions for
eachfailuremodearecalculatedasfollows:
Sliding:
Overturning:
Bearingcapacity:
In the limit state function for sliding failure, f , W,
C, Q, P
U
, P, S and k
h
are coefficients of friction,
135
Figure1. Crosssectionsfor theanalysis.
self weighof caisson, craneloading, surcharge, uplift
pressure, waveforce, collisionloadingandhorizontal
seismiccoefficient, andsubscriptV andH meanver-
tical and horizontal component, respectively. In the
limit state function for overturning failure, a
i
is the
distancebetween center of moment andactingpoint
of loading. At failure, q
a
, F
V
, B
C
ande areallowable
bearingcapacity, total vertical force, widthof caisson
andeccentric coefficient, respectively. B, E, R andD
arerefertoascommonmeanbuoyancy, earthpressure,
residual water pressureand dynamic water pressure,
respectively.
2.2 Design conditions
Inthisstudy,targetreliabilityindex(
T
)was2.05.Two
sections whichhavedifferent geometries of thecais-
son, designtidal levelsandsurchargeswereanalyzed,
respectively. Figure 1 and 2 show cross sections of
thequaywalls.Theloadcombinationconsistsof ordi-
nary and seismic case. Dynamic water pressure and
inertiaforcewereonlyappliedtoseismiccase. Coef-
ficientsof variation(COV) for mainrandomvariables
arepresentedinTable1.
2.3 Results for reliability analysis
Table 2 shows the results of the most critical load
cases of the reliability indices estimated by FORM.
Figure2. Correlation between reliability index and sensi-
tivityfactor.
Table1. COV andprobability distributionfor typical ran-
domvariables.
Random Probability
variable Mean COV distribution
Unit weight (kN/m
3
)
Reinforcedconc 24.5 0.02 Normal
Plainconc 23.0 0.02 Normal
Gravel 18.0 0.04 Normal
Coefficient of friction 0.60 0.15 Normal
Staticearthpressure 0.10 Normal
Seismiccoefficient 0.077 0.25 TypeII
Table2. Reliabilityindicesfor critical loadcases.
Section
Sliding 4.199 2.201
Overturning 12.804 8.009
Bearingcapacity 1.416 3.767
Section indicates safe values, whereas section
indicates unstable values less than target reliability
index. The value of safety margin for overturning
failureisthelargest oneamongthethreecases.
3 SENSITIVITY ANALYSIS
3.1 Sensitivity factor
Sensitivity factor is thecoefficient of linear approxi-
mationof limitstatefunction. If thereisnocorrelation
amongrandomvariables, Sensitivityfactor isdefined
asfollowing:
136
Figure3. Normalizedsensitivitydistributionfor ordinarycase(upper) andseismiccase(lower).
Equation (4) means a direction cosineof reliability
indexontheaxisof eachrandomvariableof thestan-
dardizedspace.Whenlimitstatefunctionisdefinedas
Z =X
2
X
1
, equation(4) isrepresentedinFigure2.
Incaseof nocorrelationamongeachrandomvariable,
thesensitivityfactortakespositivevalueforresistance
andnegativevaluefor loadeffect.Andthesummation
of itssquareisunity.
On the other hand, if correlation exists among
randomvariables, coefficientof correlationisconsid-
eredusingstandarddeviationandsensitivityfactor of
limit statefunction. This is represented as following
equations:
3.2 Results for sensitivity analysis
To find the effects of a specific design variable on
safetyof eachstructure, sensitivityanalyseswereper-
formedfor all loadcasesusingFORM. Only aset of
results representing the characteristics of sensitivity
for designvariableswerepresented.
Figure 3 shows normalized sensitivity factors for
each failure mode of ordinary case. When sliding
failure occurs, the most affecting design variable
was coefficient of frictionbetweencaissonandbase
ground, followed by horizontal earth pressure (E
h
),
self weight(W), buoyancy(B), residual waterpressure
(P
wr
) and vertical earth pressure (E
v
), respectively.
Similarly, in theseismic case, coefficient of friction
andhorizontal earthpressurewerethemost sensitive
ones, followedbyself weight, buoyancyanddynamic
waterpressure(P
wd
), inertiaforcebyearthquakes(F
i
),
vertical earth pressure, and residual water pressure,
respectively. For overturning failure which is deter-
minedbythemagnitudebetweenoverturningmoment
and resistance moment, the moments due to its self
weight of caissonandbuoyancy werethemost sensi-
tive, followedby themoment dueto horizontal earth
pressure, vertical earth pressure and residual water
pressure, respectively. Incaseof theseismiccase, the
momentduetohorizontal earthpressurewasthemost
sensitive, followedbythemoment duetoself weight,
buoyancy, vertical earthpressure, dynamicwaterpres-
sure, inertiaforceby earthquakes andresidual water
pressure, respectively. For bearing capacity failure,
maximumsensitivityfactorswereresistancemoment
(M
y
) andtotal vertical force(F
v
).
Therefore, they highly effects the safety as the
designvariablechanges. Incaseof theseismic case,
similar resultswereobtained.
To investigate the effect of reliability indices on
the sensitivity factors, the trends of changes of the
sensitivityfactorswereobservedshowingthefluctua-
tionof widthof caissons.Figure4showsthesensitivity
changeswithreliabilityindex.
137
Figure4. Sensitivitychangeswithreliabilityindexfor ordinarycase(upper) andseismiccase(lower).
For slidingfailure, thesensitivity of coefficient of
frictionandhorizontal earthpressurewereincreased
asreliability indicesincrease, whereasthesensitivity
of otherswereslightlydecreasedor constant.Thesen-
sitivityof coefficientof frictionwasnearlyapproached
tounity, sothatithighlyaffectedthereliabilityindices.
Whereas, forresidual waterpressure, theyapproached
to zero, so that it had no effect. The similar results
wereobtained in ordinary case, too. For overturning
failure, the sensitivities of the moment due to self
weightandhorizontal earthpressurewereremarkably
increased as the reliability indices increase. On the
otherhand, thesensitivitiesof themomentduetoiner-
tia force, vertical earth pressure and buoyancy were
decreasedwithrespecttoreliabilityindices. However,
thesensitivitiesof themoment duetodynamic water
pressurewerenotchangedwhenthereliabilityindices
wereincreased. For bearingcapacity failure, increas-
ingreliabilityindiceswereobtainedasdecreasingthe
sensitivity of theoverturningandresistancemoment.
However, increasingreliability indiceswereobtained
asincreasingthesensitivityof thetotal vertical force.
Fromtheresults,asignificantpointisthatalthoughthe
inertiaforceanddynamicwater pressurearethevari-
ables affectedby seismic coefficient withhighCOV,
thechangesof thesensitivityarenotsignificantasthe
reliabilityindiceschange.Theresultsshowedthesame
correlationbetweentidal level andsensitivityfactorof
breakwatersinvestigatedbyNagao et al. (2004).Thus,
all thesensitivitiesof thevariableswithhighCOV are
not significant.
4 ESTIMATIONOF PARTIAL SAFETY
FACTORS
Partial safetyfactorsforthedesignbylevel 1approach
wereestimated. Threecomponentswereemployedin
Table3. Partial safetyfactorsfor slidingmode.

Section
Condition Ordinary Seismic Ordinary Seismic
f 0.273 0.543 0.442 0.923
E
h
1.077 1.269 1.061 1.071
E
v
0.997 0.979 0.996 0.990
W 0.994 0.994 0.996 0.999
P
wr
1.001 8.149 1.009 1.002
B 1.009 1.009
F
i
1.004 1.044
P
wd
1.053 1.030
Table4. Partial safetyfactorsfor overturningmode.

Section
Condition Ordinary Seismic Ordinary Seismic
E
h
1.698 3.527 2.640 2.313
E
v
0.611 0.229 0.133 0.328
W 0.661 0.815 0.456 0.951
P
wr
1.015 1.008 1.286 1.048
B 1.493 1.269
F
i
1.054 2.089
P
wd
1.511 1.602
the partial safety factors. They are target reliability
index (2.05), weight applied to each design variable
accordingto sensitivity factors, andthedegree(how
designvariablesaffect stabilityof structures).
As shown in Table 3 and 4, the values of partial
safety factors for sliding mode are not significantly
scatteredbetweenordinaryandseismiccases, whereas
thevalueof coefficientof frictionisincreasedbetween
ordinary and seismic case. However, the values of
138
coefficients of friction in each quay wall are much
scattered. The values of coefficients of inertia force
and dynamic water pressure related to earthquake
approachtounityandtherearenodifferencesbetween
thevaluesof coefficientsinthetwosections.
For overturningmode, thepartial safety factors of
thedesign variables related to earthquakewererela-
tivelylargeandthepartial safetyfactor for self weight
in seismic case was increased comparing with the
ordinarycase.
5 CONCLUSIONS
Toinvestigatethesafetymarginfor eachfailuremode
of caisson typequay walls and effects of thedesign
variablesonsafety, reliabilityandsensitivityanalysis
wereperformedfor twostructures.
Asaresultof reliabilityanalysis, Safetymarginfor
overturningfailurewas thelargest. Usingsensitivity
analysis, in the ordinary case, the design variables
mostly affecting failure of the structures were coef-
ficient of friction between caisson and base ground
for sliding failure followed by the moment due to
self weight of caisson and buoyancy for overturn-
ingfailure, resistancemoment andtotal vertical force
for bearingcapacity failure. For thedesignvariables
relatedtoearthquake, theresultshowedthattheeffect
of dynamicwater pressurewassignificant.
Sensitivity analysis with respect to the reliability
index was performed. Althoughtheinertiaforceand
dynamic water pressurearethevariables affectedby
seismiccoefficient withhighCOV, thechangesof the
sensitivityarenot significant asthereliabilityindices
increase. This corresponds to the existing research
result for thebreakwater structure.
REFERENCES
EN. 1998. Eurocode 8: Design of structures for earthquake
resistance.
Nagao, T., Yoshinami, Y., Sanuki, T., & Kamon, M. 2001.
Applicationof reliabilitybaseddesigntoexternal stability
of caissontypequaywall, Structural engineering journal
47(A) (InJ apanese).
Yoneyama, H., Shiraishi, S. & Uwabe, T. 2000. A Study on
Load Factors of Seismic Loads on Limit State Design
MethodforPortandOffshoreStructuresinJ apan, Proc. of
8thASCE Specialty Conference on Probabilistic Mechan-
ics and Structural Reliability.
Yoon, Gil. 2005. Development of next generation port & har-
bour design (V). Ministryof maritimeaffairs& fisheries
(MOMAF) (InKorean).
139
Evaluatingthereliabilityof aleveeagainst seepageflow
Y. Shimizu&Y.Yoshinami
Fukken Co., Ltd., Hiroshima, Japan
M. Suzuki
T. Nakayama& H. Ichikawa
Hiroshima Institute of Technology, Hiroshima, Japan
ABSTRACT: Levees may collapseduringanabnormal downpour, andtherisk of floodinghas increasedin
recent years. To date, thestability of levees has beenevaluatedusingasafety factor by seepageflowanalysis
andthecircular arc method. Becausethesafety allowanceis estimatedfromthestability of theembankment
whichdepends ontheundergroundwater, it is necessary to quantitatively evaluatetheprobability of seepage
flowandtheuncertaintyof theresistanceof theembankment. Inthisstudy, theuncertaintyof thegeotechnical
propertiesof embankmentswasquantitativelyevaluated, andaprobabilitytheoryevaluationwasperformedfor
local failureandglobal failureassumingacircular slipsurface. Specifically, thesafetymarginwasevaluatedby
thefailureprobabilityand/or reliabilityindexbyexpressingthegeotechnical parametersusingtheprobabilistic
model andthestochastic finiteelement method. Theactionof seepageflowandtheriseof water level of the
river duetorainfall weredeterministicallycalculatedbytransient analysis, andbuoyancyandfluidpower were
estimatedineachtime, namely, thestabilityfor aparticular undergroundwater level. Finally, theresultsof the
proposedmethodarecomparedwiththoseof theconventional method.
1 INTRODUCTION
The risk of flood has increased these years due to
increaseof extremerainfall eventsincreasingtherisk
of river levee failure. The stability of levee against
seepagehasbeenassessedbasedonseepageflowanal-
ysis and safety factor calculated by the circular arc
method. However, sincetheallowanceinsafetywould
innaturebedeterminedbasedonthestabilityof thefill
againstgroundwater flowwithinthelevee, itcouldbe
requiredtounderstandtheprobability of theground-
water flowandtheuncertaintyinresistanceof thefill.
Inthis study, thesafety allowanceagainst thefull
scalefailureassumingitsslipcirclefailureisevaluated
bythefailureprobabilityandreliabilityindexobtained
bytheanalysisof stochasticfiniteelement methodby
meansof assessingquantitativelytheuncertaintiesof
thephysical valuesof leveefill. A comparisonof the
method proposed in this paper and theconventional
methodisalsodescribed.
2 ANALYTICAL METHODS
2.1 Overview
A stochastic finite element analysis was performed
by setting the circular arc at the failure mode using
theconventional circular arc method. A flowchart of
theanalysis is showninFigure1. Groundwater flow Figure1. Flowchart of theanalysis.
141
accompanyingrainfall andariseinthewater level of
the river was calculated by a transient seepage flow
analysis, inwhichtimehistorical buoyancy andfluid
power valueswereregardedtobeexternal forces.
First, atimesequential groundwater analysis was
executed. Next, a stochastic finite element analysis
was performed using the analytically determined
groundwater level and flow velocity vector at each
timepointasexternal forcesandconsideringtheuncer-
tainty of the ground parameters. The allowance in
safety against local failureand global failure, which
wasobtainedasaresultof theanalysis, wasevaluated
usingtheprobabilitytheory.
2.2 Seepage flow analysis
Thegroundwaterbehaviorwascalculatedbyperform-
ing transient saturated and unsaturated seepageflow
analysisthat canreproducetheactual phenomena.
A two dimensional expression of seepage flow is
shownbelow:
wherex =horizontal coordinate; z =vertical coordi-
nate; k =hydraulicconductivity(m/hr); =pressure
head(m); C =specificmoisturecapacity(l/m); a =1
for saturated zone and 0 for unsaturated zone;
S
s
=specificstorage(l/m); t =elapsedtime(hr)
The specific moisture capacity, C, is the tangent
inclinationof thewater retentioncurve. Thespecific
storages, S
s
, areS
s
=110
4
(l/m) for thesandysoil
andS
s
=110
3
(l/m) for thecohesivesoil.
2.3 Stochastic finite element method
Thefiniteelement methodhasbeenasuccessful ana-
lytical tool insolvingvariousgeotechnical engineering
problems, since the method can deal with complex
shapeandboundaryconditionof asystem, varioussoil
parameters, etc. However, thedeterministic FEM has
several limitations insystemmodeling; for example,
soil propertiesmustbedecidedasdeterministicuncor-
related values. To overcome such limitations, many
studieshavebeenmadeonthestochasticfiniteelement
method(SFEM) byapplyingtherandomtheory.
For local failure, two performancefunctions, both
withandwithout assumptionof potentiallymobilized
plane,wereconsideredtoevaluatetheprobabilitylevel
for slope failure. For global failure, another perfor-
mancefunctionwasconsideredonapotential failure
plane
2.3.1 Local failure
(a) Local failure without assumption of a potential
mobilized plane
Theperformancefunction for shear failurein the
i-thelement isdefinedasfollows:
where
fi
=distancebetween thecenter of theMohr
circle and the failure surface; c
i
=effective cohe-
sion; and
i
=effective internal friction angle.
1
and
12
=maximumandminimumprinciplestresses,
respectively.
(b) Local failure assuming a potential mobilized plane
Theperformancefunction for shear failureof the
i-thelement isdefinedasfollows:
where
i
and
I
=givenas theorthogonal stress and
shear stressonthepotential mobilizedplane.
(c) Reliability index and probability failure of the i-th
element
The reliability index i for local failure of i-th
element isdefinedasfollows:
whereE[g
i
] andVar[g
i
]=meanandvarianceof per-
formancefunction, respectively.
Accordingtothereliability index, probability fail-
uresof thei-thelement aregivenasfollows:
where+( )=standardnormal distributionfunction.
2.3.2 Global failure
Global failureof aslipcirclewas modeledby afail-
ureonapotential mobilizedplane. Theperformance
functionof thisfailurewasdefinedasfollows:
whereLli =i-th element of thelength of theplane,
andN is thenumber of elements across thepotential
mobilizedplane.
Thereliabilityindex for global failureisdefined
asfollows:
whereE[G] andVar[G]=meanandvarianceof per-
formancefunction, respectively.
Accordingtothereliability index, probability fail-
ureisgivenasfollows:
where+( )=standardnormal distributionfunction.
Figure2. Slipsurfacepassingthei-thelement.
142
2.4 Actions with groundwater
Stressesinleveecanberoughlyclassifiedintoeffec-
tivestress of theself weight andthestress causedby
seepageflow. Theunit weight of soil wasthustreated
assaturatedunitweightinsectionsbelowtheground-
water level and as wet unit weight in sections above
thegroundwater level.
Seepage force was assumed to act as an element
body force. Thebody forcewas determinedby mul-
tiplyingunit weight
w
of water bythehydraulicgra-
dient, whichisexpressedwithcoordinatecomponents
inEq. (9).
In the finite element method, Eq. (9) is shown in
triangular element as:
where L=area of triangular element, x
i
and
z
i
=coordinates of node i, and H
i
=total head of
nodei.
Figure3. Analytical model.
Table1. Soil parameters.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Wet weight 19.0 19.0 19.0 19.0 18.0 19.0 20.0 18.0 19.0 20.0 19.0 20.0 18.0 19.0 20.0 20.0 18.0
t (kN/m
3
)
Saturationweight 19.0 19.0 20.0 20.0 18.0 20.0 21.0 18.0 20.0 21.0 20.0 21.0 18.0 20.0 21.0 21.0 18.0
sat (kN/m
3
)
Cohesion 25.0 17.0 20.0 0.0 30.0 94.0 0.0 0.0 94.0 0.0 94.0 0.0 50.0 0.0 0.0 0.0 30.0
c(kN/m
2
)
Internal friction 0.0 22.0 30.0 35.0 0.0 15.0 35.0 50.0 15.0 35.0 15.0 35.0 0.0 30.0 35.0 40.0 0.0
angle
Permeability 1.0 1.0 3.0 2.0 2.0 3.0 2.0 1.0 5.0 1.0 3.0 1.0 2.0 3.0 5.0 1.0 1.0
coefficient k(cm/s) 10
3
10
5
10
3
10
1
10
5
10
3
10
5
10
2
10
3
10
1
10
4
10
5
10
1
10
3
10
3
10
6
10
1
3 ANALYTICAL CONDITIONS
3.1 Analytical model
Themodel usedintheanalysiswaspreparedbymodel-
inganactual leveeandisshowninFigure3. Elements
of thefiniteelement methodarestraintriangular ele-
ments, and the number of elements are 4,991. Soil
parametersareshowninTable1.Theparametervalues
inthetable(unitweight, cohesionc, internal friction
angle,andhydraulicconductivityk)areall meanval-
ues. Thecoefficientsof variationweredeterminedas
showninTable2. (Matsuo, 1984).
3.2 External force conditions
Thetimehistorical conditionsof external forces(rain-
fall and water level of the river) used in seepage
Table2. Coefficient of variation.
Coefficient of variation
Unit weight () 0.020.08
Cohesion(c) 0.20.4
Internal frictionangle() 0.10.2
143
Figure4. External forcemodel.
analysis are shown in Figure 4. (J apan Institute of
ConstructionEngineering, 2002).
Thewaterlevel of theriverwasassumedtoincrease
fromthelowwater level tothemaximumwater level
in about 100 hours, stay at thelevel for 1 hour, and
suddenlydroptotheinitial lowwaterlevel in15hours.
AlthoughtheGuidelinementionstouseaprelimi-
naryrainfall of 1mm/huntil themaximumwaterlevel
andthenarainfall intensityof 10mm/hruntil thewater
level startstosuddenlydrop, therainfall intensitywas
decidedtobe1mm/hr until thesuddendropinwater
level in this study, as amajor objectiveof thestudy
was to investigatethetimehistorical changes in the
water level intheleveeaccompanyingthechangesin
thewater level of theriver.
3.3 Circular arc
Theglobal failurewas investigated by assuming the
circular arc at the smallest safety factor that can be
determined by the circular arc method. The circular
arc of theback slopewasanalyzedsincetheanalysis
mainly aimed to investigate the stability at the high
water level.
4 ANALYTICAL RESULTS
4.1 Underground water level in the river levee
Thewater level distributioninthefill at anarbitrary
timepoint, whichwasdeterminedbytheseepageanal-
ysis, is showninFigure5. Thefigureshows that the
water level intheleveechangedalongwithchangesin
thewater level of theriver.
4.2 Central safety factor and reliability index
Theanalyzedglobal failureonaslipsurfaceatanarbi-
trarytimepoint isshowninFigure6. Thefigurealso
showsthesafetyfactor Fsdeterminedusingthecircu-
lararcmethod,thecentral safetyfactor (ratiobetween
the mean shear strength and the mean shear stress),
and the reliability index . Flow velocity vectors
determinedbytheseepageanalysisarealsoshown.
Figure5. Distributionof undergroundwater level inlevee.
Figure 6. Results of SFEM analysis and flow velocity
distribution.
Thefollowingconclusions werededucedfromthe
figure:
(1) WhenFsand werecompared, Fswaslarger than
. The trend was more notable when the water
144
Figure 7. Comparison between central safety factor and
reliabilityindex.
level of the river was rising than when it was
dropping. Thiswasattributabletothedifferences
between the circular arc method and SFEM, in
which seepage force acted as an element body
force.
(2) When and were time-historically compared,
unlikeFsand, and showedthesamebehavior
trendregardlessof thewater level of theriver.
To understand the states described in (2) to the
detail, time-sequential changesin and areplotted
inFigure7.Theplotshowsthattheybehavedsimilarly
andwerecloselycorrelated. Thus, thesafetyof levee,
whichchangesalongwiththewater level of theriver,
etc., couldbeassessedusing.
4.3 Local and global failures
A contour diagramof thedistributionof local failure
probability at thetimepoint describedinSection4.2
is shown in Figure8. Thefigureshows that theslip
surfaceof theglobal failurepassednear theboundary
of thesectionof largelocal failureprobability.
4.4 Effects of design factors on reliability index
Tounderstandtheeffectson bythecohesion, inter-
nal frictionangleandunitweight, whicharethedesign
factors, the time sequential changes of were cal-
culatedusingthecoefficients of variationof thesoil
parametersshowninTable1.Thecalculationinvolved
usingthemaximum, meanandminimumvaluesof the
coefficient of aparameter andusingthemeanfor the
coefficientsof theother twoparameters.
Theresultsof thecalculationareshowninFigure9,
fromwhichthefollowingconclusionswerededuced:
(1) Theeffectson by thecoefficientsof variations
of the soil parameters were relatively small in
unit weight andcohesionandlargeintheinternal
frictionangle.
(2) This was likely becausethecoefficient of varia-
tionsof unitweightwassmall (i.e., theunitweigh
wasrelativelyconstant). Thetrendsof theeffects
bycohesionandinternal frictionangledifferences
Figure8. Contour diagramof local failureprobability.
were likely because the ground of value was
dominant wheretheslipsurfacepassed.
The calculation showed that unit weight, which
fluctuates littlein most fills, had small effects on .
The effects by cohesion and internal friction angle,
whose coefficients of variations are relatively large,
werefoundto dependonthegroundconstant that is
dominant inthezoneof analysis.
5 SUMMARY
Thisstudygavethefollowingresults:
(1) A comparisonbetweenthetimehistorical behav-
iorsof andshowedsimilarbehaviorsregardless
of changes in thewater level of theriver, unlike
betweenFsand.Thus, thesafetyof levee, which
145
Figure9. Effectsof thecoefficientsof variationsonrelia-
bilityindex.
changes along the water level of the river, etc.,
couldbeassessedusing.
(2) Froma contour diagramof local failure proba-
bility distribution, the slip surface of the global
failure was found to pass near the boundary of
zoneswheretheprobability of largelocal failure
waslarge.
(3) To investigate the effects on by variations of
cohesion, internal frictionangleandunit weight,
wascalculatedusingwidelyusedcoefficientsof
variationsof thesoil parameters. Theunitweight,
which is believed to fluctuate little, had small
effects on . The effects by cohesion and inter-
nal frictionangle, whosecoefficientsof variations
arerelatively large, werefoundto dependonthe
soil properties that is dominant in the zone of
analysis.
Theresultsshouldbeuseful forassessingthesafety
of river levee. The only remaining problem is the
uncertainty of permeability, which fluctuates largely
and is likely to greatly affect the results of seepage
analysis (time historical changes in the water level
withinlevee). Theauthorswill proceedinvestigations
byimprovingmethodsof seepageanalysis.
REFERENCES
Ishii, K. &Suzuki, M. 1987. Stochastic finite element method
for slope stability, Structural Safety, 4: 111129.
J apan Institute of Construction Engineering. 2002. Kasen
Teibou no Kouzou Kentou no Tebiki[Guideline for Investi-
gating the Structure of River Levee].Tokyo:J apanInstitute
of ConstructionEngineering.
Komada,H.&Kanazawa,K.1975.Analysis of unsteady seep-
age flow and stability of fill dams under rapid drawdown
of the water surface level of reservoirs, Proc. of J SCE,
J apan, 240: 5162.
Matsuo, M. 1984. Jiban Kougaku Sinraisei Sekkei no Rinen
to Jissai[Geotechnology Actually with the idea of the
reliability based design]: 6271. Tokyo: Gihodoshuppan.
Nagase, M. Shirai, K. Segawa, A. & Hukunari, K. 2007.
KasenTeibougaku[Embankment study]. Tokyo: Sankaido.
Nagao,T.Yoshinami,Y. Mukai, M. &Shimizu,Y. 2000. Eval-
uation of Safety against Foundation Failure for Breakwa-
ters with Probabilistic Method, Proc. of JCOSSAR2000,
Japan, 479486.
Suzuki, M. &Ishii, K. 1986. Probabilistic optimumdesign of
drainage pipes on slope stability, Proc. of JSCE, Japan,
370(III-5): 209216.
2007. Kouwan no Shisetsu no Gijyutsujyou no Kijyun
Dou Kaisetsu[Technical basis and explanation of facil-
ities in harbors]. Tokyo: The J apan Port and Harbour
Association.
146
Determinationof partial factorsfor theverificationof thebearingcapacity
of shallowfoundationsunder openchannels
A. Murakami & S. Nishimura
Okayama University, Japan
M. Suzuki
Shimizu Co., Japan
M. Mori
National Agriculture and Food Research Organization, Japan
T. Kurata&T. Fujimura
NTC Consultants, Japan
ABSTRACT: Thelimit statedesignhas beenintroducedinto thedesigncriteriafor geotechnical structures.
Thecurrent paper attempts to apply thereliability-baseddesignmethodat Level II to thebearingcapacity of
thefoundationsof openchannelsfromtheviewpointof thelimitstatedesign.Toexaminetheapplicabilityof the
proposedprocedurefor practical structures, thereliability index iscomputedfor evaluatingthestability of the
foundationsof existingopenchannelsdesignedbytheusual method. It hasbeenproventhat theexistingdesign
methodisonthesafetysidebyobtaininglarger valuesfor thereliabilityindextobecomputedas5.0for sandy
soil and3.0for clayeysoil.
1 INTRODUCTION
Theformationof theWorldTradeOrganization(WTO)
and thesubsequent adoption of theTechnical Barri-
ers to Trade (TBT) Agreement placed an obligation
ontheInternational Organizationfor Standardization
(ISO) to ensure that international standards would
be globally relevant. For this purpose, the struc-
tural designcodefor agricultural facilitiesisrequired
to fulfill international standards such as ISO2394
(J SIDRE 2008). The use of a performance-based
design approach is widely recognized as supporting
thedevelopment of globallyrelevant ISOstandards.
The J GS published the J apanese Geotechnical
Standard, i.e. J GS4001-2004, entitledPrinciples for
Foundation Designs Grounded on a Performance-
based Design Concept (J GS 2004) to introduce a
performance-based design for foundation structures,
and developed adesign codefor geotechnical struc-
tures. Inthis standard, thelimit statedesignis intro-
ducedintotheir designcriteria.
Thecurrent paper attemptstoapply thefirst order
reliabilitymethod(FORM) tothebearingcapacityof
the shallow foundations of open channels fromthe
viewpoint of thelimit statedesigninorder to follow
theperformance-baseddesignframework. As is well
known, thereliability analysis can beclassified into
threedifferent levels, and theapproach described in
this paper belongs to thecategory at Level II which
considers first and second order statistical moments
of aparameter, e.g., theaverageandthevariance.
Several studies have been conducted for the reli-
ability analysis of the bearing capacity of shallow
foundations. For example, Babu et al. (2006) car-
ried out areliability analysis of afoundation resting
on cohesive soil based on Prandtls solution. Larkin
(2006) proposed a reliability analysis for founda-
tionssubjectedtomultidirectional seismicloadingby
considering the shear strength and the probabilistic
distributionof groundacceleration. FORM isapplied
to the bearing capacity of strip footings under the
assumptionthattheshearstrengthisvariable(Masshih
et al. 2008). Griffith et al. (2001, 2002) and Fenton
and Griffith (2003) carried out aprobabilistic study
onthebearingcapacity of aroughrigidstripfooting
onaweightlesscohesivesoil toassesstheinfluenceof
randomlydistributedundrainedshear strength.
Inthispaper, amodifiedTerzaghisbearingcapac-
ity formulais employedas anevaluationmethodfor
the bearing capacity. The shear strength parameters,
cohesionc andinternal frictionangle, andsoil den-
sity aredealtwithasprobabilisticparameters, while
theloaddueto self weights of theconcretestructure
of thechannel andtheinsidewater, is consideredas
astaticanddeterministicparameter, sincetheimpact
of theloadis relatively small for thedesign of open
channels. Thestatistical momentsof thesoil parame-
ters aredeterminedfromthepublisheddatarecords,
147
and thedatabaseof thesoil test results is organized
for thedesignof agricultural infrastructuresinJ apan.
It has beenproventhat theexistingdesignmethodis
onthesafety side, sincethecomputedvalues for the
reliabilityindexarelarger than5.0for sandysoil and
3.0for clayeysoil.
The partial factors to satisfy the target reliability
indicesarethendeterminedforexistingopenchannels.
Thedeterminedpartial factorsfor cohesionc, internal
friction angle, and soil density areaveraged for
sixteencases.Asthetargetreliabilityindices, =2.0,
3.0, and4.0arehereinadopted. Thedeterminedpar-
tial factorsarecalibratedwiththere-calculationof the
reliabilityindicesforthesixteenchannels,andverified
tobeappropriatefor thenewdesigncodefor shallow
foundationsunder openchannels.
Theremainder of thepaper isorganizedasfollows.
Inthenext section, wewill brieflyreviewthereliabil-
ityanalysismethodfor thedirect foundationsof open
channels. In Section 3, the reliability index is com-
putedfor evaluatingthestabilityof thefoundationsof
sixteenexistingopenchannels designedby theusual
methodto examinetheapplicability of theproposed
proceduretopractical structures. Thepaper endswith
asummaryof our mainconclusions.
Most modern bearing capacity predictions involvea
relationshipof theform(Terzaghi 1943)
where c: cohesion of the soil below the foundation
(kPa),
1
: unitweightof thesoil belowthefoundation
(kN/m
3
),
2
: unit weight of the soil in the embed-
ment portion (kN/m
3
), N
c
, N
, and N
q
: coefficients
of thebearingcapacity, D
f
: theembedment depthof
thefoundation(m), B: thelengthof thefoundations
shorter side(m), : correctionfactor duetothescale
effect of thefoundation.
Theperformancefunctionisdefinedusingthefollow-
ingequations.
whereq
max
: themaximumloadduetoself weightsof
the concrete structure of the channel and the inside
water anddeterministicvariable, c, =tan,
1
, and
2
: probabilisticvariables.
The Taylor series expansion of the performance
function at design points, e.g., c
1
, and
2
, is
obtainedas
Four probabilistic variables are normalized as
definedinthefollowingequationandhaveanormal
distributionof N(0,1) whenc, ,
1
, and
2
followthe
normal distribution:
The following definitions are also used at design
points(c
1
, and
2
):
wherej
c
, j
, j
1
andj
2
aretheaveragesforc, ,
1
,
and
2
,
c
,
1
, and
2
arethestandarddeviations
for c, ,
1
, and
2
.
Thederivativeof theperformancefunctionshould
satisfythefollowingequationfromequation(11):
148
The expected value of the performance function
is approximated with equation (14) derived from
equation(6) as
Thestandarddeviationof theperformancefunctionis
writteninequation(15) as
The reliability index is computed using the average
andthestandarddeviationfromequation(16).
where
c
,
,
1
, and
2
arecalled thesensitivity
andaredefinedasfollows:
3 RELIABILITY ANALYSISFORTHE
FOUNDATIONOF OPENCHANNELS
3.1 Statistics of parameters
The statistical values of soil parameters, e.g., c,
=tan,
1
, and
2
areobtained by collecting data
fromthereferences, for example, Matsuo(1984) and
J GS(1988).Table1liststheaverage,thestandarddevi-
ation, andthecoefficientof variationobtainedthrough
astatistical analysisof thedata.
FromTable1, thefollowing coefficients of varia-
tionfor different variablesareadoptedfor subsequent
analyses.
Unit weight,
1
and
2
: 0.06(
=0.055)
Coefficient of friction, =tan: 0.15(
=0.153)
Cohesion, c: 0.30(
=0.302)
Maximumbearingstress, q
max
, is treatedas being
static and deterministic. Sincetheexpected valueof
q
max
isrelativelysmall inthisproblem, comparedwith
theq
u
value, thevariability of this quantity does not
significantlyaffect theresultsof thecomputation.
3.2 Dimensions of open channels to be analyzed
Table2liststhedimensionsof sixteenopenchannels
andtheaveragestrengthparametersandload. Figure1
Table1. Statistical valuesof thesoil parameters.
Standard Coefficient of
Parameter Average deviation variation
c (kPa) 25.0 7.35 0.302
tan 0.65 0.10 0.153
1
,
2
(kN/m
3
) 16.9 0.98 0.055
Table2. Profilesof theopenchannels.
WidthB and Strength
height H of Unit parameter Load
openchannels Soil weight c (kPa), q
max
# (m) type* (kN/m
3
) (
) (kPa)
1 B=2.96,H=1.62 S 19.8 =35 22
2 B=3.00,H=1.80 S 18.0 =23 27
3 B=2.32,H=1.20 C 14.0 c =13 21
4 B=1.70,H=0.90 C 19.8 c =18 19
5 B=1.90,H=1.65 S 20.0 =25 25
6 B=2.00,H=1.70 S 20.0 =25 33
7 B=2.00,H=1.20 S 20.0 =25 13
8 B=4.50,H=1.80 S 19.8 =23 22
9 B=2.85,H=1.65 S 18.8 =29 24
10 B=7.90,H=2.48 S 20.0 =25 37
11 B=2.80,H=1.00 S 20.0 =25 15
12 B=2.00,H=2.20 S 20.0 =30 21
13 B=3.30,H=1.80 S 18.0 =15 23
14 B=3.30,H=1.80 C 15.0 c =8.0 23
15 B=3.40,H=1.00 S 20.0 =20 16
16 B=2.20,H=1.20 S 20.0 =20 16
*S: Sand, C: Clay
149
Figure1. Exampleof anopenchannel.
Table 3. Reliability indices and sensitivities of the soil
parameters.
Sensitivityfor parameters
Soil
# type* c =tan
1

2
1 11.0 S 0.985 0.000 0.171
2 6.1 S 0.963 0.003 0.268
3 3.2 C 0.997 0.000 0.071
4 3.2 C 0.999 0.000 0.041
5 10.0 S 0.963 0.000 0.268
6 5.9 S 0.975 0.003 0.224
7 12.7 S 0.956 0.000 0.294
8 5.2 S 0.988 0.016 0.154
9 17.0 S 0.986 0.000 0.167
10 7.5 S 0.968 0.000 0.251
11 9.5 S 0.963 0.000 0.271
12 17.6 S 0.961 0.000 0.275
13 6.5 S 0.887 0.003 0.462
14 2.9 C 0.990 0.000 0.140
15 6.3 S 0.957 0.003 0.290
16 14.0 S 0.838 0.000 0.545
*S: Sand, C: Clay
showsadefinitionof thevariablesusedintheequation
for thebearingcapacity throughCase1. Thesixteen
casescoverthirteencasesforsandysoil underthecon-
ditionof c =0andthreecasesforclayeysoil underthe
condition of =0. The unit weight considering the
buoyancy forcebelowthewater level is adopted for
thecomputation.
3.3 Reliability analysis and discussion
It isrevealedthat thereliabilityindicesfor sandysoil
arebetween5.2and17.6withsignificant variability
fromTable3inwhichthereliability indicesandsen-
sitivities for eachgroundparameter arelistedfor the
sixteencasestudies. Thereasonsareasfollows.
SinceparametersN
q
andN
g
inequation(1) arevery
sensitive to the internal friction angle and the max-
imumload q
max
is relatively small compared to the
valueof q
u
intheproblemsof theopenchannels, the
valueof theperformancefunctionbecomesextremely
large easily. In the cases of the clayey soil, friction
angle is zero andthevalueof q
u
has alinear rela-
tionshipwithcohesionc, namely, thereisnoextreme
change in bearing capacity q
u
for the change in c.
Consequently, thereliability indices arevery similar
amongthreecases.
The sensitivity of the internal friction angle is
dominant for thesandy grounds, andcohesionc has
dominant sensitivity for theclayey grounds. Theunit
weights,
1
and
2
, havesmall sensitivities. Sinceunit
weight
1
isusuallytreatedasasubmergedunitweight,
thesensitivity to thebearingcapacity is smaller than
for unit weight
2
.
Consequently,thereliabilityindicesaregreaterthan
5.0 for the sandy grounds. Although the reliability
indices for theclayey grounds arealmost 3.0, which
soundslikeasmall value, thecorrespondingprobabil-
ity of failureis 0.1%, andthus, thestructures onthe
groundaresufficientlysafe.
4 DETERMINATIONOF PARTIAL FACTORS
FORTHE FOUNDATIONSOF OPEN
CHANNELS
4.1 Determination of partial factors
ThefollowingmethodsarelistedinISO2394:
(1) Cases where the ground parameters follow a
normal distribution
Whenprobabilistic variables for thegroundparame-
ters followanormal distributionandtheir character-
isticvalueistheaverage, partial factor isdefinedas
follows:
where, f
k
: thecharacteristic valuefor theparameter,
usually, f
k
=j, f
d
: thedesignvaluefor theparameter,
for thereliability analysis, V: coefficient of thevari-
ationintheparameter, : sensitivityof theparameter,
and
t
: target reliabilityindex.
(2) Cases where the ground parameters follow a
logarithmicnormal distribution
When probabilistic variables follow a normal loga-
rithmic distribution, partial factor is writtenas the
followingequation:
where, : the average normal logarithm for prob-
abilistic variables, =ln(j,
1+V
2
), : standard
deviation of the normal logarithm for probabilistic
variables, =
ln(1+V
2
), andj: theaverageof the
probabilisticvariables.
4.2 Calibration of the partial factors
A series of partial factors, , is defined for each of
the four parameters, and m sets of partial factors,
i
(i =1,2,,m) are prepared. For each of the six-
teen cases listed in Table 2, the reliability indices
150
ij
=
ij
(
j
) (i =1, 2, . . . 16) are calculated with par-
tial factor
j
, ( j =1, 2, . . . , m). Consequently, 16m
reliability indices are obtained and the summations
of thesquareddeviationfor
ij
for thetarget reliabil-
ityindex,
t
, D
j
( j =1, 2, . . . , m) arecomputedinthe
manner of thefollowingequation:
Theoptimumpartial factorsareselectedfor themini-
mumD among D
j
( j =1, 2, . . . , m), so that thecal-
culated reliability index mostly approaches target
reliabilityindex
t
.
In the reliability analysis in this section, internal
frictionangle followsthenormal distribution, while
cohesionc isassumedtodistributelog-normally, since
thecoefficient of variationinc hasthegreat valueof
0.3, andthepartial factor cannot bedefinedfor the
valueof thetarget reliabilityindex,
t
=4.0.
4.3 Performance function for calibration
analysis
Incurrentdesigncodeof theopenchannels(Ministry
of Agriculture, ForestryandFisheries. 2001), follow-
ingequationisemployedtocheck thestability of the
foundations, inwhichthesafety factor of 3.0is con-
sideredfor thebearingcapacity. Thesafetyfactor has
norelationshipwiththeloadq
max
.
Sincethemaximumloadq
max
valuesarerelatively
small as seen in Table 2, the calculated reliability
indices have great values as shown in Table 3. The
values of q
max
, however, are different for each site,
and therefore, the actual q
max
values are not used
for the determination of the partial factors. As a
performancefunction, Equation (22) is employed in
followingsectionsinsteadof Equation(5).
whereq
d
isthedesignbearingcapacity,andadjustedso
thatthecomputedreliabilityindexbasedontheEqua-
tion (22) exactly coincides with thetarget reliability
indexinthecalibrationanalysis.
4.4 Computation of partial factors by
design values
Thepartial factors for eachcaselistedinTable3are
computed for thetarget reliability indices of
t
=2,
3, and 4 based on equations (18) and (19). Among
the ground parameters, the partial factors for unit
weight and coefficient of internal friction angle
tan arecomputedbasedonequation(18) by adopt-
ing thecoefficients of thevariation of 0.06 for both
1
and
2
, and 0.15 for tan under the assumption
thattheir probabilityfollowsanormal distribution.As
Table4. Expectedvaluesof partial factorsforsixteencases.
Expectedvaluesof partial factor,
Target Sandysoil Clayeysoil
reliability
index =tan
1

2
c
1

2
t
=2 1.42 1.01 1.02 1.86 1.00 1.02
t
=3 1.79 1.01 1.03 2.46 1.00 1.04
t
=4 2.42 1.01 1.05 3.22 1.00 1.07
Table5. Standard deviations of partial factors for sixteen
cases.
Standarddeviationof partial factor,
Target Sandysoil Clayeysoil
reliability
index =tan
1

2
c
1

2
t
=2 0.009 0.005 0.009 0.021 0.000 0.010
t
=3 0.023 0.004 0.013 0.057 0.000 0.020
t
=4 0.059 0.005 0.018 0.144 0.000 0.040
for cohesion, differentpartial factorsfor eachcaseare
evaluatedby equation(19) dependingoncharacteris-
tic values, becausecohesion c follows alogarithmic
normal distribution.
Theexpectedvalues andthestandarddeviationof
thepartial factors for thetarget reliability indices of
t
=2,3,and4arelistedinTables4and5,respectively.
Thefollowingitemscanbepointedout fromthese
tables.
1) Theobtainedpartial factorsof unit weights
1
and
2
arecomparativelysmall between1.00and1.07.
2) Thestandarddeviationof thepartial factorsisquite
small for every case. Thisleadstotheadoptionof
the expected value of the partial factors for each
case.
4.5 Examination of partial factors by calibration
Inorder to examineanoptimal set of partial factors,
based on their expected values evaluated in thepre-
vious subsection, trial partial factors areproposedas
multiples of 0.05 to cover theexpected valueof the
partial factors appearinginTable4. Table6shows a
setof partial factorsforeachtargetreliabilityindex,
t
.
The calibration has been made based on the set
of partial factors in Table 6 by equation (20), and
theresultant deviationof thereliability index ineach
case for the target reliability index,
t
, is obtained.
Asaresult of suchacalibration, Table7listsanopti-
mal set of partial factors to minimizethesumof the
squareof thedeviationsineachcasefordifferenttarget
reliabilityindices.
Table8 presents theaveragesafety factors for 16
cases. For theclayey grounds, thesafety factors are
151
Table6(a). Set of partial factorsfor thecaseof
t
=2.
Average Examined
Parameters factors factors
Sandy Unit weight
1
,
2
1.01,1.02 1.00, 1.05
soil Internal =tan 1.42 1.40, 1.45
friction
angle
Clayey Unit weight
1
,
2
1.00,1.02 1.00, 1.05
soil Cohesion c 1.86 1.85, 1.90
Table6(b). Set of partial factorsfor thecaseof
t
=3.
Average Examined
Sandy Unit weight
1
,
2
1.01,1.03 1.00, 1.05
soil Internal tan 1.79 1.75, 1.80
friction
angle
Clayey Unit weight
1
,
2
1.00,1.04 1.00, 1.05
soil Cohesion c 2.46 2.45, 2.50
Table6(c). Set of partial factorsfor thecaseof
t
=4.
Average Examined
Sandy Unit weight
1
,
2
1.00,1.05 1.00, 1.05
soil Internal tan 2.42 2.40, 2.45,2.50
friction
angle
Clayey Unit weight
1
,
2
1.00,1.07 1.00, 1.05, 1.10
soil Cohesion c 3.22 3.20, 3.25,3.30
Table7(a). Optimumset of partial factors for thecaseof
t
=2.
Parameters Sandysoil Clayeysoil
Unit weight
1
,
2
1.05 1.05
Internal friction tan 1.40
angle
Cohesion c 1.85
Table7(b). Optimumset of partial factors for thecaseof
t
=3.
Unit weight
1
,
2
1.05 1.05
angle
Cohesion c 2.50
Table7(c). Optimumset of partial factors for thecaseof
t
=4.
Unit weight
1
,
2
1.05 1.10
angle
Cohesion c 3.30
Table8. Averagesafetyfactors.
Target Sand Clay
t
=2 2.3 1.6
t
=3 3.5 1.9
t
=4 5.3 2.3
*Safetyfactor=q
u
,q
d
smaller thantheconventional factor of safety, 3.0for
all targetreliabilityindices, whileforthesandyground
the safety factors are greater than 3.0 for the target
reliability index
t
=3and4. Theseresults indicates
that the conventional safety factor 3.0 creates safer
sidedesignfor theclayey groundsthanfor thesandy
grounds.
5 CONCLUSIONS
This paper has evaluated reliability indices for the
foundationsof existingopenchannelstoexaminethe
safety of the current design method for the bear-
ing capacity and theeffect of theuncertainty of the
ground parameters. It is revealed that the current
designmethodissafeforthebearingcapacityof foun-
dations withareliability index over 3.0, andthat the
internal friction angle and the cohesion dominantly
affect thesafety.
Thereliabilityindexobtainedforsandisover5.0; it
cannotcontrol thesafetymarginbecauseof itsdepen-
denceontheunderlyinglayersandtheindex for clay
isabout3.0smaller thanthatfor sand. Thepartial fac-
torsarethenobtainedbyconsideringthesensitivityof
thevariability inthegroundparametersandthelimit
bearingcapacity.
Theresultsobtainedinthecurrentpaperarelimited
to shallowfoundations under open channels, but the
evaluation of the reliability indices shown herein is
effectivefor anytypeof structure.
ACKNOWLEDGEMENT
The authors are grateful to Mrs. Hitoshi Yano and
TaroSetoof J apaneseMinistryofAgriculture,Forestry
andFisheries, anditsTokai Branchfor their helpfor
collectingthedataandprovidingsomearrangements
andsuggestions. Themembersof performance-based
designcommitteeof theJ apaneseInstituteof Irriga-
tion and Drainage are also acknowledged for their
discussion.
152
REFERENCES
Babu, G.L.S., Srivastava, A. & Murthy, D.S.N. 2006. Reli-
ability analysis of the bearing capacity of a shallow
foundationrestingoncohesivesoil, Can. Geotech. J. 43:
217223.
Fenton, G.A. & Griffiths, D.V. 2003. Bearingcapacity pre-
dictionof spatiallyrandomc- soils, Can. Geotech. J. 40:
5465.
Griffiths, D.V. &Fenton, G.A. 2001. Bearingcapacityof spa-
tially randomsoil: The undrained clay Prandtl problem
revisited, Geotechnique 51(4): 351359.
Griffiths, D.V., Fenton, G.A. & Manoharan, N. 2002. Bear-
ingcapacityof roughrigidstripfootingoncohesivesoil:
Probabilisticstudy, J. Geotech. Geoenviron. Eng. 128(9):
743755.
J apaneseGeotechnical Society. 1988.Variability of Soil Data
and Design (inJ apanese).
J apanese Geotechnical Society. 2004. J GS4001-2004:
Principles for Foundation Designs Grounded on a
Performance-Based Design Concept.
J apanSocietyof Irrigation, DrainageandReclamationEngi-
neering (J SIDRE). 2008. Introduction to Performance-
Based Design for Functional Maintenance of Agricultural
Facilities, J SIDRE (inJ apanese).
Larkin, T. 2006. Reliability of shallow foundations sub-
jected to multidirectional seismic loading, J. Geotech.
Geoenviron. Eng. 132(6): 685693.
Massih, D.S., Soubra, A.H. & Low, B.K. 2008. Reliability-
basedanalysisanddesignof stripfootingsagainstbearing
capacity failure, J. Geotech. Geoenviron. Eng. 134(7):
917928.
Matsuo, M. 1984. Geotechnical Engineering Theory and
Practice of Reliability Theory, Gihodo(inJ apanese).
MinistryofAgriculture, ForestryandFisheries. 2001. Design
codefor landimprovement Channels.
153
Applicationof concept inGeo-code21 toearthstructures
M. Honda
Nikken Sekkei Civil Engineering Ltd., Tokyo, Japan
Y. Kikuchi
Port and Airport Research Institute, Yokosuka, Japan
Y. Honjo
ABSTRACT: Geo-code21 is a comprehensive foundation design code that can harmonize all the major
foundationcodesinJ apan. ThiscodewaspublishedbyJ GS(J apaneseGeotechnical Society) in2004andprin-
ciplesfor foundationdesignsaredescribedbasedonperformancebaseddesign(PBD) concept. Inorder toadd
somechaptersforearthstructuredesigntoGeo-code21, J GShassetupaworkinggroupunderStandardization
committeeof designandconstructioninStandardizingdepartmentsince2006. Inthispaper, outlineof thedraft
isreported.
1 INTRODUCTION
1.1 Background
Members of theworkinggrouparelistedinTable1.
TheworkinggrouphasbeensetupunderStandardiza-
tioncommitteeof designandconstructioninStandard-
izingdepartmentof J GSsince2006andhascompleted
thedraft of comprehensivedesigncodefor embank-
ments and cut slopes based on performance-based
designconcept.
The draft is now under consideration in J GS and
outlineof thecodeisreportedinthispaper.
Table1. Membersof theWorkingGroup.
(Chair)
Y. Honjo GifuUniversity
(SecretaryGeneral)
Y. Kikuchi Port andAirport ResearchInstitute
(Secretary)
M. Honda NikkenSekkei Civil EngineeringLtd.
(Members)
M. Hachiya ChuoFukkenConsultantsCo., Ltd.
S. Kato PublicWorksResearchInstitute
K. Kojima RailwayTechnical ResearchInstitute
A. Murakami OkayamaUniversity
K. Ohkubo NipponExpresswayResearch
InstituteCo., Ltd
M. Suzuki ShimizuCorporation
S. Tani National Institutefor Rural Engineering
A. Wakai GunmaUniversity
Y. Wakatuki FukkenCo., Ltd.
1.2 Framework of the code
Figure 1 is the hierarchy of code description of
Geocode21 and this is also applied to thecodefor
earthstructures.
Thecodeconsists of specificationof performance
requirements that achieve the objectives of struc-
tures, performancecriteriathat specifytheguidelines
to design specifically and verification of theperfor-
mance. Two types of the verification methods are
introduced.
Verification approach A is the fully performance
baseddesignapproachwheredesignersareonlygiven
the performance requirements of the structures; the
designersarerequestedtoverifytheir design, andthe
resultswouldbecheckedbyauthorizedorganizations
et al.
Approach A
Approach B
Specific Base
Design Code
Specific Design Code
Performance Criteria
Comprehensive Design Code
Performance
Requirements
P
e
r
f
o
r
m
a
n
c
e

R
e
q
u
i
r
e
m
e
n
t
s

Objective
Requirements
Code,Approach
V
e
r
i
f
i
c
a
t
i
o
n
Figure1. Hierarchyof codedescription.
155
Table2. Tableof contentsof Geo-code21.
0BASESOF STRUCTURAL DESIGN
1BASESOF FOUNDATIONL DESIGN
2GEOTECHNICAL INFORMATION
3DESIGNOF SHALLOWFOUNDATION
4DESIGNOF PILE FOUNDATION
5DESIGNOF COLUMNTYPE FOUNDATION
6DESIGNOF RETAININGSTRUCTURE
7DESIGNOF TEMPORARY STRUCTURE
(Sectionsof chapter 07areomittedhere)
8DESIGNOF EMBANKMENT
8.1Scope
8.2Objective
8.3PerformanceRequirements
8.4PerformanceCriteria
8.5Investigations
8.6Itemstobeconsideredindesign
8.7Predictingthebehavior of Embankment
8.8Verification
8.9Designreport
8.10Construction
9DESIGNOF CUT SLOPE
9.1Scope
9.2Objective
9.3PerformanceRequirements
9.4PerformanceCriteria
9.5Investigations
9.6Itemstobeconsideredindesign
9.7Predictingthebehavior of Cut slope
9.8Verification
9.9Designreport
9.10Construction
VerificationapproachB isaverificationprocedure
basedondesigncodes:Thesecodesmaybeestablished
for eachcategoryof structures(e.g. highwaybridges,
buildingsetc.) bytheauthoritieswhoareeither owner
or responsiblefor theadministrationandsafetyof the
categoryof structures.
Table of the contents of Geo-code21 is shown in
Table 2. The code has been developed as additional
code of Geocode21 (i.e. chapter 07). Design of
embankmentisdescribedinchapter 8andcutslopeis
describedinchapter9.Also,frameworkof thecontents
of theGeocode21 isappliedtoearthstructuredesign.
2 OUTLINESOF THE CODE
2.1 Scope
Scope of application is embankment and cut slope.
Thecodeis describedfor not only thestructures that
supportvertical load, e.g., roador railwayfill, butalso
thestructuresthatblockwater,e.g.,riverdike,dametc.
2.2 Objective
Inthebeginningof design, thenecessityof thestruc-
tureshall bespecified. Earthstructures havevarious
objectives. For example, theroador railway fills and
cutslopestogivefirmgroundtorunvehiclesor trains
safety, water storage dams that, through reservoirs,
stablewater supplyisachievedincaseof water short-
age. River diketo prevent thefloodto inlandincase
of flooding, thefloodcontrol dams or theregulating
reservoirs to prevent the breaking dikes by control-
lingtheoutflowincaseof heavyrain. Coastal leveeto
defendthecoastal areafromstormsurgeorTsunami.
2.3 Performance requirements
Performancethat isrequiredtoachievetheobjectives
shall bedescribed. In thecaseof complex structure,
whoseobjectiveisachievedbycombinationof differ-
enttypesof structureandtheearthstructurecomposes
apartof thecomplexstructure, e.g., road, railwayetc.,
theperformancerequirementshall bedescribedincon-
siderationwithother types of structure, e.g. bridges.
Thisisbecausethecollapseof apart of thestructure
maycausefailureof thewholecomplexstructure.
2.4 Performance criteria
Coursestoestimatetheperformanceshall bedescribed
specifically.
Degreeof achievement of theperformanceiseval-
uated by comparing the predicted behavior with the
limit state.
Limit state that divides intended condition and
unintendedconditionshall bespecifiedfor eachper-
formancerequirement.
Actions that are predicted to act on the earth
structure during design working life shall also be
determinedtoestimatethebehavior.
Inadditiontotheexternal loads, hydraulic actions
shouldalsobetakenintoaccountinthedesignbecause
soil material maycollapseunder hydraulicaction.
Infact,someearthstructurecanbeusedindefinitely
withappropriatemaintenance,e.g.,riverdikes,anditis
difficulttodeterminethedesignworkinglifefor earth
structuresspecifically. Thisisbecausesoil material is
different fromsteel or concrete in that deterioration
andcorrosiondoesnt occur.
In principle, the following three limit states shall
bespecified, althoughother alternatives arenot nec-
essarily excluded and not all limit states shall be
determined.
2.4.1 Serviceability limit state
thelimitstateinwhichtheregularusetoachievethe
objectiveof thestructureispossible, if thedamage
tothestructureoccurred.
2.4.2 Reparability limit state
thelimit stateinwhichthedamagetothestructure
hasoccurred, however regular useof thestructure
ispossibletoalimitedextent andtherearereason-
ableprospectforfull functionalityof thestructureif
economically-feasiblerepairsareperformedwithin
afeasibleperiodof time.
156
Figure2. Concept of thelimit statedesign(J SCE(2002)).
Figure3. Concept of thelimit statedesignfor earthstruc-
tures.
2.4.3 Ultimate limit state
thelimit stateinwhichthestructuremayhavesus-
tained considerabledamage, but not to theextent
thatthestructurehasreachedlargefailureor tothe
extent that would result in serious damageto the
nearbystructures.
Figure2showstheconcept of thelimit state. This
concept canbeappliedto thestructures that consists
of themembersof steel orreinforcedconcretebecause
eachlimit statecanbedeterminedby yieldpoints of
thematerial andphysical meaningof thelimit states
is clear. However it is difficult to apply this concept
toearthstructures directly becauseyieldpoint of the
material isnotclear. Besidessomeearthstructurescan
berepaired easily if thematerial has been deformed
afterpeakstrength.Thisisbecausesoil material shows
a certain strength after peak strength (see Figure 3)
andconstructionof earthstructures is comparatively
easier. Thereforeit isdifficult todistinguishbetween
reparabilitylimit stateandultimatelimit state.
2.5 Investigations
Prior to the design, a geotechnical investigation
and a survey of the surrounding conditions shall
Figure4. Concept of thefoundationdesign.
be performed. In the design of foundations, the
mechanical behavior of bearinglayer shouldbeesti-
mated. Also, interactionbetweenfoundationandsur-
roundinggroundshouldbeestimatedinthedesignof
pilefoundation(seeFigure4).
Generally, bearing layer is firm in the founda-
tion design and the behavior of soft ground is esti-
matedonly by theinteractionwithpile. Becausepile
foundation is chosen when bearing layer of shallow
foundationdoesnot haverequiredbearingcapacity.
However, inthedesignof embankment, thebehav-
ior of thesoft groundshouldbeestimatedasbearing
layer. Geotechnical investigationshouldbeconducted
inconsiderationwiththechangeinstressstatebecause
self weight of theembankment isheavyandit affects
thestressstateintheground. Besides, themechanical
behaviorof softgrounddependsoneffectiveconfining
pressurewhichvarieswithtimewhenthepermeability
of thegroundislow.
In addition to the bearing layer, mechanical
behavior of earth material should also beestimated.
Embankmentisconstructedbythecompactioncontrol
standardandthevalidityof thismethodhasbeencon-
firmedempirically. Compactioncharacteristicsof the
earthmaterial shouldbeinvestigatedandcompaction
control standardshouldalsobeexaminedintheinves-
tigation. When themechanical property is examined
by test, condition of the specimen should be deter-
mined in consideration with the compaction control
standardbecauseearthmaterial isinunsaturatedstate
ingeneral andmechanical behavior of earthmaterial
dependsonthedrydensityanwater content.
In the design of cut slopes, the behavior of the
slope should be estimated after the soil is removed.
Thusinvestigationshouldbeperformedtoexaminethe
mechanical behavior of thesoil material intheslope
after unloading. However inclinationof slopethatsat-
isfiesthesafetycanbedeterminedbytypeof thesoil,
height of the slope etc. for each category of struc-
turesandvalidity of thismethodhasbeenconfirmed
empirically.
157
Figure5. Conceptof theembankmentdesignandcutslope
design.
The foundation design and embankment design
are different. In the design of foundation and
embankment, thestructuresarecreatedby loadingto
theground, but inthedesignof cut slope, thespaceis
createdbyunloadingtheground(seeFigure4,5).
2.6 Items to be considered in design
Structureshall bedesignedbasedonthebehavior that
iscausedbypredictedactions.
The behavior that should be studied shall be
extracted as items to be considered in design. In
thedesignof earthstructure, stability of theground,
embankmentandslopeshall beexamined.Also,settle-
mentor deformationof thegroundshall beexamined.
Besides stability against the seepage action shall be
examinedinthedesignof watertight structures.
2.7 Predicting the behavior
Theglobal stability, deformationandseepagefailure
of earthstructuresshall bepredicted.
Circulararcmethodcanbeusedtostudythefailure
of the embankments and cut slopes. However incli-
nation of slope that satisfies the safety can also be
determinedbytypeof thesoil, heightof theslopeetc.
Seepageflow analysis or flow net can beused to
studytheseepagebehavior.
If theonedimensional behavior of groundis pre-
dicted, settlement can be predicted by one dimen-
sional consolidationtheory, e.g., Terzaghis model of
consolidation.
Howeverif twoorthreedimensional behaviorispre-
dictedor affecttonearbystructuresshouldbestudied,
numerical methods, e.g., finiteelement method, can
be used. However this method is developing and it
ismoreadvisablewhendeformationof thegroundis
observedafterconstructiontoconfirmtheaccuracyof
theprediction. Inthepredictionof dynamicbehavior,
thereferenceof theconditionof verifiedexampleis
veryimportant.
2.8 Verification
Theperformanceof thestructureshall beverifiedby
the comparison between predicted behavior and the
limitvaluethatisdeterminedinconsiderationwiththe
limit state. Someperformances of earthstructureare
not easytoverifydirectlyandother indirect verifying
method whose validity has been confirmed empiri-
callycanbeusedasdeemedtosatisfy. For example,
most of thedesign of cut slope, inclination of slope
doesnot determinedbycircular arcmethodbut deter-
minedbytypeof thesoil, height of theslopeetc. This
is becausethesoil material intheslopeis inunsatu-
ratedstateanditisdifficulttoevaluatethemechanical
property. Besides, thestability of theslopehas been
evaluated by the inclination of the slope based on
experiments indirectly becausethestability of slope
increaseswiththedecreaseintheinclinationof slope.
3 FINAL REMARKS
Outline of the comprehensive design code of earth
structureisreported. Thedraftisnowunder consider-
ationinJ GSandit will beapprovedandpublishedin
2009.
REFERENCES
J GS2004. Principles for Foundation Designs Grounded on a
Performance-based Design Concept (nickname Geocode
21)
J SCE2002.Basis of Structural Designfor Buildings andPub-
lic Works, Ministry of land, infrastructureandtrans-port:
http://www.mlit.go.jp/kisha/kisha02/13/131021/131021_
e.pdf
158
Limit statedesignexample Cut slopedesign
Wan-KwanLin& LiminZhang
Department of Civil Engineering, Hong Kong University of Science and Technology, Hong Kong
ABSTRACT: Thispaperpresentsthedesignof acutslopeusinglimitstatedesign(LSD), loadresistancefactor
design(LRFD), andtraditional allowablestressdesign(ASD). A soil cut slopewithamaximumheight of 15m
isconsidered.Theslopeinvolvesfourmajorsoil/rocklayersfromthegroundsurfacedownward: fill, completely
decomposedgranite(asiltysand),highlydecomposedgranite,andmoderatelydecomposedgranite.Theobjective
of designistofindasuitablecutslopeanglethatsatisfiesall designrequirements.InLSD,threedesignapproaches
(DA1, DA2, andDA3) totheapplicationof partial factorsfor actions, materials, andearthresistancesarealso
considered. Finally, thecut slopeanglesdeterminedusingthesedifferent designmethodsarecompared.
1 INTRODUCTION
Except for thedesignof retainingstructuresandrein-
forced fills, geotechnical design in Hong Kong is
primarilybasedontheAllowableStressDesign(ASD)
methodology, whichapplies aglobal factor of safety
toaccommodateuncertaintiesinbothresistancesand
loads. Many codes of practice in Hong Kong are
formedbasedontheBritishStandards.Recently,many
European countries, including the UK, adopted or
areinatransitionto theEurocodes, whicharebased
onLimit StateDesign(LSD). Besides Europe, many
other countriesincludingChina, USA, Canada, J apan
and Australia are either using or in a transition to
theLSDmethodology. Thedirectionof HongKongs
future geotechnical code developments is a subject
worthof studying.
Thetransitionto LSD, however, has met withcer-
tain difficulties. Onereason is that thedevelopment
of partial factors for LSD requires someprobability
knowledgeandstatistical data.Anotherimportantrea-
sonisthatthedesignproceduresof LSDarerelatively
newtopracticingengineers. Somewell-workeddesign
examplesmaybeuseful tohelpthetransition. There-
fore, theobjectiveof thispaper istoexploretheuseof
LSDmethodologyingeotechnical designintheHong
Kongcontext.Asslopestabilizationisoneof themost
important geotechnical elements in Hong Kong, cut
slopedesignexamplesaredemonstratedinthispaper
usingASD, LSD, andloadandresistancefactordesign
(LRFD). Asnecessarypartial factorsfor geotechnical
designsinHongKonghavenot beenfullydeveloped,
intheLSDexamples, therecommendedpartial factor
valuesinEurocode7areapplied.
2 PHILOSOPHY OF GEOTECHNICAL
DESIGNS
Thebasic criterionfor geotechnical designisthat the
resistance must be greater than the design loading
conditionwithasufficient safetymarginagainst fail-
ure of the structure or excessive settlement of the
structure. Factors of safety areusually used to com-
pensatefor uncertaintiesinsimplifieddesignmodels,
ground conditions, material strengths, construction
methods, anddistributionof loads. Becauseof these
uncertainties, the ultimate resistance of foundations
should be factored down, while the loads should be
factored up by certain amounts. Therequired safety
margin should take into account consequences of
failure and construction tolerances. There are three
commondesignmethods usedto ensureanadequate
margin of safety under consideration, namely ASD,
LSD, andLRFD.
2.1 Allowable Stress Design
The conventional design method for geotechnical
worksinHongKongisASD. A global factor of safety
isappliedtocompensatefor theuncertaintiesinboth
loadsandresistances:
where Q
i
=nominal load component; R
n
=nominal
resistance; FS =factor of safety. The design proce-
duresof ASDarerelativelystraightforward. However,
noconsiderationisgiventothefactthatdifferentloads
or differentdesignanalysismodelshavedifferentlev-
elsof uncertainty. Thefactor of safetydependsonthe
typeof problembutthesafetylevelsfordifferenttypes
of structuresmaynot beconsistent.
2.2 Limit state design
LSD considers the functional limits of strength and
serviceability of geotechnical works by considering
possibleultimatelimit states (ULS) and serviceabil-
ity limit states (SLS) explicitly. One of the most
159
comprehensive LSD frameworks has been imple-
mented in Eurocode 7, which considers risk and
complexityusinggeotechnical categoriesandcharac-
teristicvalues. LSDmayaccordmorelogicallywitha
performancebasedDesignApproach(DA). Itisbased
ontherequirement that theresistanceof ageotechni-
cal workshouldexceedtheloadeffectfor all potential
modesof failure, allowinguncertaintiesinloadeffects
andthevariability inresistanceandmaterial proper-
ties. Therefore, partial factors are applied to loads,
resistances, andmaterialsinLSDseparately:
whereR
d
=design resistanceobtained by using fac-
tored material properties; Q
i
=design load affect;
i
=partial factor for action;
R
=partial factor for
resistance. The partial factors are statistically-based
designparameters. Theseparameters canbeselected
to achieveaspecified target reliability level, so that
theuseof LSD canachieverelatively uniformlevels
of safetyfor variousgeotechnical works.
2.3 Load and resistance factor design
The AASHTO LRFD Specifications are based on
LoadandResistanceFactor Design(LRFD). LRFDis
anotherwidelyusedformatof LSD. InLRFD, strength
limit states, service limit states, and extreme event
limit states areintroduced as in LSD. Themain dif-
ferencebetween LRFD andLSD (as in Eurocode7)
istheuseof factors. LRFDappliesaresistancefactor
to thenominal resistancebut does not apply factors
directlytothematerial properties:
where
i
=load factor;
R
=resistance factor; R
n
=
nominal resistance obtained without factoring down
thesoil parameters.
Although LSD and LRFD involvesomeconcepts
of probability, no complex probability and statisti-
cal analyses are required when the load factors and
resistancefactorsprovidedindesigncodesareused.
2.4 Design approaches
In contrast to the checking of structural designs,
geotechnical actions and resistances of the ground
cannot beseparated. Geotechnical actionssometimes
dependonthegroundresistance, e.g. activeearthpres-
sure, and the ground resistance sometimes depends
on actions. For example, the bearing resistance of a
shallow foundation is afunction of theloads on the
foundation (Ovensen 2002). In order to account for
thisinteractingrelationship, EN1997-1proposesthree
DAs for checking against the failure in the ground
(GEO) andthestructure(STR) (Table1). Thechoice
of aDA is for national determination, andshouldbe
stated in theNational Annex. Different design prob-
lems may begoverned by different DAs. Thepartial
factors for theDAs aregrouped in sets and selected
accordingto theDA usedandverifiedfor structural
andgeotechnical limitstatesinpersistent andtransient
situations(Tables2to4). SetA1orsetA2of thepartial
factorsonactionsshouldbeappliedtothepermanent
and variableunfavourableor favourableactions. Set
R1, R2, R3or R4of thepartial factors onresistance
shouldbeappliedtothecorrespondingresistancecon-
ditions, such as base and shaft resistances. Set M1
or set M2 of the partial factors on soil parameters
should beapplied to thecorresponding soil strength
parameters, such as effective cohesion and friction
angle.
Table1. DesignApproaches recommended in EN1997-1.
(CEN2004).
Designexcept
Design axiallyloaded Axiallyloaded
Approach pilesandanchors pilesandanchors
DA1C1 A1+ M1+R1 A1+ M1+R1
DA1C2 A2+ M2+ R1 A2+ (M1or M2) + R4
DA2 A1+ M1+R2
DA3 (A1* orA2
#
) + M2+R3
where + implies to be combined with; *on structure
actions; #ongeotechnical actions
Table 2. Partial factors on actions (
F
) or the effects of
actions(
E
) inEN1997-1. (CEN2004)
Set
Action Symbol A1 A2
Permanent Unfavourable
G
1.35 1.0
Favourable 1.0 1.0
Variable Unfavourable
Q
1.5 1.3
Favourable 0 0
Table 3. Partial resistance factors (
R
) for slopes and
overall stabilityinEN1997-1. (CEN2004)
Set
Resistance Symbol R1 R2 R3
Earthresistance
R;e
1.0 1.1 1.0
Table 4. Partial factors for soil parameters (
M
) in EN
1997-1. (CEN2004)
Set
Soil parameter Symbol M1 M2
Angleof shearingresistance
1.0 1.25
Effectivecohesion
c
1.0 1.25
Weight density
1.0 1.0
160
3 DESIGNMETHODSFOR SLOPE STABILITY
SLOPE/W is computer programcommonly used to
analyzethestabilityof earthslopes(Geo-Slope,1991).
The slope stability is analyzed using the method of
slices. Thefactor of safety is defined as theratio of
availableshear strengthtotheshear stressrequiredto
maintainequilibrium. Thefactor of safety istakento
bethesameforall slices.Theproblemisnormallycon-
sideredintwodimensions, i.e. theconditionof plane
strain. SLOPE/W allows stability analysis using the
Ordinarymethod, Bishopssimplifiedmethod, J anbus
simplifiedmethod, theMorgenstern-Pricemethodetc.
A varietyof interslicesideforcefunctionscanbeused
inthecaseof theMorgensternandPricemethod. In
thisstudy,theMorgenstern-Pricemethodwasusedand
theintersliceforcefunctionwasassumedasahalf-sine
function.
4 DESIGNEXAMPLE CUT SLOPE
A soil cutslope, whichhasamaximumheightof about
15mandatotal lengthof about 60mmeasuredalong
theslopetoeis considered. Thecross section of the
slopeisshowninFigure1. Theslopeangleisinclined
at 61.4
to the horizontal. The groundwater level is

2mabovetherock head, representingascenariocor-
responding to a 1 in 10 year return period rainfall.
Residential buildingsarelocatednear theslopetoe.
Three layers of soils are present: fill, completely
decomposedgranite, andhighlydecomposedgranite.
The shear strength parameters of the three soil lay-
ers areshowninTable5. A stratumof moderately to
Figure1. Cross-sectionof theslope.
Table5. Soil Parameter of theslope.
Soil type
(kN/m
3
) c
(kPa)
)
Fill 16 0 33
CDG 19 5 40
HDG 19 8 42
slightlydecomposedgraniteisencounteredbelowthe
slopetoe.
Supposetheslopeis identifiedas not sufficiently
stableandneedstobecut.Thedesignproblemweface
nowistodeterminetheangleof slopecuttosatisfyall
designrequirements.
4.1 Design procedures
Theprocesses used for slopestability analysis using
LSDorLRFDaresimilartotheprocessusedforASD.
Thedesignproceduresof thethreedesignmethodsare
summarizedinFigure2.
Intheconventional design, thestabilityproblemis
definedby creatingtheprofileof theslope. Ananal-
ysis method, such as theMorgenstern-Pricemethod,
should be chosen. Soil properties, pore water pres-
sures, line loads are input respectively. Then, slip
surfacesof theslopearespecifiedtosolvetheproblem.
Theminimumfactor of safety andthecorresponding
slip surfacearedisplayed. Contours, thecritical slip
surface, free body diagrams can be obtained. If the
factor of safety is larger thanaspecifiedvalue, then
theslopeisconsideredtobesatisfactory.
The basic procedures for slope stability analysis
in LSD and LRFD arethesame, such as thedefini-
tionof problem, specificationof amethodof analysis
and sketching the problem. The soil parameters are
requiredtobefactoredinLSD, butnotinLRFD.Atthe
endof theanalysisinLSDor LFRD, theFSobtained
fromastability analysis actually represents theratio
of theresisting moment (or force) to themobilizing
moment (or force) (M
R
,M
D
), whichshouldbelarger
than the value of partial resistance factor for slope
stability(Table3).
4.2 Conventional design
GEO(2000)suggestsafactorof safetyof 1.4forslopes
thatmayhaveseriousconsequencestohumanlifeand
properties. Therefore, the computed factor of safety
fromSLOPE/Wshouldbelarger than1.4.
Theoriginal angleof theslopeinFigure1is61.4
.
Thestabilityanalysisresultsfor theexistingslopeare
showninFigure3andTable6. Most of thefactorsof
safetyof theslipsurfacesaresmaller than1.4, which
indicatestheslopeisnotsufficientlysafe.Assumethe
cutslopeangleis35.7
.Thecomputedfactorsof safety
areagainlistedinTable6andtheminimumfactor of
safetyis1.40.
4.3 Limit state design
AlthoughSLOPE/Wcalculatesthefactorof safety, the
programcanstill beusedtolimitstatedesignanalysis.
Theinputparameters, suchasloads, cohesion, friction
angle, shouldbefactoredbeforeanalysis. Thefactor
of safety obtained would be equal to M
R
,M
D
. This
ratioshouldbelarger thantherecommendedvalueof
partial resistancefactor for slopesandoverall stability
inEurocode7(Table3).
161
Figure2. Flowchart for slopestabilityanalysisbyASD, LSDandLRFD.
Figure3. Specifiedslipsurfacesbeforecuttingtheslope.
Theoriginal slopeangleis61.4
, thestabilityanal-
ysisfor theexistingslopeisshowninTable6. Almost
half of thevaluesof M
R
,M
D
aresmallerthanrequired,
which means the slope is not up to the standard.
However, thevalueof thecut slopeangleshould be
steeper as the differences between the ratio and the
partial resistancefactor arerelativelysmall compared
withtheconventional design.
4.3.1 Design Approach 1 Combination 1
Assumethecutslopeangleis41.5
. Thepartial factor
for friction angle and cohesion are 1.0. The partial
resistancefactor for slopesandoverall stabilityis1.0.
However, thevalueof partial resistancefactor of DA2
is1.1, whichislarger thanthat of DA1C1. Therefore,
theanalysisusingthesetwoDAsfor thecut slopecan
becombined.
4.3.2 Design Approach 1 Combination 2
Assumethecut slopeangleis 41.5
. Thepartial fac-
torsfor frictionangleandcohesionareboth1.25. The
minimumvalueof M
R
,M
D
forall slipsurfacesislarger
than1.0. Thepartial factors of DA3areequal to the
partial factorsof DA1C2.Therefore,therequiredangle
of slopeisthesame.
162
Table6. Factorsof safetyor M
R
/M
D
computedusingSLOPE/W.
M
R
/M
D
(LSD)
Factor of safety(ASD) Beforecuttingslope After cuttingslope
Slipsurface Beforecutting After cutting DA1C1, DA1C2, DA1C1, DA1C2,
No. slope slope DA2 DA3 DA2 DA3
1 1.00 2.57 1.00 0.80 1.81 2.27
2 1.31 1.77 1.31 1.05 1.29 1.62
3 1.34 1.40 1.34 1.08 1.06 1.33
4 0.83 2.77 0.83 0.67 1.50 1.88
5 0.90 2.44 0.90 0.72 1.54 1.93
6 1.59 2.35 1.59 1.28 1.66 2.07
7 0.74 0.74 0.59
8 1.20 1.20 0.96
9 1.60 1.60 1.28
Table7. Desinganglesof thecutslopefromdifferentdesign
methods.
Computed Required
Angleof FS or FS or
Designmethods slope(
) M
R
,M
D
M
R
,M
D
Conventional design 35.7 1.40 1.4
LSD DA1C1 41.5 1.33 1.0
DA1C2, 41.5 1.06 1.0
DA3
DA 2 41.5 1.33 1.1
LRFD 41.5 1.33 1.17
4.4 Load and resistance factor design
As no factors areapplied on materials in LRFD, the
analysisof slopestabilityusingSLOPE/Wisthesame
asinASD, except that theloadsarefactored. TheFS
obtainedwouldbeequal toM
R
,M
D
, whichshouldbe
larger thanthereciprocal of 0.85, theresistancefactor
for overall stability. Becausethereis no appliedload
ontheslopeandtheonlyloadingistheself-weight, the
analysis of theoriginal slopeis thesameas inASD.
Theresistancefactor inLRFD is basically arecipro-
cal of thepartial resistancefactor inLSD, so theFS
computed by SLOPE/W should be larger than 1.17.
Comparing the factors of LRFD and DA2, the par-
tial factor for soil parameter is equal to 1.0, thus the
soil parametersarenot factoredinbothdesigns. Asa
result, theanalysisof thecut slopedesigninLRFDis
thesameasDA2inLSD.
4.5 Summary of cut slope design
Thedesignanglesof thecut slopearesummarizedin
Table7. Theanglesof thecut slopeobtainedfor LSD
andLRFD arelarger thanthosefor theconventional
design, thusasmalleramountof soil needstobecutin
order to maintainthestability of theslope. Although
thecomputedmoment ratiosinDA1C1andDA2are
larger thantherequiredvalues, theslopeanglecannot
bereducedanymore. If theangleisslightlylargerthan
41.5
, theslopesurfacewill containathinlayerof fill,

andthefactor of safety woulddropto about 0.7. So,
thestabilityanalysisisgovernedbysoil layering.
Duringthedesignof aslope, theself-weight of the
soil istheweight of theslope. However, it isinappro-
priateto apply partial factors for permanent loads to
theweight of soil. It is difficult in geotechnical cal-
culations to realizeto which sideof theequilibrium
equationtheweight of agivenvolumeof groundcon-
tributes. Thereforethepartial factor for self-weight is
takenasunity(Ovesen2002).
5 CONCLUSIONS
LSDusespartial factorsor loadandresistancefactors
to account for thevariations of both resistances and
loads, andachievesrelativelyuniformlevelsof safety
forvariousgeotechnical works. Inpractical geotechni-
cal design, theuseof LSDissimilartotheuseof ASD,
andnocomplexprobabilityandstatistical analysesare
required.
The outcomes of the design example show that
the LSD designs of the cut slope require a smaller
reduction of the slope angle and therefore are more
economicthantheconventional design.Thisreduction
of cost doesnot sacrificethesafetyof theslope.
ACKNOWLEDGMENT
Theworkpresentedinthispaperwassubstantiallysup-
portedbytheResearchGrantsCouncil of HongKong
(ProjectsNos. 622207andHKUST6126/03E).
REFERENCES
AASHTO (1998). Standard Specification for Highway
Bridges 1998 edition. American Association of State
HighwayTransportationOfficials, Washington, D.C.
Chu, L. F. (2007). Calibration of Design Methods for Large-
diameter Bored Piles for Limit State Design Code Devel-
opment. MPhil thesis, HongKongUniversity of Science
andTechnology, HongKong.
163
EuropeanCommitteefor Standardization(1994). Eurocode
1 Basis of designation on structures Part 1: Basis of
design. EuropeanCommitteefor Standardization(CEN).
EuropeanCommitteefor Standardization(2004). Eurocode
7 Geotechnical Design. EuropeanCommitteefor Stan-
dardization(CEN).
Frank, R., Bauduin, C., Driscoll, R., Kavvadas, M.,
Ovesen, N. K., Orr, T., and Schuppener, B. (2004).
Designers Guide to EN 1997-1. Thomas Telfored Ltd.,
London.
Geotechnical EngineeringOffice(1984). Geotechnical Man-
ual for Slopes, 2nd Edition. Geotechnical Engineering
Office(GEO), HongKong.
Geo-Slope(1991). Users Guide, SLOPE/W for Slope Stabil-
ity Analysis, version 2. Geo-SlopeInternational, Calgary,
Canada.
Honjo, Y. (2007). Development of a basic specific
design code in performance based specification con-
cept: The technical standards for port and harbor facil-
ities. ISGSR2007. First Unternational Symposium on
Geotechnical Safety andRisk. Tongji University, China,
pp. 105115.
Morgenstern, N. R., and Price, V. E. (1965). The analysis
of the stability of general slip surfaces. Geotechnique,
Vol. 15, pp. 7993.
Orr, T. (2005). Evaluation of Eurocode 7. Trinity College,
Dubin.
OvensenN. K. (2002). LimitstatedesignTheDanishexpe-
rience. Foundation Design Codes and Soil Investigation
in view of International Harmonization and Performance
based Design.Swets&Zeitlinger,Lisse,TheNetherlands,
pp. 107116.
Rosenblueth, E., andEsteva, L. (1972). Reliability basisfor
someMexicancodes. ACI Pub. SP-31, pp. 141.
Schuppener, B. (2007). Eurocode 7 and its latest develop-
ments. ISGSR2007, First International Symposium on
Geotechnical Safety andRisk. Tongji University, China,
pp. 117132.
Zhang, J., Tang, W. H., Zhang, L. M., andZheng,Y. R. Cali-
brationandcomparingreliabilityanalysisproceduresfor
slop stability problems. ISGSR2007, First International
Symposium on Geotechnical Safety and Risk, Tongji
University, China, pp. 205216.
164
Probabilisticchartsfor shallowfoundationsettlementsongranular soil
Politecnico di Bari, Bari, Italy
ABSTRACT: Thecalculationof settlement is oftenconditional onthedimensioningof shallowfootings. In
loosesoils, particularlyinsandydeposits, theimpossibilityof collectingundisturbedsampleshasledtodefinea
setof formulasforcalculatingthesettlementsonthebasisof in-situmeasures. Moreover, theinherentvariability
of themechanical propertiesof thesoil aswell asmodel errorsinprovidedformulasfor calculatingsettlements
requirenewproceduresfor assessingthereliability of calculatedsettlementsonthistypeof soils. A statistical
analysisonnineformulascommonlyusedfor calculatingthesettlementsonsandysoilshasbeencarriedout in
this paper. Thecalculations of thesettlements havebeencomparedwiththevalues of thefieldtest settlement
measures by Burland & Burbidge (1985) and their accuracy, precision and entropy have been investigated.
Besides, probabilistic charts of acceptablesettlements havebeen designedby means of themost preciseand
accurateformulafor settlement resultingfromthisstudy.
1 INTRODUCTION
Past studiesshowedhighvariabilityof calculatedset-
tlements estimated by different settlement formulas.
Such variability, mostly attributed to model errors,
has been studied comparing predictions with actual
settlements fromfieldtests by means of theratio R,
definedas:
whereS
c
represents themeasuredsettlement andS
m
thepredictedonecorrespondingtothesameloadlevel.
In this paper, starting fromprevious works on such
topic(Cherubini &Greco1998, Sivakugan&J ohnson
2004, Cherubini &Vessia2005) statistical toolshave
beenusedinorder toreducethescatter intoRdatasets
relatingtodifferentformulasandtoestimatethemost
efficient settlement formulation. Finally, onthebasis
of this latter some charts of acceptable settlements
havebeensuggestedfor designpurpose.
2 ESTIMATIONOF MODEL ERRORSIN
SETTLEMENT PREDICTIONS
2.1 The R ratio method
This study aims at investigating how nine different
formulas for settlement calculation, based on N
SPT
measures, canefficiently envisageinfieldmeasured
ones. TheR ratioisdeterminedasinEquation(1). It
canbedefinedasbiasfactorsinceitisanindexof the
deviationof thesettlement estimationfromits actual
value; this means that R takes into account different
modelsof nonlinearityinsoil responsetotheapplied
load and of dependency of soil stiffness indexes to
settlement magnitude
Inliteraturethisratioisoftenexpressedasthemea-
sureof thevariabilityof thesoil deformationmodulus
(Wu 1974, Kaggwa et al. 2002) when each single
expression is considered for calculating the settle-
ments. It is impossible to give a numerical demon-
stration that one formula is more correct, in mean
terms, than another on account of thestrong depen-
denceof theestimationsontheempirical relationsof
in-situ deformability parameters. Therefore, the fol-
lowing study does not estimate the correctness of
calculation formulas in relation to the ability of the
model toexplainthetheoretical behaviour of thesoil
but it considersall sourcesof uncertaintythat heavily
affect the estimations by means of one ratio. The R
valueshavethereforebeencalculatedandstatistically
analysedinordertoquantifytheefficiencyof ninecal-
culationmodels for settlements andto suggest some
designchoicesbasedonstatistical analysis.
Past studies on nine formulas have been car-
ried out by Cherubini & Greco (1998). They used
thefollowingformulations: Terzaghi & Peck (1967),
Meyerhof (1965), Meigh & Hobbs (1975), Arnold
(1980), Burland&Burbidge(1985),Anagnostopoulos
et al. (1991), Schultze & Sherif (1973), Berardi &
Lancellotta (1991). The above-named authors have
takenintoaccount thevaluesof thesettlementsmea-
sured in-situ by Burland & Burbidge (1985) in 192
casesof different-sizeshallowfoundationswhichlay
on different kinds of soils and determined the fre-
quencyof theRratiogiveninEquation(1).Afterwards
Sivakugan& J ohnson(2004) havefoundthat thebeta
probability density function represents the sampling
distribution of the R ratio for settlements estimated
through the following four equations: Terzaghi &
165
Table1. Scaleof conservationbasedonthevaluesof theR
ratio.
Level of conservation
Degreeof (or safety) of theformula
conservation Valuesof R adoptedfor thecalculation
5 >1.5 Highlyconservative
4 1.21.5 Conservative
3 0.81.2 Neutral
2 0.60.8 Non-conservative
1 -0.6 Highlynon-conservative
Peck (1967), Schmertmann et al. (1978), Burland &
Burbidge (1985) and Berardi & Lancellotta (1994).
Insuchastudy themeaningfulnessof thesamplesof
R taken into account has not accurately investigated
at all.
The model error of nine R datasets is studied
accordingtothefollowingsteps:
1) statistical toolsfor outlier detection;
2) the quality control of datasets in terms of
precision andaccuracy.
Tothisend, atfirst, someconsiderationsonapossi-
blescaleforRvaluesmustbedone.Table1proposesa
fivelevel scalewhichidentifiesthedistancefromthe
ideal valueof R inaqualitativeandquantitativeway.
Accordingtothisscale, theideal valuefor R is1
anditfallswithintheDegree3of conservation,marked
asneutral. Thevaluesof R lessthan1(Degrees1-2)
correspond to non-conservative formulas while the
valuesof Rgreaterthan1(Degrees4-5) correspondto
conservativeformulas. A safety-orienteddesigndoes
not consider the values of R falling within Degrees
1and2, but it tends to stabilizeinhigher degrees of
conservation. However, theseconsiderationsmadeon
thevalues takenby R do not helpto choosethebest
formulationfor thesettlement estimation.
Such a scale has been applied to the complete
datasetof Rvaluesinvestigatedthroughoutthispaper.
Such dataset cannot beinserted in this paper dueto
the limited number of pages available. Nevertheless
the5pagedatasetsarereportedinCherubini &Vessia
(2005): Rtakes, ingeneral, valuesbelongingtoall the
5degreesof conservationdetected. Itisthereforenec-
essarytointroducestatistical toolsforgatheringuseful
informationonthequalityof R values.
2.2 Analysis of the datasets
TheExploratory DataAnalysis (EDA) is usedinthe
first step. Infact statistical properties, suchascentral
tendencyanddispersion, canbegreatlyinfluencedby
differentkindsof anomalousdatawhichshouldalways
beidentifiedand, incase, discardedinorder to esti-
mateanysignificant statistical quantity. Thesevalues
are too low or too high compared with the interval
withinwhichthemeasuresof adatasetfall.Theseout-
liersaffectthemeanandthedatasetstandarddeviation,
andthey canalso affect thechoiceof theprobability
density function (PDF) which explains thesampling
distributioninthebest possibleway.
There are many methods for detecting the
anomalousvaluesof adataset, amongwhichtheBox
plot method. However, this method is not suitable
instrument for deciding on the statistical nature of
thesevalues. In fact, in a set of measures, thereare
many reasons whichcanexplainthepresenceof too
lowortoohighvaluescomparedwiththeaveragemea-
sures. TheBoxplot methodhasthereforebeenjoined
totheFergusontest whichcanfindout anddecideon
theanomalousvaluesonastatistical basis.
2.3 The box plot method and the Ferguson test
This methodcalculates thosevalues belongingtothe
dataset, markedas extremeadjacent values, which
mark the boundary of the range of dataset mea-
sures being considered as acceptable. Outside this
rangeof values, datasetmeasuresareconsideredtobe
anomalousandsothey areproperly examinedand/or
discarded.
Fromtheanalyses doneby means of theBox plot
method on nine datasets of R (the APK formula
givestwopossibleexpressionsforthesettlements: one
expressionchangesaccordingtothesmallersizeof the
foundationB andtheother onechangesasthevalues
of N
SPT
increase) forty-eight anomalous values have
beenfound.
Asmentionedbefore, detectingtheanomalousdata
in astatistical way is only anecessary condition for
beingdiscardedfromthedataset, but it isnot enough.
Any exclusion is decided also according to a back-
wardspath, i.e. theconditionsunder whichmeasures
are done so as to explain their anomaly in terms of
measureerror. Inour caseonly basic dataareknown
whiledatagatheringandany possibleprocessingare
unknown.
Experiencegainedinanalysisof datasetsfrommea-
suredpropertiesinhomogeneoussoil unitsshowsthat
somevalues arevery far fromthemean, whichmay
beduetothefollowingreasons(Rethati 1988):
thelayer containsforeigninclusions;
somesamplescamefromanadjacent layer (not the
onetobetested)duetotheinaccuratedetermination
of theboundariesbetweenthelayers;
thestructureor thestateof thesamplegiving the
outlying result, has changed during the sampling,
thetransport or thestorage;
somesamplesweretestedunderdifferentconditions
withrespect totheothers;
thetestedlayer wassubjectedtosomelocal effect;
a part of the tests was performed by another lab-
oratory assistant, or by adifferent method, or less
carefully, thoughbythesameassistant;
the evaluation of test results is affected by gross
error.
Consideringthechangeablestructureof theanoma-
lous data, Grubbs (1969) suggests two procedures
according to whether the physical reasons for the
anomalousvaluesareknownor not. Inthecaseunder
examination, sincetheconditionsunderwhichthesoil
166
Table2. Referencevaluesfor theindex
b
1
asthesample
sizevaries(Ferguson, 1961).
n
30 35 40 50 60
b
1
0.01 0.98 0.92 0.87 0.79 0.72
0.05 0.66 0.62 0.59 0.53 0.49
Table3. Valuesof theFergusonIndex(1961) calculatedfor
theninedatasetsof R.
Fergusonindexfor dataset %of discarded
Dataset complete corrected measures
TP 1.46 0.71 3.4
M 1.22 0.71 7.4
MH 1.04 0.71 5.1
A 1.19 0.71 5.1
BB 1.62 0.72 3.5
APK1 1.86 0.72 3.7
APK2 2.91 0.71 1.1
SS 0.98 0.70 1.7
BL 1.04 0.67 2.9
propertiesaremeasuredareunknown, decisionsabout
theanomalousvaluescanbetakenonly accordingto
thestatistical tests.
The Ferguson test (1961) is useful to this end
becauseit applies to thosedatasets whosevariances
areunknownandthepopulationtheybelongtoissup-
posedtohaveastandardarrangement.Moreover,when
moreanomalousvaluesareall putonthesamesideas
themean value(thesevalues being greater or lower
thanthemean), theFergusonindex iscalculatedasa
functionof skewness:
Its values are reported in Table 2 according to the
numerousnessof thesample.
Considering the first column of Table 3, the 9
datasets, whichincludetheanomalousvalues, donot
satisfy Equation(3) andso all theanomalous values
detectedthroughthebox plot havebeendiscardedin
order to get the corrected values given in the third
columnof Table3.
If the skewness ratio is over the values given in
Table2, theanomalousvalueswill bediscardedstart-
ingfromthosewhicharefarther fromthemeanvalue
andtheindexinEquation(2) will becalculatedagain
until thefollowinghypothesisissatisfied:
For thecompletedataset inthe9formulastakeninto
account, the values concerning the index

b
1
are
summarizedinTable3.
Each time this procedure has had a negligible
impact on the remaining number of values in the
datasets. In the following the datasets of R will be
corrected by means of theFerguson test and will be
referredtoascorrecteddatasets.
2.4 Accuracy, precision and entropy in the
settlement estimation
In addition to the notion of conservation previously
indicatedandclassified, other notions arehereintro-
ducedinorder togiveanobjectivebasistothechoice
concerning the ratios between calculated and mea-
sured settlement, indicated as R: accuracy, precision
andentropy.
Inparticular, precision representsthedispersionof
thevaluesof aninstrumentduringthecalculationof a
specificphysical quantity. It canbeconsideredasthe
dispersionmeasure. Onthecontrary, accuracy refers
tothegreater or smaller closenessof asetof measures
tothereal valuewhichissupposedtoberepresentedby
themeasures. Checkingtheaccuracyof adataset can
beassociatedtoanindicator of central trend, suchas
themean. Theabove-mentionedfeatures of adataset
areoften neglected becauseif, on onehand dataare
very few, ontheother handtheinfluenceof thehigh
inaccuracy of thestartingdataset onthegeotechnical
designing is underestimated even in the presence of
state-of-the-art physical andnumerical models.
A synthetic index which takes into account at the
sametimeaccuracyandprecisionof adataset of R is
theRanking Index suggested by Briaud & Tucker
(1988):
wherej andsstandfor themeanandstandarddevi-
ationof theabsolutevalueof thenatural logarithmof
R, definedbeforeas theratiobetweenthecalculated
andmeasuredsettlement. Thevaluesof thisindex let
usdefinetheconservationof thedataset havingmean
j andstandarddeviations. However, indefiningRI
thevariableR is assumed to bedistributed in alog-
normal way. If we want to generalize the problem,
Cherubini &Orr (2000) havesuggestedtheRanking
Distance index, definedasfollows:
wheresymbolshavethesamemeaningof Equation4.
Theideal valueof RDis0, whichrepresentsthemea-
suresof theratioR whosemeanisequal to1andthe
deviationtendsto0.
A geometrical interpretation of the Ranking
Distance can be given by a Cartesian plane where
the mean values of R are put on the x-axis and the
standarddeviationsof R areonthey-axis(Cherubini
& Orr 2000). It represents the distance of the mea-
suredR valuefromtheideal valuewhosecoordinates
are(1,0).
167
Lastly, the entropy, like the variance, is another
measureof dispersionalthoughthevariancemeasures
thedispersionaroundthemeanvalueof arandomvari-
ablewhereastheentropy measurestheoverall uncer-
tainty associated with the randomvariable (Kinsley
1983). This measure can be more significant than
thevariancewhenever bimodal randomvariables are
dealt with. Moreover, its valueis slightly affectedby
extreme values as outliers, which are characterized
generallybylowprobabilityvalues.
If a randomvariable x has n possible outcomes
x
i
(i =1,2,..n) and the probability assigned to each
outcomeis P
i
, then theentropy associated with x is
(Shannon1948):
According to the entropy definition, the higher the
entropythelessreliabletheestimation.
The ratio of the calculated entropy to the maxi-
mumentropyiscalledrelative entropy H
r
(x) (Kinsley
1983):
whereH
max
is themaximumentropy of therandom
variable. Therelativeentropy of R canprovideaway
of relatingtheabsolutedispersionof eachRdatasetto
the8remainders.
2.5 Reliability estimation of the R datasets
Accordingtowhat hasbeenpreviously indicatedand
elaborated, synthetic statistical indicators relating to
the datasets of the ratios between calculated and
measured settlements can beused for estimating the
reliability of thesettlement calculations. Theway of
achievingthishasbeensubdividedintotwostages:
location of the anomalous values that might be
presentedinthesamplesof R andtheir removal;
definitionof theaccuracy, precisionandentropyof
eachdataset of R.
In stage 1, as already seen in the previous para-
graphs, theanomalous values havebeenlocatedand
discardedthroughtheBoxplotandFergusonmethods.
Instage2, theerrorsassociatedwiththecorrected
samples of R have been described in terms of both
precisionandaccuracy, throughtherankingdistance,
andabsoluteandrelativeentropy.
Table4givesthevaluesof RDfortheninecorrected
datasetsof R. Only theRDindex hasbeentakeninto
account sinceitsvaliditydoesnot dependonhowthe
dataset isdistributed.
Figure1showsthevaluesof themeanonthex-axis
and the values of R deviations for the nine datasets
reported inTable4. Thepoints corresponding to the
correcteddataset areshapedlikeempty circleswhile
thedarkcirclesrefer tothemost efficient formulas.
Table4. Values of RD calculatedfor themodifieddataset
of R.
Standard
Dataset Mean deviation Skewness Kurtosis RD
TP 3.17 1.9 1.83 0.2 3.0
M 1.53 1.1 1.84 0.34 1.5
MH 1.53 1.1 1.83 0.33 1.3
A 0.98 0.5 2.02 0.19 0.7
BB 1.58 0.8 1.90 0.13 1.3
APK1 1.16 0.6 1.99 0.39 0.6
APK2 1.24 0.6 1.99 0.20 0.6
SS 1.15 0.6 1.98 0.30 0.7
BL 0.82 0.5 2.02 0.39 0.5
Figure1. Planeshowingbothmeanvalues (onthex axis)
andstandarddeviations(ontheyaxis) of Rcorrecteddataset
for theninesettlement formulas.
If the curves of RD having a constant value are
represented on a graph, they have the shape of arcs
of circles drawn around theideal point (j
R
=1.0;
s
R
=0.0) (Figure1). Moreover, thesamefigureshows
that aroundtheideal point it canbepossibletosee
threeareas having different ratios between accuracy
and precision: a sector of circle within which more
precise and accurate datasets fall can be set in a 30
degreeanglefromthevertical; theequallypreciseand
accurate datasets fall within 30 to 60 degrees from
thevertical; lastly, thedatasetswhicharemoreprecise
thanaccuratearebetween60and90degreesfromthe
vertical. Insidetheseareas, ontheright of theideal
point, lowvaluesof RD correspondtohighaccuracy
and precision; high values of RD correspond to low
accuracyandprecision.
Asyoucanseethevaluesof RDarehighlyaffected
by thepresenceof anomalous values andtheir influ-
enceisunpredictableapriori withregardbothtowidth
and sign. For the nine datasets of R the points cor-
respondingto calculatedsettlements greater thanthe
measuredones(R1) fall ontheright of theideal
168
point: under theseconditions, if RDvaluesaresmall,
thecloser thevaluesof R aretotheideal point the
morereliabletheyare.
Figure1caneasily showtheratios betweenaccu-
racy and precision through the arcs of the circle.
Moreover, whenchoosingthemostefficientformulas
accordingto RD, datasets havinganR meansmaller
than1must bediscarded. Infact, valuesof R smaller
than1, andmorespecificallysmaller than0.8, corre-
spondto underestimations of thesettlement sizethat
arenot safevaluesingeotechnical designing.
For thenineformulas corrected by means of Box
plotandFergusontestapplication, referringtoTable4,
Berardi &Lancellotta(BL) formulawill bediscarded.
Theformulas correspondingtosmaller values of RD
areshapedas dark circles, that is: thetwo ones sug-
gestedbyAnagnostopoulos et al. (1991) andtheone
suggested by Arnold (1980) and Schultze & Sherif
(1973).All theseformulasfallsintheareamoreaccu-
ratethanprecise. Theseformulascoincidewithmean
values andstandarddeviations of R whicharelower
thantheother ones.
In fact, themost conservativeformulaamong the
ones taken into examination is theTerzaghi & Peck
(TP) whose RD value is equal to 3, then comes the
Meyerhof (M) which is also themost datedtogether
withtheTP one. TheMeigh& Hobbs (MH) andthe
Burland&Burbidge(BB) formulashavethesameval-
ues of RD (that is 1.3), thoughMH is moreaccurate
thanprecisewhileBB isbothaccurateandprecise.
Nevertheless, theauthorsthinkthattheactof refer-
ringtoabsolutequantitativescalesforthevaluesof RD
isacomplicatedpoint duetothe: (1) kindof problem
takenintoaccount and(2) thenumber of independent
variablesaccordingtowhichtheratioR iscalculated.
Infact, Orr & Cherubini (2003) haveintroducedthe
RankingDistance(RD) for estimatingtheprecision
andaccuracy of different formulas usedfor calculat-
ing therest thrust coefficient. In that casevalues of
RDlower than1.5havebeenrecordedbut theformu-
las had only one independent variable, i.e. angle of
internal friction(). Onthecontrary, inthisstudythe
formulas for settlement calculation depend on many
parameters (N
SPT
, q
c
, Shape factors, Relative den-
sity,empirical coefficients).Nevertheless,considering
that the value of RD calculated in the ideal point
is RD=0, then thevalues falling to theright of the
ideal point onthegraphrepresent acceptablemean
andstandarddeviationvalues. Oncewegoawayfrom
theidealpointthesevaluesincreaseRDandsothey
aremoreconservative.
FromTable4andthedarkcirclesinFigure1canbe
inferredthatAPK1, APK2, A andSSformulascorre-
spondtothelowest acceptablevalueof RDcompared
withall other expressions, followedby BB, MH and
M formulas. Finally, TP formulaisextremelyprecau-
tionary. On the contrary, BL formula, though it has
small valuesof RD, fallstotheleftof theidealpoint
on the graph and it is not on the safe side. In view
of theseremarks on theranking distance, dispersion
canbefurther investigated, that is precisionof theR
Table5. Estimationof bothentropyandrelativeentropyfor
theninedatasetsof R.
Dataset Entropy(H) Relativeentropy(Hr)
TP 1.5 1.0
M 1.2 0.8
MH 1.2 0.8
A 1.0 0.7
BB 1.2 0.8
APK1 1.0 0.7
APK2 1.1 0.7
SS 1.1 0.7
Table6. Synoptic tableof bothmomentsanddistributions
of thedatasetsof R.
K Pearsons Probability
Dataset coefficient distribution
TP 0.048 Normal
M 0.11 Normal
MH 0.039 Normal
A 0.047 Normal
BB 0.048 Normal
APK1 0.035 Normal
APK2 0.046 Normal
SS 0.04 Normal
datasetsthroughthecalculationof theabsoluteHand
relativeH
r
entropy, introducedinsection2.4. Table5
givesthevaluesof entropyforeightoutof ninedatasets
of R.
Asyoucansee, themost disperseformulasareTP,
MH andM, whiletheless disperseformulas areBL
andA. Theseresultspoint out that, referringtothese
nineformulasfor calculatingsettlements, theexpres-
sions whichareless dispersearoundtheaverageare
alsolessdisperseinabsoluteterms. Thisresult isalso
confirmed by the calculation of the relative entropy
accordingtoEquation(7).
2.6 Estimation of R distributions
Oncethemostaccurateandpreciseformulashavebeen
found we can go on estimating the sampling distri-
butions relativeto thedatasets of R for thedifferent
formulas we have considered. By exactly determin-
ingthedistributionsof threesamplesof Rconcerning
APK1,APK2andSSformulasitispossibletoprovide
areliability-baseddesign. A formulationhasnot been
reportedbecauseof resultsaresimilar toSSformula.
To this end, it is crucial to know how the three
datasetsof Rarearrangedtodevelopexceedingprob-
ability charts of settlements, which give values for
a quick design, once the acceptable settlement size
is defined. The estimation of the sampling distribu-
tions has been made by Pearson chart (Pearson &
Hartley1972).Table6showsthevaluesof K Pearsons
coefficient calculatedfor all thedatasetsconsidered.
169
It canbeseenthat all theformulasconsideredfol-
lowthenormal distribution. Theseresultsdiffer from
Sivakugan&J ohnson(2004) showingthatanomalous
valuescanhaveagreatroleontheidentificationof the
samplingprobabilitydistributions.
As amatter of fact Sivakugan & J ohnson (2004),
donot carry out preventivestudiesonthedatasetsof
availablesettlements and they find abeta-typesam-
plingdistributionof R for BB, TP andBL formulas.
Moreover, theestimationof thesamplingdistribution
is affectedby thenumerousness of thesampleunder
consideration. Unfortunately, Sivakugan & J ohnson
(2004)dontreportthenumberof valuesof settlements
consideredsothatnocomparisoncanbedonewiththe
datasetsfromthisstudy, whoseminimumsizeis159
values.
3 SETTLEMENT PROBABILITY CURVES
Thefinal aimof suchathoroughstudyonthedatasets
of Rrelativetoseveral formulasfor settlementpredic-
tionunder shallowfootingsistoprovideaquick tool
fordesignandconsultation. Infact, curvesof themea-
suredsettlementsforthethreeformulaschosenbefore
havebeendrawnfor thedifferentlevelsof probability.
Theselevels of probability havebeendeterminedfor
thetwoexpressionsof Anagnostopouloset al. (1991)
andfortheoneof Schultze&Sherif (1973), whichare
themost preciseformulas according to theavailable
full-scaledata.
Asfar asthevaluesof thepossiblesettlementsare
concerned, bothEurocode7andtheItalianProvisions
(TestoUnitario2008) donotgiveacceptablereference
values. Nevertheless, somesuggestionscomefromlit-
eraturefor different building standards (for example
Sowers & Sowers 1970). Inpractice, settlements can
beconsideredasacceptableif they donot causeboth
staticandfunctional problemstothebuilding. Inthis
study wehavemadereferenceto fivevalues of set-
tlements considered as acceptable: 5, 15, 25, 35and
45mm. Theexceeding probability curves havebeen
drawnaccordingtothenormal distributionlawfor the
valuesof R, sothat:
Thismeansthat theprobability that avalueof set-
tlement ishigher thanthecorrespondingactual value
comesfromthecalculationof theprobabilitythat any
value of R is lower than a fixed value of R. Owing
tothisconsiderationwehaveusedthreenormal den-
sitydistributionsrelatedtothemodifieddatasetsof R
for APK1, APK2andSS formulas (Figures 24) for
calculatingtheprobabilitycurves.
The values corresponding to the probability of
exceedance can be obtained from these settlement
probabilitycurves.
For example, consideringthat a20mmsettlement
has been estimated by means of the SS formula,
Figure 2. Normal-distribution probability curves corre-
sponding to different values of acceptable settlements
calculatedbyAPK1formula.
calculatedbyAPK2formula.
calculatedbySSformula.
Figure 4 shows that there is a 46% probability that
theactual settlement under footingwill bemorethan
25mm, thatistosaytheprobabilityof calculatingset-
tlements below25mmwill be70%. Similarly, there
170
will bea26%probabilitythat thesettlement isbelow
15mm, i.e. an74%probability that thesettlement is
morethan15mm.
For all three formulas the exceeding probability
curvesforvaluesof settlementsoccurringbetweenthe
curvesplottedonthegraphwill beobtainedbyinter-
polating thevalues of probability for thetwo curves
whichmarktherange.
Theseprobability charts let theengineer knowthe
probability that a certain value of the actual settle-
mentisexceededprovidedthatthesettlementvalueis
known, i.e. theyprovidetheconditionedprobabilityof
theactual settlements:
whereS
c
andS
m
arethecalculatedandmeasuredset-
tlements, respectively; s andS arethevalues of their
respectivesettlements. Sotheconditionedprobability
intermsof not-exceedancewill bethefollowing:
Therefore, as a result of a reliability-based design-
ing in terms of settlements (calculated by means of
oneof thethreeformulasAPK1, APK2andSS), the
probability that the actual settlement of the footing
doesnot exceedagivenvaluecanbeeasilygot bythe
conditionedprobabilitytheorem:
wherethefirsttermcomesfromtheprobabilitycharts
(Figures 24) whilethesecondonederives fromthe
reliability-baseddesigning. Infact, if thesettlements
areestimatedbytheSSformula, andtheprobabilityof
notexceedanceof a20mmsettlementisequal to10
4
,
theprobability that thesettlement actually measured
isbelow25mmwill beeasilygot fromFigure4:
If the calculation is made by means of the two
other formulas, for thesameestimatedsettlement we
will get:
4 CONCLUSIONS
Thispaper showstheresultsof astatistical studycar-
riedoutonsettlementsundershallowfootingsonsand
by theso-calledbiasfactor R. Settlementshavebeen
calculatedaccordingtonineformulascommonlyused
in geotechnical designing, while the values of full-
scalesettlementshavebeenobtainedfromapaper of
Burland& Burbidge(1985).
This kind of approach has led to choose three
expressionsfromthenineformulaswhichhaveturned
out tobemorepreciseandaccurateaccordingtoboth
processingandavailabledata. Thisselectionhasbeen
madeby detectingtheanomalous values for thenine
datasets of R andby calculatingtherankingdistance
inorder toestimatethereliabilityof theninedatasets
of R. Moreover, thisstudyhasalsoestimatedthebest
statistical distribution for each dataset by means of
Pearsonchart.
Theaimof theseanalyseswastodeveloptheprob-
ability charts of the settlements measured for the
three formulas chosen according to precision and
accuracy criteria (APK1, APK2 and SS). Thanks to
thesecharts andto theconditionedprobability theo-
rem, the exceeding probability of actual settlements
canbedeterminedfromtheexceedingprobability of
calculatedsettlements.
REFERENCES
Anagnostopoulos, A.G., Papadopoulos, B.P. & Kavvadas,
M.J. 1991. Direct estimationof settlement onsand, based
on SPT results. In Proceedings of the 10th ECSMFE
Florence, Italy: 293296.
Arnold, M. 1980. Predictionof footingsettlementsonsand.
Ground Eng. 2: 4049.
Berardi, R. &Lancellotta, R. 1991. Stiffnessof granularsoils
fromfieldperformance. Geotechnique 41(1):149157.
Berardi, R. & Lancellotta, R. 1994. Prediction of settle-
ments of footings on sand: accuracy and reliability. In
Proceedings of Settlement 94, (1): 640651.
Briaud, J. & Tucker, L.M. 1988. Measured and predicted
axial loadresponseof 98piles. Journal of Geotechnical
Engineering 114(9): 9841001.
Burland, J.B. & Burbidge, M.C. 1985. Settlement of foun-
dations onsandandgravel. InProceedings of Inst. Civil
Eng. part 1, 78: 13251381.
Cherubini, C. & Greco, V. R. 1998. A comparisonbetween
measured and calculated values in geotechnics. In
Proceedings of the 21st International Conference on
Probabilities and Material (PROBAMAT): 481498.
Cherubini, C. & Orr, T.L.L. 2000. A rational procedurefor
comparing measured and calculated values in geotech-
nics. In Proceedings of the International Symposium on
Coastal Geotechnical EngineeringinPractice,Yokohama:
261265. Rotterdam: Balkema.
Cherubini, C. &Vessia, G. 2005. Studi di Affidabilitnel cal-
colodi cedimenti interreni sabbiosi.Atti del III Convegno
Scientifico Nazionale Sicurezza nei Sistemi Complessi,
1921 ottobre, Bari.
Kaggwa, W.S., Cheong, M.T. & J aksa, M.B. 2002. Assess-
ment of the luck associated with the settlement pre-
dictions that are based on elastic theory. In Pttler,
KlapperichandSchweiger(ed.),International Conference
on Probabilistics in geotechnics.Technical and Economic
risk estimation, Graz, Austria, 1519 September.
Kinsley, H.W. 1983. Some geotechnical applications of
entropy. In Augusti, Borri and Vannucchi (ed.), Fourth
171
international conference on applications of statistics and
probability in soil and structural engineering, Florence,
13-17 June, Italy.
Eurocodice 7, Progettazione geotecnica Parte 1: Regole
generali, UNI ENV 1997 1-Aprile1997.
Ferguson,T.S. 1961. Rulesfor rejectionof outliers. Rev. Inst.
Int. Stat. 29: 2943.
Grubbs, F. 1969. Procedures for DetectingOutlyingObser-
vationsinSamples. Technometrics 11(1): 121.
Meigh, A.C. & Hobbs, N.B. 1975. Soil mechanics, Section
8, Civil Engineers Reference Book (3rd ed.). Newnes-
Butterworth: London.
Meyerhof, G.G. 1965. Shallowfoundations. Journal of Soil
Mechanics and Foundation Division 91(SM2).
Orr, T.L.L. & Cherubini, C. 2003. Use of the ranking
distanceas anindex for assessingtheaccuracy andpre-
cisionof equations for thebearingcapacity of piles and
at-restearthpressurecoefficient. Canadian Geotechnical
Journal: 12001207.
Pearson, S. & Hartley, H. 1972. Biometrika, Tables for
Statisticians(2).
Rethati, L. 1988. Probabilistic solutions in geotechnics.
Developments in geotechnical engineering 46. Elsevier
(ed.).
Schmertmann, J.H., Hartman, J.P. & Brown, P.R. 1978.
Improved strain influence factor diagrams. Jour-
nal of Geotechnical Engineering Division 104(GT8):
11311135.
Schultze, F. &Sherif, G. 1978. Predictionof settlementsfrom
evaluatedsettlementobservationforsand. InProceedings
of 8th ICSMFE, 3, Moscow, URSS: 225230.
Shannon, C.E. 1948. A mathematical theoryof communica-
tion. The bell system technical journal 27: 379423.
Sivakugan, N. & J ohnson, K. 2004. Settlement prediction
ingranular soils: aprobabilisticapproach. Geotechnique
54(7): 499502.
Sowers, G.B. &Sowers, C.F. 1970. Introductory soil mechan-
ics and foundations. Collier-MacMillan: London.
Terzaghi, K. &Peck, R.B. 1967. Soil mechanics in engineer-
ing practice. NewYork: J. Wiley& Sons.
Testo Unitario 2008. D.M. Infrastrutture 14 gennaio 2008.
NuoveNormeTecnicheperleCostruzioni.Ministerodelle
Infrastrutture, Ministero delInterno, Dipartimento della
ProtezioneCivile.
Wu, T.H. 1974. Uncertainty, safety and decision in soil
engineering. Journal of the geotechnical engineering
100(GT3): 329348.
172
Resistancefactor calibrationbasedonFORM for drivensteel pipepiles
inKorea
J.H. Park, J.H. Lee, M. Chung& K. Kwak
Korea Institute of Construction Technology, Goyang, Gyeonggi, Korea
J. Huh
Chonnam National University, Yeosu, Chonnam, Korea
ABSTRACT: Resistancefactors for static bearing capacity of driven steel pipepiles werecalibrated in the
framework of reliabilitytheory. Twotypesof First Order ReliabilityMethod(FORM), MeanValueFirst Order
SecondMoment(MVFOSM)methodandAdvancedFirstOrderSecondMoment(AFOSM)methodwereapplied
for thereliabilityanalysesandresistancefactor calibration. Thetarget reliabilityindicesareselectedas2.0and
2.33for grouppilecaseand2.5for singlepilecaseinconsiderationof thereliabilitylevel of thecurrent design
practices, redundancy effect of thepilegroup, acceptablerisk level, and significanceof individual structure.
Resistancefactorswererecommendedforthepilefoundationdesignandconstructionpracticeandthesubsurface
conditionsinKorea.
1 INTRODUCTION
Geotechnical design practice has been gradually
changing in recent years fromAllowable Strength
Design (ASD) to Limit State Design (LSD), known
as Load and Resistance Factor Design (LRFD) in
the United States and Canada. Geotechnical resis-
tancefactorsaregreatlyinfluencedbythesitegeology
and the design and construction practice, and it is
important to incorporate local ASD experience and
practiceaccumulatedover many decades indevelop-
inggeotechnical LRFDcodes. Foundationdesignand
constructionpracticeinKoreaisquitedifferent from
that of other countries, and theKorean geotechnical
community realizedtheneedof developingour own
LRFDcodes.
Thispaperpresentstheresistancefactorscalibrated
for thestatic bearing capacity design methods com-
monlyusedinKoreafor drivenopen-endedsteel pipe
piles. This is part of theresearch effort in Koreafor
LRFDimplementationingeotechnical engineering.
2.1 Resistance bias factor statistics
Staticloadtest reportsondrivensteel pipepileswere
collectedfromall aroundKoreaalongwithavailable
records of subsurface investigation, and laboratory
testing. Whileatotal of over 2,000staticloadtestdata
was collected and reviewed, 57 static load test data
werefoundtobereliableanduseful for thisstudy. All
of thesetest pilesweresortedintotwocasesfor reli-
ability analysis: 1) SPT N-valueat piletip less than
50(N-50), and2) SPT N-valueat piletipequal to
or morethan50(N50). Andreliabilityanalysisand
resistancefactor calibrationwereperformedonly for
total pilecapacity. Fivedifferent failurecriteriawere
employed in this study to evaluatethepiles bearing
capacity, and the Davissons criterion was proven to
performthe best in the previous study (Kwak et al.
2008).
TheKoreanDesignStandardsforFoundationStruc-
tures(2003)adoptedtwostaticbearingcapacityanaly-
sismethods.Thesetwomethodswereusedtoestimate
thestaticultimatecapacity(Q
u
)of the57testpiles.The
first methodisexpressedinthefollowingformula.
where,
v
: effective overburden pressure at the pile
tipwiththedepthlimit to20B, N
q
: bearingcapacity
factorasafunctionof soilsfrictionangle(), c: cohe-
sionof soil, N
c
: bearingcapacityfactor for cohesion,
N
c
=9 for soils with =0, A
p
: cross section area
of pile assuming 100%plugging, f
s
: unit frictional
resistancealongtheshaft (=K
s

v
tan for cohesion-
lesssoils),K
s
=1.4(1 sin)and =20degrees,
v
:
averageeffectiveoverburdenpressurealongtheshaft.
f
s
=c
u
forcohesivesoils, : adhesionfactor, andc
u
:
undrainedshear strengthof soil,A
s
: pileshaftsurface
area.
Thesecondmethodistheempirical formulawhich
isrevisedafter Meyerhof (1976) aspresentedbelow.
where, m=3 L
b
/B30 (L
b
: pile embedment depth
into bearing stratum), N
p
: average SPT N value
173
(uncorrected) inthevicinity of piletip, mN
p
1500
metric tons/m
2
, n=0.2, N
s
: average SPT N value
(uncorrected) along pile shaft, n N
s
10 metric
tons/m
2
.
Resistancebiasfactor isdefinedastheratioof the
measured ultimatebearing capacity fromaload test
overthepredictedultimatebearingcapacitybyastatic
bearingcapacity formula. Theresistancebias factors
for the selected test piles are computed for the two
static designmethods describedabove. Thestatistics
of theresistancebiasfactorsarecomputedforreliabil-
ity analysis andresistancefactor calibration, andare
presentedinTable1. For thecaseof SPT N-valueat
piletiplessthan50, Equation(1)appearstopredictthe
bearingcapacitymorecloselytothemeasuredcapac-
itythanEquation(2). Distributionof theseresistance
bias factors was alsoexaminedandlognormal distri-
bution was found to most closely represent the bias
factor distributions(Kwaket al. 2008).
2.2 Reliability analysis results
AASHTOLRFDSpecificationStrengthCaseI (2007)
isconsideredasthecritical loadingcasefor pilebear-
ingcapacity. Two types of theFirst Order Reliability
Methods(FORM), theMeanValueFirstOrderSecond
Moment (MVFOSM) methodandtheAdvancedFirst
Order SecondMoment (AFOSM) methodwereused
toevaluatethereliabilityof thestaticbearingcapacity
designmethods.
Two randomvariables, theload(Q) andtheresis-
tance(R), areconsideredandthey areassumedtobe
statistically independent andlognormally distributed.
Thelimitstatefunctioninthiscaseisdefinedas: g(R,
Q)=ln(R)ln(Q)=ln(R/Q). If theloadeffectsto
beconsideredareonly deadandliveloads, therelia-
bilityindex() canbecalculatedbythefollowingthe
MVFOSM formula.
where,
R
,COV
R
:meanandcoefficientof variationof
resistancebiasfactors, andFS: factor of safety. Inthe
AFOSM analysis, thelimitstatefunctionislinearized
atapointonthefailuresurface.Thelimitstatefunction
g(R, Q) canbeexpressedinthefollowingformat.
Table1. Resistancebiasfactor statistics.
Resistance Nat tip-50 Nat tip50
biasfactor
(
R
) Eq.(1) Eq.(2) Eq.(1) Eq.(2)
Mean 0.975 1.750 0.726 1.317
COV* 0.511 0.755 0.411 0.743
No. of data 27 30
*COV=coefficient of variation
The computed reliability indices by both the
MVFOSM and AFOSM methods are shown in
Figures12. TheKoreanDesignStandardsfor Foun-
dationStructuresrequiresaminimumsafetyfactor of
3.0forpilebearingcapacitydesign, andinsomecases
a safety factor larger than 3.0 has been used in the
Koreanpractice. Accordingly, reliabilityanalysiswas
performedfor safetyfactorsof 3.05.0. Reliabilityof
the static design methods seems to be within a rea-
sonablerange. TheAFOSM methodresultedinlarger
reliabilityindicesthantheMVFOSMmethodby3.1%
to 6.7%for SPT N at piletip less than 50 case, and
2.9%to 9.8%for SPT N at piletipequal to or more
than50case.
Figure1. Reliabilityindices(N-50).
Figure2. Reliabilityindices(N50).
174
3 RESISTANCE FACTOR
3.1 Target reliability index
Target reliabilityindexisaveryimportant element in
resistancefactor calibration, andit shouldbeselected
considering various sources of uncertainties in load
andresistanceevaluation, safetylevel requiredbythe
society, cost-effectivenessaswell asthecurrentdesign
andconstructionpractice. Meyerhof (1970) proposed
thattheprobabilityof failureof foundationsshouldbe
between 10
3
and10
4
, which corresponds to relia-
bility index between3.1and3.7. Barker et al. (1991)
suggestedtarget reliabilityindexas2.0to2.5intheir
researchfortheAASHTOresistancefactorcalibration
fordrivenpiles, andWhithiametal. (1998) confirmed
that thisrangeof target reliabilityindexisreasonable
for asinglepiledesignconsideringtheredundancyof
pilegroups.TheNCHRP507report(Paikowsky2004)
showedthat theevaluationof resistancefactorsusing
variousreliabilityindicesbeforeestablishingthefinal
targetreliabilityindexcontributedtodevelopingarea-
sonabletargetreliabilityindex. Itrecommendedtarget
reliability indices in conjunction with group redun-
dancy: 2.33 for redundant piles (5 or morepiles per
pilecap) and3.0for non-redundant piles (4or fewer
pilesperpilecap)andthesevalueswerefinallyapplied
todeveloptheAASHTOLRFDresistancefactorsfor
drivenpiles.
As statedabove, target reliability index of 2.03.0
appears to bereasonablefor pilefoundations. Relia-
bilityindicescomputedusingall 57datainthisstudy
rangefrom1.5to2.9for correspondingsafetyfactors
of 3.0to5.0, andthetwodesignmethods(Equations
[1] and[2]) havefairlyuniformreliabilityindicesfor
bothpiletipcases(N-50andN50).Therefore, the
sametargetreliabilityindicescanberecommendedfor
bothdesignmethodsandbothpiletipcases. Consider-
ingthat aminimumsafetyfactor of 3.0hasbeenused
forsteel pipepiledesign, recommendtargetreliability
index of 2.0to2.5iswithinareasonablecompatibil-
itywiththereliabilityof thecurrentdesignpracticein
Korea. Basedonthereliabilityanalysisof thecurrent
static design methods presented above and consid-
ering the target reliability indices recommended in
the literature or selected in other research projects,
thetarget reliability indices for static bearingcapac-
ity designof steel pipepiles inKoreaareselectedas
presented in Table 2. Two different target reliability
indices are recommended for the group pile case to
account for acceptable risk level or significance of
individual structure. Reliabilityindexof 2.0to2.5cor-
responds to probability of failure of 2.3% to 0.6%,
respectively.
Table2. Target reliabilityindices.
Type Target reliabilityindex(
T
)
Grouppilecase 2.0and2.33
Singlepilecase 2.5
3.2 Calibration of resistance factors
Resistancefactorswerecalibratedby thethreemeth-
odsusedinthereliabilityanalysisandtheresultswere
compared. The following formula was used for the
MVFOSM method:
where,
T
: targetreliabilityindex.Thebasicalgorithm
of theAFOSMandMCSmethodsforresistancefactor
calibrationissimilartothatof therespectivereliability
analysis. Thelimit statefunctionforAFOSM is:
Calibrationwasperformedforthetwostaticbearing
capacity analysis methods, and the results are pre-
sentedinTable3. Calibratedresistancefactorsfor the
caseof N at tip -50 arerelatively larger than those
for theN at tip 50 case. Theresistancefactors for
Equations (1) and(2) arealso different for bothpile
tipcases.
Calibratedresistancefactorsfor thecaseof Nattip
-50 are relatively larger than those for the N at tip
50case.Theresistancefactorsfor Equations(1) and
(2) arealsodifferentforbothpiletipcases. Resistance
factorsrecommendedinthisstudyarespecificfor the
pilefoundationdesignandconstructionpracticeand
thesubsurfaceconditionsinKorea.
Thisstudypresentsthefollowingfindingsandconclu-
sions.
1. Targetreliabilityindicesfor staticbearingcapacity
designof steel pipepilesareselectedas2.0and2.33
for grouppilecase(5or morepilesinagroup) and
2.5forsinglepilecase(4orfewerpilesinagroup).
2. The difference between computed reliability
indicesbytheMVFOSMandAFOSMmethodscan
besignificant andtheMVFOSM methodmay not
bereliablewhenthelimitstatefunctionisnonlinear.
Table3. Calibratedresistancefactors.
Nat tip-50 Nat tip50
Eq.(1) Eq.(2) Eq.(1) Eq.(2)
T
MvF* AF** MvF AF MvF AF MvF AF
2.00 0.41 0.44 0.46 0.48 0.37 0.40 0.35 0.37
2.33 0.34 0.37 0.37 0.39 0.32 0.35 0.28 0.30
2.50 0.31 0.34 0.33 0.35 0.30 0.33 0.25 0.27
*MvF : MVFOSM , **AF : AFOSM
175
3. Calibrated resistance factors for the case of N at
tip-50arerelativelylarger thanthosefor theNat
tip50case. Theresistancefactorsfor Equations
(1) and(2) arealsodifferentfor bothpiletipcases.
REFERENCES
AASHTO2007. LRFD Bridge Design Specifications, Amer-
ican Association of State Highway and Transportation
Officials, FourthEdition, Washington, D.C.
Barker, R., Duncan, J., Rojiani, K., Ooi, P., Tan, C., and
Kim, S. 1991. Manuals for the Design of Bridge Foun-
dations. NCHRP Report 343, Transportation Research
Board, Washington, D.C.
Korean Society of Geotechnical Engineers 2003. Design
Standards for Foundation Structures. Ministry of Con-
structionandTransportation, Seoul, Korea.
Kwak, K., Kim, K. J., Huh, J., Park, J. H., Chung, M., and
Lee,J.H.2008.TargetReliabilityIndicesof StaticBearing
CapacityEvaluationof DrivenSteel PipePiles. Proceed-
ings of the 87thAnnul Meeting of Transportation Research
Board, TransportationResearchBoard, Washington, D.C.
(CD-ROM)
Meyerhof, G. G. 1970. Safety Factors in Soil Mechanics.
Canadian Geotechnical Journal, 7(4): 349355.
Meyerhof, G. G. 1976. BearingCapacity andSettlement of
PileFoundations.J ournal of theGeotechnical Engineering
Division, 102(GT3), ASCE: 197228.
Paikowsky, S. G. 2004. Load and Resistance Factor Design
for Deep Foundations. NCHRP Report 507, Transporta-
tionResearchBoard, Washington, D.C.
Whitiam, J., Voytko, E., Barker, R., Duncan, M., Kelly, B.,
Musser,S.,andElias,V.1998.Load and Resistance Factor
Design (LRFD) of Highway Bridge Substructures. Pub-
lication FHWA HI-98-032, FHWA, U.S. Department of
Transportation.
176
Anevaluationof thereliabilityof verticallyloadedshallowfoundations
andgrouped-pilefoundations
TetsuyaKohno
Bridges and Structures Research Group, Center for Advanced Engineering Structural Assessment and
Research (CAESAR), Public Works Research Institute(PWRI), Tukuba, Japan
Takashi Nakaura
Chodai CO.,LTD. Tokyo, Kita-ku, Japan
MasahiroShirato& Shoichi Nakatani
Bridges and Structures Research Group, Center for Advanced Engineering Structural Assessment and
Research (CAESAR), Public Works Research Institute(PWRI), Tukuba, Japan
ABSTRACT: This paper estimates the reliability of the ultimate bearing capacity of shallow foundations
and grouped-pilefoundations. First, model uncertainty in theestimation for thebearing capacity of shallow
foundations is evaluated using small and large-scale plate-load test results and design formulas. A similar
evaluationis thenconductedfor singlepiles usingin-situloadtest results, andfurthermore, thereliability of
apilegroupasasystemisestimatedconsideringtheincreaseinthereliability withincreaseinthenumber of
piles, becausemost pilefoundationsinJ apanaregrouped-pilefoundationswithmorethan5piles. Finally, this
paper shows that thereliability indices for shallowfoundations andgrouped-pilefoundations withmorethan
4pilesthat aredesignedfollowingcurrent designspecificationsareequivalent.
1 INTRODUCTION
The advent of performance-based design specifica-
tions for bridges occurredin2002withtheJ apanese
Specifications for Highway Bridges (J apan Road
Association, 2002).Therequiredbridgeperformances
are clearly described for Normal Situations, Fre-
quent Earthquake Situations, Rare Earthquake Situ-
ations, andsoforth. TheseSpecificationsalsocall for
limit statesof structural componentssothat integrat-
ing the components that are verified for their limit
statecriteriacan beconsidered to haveachieved the
requiredperformanceof total bridgesystems.Accord-
ing to performance-based design, design require-
mentsareclearlyspecified, whiletheexistingdetailed
designmethods,includingcomputational methodsand
acceptablelimits, arespecified as typical acceptable
solutions. Thus, the introduction of performance-
baseddesignencouragesdesignerstoapplyadvanced
or better solutionsinpractice, evenif thesesolutions
arenot presentedintheSpecifications.
However, most of theSpecificationsarenot based
on a probabilistic background. Rather, the typical
acceptablesolutionis achievedwithempirical safety
factors.Theadoptionof areliabilitydesignconcepthas
been enthusiastically encouraged so that newdesign
approachesor materialscanbecomparedwithcurrent
practicesintermsof reliability.Accordingly, theSpec-
ificationsarebeingrevisedtowardtheimplementation
of theloadandresistancefactor design(LRFD) for-
matwithareliabilitydesignconcept. Anintermediate
draftandthefinal draftareplannedforreleasein2009
and2011, respectively.
This paper reports an evaluation of theresistance
factors for vertically loadedshallowfoundations and
grouped-pilefoundations. This study is apart of the
project for therevisionof theJ apaneseSpecifications
for HighwayBridgesbyCAESAR, PWRI. Inparticu-
lar, wewill consider thegroupeffectonthereliability
of pilefoundations and attempt to achieveaconsis-
tent reliabilitylevel betweenshallowfoundationsand
grouped-pilefoundations.
2 RELIABILITY INDEX ANDRESISTANCE
FACTOR FOR SHALLOWFOUNDATIONS
Thetheoryof bearingcapacityof shallowfoundations
has been established, but it is well known that theo-
retical bearingcapacityformulastendtooverestimate
bearing capacity for larger sized footings. However,
inthepasttwentyyears, extensive, careful experimen-
tal and numerical research was conducted in J apan,
and now the J apanese Specifications for Highway
Bridges incorporatethescaleeffect into thebearing
capacityformulaof shallowfoundation.Therefore,the
model uncertainty inthebearingcapacity of shallow
177
Figure1. Definitionof ultimatebearingcapacity.
foundationsisnot expectedtobesensitivetofooting
sizeand wemay useeven small scaletest results to
estimatemodel uncertaintyfor large-scalefootings.
2.1 Error in estimating the bearing capacity
of shallow foundations
Inthis study, themodel uncertainty is definedas the
ratio of X/Y, in which X is the observed bearing
capacity in a load test and Y is the corresponding
calculated value using the bearing capacity formula
in the J apanese Specifications for Highway Bridges
andsoil investigationresults. This ratio involves two
uncertainties: the modeling of bearing capacity the-
ory and the evaluation of geotechinical parameters.
Hereinafter, theratioX/Y is referredtoas themodel
error.
For shallow foundations, we use a database of
plateloadtests inwhicharigidplatethat modeleda
shallowfoundationthatwassubjectedtocenteredver-
tical loads. Weevaluatetheultimatebearingcapacity
based on a load-displacement curve with three dif-
ferent methods as shown in Figure 1. As shown in
Figure1(a), whenthepeak point is clearly identified
in a load-displacement curve, the load at the peak
point is definedas thebearingcapacity. As shownin
Figure1(b), whenaload-settlement curvecontains a
clear break point andshowsasimilar curvetoelasto-
perfectlyplasticbi-linear curves, theloadatthebreak
point is definedas thebearingcapacity, eventhough
there is no clear peak in the load-settlement curve.
As showninFigure1(c), whenatest was terminated
before reaching the ultimate state, we approximated
theload-settlement curveusinganexponential func-
tion and used theapproximated ultimateload as the
observedbearingcapacity.
inwhichR=load, s =settlement, R
u
=approximated
ultimateload, ands
y
=approximatedyieldsettlement.
Note that the tests that were terminated before the
load reached 1.2 times the approximated yield load
areomittedbecausetheapproximatedultimateloadis
consideredtobehighlyunreliable.
A summaryof thedatasetsusedhereinisshownin
Table1, inwhichB=footingwidthandB
=effective
footing width=squareroot of thefooting basearea
for arectangular footingor widthfor astripfooting.
Bothin-situandlaboratorytestsaredealtwithherein.
Triaxial compressiontestswereconductedfor in-situ
test cases to estimate soil strength parameters. The
considered confining pressures were also displayed.
Subgrade reactions on the footing base of highway
bridgepiers aretypically 100500kN/m
2
basedona
databaseof past designresults. Accordingly, atypical
lateral stress level at a depth of a half of the foot-
ingwidthcanrange50250N/m
2,
assumingtheearth
pressureat rest to be0.5, which is equivalent to the
confining pressures of thetriaxial compression tests
consideredinthein-situtestdata.Thelaboratoryload-
ingtestsconsideredhereinwereconductedby ajoint
researchproject betweentheUniversityof Tokyoand
thePWRI (Tatsuoka, etal. 1989, Okahara, etal. 1992),
whocarefullypreparedthetestfacilitytominimizethe
influenceof friction between thesand box and sand
ontheexperimental bearingcapacity.Accordingly, we
canneglect thesidefrictioneffect onthetest results.
Itisalsoworthnotingthatthetestswereconductedin
planestrainconditions.
Thecalculatedbearingcapacityisobtainedassuch:
in which Q
u
=ultimate bearing capacity (kN),
A
=footing base area and is equivalent to A of

footing base area in this study (m
2
), c =cohesion
(kN/m
2
), q =surcharge (kN/m
2
), =unit weight of
soil (kN/m
3
), and B=strip footing width (m) or, if
the shape of the foundation is rectangular, B is the
shorter sidelength(m). N
q
, N
c
, andN
=coefficients
of bearingcapacitythatarederivedbyKomada(1964),
and S
q
, S
c
, and S
=modifiers for the scale effect

on the bearing capacity coefficients (J apan Road
Association),
in which c
=c,c
0
(1c
10), c
0
=10 (kN/m
2
),
q
=q,q
0
, (1q
10), q
0
=10 (kN/m
2
), and
178
Table1. Statistical valueof model error inthebearingcapacityof shallowfoundations.
(a) Squareandrectanglefootings
Ground Number of Number of Variability Soil test andconfining
type datasets grounds B
(mm) Bias
p
COV
p
pres-sure(kN/m
2
) Reference
Gravel 8 3 3001300 1.479 0.171 Triaxial Maeda, et al. (1990),
compression Maeda, et al. (1991)
tests(300)
Sand 13 3 2003000 0.894 0.257 Triaxial compression Chimi, et al. (1996),
test (100300) Nakano, et al. (1996),
Yamamoto, et al. (1996),
Ouchi, et al. (1993),
Onodera, et al. (1993),
Okahara, et al. (1987),
Briaud, et al. (1997)
Silt and 4 1 300750 4.597 0.474 Boxshear test UtsunomiyaNational
clay 6 2 300750 0.537 0.285 Triaxial compression HighwayOffice(1976)
test (Noinformation)
Soft rock 6 1 300 0.814 0.254 Triaxial compression Okahara, et al. (1987)
test (Noinformation)
(b) Rigidstripfootingintheplatestrainbox
Ground Number of Variability Soil test andconfining
type datasets B
(mm) Bias
p
COV
p
pressure(kN/m
2
) Reference
Sand(experiment) 14 230500 0.722 0.307 Torsional simpleshear Okahara, et al. (1992)
test (100)
, , andj=1/3. Thebearingcapacitycoefficients
andthescaleeffect modifiers aredimensionless. For
rectangular footings, and are the dimensionless
foundationshapefactorsthat aregivenas:
in which B and D=short and long side lengths of
footing, respectively. For strip footings, = =1.
isthemodifier for theembedment effect,
inwhichD
f
=effectivefoundationembedment depth
(m). isalsodimensionless. Thesoil strengthproper-
tiesof cohesion, c, andinternal frictionangle, , that
arederivedviasoil laboratory tests showninTable1
areused in thecalculation. When c appeared in CD
triaxial compression test results even in the case of
sand and gravel, the obtained c is incorporated into
thecalculatingof bearingcapacity.
Figures2and3showtherelationshipbetweenthe
model error andtheeffectivefootingwidthfor all test
results. With all data, the bias of model error X/Y,
p
, iscloseto1.0andthemodel error, X/Y, doesnot
haveatendencytochangewitheffectivefootingsize,
suggestingthat themodifier for thesizeeffect tothe
bearingcapacitycoefficient workswell. Accordingly,
we can assume that the statistical values examined
herein can also be used as the model uncertainty in
thebearingcapacityof larger sizefootings.
Figure2. Relationship between model error and effective
footingwidthfor sandandgravel.
Figure 3. Relationship between the model error and the
effectivefootingwidthfor silt andclay.
179
Figure4. Frequencydistributionof themodel error withall
dataexcept for silt andclaycases.
However, thebiaschangeswithdifferencesinsoil
investigationmethodsforderivingsoil strengthparam-
eter values. For example, for sands, calculatedvalues
tend to overestimate test values, especially when a
torsional test is adoptedto derivesoil parameter val-
ues. It isworthnotingthat thelaboratoryfootingload
test wasconductedintheplanestrainconditionwhile
in-situ tests were not, and this is another reason to
account for the difference in the bias between the
in-situ and laboratory tests. When using abox shear
test toobtainsoil parameter valuesfor cohesivesoils,
thecalculatedbearingcapacityisunderestimatedbya
largedegreeat 0.474.
However, when using a triaxial compression test,
thebiasbecomescloser to1.0andisequal to0.573.
Figure 4 shows the frequency distribution of the
model error when the soil parameter values are
obtainedwithtriaxial compressiontestsonlyonsand
andgravel.Thefrequencydistributionseemstobecon-
sistentwithbothnormal andlog-normal distributions.
Inpractice, shallowfoundations arerarely placedon
clayey soils for highway bridges in J apan. Based on
theseresults, wefinallyestimatethat themodel error
of thebearingcapacityfollowsalog-normal distribu-
tionwithabias
p
of 0.85andacoefficientof variation
(COV
p
) of 0.30under thefollowingconditions:
(1) Thesoil parameter valueshouldbeestimatedby
triaxial compressiontestsat aconfiningpressure
of approximately250kN/m
2
.Thisconfiningpres-
surevaluesareexpectedtoapproximatethelateral
confiningstress at adepthof thehalf depthof a
shallow foundation. However, the shallow foun-
dationwidthisunknowninthesoil investigation
stage before design. In addition, triaxial com-
pression tests may become difficult in a lower
confining pressure. Therefore, we suggest that
250kN/m
2
shouldberelevantbasedontheearlier
designexperiencedescribedabove.
(2) For cohesivesoils, morecompleteinvestigations
arerequiredbecausethemodel uncertaintytends
to besensitiveto thereliability of theestimation
of cohesionc.
Indesign, aplateloadingtest usingacircular plate
withadiameter of 0.3missometimesusedtoderive
soil parameter values indirectly. When the ultimate
Figure5. Differenceinthemodel error withchangingthe
effectivefootingwidthB
.
loadQ
u
isobtainedusingtheplateloadingtest, can
beestimated fromback analysis using thefollowing
equation, assumingthevalueof cohesionc:
Sinceoneof thetestcasesconductedaseriesof loading
tests at a particular site using square and rectangu-
lar plateswithdifferingdiametersrangingfrom0.3m
to 1.30mon gravel (Maeda, et al. 1991) as summa-
rizedinTable1,wewill estimatethemodel uncertainty
whenusingaplateloadingtest toderivesoil parame-
ter values. Inthiscase, weconsideredacohesionc of
100kN/m
2
that wasestimatedfromCDtriaxial com-
pressiontest results, becausethegravel onsiteseems
to have cohesion because of moisture content. With
this assumption, weperformedaback calculationof
usingEq. (6) andc =100kN/m
2
. Thebackcalcula-
tionof givesaresult of 45degrees. Figure5shows
the relationship between the model error and effec-
tivefootingwidths. Figure5showsthatthebiasof the
model error approaches1.0at everyeffectivefooting
widthwhenusingaplateloadingtest together witha
triaxial compressiontest toestimatethesoil parame-
ter values. Thisresult suggeststhat, if weevaluatethe
geotechnical parameters with both triaxial compres-
siontests andaplateloadingtest, wecanestimatea
morereliablebearingcapacity.
2.2 Safety margin and resistance factor
When a load Q and a resistance R are probabilistic
variablesassumedtofollowalog-normal distribution,
asafetyfactor j
FS
inthecurrent designisassociated
withthereliabilityindex asfollows:
in which,
Q
and COV
Q
arethebias and coefficient
of variationof loaddistribution, Q, (or calculatedload
acting on the head of the pile), and
R
and COV
R
180
Table2. Reliabilityindex of existingshallowfoundations,
target reliabilityindex
T
, andresistancefactor.

T
+
3.04 3.10 0.33
arethebias andcoefficient of variationof resistance
distribution, R. For simplicity, thisstudyshowsacon-
ditional reliability analysis, disregarding thestatistic
issues of load, Q, and regarding loads in thecurrent
J apaneseSpecifications for Highway Bridges as the
deterministicfactoredvalues. Weendedupassuming
Q
=1.0andCOV
Q
=1.0. Eq. (7)canbereformulated
asfollows:
When we apply the safety factor value of the cur-
rent Specifications, j
FS
=3.0, and
R
=0.85 and
COV
R
=0.30asevaluatedabove, thereliabilityindex
iscalculatedas =3.04.
Basedonthisresult, wedeterminedthetarget reli-
ability index
T
=3.10. The resistance factor + is
calculatedby:
inwhichQ
d
isthedesignvertical load,R
ud
isthedesign
vertical bearingcapacity, andR
uc
isthecharacteristic
valueof thevertical bearing capacity, obtained with
Eq. (2) andc and measuredbyaprescribedsoil test
or acombinationof aprescribedsoil test andaplate
loading test. Rearranging Eq. (2) and replacing j
FS
with +, where+=theinverseof thecurrent safety
factor j
FS
leadstothefollowingequation:
Finally, +iscalculatedtobe0.33.Table2summarizes
thereliabilityindex of existingshallowfoundations
andthetarget reliabilityindex
T
, resistancefactor +
to fulfill the
T
. However, as statedabove, thereis a
largebias for cohesivesoils ontheunfavorableside.
Accordingly,itisnecessarytoapplysomepartial factor
tothetermof N
c
inEq. (2) inorder tocompensatefor
theassumedbiasintheestimationof + above.
3 RELIABILITY INDEX ANDRESISTANCE
FACTOR FOR GROUPED-PILE
FOUNDATIONS
Many load test results are available for single piles
andthemodel uncertaintyinthebearingcapacitycan
bedirectly estimatedusingthesetest data. However,
most pilefoundations areusedas groupedpiles that
are connected though a rigid pile cap. Accordingly,
it is preferred to estimate the model uncertainty of
the bearing capacity of a foundation systemrather
than that in individual singlepiles in agrouped-pile
foundation. For example, theAASHTOLRFDbridge
designspecifications(AASHTO, 2007), inthecaseof
pilefoundations, considersthatthebearingcapacityof
pilegroupsshouldbelarger thansinglepilesbecause
soil betweenpilescanbecompactedduringconstruc-
tion, and it adopts different target reliability indexes
between pile groups with smaller numbers of piles
andthosewithlarger numbers(morethan5) of piles
(Paikowskyetal., 2004, Zhangetal., 2001). However,
this compaction effect can be considered to change
according to piling method. In Europe, Eurocode 7
(CEN, 2004) doesnot describeaspecificredundancy
factor that is similar to theAASHTO specifications.
However, Bauduin (2002) has suggested a factor of
1.1 that can bemultiplied with thebearing capacity
of groupedpiles withastiff pilecapconsideringthe
redundancy that stems fromaredistributionof loads
within the piles when one of the piles reaches its
maximumresistance.
Inthis paper, wewill first attempt to estimatethe
difference in the model uncertainty in the bearing
capacity of grouped piles. Wewill then evaluatethe
reliabilitylevel andresistancefactorsforgrouped-pile
foundationsasafunctionof thenumber of pilesinthe
system.
3.1 Source of model uncertainty under particular
consideration in this study
We are particularly considering the following two
kindsof uncertaintyherein: 1) uncertaintyinestimat-
ingtheaveragebearingcapacityof several singlepiles
and2) uncertainty inthebearingcapacity becauseof
thedifferenceinsite.
As for thefirst source, let us consider anexample
inwhichthereareseveral piles withidentical details
on a particular site and their bearing capacities are
all tested. Thebearingcapacity of eachpilewill dif-
fer. However, asillustratedinFigure6, whenchoosing
npilesamplesfromall pilesat thesite, obtainingthe
averagebearingcapacity, andrepeatingall thisprocess
againandagain, wecanendupfurnishingthedistri-
butionof theaveragebearingcapacityof npilesatthe
site. Asaresult, it isexpectedthat thevariationof the
averagebearingcapacity will decreasewithincrease
inthenumber of samplepilesnbecausethevariation
intheindividual bearingcapacitywill becanceledout
amongthesamplepiles.Accordingly, thereliabilityof
bearingcapacity asagroupedpilefoundationshould
belarger thanthatof asinglepileandweshouldeval-
uatethevariabilityinthebearingcapacityof grouped
pilefoundationsatasitewithafunctionof thenumber
of pilesn.
As for the second source, the bearing capacity is
calculatedusingthefollowingequation.
181
Figure6. Differenceintheuncertaintyof bearingcapacity
betweensinglepilesandpilegroupsat aparticular site.
Table3. Variability inmodel error of singlepilesat asite,
COV
L
.
Site 1a
1b
2 3 4 5
Number of single 2 2 2 3 2 3
pilesinasite
COV
L
0.01 0.01 0.05 0.05 0.04 0.04
Site-1aand Site-1b tests wereconducted at thesamesite,

but their pilelengthsandbearinglayersaredifferent.
inwhichR
u
=bearingcapacity of asinglepile(kN),
q
d
=ultimateend bearing capacity intensity per unit
area (kN/m
2
) that is a function of SPT-N value and
specifiedintheJ apaneseSpecificationsfor Highway
Bridges, A=areaof pilebase(m
2
), U =perimeter of
pile(m), L =thicknessof soil layer consideringshaft
resistance(m), f =maximumshaft resistanceof con-
sideredsoil layer(kN/m
2
)asafunctionof SPT-Nvalue
andisspecifiedintheSpecifications.Theultimateend
bearingcapacityintensityq
d
andmaximumshaftresis-
tancef intheSpecifications arebasedonregression
analyses using pile load test results that were con-
ducted all over J apan. Accordingly the model error
changesbysite.
3.2 Variability in the average bearing capacity
of several piles at a particular site
In thePWRI database, as tabulated inTable3, there
aresixloadtestcasesinwhichtwopileswithidentical
detailsweresupportedbythesamebearingstratumat
aparticular siteandtested. Whilethepileresistanceis
comprisedof thebaseandshaft resistances, thetests
at site Nos. 25 were conducted using piles with a
frictionreductiontreatmentover80%of theembedded
pile length. Accordingly, while the variability in the
Figure7. RelationshipbetweentheCOV of model errorand
theeffectiveembedment ratio, LL,D.
Table4. Variability inmodel error of grouppilesat asite,
COV
L
.
Pilenumber
inafooting 1 2 3 4 5 6 9
COV
L
0.050 0.035 0.029 0.025 0.022 0.020 0.017
bearingcapacity for siteNo. 1results fromboth the
variationinthesideandbaseresistancesthat for site
Nos. 25resultsalmostentirelyfromthevariabilityin
thebaseresistance.
WeestimatedthebiasandCOV of themodel error
foreverytestcase,assumingnosideresistanceoverthe
frictionreductiontreatmentpartinthecalculation.The
resultisshowninTable3andFigure7. Figure7shows
therelationshipbetweenvariationandtheratioof the
effective embedded pile length LL, where the fric-
tionreductiontreatmentpartisdisregarded, tothepile
diameter D. Thevariabilityfor siteNos. 25islarger
thanthatforsiteNo1.Thistendencyshouldbereason-
able, becausethespatial variationinthesideresistance
issupposedtobesmallerthanthepointvariationinthe
shaft andbaseresistancesandthevariationinthepile
bearingcapacitycandecreaseastheratioof theshaft
resistancetothetotal pileresistance. Finally, weesti-
matethatthesitedependentvariabilityfor twopilesat
aparticular siteisof theorder of smaller than10%.
Finally, aCOV
L
valueof 0.05isadoptedasarep-
resentative value in this study. When assuming that
thevariation in thedistribution of theaveragebear-
ingcapacityof several pilesataparticular sitefollows
thet-distribution, thevariation in themean valueof
themodel error COV
L
decreasesasshowninTable4,
withincreaseinthenumberof piles, inwhich2, 3, 4, 5,
6, and9pilesareincludedincommonlyusedhighway
bridgegroupedpilefoundations.
3.3 Variability in model error due to site
differences (COV
m
)
Asstatedabove,themodel errorinthebearingcapacity
of groupedpilesshoulddiffer duetositedifferences.
Accordingly, wetriedtoestimatethevariabilitydueto
182
sitedifferenceCOV
m
using an availabledatabaseof
singlepileloadtestsat different sites.
Theuncertainty inthemodel error for singlepiles
isalsoafunctionof boththevariationwithinthecon-
sidered site and the variation due to site difference.
Accordingly, thevariationinthemodel errorforsingle
piles, COV
P
, is obtained by combining thevariation
for singlepileswithintheconsideredsite, COV
L
, that
hasalreadybeenestimatedaboveandgiveninTable3,
withthevariationduetositedifference, COV
m
:
WhileCOV
p
and COV
m
changewith thenumber of
piles, COV
m
ispresumedtobeunaffected.
To estimate the value of COV
p
, we analyze the
model error for singlepilesusingthePWRI database.
The test value of bearing capacity of single piles is
obtainedasfollows:
(i) The peak load is considered as the bearing
capacity whenthepeak loadappears beforethe
settlement reachesat 10%thepilediameter.
(ii) In the case that the peak load does not appear,
theloadwhenthesettlement reaches10%of the
pilediameter is definedas thebearingcapacity,
asstatedintheSpecifications.
(iii) When the pile load test was terminated before
thesettlement reaches10%of thepilediameter,
the exponential fitting is applied to the load-
displacement curveandtheloadat asettlement
of 10%of thepilediameter iscalculated. Thisis
becausemost loadtest dataindicatethat thepile
bearingcapacity canbeequivalent to themobi-
lizedloads at asettlement level of 10%thepile
diameter (Okaharaet al., 1999).
Note that we use only that data that meets the
followingconditions:
Notreatment toreducethesideresistance.
Whenusingtheexponential fitting, themaximum
loadduringthetest was larger thantheyieldload
calculatedbythefittedcurve.
Table5showsthevaluesof COV
p
for singlepiles.
Fromtheseresults,wesetthevariablemodel of thesin-
glepileas
p
=1.0andCOV
p
=0.35. This indicates
thatthebiasof themodel error for grouped-pilefoun-
dationsalsowill be
p
of 1.0.Finally,substitutingthese
valuesintoEq. (8) givesCOV
m
=0.346(Table6).
3.4 Error in estimating the bearing capacity
of grouped-pile foundations
COV
p
is estimated by substituting COV
L
, which
changesaccordingtothenumber of piles, andCOV
m
indicated in Table 6 into Eq. (8). We set the values
indicatedinTable7.
3.5 Safety margin and resistance factor
Weevaluatedthesafety marginandresistancefactor
for grouped-pilefoundationsashasbeendonein2.2.
Table5. Statistical valuesof themodel error inthebearing
capacityof singlepiles.
Piletypeand Number of Bias Variability
pileinstallationmethod test results
p
COV
p
Cast-in-placePiles 16 1.034 0.315
Driven-in-steel piles 11 0.928 0.327
Inner-soil-excavation 9 1.225 0.323
Hollowsteel pile
Table6. Variabilityduetothedifferenceof thesites, COV
m
.
COV
m
0.3464, not dependent onthenumber of piles
Table7. Variabilityinthemodel error of pilegroup, COV
p
.
Number of piles 1 2, 3 4or more
COVp 0.350 0.348 0.347
Table8. Reliabilityindex of pilegroup.
3.06 3.08 3.09
Table9. Resistancefactor +.
+ 0.329 0.331 0.333
First, weevaluatedthereliabilityof existinggrouped-
pilefoundationstodefinethetarget reliability index.
Thereliabilityindexof existinggrouped-pilefounda-
tionsisobtainedbysubstitutingtherelevantstatistical
values and the resistance factor +=3 into Eq. (6).
Table8showstheresult. of singlepilesis3.06, and
increasesaccordingtothenumber of piles, resulting
in of grouped-pilefoundationswithmorethanfour
pilesbeing3.09.
Most existing grouped-pilefoundations havefive
or morepiles. AsshowninTable8, of grouped-pile
foundationswithmorethanfourpilesisassumedtobe
3.09. This is almost thesameas the valueof exist-
ingshallowfoundations. Therefore, thisstudyreveals
thatthereliabilityof shallowfoundationsandgrouped-
pilefoundationsarealmostequal inthecurrentdesign
norm.
Basedontheresults of Table8, wethenset
T
of
pilefoundation systems having morethan four piles
to 3.10 to make it equal to that of shallow founda-
tions, 3.10.Table9showstheresistancefactors+that
183
satisfies
T
. Theresistancefactor, +, increases with
increasedinthenumber of piles. This is becausethe
variationintheaveragebearingcapacity of theindi-
vidual piles in a grouped pile foundation decreases
withincreaseinthenumber of piles. However, thedif-
ferencein + dueto thedifferencein thenumber of
piles is not large. This indicates that theaccuracy of
thecalculatedbearingcapacity is poor andCOV
m
is
large. Accordingly, there is an ample roomto allow
forapplicationof alargerresistancefactorwhenapile
loadtest isconducted.
4 CONCLUDINGREMARKS
Weexamined theresistancefactor for shallowfoun-
dationsandpilefoundationsystems.
Inthenormal designsituation,thereliabilityindexes
of both shallow foundations and pile foundation
systemsarealmost thesame: 3.10.
Forpilefoundationsystems,weshowedthatthereli-
ability index of existingpilefoundations that have
4or morepilesis3.10.
Weproposedaresistancefactor for pilefoundation
systemsasafunctionof thenumber of pileswitha
consistent target reliabilitylevel.
REFERENCES
AASHTO. 2007. LRFD Bridge and Construction Specifica-
tions. 4th edition.
Bauduin,C.2002.Designof axiallyloadedcompressionpiles
accordingtoEurocode7. Proceedings of the Ninth Inter-
national Conference on Piling and Deep Foundations.
Nice,PressesdelEcoleNationaledesPontsetChaussees:
301311.
Briaud J, L. and Gibbens, R. 1997. Large scale load tests
anddatabaseof spreadfootingsonsand, PublicationNo.
FHWA-RD-97-068, Federal HighwayAdministration.
CEN/TC250. 2004. EN1997-1 Eurocode 7 Geotechnical
Design Part 1: General Rules.
Chimi, K., Ouchi, M.,Takada, S. andTatsuoka, F. 1996. Foot-
ing settlement in central and eccentric vertical loading
tests on a Pleistocene sand deposit, Proceedings of the
Geotechinical Conference, JGS, No.31: 15771578.
Chimi, K., Ouchi, M.,Takada, S. andTatsuoka, F. 1996. Foot-
ing settlement in central and eccentric vertical loading
tests on a Pleistocene sand deposit, Proceedings of the
Geotechinical Conference, JGS, No.31: 15771578.
J apan RoadAssociation. 2002. Specifications for Highway
Bridges.
Komada, K. 1969. Bearingcapacitycalculationdiagramsfor
shallowfoundationswithinclinedloadinthetwodimen-
sional ground. Technical Report of PWRI(135). Public
WorksResearchInstitute. InJ apanese.
Maeda, Y., Kusakabe, O. and Ouchi, M. 1991. Largescale
in-situloadingtests of square, rectangular footings ona
densescoria, Journal of Geotechnical Engineering, JSCE,
No.430/III-15: 97106.
Maeda,Y., Kusakabe, O. andShigeki, K. 1990. Therelation-
shipbetweentheloadedwidthandcoefficientof subgrade
reactionobtainedwithlarge-scaleloadtests, Proceedings
of the Annual Conference, JSCE, No. 3: 10101011.
Nakano, M., Tasaka, M., Oyake, T. andTatsuoka, F. 1996.
Shape factor and effects of load eccentricity in bearing
capacityof aPleistocenesanddeposit, Proceedings of the
Geotechinical Conference, JGS, No. 31: 15791580.
Okahara, M., Kohata, H., Mori, H. andTsugawa, Y. 1987.
A study on the approximation method of the bearing
capacityof shallowfoundationsonrock,Technical Memo-
randumof PWRI (2512), PublicWorksResearchInstitute.
InJ apanese.
Okahara, M., Takagi, S., Kimura, Y., Mori, H., Asai, K.,
Watarai, M. and Inoue, A. 1992. An experimental study
on the bearing capacity of rigid foundations, Technical
Memorandum of PWRI (3087), Public Works Research
Institute. InJ apanese.
Okahara, M., Takagi, S., Kimura, Y., Mori, H., Asai, K.,
Watarai, M., Inoue, A. andTatsuoka, M. 1992. Anexper-
imental research of the bearing capacity of rigid foun-
dations, Technical Memorandum of PWRI (3087), Public
WorksResearchInstitute. InJ apanese.
Okahara, M., Takagi, S., Nakatani, S., & Kimura, Y. 1991b.
A studyonthebearingcapacityof singlepilesanddesign
methodof columnshapedfoundations. Technical Memo-
randumof PWRI (2919), PublicWorksResearchInstitute.
InJ apanese.
Onodera, I., Miura, Y., Shiraki, T. and Ouchi, M. 1993.
Comparisonof modulusof deformationbetweenlabora-
tory and in-situ tests, Proceedings of the Geotechinical
Conference, JGS, No. 28: 18931894.
Ouchi,M.,Abe,S.,Kusakabe,O.andMaeda,Y.1993.Results
and observation of in-situ loading tests of square foot-
ings on Hayakawariver mouth sand, Proceedings of the
Geotechinical Conference, JGS, No. 28: 15831584.
Paikowsky, S, G., Brigisson, B., McVay, M., Nguyen, T.,
Kuo, C., Baecher, G.,Ayyub, B., Stenersen, K., OMalley,
K., Chernauskaas, L., ONeill, M. 2004. NCHRP
REPORT507 Load and Resistance Factor Design (LRFD)
for Deep Foundations, TransportationResearchBoard.
Tatsuoka,F.,Tani,K.,Okahara,M.,Morimoto,T.,Tatsuta,M.,
Takagi, S. and Mori, H. 1989. Discussion on influence
of the foundation width on the bearing capacity factor
by Hettler and Gudehus, Soils and Foundations, 29(4):
146154.
UtsunomiyaNational Highway Office. 1976. Report of the
analysisof in-situtest for caissonfoundation.
Yamamoto, N., Nakano, M., Oyake, T. andOuchi, M. 1996.
BearingcapacitytestsinapnenumaticcaissononaPleis-
tocene sand deposit, Proceedings of the Geotechinical
Conference, JGS, No. 31: 15811582.
Zhang, L., Tang, W. H. andNg, C. W. W. 2001. Reliabilityof
axiallyloadeddrivenpilegroups, Journal of Geotechnical
and Geoenvironmental Engineering, ASCE: 10511060.
184
Studyonrational groundparameter evaluationmethodsfor thedesignand
constructionof largeearth-retainingwalls
Y.Yamamoto, T. Hirose& M. Hagiwara
East Nippon Expressway Co., Ltd., Tokyo, Japan
Y. Maeda
Kyushu Kyoritsu University, Kitakyushu, Japan
J. Koseki
The University of Tokyo, Tokyo, Japan
J. Fukui
Advanced Construction Technology Center, Tokyo, Japan
T. Oishi
Expressway Technology Center, Tokyo, Japan
ABSTRACT: Tokyo Outer Ring Expressway is an important trunk road that circularly connects the areas
locatedabout 15kmaway fromthecenter of Tokyo, J apan. In thevicinity of theTokyo-Chibaboundary, the
expresswayextendsforabout12.1km, wheresectionsapproximately9.5kmlongmakeuseof semi-underground
ditchstructuresthat will mostlybeconstructedwithanopen-cut systemusingearth-retainingwalls.
Sincetheconstructioncost of earth-retainingwalls accounts for themajor portionof thetotal project cost,
theevaluationof groundparameterswill greatlyaffect thecost. Aimingfor cost reduction, thispaper presentsa
studyonrational methodsfor evaluatinggroundparameterstobeusedfor designpurposes.
Thefollowingsummarizesthetechnical characteristicsandresultsof thisstudy.
1) Up until now, most ground parameters to beused for design purposes estimated with empirical methods
using, for example, SPT N-valueshadconservativevalues, whichresultedinquiteuneconomical designs. In
thisstudy, aseriesof highlyaccuratetriaxial compressiontestswereconductedonundisturbedsamplesand
groundstrengthparametersbasedonsoil dynamicswereinvestigated.
2) Theinvestigation results revealed that in thecaseof Pleistocenesand layers, alarger valuefor undrained
shear strengthc
u
is obtainedas comparedto thevalueestimatedusingSPT N-valueandunconfinedcom-
pressionstrength, q
u
. Inthecaseof Pleistocenesandlayers, theexistenceof truecohesionc
d
, duepossibly
tocementation, wasconfirmed, sothat it canbeeffectivelyconsideredinthedesign.
3) A rational groundparameter evaluationmethodwasinvestigatedviastatisticsbasedontriaxial compression
test dataof varioussoil types.
4) Soil layerswereclassifiedintoseveral categoriesbasedontheirfinescontentandplasticityindex, andground
parameterswereassignedfor eachcategorizedzone.
5) The deformation of earth-retaining walls during excavation was compared between the computed values
usingdifferent setsof groundparametersandthemeasuredvaluefromtrial construction. Asaresult, it was
foundthat thegroundparametersobtainedfollowingtheproceduresproposedinthisstudy aresufficiently
conservative, andtheycanprovidegreat economical effect.
1 INTRODUCTION
1.1 Scope of project
Tokyo Outer RingExpressway is an important trunk
road that circularly connects theareas located about
15kmfromthecenterofTokyo(seeFig.1).Thesection
in the vicinity of theTokyo-Chiba boundary, which
extends for about 12.1km, is under construction.
Urbanization is in advanced stages along the road
andapproximately9.5kmisalsoNational Route298
located on the surface for public use and the rest,
exclusive to the Tokyo Outer Ring Expressway, is
underground. Itmakesuseof semi-undergroundditch
structureswheremostof itwill beconstructedwithan
open-cut systemthat uses earth-retaining walls (see
Fig. 2).
185
Figure1. Locationmap.
Figure2. Typical ditchstructurecross-section.
1.2 Project Issues
The ground of the construction site is composed of
Holocenedeposit, Pleistocenedeposit andalternating
layersof sandandcohesivesoil.Aquiferiscompressed
on the waterside near the surface. Thus, the ground
condition is relatively inconvenient. Under these
environmental andconstructionworkconstraints, nec-
essarymeasuresfor groundexcavationsuchassubsi-
dence restraint of peripheral ground and prevention
of heaving bottomare necessary. Temporary earth-
retainingwallsareplannedtobeusedfor theopen-cut
method.
Inadditiontotheweaknessof theground, thewidth
and depth of theexcavation arelarge, so large-scale
breastingisrequired.
Variouskindsof studiesarebeingcarriedout with
thegoal of reducingprojectcosts. Oneof whichisthe
reductionof thetemporaryearth-retainingwalls.
Onemeasurefor constructingearth-retainingwalls
economically is shortening the design embedment
length.
In particular, in terms of designing the tempo-
rary earth-retainingwalls, groundparameter settings
greatly influence the earth pressures of both action
andreaction, andthestability of theexcavationbase.
Therefore, proper evaluationof soil survey results in
settinggroundparameters greatly contributes tohow
economical theproject is.
This report presents thestudy results ontheratio-
nalizationof earth-retainingwallsdesignbyfocusing
on ground parameters (cohesion, c, angle of shear-
ingresistance, , andmodulusof deformation, E) and
the appropriate evaluation of natural ground having
confirmeditssafety.
2 BACKGROUNDOF STUDY
2.1 Characteristics of section
Figure 3 shows soil components in vertical section.
Thecharacteristicsareasfollows.
Thissectionconsistsof:
(1) alternatinglayersof variouskindsof stratum.
(2) soft alluvium deposit up to the vicinity of the
structuresdeck.
(3) ahighgroundwaterlevel andawatersidethatrises
near thesurface.
2.2 Conventional method for setting ground
parameters
Groundparameterscanbeestimatedfromlaboratory
soil test aside from using the N-value of standard
penetrationtest results.
In most cases the shearing resistance angle of
cohesivesoil and thecohesion of sandy soil arenot
considered in a design. Instead of using the results
of triaxial compression tests, safevalues for ground
parameterssuchas =0for cohesivesoil andc=0for
sandysoil areapplied.
Thegroundparameters aredividedinto fiveparts
asshowninTable1. Thesepartsthat characterizethe
soil stratumaretheplateau, theseaerosionplatform
inthenorth, thesanddune, theseaerosioninthesouth
andtheundergroundtrough.
(1) Thecohesionof cohesivesoil istheaveragevalue
of triaxial compressiontest resultsobtainedfrom
boring log sampling. Cohesion is assumed to be
zerofor sandysoil.
(2) Theangleof shearingresistanceof sandy soil is
alsoset as theaveragevalueof triaxial compres-
siontest results. Theangleof shearingresistance
of cohesivesoil isassumedtobezero.
(3) Themodulus of deformation is theproduct of a
constantandN-value(28N) for bothcohesivesoil
andsandysoil.
Although the ground parameters set according to
the above method provide a safe design in terms of
strength, there is roomfor improving its economic
impact.
In order to confirmthat this tendency exists, we
investigated the actual behavior of earth-retaining
walls.
The test work of the open-cut method (approxi-
mately90m), intheseaerosionplatforminthenorth,
186
Figure3. Soil componentsinvertical section.
Table 1. Conventional method for setting ground
parameters.
Angleof shear Modulusof
Soil Cohesionc resistance deformationE
Sandy Triaxial Standard
soil compression penetration
test test (28N)
Cohesive Triaxial Standard
soil compression penetration
test test (28N)
Table2. Groundparametersfromtest site.
Soil type c(kN/M
2
) (
) F (MN/m
2
)
Ap 20.0 0.0 6.8
Ac2 25.0 0.0 8.0
As2 0.0 35.0 9.2
Dc2 125.0 0.0 80.0
Ds2u 0.0 35.0 124.0
is under construction for the purpose of verifying
theconstructionmethod, gaininganunderstandingof
possibleproblems duringconstruction andapply the
findingstoactual designandconstruction.
Displacement of earth-retainingwallsismeasured
andcomparedtothedisplacementcalculatedusingthe
past groundparameters(seeTable2) set accordingto
themethodmentionedearlier (seeFig. 4).
The result was that the calculated displacement
according to the conventional method exceeds the
measuredvalue. It is thus confirmedthat theground
parameters used in the design are sufficiently safe
values.
Therefore, the authors believe that it would be
economical to establish new values for the ground
parameters. Economical values aresuggested herein
bystudyinggroundparametersthatareclosetonatural
groundandconfirmingsaferesults.
3 COHESIONOF PLEISTOCENE SAND
3.1 Composition of the soil
Thisstudyprovidesgroundparametersusedindesign
andconstruction. Inorder to obtainpractical values,
thisstudyfocusesonsoil compositionandparameters
that haveanimpact onearth-retainingwallsdesign.
It wasfoundthat thedesignresult wassensitiveto
thecohesionof sixPleistocenelayers(i.e., Ds1, Dc1,
Ds2, Ds2u, Dc2, Ds21) near thelevelingstructure.
3.2 Triaxial compression test results
Althoughcohesionisgenerallyignoredinsandysoil,
therearecases of Pleistocenesandlayer that exhibit
cohesion.
Thetriaxial compressiontest resultsgatheredfrom
thesitewererearrangedinorder toexaminewhether
cohesioncouldbeconsidered.Cohesionisfoundusing
afailureenvelopedrawnusingleast squares method
of Mohrcirclesobtainedfromthetriaxial compression
test.
Asanexample, thetriaxial compressiontest result
under theCDconditionof thetest sitesDs2ulayer is
presentedinFig. 5.
187
-25.0
-23.0
-21.0
-19.0
-17.0
-15.0
-13.0
-11.0
-9.0
-7.0
-5.0
-3.0
-1.0
-40 0 40 80 120 160 200
D
e
p
t
h
(
m
)
Measured value
Conventional design value
Excavation depth
Bs
Ap
Ac2
As2
Dc1
Ds2u
Second strut
Third strut
Final Excavation - Comparison between the Measured
Deformation Value and the Conventional Design Value
Deformation(mm)
Excavation depth GL-15.4m
Fourth strut
First strut
Figure 4. Comparison of earth-retaining walls displace-
ment. (Measured values and calculated values using the
conventional method).
Figure5. Triaxial compressiontestresultof Ds2ulayerfrom
test site.
Resultsreveal that althoughcohesionisignoredin
most cases that involvesandy soil, it is confirmedin
thissite(c=27.8kN/m
2
). PleistocenesandDs1, Ds2u
andDs2l showedsimilar results.
Figure6. Unconfined compression test results of thetest
sitesDs2ulayer.
3.3 Unconfined compression test results
Figure6showstheunconfinedcompressiontestresult
of test sites Ds2u layer. Unconfined compressive
strength was confirmed in both saturated and par-
tially saturatedsamples. This result also implies that
cohesionexistsinsandysoil of thetest site.
4 RESETTINGOF ZONES
4.1 Reclassification stratum
The stratum classification of the test site is rear-
rangedintoninezonesbasedonthegroundsformation
processandparticlesizedistribution(seeFig. 7).
4.2 Reclassification of soil composition
Inadditiontocohesivesoil andsandysoil, intermedi-
atesoil whichexhibitscharacteristicsof bothcohesive
soil andsandy soil istakenintoconsiderationduring
classification.
Soil isclassifiedaccordingtosandcomponentrate,
finescontent andplasticityindex. Basically, cohesive
soil isdefinedtohavesandcontent of lessthan50%,
fines content of 50%or moreand aplasticity index
of 30 or more; intermediate soil is defined to have
sandcontent of 50%to80%, finescontent of 20%to
50%andaplasticityindexof lessthan30; sandysoil
is definedto havesandcontent of 80%or moreand
granulecontent of lessthan20%(seeFig. 8).
5 STATISTICAL APPLICATIONOF TRIAXIAL
COMPRESSIONTEST RESULT (COHESION)
Usually, groundparameters areset statistically using
the average value of triaxial compression results of
boringsamplesexceptoutliers. However, thisrequires
statistical application which considers ground and
spatial dispersions.
Boring sampling can be done before excavation.
Thus, several tests were conducted and much data
was accumulated (approximately 140 UU or uncon-
solidatedundrainedtestsand190CDor consolidated
drainedtests).Thisstudyproposesnewgroundparam-
etersbasedonthesedata.
188
Figure7. Newsoil composition.
Figure8. Soil component classificationmethod.
The new design value for cohesion is calculated
according to thefollowing equation using mean (m)
andstandarddeviation() of data.
Figure9Showsanexampleof settingdesignvalue.
6 COMPARISONOF TRIAXIAL
COMPRESSIONTEST VALUE OF BLOCK
SAMPLINGANDDESIGNVALUE
Thetriaxial compressiontestresultsof blocksampling
takenfromthesitearecomparedtocalculateddesign
valuesmentionedearlier.
High-qualityundisturbedsamplescanbecollected
by block samplingbecausesoil iscut andtakenfrom
thegroundinclumps. Thetest resultsobtainedherein
are believed to more accurately evaluate the ground
cohesion.
Figure9. Exampleof designvalue(Ds2u(s)).
Figure 10 shows the model for soil composition
in vertical section of thetest site. In this study, Dc1
is divided into two layers, cohesive soil and sandy
soil, based on results of grain size analysis. Block
sampling is conducted using Dc1, Dc1 (sand) and
Ds2u soils. Comparison diagramof statistic values
189
computed according to Equation 1 using the boring
sampling triaxial compression test and the result of
theblocksamplingtriaxial compressiontest isshown
inFigures11and12.
Figure11showstheresultforsandysoil of Dc1.Tri-
axial compressiontestswereconductedunder CDand
CU (consolidated undrained) conditions sincesandy
soil isunder consideration. Resultsreveal thatthesta-
tistical valueof boringsamplingandthetest valueof
blocksamplingwereclose.
Figure 12 shows the results for intermediate soil
Ds2u. Triaxial compression tests were conducted
under all conditions (i.e., UU, CD and CU), since
intermediatesoil exhibitscharacteristicsof bothcohe-
sive soil and sandy soil. Results reveal that all test
valuesof blocksamplingexceedstatisticvaluesof bor-
ingsampling. Therefore, it isthought that thestatistic
valueof boringsamplingproducesafer designresults
comparedtotest valuesof blocksampling.
Figure10. Soil componentsinvertical sectionof site.
Figure11. Comparisonof blocksamplingandboringsamplingof sandylayer Dc1(C).
7 SHEARINGRESISTANCEANGLEAND
DEFORMATIONMODULUS
Shearing resistance angle and deformation modulus
hasnosignificanteffectondesignbasedontheresults
of the sensitivity analysis. Their new values are set
followingtheproceduresthat aresimilar tothosethat
wereemployedfor settingcohesion. Itisassumedthat
thesamefactors as usedfor settingthecohesioncan
be considered in soil stratumclassification and soil
compositionclassification. Shearingresistanceangle
anddeformationmodulusforeachsoil issetasfollows.
7.1 Shearing resistance angle
Inthecaseof sandysoil andintermediatesoil, statis-
tic application of triaxial compression test of boring
samplingisconductedusingEquation1. For cohesive
soil, =0is assumedsinceits sensitivity is small in
earth-retainingwallsdesign.
7.2 Deformation modulus
Statisticapplicationof triaxial compressiontestresults
of boringsamplingisconductedusingEquation1for
all typesof soil.
Although the deformation modulus can be deter-
mined using N-values and unconfined compression
testresults, inthisstudyitissetusingtriaxial compres-
siontests, similar tocohesionandshearingresistance
angle.
8 EVALUATIONOF GROUNDSTRENGTH
DEGRADATION
The ground inside the earth-retaining walls is exca-
vatedinorder tobuildthestructure(seeFig. 13). As
excavation continues, covering pressureis unloaded,
190
the floor swells and eventually ground strength
reduces. Swellingtestsareconductedinordertodeter-
minethestrengthreductionof ground. Inthistest, CU
test is executedafter leavingthesaturatedsampleto
swell.
Thedegreeof strengthreductionisstudiedbycom-
paringtheinitial strengthintermsof thepeakdeviator
stress (q0w) obtained fromthe CU test of the non-
swelled sample with those (q) of the swelling test
results. The relationship of the normalized strength
and elapsed timeis shown in Figures 14 and 15 for
cohesivesoil andsandysoil, respectively.
With respect to initial strength, cohesive soil
reduced its strength to approximately 50%after one
to two weeks. On theother hand, sandy soil showed
almost nostrengthreduction.
Therefore, thepassiveshearingresistanceof cohe-
sivesoil (excavationside) reduceswithelapsedtime.
Heretheshearingstrengthis equal tocohesionsince
shearingresistanceangleisnotconsidered. Thecohe-
sionusedindesignisset to50%of thestatisticvalue
sincethenormalizedshearingstrengthalsoconverges
to50%.
Thestatisticvalueforsandysoil isnotreducedsince
nosignificant strengthreductionwasfound.
Cohesion is reduced to 50%, similar to cohesive
soil, for thesafedesignof intermediatesoil.
9 ESTABLISHMENT OF NEWGROUND
PARAMETERS
Basedontheaboveresults, newgroundparametersin
thevicinityof thestructuresfloor areset below.
1) Groundparametersareset accordingtotheresults
of boringsamplingtriaxial compressiontests.
Figure12. Comparisonof blocksamplingandboringsamplingof sandylayer Ds2u(I).
2) Soil isclassifiedintothreetypes, i.e. cohesivesoil,
intermediatesoil andsandysoil.
3) Cohesion and deformation modulus of cohesive
soil uses thevaluecalculated according to Equa-
tion1. Inthiscase, cohesioninthepassivesideis
assumedtobe50%of that intheactivesidesince
Figure13. Illustrationof groundexcavation.
Figure14. Strengthreductionof Dc1soil.
191
Figure15. Strengthreductionof Ds2usoil.
Figure16. Comparison of earth-retaining walls displace-
ment. (Actual values vs. estimated values of new ground
parameters)
cohesion decreases with time after excavation.
Shearingresistanceangleisnot considered.
4) Cohesion and the deformation modulus of inter-
mediatesoil usesthevaluecalculatedaccordingto
Equation1. Inthiscase, cohesionisnotreducedin
thepassiveside.
5) Cohesion, shearingresistanceangleanddeforma-
tionmodulusof sandysoil usesthevaluecalculated
accordingto Equation1. Inthis case, cohesionis
not reducedinthepassiveside.
Theactual displacementof testsitesearth-retaining
walls and theestimated displacement using thenew
groundparametersarecomparedinFig. 16.
Both displacement modes are close and almost
similar in behavior. However, at the top of earth-
retaining walls, the displacement according to new
ground parameters is safer. Therefore, it can besaid
that thenewgroundparametersproperlyevaluatethe
natural groundconditionsandsecuresaferesults.
10 CONCLUSIONS
Thefollowingconclusionsaredrawnfromtheresults
of the study on the cohesion evaluation method for
design using statistical application of triaxial com-
pression test results of boring sampling and block
sampling.
1) Theexistenceof cohesioninPleistocenesandwas
identifiedfromthetestsconductedherein.
2) Design value is set using the mean and standard
deviation of triaxial compression test data taken
fromseveral boringsamples.
3) Thecohesioninthepassivesideisreducedto50%
forsafedesignconsideringthereductionof ground
strength with timeafter excavation. Based on the
comparative study of actual behavior of the test
sitesearth-retainingwallsandthecalculatedvalue
of thedesign, both almost coincided and thecal-
culatedvalueissafesinceit exceedsthemeasured
value.
4) Detailed soil classification is established using
finescontentandplasticityindex, andcorrespond-
inggroundparametersweresuggested. Theresults
of earth-retainingwallstrial designusingthesug-
gested ground parameters imply that penetration
length and steel quantity can be reduced and so
that theconstruction cost of earth-retaining walls
canbereducedbyapproximately20%.
REFERENCES
J apan RoadAssociation, 1993. Road earthwork: Guide on
Temporary Structures Engineering (inJ apanese).
The J apanese Geotechnical Society, 1992. Geotechnical
notes: Two intermediate soils, sand or clay (inJ apanese).
192
Systemreliabilityof slopesfor circular slipsurfaces
J ianyeChing
Yu-GangHu
Kok-KwangPhoon
ABSTRACT: Evaluating thereliability of aslopeis achallenging task becausethepossibleslip surfaceis
not knownbeforehand. Approximatemethods viathefirst-order reliability method(FORM) provideefficient
ways of evaluating failureprobability of themost probable failuresurface. Thetradeoff is that thefailure
probability estimatesmay bebiasedtowardstheunconservativeside. TheMonteCarlosimulation(MCS) isa
viableunbiasedway of estimatingthefailureprobability of aslope, but MCS is inefficient for problems with
small failureprobabilities. This study proposes anovel way based on theimportancesampling techniqueof
estimatingslopereliabilitythat isunbiasedandyet ismuchmoreefficient thanMCS. Inparticular, thecritical
issueof thespecificationof theimportancesamplingprobability density functionwill beaddressedindetail.
Several numerical examplesareinvestigatedtoverifytheproposednovel approach. Furthermore, thesimplified
Bishopmethodof slicesistakenastheslopestabilitymethodfor thisexample.
1 INTRODUCTION
Slopes with nominal safety factor more than 1 are
not necessarily safe, becauseof underlying geotech-
nical variabilities and inherent uncertainties in the
predictive methods. Traditionally, this limitation is
well recognized, buthandledbyempirical prescriptive
rules,suchasimposingafactorof safetyof 1.2fortem-
poraryslopesand1.5for permanentslopes. However,
theprescribedfactorsof safetydonot changeregard-
lessof thedegreeof variabilitiesand/or uncertainties
intheproblem. Reliability, namely oneminusfailure
probability, mitigates this limitation. Over theyears,
analysismethodshavebeenproposedtoevaluatereli-
abilityof slopes, e.g.: Vanmarcke(1976), Chowdhury
andXu(1993), Christianet al. (1994), LowandTang
(1997), Lowet al. (1998), Bhattacharyaet al. (2003)
andGriffithsandFenton(2004). However, evaluating
reliabilityof aslopeisnot aneasytask, primarilydue
tothelackof knowledgeof thefailuresurface.
A standard procedureof evaluating thereliability
of aslopeis throughMonteCarlosimulation(MCS)
(Rubinstein1981;AngandTang1984).Thisapproach
was taken by Griffiths and Fenton (2004) wherethe
MCS method is implemented with a random field
model for spatial distributionof shear strengths. The
procedureforMCSisstraightforward. LetZ denoteall
uncertainvariablesintheslopeof interest.Withoutloss
of generality, let usassumeZ isindependent standard
Gaussian. Inthecasethat Z isnot standardGaussian,
proper transformations can be taken to convert the
problemintothestandardGaussianinputspace.Monte
Carlosimulationcontainsthefollowingsteps:
a. DrawZ samples {z
i
: i =1, . . . , N}fromthestan-
dardGaussianPDF.
b. For eachsamplez
i
, conduct adeterministic slope
stability analysistofindthemost critical slipsur-
faceamongall trial surfaces. If thesafetyfactor of
themost critical surfaceis less than 1, theentire
slopeisconsideredtofail for that z
i
sample.
c. Repeat Step2for i =1, . . . , N. Theaverageof the
failureindicatorsissimplyanestimateof thefailure
probabilityof theslope.
Mathematically, the MCS procedure can be sum-
marizedasthefollowingequation:
whereF is thefailureevent of theentireslope; I [.]
is theindicator function; O is theset of all trial slip
surfaces; is oneof thetrial surface, andFS
is the
safety factor for that trial surface, which is clearly a
functionof Z.
Anotherwayof interpretingtheoverall failureprob-
ability in (1) is as follows: let F
denote the event

FS
(Z)-1, i.e.: thefailureevent of thetrial surface

.Theoverall failureeventF issimplytheunionof all
individual failureevents
O
F
. This interpretation
is graphically depicted in Figure1. Theoverall fail-
ureprobabilityisthereforethevolumeunder f (z) (the
standardGaussianPDF) withintheF region.
193
Figure 1. Illustration of various failure regions in the
standardGaussianspace.
The MCS method provides an unbiased estimate
of the actual failure probability. However, it can be
verytimeconsuming, especiallyfor slopeswithsmall
failureprobabilities. Thisisbecausethecoefficientof
variation(c.o.v.; standarddeviationdividedby mean
value) of theMCSestimator P
MCS
F
isequal to
where (.) denotes the c.o.v. Therefore, in order to
makethec.o.v. of theMCS estimator to beas small
as 30%, it roughly requires 10/P(F) MCS samples,
i.e.: 10/P(F) deterministicslopestabilityanalyses.
A more efficient method based on the first-order
reliability method(FORM) wasproposedtoestimate
the failure probability. The idea is to solve for the
followingoptimizationproblem:
Notethattheslipsurfacevariable isaugmentedinto
theoriginal FORMoptimizationproblem.Letthesolu-
tion of theoptimization problembez
and
, then
+(||z
||) istheestimatedfailureprobabilityand
isthemostprobableslipsurface.Thisapproachwas
takenby LowandTang(1997) andLowet al. (1998)
for theevaluationof slopefailureprobability.
The FORM technique is efficient and convenient
because the repetitive deterministic slope stability
analyses requiredby MCS arenot needed. However,
thetradeoff is that theFORM methods may provide
biasedandunconservativeestimatesof theactual over-
all failure probability. This can be seen in Figure 1,
where the thick curve indicates the limit state func-
tion of themost probableslip surface
. It is clear
that the volume under f (z) within the union region
O
F
(the actual overall probability) is always

greater thanor equal tothatwithintheregionF
(the
FORM-estimated failure probability) because the
latter isasubset of theformer.
What ismissingintheliteratureisatechniquethat
can provide unbiased estimate of the actual overall
failureprobabilityandyet onlyrequiresasmall num-
ber of repetitivedeterministicslopestabilityanalyses.
Thepurposeof this study is to demonstratethat it is
possibleto implement theimportancesampling (IS)
technique(Rubinstein1981) toachieveso. Moreover,
it is shown by examples that thec.o.v. of thefailure
probabilityestimator madebytheIStechniquecanbe
assmall as0.2withonly100deterministicslopestabil-
ity analyses for practical rangeof failureprobability.
In thepaper, thediscussion of theIS techniquewill
bemadein thecontext of circular trial surfaces and
methodof slicesalthoughitsuseisobviouslynotlim-
itedtothisscenario.Thelimitationof theIStechnique
will alsobeaddressed.
2 METHODSOF SLICES
Some popular methods of determining the safety
factorforagiventrial slipsurfaceandfixedsoil param-
eters arebriefly reviewed. This presentationdoes not
aimto give a complete review of the slope stability
methods but just to giveenough background for the
forthcomingpresentationof theimportancesampling
technique. Moreover, the presentation is limited to
methodsof sliceswithcircular trial surfacesalthough
the importance sampling technique may apply to
general methodsandnon-circular trial surfaces.
2.1 Ordinary method of slices
AsshowninFigure2, acircular trial surfaceisunder
consideration, and the soil parameters Z are fixed
at some values (e.g.: for MCS, they are fixed their
sampledvalues for eachdeterministic slopestability
analysis).Thegoal istodeterminingthesafetyof apre-
definedslipsurfaceindexedby.Asimplifiedmethod
that assumesnointeractingforcesbetweenslicescan
beusedtocomputethesafetyfactor:
wherec
k
and
k
arethecohesionandfrictionangleof
thek-thslice; W
k
isthetotal weight of thek-thslice;
u
k
istheporewater pressureatthemiddlepointof the
slicebottom; Ll
k
isthelengthof theslicebottom;
k
is theinclinationangleof theslicebottom. Equation
(4) is obtained based on theequilibriumequation of
theoverall momentfor all slicesaboutachosenpoint.
Thismethodof determiningthesafetyfactor iscalled
theordinarymethodof slices(OMS) (Fellenius1936).
TheOMSisthesimplestmethodamongall methods
of slices. It does not satisfy forceandmoment equi-
libriumof individual slices, and it usually provides
conservativeestimatesof safety factors. It isalsothe
only method of slices that does not require iterative
calculations to obtain the safety factor. In practical
applications, many trial surfaces are randomly gen-
erated, and the trial surface with the smallest safety
194
Figure2. Illustrativeslopeexample.
factor isthecritical slipsurface. If thesafetyfactor of
thecritical slipsurfaceislessthan1, theentireslope
isthenconsideredasunstable, or tohavefailed.
2.2 Simplified bishop method of slices
The simplified Bishop method of slices (SBMS)
(Bishop 1955) takes a different assumption that all
inter-slice forces are horizontal. Equilibriumof ver-
tical forcesinall slicesandtheoverall moment gives
thefollowingexpressionfor thesafetyfactor:
The safety factor FS
can be found by iteratively

solving(5). Again, many trial surfaces arerandomly
generated, andthetrail surfacewiththesmallestsafety
factor isthecritical slipsurface. If thesafetyfactor of
thecritical slipsurfaceislessthan1, theentireslope
isconsideredtohavefailed.
2.3 Transformation to standard Gaussian space
Previously, it is assumed that all uncertain variables
aretransformedtothestandardGaussianinput space
for theeaseof presentation. To demonstratehowthe
expressionsof safety factorscanbetransformedinto
thestandardGaussianspace, consider theexamplein
Figure2, wherethetrial surfaceis, bothsoil layers
arehomogeneous, andthec, and parameters for
bothlayers areuncertainandindependent. Notethat
thefirst fewslicessharethesameshear strengthsand
unit weights, andsimilar for thelast fewslices. Inthe
casethat theOMS is taken, thesafety factor can be
expressedas:
wherec
n
,
n
and
n
arethecohesionand(average) unit
weightof then-thsoil layer;
n
=tan(
n
)isthetangent
of thefrictionangleof then-thsoil layer; thesumma-
tionoverknindex(n=1or2) isforthesliceswhose
bases arewithinthen-thsoil layer; h
k
is theaverage
height of the k-th slice. We assume without loss of
generalitythat theslicebasesliefullywithinonesoil
layer. This is always achievable by appropriate divi-
sion of slices. Equation (6) can be transformed into
standard Gaussian space by proper transformation,
e.g.: Rosenblatts transformation (Rosenblatt 1952).
The transformed safety factor will be denoted by
FS(Z), andthestandardGaussianrandomvariablecor-
respondingtoc
1
will bedenotedbyZ
c1
andsimilarfor
other randomvariables.
2.4 Approximate location of the design point of a
trial surface
Given the trial surface , an approximate way of
finding its design point, denoted by z
, is proposed.
Finding z
is highly non-trivial because most slope

stabilityanalysismethodsareiterative,e.g.:thesimpli-
fiedBishopmethod. Usually, anoptimizationproblem
similar toEq. (3) needstobesolved(with fixedat
theprescribedtrial surface) inordertofind z
.Among
theslopestability analysis methods, theOMS is the
only non-iterative method. Therefore, it is proposed
in this research to implement the OMS to find the
approximatelocationof thedesignpoint z
.
Onecanre-writeEq. (4)asthefollowingOMSlimit
stateequation:
where g
OMS
is the OMS limit state function for the
failureeventof thetrial slipsurface: if g
OMS
-0, the
surfacefailsunder theOMScriterion, andviceversa.
Thecrucial assumptionshereare: (a) thegeometryof
thecritical slipsurfacedoesnot change, althoughthe
soil parametersareallowedtovaryand(b)b
1
andb
2
are
constants givenby therespectivemeanunit weights.
Theseassumptionsareobviouslyerroneous,butweare
onlylookingfor anapproximatelocationof theactual
designpoint z
.
Let us consider thesoil parameters c
1
, c
2
,
1
,
2
,
2
,
2
tobeuncertain. Notethatif c
1
, c
2
,
1
,
2
and
2
,
2
areindependent Gaussian, thelimit statefunction
inthestandardGaussianspacecanbeapproximatedas
wherej and arethemeanvalueandstandarddevi-
ationof thesubscriptedvariable. Bydoingso, Eq. (8)
isnowalinearfunctionof thestandardGaussianinput
Z. It is soon clear that theOMS design point in the
standardGaussianspaceis
195
where
istheOMSreliabilityindexof thetrial surface.This
OMSdesignpoint
Note that the derivations of the design point in
Eq. (9) are based on the following assumptions:
(a) c
1
, c
2
,
1
,
2
and
2
,
2
areindependent Gaussian
and(b)theOMSistheadoptedmethodof slices.Those
assumptions aretakenbecausetheresultingapproxi-
matesolution z
OMS
of theactual designpoint z
has
a simple analytical form. Clearly, such an approxi-
mate design point z
OMS
in general is not the actual

designpoint z
becausefortheactual applicationthose
assumptions may not betrue, i.e.: theactual adopted
slopestability method may not betheOMS and the
soil parametersmaynotbeindependentGaussian. Itis
alsonotequal to z
evenunderassumptions(a)and(b),
becauseEq. (8) iserroneous. Nonetheless, it isfound
empiricallythat theapproximatedesignpoint z
OMS
is
usually reasonably closetotheactual designpoint z
even when the two assumptions are violated in the

actual application. At theend, theISmethodwill pro-
videunbiasedfailureprobabilityestimatesregardless
of theaccuracyof thedesignpoint.
3 IMPORTANCE SAMPLINGTECHNIQUE
As mentioned earlier, reliability of a slope is not
component-wisebut system-wise. Therefore, finding
thelocationof thedesignpointforasingleslipsurface
doesnot sufficetodeterminethesystemreliabilityof
aslope. AlthoughMCS isaviablemethod, it iscom-
putationally inefficient. This is primarily due to the
fact that the high probability-density region of f (z),
i.e.: standard Gaussian PDF, is quite far from the
failureregionF =
, especially whentheactual
failureprobabilityissmall.Thepractical consequence
is that MCS requires many samples beforeafailure
sampleisobtained.
Thebasicideaof theproposedimportancesampling
methodistoadopt theso-calledimportancesampling
PDF (IS PDF) q(z) whose high probability-density
region is closer to thefailureregion F. In this study,
thefollowingmixtureof MGaussianPDFsistakenas
theISPDF:
where
j
isthej-thslipsurfaceamongtheM repre-
sentativeslipsurfaces; d isthedimensionof z; z
OMS
is the center of the j-th Gaussian PDF, taken to be

the approximate design point z
OMS
of the j-th rep-

resentativeslipsurfaces; w
j
is theweight of thej-th
GaussiansPDF.
Asdiscussedlater, theseM representativeslipsur-
faces canbedeterminedby usingtheOMS although
theadoptedstabilitymethodfor definingfailuremay
notbetheOMS. SincetheOMSisnon-iterative, sothe
determinationof theseM representativeslipsurfaces
doesnotrequiremuchcomputation.Thedetermination
of theMrepresentativeslipsurfacesandtheir weights
will bediscussedlater indetail. Theprinciplehereis
tochoosetheM representativeslipsurfacessuchthat
thehighprobability-densityregionof q(z) iscloser to
thefailureregionF.
The IS technique is based on the following
observation:
where {z
i
: i =1, . . . , N} are independent samples
fromq(z); E
f
and E
q
denote the expectations with
respecttothePDFsf (z) andq(z), respectively. There-
fore, theISsimulationcontainsthefollowingsteps:
a. DrawZ samples{z
i
: i =1, . . . , N}fromq(z).
b. For eachsamplez
i
stability analysis to find the critical slip surface
amongall trial surfaces. If thesafety factor of the
most critical surface is less than 1, the slope is
consideredtofail for that z
i
sample.
c. Repeat Step2for i =1, . . . , N. TheISestimateof
theoverall failureprobabilityissimply
Note that P
IS
F
is an unbiased estimator of the
actual overall failure probability regardless the
choice of {(
j
,w
j
):j=1,, M}. More importantly, if
{(
j
, w
j
):j =1, . . . , M}iscarefullychosensothatthe
highprobability-density regionof q(z) isclosetothe
failureregionF =
, thec.o.v. of P
IS
F
canbequite
small. Thec.o.v. of theestimator P
IS
F
issimply
3.1 Choice of {(
j
, w
j
):j =1, . . . , M}
Although P
IS
F
is unbiased regardless the choice of
the representative slip surfaces and their weights
{(
j
, w
j
) : j =1, . . . , M}, thechoicemay affect the
efficiencyof theISestimator. Anideal choiceissuch
that the high probability-density region of q(z) can
cover thefailureregionwithhighprobabilitycontent,
i.e.: cover all failuremodes.
For instance, if thereisonlyonefailuremodeinthe
slopeof interest, asensiblechoiceistolet M equal 1,
1
bethemost probablefailuresurfaceandw
1
equal
1.0, i.e.: theIS PDF q(z) is simply aGaussian PDF
centeredat thedesignpoint of themost probablefail-
ure surface. In the case where there are two failure
modes, it ispreferabletolet M equal 2,
1
and
2
be
thetwomost probablefailuresurfaceswithinthetwo
196
failuremodes, andw
1
andw
1
equal thefailureproba-
bilitiesof
1
and
2
, respectively, i.e.: theISPDF q(z)
isthemixtureof twoGaussianPDFsweightedbythe
failureprobabilitiesof
1
and
2
andcenteredat the
designpoints of
1
and
2
. However, findingdesign
points is itself achallengingtask especially whenan
iterativestability method(e.g.: simplifiedBishop) is
adoptedtodefinefailures.
Nevertheless, theISmethodisunbiasedregardless
thechoiceof {(
j
, w
j
):j =1, . . . , M}.Thisimpliesthat
a rough estimate of design points may suffice. The
key ideain this paper is thereforeto obtain aset of
approximate design points through the OMS, called
theOMSdesignpoints, althoughtheadoptedstability
analysismethodmaybethesimplifiedBishopmethod.
BecausetheOMSproducesconsistentlyconservative
safety factors, it is expected that the resulting q(z)
can still cover thefailureregion to acertain degree.
Thedeterminationof theOMSdesignpointsdoesnot
requiremuchcomputationbecausetheOMS is non-
iterative. Theproposedsimpleprocedurebasedonthe
OMSisdescribedasfollows.
First of all, acircular trial slipsurfacecanbechar-
acterized by three variables, u, v and in Figure 2,
where u and v are the horizontal coordinates of the
two points where the slip surface and the ground
surface intersect, and is the tangent angle of the
slipsurfaceat thex
low
location. Theupper andlower
bounds of thethreevariables, denotedby [u
low
, u
up
],
[v
low
, v
up
] and [
low
,
up
], should bespecified apri-
ori based on engineering judgments. As aresult, the
spaceof slipsurfaces is thethree-dimensional space
{u
low
uu
up
, v
low
vv
up
,
low

up
}. Sec-
ondly, equally spacing grid points arelocated in the
{u, v, }space, e.g.: if theuaxisisdividedintothree
equally spacinggridpoints {u
low
,(u
low
+u
up
)/2, u
up
}
andsimilarfortheothertwoaxes, therewill be3
3
=27
equallyspacinggridpointsinthe{u, v, }space, and
eachgridpoint representsatrial slipsurface.
Thetrial surfacescorrespondingtothesegridpoints
are directly taken as the representative slip surfaces
{
j
:j =1, . . . ,M}except theinadmissibleones, e.g.:
trial surfaces intersecting the ground surface. It is
hopedthat all failuremodes canbedetectedthrough
thesewidelyspreadingrepresentativesurfaces. Equa-
tion(10) canthenbeusedtofindtheirOMSreliability
indices, denotedby {
j
:j =1, . . . ,M}. Theweight of
thej-thetrial surface
j
is simply taken as its OMS
failureprobability, i.e.: w
j
=+(
j
). Noticethat the
calculations for determining {(
j
, w
j
):j =1, . . . ,M}
are fast even for a large number of trial slip sur-
faces becauseonly thenon-iterativeOMS equations
areinvolved.
4 PROCEDURE OF PROPOSEDAPPROACH
Theprocedureof theproposedIS methodis summa-
rizedasfollows:
a. Divid the{u,v,} spaceinto equally spacing grid
points to obtain the admissible representative
slip surfaces {(
j
, w
j
):j =1, . . . , M}. Use Eqs.
(9) and (10) to find their OMS design points
{ z
OMS
j
: j =1, . . . , M} and reliability indices
{
j
: j =1, . . . , M}, and let the weights be
w
j
=+(
j
).
b. Draw Z samples {z
i
:i =1, . . . ,N} fromq(z), the
mixture of M Gaussian PDFs centered at { z
OMS
j
:
j =1, . . . , M}weightedby{
j
:j =1, . . . , M}.
c. For eachsamplez
i
stabilityanalysis(notnecessarilytheOMS) tofind
thecritical slipsurfaceamongall trial surfaces. If
thesafety factor of thecritical surfaceislessthan
1, the slope is considered to have failed for that
z
i
sample, i.e.: I [minFS
(z
i
)-1]=1; otherwise
I [minFS
(z
i
)-1]=0.
d. Repeat Step4for i =1, . . . , N. TheIS estimateof
theoverall failureprobabilityissimply
5 NUMERICAL EXAMPLE
Example1AnexampleinSTABL manual
Consider thefollowingexampleextractedfromthe
STABL user manual (Siegel 1975): theslopeinFig-
ure 3 underlain by a hard layer. The shear strength
parameters c and of the soil are uncertain. It is
assumed that tension cracks arepresent near thetop
surfaceof theslopeuptoadepthof 3.35m.Theuncer-
taincohesionc is lognormally distributedwithmean
j
c
=24kN/m
2
andc.o.v.
c
=20%, while=tan()
Gaussian with mean j
=0.176 (corresponding to
j
=10
o
) and standard deviation
=0.018. There
isanegativecorrelationof 0.3betweenZ
c
andZ
.
ThesimplifiedBishopmethodof slicesistakenasthe
slopestabilitymethodfor thisexample.
Themainpurposeof thisexampleistodemonstrate
theadvantageof theISmethodover theMCSmethod
inevaluatingsmall slopefailureprobability.TheMCS
method with sample size N =1000 is taken to esti-
matethefailureprobability of theslope. For each Z
sample, aglobal searchalgorithmistakentofindthe
mostcritical slipsurface, i.e.: theslipsurfacewiththe
least Bishop-basedsafety factor, thenEq. (1) canbe
employedtofindP
MCS
F
. Figure4showsthelocations
of all MCS samples in thestandard Gaussian space,
Figure 3. The slope considered in the first example. The
greylinesaretheadmissiblerepresentativeslipsurfaces.
197
Figure4. TheMCS(left; N=1000) andIS(right; N=100)
samples in thestandard Gaussian space(j
=0.176). The
grayregionistheactual failureregionF, andthetrianglesare
thelocationsof theOMSdesignpointsof therepresentative
slipsurfaces.
whereao markindicatesanon-failuresample, while
ax indicatesafailuresample.Onecanseemostof the
samplesarenon-failureones. Theanalysisshowsthat
P
MCS
F
is0.096andthec.o.v. of theestimator is9.7%.
Asaverification, theactual failureregionFisfound
by an exhaustiveanalysis described as follows. First
divide the standard Gaussian space into dense grid
points, eachgridpoint correspondingtoaset of fixed
values of c and . At each grid point, a simplified
Bishop stability analysis is taken to see if the slope
fails(again, aglobal searchalgorithmishereintaken
tofindtheslipsurfacewiththeleastsafetyfactor). Do
sofor all gridpointstoobtaintheactual failureregion
F.ThegrayregioninFigure4istheactual failureregion
F foundbytheexhaustiveanalysis.
FortheIStechnique,therepresentativeslipsurfaces
arefound by dividing the{u,v,}spaceinto 3
3
=27
equallyspacinggridpoints. Amongthem, 21of them
areadmissible, andthey areplottedinFigure3. The
IS PDF thereforeconsists of 21GaussianPDFs cen-
teredat their OMSdesignpoints, andtheweightsare
their OMSfailureprobabilities. Noticethat thecalcu-
lationsofarisfastbecauseonlythenon-iterativeOMS
equationsareinvolved.
Thelocationsof theOMSdesignpointsfor therep-
resentativesurfaceareshownasthetrianglesinFigure
4. Therearesomedesignpointsthatareveryfar away
fromthe actual failure boundary, e.g.: the left-most
triangle in the figure (the other such design points
cannot be seen because their horizontal coordinates
areless than5). Thesedesignpoints correspondto
therepresentativeslipsurfacesthat intersect thehard
layer. It is clear that someapproximatedesignpoints
arefairlyclosetotheactual designpoint eventhough
the actual stability method is the simplified Bishop
method, not the OMS, and even though in reality c
and arenot independent Gaussian.
FollowingtheISprocedure, only100samplesof Z
aredrawnfromq(z) toobtainP
IS
F
viaEq.(13). Those
samplesareplottedinFigure4. It isclear that alarge
portionof thesamplesarefailuresamples. Theresult-
ing P
IS
F
is 0.0913, and its c.o.v. is estimated to be
13.2%viaEq(14). To demonstratethesuperiority of
the IS approach for small failure probability, j
is
increased to 0.268 (corresponding to j
=15
) and
Figure5. TheMCS(N=10000) andIS(N=100) samples
inthestandardGaussianspace(j
=0.268).
standarddeviationischangedto
=0.028, andboth
theMCS andIS methods areappliedto estimatethe
correspondingfailureprobabilities.Figure5showsthe
samplesobtainedbythetwomethods,andTable1sum-
marizestheanalysisresults. Forasensiblecomparison
betweenMCSandIS,therequirednumbersof samples
toachieveac.o.v. of 20%areestimatedbasedonthe
rulethat c.o.v. is proportional to 1/
N. Thesenum-
bers showthat theMCS methodbecomes inefficient
whenthefailureprobabilitygetssmaller, whiletheIS
methodseemsmorerobust onthat aspect.
Comparisons with FORM arealso made. Table1
showstheresultsfromtheFORM method, i.e.: solve
Eq. (3)tofindthedesignpoint z
andestimatethefail-
ureprobability as +(||z
||). In this paper, aglobal

search algorithm is adopted for solving Eq. (3) to
avoidlocal solutions. TheadoptedFORM methodis
basedonanalgorithmsimilar tothegeneticalgorithm
in finding the most probable slip surface (e.g.: Xue
andGavin(2007)). Basically, theHasofer-Lindrelia-
bility index
HL
is first foundwiththegenetic algo-
rithmbysolvingaconstrainedoptimizationproblem.
ThentheFORM failureprobability canbeestimated
as +(
HL
). For this example, the adopted FORM
methodperformssatisfactorily, andtherequiredcom-
putational costismuchlessthanthoserequiredbythe
MCSandISmethods.
Example2A slopeintwoclayeysoil layers
Consider theslopeintwoclayeysoil layersunder-
lain by a hard soil layer shown in Figure 6. The
undrainedshear strengthss
u1
ands
u2
of thetwoclayey
layersareuncertainandindependent.Theyarelognor-
mallydistributedwithmeanvaluesj
su1
=120kN/m
2
andj
su2
=160kN/m
2
andwithc.o.v.s
su1
=30%and
su2
=30%. ThesimplifiedBishopmethodof slicesis
takenastheslopestabilitymethodfor thisexample.
The main purpose of this example is to demon-
stratetheadvantageof theISmethodover theadopted
FORM methodinevaluatingslopefailureprobability
whenmultiplefailuremodes arepresent. By solving
Eq. (3), thefailureprobability estimatemadeby the
adoptedFORM methodisfoundtobe0.0016.
For the IS technique, 3
3
=27 representative slip
surfaces aregenerated, among them, 18 of themare
admissible, and they areplotted in Figure6. TheIS
PDF thereforeconsistsof 18GaussianPDFscentered
at their OMS designpoints, andtheweightsaretheir
198
Table1. TheanalysisresultsfromtheMCS, ISandFORM
methodsfor variousvaluesof j
.
MCS IS FORM
Method (
=0.176)
Number of sampleN 1000 100
Computer runtime(minutes) 282 12 1
Estimatedfailureprobability 0.0960 0.0913 0.097
Estimator c.o.v. (%) 9.7% 13.2%
RequiredNtoachieve 235 44
c.o.v. =20%
Method (
=0.268)
Number of sampleN 1000 100
Computer runtime(minutes) 282 12 1
Estimatedfailureprobability 0.0960 0.0913 0.097
Estimator c.o.v. (%) 9.7% 13.2%
RequiredNtoachieve 235 44
c.o.v. =20%
Figure6. Theslopeconsideredinthesecondexample. The
greylinesaretheadmissiblerepresentativeslipsurfaces.
OMS failure probabilities. Notice that the calcula-
tionsofar isfast becauseonlythenon-iterativeOMS
equationsareinvolved.
Figure 7 shows the actual failure region F deter-
minedbytheexhaustiveanalysis. It isclear that there
aretwofailuremodesfromthegeometryof theactu-
ally failureregion. Unfortunately, theFORM method
isonly abletoidentify onemode. Moreinterestingly,
thedistancestotheoriginof thelimit statefunctions
for thetwo modes aresimilar, indicating thefailure
probabilities of the two failure modes are compara-
ble. Thelocations of theOMS design points for the
representativesurfaces areshown as thetriangles in
Figure7. Itisinterestingtoseethatbothfailuremodes
arecapturedbythetriangles. FollowingtheISproce-
dure, 100samplesof Zaredrawnfromq(z) toobtain
P
IS
F
via Eq. (13). Those samples are plotted in Fig-
ure7. Itisclear thatalargeportionof thesamplesare
failure samples. The resulting P
IS
F
is 0.0038, and its
c.o.v. isestimatedtobe20.9%viaEq. (14). TheMCS
method with samplesizeN=10000 is also taken to
estimatethefailureprobability. Thefailureprobabil-
ityestimatefortheMCSmethodisfoundtobe0.0044
withc.o.v. =15.0%. Tofurther verifytheconsistency
of theISmethodwithlarger samplesize, anISanaly-
siswithsamplesizeof 1000istaken, andtheresulting
P
IS
F
is 0.0041. This result is fairly closeto theMCS
result. TheseresultsarelistedinTable2together with
theFORM results. It isclear that theadoptedFORM
Figure7. TheMCS(N=10000) andIS(N=100) samples
inthestandardGaussianspace(Example2).
Table2. Theanalysisresultsfromvariousmethods(Exam-
ple2).
Method MCS IS FORM
Number of sampleN 10000 100 1000
Computer runtime 2137 12 105 1
(minutes)
Estimatedfailure 0.0044 0.0038 0.0041 0.0016
probability
Estimator c.o.v. (%) 15.04% 20.9% 6.62%
RequiredNtoachieve 5655 109 109
c.o.v. =20%
methodsignificantlyunderestimatesthefailureprob-
abilityalthoughit iscomputationallycheap. Boththe
IS andMCS methods provideunbiasedestimates for
thefailureprobability, but theISmethodisobviously
moreefficient.
6 CONCLUSION
A newmethodbasedontheimportancesampling(IS)
technique is proposed to efficiently estimate failure
probabilityof slopestabilityfor circular slipsurfaces.
Themainnovelty istoutilizetheordinary methodof
slices to determine suitable locations of the impor-
tancesamplingprobabilitydensityfunction(ISPDF)
sothattheISPDF ismuchcloser tothefailureregion.
Theresulting algorithmis much moreefficient than
the Monte Carlo simulation although both methods
provide unbiased estimates for failure probabilities
of slopes. From the results of the three numerical
examples, it is concluded that themethods based on
First-OrderReliabilityMethod(FORM) mayunderes-
timatethefailureprobability,dependingonthenumber
of failuremodes. However, theproposed IS method
alwaysprovidesunbiasedfailureprobabilityestimates.
REFERENCES
Ang, A.H.S. andTangW.H. (1984). Probability Concepts in
Engineering Planning and Design, Volumn II Decision,
Risk, and Reliability, J ohnWileyandSons.
199
Bhattacharya, G., J ana, D., Ojha, S. and Chakraborty, S.
(2003). Direct search for minimumreliability index of
earthslopes. Computers and Geotechnics, 30, 455-462.
Bishop,A.W. (1955).Theuseof theslipcircleinthestability
analysisof slopes. Geotechnique, 5, 717.
Chowdhury, R.N. andXu, D.W. (1993). Rational polynomial
techniquein slope-reliability analysis. ASCE Journal of
Geotechnical Engineering, 119(12), 19101928.
Christian, J.T., Ladd, C.C. andBaecher, G.B. (1994). Relia-
bilityappliedtoslopestabilityanalysis. ASCE Journal of
Geotechnical Engineering, 120(12), 21802207.
Fellenius, W. (1936). Calculation of the stability of earth
dams. Transactions of 2nd Congress on Large Dams,
Washington, DC, 4, 445462.
Griffiths, D.V. andFenton, G.A. (2004). Probabilistic slope
stability analysis by finite elements. ASCE Journal
of Geotechnical and Geoenvironmental Engineering,
130(5), 507518.
Low, B.K. andTang,W.H. (1997). Probabilisticslopeanalysis
using J anbus generalized method of slices. Computers
and Geotechnics, 21(2), 121142.
Low, B.K., Gilbert, R.B. and Wright, S.G. (1998). Slope
reliability analysis using generalized method of slices.
ASCE Journal of Geotechnical and Geoenvironmental
Engineering, 124(4), 350362.
Rosenblatt, M. (1952). Remarks onamultivariatetransfor-
mation. Ann. Math. Statist., 23, 470472.
Rubinstein, R.Y. (1981). Simulation and the Monte-Carlo
Method, J ohnWiley& SonsInc., NewYork.
Siegel, R.A. (1975). STABL User Manual, J oint Highway
Research Project 75-9, School of Engineering, Purdue
University, West Lafayette, Indiana.
Vanmarcke, E.H. (1976). Reliability of earth slopes. ASCE
Journal of Geotechnical Engineering, 96(2), 609630.
Xue, J.F. andGavin, K. (2007). Simultaneousdetermination
of critical slip surface and reliability index for slopes.
ASCE Journal of Geotechnical Engineering, 133(7),
878886.
200
Correlationof horizontal subgradereactionmodelsfor estimating
resistanceof pilesperpendicular topileaxis
Y. Kikuchi
Port &Airport Research Institute, Yokosuka, Japan
M. Suzuki
ABSTRACT: A beamonanelastic-mediummodel isconventionallyusedfor designingpilesloadedperpen-
diculartothepileaxis.Threetypesof subgradereactionmodel: Changsmodel, theS-typemodel, andtheC-type
model, aregenerallyusedinthedesignof port facilities. Theestimatedcoefficient of subgradereactionvaries
fromgeotechnical investigationresultsquitemarkedlyfor eachmodel. It isthereforeimportant toimprovethe
accuracyof estimatingthecoefficientof subgradereactioninreliability-designmethods. Todothisfor Changs
model, themost widely usedmodel, anewrelationbetweentheSPT-N valueandthecoefficient of subgrade
reactioninChangsmodel, k
CH
, isproposed.
1 INTRODUCTION
InJ apan, whenestimatingtheresistancecharacteris-
tics of apileagainst aforceacting perpendicular to
thepileaxis, thepileisusuallymodeledasabeamon
anelasticmedium. Thismodelingapproachissimple,
yet has enough accuracy for practical design situa-
tions. Whenapileismodeledasabeamonanelastic
medium, the modeling of the subgrade reaction is
the most important part of design fromthe view of
geotechnical engineering.
Subgradereactionperunitarea,p,isgenerallygiven
asp =k x
m
y
n
, wherek isthecoefficientof subgrade
reaction, x isdepth, y isdeflection, andm, n areexpo-
nential indices. For example, inChangsmodel, which
isoneof themostwidelyusedmodelsinJ apan, mand
n areset asm=0, n =1andthesubgradereactionis
expressedasp = k
CH
y. InJ apanesetechnical stan-
dards and recommendations for port facility design
(J TSP) (MLIT, 2007), PHRI models (PHRI =Port
& Harbour Research Institute, the former name of
the Port & Airport Research Institute) also are rec-
ommended as useful models for subgrade reaction
models for pile design. They consist of two types:
the S-type model and the C-type model. In the S-
typemodel, m andn areset asm=1andn =0.5and
thesubgradereaction is expressed as p =k
s
x y
0.5
.
TheS-typemodel isalsocalledKubosmodel. Inthe
C-typemodel, m andn areset as m=0, n =0.5and
thesubgradereactionisexpressedasp =k
c
y
0.5
. The
C-type model is also called Hayashis model. The
S-type model is mainly applied to normally consol-
idatedground, andtheC-typemodel ismainlyapplied
to over-consolidated ground. The PHRI models are
proposedinaccordancewithexperimental results, and
can explain a piles behavior. Shinohara and Kubo
(1961), whoproposedthemodels, concludedthat the
coefficients of themodels, k
s
and k
c
, arefundamen-
tally constant and that their values depend solely on
thegroundcondition.
Changs model is appliedto awiderangeof anal-
ysis because the governing equation for a beamon
anelasticmediumcanbeanalyticallysolvedbyusing
Changsmodel. Themodel assumeslinearitybetween
thesubgradereaction, p, andthedeflection, y. How-
ever, thisassumptiongivesimpropercharacteristicsof
theresistanceof apileloadedperpendiculartothepile
axis.
Changsmodel andthePHRI modelsareconsidered
for subgrademodels for estimating theresistanceof
piles perpendicular to pileaxis in this paper. In this
research, first, thevarianceof therelationbetweenthe
SPT-N valueandthecoefficient of subgradereaction
of each model is surveyed. Then the correlations of
thecoefficient of each model arepresented. Finally,
newrelationsfor estimatingthecoefficientsfromthe
SPT-N valueprofilearedeveloped.
2 COMPARISONOF BEHAVIOR OF THREE
SUBGRADE REACTIONMODELS
2.1 Fitting of laboratory loading test results
A series of experiments (Kikuchi et al, 1992) is con-
ducted in asand box with alength of 6m, width of
3m, anddepthof 3m, whichis filledwithdry sand
to a depth of 2.25m. Thesand used in this study is
Sengenyamasand, whichhasadensityof 2.746g/cm
3
,
andD
50
of 0.2mm. Therelativedensityof thelayer is
201
40%. A pile0.2mwideand2.5mlong, withaflexu-
ral rigidity2.94kNm
2
, isusedintheexperiments.The
embedded length of thepileL is set to 2.1m. Hori-
zontal static loading on thepileis conducted with a
loadingheight hof 0.25m. Measureditemsare: time,
load, horizontal andvertical displacementsof theload-
ingpoint, piletop, horizontal displacementof thepile
at the ground surface, and bending moments of the
pile.
Figure1showsthecomparisonbetweentheexper-
imental results and the fitting results using Changs
model andtheS-typemodel.
Figure1(a) comparesdeflectiondistributions. The
coefficient of subgrade reaction on each model is
selectedtofit thedeflectionof thepileat theground
surface. Thedeflections at theprotrudingpart arein
goodagreement ineachmodel. However, thedeflec-
tionof thedeepareainChangsmodel doesnotfitwell.
Values of k
s
for fittingat bothloadinglevels arethe
same,althoughthevalueof k
CH
changesinaccordance
withtheloadinglevel.
Figure1(b) compares thebending moment distri-
butions. Thecoefficientsof subgradereactionusedin
thisfigurearethesameasthevaluesusedinFig. 1(a).
TheseresultsfromtheS-typemodel showgoodagree-
mentwiththeexperimentdata, whereastheresultsfor
Changs model have less agreement. In the case of
Changs model, maximumbending moment is very
small anddecrement of bendingmoment isveryslow
at depthsfromaround0.8mto1.5m.
Inthesefigures, thebest fittingcoefficient of the
subgradereactionvalueof theS-typemodel is inde-
pendent fromthe loading level, but that of Changs
model decreasesalongwithanincreaseintheloading
level.
2.2 Comparison of features of three subgrade
reaction models under other conditions
As the model experiment is conducted under a pile
headfreecondition, calculationresults of theS-type
model and Changs model are compared under pile
headfreeconditionsinSection2.1. Thoseunder other
conditionsarecomparedinthissection.
Tobeginwith, thecoefficientsof subgradereaction
of thePHRI model andChangsmodel aredetermined
under the same load, the same pile head condition,
and the same deflection at ground surface with the
samepileconditionandasimilar deflectionatground
surface presented in Section 2.1. Calculated deflec-
tion distributions and bending moment distributions
of thesemodelsarecompared.
Comparisonof thecalculationresultsof theS-type
model andChangsmodel under pileheadfixedcon-
ditionareasfollows:Thereislessdifferenceobserved
indeflectiondistributions, but themaximumbending
moment fromChangsmodel wasquitesmall andthe
decrementof bendingmomentisveryslowcompared
totheS-typemodel.Thesebehaviorsaresimilartothe
resultspresentedinSection2.1.
Figure1. Comparisonof thedifferencesinsubgradereac-
tionmodelsinacaseinwhichthereisthesamedeflectionat
groundsurface.
Comparisonof thecalculationresultsof theC-type
model andChangsmodel under bothpileheadfixed
and pile head free conditions are as follows: There
arefewer differences observed in thedeflection dis-
tributions, but the maximumbending moment from
Changsmodel issmall andthedecrement of bending
202
momentisslowcomparedtoC-typemodel.Thesedif-
ferencesarerather small comparedtothedifferences
in the comparison in the S-type model and Changs
model.
The value of maximumbending moment and the
depth appearing at the maximumbending moment
areimportant factorsinpilefoundationdesignwork,
in addition to deflection at ground surface and the
protrudingpart.
BecausetheS-typemodel quitewell representsthe
bending moment distributions of the experiment as
showninFig. 1, thecoefficient of thesubgradereac-
tionof Changsmodel isdetermined, soastohavethe
same maximumbending moment as that calculated
usingtheS-typemodel under thesamepileandload-
ingconditions presentedinSection2.1. Comparison
of thecalculationresultsisshowninFig. 2.
The coefficient of subgrade reaction of Changs
model need to be greatly reduced to have the same
maximum bending moment as that in the S-type
model. As a result, distributions of deflection and
bending moment from Changs model were quite
different fromtheexperimental results.
A comparison of calculation results in theS-type
model andChangsmodel under apilefixedcondition
determinethecoefficient of subgradereaction, so as
tohavethesamemaximumbendingmoment, iscon-
ducted. SimilarresultsasshowninFig. 2areobserved;
thecoefficientof subgradereactionof Changsmodel
needtobegreatlyreduced, anddistributionsof deflec-
tion and bending moment fromChangs model are
quitedifferent fromthat of theS-typemodel.
ThesamecomparisonsareconductedfortheC-type
model andChangs model. Inthecaseof theC-type
model and Changs model under a pile head-fixed
condition, thedepthappearingat themaximumbend-
ing moment and decrement behavior of thebending
moment arequitedifferent.
3 RELATIONBETWEENSPT-NVALUEAND
COEFFICIENTSOF SUBGRADE REACTION
Figures 3 to 5 showtherelation between theSPT-N
valueprofileandthecoefficient of subgradereaction
of eachmodel for estimatingthepileresistanceper-
pendiculartothepileaxis(MOT,1999).Groundwhere
theSPT-N valueisconstantwithdepthcanbemodeled
as C-typemodel groundwhenusingthePHRI mod-
els, while ground where the SPT-N value increases
withdepthcanbemodeledas S-typemodel ground.
Theabscissaaxis of Fig. 4 is theaverageincrement
of SPT-N valueinameter. If usingFig. 5toestimate
k
CH
, theaverageof theSPT-N valuefromthesurface
tothedepthof 1/ isused.
Thesefigureswereproducedbefore1967inaccor-
dance with the existing relation between field load-
ingtest resultsandgeotechnical investigationresults.
Measurementinbothclayeygroundandsandyground
wereusedtomakethesefigures, soastocovertherela-
tionsbetweenk
s
, k
c
,k
CH
andtheSTP-N profileinboth
Figure2. Comparisonof thedifferenceof subgradereaction
modelsincaseswiththesamemaximumbendingmoments.
clayeyandsandyground.Dottedsquaresshowthedata
fromclayeyground, andsolidsquaresshowthemfrom
sandy groundinFig. 3. Whitesquares showthedata
fromclayey ground, andsolidlines showthemfrom
203
Figure3. RelationbetweenSPT-N valueandk
c
.
Figure4. RelationbetweenN andk
s
.
sandy groundinFig. 4. Whitesquaresandshadowed
squaresshowthedatafromclayeyandsandyground,
respectively, inFig. 5. Thesefiguresarepreparedfor
the convenience of designers to estimate coefficient
with conventional geotechnical investigations. Actu-
allythequalityof theloadingtestsandthegeotechnical
investigations used for producing the figures varies
greatly, sotherelationhassomevariance.
The coefficient of subgrade reaction of Changs
model needs to bechanged in accordancewith load
level or deflectionlevel for estimatingpilesbehavior,
becauselinearity between thesubgradereaction and
the deflection is assumed in Changs model and the
actual relation is nonlinear. Thevalueof k
CH
shown
inFig. 5is back-calculatedfromtherelationof load
andsurfacedeflectionwherethesurfacedeflectiony
0
is1cm(MOT, 1999).
Figure5. RelationbetweenSPT-N valueandk
CH
.
Table1. RelationbetweenSPT-N valueprofileandcoeffi-
cient of subgradereaction.
Approximation Coefficient of Coefficient
formula correlation of variance
k
c
=540N
0.648
(kN/m
2.5
) 0.872 0.111
k
s
=592N
0.654
(kN/m
3.5
) 0.966 0.077
k
CH
=3900N
0.733
(kN/m
3
) 0.917 0.754
From Figs. 3 to 5, the correlation and variance
betweentheSPT-N valueandthecoefficients of the
modelsarecalculated.TheresultsareshowninTable1.
As showninTable1, thereis abigvarianceinthe
relation between SPT-N valueand k
CH
. It is hard to
precisely estimatek
CH
fromFig. 5, whentheSPT-N
valueis less than10. Thecoefficients k
s
andk
c
esti-
mated fromtheSPT-N valuehavea small variance.
This means that k
CH
canbegivenabetter estimation
by consideringtherelationbetweenk
CH
andk
c
or k
s
thandirectlyfromtheSPT-N value.
4 RELATIONBETWEENK
S
ANDK
CH
OR K
C
ANDK
CH
Here, a new relation between the SPT-N value and
k
CH
isproposed, althoughit cannot solvetheinherent
problems of Changs model as showninFig. 2. Nev-
ertheless, it will useful in producing amorereliable
estimationof k
CH
for design.
Themain factors affecting k
CH
areground condi-
tion, pile diameter, flexural rigidity of pile, loading
level, and loading height. On theother hand, k
s
and
k
c
aredefinedso as to beconstants only affectedby
ground condition. This kind of difference gives the
differenceof varianceof therelationbetweenSPT-N
values and each coefficient of subgrade reaction as
showninTable1.
204
Tobeginwith, therelationamongk
s
or k
c
andk
CH
isexamined. Here, wetrytodeterminetheir relations
when surfacedeflection is thesameunder thesame
pileandsameloadingcondition.
If theloadingheight is 0, suchas surfaceloading,
even thePHRI models havean approximateanalyti-
cal solutionandthefollowingequationsarepresented
(MOT, 1999):
for theS-typemodel:
Pileheadfreecondition:
Pileheadfixedcondition:
for C-typemodel:
where, EI : flexural rigidity, D: pilediameter, k: coeffi-
cient of subgradereaction, T: horizontal loadapplied
topiletop, y
0
: deflectionat groundsurface.
Therelations among k
s
or k
c
and k
CH
for surface
loading arecalculated using Eqs. (1) and (2) and an
analytical solutionof Changsmodel (y
0
=T,(2EI
3
):
whenthepiletopis free, andy
0
=T,(4EI
3
): when
thepiletopisfixed):
Thiskindof relationisalsoexaminedfor pilespro-
trudingaboveground. Insuchacase, thePHRI models
cannot giveananalytical solution. Thecalculationof
Table2. Rangeof pileconditionsfor calculation.
Pilediameter D 0.51.8m
EI /D 1206400MNm
Loadingheight h 120m
Surfacedeflectiony
0
0.010.1m
k
s
or k
c
30010000kN/m
3.5
(or kN/m
2.5
)
thePHRI modelsareconductedusingalawof similar-
ity(MOT, 1999).About4800casesof calculationsare
conductedwithvaryingparametersfor aconsiderable
rangeof steel pipepilesasshowninTable2.
Calculationconditions for determiningthecoeffi-
cientof subgradereactionforeachcasewere: thesame
pile condition, the same loading condition. Another
conditiontobeconsideredisseparatedintotwocases
todeterminethecoefficient
1) in order to have the same deflection at ground
surface.
2) in order to have the same maximum bending
moment.
In thecaseof 2), thetotal behavior of thepileis
quitedifferent inbothmodelsasseeninSection2.2.
Thereforethecorrelationof k
s
andk
CH
, andk
c
andk
CH
areconsideredunder theconditionof 1).
Multipleregressionanalysesareperformed, result-
inginthefollowingmultipleregressionequations.
whereh: loadingheight.
Theseregressionequationshavemorethan0.9999
of multiple correlation coefficients. Figure 6 shows
theexampleof therelationbetweenk
CH
andthecal-
culationresultsof regressionequationunderanS-type
model withapileheadfreecondition.
Relations between k
s
and k
CH
under theconditions
showninFig. 1arecalculatedfromEq. (5a) inorder
toverify theaccuracy of thisregressionequation. As
showninFig. 1, thebestfittingcoefficientof subgrade
205
Figure6. Relationbetweenk
CH
andtheregressionequation
inS-typeground, pilehead-freecondition.
reaction in the S-type model k
s
was 1200kN/m
3.5
.
Deflectionsat groundsurfaceunder ahorizontal load
of 2.11kN and 4.19 kN was 0.065 m, and 0.145 m,
respectively. Loading height h is 0.25m, and EI/D
is 14.7kNm in this case. The coefficients of sub-
grade reaction of Changs model fromEq. (5a) are
980kN/m
3
and710kN/m
3
under aloadingcondition
of 2.11kN and 4.19kN, respectively. These values
coincidewiththevaluefromthefitting.
Because no suitable experimental data are pre-
pared in the case of the C-type model and Changs
model, comparing calculation results are presented
here.AssumeapilewithEI of 9.5610
5
kNm
2
,diam-
eter D of 1m, inserted20minto thegroundwithan
SPT-N of 8, andloadedat 3minheight h. Whenthe
SPT-N valueis8,k
c
iscalculatedas2080kN/m
2.5
from
Table1. If 270kNand1200kNof horizontal loadare
applied to the pile, deflection at the ground surface
calculatedbytheC-typemodel are0.010mand0.096
mrespectively. Thecoefficients of subgradereaction
of Changs model k
CH
calculated with Eq. (??) are
26400kN/m
3
and 8760kN/m
3
respectively. Figure7
showscalculateddeflectionandbendingmomentdis-
tributionof bothbytheC-typemodel andbyChangs
model. Calculationresultsof bothmodelsareingood
agreement inthiscase.
As designers want to know about the relation
betweengroundinvestigationresultsandcoefficientof
subgradereactionindesignwork, therelationbetween
k
CH
andSPT-N orSPT-N areleadfrombothwithEqs.
(5), (6) andwithTable1. Therelations arepresented
asfollows:
If SPT-N valueincreaseswithdepth:
Figure7. PiledeformationinC-typeground.
206
Figure8. NewrelationbetweenSPT-Nvalueandk
CH
.
If SPT-N valueisconstant alongthedepth:
Inthiscalculationform, COVsof k
CH
arethesameas
thoseof k
s
andk
c
showninTable1, becausemultiple
correlationsof k
CH
andk
s
or k
c
arealmost 1.
Figures8and9showanexampleof theserelations.
Therelationbetweenk
CH
andSPT-N proposedin
J TSP as shown in Fig. 5 is arrived at by fitting the
loading experiment results of deflection at aground
surfaceof 1 cm. If thedataused for producing Fig.
5 are fromthe ground condition of constant SPT-N
alongthedepth, thevarianceof therelationbetween
SPT-N andk
CH
issmall asshowninFig. 8. But, if the
SPT-N valueincreases withdepth, fittedk
CH
will be
affectedby thefactor (EI/D). Thisisthemainreason
therelationbetweentheSPT-N valueandk
CH
shows
alargevariety.
5 CONCLUSION
This paper reviewed the variances of the estimated
valuesof thecoefficient of subgradereactionfor esti-
matingtheresistanceof apileperpendiculartothepile
Figure9. NewrelationbetweenN valueandk
CH
.
axis fromtheprofileof SPT-N values. Theproposed
relationbetweenSPT-N valueandk
CH
showedalarge
variance, althoughthosebetweentheSPT-N valueand
k
c
or N andk
s
showedsmall variances.
Tomakecorrelationsof subgradereactionmodels,
it was foundthat comparingeachcoefficient of sub-
gradereactionshouldbedoneinthesamedeflection
atgroundsurfaceunder thesamepileandgroundand
loadingconditions.
New relations between the profile of the SPT-N
valueand k
CH
or k
c
or k
s
wereproposed based on a
parametric study andmultipleregression analysis. It
wasfoundthat thelargevarianceof theexistingpro-
posal of therelationbetweenSPT-N valueandk
CH
was
causedbydisregardingtheeffectsof (EI/D) whenthe
SPT-N valuesincreasewithdepth.Thenewlyproposed
relationshipswill minimizethevarianceof estimated
valueof thecoefficient of subgradereactionineach
model.
REFERENCES
Kikuchi, Y., Takahashi, K., and Suzuki, M. (1992)."Lateral
resistance of single piles under large repeated loads."
Report of PHRI, 31(4), 33-60. (inJ apanese).
Shinohara, T. andKubo, K. (1961). "Experimental study on
thelateral resistanceof piles(part1) lateral resistanceof
singlefreeheadpilesembeddedinuniformsandlayer ."
Monthly Report of TTRI, 11(6), 169-242. (inJ apanese).
Ministryof Land, Infrastructure, andTransportation(2007).
Technical standards and commentaries for port and har-
bor facilities in Japan, The J apan Port and Harbour
Association. (inJ apanese).
Ministry of Transportation(1999). Technical standards and
commentaries for port and harbour facilities in Japan,
TheJ apanPort andHarbourAssociation. (inJ apanese).
207
Risk management in geotechnical engineering
Thelongandbigtunnel fireevacuationsimulationbasedonanacceptable
level of riskandEXODUSsoftware
Shu-qingHao, Hong-wei Huang&YongYuan
Department of Geotechnical engineering, Tongji University, Shanghai
ABSTRACT: With thelong and big tunnel extensiveapplication of modern civilization, fireas oneof the
major disasters, areoftenconcerned. Inthepast, someexpertsandresearchershavedonealotof researchworks
aboutthefiretemperatureandsmokediffuseruleoccurredinthelongandbigtunnel fireandhavegottenmany
valuableconclusionsforfireevacuationinbigtunnels. However, theconclusionshavesomeblindnessbecauseit
hasnot beenguidedbasedtheriskacceptancelevel. Howtocontrol thefirelossunder theriskacceptancelevel
andrealizethepeopleevacuationafter fireisveryimportant. Theauthorsmadetheacceptancerisklevel bythe
society after investigationandresearch. Then, usethelevel 2risk to guidethesimulationof thelongandbig
tunnel evacuationtechnology. EXODUSisasuiteof softwaretoolsdesignedtosimulatetheevacuationof large
numbersof individualswithincomplexstructures.TheEXODUSfamilyof evacuationmodelscurrentlyconsists
of airEXODUS, buildingEXODUS and maritimeEXODUS. TheEXODUS softwaretakes into consideration
people-people, people-fireandpeople-structure. TheEXODUSsoftwarehasbeenwritteninC++ usingobject
orientated techniques and rule-baseconcepts to control thesimulation. Thus, thebehavior and movement of
eachindividual isdeterminedbyaset of heuristicsor rules. Themodel tracksthetrajectoryof eachindividual
asthey maketheir way out of theenclosure, or toovercomefirehazardssuchasheat, smokeandtoxic gases.
Developed by theFireSafety Engineering Group (FSEG) at theUniversity of Greenwich, Smatfireaims to
makefirefieldmodel techniquesaccessibletonon-Computational FluidDynamics(CFD) experts. Suchasfire
fighters, architectsor safetyengineers. Smatfireachievestheseaimsbyintegratingknowledgebasedtoolsand
techniqueswithinmanyof themodules. Smartfireisacompletefirefieldmodellingenvironment that employs
stateof theartfinitevolumemethods. Smartfireprovidesdynamicuser interactionandsophisticatedcaseset-up
tools to facilitatesimulationandto helpensuresolutioncorrectness. Theauthors useSMARTFIRE software
to simulatethefireinthelongandbigtunnels andimport thehazardmodel to buildingEXODUS evacuation
softwareto do theevacuationsimulationthefire. By comparingthedifferent numbers anddifferent widthof
evacuationholes, get thefireevacuationtunnel structureparametersandtheventilationtechnologyparameters
under riskacceptablelevel 2.
1 INTRODUCTION
Tunnel asoneof thethree-dimensional modesof trans-
portationcanbeusedonurbantrafficeasily. However
as one of the great disasters in tunnel, fire disaster
riskisaveryhorribleandlargelossone. Studyof the
evacuationtechnologywhenfiredinthetunnel isvery
important.
2 BUILDINGEXODUSMETHOD
EXODUS is a suite of software tools designed
to simulate the evacuation of large numbers of
individuals within complex structures. The EXO-
DUS family of evacuation models currently consists
of air-EXODUS, building-EXODUS and maritime-
EXODUS. The EXODUS software takes into
considerationpeople-people, people-fireandpeople-
structure. TheEXODUSsoftwarehasbeenwrittenin
C
++
languageusingobject orientatedtechniquesand
rule-base concepts to control the simulation. Thus,
the behaviour and movement of each individual is
determinedby aset of heuristicsor rules. Themodel
tracks thetrajectory of eachindividual as they make
their way out of theenclosure, andhowto overcome
fire hazards such as heat, smoke and toxic gases.
Developed by the Fire Safety Engineering Group
(FSEG) at the University of Greenwich, Smartfire
aims to makefirefield modelling techniques acces-
sible to non-Computational Fluid Dynamics (CFD)
experts, suchasfirefighters, architectsorsafetyengi-
neers. Smartfire achieves these aims by integrating
knowledge based tools and techniques within many
of the modules. Smartfire is a complete fire field
modelling environment that employs state-of-the-art
finite volume methods. Smartfire provides dynamic
user interaction and sophisticated case set-up tools
to facilitate simulation and to help ensure solution
correctness. Smartfire uses state-of-the-art physics,
numericandintuitiveconfigurationoptionstogivethe
most true-to-lifesimulationmodels.
211
Figure1. EXODUSmodel of submodels.
BuildingEXODUS model is not only asylum, but
inemergencysituations, under normal circumstances,
computer-basedevaluationof populationmovements,
laboratory operations. Through GMT University of
earlier research and development, fire safety Engi-
neeringGroup(FSEG) developedasimulationof the
buildingEXODUS is people, people with fire, the
interactionof peopleandstructures.Thismodel issub-
ject to thepursuit of heat, smokeandtoxic gas after
theimpact, escapedfromtheinterior of eachpersons
pathasylum.
3 LONGANDBIGTUNNEL RISK
ACCEPTABLE LEVEL
Generallyspeaking,theriskcanbecharacterizedasthe
probabilityof riskof accidentsandlossof theproduct
of theaccidents. Thefollowing aretherisk gradeof
standards(seetable1andtable2).
Table1. Riskclassificationevaluationcriterion
Intable1Character P presentsprobability; Character L
presentsloss. And1ispresentsnegligible, 2istobeconsid-
ered, 3isserious, 4isveryserious, 5isdisastrous. Inorder to
makeriskassessmentlooksmoreintuitive, wemakedifferent
color topresentsdifferent risklevels.
Table2. Riskratingstandardcolor logo.
Inthearticletheriskincluding2points,theyarerisk
probabilityandlossbecauseof risk. Whenconsidered
therisk level. Whentherisk level is 2, accordingits
concept, thepeoplecasualtiesprobabilityisbellow0.1
percent. Hereit isastructureoptimizationcriterion.
4 EVACUATIONTHEORY ANDEVACUATION
STRUCTURE SITUATIONABOUT LONG
ANDBIGTUNNEL
Safe evacuation time is a very important parameter.
Theevacuationtimeisaffectedmainlybyunderground
spacedensity of staff, theevacuation holes rational-
ization and the concentration of toxic gas, etc. The
greater of theevacuationchannel density, thegreater
themouth, theeasier toevacuate. However, fromthe
perspectiveof economic riskstoconsider, it doesnot
suitableto design theholes too denseand causethe
greater economic costs. Therefore, combination the
risk considerations, to meet theoptimal level of risk
evacuationcircumstances, theevacuationchannelsto
findtheoptimal designof thedesignparameters.
The conditions for safe evacuation time: Secu-
rity officers evacuated thetimecriterion is to assess
whetherthefiresafetycriteriaisoneof themainobjec-
tives. Guarantee the safety of underground space in
thefireevacuation,thekeytechniqueistocompletethe
evacuationof all personnel timerequiredtoallowthe
safeevacuationtimeisless. Nowweconsider therisk
level 2for thedesigncriterion.
Trapped persons of theindividual as amodel is a
collection of attributes that are widely divided into
four categories: physical (suchasage, gender, mobil-
ity, etc.), psychological (such as patience, power,
etc.), experience (such as distance, individual time-
consuming, andsoon) andtheimpact of disaster risk
(forexample, FIN, FICO
2
, FIH).Theseattributeshave
adual effect; all thetrapped persons can bedefined
asindividualsandthroughthedifficultiesof tracking
themtounderstandtheprocess. Simulationof thepro-
cess, someattributeisfixed, andsomeother attributes
as theother modeof input tochangetheoutcome, is
dynamic. Inaddition, somepropertiesrequireusersto
manuallyor attributedefinitionof value, or acceptthe
default settings, other attributeis in themodulewas
calculated.
5 THE COMPUTATIONAL PROGRESS
ABOUT THE BUILDINGEXODUS
(1) Geometrymodel
The establishment a length of 1000m, width of
15m, high up to 8mof the tunnel, similar to a
squaresectionand500minlength, incaseof fire,
firethreedimensionsare3mof gasburner of the
gas burner alinear rampof heat releasecurve, a
slopeof 34kwthefirst20s. 6radiationfluxmodel
isusedtodesignaradiantheat.Aassumptionabout
theheat transfer model is that thetunnel wall to
wall of theturbulentheattransfer. Longtunnel fire
212
burningpoweris35kw.Thefireparametersforthe
curve: B=(34/20)=1.7kw.
(2) Mesh
Use mesh tools of the Smartfire to mesh the
geometrymodel andget theresult asfig. 2
Figure2. Meshfigure.
(3) CFDcalculationenginerunning
Figure3. Thedevelopment of fire.
Figure4. Gasflowpattern.
Summary: In comparison, when the ventila-
tion rate is 3.0m/s, will effectively curb return.
Below import EXODUS evacuation simulation
conducted by the circumstances of the effective
evacuation timeto reach theevacuation and fire
safetyriskrating.
6 EVACUATIONBASEDONTHE EXODUS
6.1 Set up two export spreadsheet (not consider the
impact of disasters)
Figure5. Theevacuationsituationwhentime=00:05:06.
When experienced five minutes six seconds, the
evacuationisstill not completed, 5,000peoplein833
evacuated. Withthefirescenecontrol, andpersonnel
cannotbeeffectiveintimetoevacuate, theevacuation
of failure.
6.2 Set up 12 export spreadsheet (not consider
the impact of disasters)
Through 6 minutes 37 seconds, evacuation is still
not completed, the crowd appeared in exports arch
distribution, and through the control of fire scenes,
thestaff still cannot beeffectiveintimetoevacuate,
theevacuationof failure. Analysis of thereasons for
thefailureof theevacuation, setuptoomanystaff, set
upatotal of 20,000people, of whichonly8,133people
in6minutes37secondstoevacuate. Therest wereall
killedwereburnedtodeath.
Figure7. Geometricpatternsall thetimefledthefunction.
Figure8. All fled thegeometric patterns of afunction of
time.
6.3 Set up two export-100m
Oneminutefour secondslater, 2,000peoplein180
toevacuation.
213
11seconds23,all peopleescaped.Fireburningsim-
ulation calculation based on SMARTFIRE software
with the heat generated by the speed of the spread
of toxic gases contrast. Note75mhereto escapethe
holedesignedto meet thefull demandfor thesafety
of escape.
6.4 50m spacing set to escape the hole, a long
100m, the number of 800 people, in the escape
of the exit of the arch there, and three seconds
in 39 time-sharing, to escape the end.
Set thewidthof doorsfor 1.5m.
Figure11. Evacuationsituation.
Figure12. Whentime=00:01:28.
Figure13. All fledthegeometricpatternsof timefunction.
Figure14. All fledthegeometric patterns of afunctionof
time.
Figure 15. Relationship between Elapsed time and Total
FlowRate.
7 CONCLUSIONS
(1) When the fire broke out inside the tunnel when
thefirefromthefirealongthetunnel wall space
to spread, different rates of thermal power ledto
thereleaseof spacewithintheunacceptablehigh
temperatureof different times, thehigher therate
of heat release, thesooner thespacetemperature
changes, Releaseto35KWof power, spacewithin
the tunnel will be 7min within the scope of the
temperaturereached200in60
C, andthefluegas
will alsospreadto200mmof thetunnel.
(2) thecrowdthroughthedoorsof thetimecalculated
by EXODUS be, we can see that when the heat
releaserateof 35KWof whenthefirebrokeout, if
thetunnel every75msetupanescapehole, will
enablethestaff within15minutes safety Escape,
thefirerisk rating will also becontrolled at two
levels. Therisklevel isacceptable.
ACKNOWLEDGEMENTS
Thework describedinthearticleis supportedby the
Keyproject(2006BAJ 27B04) supportedbyNational
Science&TechnologyPillar Program.
REFERENCES
1 A Study on the Evacuation of People in a Hall Using
theCellular automaton Model. International Journal of
Modern Physics C (IJMPC).2007Vol (18) I: 359367.
2 Cellular automaton Simulation of the Escaping Pedes-
trianFlowinCorridor. International Journal of Modern
Physics C (IJMPC). 2005. Vol(16).Page: 225235
3 Preston, RJ ; Marcozzi, D; Lima, R, et al. The effect
of evacuation on the number of victims following haz-
ardous chemical release. Prehospital Emergency Care.
2008. Vol(12): 1823
4 Wang, J H; Lu, SX; Wang, DR, et al. Risk analysis of
evacuation under building fire based on point estimate
method. 1st International Conference on Risk Analysis
and Crisis Response, Date: SEP 2526, 2007 Shang-
hai Maritime Univ Shanghai PEOPLES R CHINA.
214
Source: PROCEEDINGS OF THE 1ST INTERNA-
TIONAL CONFERENCE ON RISK ANALYSIS AND
CRISISRESPONSE 2007Vol(2): 312316.
5 Zhang, Q; Liu, M; Liu, J, et al. Modification of evac-
uation time computational model for stadium crowd
riskanalysis.ROCESS SAFETYANDENVIRONMENTAL
PROTECTION. 2007Vol(85): 541548
6 Liu,YZ; Zhang, XN; Li, FW, etal. Integratingevacuation
trafficsimulationmodel withfloodrisk systeminaGIS
environment. ConferenceInformation: 2nd International
Symposium on Intelligence Computation andApplication
(ISICA 2007), Date: SEP2123, 2007WuhanPEOPLES
R CHINA. Source: PROGRESS IN INTELLIGENCE
COMPUTATION AND APPLICATIONS, PROCEED-
INGS. 2007:662666
7 Conventional RSK Roleat Risk, STKE 2006(345):247
259.
8 A.P. TEIXEIRA; C.GUEDES SOARES. Reliability of
load Bearing Steel Plates Subjected to Localized Heat
Loads. International journal of reliability, Quality and
Safety engineering (IJRQSE). 2006. Vol(13)2: 97133
215
Riskbaseddecisionsupport systemfor thepumpingprocessin
contaminatedgroundwater remediation
ToshiroHata
Nagano National College of Technology, Nagano, Japan
YoshihisaMiyata
National Defense Academy, Yokosuka, Japan
ABSTRACT: Newrisk-baseddecisionsupportsystemisproposedforthepumpandtreatsmethodtoremediate
contaminatedgroundwater withVOCs. Thesystemcancontrol thepumpingprocessincontaminatedground-
water remediation by considering not only theremediation efficiency but also environmental impact. In the
evaluationof theremediationefficiency, humanriskiscountedbyusingASTMRBCA model.Thisachievement
maycontributefor low-cost andlow-riskremediationworks.
1 INTRODUCTION
Thepumpandtreatmethodhasbeenappliedtoalotof
contaminatedsiteswithVOCsinJ apan. Inthisreme-
diation works, it is difficult to control the pumping
ratebecausetherearemanyrestrictionssuchashuman
health and environmental risk. Authors propose the
newdecisionsupportsystemfor therational pumping
processincontaminatedgroundwater remediation.
2 PUMPANDTREAT REMEDIATION
2.1 Outline
Pump and treat systemcan be used to remove con-
taminant (suchasVOCs) massfromaplume, causing
theplumetoshrinktowarditssource. Thus, it maybe
possibletocontrol riskstopotential receptorsarising
fromdissolvedcontaminantsinplume. For control of
plumesat depthof 4050mor more, pump-andtreat
maybethemost viableandcost effectivetechnology.
Figure 1 illustrates the conceptual effluent con-
centrations with timefor an array of pumping wells
in apump-and-treat systemassociated with aplume
emanatingfromalongtermsourceof contamination
below thewater table. For all wells, thereis arapid
declineineffluentconcentrationsimmediatelyfollow-
ingtheonset of pumping, followedby alongperiod
of graduallydecliningconcentrations.
The quality of the effluent from the source-
contaminant wells is influenced by the groundwater
qualityclosetothesource, andtheeffectof dilutionby
uncontaminatedgroundwaterdrawnfromsurrounding
regionsaspumpingproceeds.Theeffluentdoesnot, in
apractical timeframe, attainaqualitysatisfyingenvi-
ronmental guidelines, adirect consequenceof contin-
uedreleaseof contaminantsfromthesourcezone. For
Figure1. Theconceptual remediationof pumpandtreat.
the plume extraction wells, contaminant concentra-
tionsmaydecreasetoacceptablelevelsintimeframes
extending fromyears to decades or longer. Decreas-
ingcontaminantconcentrationsineffluentfromplume
extractionwells areconsequenceof gradual removal
of contaminantsfromtheplumeincombinationwith
theintroductionof uncontaminatedgroundwater from
adjacentregionsof theaquifer systems. Andriskcon-
trol wells can reduce the groundwater contaminants
under atarget suchascancer risk.
2.2 Target of remediation
Oneof themoreimportant issuesconcerningground
water remediation is to set the reasonable target in
the remediation work. Persuasive target is to repro-
ducethepre-remediatedconditionwhichmeanszero
217
contaminant concentrations. However, this target is
alwaysimpossibletoachievebecauseof thehighcost
involved. InJ apan, therearenocomparablestandard
for allowableconcentrationsincontaminatedground-
water.Humanhealthriskassessmentapproachisatool
that permits establishment of remediationlevels (tar-
get) thatareprotectiveof humanhealth, dependonsite
conditions, remediationtechniques, chemical charac-
teristics, andutilizeacceptable risk levels that can
beselectedtobemoreor lessrestrictive. Humanrisk
assessment considers for mainsteps: 1) hazardiden-
tification, 2) dose-response or toxicity assessment,
3) exposureassessment and 4) risk characterization.
Thehazardidentificationisanidentificationof poten-
tially harmful chemicals present in different media
at a site. The toxicity assessment examines the tox-
icity, or harmfulness of each chemical found at the
site. Theexposureassessment stepconsistsof identi-
fyingconditionsunder whichpeoplecomeincontact
withthechemicals andcharacterizingthemagnitude
of exposure (exposure dose, mg/kg/day). In the risk
characterizationstep,informationfromtheabovethree
steps is combined to estimate an additional risk to
humanhealthcausedby exposureto thetoxic chem-
icals assessed. For chemicals causing cancer, risk is
understood to be the likelihood of cancer resulting
fromexposuretoachemical andisexpressedasaprob-
ability. If theriskobtainedfromtheriskassessmentis
largerthanthatestablishedasacceptable(usuallyset
110
5
of drinkingwaterinJ apan), thenitissaidthat
thepopulation is at risk and its necessary to reduce
mediaconcentrationstoAcceptablelevels. For non-
carcinogen chemicals, the Hazard Quotient (HQ) is
calculated, whichistheratioof theexposuredosefor
thatchemical totheRfD(ReferenceDose). If HQ>1,
it meansthat thepopulationisreceivingadoselarger
thantheRfD(maximumdoseatwhichitisknownthat
noadverseeffectsoccurs) and, consequently, thereisa
possibilitythatthepopulationwill experienceadverse
effects.
Forthispaper, newDecisionSupportSystem(DSS)
was proposed based on rational pumping process
basedof fuzzyinferenceandRBCA riskassessment.
3 RATIONALIZATIONOF PUMPINGRATE
BASEDONFUZZY INFERENCE
3.1 Fuzzy inference
Inthispaper, newrisk-baseddecisionsupport system
is proposedby introducingFuzzy inferenceconcept.
This concept is proposed by Zadeh in 1970 (Zadeh,
1968; Zadeh, 1973) and it has already applied to
various everyday electronic devices. In the field of
environmental geotechnics, the application to opti-
mumarrangement of a monitoring well etc. is pro-
posed (Morisawa, 1991). In this paper, inference
method proposed by Mamdani is used (Mamdani,
1974; Mamdani, 1976). This technique is a method
whichMamdani usedfor thefirst fuzzy control. The
sequencesof fuzzyinferenceareasfollowings.
Figure2. Outlineof thefuzzyinferencemodel.
1) Fuzzy logic control is defined in theformof the
following.
If X
1
isA
11
andY
2
isA
21
thenZisB
1
If X
1
isA
12
andY
2
isA
22
thenZisB
2
WhereX
i
is pumpingrate, Y
i
isVOCs concentra-
tion, Z
i
isremediationefficiency, A
ij
isfuzzylevel
andB
i
isfuzzylevel.
2) The degree of conformity (w
i
) of each rule is
calculated
3) Theinferenceresult of eachruleiscalculated.
4) ResultB
i
of eachruleisunifiedanditisthewhole
rule. Next, B
iscalculated.
5) Thecenter of gravityof result B
iscalculated.
AfterX
1
andX
2
aregivenbythisreasoningmethod
beforeaskingfor y
, sincemax (=) isbeingcalcu-

lated in Step 3 and it is finally calculating for rep-
resentationvaluey
by thecenter-of-gravity method,
min(=) iscalledmin-max compositioncalculating
methodbyStep1and2.
3.2 Rational pumping rate model
In order to keep a constant high effective remedia-
tion, it isexpectedthat maintainingthepumpingrate
withVOCsconcentrationandgroundwaterlevel main-
taining suitable is effectiveness in the remediation
period. Inthis study, theeffects of remediationwere
examinedby numerical simulation. Concept of fuzzy
inferencemodel isshowninFig.2.Thethreefollowing
werechosenas input datafromthedaily monitoring
information:
1) VOCs concentration was chosen the plume
situation.
218
Figure3. Sampleof themembershipfunction.
Table1. Sampleof thefuzzylogiccontroller.
Impactongeological environment
Groundwater level
Small Medium Large
Pumping Small Small Small Medium
quantify Medium Small Medium Medium
Large Medium Medium Large
Contributiontodecontamination
TCE concentration
Small Medium Large
Pumping Small Small Medium Medium
quantify Medium Medium Medium Large
Large Large Large Large
2) Groundwater level waschosentomonitor therisk
of settlement.
3) Pumpingratewaschosentorepresent theremedi-
ationeffect.
Each item is expressed by five membership func-
tions. Next, each55Fuzzy Logic Controller (FLC)
usingpumpingrateandVOCsconcentrationexpresses
environmental impact; pumpingrateandgroundwater
level expressesremediationefficiency.
Next, twin55FuzzylogicController(FLC) using
pumping rate, VOCs concentration and groundwater
level expressesenvironmental impactandremediation
efficiency. Thesamples of themembership function
and fuzzy logic controller are shown in Fig. 3 and
table1.
4 HUMANRISK CALCULATIONBASEDON
REFERENCE-DOSE MODEL
4.1 Human health risk assessment model
To performHuman health risk assessment, the four
stepsdescribedintheintroductionwerefollowed. For
theHazardIdentificationsteps, informationregarding
tolanduseetc. andchemical propertiesof theground-
waterwasmeasured.Groundwatersweresampleswere
takenat saturatedzoneandnear thesourcezoneand
risk control zone. For thedose-responseassessment,
toxicological informationwasusedtoUS-EPAsguide-
line value. For the off-site receptors the land use
was set as residential and for the on-site receptors
landusewas consideredas industrial. Themigration
pathwasconsideredfor groundwater remediationsite
were on groundwater migration. New human health
assessment model for pumpandtreat remediation.
Exposure pathway was only considered by the
groundwater ingestion. This model considered only
groundwater ingestion .To provides a overview of
the calculations done to determine if the exposed
populations are at risk equations (5) and (6) are
presented:
WhereHQ(nc)=hazardquotient for non-carcinogen
compounds; Risk(c)=risk for carcinogen com-
pounds; Conc =groundwater concentration; RfD=
Reference Dose (mg/kg/day); SF =Cancer slope
factor; IR
w
=Ingestion rate of Water (L/day);
ED=Exposure duration(year); EF =Exposure fre-
quency (days/year); BW =Body weight (kg), AT
n
=
Averaging time for non-carcinogens (year), AT
c
=
Averaging time for carcinogens (year). For the non-
carcinogen contaminants (equation 1), theHQ value
for a particular contaminant would be the exposure
dosedivideby RfD, if HQ>1, thenit wouldsuggest
thepossibilityof developingadverseeffects.
Theriskiscalculatedmultiplyingtheexposuredose
by the SF (cancer slope factor) which is the slope
for thedose-responsecurveobtainedinthelaboratory
forparticularchemicals, andrepresentsthepotencyof
thechemical carcinogen. If risk>acceptablerisk, it
wouldsuggest ahighriskthat thepopulationexposed
will developcancer.
Individual(for a single chemical) acceptable risk
wasset to110
5
(carcinogencompounds); accept-
able Hazard Quotient (HQ, single chemical, non-
carcinogeneffects) wassetto1.0aswell astheHazard
Index(HI ), thesevalueswereestablishedaccordingto
ASTM recommendations.
5 NEWDECISIONSUPPORT SYSYTEMWITH
FUZZY-RISK CALCULATIONMODEL
Thegoal of aPumpandPumpandTreat remediation
work shouldbeto protect humanhealthfor decision
making. Wedevelopedthenewdecisionsupport sys-
temfor groundwater remediationby usingPumpand
treatmethod.ThisDSScanpotentiallybeareasonable
219
Table2. Parameter of riskevaluation.
Parameter Value Unit Reference
NonCarcinogens Referential dose(RfD0) 0.006 mg/(kgday): TCE U.S.EPA (1996)

Bodyweight (BW) 50 kg J EA (1999)
Averagingtimefor non- 30 year J EA (1999)
carcinogens(ATn)
Ingestionrateof water (IRw) 2 L/day J EA (1999)
Exposureduration(ED) 3010 year Assumedinthisstudy
Exposurefrequency(EF) 350 days/year J EA (1999)
Carcinogens Bodyweight (BW) 50 kg J EA (1999)
Averagingtimefor 70 year J EA (1999)
carcinogens(ATc)
Slopefactor (SF0) 0.011 [mg/(kgday)]
1
: TCE U.S.EPA (1996)
Ingestionrateof water (IRw) 2 L/day J EA (1999)
Exposureduration(ED) 3010 year Assumedinthisstudy
Exposurefrequency(EF) 350 Days/year J EA (1999)
*USEPA: U.S. Environmental ProtectionAgency, J EA: J apanEnvironmental Agency
Figure4. Flowof adecisionsupport system.
remediationwork andprotectinghumanhealth. Con-
cept of decision support systemis shown in Fig. 4.
Human health risk assessment was calculated based
ontheASTM RBCA model. And, environmental risk
andrational pumpingrateisexpressedbyfuzzymem-
bershipfunctions. Thesamplesof theriskcalculation
parametersareshowninTable2.
6 SUMMARY
The concept of new fuzzy inference model, which
is effective for groundwater contaminated with low
levels of VOCs, was proposed. In this remediation
technique, remediationefficiency andenvironmental
impact are considered in the remediation period by
fuzzyinference.Theproposedmethodmaycontribute
for low-cost and low-risk remediation works. Appli-
cationtoactual siteandmodificationof basicconcept
shouldbeneededtofurther development.
REFERENCES
ASTM (2000) Standard guide for risk-based corrective
action, ASTM E 2081-00.
J apan Environmental Agency (1999) Survey and counter-
measureguidelinesfor soil andgroundwater contamina-
tion, Geo-Environmental ProtectionCenter.
Mamdani, E.H. (1974) Applicationsof fuzzyalgorithmsfor
control of simpledynamicplant, Proc. IEE, 121, 12, 1588.
Mamdani, E.H. (1976) Advances inthelinguistic synthesis
of fuzzy controller, Int. J. Man-MachineStudies, Vol.8,
No.6, pp. 669679.
Morisawa, S. (1991)Optimumallocationof monitoringwells
aroundasolid-wastelandfill siteusingprecursor indica-
tors andfuzzy utility functions, J ournal of Contaminant
Hydrology, Vol.7, pp. 337370.
U.S. Environmental ProtectionAgency (1996) Soil screen-
ingguidancetechnical backgrounddocument, EPA/540/
R95/128.
Zadeh, L.A. (1968) Fuzzyalgorithms, informationandcon-
trol, Vol.12, pp. 94102.
Zadeh, L.A. (1973) Outlineof anewapproachto theanal-
ysis of complex systems and decision processes, IEEE
TransactionsonSMC, SMC-3, Vol.1, pp. 2844.
220
A riskevaluationmethodof countermeasurefor slopefailureand
rockfall withaccount of initial investment
T.Yuasa& K. Maeda
Graduate School of Civil Engineering, Nagoya Institute of Technology, Japan
A. Waku
Ltd. Takashima technology central adviser, asset management department general manager
ABSTRACT: InJ apan, withtheasset management andtheadvancement of risk management research, soil
structures, suchasslopes, arenowbeingtakenintoconsiderationasinfrastructureproperties. Inordertoexecute
effectivecountermeasurefor slopefailureandrockfall, it isnecessarytoquantityeffectsof countermeasureby
numerical analysis,suchasDEMrockfall simulation.And,atpresent,methodforevaluatingtheinitial investment
incountermeasureshasnot yet beenestablished. Toallowfor thedeterminationof anappropriateinvestment,
weproposeanewmethodof calculatingtheinvestment basedonsloperiskinthispaper.
1 INTRODUCTION
In J apan, with the development of risk manage-
ment technology, earthstructures, suchasslopesand
tunnels, are now being taken into consideration as
infrastructureproperties. Inthisframework, thereare
circumstances in which it is necessary to execute
slopemeasuresmoreefficiently. Thisisbecauselocal
governments face strict financial limitations, while
thenumber of dangerous slopes is increasing dueto
expandingroadnetworksinmountainareas.
Slope risk management presumes the slope risk
valuebeforehand, andsupportstheexecutionof strate-
gic measures corresponding to the slope risk value.
Some risk quantification techniques have been pro-
posed,andinpart,thesehavebeenstudiedincreasingly
for business use with GIS. However, many large
problemsshouldbeaddressed.
The first problem is improving the precision of
slope risk value. This is particularly the case with
the countermeasures; it is necessary to adequately
reflecttheirpositiveeffectinthesloperisk.Atpresent,
this effect has not beenquantifiedandanevaluation
technique to quantify the above effect has not been
established.
Second, anevaluationtechniquefor calculatingthe
investmentfornewslopemeasureshasnotbeenestab-
lished. Inalotof infrastructureproperties, investment
appraisal is executed, includingcost-benefit analysis
(B/C). However, amethodto evaluatethevalidity of
slopemeasurecost hasnot beendetermined. Thisisa
critical problemfor efficient investment.
The purpose of this paper is, first, improving the
precisionof sloperiskquantificationwithslopemea-
sures. Second, the paper proposes a new method to
evaluatethevalidityof theslopemeasurecost.
Inthisstudy, deal withslopefailureandrockfall as
aslopecollapsedisaster at thesametime. Probability
distributionsareusedinamechanical stabilityanalysis
model to quantify thesloperisk after eachoccurring
mechanismis presumed. Moreover, weusetherock-
fall simulation DEM (Discrete Element Method) to
quantifytheeffect of counter measures.
2 QUANTIFICATIONOF SLOPE RISK
2.1 Definition of slope risk
Inorder to quantify thesloperisk R, R is definedas
follows.Thisshowstheexpectedvalueof amountlost.
wherep is theprobability of slopedisaster, D is the
amount lost duetodisaster.
2.2 Calculation of occurrence probability
Weindicatethecalculationmethodfor eachprobabil-
ityinthissectionbecausetheoccurringmechanismis
different betweenslopefailureandrockfall.
(1) Probabilityof slopefailure
The calculation method for slope failure probability
is dividedroughly into astatistical andamechanical
technique. Here, we will use the mechanical model
that Ohtsu proposedthemethodwhichis usedslope
stabilityanalysis.
In this method, first, rainfall is assumed to bean
exogenous factor of slope disaster. We calculate the
slopefailureprobabilityp
a
fromtheexcessprobability
of annual probablerainfall () in rainfall intensity
221
Figure1. Infiniteslopemodel.
andtheslopefailureprobability p
f
in caseof that
probablerainfall.
If it issupposedthat therainfall hazardfollowsthe
Gumbel distribution, theexcessprobability() will
bethefollowing.
Inthisequation, a andb areconstantnumbersobtained
fromtherainfall history.
Next, it is supposed that a collapse type is infin-
ity slopemodel (Fig. 1) in order to calculatethep
f
.
Then, thesafetyfactor of theslopestabilityanalysisis
indicatedasfollows.
where is thesoil unit weight,
w
is theunit weight
of water, H isthethicknessof theslidinglayer, H
w
is
groundwater level, istheslopeangle, c and arethe
soil strengthparameters.
Moreover, it isnecessarytorelate andH
w
, sowe
decidedto useEq. (4) fromreference. (J CCA kinki,
2006)
Now, toquantifyslopefailureprobability, wecalcu-
latetheprobabilityof F -1 becausethisshowsthe
unstable(collapse) condition. To do this, weassume
thesoil cohesionc andinter frictionangle toberan-
domvariablesaccordingtoanormal distributionasin
Eq. (5). Thisshowstheuncertainlyof soil parameters
inphysical terms.
As a result, the slope failure probability p
f
is
calculatingfromthefollowingequation.
Figure2. Typesof rockfall.
Therefore,theslopefailureprobabilityp
a
isasfollows.
Thefeatureof thismethodisthatprobabilitydistri-
butionis takeninto themechanical stability analysis
model. However, itisdifficulttodeterminethevolatil-
ityof therandomvariable.Inaddition,itisnecessaryto
notethatweuseasimpleequationfor easeof calcula-
tion, thoughitisknownthattheoccurringmechanism
isinfluencedbythesaturation, groundwater level, and
soon.
(2) Probabilityof rockfall
Theoccurringmechanismof rockfall isverycomplex
andhasnotyetbeenclarified.However,thecalculation
methodforrockfall probabilityisalsoroughlydivided
into astatistical andamechanical technique. Weuse
themechanical model that Okimura proposed.
Fig. 2showsthecasesof theoverhangtypeandfall
off typeof rockfall, andthesafetyfactor equationfor
eachtype.Then,therandomvariablesc and aretaken
intotheseequations. Itisassumedthattheprobability
fromF number of fallsbelowone(N
) isdividedinto
thenumberof all trials(N)byMonteCarlosimulation.
Thismechanical model isverysimple, soit iseasy
toexecute.Ontheotherhand,itseemsthatendogenous
factors such as geological situation and exogenous
factors suchas rainfall, snow, freeze-thaw, wind, and
earthquakes cannot bereflectedinthemodel. There-
fore, thesepoints will becomeresearch topics in the
future.
2.3 Calculation of loss amount due to disaster
Itthisstudy,wedefineDasthesumof D1,thepersonal
loss; D2, theroadrestorationcost; andD3, thetraffic
detour loss. In addition, theloss amount is changed
bywhether thereareexistingcountermeasuresor not.
However, this section does not deal with the effects
of countermeasures; thesedetailswill begiveninsec-
tion4. Collapsetypes, whichareneededtocalculate
D1, aremodeledasinFig. 3.
222
Figure3. Typeof collapse.
(1) Personal loss, D1
Incaseof slopefailure, itcalledaburiedcasewhen
sand exceeds the height of a car; the buried case
seems to result in death, incurring the human loss
I (yen) in this case. The loss amount decreases lin-
early inrelationtothereductionof thedisaster level.
Incaseof rockfall, it seemsthat thecaseof beingsit-
uated right under thefalling rock results in death,
and this is calculated in the same manner as before
(Fig. 3).
Thenumber of victimsiscalculatedfromthedaily
trafficvolumeof theobjectroadandtheaveragenum-
berof passengers. Furthermore, thenumberof victims
iscalculatedseparatelyinrelationtocompactcarsand
large-sized cars in order to reflect the difference in
thecars height. Incidentally, thehumanloss usedis
I =29,764,000(yen), asobtainedfromthesurvey of
theJ apaneseCabinet Office.
(2) Roadrestorationcost, D2
Ineachcaseof slopefailureandrockfall,weuseEq.(9)
to calculate D2 (Yen). This equation is a regression
functionbetweentherestorationcostandthearchived
volumeof sandV(m
3
) basedonpast disaster records.
(PWRI, 2004)
(3) Trafficdetour loss, D3
Thetraffic detour loss refers to theclosing of roads
whenslopefailureor rockfall occurs, andD3consists
of twokindsof loss, theCostlossof timegenerated
by increases of running timeand theRunning cost
loss generatedbyincreasesinmileage. Thesecanbe
calculatedaccordingto themileagedistanceandthe
dailytrafficvolume.
Recovery time has a big influence on the detour
loss, and Fig. 4 shows relations between the recov-
ery timeN(days) andthecollapsevolumeV(m
3
). In
this study, weusetheregressionfunctionas givenin
Eq. (10) However, it seems that extensivelosses that
arenot reflected in theequation occur if thereis no
detour or theevent occursnear anisolatedvillage, so
it is necessary to evaluateD3ineachslopefor these
cases.
Figure4. Relationshipbetweencollapsevolumeandrecov-
erytime(Kohashi, Het al. 2007).
Table1. Slopefailureconditions.
Slopeno. 1 2 3 4 5
V(m
3
) 2,100 600 900 800 1,500
(deg.) 30 35 30 35 25
(deg.) 41 38 43 36 38
(kN/m
3
) 18 18 18 18 18
c (kN/m
2
) 16.7 2.7 10.4 4.5 13.2
H (m) 3.5 1 2 1.5 2.5
Trafficvolume 4,502 4,502 4,502 4,502 4,502
(/days)
Mixratesof 10 10 10 10 10
large-sizedcar (%)
Distanceof original/ 20/40 10/25 15/25 15/50 15/50
detour road(km)
Speedof original/ 50/30 50/30 50/30 50/30 50/30
detour road(km/h)
3 RESULTSOF CASE STUDY
Inthissection, thecalculationsof sloperiskusingthe
abovemethodareshown. Inthispaper, weset the10
slopeconditionsasinTable1.Thesearenotreal slopes,
butwecanassesswhatfactorsinfluencetheresults. In
addition, probablerainfall isbasedonrainfall history
for 19452006intheTakayamaGifuobservatory.
Previously, therewereno methods of indicatinga
sloperisk whichis expressedby themonetary value
intermsof slopefailureandrockfall atthesametime.
Now, however, we indicate that each type of slope
risk canbeevaluatedsimultaneously by theindex of
sloperisk. Moreover, itisunderstoodthatthecollapse
probability by useof stability analysis that has been
used does not necessarily correlate the slope risks.
This is because the risk includes not only an index
that shows thedanger of slopecollapsebut also that
considerstheamountlosswhenthedisasterwill occur.
Forinstance, comparingslopeNo. 3andNo. 9, thecol-
lapseprobabilityisalmostatthesamelevel. However,
it is understoodthat slopeNo. 3has doubleor more
theriskof slopeNo. 9(Fig. 5).
223
Table2. Rockfall conditions.
Slopeno. 6 7 8 9 10
Typeof collapse* (ov) (ov) (rf) (rf) (rf)
Weight of rock 12.9 2.76 6.9 86.3 4.1
W(kN)
Angleof sliding 80 95 65 50 55
surface (deg.)
Lengthof sliding 0.8 0.4 0.5 2.5 0.2
surfaceY (m)
Lengthof crack 0.0 0.2
LY (m)
(deg.) 35 35 35 35 35
c (kN/m
2
) 15 15 15 15 15
Trafficvolume 4,502 4,502 4,502 4,502 4,502
(/days)
Mixratesof 10 10 10 10 10
large-sizedcar (%)
Distanceof original/ 10/25 20/30 20/70 5/20 10/30
detour road(km)
Speedof original/ 50/30 50/30 50/30 50/30 50/30
detour road(km/h)
*(ov)OverhangType(rf)Rockoff Type.
Table3. Resultsof sloperiskanalysis.
Slopeno. p
a
D1 D2 D3 D R
1 0.0621 1,446 2,157 2,258 5,862 364
2 0.9960 57 714 807 1,578 1,571
3 0.2001 803 1,002 761 2,567 515
4 0.3384 0 906 1,437 2,343 793
5 0.1041 1,307 1,580 2,719 5,605 583
6 0.3870 566 137 485 1,188 460
7 0.4650 566 136 463 1,165 542
8 0.0200 566 136 976 1,678 34
9 0.2050 566 140 466 1,172 240
10 0.0350 566 136 549 1,252 44
Figure5. Resultsof sloperiskandfailureprobability.
A methodology to determine the priority level of
slopemeasureshadnot beenclarified. However, it is
thought that theseprioritylevelscanbedecidedmore
reasonably by arranging slopes in order in terms of
sloperisk value. (Fig. 6) Thisalsomeansthat wecan
order slopes in terms of their impact on society. In
Figure6. Priorityof executingslopemeasures.
addition,itseemsthatsloperiskcanbeusedasameans
of determining theaccountability for residents as to
whenslopemeasureswill beexecutedunderthebudget
reductions.
4 QUANTIFICATIONOF SLOPE MEASURE
EFFECT
Inthissection, itdescribedhowtoquantifytheeffects
of countermeasure that either exist or will be built
up. Inthis study, it pays attentionto therockfall dis-
aster; the effect of countermeasure is quantified by
calculatingarockfall behavior usingtwo-dimensional
DEM.
4.1 Concept of DEM rockfall simulation
DEM is a numerical analysis method used to solve
eachelement progressivelyinanindependent motion
equation. At present, this methodis most commonly
usedfor rockfall simulations, inJ apan.
Thegroundslopeisanactual sectionof areal site,
andgroundsurfaceisapproximatedby asinglelayer
of someparticles. Often, groundslopeisexpressedby
particleassembly.Inthisstudy,onlyonelayerwasused
becauseitreducesthelengthyanalytical timerequired
tocalculatemanycases. Detailswill beprovidedlater.
Basedonpreliminarysurveys, thelocationof rock-
fall generation was determined. To simplify calcula-
tions, rock particle was assumed to have a circular
shape. Furthermore, theshapeandlocationof counter-
measure, suchasaretainingwall, isset uparbitrarily.
Thisrepresentsthesituationof howtoestablishanew
countermeasure.
To quantify theeffect of slopemeasure, it is nec-
essary to judgewhether theroadwouldbestruck as
aresult of rockfall simulation. Twocasesarethought
toserveasjudgmentstandards, andarerepresentedin
Fig. 7andTable4. Incase(B), thejudgment standard
ispresumedtobeanamount greater thanthepossible
absorptionenergyof thecountermeasure.
In order to describe uncertain rockfall behavior,
causedbyinitial conditions, thegenerationlocationis
regularlychangedandrockfall simulationisexecuted
38times.Asaresult, theprobabilityof beingstruckby
224
Figure7. J udgment standardsfor roadbeingstruck.
Table4. J udgment standardsfor roadbeingstruck.
(A) A rockexceedstheretainingwall
(B) A rockdestroystheretainingwall
Table5. Analytical parametersof DEM.
Springconstant (normal) 5.010
6
Springconstant (shear) 5.010
6
1,4
Dampingfactor (normal) 0.3
Dampingfactor (shear) 0.3
Coefficient of particlefriction 0.477
therockfall iscalculatedasthenumber of timeseither
judgment standard(A) or (B) is met dividedinto the
total numberof simulations. Itisalsopossibletoverify
rockfall behavior width.
Inaddition, anothermethodof describinguncertain
rockfall behavior exists. It istheprobability distribu-
tion that takes into analytical parameters of rockfall
simulation. Becausetheinfluenceof analytical param-
etersonfallingrockbehaviorhasnotyetbeenclarified,
theformer methoddescribedisusedinthisstudy.
4.2 Results of DEM rockfall simulation
Analytical parameters DEM are set according to
Table5.
Fig. 8representsoneexampleof asimulationresult.
The solid line shows tracks of two or more falling
rocks. Thefiguresscalesareequivalent totheactual
site. Inorder todeterminethenumber of timesjudg-
ment standards(A) or (B) arefilled, it isnecessaryto
obtaintwonumbers. First, thenumber of timesrock-
fall height exceedstheretainingwall height of 2.0m;
and, second, thenumber of timesthat rockfall energy
exceeds thepossibleabsorptionenergy of theretain-
ingwall. Tobemorespecific, thepossibleabsorption
Figure8. Resultsof simulation(wall height=2.0m).
Figure9. Resultsof simulation(wall height=2.5m).
energy isset at 300kJ, andit isassumedthecounter-
measureisdestroyedif itreceivedmorethan300kJ of
kineticenergyinthex-direction.
In this study, that number was 21; therefore, the
probability of being struck by rockfall is estimated
to be55%. Next, it is necessary to calculatetherisk
decrease rate in order to correlate this result to the
sloperisk. Inthiscase, risk decreaserateisestimated
at 45%, calculatedas(10055)%.
In addition, rockfall simulation may be used not
only to examinetheeffects of countermeasures, but
also to decidethebest scaleandshapefor them. For
instance, Fig. 9 represents the results of changing
theheight of theretainingwall, by only 0.5m, when
exactly thesamesimulation as beforewas executed.
Asaresult, theriskdecreaserateincreasedto70%.
Theriskdecreaserate, relativetothearbitraryscale
and shape, is estimated by DEM rockfall simulation
bothinexistingandnewlymeasuredcases.
4.3 Assignment of rockfall simulation
DEMisusedmostcommonlyforrockfall simulations;
however, it is not clear howmuchit is influencedby
analytical parameters, the effects of talus, and rock
shapesoforth. Especially, theeffectsof taluswhichis
depositedmaterialsor weatheredslopesurfaceswhen
225
Figure10. Meaningof sloperisk.
falling rock is digging into them, how breakage of
fallingrockwhentherockiscrushedandotherimpor-
tant elements have on rockfall behavior. Moreover,
when afalling rock is modeled as acirclein 2D, or
aspherein3D, theinfluenceof shapeandinteraction
betweenrotational andtranslational motionneeds to
beconsidered.
Thepointsthathavebeendescribedsofarrepresent
problemstosimulaterockfall behavior adequately. At
the same time, however, energy dissipation effects
suchas rock crushinganddigginginto thetalus, for
instance,mustbeconsideredwhendesigningeffective,
natural countermeasures. Therefore, it is important
to continueto improvetheprecision of categorizing
fallingrockbehavior tobeabletoachievemoreeffec-
tiveyet inexpensivemeasures. It will alsobepossible
tomeasuretheeffect of slopeandtoevaluateinvest-
ment decisionsbyusingtheriskdecreaserate. Thisis
describedfurther inthenext section.
5 EVALUATIONMETHODOF INVESTMENT
FOR COUNTERMEASURES
Inthis section, weproposeanevaluationmethodfor
theinitial investmentfor countermeasures.Atpresent,
atechniquefor evaluatingthevalidityof countermea-
sure cost has not yet been established. The biggest
reasonfor thisisthatitisquitedifficulttoforecastthe
degreesof losswhenadisaster occursandtoquantify
the effect of damage reduction by countermeasures.
However, as seen up to now, it is becoming feasible
to usesloperisk for theformer andquantification
of slopemeasureeffect for thelatter. Thus, wepro-
poseanevaluationmethodfor thevalidityof aninitial
investment for countermeasuresbasedonsloperisk.
5.1 Slope LCC (Life Cycle Cost)
To examinetheamount of investment, wemust con-
sider the cost generated during the use period, and
theconcept of LCC is necessary for this, as well as
ageneral infrastructure.
Thedamagecost ontheslopeis generatedonly at
thedisaster points, asshowninFig. 10(Left). Onthe
other hand, slope risk shows the expected value per
year of the cost that may be generated, as shown in
Fig. 10(Right). Inother words, it canbeassumedthat
Figure11. Comparisonof thetwotypesof LCC(w=65%).
thesloperisk is theslopes cost per year. Therefore,
slope LCC canbecalculatedbyintegratingsloperisks
for theuseperiod. Inaddition, it ispossibletoadjust
theevaluationtechniquesrelatedtoinvestmentamount
that havebeendevelopedwithgeneral infrastructures
bytheconceptof slope LCC. Therefore, slope LCC is
definedasfollows.
where R
i
is the slope risk of the ith year, N is the
useperiod, C
0
is theinitial investment, w is therisk
decrease rate by investment, and r is the J apanese
social discount rate(4%).
TheOM cost (operationandmaintenancecost) is
notincludedinEq. (11) becauseweassumethatslope
checkandresearchcostscanbedisregardedcompared
withsloperisk. Inaddition, itisnecessarytopresume
w based on the DEM simulation result indicated in
section4, becausethis differs dependingonthetype
of measurement andscalesineachslope.
5.2 Evaluation index of investment, W
In order to examine the amount of the investment,
it is necessary to compare LCC
in case of making
theinitial investment (countermeasureisexecuted) to
LCC in the case of not making it (unmeasured). If
LCC
-LCC, it will bejudgedthat It isnecessaryto

executetheproject. This means that thestandardof
theevaluationhastodowiththerelationshipbetween
bothLCCs, asshowninFig. 11.Therefore, wepropose
anewindexW asinEq. (12) thatpaysattentiontothe
ratio of both LCCs in order to simplify investment
decisions.
Withtheindex of W, theproject canbejudgedIt
is necessary toexecuteit if W is positive, or canbe
judgedShouldnot executeit if W isnegative.
Table6showstheexampleof investment decisions
when some measures plan is shown. The decision-
makercandeterminethemosteffectiveplanbychoos-
ingtheplaninwhichW hasthebiggest value. Inthis
case, Plan (B) is themost effectiveinvestment plan.
226
Table6. Exampleof investment judgment for eachplan.
Non Plan(A) Plan(B) Plan(C)
Measurescost 0 100 200 400
Riskdecrease 0 50% 65% 90%
rate(%)
LCC or LCC
500 350 375 350

(100+250) (200+175) (400+50)
W (%) +20 +25 +10
Table7. Volatilityfactorsinthefuturesloperisk.
Volatilityof (A) Volatilityof probablerainfall
probability (B) Volatilityof geometrical parameter
Volatilityof (C) Volatilityof trafficvolume
lossamount
Figure12. Trafficvolumepredictionbyuseof Arithmetic
Brownianmotion.
And, for thefurther discussion, wecall initial invest-
ment when becoming W =0 Amount of the limit
investment, C
sup
.
5.3 Evaluation of investment amount considering
the risk volatility
In the previous section, the slope risk volatility is
not considered. However, it is necessary to consider
therisk volatility accordingtopassageof timeinthe
use period when we calculate slope LCC. Then, we
describetheinvestmentamountevaluationinthecon-
dition of uncertainty. The risk volatility factors are
shown in Table 7, but the governing factor is (C).
Therefore, inthisstudy, weassumethat(C) istheonly
volatilityfactor inthefuturesloperisk.
Then, it is necessary to model the traffic volume
in the future. This estimation is usually carried out
accordingtothesomescenariobasedonforwardroad
planning, but this has been criticized in that there
is uncertainty in prediction and a constant width in
thepredictivevalue. Becauseof this, weconsidered
the annual volatility of traffic volume to be Arith-
metic Brownian motion, which is used in the field
of theFinancial Engineering; this is expressedusing
Figure13. Result of Slope LCC.
Figure 14. Relationship between Wand C
0
for
decision-maker.
theMonteCarlosimulation. Fig. 12showsoneexam-
ple of these results. In addition, the feature of the
arithmeticBrownianmotionisitsaccordancewiththe
Markovprocessinthatthefollowingpresenttermtraf-
ficvolumedependsonlyonthetrafficvolumebefore
oneterm.
LCC can be fined when the traffic volatility is
given; theresultisshowninFig. 13.Theresultdemon-
strates that LCC
50
shows distribution with a certain
widththat centers onthemeanvalue. Here, wecon-
sidertheconfidenceinterval of distributiontobe90%,
and it is assumed that LCC
DownSide
50
when the cumu-
lative probability is 5% and LCC
DownSide
50
when the
cumulativeprobabilityis95%. Fromthis, wecancal-
culatetheAmount of thelimit for investment.With
this result, the W and initial investment C
0
relation
are led as in Fig. 14. It can be determined whether
the initial investment under examination is in the
safety zone or in a dangerous zone, and the width
of the middle zone generated by the uncertainty of
the traffic can also be understood. For instance, if
it was proposedtobuildnewly countermeasurefrom
150,000thousandyen, decision-maker canevaluated
thatthisproposal isnoteconomical butisnotfar apart
fromC
sup
. And if it was proposed more expensive
plan, they might requireto changethis measureplan
fundamentally.
Fromthis, it isbelievedthat theinvestment evalua-
tionfor uncertainconditionswill becomepossibleby
theindexof W.
227
6 CONCLUSION
In this paper, it has been made clear that slope risk
can bequantified both in terms of slopefailureand
rockfall when slope risk management is executed.
Furthermore, it has beendemonstratedthat it is pos-
sible to reasonably determine the priority level of
countermeasuresbysloperisk.
Moreover, in order to quantify theslopemeasure
effect, rock fall simulation by DEM is an effective
method.Anditisabletocalculateriskdecreaserateof
countermeasure by judging the road struck in many
calculations.
Wehaveproposedanewmethodfor evaluatingthe
validityof aninitial investmentbyleveragingtheindex
W basedontheconceptof slope LCC. Inaddition, this
methodwill beabletosupportdecisionmakingunder
uncertainconditions, asithasbuiltintheriskvolatility
of trafficvolumeinthefuture.
Futuretasks includedetermining howto treat the
statistical andmechanical errormarginforeachmodel.
Furthermore, therequiredriskprecisionisdifferentin
the steps of risk management as compared to those
of decision-makers, so it is necessary to construct a
complete systemof risk management. For instance,
the decisions of measuring priority level are based
onrelativeriskevaluations; ontheother hand, invest-
ment decisions are based on an absolute risk value.
Inthiscase, ahighlevel precisionisneeded. Briefly,
it is necessary to clarify therisk quantificationtech-
niquecorresponding to thenecessary risk precision.
In addition, in relation to the rock fall simulation
DEM, etc., futureresearchtoimproveitsprecisionis
necessary.
REFERENCES
Cundall, P.A. 1971. A computer model for simulating pro-
gressive large scale movements in blocky rock system,
Proceedingsof theSymposiumof theInternational Soci-
etyof RockMechanics(Nancy, France), Vol. 1, NoII-8
J apanese civil engineering consultants association Kinki
branch, J CCA kinki, J uly2006. Introduction of deteriora-
tion concept for slope stability evaluation. (inJ apanese)
J apanese Society of Civil Engineers, 2003, Guidebook on
Traffic Demand Forecasting, Vol. 1(inJ apanese)
Kohashi, H., Kato, S., Ishihara, H & Furuya, A. September
2007. Evaluation of recovery time of road slope disas-
ter, J SCE 2007Annual meeting, Hiroshima: 877878(in
J apanese)
Komura,T.,Muranishi,T.,Nishizawa,K.&Masuya,H.2001.
Impactof fallingrockonfieldslopesusedinrock-fall sim-
ulationmethod, Journal of Structural Engineering, J CSE,
VI-47A: 16131620. (inJ apanese)
Okimura, T., Torii, N., Hagiwara, S. &Yoshida, M. 2002. A
proposal for risk evaluationmethodfor rock fall onroad
slope, Journal of the Japan Landslide Society, Vol. 39,
No. 1, J une: 2229(inJ apanese)
Otsu, H., Onishi,Y., Nishiyama, S. &Takeyama,Y. 2002.The
Investigationof RiskAssessmentof RockSlopesConsid-
eringtheSocioeconomic LossduetoRock Fall, Journal
of JSCE, No.708/III-59: 187198. (inJ apanese)
Otsu, H., Supawiwat, N., Matsuyama, H. & Takahashi, K.
2005. A Study on Asset Management of Road Slopes
Considering Performance Deterioration of Groundwa-
ter Counter-measurement System, Journal of JSCE,
No.784/VI-66: 155169. (inJ apanese)
Public works research institute, PWRI. 2004. Manual that
supports risk analysis and management for road slope
disaster (idea). (inJ apanese)
Usiro, T. 2001. Rockfall numerical simulation software
DRSP, Kochi Prefecturebridgeassociation. (inJ apanese)
228
Riskassessment ontheconstructionof interchangestationof Shanghai
metrosystem
Z. W. Ning, X.Y. Xie& H. W. Huang
Key laboratory of Geotechnical and Underground Engineering of Ministry of Education,
Tongji University, Shanghai, P. R. China
Department of Geotechnical Engineering, Tongji University, Shanghai, P. R. China
ABSTRACT: Astherapidgrowingof Shanghai metrosystem, risksalongwiththeconstructionof interchange
stationaregivengreatconcerns. Inthispaper, anoverviewof interchangestationin2012Planisfirstpresented
bythoroughstatistics. Thenboththeobjectiveandthesubjectiveriskfactorsof theconstrucitonof interhchange
stationarewell analyzed. Finallyariskassessmentmodel isestablishedbyusingtheFuzzySyntheticEvaluation
MethodandtheAnalyticHierarchyProcesstoevaluatetherisk level of the4-lineCenturyAvenueinterchange
station.
1 INTRODUCTION
The Shanghai metro system which currently oper-
ates 8 lines with a total length of 228.4kmis one
of the fastest growing metro systems in the world.
Accordingtotheplan, thesystemwill include13lines
with a total length over 500 kmin 2012. Therefore,
Shanghai metro systemhas stepped into aperiod of
network-scaleconstructionfromprevious single-line
scale construction, which inevitably raises new risk
issues.
Among those, the risks along with the construc-
tionof interchangestationsaregivengreat concerns.
Generally speaking, their morecomplicatedconfigu-
rationcomparedwithnormal stationswouldincrease
the chances of failure when excavation, and there-
forecausesevereconsequences as they arenormally
located on central spots of the city. Moreover, a
large potion of interchange stations are constructed
by expanding or reconstructing theexisting stations,
which brings about great threatens to the safety of
the operating metro lines. Apart fromthese techni-
cal concerns, poorly risk management insite, lack of
experiencedcontractorandqualifiedpersonnel aswell
asequipment problemsareall thepotential causesfor
accident or delayintheprocessof interchangestation
construction.
2 STATISTIC OF INTERCHANGE METRO
STATIONINSHANGHAI 2012PLAN
According to the 2012 Plan, nearly 60 interchange
stations will be built. Before risk analysis, we con-
ducted statistic study to obtain an overviewof these
interchangestations insomeaspects as stationscale,
interchangemodeandconstructionmode.
Figure1. Distributionof interchangestationswithdifferent
scale.
2.1 Statistic study on station scale
Thereare59interchangestationsaccordingtothe2012
Plan. Among those, there are 43 2-line interchange
stations, 15 3-line interchange stations and 2 4-line
interchange stations, making up 73%, 25% and 2%
respectivelyintotal.
Figure1showsthedistributionof interchangesta-
tions with different scale among all the 13 metro
lines.
It is found that Line 4 has the most interchange
stationsasitisacircularlinethatintersectswithmany
other lines. The early built lines (Line 1 to Line 4),
whichmakeupof theframeworkof theShanghai metro
system, consistof morelargescalestations(3-lineand
4-line) thanthosenewly built or inplan. It indicates
that,inthefollowingfewyears,therearemanystations
inoldlinesneedtobeexpandedor reconstructedfor
interchangefunction.
229
Figure 2. Distribution of interchange pairs with different
interchangemodes.
2.2 Statistic study on interchange mode
InShanghai metrosystem, therearethreeinterchange
modesas: parallel-interchange, cross-interchangeand
passage-interchange(interchangingviapassagecon-
nectingseparatestationsof different lines).
For every pair of two different lines in an inter-
changestation, they belong to atypeof interchange
mode. Thus, whenwecalculatingthenumber of inter-
change modes, there are C
n1
n
interchange pairs for
a n-line interchange station. Among all the inter-
changepairsinthoseinterchangestations, thereare18
pairsof parallel-interchangemode, 34pairsof cross-
interchangemodeand42pairsof passage-interchange
mode, makingup19%, 36%and45%respectivelyin
total.
Figure2showsthedistributionof interchangepairs
with different interchange modes among all the 13
metrolines.
From Figure 2, high percentage of passage-
interchangemodeisfoundinLine1, Line2andLine
3. That isbecauseacompletemetronetwork hadnot
formed yet in early age of metro construction and
less considerations for futureinterchanging needs in
design. Under thiscircumstance, passage-interchange
becomes a low cost and low risk option for exist-
ing lines to connect with new lines. In those newly
builtlinesorlinesstill inplan, theparallel-interchange
andthecross-interchangemodes arewidely adopted
with higher efficiency and more convenience for
passengers.
2.3 Statistic study on construction mode
As old saying goes that, Rome was not built in one
day, it isimpossibletobuildall themetrolinessimul-
taneously. Besides, the alignment of lines could be
adjustedafter original planning. Therefore, thereare
currentlythreedifferent constructionmodesfor inter-
changestationsas: reconstruction, built at stagesand
builtatonetime. Forthereconstructionmode, itrefers
toreconstructingor expandingastationwithoutinter-
changingfunctiontoaninterchangestation. Astothe
builtatstagesmode, itmeanstheinterchangestationis
designedasawholecomplexwhilethesectionsbelong
Figure3. Distributionof interchangestationswithdifferent
constructionmodes.
to different lines arebuilt at different stages. For the
built at onetimemode, it literallymeansthesessions
belongtodifferent lineswithinaninterchangestation
arebuilt at onetime.
Among all the interchange stations, there are 24
stationsof reconstructionmode, 23stationsof builtat
stagesmodeand12stationsof builtatonetimemode,
makingup41%, 39%, 20%respectivelyintotal.
Figure3showsthedistributionof interchangesta-
tionswithdifferent constructionmodesamongall the
13metrolines.
Figure 3 demonstrates high percentage of recon-
structionmodeinearly built lines as Line1, Line2,
Line3andLine4. It is also for thereasonthat there
was less consideration for futurenetwork expansion
whenthesestationsweredesigned. Astointerchange
stationsinnewlines, exceptingforthoseintersectwith
existinglines, theyaremostlybuiltatstagesor builtat
onetimewithanoverall interchangingdesignconcept.
3 RISK ANALYSISOF INTERCHANGE
STATIONCONSTRUCTION
In this paper, the authors divide the risk factors of
interchangestation construction into two categories:
objectiverisk factors andsubjectiverisk factors. The
former arecommonly recognizedastechnical or nat-
ural issueswhichaddinherent difficultiesanduncer-
taintiestoproposedprojectswhilethelatterareusually
related to human errors or organizational deficiency
inmanagement, whichmayleadtofailureor delayof
projects.
3.1 Objective risk factors of interchange station
construction
Fromprevious statistic study ontheinterchangesta-
tions of Shanghai metro system, it canbefoundthat
thecomplicated configuration and various construc-
tionmodesareuniquefeaturesof interchangestation
comparedwithnormal station. Toagreat extent, they
would affect the risk level of projects. Besides, as
ground-relatedproject, thegeological conditionsand
thesurroundingenvironmentsshouldalwaysbetaken
intoconsiderationasobjectiveriskfactorsaswell.
230
Table1. Risk level of interchangestation construction in
termsof stationconfiguration
Risklevel Low High
Interchange
mode Passage Parallel Cross
Depth shallow/ medium/ deep
medium deep
Area normal large medium
Layout regular regular complicated
3.1.1 Station configuration
Unliketheuniformconfigurationof normal metrosta-
tions,theinterchangestationsaredistinctinexcavation
depth, excavationareaandplanlayoutduetodifferent
interchangemodes. Therelationshipbetween station
configurationandconstructionrisk will bediscussed
in terms of threedifferent interchangemodes stated
above.
1 Cross-interchange
As to the cross-interchange mode, because the
platformsof differentmetrolinesshouldbelocated
at different levels, thestationsarenormally three-
story or even four-story with excavation depth
rangingfrom20mto25minShanghai. Suchdeep
excavation would increase the chances of failure
of theretainingstructureandexcessdisplacement
of the adjacent ground. Besides, the plan layout
of cross-interchangestations tend to beirregular,
which is likely to result in poor arrangement of
bracingandweak jointsinthecornersof retaining
walls. Finally, if thestationisnot built at onetime,
thelater-built linehas to go up-throughor down-
through the existing line. That would inevitably
bringgreat threatento thesafety of theoperating
line.
2 Parallel-interchange
For the parallel-interchange mode, though the
planlayout of stations areusually rectangular, the
excavationareaaredoubleor eventripleof normal
station, which would bring higher level of risk in
excavation. However, theexcavationdepthcouldbe
well controlledless than20mas theplatforms of
different linescouldbearrangedat thesamelevel.
Similarly, if thestationisnotbuiltatonetime, risk
will increaseduetoavery closedistancebetween
theexistinglineandthelater-built line.
3 Passage-interchange
Becausethestationof passage-interchangemode
actually consists separate stations connected by
passage, its construction risks are the same with
normal station. Evenwhennewlineswill beincor-
poratedlater, theimpact will berelativelylowdue
toafar distancebetweendifferent lines.
The risk level of interchange station in terms of
stationconfigurationissummarizedinTable1.
3.1.2 Construction mode
A uniquefeatureof interchangestation construction
thatdiffersfromnormal stationisthepossibleinterval
between the construction of different parts within a
wholestation. Therefore, therisksaccompaniedwith
eachconstructionmodeasdefinedinsession2should
beanalyzed.
1 Reconstruction
Regardingtothereconstructionmode, theexist-
ing station will be partially reconstructed or
expanded to incorporate new lines. Thus, there
must beloadingor unloadingeffects ontheexist-
ingstructureinducedbyanyconstructionactivities
operated above, beneath or asideit, which would
consequently cause certain response of the early
built station. According to theprotecting require-
ments for operating metro in Shanghai, neither
vertical nor horizontal displacement of thestation
is allowedto exceed20mm. Therefore, extraordi-
naryhighrisksexist inthereconstructionprojects
of interchangestation, especiallyinshanghaishigh
sensitive and saturated soft clay ground condi-
tion. Moreover, rareexperienceinsimilar projects
multipliesthepotential risks.
Up to now, there have been some success-
fully completed reconstruction projects of inter-
change stations in Shanghai of various type as:
Century Avenue Station (parallel/up-through),
Shanghai Indoor StadiumStation(down-through)
andPeoplesSquareStation(parallel). Moredetails
will bepresentedinsession5.
2 Built at stages
Dueto thelimit of availablecontractors, labor
force and equipments as well as the endurability
of the environment and the public, the number
of metro projects carried out during the same
period should becontrolled. So thereareanum-
ber of interchangestationscouldnotbebuiltatone
time. However, for built at stagesmode, theneeds
for future expansion and measures to reduce the
impactsof later-built linesaretakenintoconsider-
ationinstructural designphase. Besides, thejoint
areasconnectingdifferentlinesareusuallybuiltand
carefullytreatedinadvancewiththefirst-builtline.
Moreover, thebidfordifferentlineswithinaninter-
changestationusuallygoestothesamecontractor
to make sure an overall organization and smooth
handover betweenseparatedconstructionstages.
For all thesereasons, thoughtherestill wouldbe
impactsof thelater-builtlinetotheexistingone,the
riskof built at stagesmodeisrelativelylower than
thereconstructionmode. Accordingtothestatistic
above,thismodeisnowwidelyadoptedinnewlines
under construction.
3 Built at onetime
Theconstructionof interchangestationof built
at onetimemodein Shanghai is thetypical deep
excavation project in soft clay areas. Generally
speaking, its risk is lower than the previous two
modeswithsimilarscaleunderidentical geological
andenvironmental conditions.
231
The risk level of interchange station in terms of
constructioinmodeissummarizedinTable2.
3.1.3 Geological conditions
InShanghai thesoft claylayersarewidelydistributed
with low strength, large compressibility and high
water content. The weakest silty clay layer and
clay layer aredistributedamong7to 20munder-
ground where most metro stations are located. It
greatly increases thechanceof failureof theretain-
ing structure and excess ground movement during
excavation.
Theconfinedwaterisanotherimportantrisksource
for deep excavation in Shanghai. It will lead to soil
up-rushat thebottomwhenitspressureisclosetothe
gravity of the above soils. Also, seepage may occur
aroundthefootof retainingwallsduetothewaterpres-
sureandhighpermeabilityof thefinesandlayer.There
arethreeconfinedwater layerswhichmightaffectthe
constructionof metrostationinShanghai asfollows:
thesub-confinedwater in sandy silt layer, the1st
confinedwater insiltyfinesandlayer andthe2nd
confinedwater insandlayer.
Table2. Risk level of interchangestation construction in
termsof constructionmode.
Interchangemode
Constructionmode Passage Parallel Cross
Built at onetime low low/medium medium
Built at stages low medium/high high
Reconstruction medium high high
Table3. Risklevel of interchangestationconstructionintermsof geological conditions.
Risklevel Low High
Confined
water zone ZoneI ZoneII ZoneIII ZoneIV ZoneV
Description thinor missing 1st confinedwater 1st confinedwater 1st confinedwater thicksub-confined
sub-confined layer locatesat layer locatesat layer islocatedat water layer.
water layer. around30m. around30m. shallowlevel. 1st and2nd
1st confinedwater 1st and2ndconfined water isconnected. confined
40mto50m. water isconnected.
layer locatesat
Table4. Risklevel of interchangestationconstructionintermsof environmental conditions.
Risklevel Low High
Protectiongrade 1st-grade 2nd-grade 3rd-grade
Environmental Therearemetro, municipal Thereareimportant pipelines, Thereisnoimportant
conditions commontrenches, gaspipes, buildingsor structureswithin pipelines, buildingsor
mainwater pipes, important therangeof H2Hfromthe structureswithintherange
buildingsor structureswithin excavation. of 2Hfromtheexcavation.
therangeof 0.7H* fromthe
excavation.
*H: excavationdepth.
Based on the depth and the relationship between
different confined water layers, it is divided into
five confined water zones in Shanghai (Liu, 2008).
Risk varies inthesezones for theconstructionof the
interchangestationsasshowninTable3
3.1.4 Surrounding environments
Asthemetronetworkgrowsrapidlyindowntownareas
inShanghai, it increasingly becomeathreatento the
safetyof surroundingenvironmentssuchasbuildings
or infrastructures during construction. Theconstruc-
tionof interchangestationbears especially highrisk
astheyareusuallylocatednear thecommercial center
or other important cityspots.
There is a control criterion of environment pro-
tectioninexcavationof metro constructionbasedon
theengineeringexperienceinShanghai , whichcould
beadopted as areferenceto therisk assessment for
theinterchangestationconstruction(Specificationfor
ExcavationinShanghai MetroConstruction, 2000). In
thisspecification,threeenvironmentprotectiongrades
are set concerning the importance of the adjacent
objectsaswell asthedistancebetweentheobjectsand
theconstructionsite.
3.2 Subjective risk factors of interchange station
construction
Indeed, objective risk factors like technical difficul-
ties and ground uncertainties increase the chances
of project failure. However, whether the failure will
occur or not is to a large extent affected by subjec-
tivefactors likemanagement andpersonal expertise.
Both Sowers (1993) and Bea (2006) have studied
232
a large number of well-documented cases of failing
civil engineeringprojects. Thoughthereisalongtime
spanbetweentheir studies, they got similar findings
that approximately 80%of thefailuresarecausedby
subjectivefactorslikehumanor organizational short-
comings. Asfollows, functioningof riskmanagement
system, contractorexperience, personnel qualification
andequipment conditionarediscussed.
3.2.1 Functioning of risk management system
A well established and effectively functioning risk
management systemisessential for riskcontrol espe-
ciallyinconstructionphase. Veryimportant areshort
andefficient communicationchannelsinconjunction
with clearly defined responsibilities (Savidis, 2007).
Andthemanagement of all informationsuchasmon-
itoring data is the core of risk management system
of geotechnical projects. Unfortunately, onlyapart of
contractorsrealizetheimportanceof riskmanagement
systemandarewillingtoputtheir resourcesintoit. In
manycases, carelessinspectionof monitoringdataor
slowresponsetoemergencyinpoor riskmanagement
systemledtoseriousaccidents.
3.2.2 Contractor experience
Underground projects present distinct regional fea-
tures becausedifferent geological conditions require
differentconstructionmethodsandparameters.There-
fore, contractorswithlessexperienceinsoftclayareas
mightbringabouthigher riskthanlocal contractorsin
Shanghai. Moreover, contractorswithaffluentexperi-
encearemoresensitivewithsignsof potential hazards.
Thus effectivemeasures canbetimely carriedout to
prevent theaccidentsfromhappeningor at least min-
imizethenegativeconsequencesasmuchaspossible.
Whereas, for the reason that the experienced local
contractorsarenot enoughfor large-scalemetrocon-
structioninShanghai, importsof contractorswithless
experienceisinevitable.
3.2.3 Personnel qualification
The metro projects are mostly ground-related with
highriskandstrict technical requirements, whichcan
only be done by specially trained staffs. According
to the terms of relevant codes in China, certificates
arecompulsoryfor staffsparticipatinginconstruction
projects. For large-scaleproject liketheinterchange
metro station discussed here, the project manager
should have the 1st-class manager qualification and
theproportionof seniorprofessional personnel should
behigher than10%inmanaginggroup. However, due
to the booming development of the metro in China,
theshortageof excellent project managers andqual-
ified technical staffs become an urgent issue. There
are situations where contractors employ unqualified
staffs or appoint inexperienced managers, which are
reportedtobethedirect or indirect causesfor lotsof
metroconstructionaccidentsinChina.
3.2.4 Equipment condition
Comparedwithother factors, althoughless structure
failures or casualty aredirectly causedby equipment
problems, it often leads to a delay in construction
procedures. Subject tothepressuresof atight sched-
uleof metro construction plan in Shanghai, overuse
andinsufficientmaintenanceof equipmentsarewidely
foundamongcontractors.
4 RISK ASSESSMENT MODEL
Inthisgeneral riskassessmentmodel fortheconstruc-
tionof interchangestation, aprimary risk valueF
o
is
first introducedconcerningtheobjectiverisk factors.
Then, inordertoconsidertheimpactof subjectivefac-
tors, an adjustment coefficient is deduced as well.
Thefinal riskvalueisobtainedasfollow:
Both F
o
and are calculated by applying the
Fuzzy Synthetic EvaluationMethodandtheAnalytic
HierarchyProcess(AHP) (Liang, 2001).
4.1 Calculation of primary risk value F
o
1 factor set
Thefactor set for F
o
is:
Whereu
1
is stationconfiguration, u
2
is con-
structionmode, u
3
is geological conditions and
u
4
issurroundingenvironments.
2 weight set
Theweight set for eachfactor is:
It canbedeterminedbyAHP method.
3 comment setThecomment set for F
o
is:
Wherev
1
isveryhigh, v
2
ishigh, v
3
ismedium
andv
4
islow. Thevalueassignedtov
1
, v
2
, v
3
, v
4
is4, 3, 2 and1 respectivelyforquantification
of theassessment result.
4 evaluationmatrix
Theevaluationmatrixor F
o
expressesas:
Wherer
ij
isthedegreeof membershipof factor
u
i
torisk level v
j
. R
F
canbeobtainedby member-
ship function, statistic or referring to pre-defined
relationshipbetweenU andV.
5 primaryriskvalueF
o
Theevaluationvector B
F
iscalculatedas:
233
Table5. Risklevel of interchangestationconstruction.
RiskLevel I II III IV
Description low medium high veryhigh
Riskvalue -1.5 1.52.5 2.53.5 >3.5
It gives thedegreeof membershipof thetarget
project to different risk levels. Then primary risk
valueF
o
is:
4.2 Calculation of adjustment coefficient
1 factor set
For thecalculationof , Thefactor set is:
Where u
1
is functioning of risk management
system,u
2
iscontractorexperience,u
3
isperson-
nel qualification andu
4
isequipment condition.
2 weight set
Theweight set for eachfactor is:
It isdeterminedbyAHP methodaswell.
3 comment set
Thecomment set is:
Wherev
1
ispoor, v
2
ismedium, v
3
isexcel-
lent. The value assigned to v
1
, v
2
, v
3
is 1.3,
1 and0.7 respectively for quantificationof the
assessment result.
4 evaluationmatrix
Theevaluationmatrixfor expressesas:
5 adjustment coefficient
isfirst calculatedas:
Itgivesthedegreeof membershipof thecombin-
ingconditionof subjectiverisk factorstodifferent
levels. Thentheadjustment coefficient is:
4.3 Calculation of risk value F
After the primary risk value F
o
and the adjustment
coefficient are both obtained, The risk value F is
determinedbyequation(1). Thefinal risklevel could
refer toTable5asfollows.
Figure4. Overviewof CenturyAvenueinterchangestation.
5 CASE STUDY
5.1 Project introduction
CenturyAvenuestationisa4-lineinterchangestation
for Line2, Line4, Line6andLine9.Theplatformsof
Line2, Line4andLine9areparallel withtheplatform
of Line6go across themabove. Theconstructionof
this station is divided into three phases. Line 2 was
firstlybuilt andput intooperationearlyin1999. Line
4wasbuilt yearslater andput intooperationin2005.
Line6andLine9arenowbeingbuilt synchronously.
Theriskof constructionphase2will beassessedinthe
followingpart.
Theexcavationfor thestationof line4is20.8min
depth with an areaof about 4300m
2
. Theminimum
distancebetweentheretainingwall andtheoperating
line2stationisonly5.4m. Becausenofutureexpan-
sionwasconsideredinpreviousdesignof line2,partof
thestationhavetobereconstructedfor theconnection
withthenewline.
As to the ground profile, the soft clay is widely
distributed30mbelowtheground. The1st confined
water layer is found about 8mto the bottomof the
excavationwithapressureheadabout 20m.
Except for theexistingmetro line, therearesome
commercial buildingsandmunicipal pipelinesnearthe
excavationwithinarangeof 5to10m.
The contractor who carried out this project holds
the top qualification and has accumulated affluent
experience in the underground projects in Shang-
hai. A web-based multilevel field monitoring and
information management systemwas applied to this
project.
5.2 Risk assessment
5.2.1 Risk assessment of object risk factors
Thestatuses and rough risk levels of relevant object
riskfactorsaresummarizedinTable6referringtothe
previousanalysis.
In order to get the weight set and the evaluation
matrix, a survey was conducted among 15 experts
engagedinthegeotechnical engineering.
234
Table6. Risklevel of interchangestationconstruction.
Riskfactors Status Risklevel
Station Depth medium medium
configuration Area medium
Layout regular
Constructionmode reconstruction high
Geological Condition zoneIII medium/high
Surroundingenvironment 1st grade high
The weight set A
F
was determined by the AHP
methodasfollows:
TheevaluationmatrixR
F
wasdeterminedinaway
that,if nexpertsconsidertherisklevel of factoru
i
isv
j
,
thenr
ij
isn/15. For example, after therisk evaluation
of stationconfiguration(u
1
), thenumber of experts
whogavethecommentof veryhigh(v
1
), high(v
2
),
medium(v
3
), andlow(v
4
) is1, 5, 8and1respec-
tively. So, theevaluationvectorR
1
is(1/15, 5/15, 8/15,
1/15). Inthisway, evaluationmatrixR
F
wasobtained
as:
F
wascalculatedaccording
toequation(2):
TheprimaryriskvalueF
o
toequation(3):
So, referringtoTable5, therisklevel ishigh based
ontheassessment of objectivefactors.
5.2.2 Risk assessment of subjective risk factors
TheweightsetA
wasdeterminedbytheAHPmethod
aswell:
TheevaluationmatrixR
wasobtainedinasimilar
waytoR
F
:
toequation(4):
Theadjustmentcoefficientwascalculatedaccord-
ingtoformula(5):
Sotheoverall conditionof thesubjectivefactorsis
excellent.
5.2.3 Final assessment result
Since the primary risk value F
o
and the adjustment
coefficient are both obtained, the risk value F is
calculatedaccordingtoequation(1):
Thus, therisk level of thisproject ismedium but
veryclosetohigh.
Fromtheoverall risk assessment, it is found that,
thoughtherisklevel ishigh consideringtheobjective
factors, theoverall risk level was effectively reduced
to medium by excellent combining condition of
subjective factors, regarding experienced contractor,
effectivemonitoringandriskmanagement infield.
6 COMMENTSONRISK COUNTERMEASURE
By therisk assessment model developedinthefore-
goingsession, all theinterchangemetrostationstobe
built couldbeclassifiedinterms of risk level. How-
ever, theultimatetarget of risk assessment, whichis
for better decision-making to avoid or minimizethe
risksbycorrespondingcountermeasureshasnot been
touched. Heretheauthorsprovidesbrief commentson
thisconcerns
Therisk assessment canbecarriedout inselecting
the suitable contractors during the tendering proce-
dure. For projects with low or medium risk level
only considering the objective risk factors, average
contractors are qualified. While for projects with
high or very high risk level inobjectiveview, top
contractor withexcellent risk management shouldbe
required to adjust the overall risk to a lower level.
When therisk assessment is carried out during con-
struction, special attentionsshouldbegiventoprojects
withhigh andvery high risk level. For high risk,
evaluation of construction methods and intensified
monitoringarerequired. Theveryhigh riskisgener-
allyunacceptable.Underthiscircumstance,temporary
cease of the project may be necessary until sound
countermeasures are achieved by special technical
meetingsor researches.
7 CONCLUSIONS
Therearealargenumberof interchangestationsgoing
to be built according to the plan of Shanghai metro
235
systeminthefollowingfewyears. Unlikethenormal
metrostation, theinterchangestationfeaturesincom-
plicated configuration, various construction modes,
whichgreatlyincreasetherisksinconstructionphase.
Therisk factors that affect therisk level of inter-
changestationconstructioncanbedividedintoobjec-
tive factors and subjective factors. station configu-
ration, construction mode, geological conditions
andsurroundingenvironments arefour mainobjec-
tive factors while functioning of risk management
system,contractorexperience,personnel qualifica-
tion, equipment condition arefour mainsubjective
factors.
FromtheriskassessmentonCenturyAvenue4-line
interchange station by using Fuzzy Synthetic Eval-
uation Method and the Analytic Hierarchy Process
(AHP), it is found that, therisk level of interchange
stationisgenerallyhighconcerningtheobjectiverisk
factors. However, a qualified and experienced con-
tractor witheffectively functioningrisk management
systemwould well decreasetheoverall risk level of
theproject.
The results of risk assessment can be used for
contractor selection and better decision-making to
reduce or minimize the risk during the construction
of interchangemetrostations.
ACKNOWLEDGEMENTS
The work presented in this paper was supported
by Shanghai Shentong Group Co., Ltd. The authors
gratefullyacknowledgethissupport.
REFERENCES
Bea, R. 2006. Reliabilityandhumanfactorsingeotechnical
engineering. Journal of Geotechnical and Geoenviron-
mental Engineering. May 2006: 631643.
Chapman, T.J.P. &VanStaveren, M.Th. et al. 2007. Ground
riskmitigationbybettergeotechnical designandconstruc-
tion management. Proc. ISGSR2007 First International
Symposium on Geotechnical Safety and Risk, Shanghai,
1819 Oct.
Liang, S. &Bi, J.H. 2001. Fuzzysyntheticevaluationmethod
of construction engineering quality grade. Journal of
Tianjin University 34(5): 664669.
Liu, J & Pan, Y.P 2008. Guideline of Confined Water Risk
Control in Rail Transit Construction. Shanghai: Tongji
UniversityPress.
Sowers, G.F. 1993. Humanfactorsincivil andgeotechnical
engineeringfailures. Journal of Geotechnical Engineer-
ing. 119(2): 238256.
Savidis, S.A. 2007. Risk management ingeotechnical engi-
neeringprojects by means of aninternet-basedinforma-
tion and cooperation platform. Proc. ISGSR2007 First
International Symposium on Geotechnical Safety and
Risk, Shanghai, 1819 Oct.
Wang, Q.G. 2008. Research on Deformation Prediction and
Construction Control in Deep Excavation of Expansion
Project near Existing Metro Transfer Station. Shanghai:
Tongji University.
236
Challengesinmulti-hazardriskassessment andmanagement: Geohazard
chaininBeichuanTowncausedbyGreatWenchuanEarthquake
LiminZhang
Department of Civil and Environmental Engineering, Hong Kong University of
Science and Technology, Hong Kong
ABSTRACT: TheGreatWenchuanEarthquakeof Ms8.0inRichter scaleon12May2008triggeredapproxi-
mately 15,000landslides. Thispaper startswithanoverviewof thedisaster chaincausedby theearthquake. It
thendescribesfour episodesof geohazardsoccurredduringandshortlyafter theearthquakeinBeichuanTown,
asmall county towninnorthSichuan. InEpisodeI, thefault rupturingandtheearthquakecausedthecollapse
of 80%of thebuildings inthetown. InEpisodeII, onelargelandslide(WangjiayanLandslide) andonerock
avalanche(J ingjiashanLandslide) weretriggeredbytheearthquake, whichtogether buriedasignificant part of
thetownandcausedthedeathof morethan2300people. InEpisodeIII, on10J une2008, thefloodfromthe
breachingof TangjiashanLandslideDam, whichislocatedapproximately3.5kmupstreamof Beichuan, flushed
thetown. Thedam-breachingfloodwas greater thana200-year returnperiodflood. InEpisodeIV, on23-25
September 2008, thelooselandslidedeposits near andinsidethetownturnedinto severedebris flowtorrents
amidastorm. Thedebrisflowedintothetownandburiedmuchof theoldtownthat wasnot affectedbythetwo
landslidesandtheTangjiashanfloodoccurredearlier. Finally, several challengestogeotechnical riskassessment
andmanagement areposedbasedontheobserveddisaster chain.
1 INTRODUCTION
A great earthquake of magnitude 8 in Richter scale
occurred in Sichuan on 12 may 2008. The strike-
slip rupturing started near Yingxiu and developed
towards the northeast along the Yingxiu-Beichuan-
Qingchuanfault zone. Therupturezonewas approx-
imately 300kmlong (Fig. 1, Chen 2008). The epi-
center depth was between 1020km. Latest inves-
tigations (Huang 2008; Xie et al. 2008) show that
approximately 15,000 landslides weretriggered dur-
ingtheearthquake. Theearthquakecausedacasualty
of approximately 80,000, of these 19,065 were stu-
dents. Onequarter of thetotal casualtywascausedby
earthquake-inducedlandslides.
I amdeeplysaddenedbecauseall thesehappenedin
theplaceI livedfor manyyears. Whatfurther touches
meisthedegreeof damagetoasinglesitebyachain
of geohazards. It is well known that aparticular site
may be affected by multiple hazards, such as rock-
falls, landslides, anddebrisflow.TheGreatWenchuan
Earthquake posed new challenges: multiple geohaz-
ards not only occurred at a particular site, but also
developedover timeasadisaster chain. Henceassess-
mentof therisksassociatedwithsuchadisaster chain
becomesadynamicproblem.
This paper starts with an overview of the disas-
ter chain caused by theWenchuan earthquake. Then
fourepisodesof catastrophesoccurredduringandafter
Figure1. RupturezoneintheGreatWenchuanEarthquake
andlocationof Beichuan. (After Chen2008).
theearthquakeinasingleplace, BeichuanTown, are
described. Finally, several challenges togeotechnical
risk assessment andmanagement areraisedbasedon
theobserveddisaster chain.
237
2 DISASTER CHAINCAUSEDBY GREAT
WENCHUANEARTHQUAKE
Figure 2 summarizes some observed disaster/event
chains along the time line caused by the Great
WenchuanEarthquake. Focusingongeohazardsonly.
Five typical types of geohazards occurred during
theearthquake, namely, landslides, topplings, ejected
deposits, loosened or cracked soil/rock masses, and
peeled terrains covered by loose dry debris. These
geohazards became more severe few days after the
main shocks, caused by numerous aftershocks with
eightof theselargerthanmagnitude6inRichterscale.
The landslides also blocked rivers and formed over
100 landslide dams. The water levels in these lakes
started to rise after the earthquake. By one to three
weeks after the earthquake, 34 large landslide lakes
had formed and posed enormous risks to the public
both downstreamand upstreamof the dams. While
many of these landslide dams overtopped naturally,
emergentengineeringdivisionmeasureshadtobecar-
riedout for several very largeones. TheTangjiashan
landslidedamisanexampleinwhichadivisionchan-
nel wasexcavatedat extremelydifficult conditionsto
reducetherisks to over onemillionpeople. Thecol-
lapseof largelandslidedams along Mianyuan River
andJ ianjiangRiveralsocausedfloodslargerthanrain-
induced, 50year-returnfloods, whichinundatedtowns
and cut off somemajor highways. Then in J uly and
September 2008, severestormsoccurredintheearth-
quakezoneand thewidespread dry landslidedebris
causedbytheearthquaketurnedintowet debrisflow.
Suchdebris flowcausednot only losses of lives and
property, but also fundamental transformationof the
natural environment. Suchtransformationisexpected
tolast for manyyears.
Figure2. Observeddisaster/event chainscausedbytheGreatWenchuanEarthquake.
3 MULTI-HAZARDSAT BEICHUANCAUSED
BY GREAT WENCHUANEARTHQUAKE
3.1 Episode I, fault rupture and building collapse
BeichuanTown is located in northern Sichuan (Fig.
1). By the end of 2006, the population of Beichuan
County was 160,156(BeichuanCounty Government
2008). Beichuan Town (Qushan Town), where the
countygovernmentislocated,wasaquietandbeautiful
town surrounded by green mountains and the J ian-
jiangriver (Fig. 3). DuringtheWenchuanearthquake,
whichoccurredat 2: 28pm, 12May 2008, theearth-
quakerupturewentacrosstheentiretown(Fig. 4).The
groundseismicintensityreachedthe11thdegree. The
earthquake caused devastating damage to buildings
(Figs. 46). Infact, 80%of thebuildingsinBeichuan
Towncollapsedandalmostall of theremainingbuild-
ings wereseverely damaged(Fig. 5). Approximately
15,640peopleinthecountylosttheirlives(9.8%of the
Figure 3. Beautiful Beichuan Town surrounded by green
mountainsandameanderingriverbeforetheGreatWenchuan
earthquake. (After BeichuanCountyGovernment 2008).
238
population) and 4,412 peoplewerereported missing
(CCTV 2008). The most tragic of all was the col-
lapse of two five-storey classroombuildings at the
Beichuan Middle School (Fig. 6, loss of more than
1000youngstudents(BeichuanMiddleSchool 2008))
and the burial of the entire New Beichuan Middle
Figure4. Thefault zoneacrosstheentiretownandsevere
building damage. (After Huang 2008). The extents of two
largelandslidescanalsobeenseen, whichaffectedasignifi-
cant part of thetown.
Figure 5. Beichuan Town shortly after the earthquake.
(After XinhuaNewsAgency, 16J une2008).
Figure 6. Ruins of Beichuan Middle School after the
earthquake. Over 1000youngstudentslost their lives.
School (lossof over 700youngstudentsandteachers,
tobedescribedlater).
3.2 Episode II, Landslides
After amoment of timedelay, onelargelandslideand
onerock avalancheoccurred. Thelandslide, Wangji-
ayan Landslideshown in Fig. 7, is about 10 million
m
3
in volume. The scar is steep, wide and very
high, showingthefeatures of atypical landslidethat
detached fromthe scar as a whole piece under ten-
silestresses and thedetached materials slid down at
ahighspeed. Thelandslidedebrisburiedalargepart
of a government officearea in theold town. As the
debrisadvanced,itpushedmanybuildings,whichwere
alreadyshatteredbytheearthquakebutnotburied, for-
ward for adistance, turning thesebuildings in ruins
(Fig. 8).Thescrapingpower of suchhighspeeddebris
wassuchthattheshallowfoundationof abuildingwas
seento havebeenbrought to thedebris surface(Fig.
9). Thislandslidecausedthelossof about 1600lives:
the most amount of casualty in a single geohazard
event.
Figure7. WangjiayanLandslideinBeichuanTown, which
killed1600people. (AfterYanget al. 2008).
Figure 8. Wangjiayan Landslide in Beichuan Town. The
landslidedebrisnotonlyburiednumerousbuildingsbutalso
pushedanumber of buildingsforwardfor adistance.
239
Figure 9. The high-speed debris fromWangjiayan Land-
slide brought the shallow foundation of a building to the
debris.
Figure10. J ingjiashanLandslideinBeichuanTown, which
buried the Beichuan Middle School, killing approximately
700people, mainlyyoungstudents.
Figure11. J ingjiashanLandslideinBeichuanTown. Large
rock blocks are seen in this picture, three blocks being
sufficient tofill abasketball court.
Therockavalanche,J ingjiashanLandslide,wasalso
about 10millionm
3
involume(Figs. 4and10).
Therock avalancheis characterizedby thefalling
of alargequantityof rockblockswithdiameterslarger
than 5 m(Fig. 11). The avalanche buried the New
Figure 12. Tangjiashan Landslide, Tangjiashan Landslide
Lake, andBeichuanTown. (AfterYanget al. 2008).
Figure13. TangjiashanLandslide, whichis 30millionm
3
involume. (AfterYanget al. 2008).
BeichuanMiddleSchool, leavingapproximately 700
youngstudentsandtheir teachersinthedarkness.
3.3 Episode III, flooding from breaching of
Tangjiashan landslide dam
Theendof earthquakewasnot yet theendof geohaz-
ardsexposure.Duringtheearthquake,alargelandslide
occurred at Tangjiashan, about 3.5kmupstreamof
BeichuanTown (Fig. 12). Similar to theWangjiayan
landslide, theTangjiashanlandslideisawhole-piece,
high-speedlandslidedetachedfromawideandsteep
scar (Figs. 12and13), withthematerialsfallingdown
for a vertical distance of approximately 500m. The
landslide deposit measures 611malong the sliding
directionand803mintheperpendiculardirection,and
approximately20.4millionm
3
involume(Liu2008).
Thelandslidecut off theJ ianjiang river and formed
a large landslide dam82124mhigh (Figs. 12 and
13). Inthethreeweeksfollowingtheearthquake, the
landslide lake was filled at a rate of approximately
110m
3
/s. By 9 J une 2008, the lake volume reached
247 million m
3
(water level =742.58m). When full
thelakecapacitywouldbe316millionm
3
(Liu2008).
240
Figure14. Breachingof TangjiashanLandslideDamon10
J une2008.Thepeakflowratereached6500m
3
/sat11:30am
andthecorrespondinglakewater level was735.81m. (After
GangLi, XinhuaNewsAgency, 10J une2008).
Figure 15. The flood fromthe breaching of Tangjiashan
Landslide Dampassed Beichuan Town, 12 noon, 10 J une
2008. Thepeakflowratereached9780m
3
/s; thefloodwater
level was 629.54m. (After GangLi, XinhuaNewsAgency,
10J une2008).
Thelakeposedenormous risks to 1.2millionpeople
downstream.
A well organizedemergencydivisionprogramwas
implemented under the leadership of Mr. Ning Liu,
Chief Engineer of theMinistry of Water Resources,
during 25 May11 J une 2008. Meanwhile, approxi-
mately 250,000 peopledownstreamwereevacuated.
The dam finally breached in a controlled manner
on 10 J une 2008 (Fig. 14). The peak flow rate
reached 6500m/s, which was similar to a flood of
200-yearreturnlevel (6970m
3
/s)(Huaxi Metropolitan
News2008). Whenthedam-breachingfloodreached
BeichuanTown, thepeakflowratereached9780m
3
/s,
whichwaslarger thana200-year returnflood.Theold
townwasseverelyflooded(Figs. 1517). Inparticular,
theJ ianjiangriver makes aturninsidethetown, cre-
atingahigher floodwater level onthesideof theold
town(Fig. 15).Thewatermarkscanbeclearlyseenon
thethirdfloor of abuildinginFig. 16. Thefloodinun-
datedmuchof theoldtown, flushingbuildingdebris
intothetown.Thedebrisjammedcompletelytheroad-
way, whichwasopenandplayedacritical roleduring
therescueperiod(Fig. 17).
Figure 16. Remains from the flood from the breaching
of Tangjiashan Landslide Dam in Beichuan Town. The
watermarkscanbeclearlyseenonthethirdfloorof abuilding.
Figure17. Buildingdebristhat wasbrought intoBeichuan
Townby thefloodfromthebreachingof TangjiashanLand-
slideDam.Alsoshowninthisfigurearethebuildingdamage
by theearthquakeand arock avalanchefront on theright.
Quake collapse, landslide and flooding debris in the same
scene: what elsecouldpossiblybemore?
Fortunately, all the people in the town who sur-
vived through the earthquake were evacuated about
two weeks after theearthquakefor infectivedisease
andbiological control; hencenocasualtywasresulted
fromtheflood.
3.4 Episode IV, debris flow
TheearthquakeinMay 2008causednumerous land-
slides androck avalanches. Muchof thelandslideor
avalanchedepositsspreadingonthehilly terrainsare
at amarginallystablecondition, andarehighlyerodi-
ble. On 2325 Sept. 2008, a severe stormbrought
about 190mmof rainfall, which caused widespread
debrisflowtorrents. A severedebrisflowfromWeiji-
agou, agully inthesouthwest of thetown, aswell as
that originatedfromtheWangjiayanlandslidedeposit
(Fig. 13), burstedintotheoldtown.A largepartof the
old town that was not affected by thelandslides and
theTangjiashandam-breachingflood, wasnowburied
241
Figure18. BeichuanTownafter themassivedebrisflowon
24Sept.2008.TheburiedpartwasplannedasaMemorial Site
to memorizethosewho lost their lives. (After ChinaNews
Agency, 25September 2008).
Figure19. BeichuanTownafter themassivedebrisflowon
24Sept. 2008. Thedebris is seento havecontributedfrom
WejiagouGullybehindthetownandWangjiayanLandslide.
(After ChinaNewsAgency, 25September 2008).
(Figs.18and19).Againnocasualtywasresultedinside
thetownbecauseall thepeoplewereevacuatedinlater
May.
4 CHALLENGESTOGEOHAZARDSRISK
ASSESSMENT ANDMANAGEMENT
4.1 Hazards Identification
A first lesson learned through the Great Wenchuan
Earthquake is the need to re-assess risks faced by
citiesandcommunitieslocatedinhighseismic zones
or exposed to high-risk geohazards. A critical task
in risk assessment is the identification of possible
hazards that may affect theelements under concern.
The time line for the four scenarios of geohazards
reported in the above sections is shown in Fig. 20.
IsthedebrisflowinEpisodeIV theendof hazardsfor
Beichuan Town? Obviously, unexpected geohazards
mayoccur inthefuture: thefour episodesreportedin
thispaper werelargelyunexpectedyearsago. Smaller
earthquakescanoccur intheforeseeablefuture. Some
largelandslidesmayreactivateduetorainfall infiltra-
tionor other triggers. Debris flowfromthelandslide
Figure20. Timelineof geohazardsinBeichuanTown.
deposits and loosen terrains is likely to continue in
years to come. Experiences with the Ch-Chi earth-
quake(M
L
=7.3) inTaiwanin1999(Linet al. 2008)
showthattheoccurrenceof major debrisflowtorrents
during two typhoon events after the Chi-Chi earth-
quakewasmorethandoubledcomparedwiththatprior
totheearthquake. Transformationof theriver system
tookplaceinthecourseof debrisflowandgeneral soil
erosion. It appearsthat our abilitytoidentifypossible
hazardscenariosisstill limitedandthatwehavetolive
withtheunexpected. Maypeaceupontheceased.
4.2 Risk assessment
Therisk assessment process answers threequestions
(e.g.Ayyub2008): (1)Whatcangowrong?(2)Whatis
thelikelihoodthatitgoeswrong, (3)Whatarethecon-
sequencesitwill gowrong.ThequestionWhatcango
wrong isaddressedinhazardsidentification.Therisk
associatedwithmultiplehazardscanbeexpressedas
where p
i
is the occurrence probability of an hazard
event i out of npossibleevents; v
i
isthevulnerability
of theelement at risk to theithhazard; andc
i
is the
element at riskgiventheoccurrenceof theithhazard.
TheBeichuanTownhazardscenariosshowthat:
(1) Thehazardeventsarehighlycorrelated. Theout-
comeof oneevent (e.g. landslide) isthecauseof
other events (formation of landslide dams, dam
breaching, debrisflowetc.), theleadcausebeing
thestrongearthquake.
(2) Theevents do not necessarily occur at thesame
time. Theyevolveasadisaster chain(Figs. 2and
20).
(3) Thevulnerability toeachevent may bedifferent.
Forexample, theearthquakeandlandslidescenar-
ios camewithout any warning and thus resulted
inenormouslossof life, whereasdam-breaching
floodanddebrisflowcausedlittlecasualtydueto
thesufficiently early warning and evacuation of
thepopulationat risk.
(4) Theselection of aproper benchmark recurrence
periodfor ahazardtypeisanimportant decision
forriskanalysis.Theoccurrenceprobabilityof the
root cause(strongearthquake) isextremelysmall
(approximately2000-yearreturnperiodinthecase
of theGreat Wenchuanearthquake). If asystem-
atic risk analysiswereconductedconsideringthe
242
possiblerun-downdistancesof thetwolargeland-
slides (Figs. 4, 7 and 10) and the debris flow
(Figs. 18 and 19), the safe distance to the fault
rupture(Fig. 4), andthezonesusceptibletoflood-
ing (Figs. 1517), then almost the entire town
would not beinhabitable. Notethat theChinese
Codefor Geotechnical EngineeringInvestigation
GB50021-94(Ministryof Construction1995)rec-
ommendsthefollowingsafedistancestostrongly
active faults with the potential to cause earth-
quakesgreater thanmagnitude7inRichter scale:
3000mfor designs to 9th degree intensity, and
10002000mfor designsto8thdegreeintensity.
The Code also recommends that important con-
structions should not beon theupper platenear
therupture. Similar situations in many towns in
thequakezonemakedecisionsonreconstruction
planningverydifficult.
4.3 Risk management
Theriskmanagementanswersthequestionsof (1)what
canbedonetoreduceor control riskand(2) what are
theimpactsof anyproposedsolutionsontheriskpro-
fileof thesystem(e.g.Ayyub2008).Thefirstquestion
canbeansweredthroughthereductionof oneor more
of thethreerisk components: occurrenceprobability
of hazard, vulnerability, and element at risk. Several
issuesor challengesindealingwiththetwoquestions
areasfollows:
(1) Reducing the occurrence probability of hazards
requirestheidentificationandstrengtheningof a
largenumber of natural or manmadeslopes to a
tolerablelevel. Whilesomelandslidehazardswill
certainlybemitigatedandsomeslopeswill besta-
bilized, it isnot likely agreat number of features
canbeidentifiedandstrengthened.
(2) Again theselection of aproper benchmark haz-
ard recurrence period for engineering design is
adifficulty decision. A largeareasuffered from
damageof 8th-11thdegreeduringtheWenchuan
earthquake(2000-year event), whiletheoriginal
design seismic intensity was mostly 7th degree
basedonahazardlevel thathasa10%probability
of exceedanceina50-year exposureperiod(475-
year return period). Although a design intensity
of 8th degree has been adopted for reconstruc-
tionworks, thereisstill ashortfall.A costanalysis
oftendoes not allowtheuseof thesevereearth-
quakeeventactuallyhappenedasabenchmarkfor
design. There must be a trade-off between risk
and costs. Theconstruction of theNew Orleans
HurricaneProtection Systemoffers an example.
While Hurricane Katrina inAugust 2005 was a
400-year event, the new systemis designed for
100-year level eventsfromacost-benefit point of
view(USACE 2006).
(3) Nowthatthehazardoccurrenceprobabilitycannot
be reduced significantly due to economic con-
cerns, amoreeffect way to risk mitigationis the
loweringof vulnerability. Sinceearthquakedisas-
ters may not be forecasted in a near future and
thepeopleareto livewith disasters in high-risk
zones, measuresfor vulnerabilityreductionat the
communitylevel (safeislandsetc.) arecalledfor.
Whileforecastingof earthquakesisaformidable
task, itispossibletomonitorandpredictthedevel-
opmentof after-quakedisaster chains(Figs. 2and
20).ThemitigationofTangjiashanLandslideDam
risks(Figs. 1217) isasuccessful example.
(4) Whenriskscannot bereducedtoatolerablelevel
or cost-benefit cannot bejustified, thereisaneed
toreducetheelementsatrisk, i.e., torelocateper-
manentlytheresidentsinareassusceptibletonear-
future disaster chains. This is often proved not
quite viable considering various negative social
impacts(e.g., separationof familymembers, loss
of familiar communities, jobmarket etc.).
(5) There is a need for effective risk education and
communicationonseveral issuesduringtherecon-
struction period: (a) Cost-benefit evaluation of
newdesignintensity whenit is smaller thanthat
actually happened; (b) Keeping earthquake vul-
nerabilityinmindincityplanning; (c) Relocation
or reconstruction at it is; (d) Use of potentially
moredangerousnewsites.
5 SUMMARY
Thegeohazard chain in BeichuanTown, particularly
four episodesof catastrophesduringandshortlyafter
theGreat Wenchuan Earthquakeis described in this
paper. The hazard events did not occur at the same
time; insteadtheseeventsevolvedasadisaster chain
with possibly unknown future events. The hazard
eventsarehighlycorrelated.Theoutcomeof oneevent
(e.g. landslide) is the cause of other events (forma-
tion of landslide dams, dambreaching, debris flow
etc.), the lead cause being the strong earthquake.
The earthquake actually happened was an extreme
event (approximately 2,000-year event) that caused
theloss of 80,000 people. Thepublic has witnessed
the cruel fact but has yet to accept smaller hazard
events for future design and construction. All these
raisechallenges to risk assessment and management
inamulti-hazardenvironment.
ACKNOWLEDGMENTS
Theauthor wouldliketo acknowledgetheassistance
fromProf.Chang-RongHeof SichuanUniversity,who
arrangedatripfor theauthor tovisitBeichuanshortly
after the Great Wenchuan Earthquake. Profs. Run-
QiuHuangof ChengduUniversityof Technologyand
Wei-LinXuof SichuanUniversity providedvaluable
information. Mr. Yao Xu, Miss Melissa Zhang, and
Ms J inhui Li proof read the manuscript. The finan-
cial support from the Research Grants Council of
HongKong(Project No. 622207) andtheHongKong
243
University of Science and Technology (Project No.
RPC06/07.EG19) isalsogratefullyacknowledged.
REFERENCES
Ayyub, B.M. (2008). A risk-based framework for multi-
hazardmanagementof infrastructure. Proc. International
Workshop on Frontier Technologies for Infrastructures
Engineering, 2315Oct. 2008, National TaiwanUniver-
sityof ScienceandTechnology, Taipei, S.S. ChenandA.
H-S. Ang(eds.), 209224.
Beichuan County Government. (2008). Beichuan county
population statistics. http://beichuan.my.gov.cn/bcx/
1658169087802474496/20080617/305295.html.
Beichuan Middle School (2008). Earthquake relief for
BeichuanMiddleSchool. Online: http://bczx.changhong.
com/.
CCTV (2008). Beichuancasualtystatistics. TheChinaCen-
tral Television. 23J une2008.
Chen, Y.T. (2008). Mechanisms of Wenchuan earthquake.
Keynotelecture, ForumonEarthquakeRelief andScience
andTechnology, ChineseAcademyof Sciences, Chengdu,
25J uly2008.
ChinaNewsAgency. (2008). 25September 2008.
Huang, R.Q. 2008. Preliminaryanalysisof thedevelopments,
distributions,andmechanismsof thegeohazardstriggered
bytheGreatWenchuanEarthquake. StateKeyLaboratory
of Geohazards Prevention and Geological Environment
Protection, ChengduUniversityof Technology, Chengdu,
China.
Huaxi Metropolitan News. (2008). Tangjiashan division
channel canpassfloodsof 200year-returnperiod. 25J une
2008.
Lin, M.L., Wang, K.L., andKao, T.V. (2008). Theeffectsof
earthquakeonlandslides A casestudyof Chi-Chi earth-
quake, 1999. LandslidesandEngineeredslopes, Chenet
al. (eds.0, Taylor & FrancisGroup, London, 193201.
Liu. N. (2008). Landslide dams in Wenchuan earthquake
andrisk mitigationmeasures. Keynotelecture, Forumon
EarthquakeRelief vs. ScienceandTechnology, Chinese
Academyof Sciences, 25J uly2008, Chengdu, China.
Ministry of Construction. 1995. Commentary Code for
Investigationof Geotechnical Engineering, GB50021-94.
Ministryof Construction, Beijing.
United States Army Corps of Engineers (USACE). 2006.
PerformanceEvaluationof theNewOrleans andSouth-
east LouisianaHurricaneProtectionSystem. Draft Final
Report of theInteragency PerformanceEvaluationTask
Force, VolumeI ExecutiveSummary andOverview. 1
J une2006.
Xie, H.P., Deng, J.H., Tai, J.J., He, C.R., Wei, J.B., Chen, J.P.,
andLi, X.Y. (2008).Wenchuanlargeearthquakeandpost-
earthquake reconstruction-related geological problems.
Chinese Journal of Rock Mechanics and Engineering,Vol.
27, No. 9, pp. 17811791.
XinhuaNewsAgency. (2008). 10J une2008.
XinhuaNewsAgency. (2008). 16J une2008.
Yang, X.G., Li, S.J., Liu, X.N., and Cao, S.Y. (2008). Key
techniquesfor emergent treatment of earthquakeinduced
landslidedams. StateKeyLaboratoryof MountainRivers
andEnvironment, SichuanUniversity, Chengdu, China.
244
General sessions
Design method (1)
A studyof thenewdesignmethodof irrigationpondsusingsheet materials
M. Mukaitani, R.Yamamoto&Y. Okazaki
Takamatsu National College of Technology, Takamatsu, Japan
K. Tanaka
Naruto surveying & designing co., ltd., Naruto, Japan
ABSTRACT: There are two hundred ten thousand of ponds in J apan. Most of all ponds became too old,
damaged by natural disastars and spread of housebuilding. When werepair theponds dike, wecan not get
soil materials fromnearby mountains. Nobody is permittedto get frommountains soil withlicense, because
of environmental problems. Manytypesof industrial sheetsaredevelopedafter 1970s. Wemust makethenew
designmethodof ponddikeusingsheet materials.
Most ponddikesslopefailuresarecausedby theshear strengthonbetweenmainsliplayerssoil andsheet
material. Then, theshearingmechanisms of dikes soils onsheet material areimportant. This paper treats the
considerationsof dikesslopebyviewpoint of theinfiniteslopemethod, applicationfor thein-situcasestudy, a
comparativestudyof ordinarydesignmethodsandprogressof seepageline. Firstly, weconsideredtheinfinite
slopemethodon simpletypeof slopestability andclarifiedthesafety factors relation between theproposal
methodandtraditional slopestability. Weformularizedtheseepagelineof progressof soil onthesheetmaterial.
Weclarifiedthat theseepagelineisdescribedaparabola.
1 INTRODUCTION
Thispaper treatsthenewdesignmethodof irrigation
ponds usingsheet materials. Thereare210thousand
of pondsinJ apan. If thepondisrepairedagainst too
old or the earthquake-resistant, we must seek dikes
materials. Thedikeof theirrigationpondisconsisted
of thecoveredsoil, thefilter soil andthecoresoil.
Itseemsthatwecantgetdikesmaterialssoeasily.
Top5prefecturesmakeupover 50%of theirrigation
pondsinJ apan.TheseareHyogo, Hiroshima, Kagawa,
Yamaguchi andOsakaprefecture.
The newly construction method of pond is made
by thecenter coretype. Theoldpondisimprovedby
thefrontcoretype. Thefrontcoretypepondneedslot
of dikes materials. Fromenvironmental view point,
it is difficult to get newly dikes materials fromthe
near-by mountain area. In solitary island and island
area, it is more difficult to get them. Because of
abovementionedreasons, wemustinvestigatethedike
improvementmethodwithlow-costandusinglowvol-
umeof newly materials. Thedikeimprovement work
isnotusedlargescaleconstructionmachinesincoun-
try area, wheretheirrigationponddoes not havethe
bigvolume. Wehavebeenusedthewaterproof sheet
material which madein factories after 1960s. There
aremany kinds of waterproof sheet materials. These
are imported fromother countries and lot of types
of structural design by each company. It is needed
that the proper design method and the check sys-
temareclarifiedagainstthepondsimprovementwork
(Tanaka(2007),Mukaitani (2008a,bandc)).Whenwe
designedtheimprovementworkof theoldpond,design
problemsareexperimentallysolvedineachengineer.
Inthispaperwewouldliketogiveouropinionabout
the newly design method of the old pond using the
waterproof sheet material whichhas lately becomea
subject of special interest.
2 SUMMARY OF THEORETCAL ANALYSIS
Fundamental formulaof theoretical analysis is based
ontheinfiniteslopestability. Inthissection, wemake
theassumptionthat thesoil conditionof thedikeon
thewaterproof sheet material isaswell astheinfinite
slope. R. M. Koener and D. E. Daniel (1997) pro-
posedthedikesanalysisusingthefiniteslopemethod
includingthewaterproof sheet material. Manufactur-
ing companies show lots of technical notes and the
soil slidingmodel withconsiderationof thematerials
tensilestrength.
Weproposedthenewlyinfiniteslopestabilitywhich
can be considered the parallel submergence ratio
calledPSR, theback sideseepagepressure, theseep-
agepressureandthecohesionof thesoil.ThePSRand
horizontal submergence ratio called HSR are deter-
minedbyKoener et al. For example, whenthePSR is
zero, thegroundwater level is equal to thebottomof
theslopesoil columnonthewaterproof sheetmaterial.
If thevalueof PSR is one, it means that theground-
water level isequal tothetopof theslopesoil column.
247
Figure1. Schematicdiagramof infiniteslopestability.
Whenwater isfilledinthepond, thevalueof PSR is
over one. In slopestability, most dangerous ground-
water conditionshowsthatthevalueof PSRisgreater
thanzeroandlessthanone.
Wedetermined thegeneral equation of thesafety
factor of theinfiniteslopeaboveconsideredproblems
asfollows;
wherePSR=h
w
/h, Zisthevertical height of thecov-
eredsoil layer,isthecoefficientof thebacksideseep-
agepressurewhichdeterminedbyc
/(
w
h tan
) ,
isslopeangleof waterproof sheetmaterial, andc
and
arestrengthparametersof thecoveredsoil.
The indicatesthevaluefromzerotoone. Wehad
better adopt the less than 0.5. The seepage water
pressuref
h
isdeterminedasfollows;
3 A TYPICAL CASE STUDY OF FAILURED
DIKE
Inthiscase, c
and areequal tozerobecauseof the

conventional analysis. Figure2 shows theschematic
diagramof the failure dike. The failure of the pond
dikewas occurred by thecontinuous rainy days and
thehighlevel waterof nearriver, whichmadetheseep-
agepressureintheslopessoil layer onthewaterproof
sheet material.
When the value of PSR is zero, the safety factor
fromequation(1) is given1.019. Whenthevalueof
PSR is0.5, thesafetyfactor isgiven0.714. Whenthe
valueof PSR isone, thesafetyfactor isgiven0.483.
If the pond is filled with water, the safety factor
is recovered to 1.019. Thepond which is filled with
water ismost stabilized.
We investigated that the influence on the safety
factor of the slope length between the conventional
infiniteanalysisandKoenersmethod.
Figure2. Schematicdiagramof atypical casestudy.
Figure3. Relationbetweensafetyfactorsandslopelength.
Figure3showsrelationbetweensafetyfactorsand
slopelength. InthispaperthesafetyfactorbyKoeners
analysis is called F. The safety factor of Fs by our
infinitemethodshowconstant valueof 1.019against
the change of the slope length. The safety factor
by Koeners method decreases with theslopelength
growing. The relation between the safety factor by
our infinitemethodandKoeners F is determinedas
follows;
Wecomparethemarginof error byeachmethodon
10mof theslopelength.ThesafetyfactorbyKoeners
methodis6.6%greaterthanourone. Ourconventional
infinitemethodgivesthelowvalueof thesafetyfactor.
Wecangetalittlesafetydesigningagainstthechanges
of environment immediately.
Figure4showstherelationbetweenthesafetyfac-
tor by our proposed infinitemethod and PSR under
variationof the fromzeroto0.3. Thisfigureshows
thatthesafetyfactor isaffectedbytheexistenceof the
back sideseepagepressureor thegroundwater level
inthesoil layer onthewaterproof sheet material. We
thinkthatmanyslopefailureusingthesheetmaterials
wereaffectedby theexistenceof or seepagewater
pressure.
When the bottom of pond is partially saturated,
thepossibility of slopefailurerises up. Thedrainage
systemis neededagainst thegroundwater nearby the
bottomof pond.
248
Figure4. RelationbetweenFsandPSRundervariationof .
The covered soil on the waterproof sheet mate-
rial has lowvalueof cohesion. Thecovered soil has
ten blows of the standard penetration test value by
compaction work at the dike slope. The 5kN/m
2
of
cohesion is needed by our infinite analysis against
thenormal slopeof thepond dike. Thecohesion on
theslopeis raisedusingthecement mixingmethod.
Weproposedthatthecohesionunder wetconditionof
unsaturatedsoil isequal tothethicknessof theslope.
Thisinvestigationwill treat another paper.
4 COMPARISONWITHTECHNICAL NOTES
BY MATERIALSCOMPANIES
Therearetwogroupsof companiesof thewaterproof
sheet materials. One is made by the rubber sheet.
Another ismadebythebentonitesheet.
4.1 Finite slope satiability using the mixed analysis
This method is adopted by companies of the rub-
ber sheet. After thecountermeasureof bottomsoils
improvement is treated, they calculatetheresistance
based on thepassiveearth pressure. This viewpoint
issimilar tothecalculationof thecantilever retaining
wall. Thetensilestrength of sheet is not considered.
Thefatal defect of this concept is not consideredthe
changingof water level intheslopesoil layer.
4.2 Infinite slope satiability considering with
reinforced sheet and soil layer
Thismethodisadoptedbycompaniesof thebentonite
sheet. Thesafety factor is calculatedwiththetensile
strengthof sheetontheslope. If thesheethastheopen
arealikethegeogrid, thesoil penetrates to thesheet
material. Thefatal defect of this concept is not con-
sidered thechanging of water level in theslopesoil
layer, too. InthiscaseKoenersmethodgives1.26of
thesafetyfactor.
Figure5. Schematic diagramof seepagelineontheslope
soil layer byrainfall.
5 CALCULATIONOF SEEPAGE LINE INTHE
PONDSDIKE UNDER RAINFALLL
Wefocusedtwopointsbeforecalculationsof seepage
lineinthepondsdikeundertherainfall. Oneisneeded
conventional assumptionsandthatformulaisabeleto
beadaptedto thegeneral slopeproblems. Another is
neededthat it is easily calculatedsafety factor inthe
constructionwork at ponds dike. Figure5shows the
schematicdiagramof theseepagelineontheslopesoil
layer under therainfall.
We focused the column on the slope. In the fol-
lowing, it shall beassumedagainst theanalysisof the
seepageline;
1) Thesedimentaryconditionisassumedthehorizon-
tal deposit.
2) The coefficient of horizontal permeability; k
h
is
greater thanthecoefficient of vertical permeabil-
ity; k
v
.Thecoefficientof storageinthecoveredsoil
layer is determined by (11/a). Therefore, if the
valueof inflowisone, thevalueof outflowis1/a.
3) The anisotropy of permeability is assumed that
k
h
/k
v
is normally 25 by theJ apaneserecommen-
dationfor designof irrigationtanks(2006).
4) Thevalueof theseepagevelocity by rainfall; v is
assumedfrom2to3mm/hr. Theseepagevelocity
by rainfall and the rainfall intensity have a sim-
pleproportional relation. Thevolumeof inflowis
assumedabout 26%for decomposedgranitesoil.
Wewill explainthisintheFigure5, thecoordinate
originisdecidedatthetopof slope.Theseepagelinein
theslopeonthewaterproof sheetmaterial iscalculated
asfollows;
249
whereDisoutput water level, Z
0
isinput water level
andX
max
isthepeakvalueof water level intheslope.
TheX
max
iscalculatedbyfollowingformula;
Theequationshowsthat theX
max
isaffectedby D,
andL.
Wemustabovementiontheseepagelinetoourpro-
posedPSR. PSR iscalculatedbyfollowingformula;
whereD
max
is thethickness of covered soil layer on
thewaterproof sheetmaterial. Theareaunder seepage
lineiscalculatedbytheintegrationof equation(4).
6 CONCLUSIONS
Many small scaleirrigation ponds must berepaired.
Itisnecessarytocalculateeasilyatin-situandeconom-
ically. This paper proposedtheconventional analysis
usingtheinfiniteslopestability. Thepondsdikewill
repair using thewaterproof sheet materials. Thesoil
layersstabilityonthewaterproof sheet material must
bereconstructedmorecarefully. InJ apan, heavyrain-
fall will behappenedinoneyear.Theaccidentindikes
constructionmust bepreventingasmuchaspossible.
Theseepagepressuremakes theslopefailurewhich
is not so filledinsoil layer. Theeffects of thebench
cut under the sheets and the taper of covered soils
layeronthesheetaremustbeinvestigated.Theseitems
areoftenconsideredinslopeworks. Wecontinuethe
applications against in-situ problems and the failure
slopedata.
REFERENCES
Koerner, R.M & Daniel, D.E. 1997. Final Covers for Solid
Waste Landfills and Abandoned Dumps, ThomasTelford
Ltd.
Tanaka, K. & Mukaitani, M. 2007. Thenewdesignmethod
of irrigationpondsusingsheetmaterials(part1), Proc. of
domestic annual conference for geotechnical engineering,
J GSof Shikokubrunch, 8788. (inJ apanese)
Mukaitani, M. &Tanaka, K. 2008a. Thenewdesignmethod
domestic conference on disaster prevention engineering,
J SCE of Shikokubrunch, 5560. (inJ apanese)
Mukaitani, M. &Tanaka, K. 2008b. Thenewdesignmethod
domestic annual conference for civil engineering, J SCE
of Shikokubrunch, 212213. (inJ apanese)
Mukaitani, M. &Tanaka, K. 2008c. Thenewdesignmethod
domestic symposium for geotechnical problems and envi-
ronment, J GSof Shikokubrunch& EhimeUniv., 6368.
(inJ apanese)
Recommendation for design of irrigation tanks, 2006 The
J apanese society of irrigation, drainage and rural eng.
(inJ apanese)
250
Researchonkeytechniqueof double-archtunnel passingthrough
water-erodedgroove
Y. Chen
Key Laboratory of Geotechnical and Underground Engineering of Ministry of Education,
Tongji University, China
X. Liu
School of Civil and Architectural Engineering, Central South University, China
ABSTRACT: Inexcavationof double-archtunnel, thereexistsabigwater-erodedgroove.Themethodof elastic
foundationbeamisusedtocopewiththesituation.Accordingtotherelationshipof foundationbeamwithtunnel
structureandscenecase, themechanical behaviour of double-archtunnel passingthroughwater-rodedgroove
is studied by using finiteelement numerical simulation method. Theresearch result shows that theinterface
of two different foundationmediums is worst point of foundationbeam, cuttingdamageandrippingdamage
easily happenhere; thephysics property of fillingmatter inwater-erodedgroovehas distinct influenceto the
foundationbeam, toconsider fullythecarryingcapacityof fillingmatter isreasonable; settinginvert isgoodto
improveinternal forcestateof tunnel structure, but theactionisnobigtosharetheloadof foundationbeam.
1 GENERAL INSTRUCTIONS
Sincethereisabigwater-erodedgrooveinexcavation
of tunnel, the choice of traversing method becomes
critical. In practical tunneling engineering, the exis-
tence, location and size of water-eroded groove is
difficult to be judged because of the limitation of
geologic investigation, so relevant traversingmethod
according to the actual condition is only when the
problemhappens instead of adesign special for big
water-eroded groove. This way is available to sin-
gletunnel becauseof its simplestructure. However,
to double-arch tunnel, a special design is necessary,
considering the difficulty in the excavation caused
bythecomplicatedstructureandmoreimportant, the
influencetothestabilitybytheconstruction.
Therefore, taking the structure characteristics of
double-arch tunnel in water-eroded groove into
account, thewriter get theavailablesolution by cal-
culation and analysis for setting foundation beam
combinedwiththedesignandconstructionof practical
engineering.
2 THE RELATIONBETWEENDESIGNAND
CONSTRUCTIONOF DOUBLE-ARCH
TUNNEL
Intheconstruction, amiddlepilot headingisfinished
firstly inorder tosupport thearchringat bothsides.
Themiddlewall isat adisadvantagestressstatesince
itstwosidesisnot tight tothesurroundingrocksthus
causingindependentdeformation. Commonly, itisnot
Figure1. Constructionstepsof double-archtunnel.
suitabletoconstruct thebothtunnelsat thesametime
inorder toavoidlong-spanat thesamecross-section
that influences the stability of surrounding rock. So
theidea of construction schemeis that excavatethe
one-sidesinglespanfirstly.Asaconsequence, thehor-
izontal forceof thefirst-finishedarchringmakesthe
middlewall bearmarkedbias,whichisdisadvantageto
themiddlewall withoutlateral rockresistance.Appar-
ently, themiddlewall isthekeyinthiskindof tunnel
construction. Constructionstepsareshowedinfig1.
Fromthe view of design, there are two kinds of
structural stress state in the construction of double-
archtunnel, namelysinglespanstateanddoublespan
state.Themiddlewall isunderadisadvantagebiascon-
ditioninthefirst-finishedsinglespanliningandthis
251
bias disappears since the final-finished double span
lining is a balanced construction, so apparently the
singlespanstateisthemost disadvantagedcondition.
Inthedesignandconstructionshouldreflectthischar-
acteristic toensurethesafety. Sotheemphasisinthe
designisthestateof singlespan.
3 DISCUSSIONOF ENGINEERINGCASE
TongyouMountaintunnel, major project inGuangXi
South2belt highways, is445minlength. Theareais
inkarstsdevelopingdistrictwherelimestoneiswidely
distributedanddissolutiongrooveandcrack is com-
mon. Dissolutiongeodeiscommonlyseenindolomite
which formanomalistic cast groove along the joint
plane. Being limited by landformand landmark, the
tunnel is designed as double-arch structure, meeting
thedemandof connectionof roadat thetunnel open-
ing, whichisexcavatedataclearwidthof 28.56mand
aclear height of 8.92m.
Inthepartof K2+312+332of thetunnel, there
is a big water-eroded groove, 20m in longitudinal
width, displaying an oblique tunnel axis along the
directionnortheast-southwestanditispassedthrough
bythecrosssectionof thedouble-archtunnel.Thepart
of water-erodedgroove, belongingto shallow-buried
segment, isabout20minburieddepth. Itscoverisclay
addingdetritusinwhich2040%belongstodolomitic
limestonewhosecorrosionsurfaceis clear. Theclay,
havingdelicateandplastic structure, obviously isthe
fillingof thegrooveandthiskindof fillingiscompact-
inganditsbearingcapacityisabout 150180Kpa. Its
natural unit weight is =1617kN/m
3
andthereis
no influenceongroundwater. To provetheinfluence
rangeof thebottomof thegroove, wedrilledto17m
under thetunnelsbottomandstoppedsinceit isalso
bottomless.
3.1 Water-eroded groove span scheme and elastic
foundation design
The main method of water-eroded groove crossing
is filling or span etc
[13,14]
. The range of this tun-
nels groove is deep, so the technical requirement
to the filler of grouting reinforcement is high and
objectiveeffect cannot beguaranteedeasily. Through
careful research, thewayof erectingbeamcrossingis
decided to adopt. This method can ensurethat there
is no longitudinal or transversecrack alongthelines
causedbytheweakeningof thebearingstrengthof the
basement in thestructureof double-arch tunnel thus
creatingconditiontoguaranteethehighqualityof the
tunnel.
To make the tunnel pass safely, there is a strip
reinforcedconcretebeam(foundationbeamfor short)
acrossthegrooveinthebaseof rightwall, leftwall and
middlewall. Thebeamis30m-longand1m-high. To
detract the stress on the base of the wall, the width
of thefoundationbeamintheside-wall isdesignedto
1.3mandthat of middlewall is2.7m, all gobeyond
thewidthof thewall bottomby 15cmto 35cm. 5m
isremainedonthebothendof thebeamtoput onthe
bedrock(IV gradewall rock) andthemiddlepieceof
thebeamisputonthegroovesfiller. Sothisisaelastic
foundationbeamsettingondifferent foundation.
3.2 Design of tunnels support structure and load
of elastic foundation beam
The tunnels support structure is compound lining.
Initial support is acombined support including wire
mesh, anchor and shotcrete. Doublelevel of 6 mat
reinforcement isset inthewholecrosssectionof the
archwall.Thespacebetweenthematreinforcementis
25cm* 25cm. Theanchor armis3m-long+22rein-
forcement anditslongitudinal spaceis1mand1.2m
inringspace. Theshot CreteisC20concreteandit is
25cmthick
[15]
.
Inorder tomakethesurroundingrockstable(espe-
cially the clay filling of the water-eroded groove),
advanced small pipe grouting support is set, 4.5m-
long42steel pipeisarrangedalongthetunnel arch
ring, one ring every longitudinal 2m and the ring
distanceis 35cm. MeanwhileI 18-typesteel arch
timberingandthelongitudinal distanceis0.5m.
Secondary liningdesignisdeterminedby analysis
of load-structuremodel calculation. Consideringthat
primary support is under acertainloadof surround-
ingrock andalso small duct groutingis takeninthe
assistingconstructionmeasuresthathelptoreducethe
stressof surroundingrock,so80%of theoverlyingsoil
weight is used as thevertical load of secondary lin-
ingandthelateral pressureis comparatively treated.
But thereis no lateral pressureinthemiddlewall in
thealtituderang of middlepilot heading. According
to theactual situation, themost unfavorableloading
stateistakennamelysingle-spanstructurecalculation
asshowninfig2.
After thecalculationof liningstructure, theresult
of foundationpressure(per runningmeter) is2460kN
inmiddlewall and2200kN insidewall. Theloadon
thefoundationbeamis2460/(2.71.0)=911kN/m
2
inthemiddlewall whosefoundationbeamswidthis
2.7mand2200/(1.01.0)=2200kN/m
2
insidewall
whosefoundationbeamswidthis1.0m.
Figure2. Liningcountingsketch.
252
3.3 Counting and analysis of elastic foundation
beam
Since the water-eroded grooves filling and bedrock
ontheendof thefoundationbeamistwocompletely
differentmedium, itisaproblemaboutsolvingelastic
foundationbeambasingondifferentfoundation.Finite
element displacement method programming calcu-
lation is adopted and the assumption of simulation
foundations sedimentation deformation by Winkler.
fig3is thecalculationgraphic formulas andTable1
isthecalculationparameter.
Put the max internal force and deflection of the
foundationbeamintotheTable2.Takethemiddlewall
asanexampleandtheinternal forceanddeflectionis
listedinthePicture4. Fromtheresult of calculation,
it canbeseenthat themax shearingforcehappenat
thefifth point, thepoint of intersection between the
bedrockandthefillingof thegroove, accordingtothe
Fig. 3anditshowsthatthishasthegreatestpossibility
of shearfailure.Absolutemaximummomenthappenat
theforthpoint andalsoat thejunction. Whatsmore,
it is negativemoment which may makethesuperior
margin of the beamhave tension failure. All these
shouldbepaidtogreat attentioninthereinforcement
design.Themaxdeflectionhappensatthemidspanof
thebeam. However, thiscannotformdangerouscross-
sectionsincethemoment andshearingforceareboth
verylittle.
Itshouldbeillustratedthatthelengthof foundation
beamonthebaserockof thetwoendshasgreatimpact
ontheinternal forcestateof thebeam, longindwelling
leadingtowasteandshorttounsafeandbothoverlong
andover short canmaketheinternal forceunreason-
abledistribution. Sincethegrooveshould betreated
assoonaspossibletoavoidcollapseaccident, thereis
notimeforfurtheranalysis. Optimizationtreatmentof
proportionbetweenindwellinglengthof thebeamon
thebaserock andaspect of crosssectionissuggested
insimilar project. Fromthetable2, it canbeknown
thatthemaxpositiveandnegativemomentisseparated
by19%thatmeansfurther adjustmentof thelengthof
thewholebeamorproportionof theaspectof thecross
sectionisallowed. However, that inthemiddlewall is
only1.4%showingthereasonabledesignparameter.
Figure3. Countingpatternof elasticfoundationbeam.
Table1. Countingparametersof foundationbeam.
Elasticfoundationcoefficient
Thesection (MPa/m)
Location of beam(m) Elastic Unit weight Vertical
of beam (longhigh) Rockof gradeIV Caveclaymedium modulus(MPa) (kN/m
3
) load(kN/m
2
)
Sidewall 1.01.0 350 100 2.8510
4
25 2200
Middlewall 2.71.0 350 100 2.8510
4
25 911
In order to discuss theinfluenceof fillings to the
internal forcestateof thefoundationbeam, comparing
calculationistakenamongdifferent fillings. Takethe
middlewall asanexample, after regressiontreatment,
andtheresult isshowninfig5.
Shown in the fig 5, there is a higher-correlation
power functionrelationshipbetweenthemaxpositive
andnegativemomentandelasticcoefficientof thefill-
ings, namely value K. When the value K becomes
small, the growth speed of the max moment gets
higher. EspeciallywhenthevalueK becomessmall to
somedegree(30MPa/m), themax moment increases
dramaticallyandthenegativemoment becomesmore
obviousthanthepositivemoment. Nomatter howthe
elasticcoefficientchanges, theactionpointof themax
negativemomentisalwaysonthefourthpoint, namely
theedgeof thegroove.
Thereforethejunctionof twodifferent foundations
mediumis theeasiest to damageplaceandthesupe-
rior margin of the beam is easier to damage. And
with the decrease of the value K, the max positive
moment gradually moves to themid span and even-
tually happens at the mid span. This shows that the
groovefillings haveimportant impact on theadjust-
ment of thedistributionof themoment of thewhole
beam.
3.4 Discussion about invert
Invert and lining is divided into first doneand later
done. If do theinvert later, thefoundationbeamwill
bear all theoverlyingpressureandthestructurecalcu-
lationiswithouttheinvert. If dotheinvertfirst, partof
thepressureisseparatedmakethestressonthebeam
decreased and now, theinvert is involved in thecal-
culation. Toensuretheactual influenceof thiskindof
Table 2. The max internal force and flexibility of the
foundationbeam.
M
max
/(kN.m)
Point 4 1662 Point 8 1410
Point 4 1663 Point 8 1687
Notice: points in the table see figure 5; M
max
Abso-
luteMaximumMoment, Q
max
AbsoluteMaximumShear,
Wmax-MaximumDeflection;
A positive/negative bending moment causes compres-
sion/extensionof thetopfibres of abeam; apositiveshear
will tendtorotateeachportionof thebeamclockwisewith
respect toitsother end, otherwiseisnegative.
253
Figure4. Internal forceandflexibilityof thebeamunder meddlewall.
Figure5. Influenceof different fillingstomax moment of
foundationbeam.
Table3. Theinfluenceof invert tofoundationbeamload.
Location Setting Pressure Actingload Load
of beam invert (kN) (kN/m
2
) decrement(%)
Middle Yes 2390 885 2.8
wall No 2460 911
Sidewall Yes 1840 1840 16
No 2200 2200
groovefillingtothebeam, comparativecalculationis
taken in this case. Theresult turns out that themain
effectof theinvertistoimprovetheinternforcestateof
theliningstructure.Takethevaultnodeasanexample,
themoment isdecreasedbyabout 70%wheninvert is
set but it do littleto reducetheload function to the
beamespeciallytothemiddlewall beam, only2.8%is
reduced. It isslightly inconsistency totheinitial esti-
matethat theinvert bears themost foundation beam
load. After analysis, it turns out that lining structure
mainly takenby thefoundationbeamsincetheinvert
islocatedontheflabbygroovefillingandthelinings
wall corner is located on the solid foundation beam
that causing a tremendous difference in their nature
leading to subsidenceof theinvert. Therefore, when
thiskindof foundationbeamisdesigned, thefounda-
tionof theinvert cannot bemagnified. Theinfluence
of invert tothefoundationbeamislistedintable3.
4 CONCLUSION
(1) When thegrooveis filled, it should beregarded
aselastic foundationrather thansimpleoverhead
beam. Byanalysis, weknowthatevencompletely
not consideringthefillingsbearingcapacity, the
effect to the intern force of the beam is obvi-
ous. Comparingthefoundationelasticcoefficient
k=5MPa/mand k=0MPa/m, it turns out that
themax moment of thebeamis29745kN.mand
70230kN.mrespectively, havingadifferenceof 2.
Therefore, when meeting the groove, we should
earnestly find out the fillings compactness and
bearingcapacityandalsothegroundwater condi-
tion. If possible, weshould do our best to think
about its bearing capacity to reduce the cost of
theengineering. Whatsmore, appropriategrout-
ingandexchangingfillingcanbecombinedwith
to enhance the bearing capacity artificially and
thenerectingbeamandspan. It ismoreeconomi-
cal thanthemethodtotallywithoutthecalculation
of foundationbeamcapacity.
(2) Generally speaking, sincetheuncertainty of the
surround rock around the tunnel, the load to
the foundation beamhas its fuzziness. However
to shallow-buried tunnel, this kind of fuzziness
becomesmuchsmaller.Therefore, thereiscertain
reasonabilitytoreachtheeffectonthefoundation
beambythewaymentionedinthiscontext.
(3) The junction where two different kinds of base
mediummeet is themost dangerous point of the
foundation beam, shear and tension crack easily
occurringthere.
(4) The function of the invert is mainly to improve
theinternforcestateof theliningstructure. How-
ever, turntothesharingthestressof thebeam, its
functionischangingwiththebearingcapacityof
thefilling. Whenthefillingisveryweak, theload
sharingbytheinvert isverysmall.
254
(5) Themiddlewall isweakestpartintheengineering
andhorizontal blankinglevel braceshouldbeset
toimprovethestressstate.
REFERENCES
[1] PENGDing-chao,YUANYong,ZHANGYong-wu.
Spatial effects onmid-partitiondueto excavation
of a double-arched runnel[J ]. ModernTunnelling
Technology, 2002, 39(1):4753.
[2] LUYao-zong, YANGWen-wu. Ressearchoncon-
structionschemeof Lianhuashandouble-archtun-
nel[J ]. China J ournal of Highway andTransport,
2001, 14(2):7577.
[3] LIU Hong-zhong, HUANG Lun-hai. Overview
of design and construction of tunnel with multi-
plearch[J ]. West ChinaExploration Engineering,
2001, 68((1):5455.
[4] XIA Cai-chu,LIUJ in-lei. Studyonthemiddlewall
stressof Xiangsilindoubledarchtunnel[J ].Chinese
J ournal of RockMechanicsandEngineering, 2000,
19(Supplement):11161119.
[5] LIUZhi-kui, LIANGJ ing-cheng, ZHUShou-zeng,
et al, Stability analysis of rock foundation wihth
caveinkarst area[J ]. ChineseJ ournal of Geotech-
nical Engneering, 2003, 25(5):630633.
[6] ZHAOMing-jie, AOJ ian-hua, LIU Xu-hua, et al.
Studyondeformationcharacter of thesurrounding
rockmassesconcerningtheinfluenceof karstcaves
in the bottomof tunnel[J ]. J ournal of Chogqing
J iaotongUniversity, 2003, 22(2):2023.
[7] ZHOUYu-hong, ZHAOYan-ming, CHENGChon-
guo. Optimumanalysisontheconstructionprocess
for joint aarchtunnelsinpartial pressure. Chinese
J ounal of RockMechanicsandEngineering, 2002,
21(5):679683.
[8] LI De-hong.Constructionmonitoringof multi-arch
tunnel anditsresultanalysis[J ], 2003, 40(1):5964.
[9] QI Zhi-fu,SUNBo.Constructionof multi-archtun-
nelswithtwinlargespansbyNATM[J ]. J ournal of
RailwayEngineeringSociety, 2002(1):6265.
[10] LIUGui-ying,WANGYu-xing,CHENGJ ian-ping,
et al. Structureanalysisandworkingoptimization
of the double-arch tunnel of the expressway[J ].
Geological Science and Technology Information,
2003, 22(10):97100.
[11] CHEN Shao-hua, LI Yong. Structural analysis for
ajoinedroof tunnel[J ]. ChinaJ ournal of Highway
andTransport, 2000, 13(1):4851.
[12] HANChang-ling. Structuredesignof double-arch
integritytypetunnel[J ]. Haiway, 2000(11):7981.
[13] TheSecondSurvey andDesignInstituteof China
Railway. Design technique handbook for rail-
wayengineeringTunnel [M]. Beijing: Pressof
Chinarailway, 1995: 426438.
[14] The Second Engineering Burean of China Rail-
way. Construction technique handbook for rail-
way engineeringTunnel, thenext volume[M].
Beijing:Pressof Chinarailway, 1995: 323329.
[15] ZHANG De-he. Design and construction of tun-
nel crossingcavernwithaccumulations[J ], Under-
groundSpace, 1999, 19(2):93100.
255
Safetymeasuresbyutilizingtheoldridgeroadandpotential risks
of thelandnear oldriver alignments
M. Okuda
Okuda Construction: Nagoya, Japan
Y. Nakane
Showa Concrete Industries Co., Ltd, Nagoya, Japan
Y. Kani
Nippon Concrete Industries Co., Ltd, Nagoya, Japan
K. Hayakawa
Ritsumeikan University, Kusatsu, Japan
ABSTRACT: Thisisastudypaper fromthehistorical viewpointsonthesafetymeasureandthepotential risks
of ridgeroadandoldriver siterespectively.
Many oldroadsinthemountainousareaaretheroadalongthecrest of themountainridge. Sincetheridge
roadislocatedfar away fromtheriver andlocatedonthetopof mountainousslope, theoldridgeroadcanbe
usedfor thepeoplecomingandgoingduringtheblockadeperiodof themodernriversideroadduetocalamity. It
canbeprovedbytheexaminationsof theexistenceof oldfacilitieswithalonganddistinguishedhistory, which
aretheoldtemple, oldtravelersguardiandeity, Goddessstone, statueof Mercy, etc. Theoldridgeroadshave
beensteadilyexistedfor alongperiodafter theexperiencesof historical calamities.Therefore, theoldridgeroad
canbeutilizedas thefail-safepathduringthetimeof calamity. This is also provedby theexperiences of the
recent calamityof floodandearthquake.
Thealignments of oldKisoRiver intheNobi Plainwereinvestigatedthroughtheliteratures at first. It was
foundthatthereweremanyoffshootsof oldKisoRiverintheNobi Plain.Theplace-namesneartheoldKisoRiver
andoffshootsarealsotheuseful informationtojudgethealignmentsof oldriverflow.Thegroundreconnaissance
was madealongthealignments of oldKiso River. Many sections of theoldoffshoots alignments havebeen
alreadyreclaimedanddevelopedastheresidential area, farmland, road, irrigationchannel, etc. Suchadeveloped
landhasnomoreretainedwhat it usedtobe. Thosedevelopedplaceswill havethepotential risksof floodand
thedamages by earthquakedueto thelower land, theexistenceof shallowgroundwater andtheexistenceof
shallowandloosesandylayer, whichwill havealiquefactionrisk.
1 INTRODUCTION
This is aqualitativestudy presentingthesafety mea-
sures by utilizing the old ridge road as the fail-safe
pathduringthetimeof calamity. AkihaOldRoadis
selected as a case study of old ridge road and also
presented herethreeexamples of old ridgeroad uti-
lized during thetimeof calamity. Also, this study is
presentingthepotential groundrisksof floodandliq-
uefactionoccurringalongtheoldoffshootsalignment
of Kiso River in the Nobi Plain. The ground recon-
naissancealong theold offshoots of Kiso River was
made.
Thefindingsof thesafetymeasuresbyutilizingthe
oldridgeroadandthepotential risks alongoldriver
flowareshownrespectivelyfromthehistorical view-
pointsinthisstudy.
2 OLDMOUNTAINOUSROADS
2.1 Status of old road in the mountainous area
ManymountainousoldroadsintheTokai areaarethe
roadalongthecrestof themountainridge, whichhave
beenutilizedsincetheJ apaneseWarringStatePeriod
(16Century) or before.
Also, thesaidmountainousoldroadsusedtobeing
locatedfar away fromriver andlocatedonthetopof
mountainousslope.
257
Photo 1. Gateway at the entrance to theAkiha Shrine in
HamamatsuCity.
The distinguish history of an old mountainous
road, AkihaOldRoad, isintroducedhereinafter asan
exampletothelonghistoryof oldridgeroad.
2.2 History of Akiha Old Road
Thedistinguishedhistory of AkihaOldRoadcanbe
proved by the existences of old shrines, old temple,
stonestatues, stoneimageof Buddha, etc. along the
alignment of AkihaOldRoadasfollows:
i) ShuyoTemple;Thistemplehasbeenfamousfor a
firefestival, locatedabout 700mabovesealevel
and founded by Gyoki Bodhisattva about 1300
yearsago,
ii) AkihaShrine;Thisshrinehadbeenintegratedwith
the Shuyo Temple up to the end of the Edo era
(1868). Theshrineis locatedat 885mabovesea
level and has been also famous for fire preven-
tion and metallurgy. Theshrineholds about 400
J apaneseswordsdedicatedby Samurais. Famous
dedicated swords are Masamune by Shingen
TakedaandOsafune by KansukeYamamotoin
1534.
iii) Stonestatues;Thelettersof theBunkaera(1804
1818)arechiseledonthestonestatuesof J izo.The
yearsof 1760, 1787, etc. arealsochiseledonthe
stoneguideposts.
iv) Yamazumi Shrine; This shrine has been famous
for wolf andwar, locatedabout 1100mabovesea
level andfoundedabout1300yearsago.A pair of
stoneguardianwolvesisprovidedat theentrance
of Yamazumi Shrine.
Thus,AkihaOldRoadhasbeenutilizedbythepeo-
plefor 1300yearsor moreasaroadof life, belief in
Akiha, transportationof volcanic glassaswell assalt
andaroadfor militaryoperation. Thoseevidencesof
long history havebeen spread over thealignment of
AkihaOldRoad.
2.3 Historical Calamity occurred near Akiha
Old Road
The main part of Akiha Old Road is located in the
HarunotowninHamamatsuCity.Accordingtoabook
Photo2. A StoneGuidepost alongAkihaOldRoad.
Photo3. A Pair of StoneGuardianWolvesat theEntrance
of Yamazumi Shrine.
writtenbyKishita(1984), thereisnodescriptionabout
thedisaster occurred along theAkihaOld Road, but
describesthefollowingcalamitiesoccurredalongthe
rivers:
i) Anearthquakeoccurredin715causedlandslide
andtheslidmaterialsblockedtheflowof Tenryu
River for several months,
ii) A floodoccurredin1715. Itwastheseriousflood
in180years,
iii) Anearthquakeoccurredin1854. 1400housesin
thecountywerecompletelydestroyed,
iv) Bridgeswerewashedaway duetotheheavy rain
occurredin1904.
The historical calamities have occurred near the
river in Haruno Town, but might benot often along
theridgeroad asAkihaOld Road. Furthermore, the
above-mentionedlonghistoryofAkihaOldRoadindi-
cates alsothat theridgeroadhas beenrelatively safe
andstable.
2.4 Recent examples of old ridge roads utilized
during the time of calamity
Example1; Heavy rainattackedonObara& Fujioka
village of Toyota City in J uly 1972. The rainfall of
downpour was85mm/hour andthedeathtoll fromthe
heavyrainclimbedto67. Nextdayafterthedownpour,
anoldridgeroadwaspassable, buttheprefectural road
near river wasnot passableasshowninFigure2.
258
Figure 1. Alignments of Old Akiha Road in Hamamatsu
City. (after Nakane).
Figure2. PassableRidgeRoadat ObaraVillageof Toyota
Cityafter HeavyRainin1972.
Figure3. PassableOldRoadsinYamakoshi Vill.of Nagaoka
CityAfter the2004MidNiigataPrefectureEarthquake.
Figure 4. Provision Road for Disaster andAn Old Ridge
RoadinOkazaki City.
Example2;The2004MidNiigataPrefectureEarth-
quakeof thrust typeearthquakeoccurredonOctober
23, 2004at near NagaokaCity inNiigataprefecture.
Themaximummagnitudewas6.8(J MA)andthedeath
toll fromtheearthquakeclimbedto68. Themainlife-
lineroadsinYamakoshi villageof NagaokaCitywere
not passabledueto damages fromearthquakeat that
time. However, someoldridgeroadswerepassableas
showninFigure3andtheywereutilizedastheaccess
roadjust after theoccurrenceof earthquake.
Example 3; A provision road against disaster in
Kobutownof Okazaki Cityhadbeenprovidedbythe
Okazaki city authority in 2005. This road alignment
was selectednear andparallel tothealignment of an
oldridgeroad(=ZemanjoRoad) asshowninFigure4.
259
Figure 5. Main Offshoots of Old Kiso River (after
KisogawaTownHistory, 1981).
2.5 Merit of ridge road
Thereasonswhy many mountainousroadshavebeen
chosen the ridge alignment and sustained for long
periodareasfollows:
i) War operationat theridgeroadhasanadvantage
over enemy,
ii) Safefromdangersof wildanimals, harmful insect
and viper, because the animal path is not over-
lappedwithridgeroad,
iii) Lesschanceacrossriver andswamp,
iv) Goodvisibility,
v) Lessobstructionof fallingstone.
3 OFFSHOOT ALIGNMENT OF OLD
KISORIVER
3.1 Search for old offshoots shown in literatures
OldKisoRiver ranalongSakai River andjointedwith
NagaraRiver upto1586. It wasnotedthat therewere
sevenoffshootsof oldKisoRiver after Studyonthe
Place-name of Owari Clan, 1916. Other literatures
describe not only seven offshoots of old Kiso River
but also eight offshoots and twelve offshoots of old
KisoRiver. (forexample, afterAsahi VillageJ ournal,
1963)
Therefore, it canbeunderstoodthat therehadbeen
many offshoots of old Kiso River, but not specific
number of offshoots.
Accordingtoareportof theresourcesinvestigation
committeebelongingto thePrimeMinisters Office,
therewerethreemainoffshootsof oldKisoRiveratthe
beginningof theEdoperiod(16031867) asindicated
intheFigure5.
The1st offshoot, 2ndoffshoot and3rdoffshoot of
old Kiso River was identified as Ishimakura River,
Hannya River andAzai River, respectively. Further-
more, Kuroda River and Ajika River were also the
Figure6. Typical Sectional Dimensionsof Okakoizutsumi
Dike(after Nishida).
Photo 4. Signboardof Okakoizutsumi alongpresent Kiso
River inKonanCity.
offshoots of old Kiso River. It was also known that
SayaRiver wasamainflowof oldKisoRiver.
TheOwari Clan,whohadruledtheleftsideterritory
of KisoRiver duringtheEdoperiod, hadconstructed
theleft dikeof about 48kminlengthalongoldKiso
River fromInuyamaCity toYatomi City betweenthe
years 1608 and 1609 in order to protect his terri-
tory against flood attacked by the old Kiso River.
The left dike of old Kiso River has been called as
Okakoizutsumi.Thetypical sectional dimensionsof
Okakoizutsumi dikeareshowninFigure6.
TheOkakoizutsumi dikehadbeenconstructedonly
alongtheleft bank of oldKiso River. Therefore, the
right bank area of old Kiso River had been often
affectedbyflooding.Also, all flowsof offshootexcept
SayaRiver had been blocked off theforks fromold
KisoRiver bytheembankment of Okakoizutsumi.
ItcanbenoticedfromFigure6thatsometreeswere
plantedonthedikethoughthemodernJ apaneseriver
codedoes principally not allowto plant trees on the
outer slopeof embankeddike.
Planting tree on the outer dike, however, could
be said a J apanese traditional method. Because, the
article 17 of theYozenryo (= a regulation of build-
ing & repairs) in theTaihoritsuryo (= old J apanese
regulationestablishedin701) specifiedthat thetrees
(= elm, willow, etc.) shouldbeplantedonthedike.
3.2 Ground reconnaissance on alignment
of old Kiso river
Somedikesectionsof Okakoizutsumi still existalong
thepresent KisoRiver andalsoalongtheleft bank of
theabandonedSayaRiver.
260
Figure7. AnExistingSectional Dimensionof Okakoizut-
sumi Dikeat FutagoTowninAisai City(after Nakane).
Figure 8. Soil Profile along Okakoizutsumi Dike from
SobueareatoTatsutaarea(after Nakane).
Anexistingsectional dimensionof abandoneddike
of Okakoizutsumi at Futago Town in Aisai City is
showninFigure7.
The soil profile along Saya River as determined
fromthreeboringdataisshowninFigure8. Thebor-
ingpointA, point B andpoint C islocatedalongold
SayaRiver offsettingfromKisoRiver around30km,
26kmand20kmupper fromthepresent Kiso River
mouth, respectively.
Thestratanameof soil profilearecomplied with
thedescriptionshownintheliteratureof Groundof
Inazawa becausetheorder of stratahas likeness to
thesoil profileof westernInazawaCity presentedin
thisliterature.
After theblockageof branch forks in 1610, some
sections of the abandoned offshoot have gradually
reclaimedanddevelopedastheresidential area, farm-
land, road, irrigationchannel, etc.
TheSayaRiver has beenalsoblockedoff thefork
from Kiso River in 1900 due to the dike burst at
Utasuof Aisai City in1897anddevelopedalongthe
abandonedSayaRiver.
Photo5. Traceof 1stOffshoot,Aoki RiveratKashiwamori.
Photo 6. Trace of 2nd Offshoot, Hannya River at Konan
City.
Photo 7. Trace of 3rd Offshoot, Nikko River at Inazawa
City.
Thedevelopedlandalongoldoffshoothasnomore
retainedwhat it usedtobeasshownabove.
Aoki River isthepresentnameof IshimakuraRiver
whichis identifiedwiththe1st offshoot of oldKiso
River.
Theplace-nameindicatesalsothetraceof oldKiso
River alignment.
For example, theplace-nameKotsuattheforkof
1st offshoot indicates thetimberland of drift-timber.
Kuraof IshimakuraRiver (=1stoffshoot) indicates
thenarrowgroundbetweenrivers. Hannyaof Hannya
River (=2nd offshoot) called originally Haniya,
261
Figure9. Zero-meter AreainNobi Plain(after Society of
Tokai-three-Pref. GroundSettlement Investigation).
whichindicatesthedrieduplandof mudflow.Azai of
Azai River(=3rdoffshoot) indicatestheswampyarea
and/or theexistenceof shallowgroundwater.
Also, theminor place-namesrelatedwithriver and
swamp can befound along theoffshoot areas of old
Kiso River, such as Furukawa (=old river), Sunaba
(=Sandyground), Kawahara(=riverbed), Hatagawa
(=old river name), Hasuike (=old swamp name),
etc. Thesegroundswhichhavetheminor place-name
relatedwithriver andswampusedtobelower thanthe
surroundingground.
3.3 Potential risks of flood at the lower ground
TheNobi Plainspreadsover about 1,300km
2
andhas
thegentleslopefromnortheasttosouthwest.TheNobi
Plainisalsoknownwithzero-meterareaatthesouth
part of theplain. Zero-meter areaspreads over about
274km
2
(after Committee of Ground Subsidence at
theTokai ThreePrefectures,1988)whichoccupyabout
21%of Nobi Plainarea. Most partsof thezero-meter
areahadbeenfloodedatthetimeattackedbytheVera
Typhoon(=IsewanTyphoon)in1959.Thiszero-meter
areainNobi Plainisthelargest zero-meter groundin
J apanandhasapotential riskof flooding.
Somelocal governmentshaveannouncedthepoten-
tial riskof floodingareasnear theoffshootof oldKiso
River, which information are available in the home
pageof local government viainternet. Anexampleof
floodingsimulationalongAoki River (=1stoffshoot)
isshownFigure10.
Thus, thezerometer areaandsomeareasnear the
offshoot alignment of oldKisoRiver haveapotential
riskof flooding.
The said flooding risk can be also recognized by
thefollowingmapwhichisoverlappedFigure5(after
KisogawaTown)inArticle3.1andamapof anticipated
Figure 10. Flooding Simulation Map of Aoki River in
40min. after Burst of Aoki River Dike at 2.2km (after
InazawaCityH.P.).
Figure11. Anticipated flooding areaand theoffshoots of
oldKisoRiver (afterAichi Pref. 1978andKisogawaTown).
flooding area when the right dike of Kiso River is
burstedatYamanainFusoTown(after FloodControl
Planof Aichi Prefecture, 1978).
3.4 Potential risks of liquefaction near the offshoots
of old Kiso River
Theauthorityof Aichi prefectureannouncedofficially
thepredictiondegreeandzoneof liquefactionagainst
thefuturemega-earthquakeasshowninFigure12.
Inordertochecktheliquefactionpotential alongthe
alignment of Okakoizutsumi, four sandysoil samples
262
Figure 12. Prediction degree and zone of liquefaction
against thefutureTonankai Earthqueke(Aichi pref. H.P.).
Table1. Criteriaof LiquefactionPotential byJ BFDC,1974.
Item J BFDC Criteria Okakoizutsumi
Finer than#200sieve <10% 2.90.5%
D
50
0.0752.0mm 0.340.47mm
Uc <10 2.33.6
were collected from Okakoizutsumi dike at Bisai,
Sobue, Saya and near Yatomi. These samples are
expectedto belongto thetopsandy layer of thesaid
places. Because, the borrow pit for the Okakoizut-
sumi earthworkunder thelabor intensiveconstruction
would be the adjacent ground and within about two
metersindepthfromthegroundsurface.
Thesesandy soil samples havethegraindiameter
corresponding to the50%finer line(D50) of 0.34
0.47cmandthecoefficientof uniformity(Uc) of 2.3
3.6. According to theJ apaneseBuilding Foundation
DesignCode(J BFDC, 1974), thefollowingconditions
havethepotential of liquefactionasshownintable1.
Hence, the sandy soil samples obtained from
Okakoizutsumi canbeevaluatedashavingapotential
riskof liquefactionunder theJ BFDC (1974) criteria.
Also, theJ apaneseMarineStructural DesignCode
(J MSDC, 1989) specifies the liquefaction potential
zonebythesoil gradation. Thegradationof thesandy
soil samplesobtainedfromOkakoizutsumi arewithin
thehigher potential zoneof liquefactionas shownin
Figure13.
On the other hand, the Saori Town History Book
presentsthepastobservationof 157liquefactioncases.
Furthermore, theAichi Deposited Culture Center
inYatomi Cityholdarelief of quicksand(=estimated
occurrence during the Tensho Earthquake in 1586)
penetrated into the ground of Kiyosu castle down-
town remains, where was located near Gojo River.
GojoRiver wasthedownstreamof 1st offshoot of old
KisoRiver. Thefollowingfigure14showsthetraces
of quicksandobservedat thearchaeological remains
of Kiyosucastledowntown.
Asmentionedabove, someareasnear theoffshoot
alignmentsof oldKisoRiverintheNobi alluvial plain
haveapotential riskof liquefaction.
Figure 13. Gradations of soil samples obtained from
Okakoizutsumi andtheliquefactionpotential zonespecified
inJ MSDC.
Figure14. Traces of QuicksandObservedat theArchaeo-
logical Remainsof KiyosuCastleDowntown(after Hattori).
4 CONCLUSION
Theoldridgeroadsarerelativelysafeandstable,which
canbeprovedby their longhistory andtheir present
shape. Therefore, someoldridgeroadscanbeutilized
as the fail-safe path during the blockade of modern
paved riverside road at the time of future calamity,
whichcanbealsoprovedbytherecentcasesmentioned
inthearticle2.4of thisreport.Therefore, theoldridge
road should be paid more attention and maintained
properly, sothatthoseoldridgeroadswill serveasthe
fail-safepathduringthefuturecalamity.
Somereclaimedgroundsandhinterlandsalongthe
offshoot alignment of oldKisoRiver havethepoten-
tial risksof floodingandliquefactionduringthetime
of futurecalamities. Because, thesaid areas used to
have(1)thelowergroundthanthesurroundingground,
(2) the existence possibility of shallow groundwa-
ter, and (3) the existence possibility of shallow and
loosesandylayer. Manyreclaimedgroundsalongthe
263
offshoot alignment of old Kiso River can be identi-
fiedbytheplace-namewhichisrelatedwithriver and
swamp.
Though some local governments have officially
announcedaboutthosepotential risksof floodingand
liquefaction, thegrounddeveloper, buildingdesigner
andpropertyowner shouldalsopayattentiontothose
potential riskswhentheyexecutetheprojectinthearea
of oldoffshoot alignmentsaswell astheir hinterland.
From the historical viewpoint mentioned in this
paper, calamity will repeat itself at theriversideroad
andalongtheoldoffshoot site. Weshouldnot forget
thispotential groundriskof thehistorical repetition.
REFERENCES
Aichi Pref. 1964. Recovery report fromthecalamity of Ise
BayTyphoon. p.221, Nagoya. Aichi Pref.
Aichi Pref. 1978. Floodcontrol planof Aichi prefecture, p.
274, Nagoya. Aichi Pref.
Aichi Pref. H.P. 2008. Damage prediction of Tokai &
Tonankai Earthquake.http://www.pref.aichi.jp/bousai/all/
all.htm. Nagoya. Dept. of Disaster Prevention. AichPref.
AkihaShrine. 1999. SwordsreservedbyAkihaShrine. pp.64,
80, Hamamatsu. AkihaShrine
Editing committee of Asahi village journal. 1963. Asahi
villagejournal. p. 111, Nagoya. IwataS.
Editing committee of Bisai city history. 1998. Bisai city
historyVol. 2. p. 8, Bisai, Bisai city.
Editingcommitteeof Kisogawatownhistory.1981.Kisogawa
townhistory. pp. 4142, Aichi, KisogawaTown.
Editing committeeof Saori town history. 1989. Saori town
history. p. 17, Aichi. Saori Town
Hattori T. 1993. Annual report of Heisei 4financial year. pp.
129-131. Nagoya. Aichi Pref. DepositedCultureCenter
Inazawa City H.P. 2008. Flooding Simulation Map.
http://www.city.inazawa.aichi.jp/kurashi/hazard_map/
simulation/index.html. Inazawa. InazawaCity.
J GC-Chubucommitteeof study ontheNobi ground. 1996.
Groundof Inazawa. pp.29,40,45, Inazawa. InazawaCity
J GS. 2006. Ground at Nobi Plain, Geotech Note 15.
pp.19,105, Tokyo. Maruzen
KishitaT. 1984. Chroniclesof Haruno. pp.11, 128, 166, 212,
Hamamatsu. KishitaT.
Kubomi M. 1916. New interpretation of Taiho RyoVol. 3.
p. 591, Tokyo. MeguroJ.
Nakane Y. et al. 2008. A consideration about the practi-
cal use of the old road on the disaster, Proceedings of
the 28th Annual Conference of Histrical Studies in Civil
Engineering. pp. 203214, Tokyo. J SCE
NakaneY. et al. 2008. A Study ontheOldRiver Channels,
No.141 Symposium for the construction and conservation
technique of historical geotechnical structure. pp. 1320,
Tokyo. J GS
NishidaS.1994.LifeontheKiso-three-river.pp.5153,Gifu,
Institutefor water andcultureof Kiso-three-river
Oka M. 1979. Farmers biography 7, Nihon Agricul-
ture Library Vol. 16. p. 293. Tokyo, Association of
farm/mountain/fishingvillagesculture
River Front MaintenanceCenter. 1999. Management man-
ual of tree in the river boundary. pp. 183193. Tokyo.
Sankaido
Society of Tokai-three-Pref. Ground Settlement Investiga-
tion. 1985. Groundsettlement andgroundwater at Nobi
Plain. p.115, Nagoya. NagoyaUni. Press
TakigawaS. 1931. StudyonRitsuryo. P.???
Tsuda. M. 1916. Studyontheplace-nameof Owari Clan. pp.
243, 245, 246, 253, 312, 497. Tsushima. Educationboard
of AmaCounty, Aichi Pref.
264
Bearingcapacityof rigidstripfootingsonfrictional soilsunder eccentric
andinclinedloads
K.Yamamoto
Department of Ocean Civil Engineering, Kagoshima University, Kagoshima, Japan
M. Hira
Department of Environmental Sciences and Technology, Kagoshima University, Kagoshima, Japan
ABSTRACT: Finiteelementanalyseswereappliedtofindtheexactlimitloadof rigidstripfootingsonfrictional
soilsundereccentricandinclinedloads. Inadditiontothevaluesof limitbearingcapacity, thecontactnormal and
shear stressdistributionsbelowthefootingwereobtained. Theresultswerecomparedwiththestandarddesign
assumptions of aneffectivewidthby Meyerhof for bearingcapacity calculations andof linear distributionof
thecontact normal stress on thebaseof thefooting. Thecomparison suggests that Meyerhofs andHansens
proceduresareunconservativefor largeeccentricities. Finally, abearingcapacityequationwasnewlyproposed
toobtainthereasonablebearingcapacityinthelargeeccentricities.
1 INTRODUCTION
Footings are frequently subjected to eccentric and
inclined loads due to moments resulting fromload-
ingsduetotheactionof wind, waterorearthpressures
among several other possible sources. In practice,
the design of such footings is usually conducted in
two parts: an eccentric vertical load; and acentered,
inclinedload.Thebearingcapacityof thefootingthen
obtained by analyzing the problemin two separate
parts: (1) thebearingcapacityof footingsubjectedto
eccentric vertical loads; and(2) thebearingcapacity
of footing subjected to centered, inclined loads. The
two values of obtained bearing capacities aresuper-
posedtoget thebearingcapacityof footingsubjected
toeccentricandinclinedloads.
Many researchers have investigated the footing
subjected to eccentric vertical loads (e.g., Meyerhof
1953; Prakash & Saran 1971; Purkayastha & Char
1977). Similarly, the footing subjected to centered,
inclined loads have also been investigated by many
researchers (e.g., Meyerhof 1953; Saran et al. 1971;
Hanna & Meyerhof 1981). A few researchers have
investigated the footing subjected to eccentric and
inclined loads with the help of model tests (e.g.,
Meyerhof 1953; Saran&Agrawal 1991). Also, Saran
& Agrawal (1991) obtained the bearing capacity of
an eccentrically inclined, loaded footing using limit
equilibriummethod. Despitethecommonoccurrence
of inclinedandeccentricloads, rigoroussolutionsfor
thebearingcapacityandthecontactstressdistribution
for footings subjected to such loads were not previ-
ously available. Instead, approximatesolutions often
based on empirical observation have been typically
used (e.g., Meyerhof 1953; Hansen 1970). A com-
monlyusedapproachfor dealingwitheccentricloads
in design was first suggested by Meyerhof (1953).
Meyerhof suggested that an effective width B2e,
definedasthefootingwidth(B) reducedbytwicethe
loadeccentricity(e), beusedinthecalculationof bear-
ingcapacity. Thisprocedurehasbeenwidelyadopted
ingeotechnical design. For eccentricallyloadedfoot-
ings, thecontactstressdistributionwiththesoil isnot
uniform. Indesign, thecontact stress distributionfor
aneccentrically loadedfootingwitheccentricities in
bothdirectionsistypicallyassumedtobelinearinboth
theB andL directions, but this is asimplificationof
thereal distribution. Pecketal. (1953) pointedoutthat
if aneccentrically loadedfootingistofail inbearing
capacity, theright comparisonwouldbebetweenthe
maximumstressat thetoeof thefootingandthelimit
bearingcapacityfor amuchsmaller effectivewidthof
thefooting.Thisisconceptuallythemostcorrectsolu-
tionsinceanybearingcapacityfailurewill takeplace
at or near thetoeof thefooting. Astotheunit bearing
capacitynear thetoeinsands, it isnot thesameaswe
would calculatefor thewholefooting in thecaseof
acentrally loadedfooting, inwhichthewholewidth
B is operative. The difficulty with this procedure is
that noresultsexist prescribingwhat fractionof B to
use for calculating the bearing capacity. Peck et al.
(1953) suggested1/3of B, butthisistooconservative.
Regardingtheapproachfor dealingwithinclinedload
indesign, Meyerhof (1953) andHansen(1961) have
proposedinclinationfactorsi
c
, i
andi
q
for reduction
of bearingcapacityfactorsN
c
, N
andN
q
.
Theobjectiveof thispaperistoapplyfiniteelement
analyses to evaluatethebearing capacity of arough
265
Figure1. Signconvention.
rigid strip footing on frictional soils under eccentric
and inclined loads, as well as the contact normal
and shear stress distributions below the footing. In
addition,thevalidityof currentdesignmethodisinves-
tigated. The results are compared with the standard
designassumptions of aneffectivewidthfor bearing
capacity calculations (Meyerhof 1953) and of linear
distribution of thecontact normal stress on thebase
of thefooting. Theresultsobtainedinthispaper may
provideimportant informationfor futuredesign.
2 FINITE ELEMENT ANALYSIS
Finiteelement analysis canbeconductedto approxi-
matetheexactlimitloadforaroughrigidstripfooting
under eccentric and inclined loads. Thecontact nor-
mal and shear stress distributions below the footing
wereobtained. Themodel followstheMohr-Coulomb
yieldcriterionwithoutthegradientsingularitiesatthe
corners of the yield surface in the deviatoric plane
(Abbo & Sloan 1995). An associated flow rule was
assumed. The mesh was composed of 6-noded tri-
angular elements with quadratic shape functions in
plain-straincondition. Everymeshusedintheanalysis
wasmadeveryfinenear thepointof loadapplication.
Thefiniteelement analysiswasconductedby first
applying gravity to a level ground to construct the
initial stressfield,thenthevertical andhorizontal load-
ings, incrementally, until a limit load was defined.
A Newton-Raphsoniterativeschemewasusedtosolve
theproblem. TheNewton-Raphsonschemeisapopu-
lar methodfor thesolutionof nonlinear elasto-plastic
problems. Itisanincremental-iterativescheme, which
uses tangent stiffness iterationtoobtainthesolution.
Inthescheme,thestiffnessmatrixisupdatedaftereach
iteration. It was confirmedthat themodulus of elas-
ticity (E) and Poissons ratio () of the soil do not
influence the value of the limit load, provided that
E and areinareasonablerange.
3 RESULTSANDDISCUSSION
Figure 1 shows the sign convention. Regarding the
signs of the load eccentricity e and the load incli-
nation angle , the left side fromthe center of the
footingandtheclockwisedirectionfromthevertical
arepositive. Figure2 shows atypical finiteelement
mesh.Thebottomboundaryof themeshwasfixedand
thelateral boundariesweremodeledwithrollers. The
lateral boundarieswereplacedfarenoughnottoinflu-
encethelimit loadandfailuremechanism. Although
thebottomboundaryappearsnot tohavebeenplaced
verydeep,itwasconfirmedthattheresultswereessen-
tiallythesameaswhenthebaseisplacedatadepthof
4B. Thethicknessof thefootingis0.1B. Thefooting
wasassumedtobeveryrigidcomparedwiththesand.
Themeshwas composedof 4160triangular 6-noded
elements. Themeshwas particularly densealongthe
direction of the applied load and near the edges of
the footing. The soil parameters were set as c=0,
=30
, 35
and40
. Intheanalysis, therangesof the

eccentricity-to-widthratio, e/B=0, 1/12, 1/6, 1/3and
theinclination angle, =0
, 5
, 10
, 15
werecon-
sidered. Figure3shows theload-displacement curve
for thecases of centeredandeccentric, vertical load
(e/B=0, 1/12, 1/6, 1/3 and =0
) obtained from
FEM. Intheanalysis, themulti-increment stepswere
usedtoget moreaccuratelimit load. Thisfigureindi-
cates that thecurves convergeto somevalues as the
vertical displacement increases. This valueof q
b
can
betakenasthelimitunitvertical loadatthebaseof the
footing. ItisfoundfromFig.3thatwhene/Bincreases,
theconvergent valuefor q
b
/B decreases.
Figure4shows thedistributionof contact normal
stressbelowthefootingunder centeredandeccentric,
vertical loadings. In FEM, the stresses at the inte-
grationpoints of elements areobtained. Thestresses
of nodal points belowthefootingareinterpolatedby
thoseof theintegrationpointsof theelements, which
constitutethenodal points. For thenodesconstituting
several elements, the average value of contact nor-
mal stresses was used. The straight lines represent
theresults obtainedfromtheFEM. Thedashedlines
represent the linear distribution of contact stresses
expressedas:
for e,B1/ 6, and
for e,B>1 / 6; where B and L are the width and
lengthof thefooting; Q isthetotal vertical load; and
e is theeccentricity. It is notedthat inEq. (1), when
e/B becomes 1/6, q
b,min
equals zero. For e,B>1,6,
q
b,min
will benegative, whichmeansthat tensionwill
develop. Sincesoil cannot takeanytension, therewill
beaseparationbetweenthefootingandthesoil.Then,
theshapeof thepressuredistribution will becomea
triangle, described by Eq. (2). It is currently consid-
eredthat theexact distributionof contact pressuresis
difficult toestimate. InFig. 4(a), whichisthecaseof
centered,vertical loading(e/B=0and=0
),thedis-
tributionis symmetrical withrespect to thecenter of
266
Figure2. Typical finiteelement meshfor FEM.
Figure 3. Relationship between q
b
/B and vertical
displacement/B fromFEM ( =35
, =0
).
thefooting, andthemaximumvalueof thedistribution
is obtainedat thecenter of thefooting. Notethat the
tensionispositivefor thecontact normal stressbelow
the footing. As the eccentricity-to-width ratio (e/B)
increasesbeyond1/6, theFEMshowsthattheextentof
thecontact stressdistributionsandthemaximumval-
uesbecomesmaller, consistentwiththelossof contact
betweenthefootingandthesoil atthetrailingedgeof
thefooting. Accordingly, q
b
/B alsodecreases. From
Fig. 4, themaximumvalueof thedistributionobtained
fromtheFEM occurs almost at thepoint of applica-
tionof theloadQ. Forthecaseof e,B=1/3and=0
showninFig. 4(d), it isobservedthat theareaof con-

tactbetweenthesoil andthefootingisentirelylocated
ontheleftsideof thefooting.AsshowninFig. 4, both
the distribution shapes and the maximumvalues of
contact normal stressareingoodagreement between
FEM andthelinear distribution.
Figure 5 shows the distribution of contact shear
stress belowthefootingfromFEM correspondingto
Fig. 4. In Fig. 5(a), which corresponds to Fig. 4(a),
thedistributionof contactshear stressissymmetrical.
Regardingthesignof thecontactshearstressbelowthe
footing, theclockwisedirection is positive. Likethe
distributionof contact normal stressshowninFig. 4,
thedistribution and thevalueof contact shear stress
becomesmaller, ase/B increases. Thepoint of appli-
cationof theload, at whichthevalueof contact shear
stress equals zero, separates negative shear stresses
on theright frompositiveshear stresses on theleft.
When e/B increases, the point at which the contact
shearstresschangesfrompositivetonegativemigrates
inthesamedirectionastheeccentricitye. Inthecase
of e,B=1/ 3and=0
, thecontact shear stressesis

mobilized at theleft sideof thefooting as shown in
Fig. 5(d). Inthecaseof eccentricandvertical loading,
the variations of the distributions of contact normal
andshear stressesbelowthefootingaremoreremark-
ableat thedirectionof theeccentricity. Thetendency
wherethezigzagis appearedinthedistributions can
befoundinFrydman& Burd(1997). This wouldbe
duetotheeffect of thediscretizationof continuumin
finiteelementmethod, theinterpolationfromtheinte-
gration points of elements and theaveragevaluefor
thenodesconstitutingseveral elements.
Figures 6 and 7 show thedistributions of contact
normal and shear stresses below the footing under
centeredandeccentric, inclinedloadings. Liketheten-
dencyshowninFigs. 4and5, whene/B increases, the
extentof thecontactstressesdistributionsandthemax-
imumvalues becomesmaller. q
b
/B also decreases.
Thezigzaginthedistributions is not muchobserved
atthedirectionof theapplied(inclined) loading, com-
paringwiththatof theoppositeside.InFig.7,theareas
wherenegativeshear stresseswork, increasewhenthe
angle of load inclination is positive (the clockwise
directionfromthevertical), comparingwithFig. 5.
Figure 8 shows the relationship between q
b
/B
and eccentricity-to-width ratio (e/B) for the cases
of =30
, 35
and 40
. In this figure, the solu-

tionobtainedfromFEM andthesolutionscalculated
usingMeyerhofsandHansensbearingcapacityequa-
tion are presented. The bearing capacity equations
proposedbyMeyerhof (1963) andHansen(1970) are:
whereB isthefootingwidth; e istheeccentricity; is
theangleof theloadactingonthefootingwithrespect
tothevertical; and aretheinternal frictionangle
andunitweightof thesoil; N
inEqs. (3) and(4) isthe

bearingcapacityfactor givenbyMeyerhof (1963) and
Hansen(1970), respectively; andB2e isMeyerhofs
effectivefootingwidth.
RegardingtheN
value, itisobservedthatthesolu-
tionobtainedfromFEM agreeswell withMeyerhofs
solution. The values q
b
/B obtained from Eq. (3)
are always larger than those obtained from FEM,
exceptwhene/B=0,whentheynearlymatchtheFEM
267
Figure 4. Contact normal stress distribution on footing
basefromFEM ( =35
).
values. Also, thevalues q
b
/B obtainedfromEq. (4)
are larger than those fromFEM for e,B>1/12. It
is observed that the Meyerhofs solution is always
larger thanHansenssolution. Whentheinternal fric-
tion angle increases, the difference between Meyer-
hofsandHansenssolutionsbecomelargeparticularly
for e,B=0. However, whentheeccentricity-to-width
Figure5. Contact shear stressdistributiononfootingbase
fromFEM ( =35
).
ratio(e/B) increases, thedifferencegetssmaller grad-
ually.TheseobservationssuggestthatMeyerhofsand
Hansens Eqs. (3) and (4) tend to overestimate the
bearingcapacitywhene/Bbecomeslarge(e,B>1/6).
Thisfactagreeswell withthatderivedbyMichalowski
&You(1998). TheyexaminedMeyerhofsconcept of
aneffectivewidth(Meyerhof 1953), andshowedthat
268
Figure 6. Contact normal stress distribution on footing
basefromFEM ( =35
).
its use may lead to upper bounds that are large for
cohesionlesssoils.Thefollowingequationisproposed
not to overestimate the bearing capacity in large
eccentricities.
Figure7. Contact shear stressdistributiononfootingbase
fromFEM ( =35
).
where N
is the bearing capacity factor given by

Hansen(1970); andthekeyof Eq. (5) isthat B2.5e
is usedinsteadof B2e. It is observedthat thepro-
posedequationiswell correspondedtoHansensvalue
for e,B1/12andis similar to thosefromFEM for
e,B>1/12.
269
Figure 8. Relationship between q
b
/B and eccentricity-
to-widthratio(e/B) for variousfrictionanglesand=0
.
Figure 9 shows the relationship between q
b
/B
and inclination angle for =35
. It is observed
that Meyerhofs and Hansens solutions are larger
than the FEM values for =0
in Fig. 9(b) and

=0
, 5
, 10
and 15
in Figs. 9(c) and (d). Par-

ticularly for the case for e/B=1/3 (Fig. 9(d)), the
values obtained fromMeyerhofs and Hansens Eqs.
(3) and (4) are much larger than the FEM values,
when is small. In Fig. 9(a), theproposed equation
agrees well with Hansens value when =0
. Also,
the equation gets closer to the FEM values when
increases. In Figs. 9(b)(d), it is observed that the
equationissimilar totheFEMvaluescomparingwith
Figure9. Relationshipbetweenq
b
/Bandtheloadinclina-
tionangle for =35
.
Meyerhofs and Hansens solutions. As a result, we
canstatethat thesolutionsobtainedfromMeyerhofs
and Hansens Eqs. (3) and (4) are not accurate and
tend to overestimate the bearing capacity when the
eccentricityislarge(e,B1/3).
270
4 CONCLUSIONS
The bearing capacity of a rough, rigid strip footing
on purely frictional soil subjected to eccentric and
inclined loads was analyzed using finite element
method. The bearing capacities obtained from the
finiteelement methodwerecomparedwiththosecal-
culatedfromthebearingcapacity equationproposed
byMeyerhof (1963) andHansen(1970).
Theconclusionsdrawninthispaperaresummarized
asfollows:
(1) Thefiniteelementanalysisproducescontactstress
distribution and maximumcontact stress values
that areinbetter agreement withthecorrespond-
ingvaluesfor thelinear stressdistribution.
(2) The bearing capacities calculated using the
Meyerhof (1963) and Hansen (1970) equations,
which is expressed in terms of aeffectivewidth
B2e, are not accurate and tend to overesti-
matethebearingcapacities, particularlywhenthe
eccentricityislarge(e,B1,3). TheMeyerhofs
solution is always larger than Hansens solution.
Whentheeccentricity-to- widthratio(e/B) or the
inclinationangleincreases,thedifferencebetween
the Meyerhofs and Hansens solutions become
small gradually. In general, foundation design
avoids large eccentricities, so this deficiency in
theMeyerhof (1963) andHansen(1970) solutions
arenot usuallypresent inasignificant way.
(3) The bearing capacity equation is proposed not
to overestimatethebearing capacity in thelarge
eccentricities. It isshownthat theproposedequa-
tioncangivethereasonablebearingcapacityeven
inthelargeeccentricities.
REFERENCES
Abbo,A.J.&Sloan,S.W.1995.Asmoothhyperbolicapprox-
imation to the Mohr-Coulomb yield criterion, Comput.
Struc., 54(3): 427441.
Frydman,S.&Burd,H.J.1997.Numerical studiesof bearing-
capacityfactor N, J. Geotech. Geoenviron. Eng., ASCE,
123(1): 2029.
Hanna, A. M. & Meyerhof, G. G. 1981. Experimental eval-
uationof bearingcapacity of footingssubjectedinclined
loads, Can. Geotech. J., 18: 599603.
Hansen, J. B. 1961. A general formulafor bearingcapacity,
Danish Geotech. Inst. Bull., 11: 3846.
Hansen, J. B. 1970. A revised and extended formula for
bearingcapacity, Danish Geotech. Inst. Bull., 28: 511.
Meyerhof, G. G. 1953. Thebearingcapacity of foundations
under eccentricandinclinedloads, Proc. of 3rd ICSMFE,
Zrich, 1: 440445.
Meyerhof, G. G. 1963. Somerecent researchonthebearing
capacityof foundations, Can. Geotech. J., 1(1): 1626.
Michalowski, R. L. & You, L. 1998. Effective width rule
in calculations of bearing capacity of shallow footings,
Comput. Geotech., 23(4): 237253.
Peck, R. B., Hanson, W. E. & Thornburn, T. H. 1953.
Foundation Engineering, J ohnWiley & Sons, Inc., New
York.
Prakash, S. & Saran, S. 1971. Bearingcapacity of eccentri-
cally loaded footings, J. Soil. Mech. And Found. Engrg.
Div., ASCE, 97(1): 95117.
Purkayastha, R. D. & Char, A. N. R. 1977. Stabilityanalysis
for eccentricallyloadedfootings, J. Geotech. Engrg. Div.,
ASCE, 103(6): 647651.
Saran, S., Prakash, S. & Murty, A. V. S. R. 1971. Bearing
capacity of footings under inclined loads, Soils Found.,
11(1): 4752.
Saran, S. &Agrawal, R. K. 1991. Bearingcapacityof eccen-
tricallyobliquelyloadedfooting, J. Geotech. Eng.,ASCE,
117(11): 16691690.
271
Uncertainty
Reliabilityanalysisof slopestabilitybyadvancedsimulation
withspreadsheet
S.K. Au,Y. Wang& Z.J. Cao
Department of Building and Construction, City University of Hong Kong, Hong Kong, China
ABSTRACT: This paper develops apackageof EXCEL worksheets andfunctions/Add-Into implement an
advancedMonteCarlomethodcalledSubset SimulationintheEXCEL spreadsheet andappliesit toreliability
analysisof slopestability.Thedeterministicslopestabilityanalysisanduncertaintymodelingandpropagationare
deliberatelydecoupledsothattheycanproceedseparatelybypersonnel withdifferentexpertiseandinaparallel
fashion. Anillustrativeexampledemonstrates applicationof theEXCEL packageto aslopewithuncertainty
onundrainedshear strengthandhighlightscomputational efficiencyof SubsetSimulationfor theslopestability
problem.
1 INTRODUCTION
Thereluctanceof geotechnical practitioners to apply
reliability methods to slope stability analysis is
attributed, among other factors, to thesophistication
of advancedprobabilisticassessment/modellingmeth-
ods, limitedpublishedstudies/workedexamplesillus-
tratingtheimplementation, andlack of user-friendly
tools. Fromanimplementationpoint of view, theless
informationrequiredfromtheengineersregardingthe
probabilistic assessment or reliability computational
algorithm, thesmaller hurdletheengineer will facein
properlyusingthealgorithm,andthemorelikelyitwill
beimplemented. Therefore, itisdesirabletodecouple
theprocessof deterministicslopestabilityanalysisand
reliabilityanalysissothat theworkof reliabilityanal-
ysis can proceed as an extension of deterministic
analysisinanon-intrusivemanner. It isalsodesirable
to implement the reliability analysis algorithmin a
softwareplatformwithwhomtheengineersarefamil-
iar. Fromthis perspective, the ubiquitous Microsoft
EXCEL spreadsheet isparticular of interest.
Low(2008) showedthat geotechnical analysisand
theexpandingellipsoidal perspectiveof theHasofer-
Lindreliability index can bereadily implementedin
a spreadsheet environment. The Hasofer-Lind relia-
bilityindexcanbeobtainedusingtheobject-oriented
constrainedoptimizationtool intheExcel spreadsheet
(LowandTang 1997, 2007). Theapproach has been
appliedtoobtainthereliabilityindexof conventional
bearingcapacityproblem(Low2008), anchoredsheet
pile design (Low 2005a&b), slope stability analysis
(Lowet al. 1998, Low2003).
This paper implements an advanced MonteCarlo
methodcalledSubsetSimulation(AuandBeck2001)
intheEXCEL spreadsheet andillustratesitsapplica-
tiontoreliabilityanalysisof slopestability. After this
introduction, theSubset Simulationalgorithmwill be
brieflydiscussed, followedbydevelopmentsof Subset
Simulationtool andslopestabilityanalysisworksheets
inEXCEL.Then,anexampleof slopestabilityreliabil-
ityanalysiswill bepresentedtoillustratetheanalysis
processusingtheEXCEL spreadsheets.
2 SUBSET SIMULATIONALGORITHM
Subset Simulation is an adaptive stochastic simula-
tion procedure for efficiently computing small tail
probabilities (Au and Beck 2001, 2003). Originally
developedfor dynamicreliabilityanalysisof building
structures, it stems fromtheideathat asmall failure
probability can be expressed as a product of larger
conditional failureprobabilitiesforsomeintermediate
failureevents, therebyconvertingarareeventsimula-
tionprobleminto asequenceof morefrequent ones.
Duringsimulation, conditional samplesaregenerated
fromspecially-designed Markov chains so that they
populate gradually each intermediate failure region
until theyreachthefinal target (rare) failureregion.
Let Y be a given critical response for which
P(Y >y) is of interest, let 0-y
1
-y
2
- -y
m
=y
bean increasing sequenceof intermediatethreshold
values.Itshouldbenotedthatconsideringasinglecrit-
ical responseleadstolittlelossof generality because
multiple failure criteria can be incorporated into a
singleone(AuandBeck 2001). Bysequentiallycon-
ditioningontheevent{Y >y
i
}, thefailureprobability
canbewrittenas
The raw idea is to estimate P(Y >y
1
) and
{P(Y >y
i
|Y >y
i1
): i =2,,m}by generatingsam-
ples of O conditional on {Y(O)>y
i
): i =1,,m}.
275
Figure1. Schematicdiagramof Subset Simulationprocedure.
In implementations, y
1
, , y
m
are generated adap-
tively using information from simulated samples
so that the sample estimate of P(Y >y
1
) and
{P(Y >y
i
|Y >y
i1
): i =2,,m} always correspond
toacommonspecifiedvalueof theconditional prob-
abilityp
0
.
Theefficient generationof conditional samples is
highly-nontrivial but pivotal in the success of Sub-
set Simulation, and it is made possible through the
machinery of Markov Chain Monte Carlo (MCMC)
simulation(Roberts & Casella1999). Markov Chain
MonteCarloisaclassof powerful algorithmsfor gen-
erating samples according to any given probability
distribution. It originates fromthe Metropolis algo-
rithmdeveloped by Metropolis and co-workers for
applications in statistical physics (Metropolis et al.
1953). In MCMC, successivesamples aregenerated
fromaspeciallydesignedMarkovchainwhoselimit-
ingstationary distributiontends to thetarget PDF as
thelength of theMarkov chain increases. An essen-
tial aspect of the implementation of MCMC is the
choiceof proposal distribution thatgovernsthegen-
erationof thenext samplefromthecurrent one. The
efficiencyof SubsetSimulationisrobusttothechoice
of theproposal distribution, buttailoringitforapartic-
ularclassof problemcancertainlyimproveefficiency.
For robustnessinapplications, thestandarddeviation
of theproposal distributionfor eachrandomvariable
is set equal to that of theconditional samples of the
current simulationlevel.
The Subset Simulation procedure for adap-
tively generating samples of O conditional on
{Y(O)>y
i
: i =1,,m} corresponding to specified
targetprobabilities{P(Y(O)>y
i
)=p
i
0
, i =1,,m}is
illustrated schematically in Figure 1. First, N sam-
ples{O
0,k
: k =1,,N}aresimulatedbydirectMonte
Carlo simulation (MCS), i.e., they are i.i.d. as the
original PDF. The subscript 0 here denotes that
thesamples correspond to conditional level 0 (i.e.,
unconditional). Thecorrespondingvaluesof thetrad-
able variable {Y
0,k
: k =1,,N} are then computed.
Thevalueof y
1
is chosenas the(1p
0
) N-thvalue
intheascendinglist of {Y
0,k
: k =1,,N}, sothat the
sampleestimateforP(F
1
)=P(Y >y
1
) isalwaysequal
top
0
.
Due to the choice of y
1
, there are p
0
N sam-
plesamong{O
0,k
: k =1,,N}whoseresponseY lies
in F
1
={Y >y
1
}. These are samples at conditional
level 1 andareconditional onF
1
. Startingfromeach
of thesesamples, MCMC isusedtosimulateanaddi-
tional (1p
0
) N conditional samplessothat thereis
atotal of N conditional samplesatconditional level 1.
Thevalueof y
2
is then chosen as the(1p
0
) N-th
value in the ascending list of {Y
1,k
: k =1, . . . , N},
and it defines F
2
={Y >y
2
}. Note that the sample
estimatefor P(F
2
|F
1
)=P(Y >y
2
|Y >y
1
) isautomat-
icallyequal top
0
.Again, therearep
0
N sampleslying
inF
2
.Theyaresamplesconditional onF
2
andprovide
seeds for applyingMCMC tosimulateanadditional
(1p
0
) N conditional samplessothatthereisatotal
of N conditional samplesat conditional level 2.
This procedure is repeated for higher conditional
levels until thesamples at conditional level (m1)
have been generated to yield y
m
as the (1p
0
)
N-th value in the ascending list of {Y
m1,k
:
k =1,,N} and that y
m
>y so that there are
276
enough samples for estimating P(Y >y). Note
that the total number of samples is equal to
N +(m1) (1p
0
) N.Approximateformulashave
beenderivedforassessingthestatistical error(interms
of coefficientof variation) thatcanbeestimatedusing
samplesgeneratedinasinglerun.
TheSubset Simulationalgorithmhasbeenapplied
to a variety of complex systems in structural (Au
and Beck 2003), aerospace(Thunnissen et al. 2007)
and fire (Au et al. 2007) engineering. Probabilistic
sensitivityandfailureanalysishavealsobeencarried
out usingSubset Simulation(Au2004, AuandBeck
2003).
3 SIMULATIONTOOLSINEXCEL
SPREADSHEET
A packageof EXCEL worksheetsandfunctions/Add-
In are developed, with the aids of Visual Basic for
Applications(VBA)inEXCEL,toimplementtheSub-
setSimulationalgorithminaspreadsheetenvironment
and apply it reliability analysis of slope stability. A
softwarearchitectureis proposedthat clearly divides
the package into three parts: 1) deterministic analy-
sis of slope stability, 2) modeling of uncertainty in
slopestability problem, and 3) uncertainty propaga-
tion by Subset Simulation. It is of particular interest
to decoupletheprocess of developing thedetermin-
istic slopestability analysisworksheetsandtheVBA
functions/Add-Infor uncertaintymodelinganduncer-
tainty propagation (Subset Simulation) so that the
work of uncertainty modeling and propagation can
proceed as an extension of deterministic analysis in
anon-intrusivemanner. Thedeterministicanalysisof
slope stability and uncertainty modeling and propa-
gationcanbeperformedseparatelybypersonnel with
differentexpertiseandinaparallel fashion.Therefore,
minimuminformationisrequiredfromtheengineers
regardingthereliabilitycomputational algorithm.
3.1 Deterministic slope stability analysis
worksheets
Deterministic analysis of slope stability is the pro-
cessof calculatingthefactor of safety FS for agiven
nominal set of values of the systemparameters .
The systemparameters generally include geometry
and stratigraphy of the slope, soil properties (e.g.,
soil unit weight, undrained shear strength, friction
angle, cohesion, and pore water pressure) and other
relevant information. Limit equilibriumanalysispro-
cedures (e.g., Ordinary Methodof Slices, Simplified
Bishop, SimplifiedJ anbu, Spencer, Morgensternand
Price, ChenandMorgensternprocedures) areimple-
mentedinaseries of worksheets andVBA functions
fortheFScalculation.Thedeterministicanalysisusing
limit equilibriumprocedures are organized into one
or a set of worksheets, although for discussion pur-
poses it is referred as a single worksheet. Froman
input-output perspective, the deterministic analysis
worksheet takes a given as input, performs limit
equilibriumcalculations and then returns the value
of FS as the output. No probability/reliability con-
cept is involved in the deterministic worksheet and
soitcanbedevelopedbypersonnel withoutreliability
background.
ToallowaseamlessintegrationwithSubsetSimula-
tion, thedeterministicanalysisworksheet isspecially
designedto befully automatedanddoes not involve
any human intervention. This is necessary for auto-
matedcalculationof theFSduringSubsetSimulation.
E.g.,if calculatingtheresponserequiresclickingabut-
ton, thensuchbuttonwill needtobeclickedasmany
timesasthenumberof samplesusedinthesimulation,
whichcouldbeintheorder of athousandandis not
acceptableinthesimulation.
3.2 Modeling of uncertainty in slope stability
problem
Uncertaintyinslopestabilityanalysisarisesfromthe
systemparameters, suchassoil properties(e.g., soil
unit weight, undrainedshear strength, frictionangle,
cohesion, and porewater pressure). Therefore, these
soil properties aretreatedas randomvariables inthe
analysis, althoughdifferent limit equilibriumanalysis
procedures may have slightly different sets of ran-
domvariables. Note that this paper focuses on the
uncertainties arising from soil properties and does
not account for other uncertainties, such as calcu-
lation model uncertainties. The spatial variability of
soil propertieswithinagivenlayer of soil ismodeled
byhomogeneousrandomfieldswithanexponentially
decayingcorrelationstructure. Anuncertainty model
spreadsheetisdevelopedforgeneratingarandomsam-
ple(realization)of theuncertainparameters.Starting
withuniformrandomnumberssupportedbyEXCEL,
transformation is performed to produce the random
samples of desired distribution. AvailableVBA sub-
routinesinEXCEL isusedtofacilitatetheuncertainty
modeling. Theuncertaintymodel worksheet isdevel-
oped in parallel with development of deterministic
analysisworksheet. Fromaninput-outputperspective,
theuncertaintymodelingworksheettakesnoinputbut
returnsarandomsampleof asitsoutputwhenever a
re-calculationiscommanded(e.g., bypressingF9 in
EXCEL).
Similartothedeterministicanalysisworksheets,the
uncertainty model worksheet isspecially designedto
befully automated and does not involveany human
intervention. In addition, the uncertainty modeling
is implemented in asingleworksheet for theconve-
nienceof Subset Simulation. TheSubset Simulation
VBA codeinstructs EXCEL to re-calculateonly the
uncertaintymodel worksheettogenerateasampleof ,
avoidingre-calculationof deterministicanalysiswork-
sheets which is not needed and which is often most
time-consuming.
3.3 Uncertainty propagation by subset
simulation
Afterthedeterministicanalysisanduncertaintymodel
worksheets are developed, they are linked together
277
Figure2. Schematicdiagramof linkbetweendeterministicanduncertaintymodelingworksheet.
Figure3. Subset SimulationAdd-In.
through their input/output cells to produce a proba-
bilistic analysismodel of theslopestability problem.
AsillustratedbyFigure2, linkingsimplyinvolvesset-
tingthecell referencefor thenominal values of in
thedeterministicanalysisworksheettobethecell ref-
erencefortherandomsampleintheuncertaintymodel
worksheet. After thistask, thevalueof showninthe
deterministicanalysisworksheetisequal tothatgener-
atedintheuncertaintymodelingworksheet, andsothe
FSvaluecalculatedinthedeterministicanalysiswork-
sheetisrandom. E.g., pressingtheF9 keyinEXCEL
generates a randomvalue of the FS. In other words
at this stage one can performa direct Monte Carlo
simulationof theproblemby repeatedly pressingthe
F9 key.
When the deterministic analysis and uncertainty
model worksheetsarecompletedoneisreadytomake
useof theSubsetSimulationalgorithmforuncertainty
propagation that can providebetter resolution at the
distribution tail (i.e., low failure probability levels).
A VBA codefor Subset Simulationisdevelopedthat
functions as anAdd-IninEXCEL andcanbecalled
by selectingfromthemainmenuTools followedby
SubSim. A user formappears uponinvokingof the
function, asshowninFigure3. Theuser shouldinput
the cell reference of the uncertain parameters and
Figure4. Slopestabilityexample.
system response Y =1/FS in the uncertainty mod-
eling worksheet. Other inputs include the following
algorithm-specificparameters: p
0
(conditional proba-
bilityfromonelevel tothenext),N (numberof samples
per level), m (number of levelstoperform).Asabasic
output, theprogramproducescomplementaryCDF of
the driving variable versus the threshold level, i.e.,
plot of estimatefor P(Y >y) versusy. Ingeneral the
CDF, histogramor their conditional counterparts can
beproduced.
4 ILLUSTRATIVE EXAMPLE
As an illustration, the worksheets and VBA
functions/Add-In are applied to assess reliability of
theslopeshowninFigure4. Thefactor of safety FS
isdefinedasthecritical (minimum) ratioof resisting
moment to theoverturning moment, and it is calcu-
latedusingtheSwedishCirclemethodinconjunction
withgeneral procedureof slices(DuncanandWright
2005). Theslipsurfaceisassumedtobeacircular arc
centered at coordinate (x, y) and with radius r. The
overturningandresistingmomentsaresummedabout
thecenter of thecircletocalculatethefactor of safety,
asshowninFigure4. Formomentcalculationsthesoil
massabovetheslipsurfaceissubdividedinto24verti-
cal slices, eachof whichhasaweight W
i
, circular slip
segmentlengthLl
i
, undrainedshearstrengthS
ui
along
theslipsegment, andanangle
i
betweenthebaseof
278
thesliceandthehorizontal.Thefactorof safetyisthen
givenby
wheretheminimumistakenover all possiblechoices
of slipcircles, i.e., all possiblechoicesof (x, y) andr.
A VBA codehasbeenwrittentocalculatetheratioof
resistant to overturning moment for different values
of (x, y) and r and then pick the minimumvalue as
thefactor of safety. Asareference, thenominal value
of FS that correspondtothecasewhereall S
u
values
equal totheir nominal valueof 20kPa, isequal to1.2.
Undrainedshearstrengthof soil istreatedasrandom
variable, andtheundrainedshear strengthatlocations
with the same depth is assumed to be fully corre-
lated. The spatial variability with depth is modeled
by ahomogeneous Lognormal randomfield with an
exponentiallydecayingcorrelationstructure. Thecor-
relationstructureisdescribedthroughthelogarithmof
theundrainedshear strength. That is, let S
u
(z) bethe
valueof undrainedshear strengthat depthz. Thenthe
correlationbetweenlogS
u
(z
i
)andlogS
u
(z
j
)isgivenby
R
ij
=exp(2|z
i
z
j
|,), where istheeffectivecor-
relationlength.Thevaluesof S
u
atdifferentdepthsare
simulatedthroughtransformationof i.i.d.uniformran-
domvariables. Thegenerationof thelatter isprovided
by thebuilt-in function Rand() in EXCEL. Specif-
ically, let S =[S
u
(z
1
), S
u
(z
2
), . . . , S
u
(z
n
)]
T
beavector
of S
u
valuesat depthsz
1
, . . . , z
n
. Then
where u and s are Lognormal parameters equal to
the mean and standard deviation of logS
u
(z); 1 is
a column vector of length n with all entries equal
to1; Z isann-dimensional standardGaussianvector;
L isalower triangular matrixobtainedfromCholesky
factorization of thecorrelation matrix R=[R
ij
] such
that LL
T
=R. Note that each component of Z can
begeneratedusingtheinverseof thestandardGaus-
siancumulativedistributionfunction(NORMINV in
EXCEL)of auniformrandomvariable.Acombination
of u =2.976ands =0.198isadoptedinthisexample
so that themean and spatial variability areapproxi-
matelyequal to20kPaand4kPa(i.e., 20%coefficient
of variation), respectively. The effective correlation
length is assumed to be2m. Thesoil between the
upper groundsurfaceand15mbelowis dividedinto
30equal layers. Inthecontext of Subset Simulation,
theset of uncertainparameters is O=Z, whichcon-
tains30i.i.d. Gaussianrandomvariablesthatareused
for characterizingthespatially varyingrandomfield
for S
u
.
Bydefault, SubsetSimulationdrivesthesamplesto
theupper tail of thedistributionof theresponseY. As
thelower tail of thedistributionof thefactor of safety
(i.e., theunsafezone) isof interest, theresponseY is
definedas thereciprocal of FS, i.e., Y =1,FS =the
ratioof theoverturningtoresistant moment.
Figure5. Simulationresults.
Subset Simulation is performed in EXCEL with
the following parameters: p
0
=10%, N =200 sam-
ples per level, m=3 simulation levels. This means
that theresults for P(Y >y) shall beproduced from
aprobabilitylevel of 1downto0.001.
Figure 5 shows a typical estimate of the comple-
mentary CDF for Y =1,FS, i.e., P(Y >y) versus y,
estimated by Subset Simulation with a total of
200+180+180=560 samples (i.e., calculations of
FS). For comparison the estimate by direct Monte
Carlo withthesamenumber of samples is also plot-
ted. It is seen that the CDF curve by direct Monte
Carloisnotaccurateatlowprobabilitylevels,whilethe
SubsetSimulationestimateprovidesconsistentresults
eveninthelowprobabilitylevel regime.Theobserved
computational efficiencyof SubsetSimulationfor the
slope stability problemis typical of those structural
reliabilityproblems(Auet al. 2007).
As mentionedbefore, thenominal factor of safety
(i.e., ignoring soil uncertainty) is equal to 1.2. For
this configuration in the presence of soil uncer-
tainty described the failure probability P(FS -1)=
P(Y >1) isabout8%.Thiscorrespondstoaneffective
reliability index of +
1
(92%)=1. 41. Notethat this
valuedependsonthelevel of soil uncertaintyassumed.
E.g., hadalarger spatial variabilitybeenassumed, the
failureprobabilitywill belargerandthecorresponding
effective reliability index will be smaller. Neverthe-
lessthenominal factor of safety remainsthesameas
it does not address uncertainty. The actual value of
spatial variabilityusedshouldof coursebeconsistent
withthesitecharacteristics.
5 CONCLUDINGREMARKS
This paper developed a package of EXCEL work-
sheets and VBA functions/Add-In to implement an
advanced Monte Carlo method called Subset Simu-
lation in the EXCEL spreadsheet and illustrated its
applicationtoreliability analysis of slopestability. A
softwarearchitectureis proposedthat clearly divides
thepackageintothreeparts: 1) deterministicanalysis
279
of slopestability, 2) modelingof uncertaintyinslope
stability problem, and 3) uncertainty propagation by
Subset Simulation. The process of developing the
deterministic slopestability analysis worksheets and
theVBA functions/Add-In for uncertainty modeling
and uncertainty propagation (Subset Simulation) are
deliberatelydecoupledsothattheworkof uncertainty
modelingandpropagationcanproceedasanextension
of deterministic analysis in a non-intrusive manner.
Thedeterministicanalysisof slopestabilityanduncer-
tainty modeling and propagation can be performed
separately by personnel with different expertise and
inaparallel fashion.Therefore, minimuminformation
is requiredfromtheengineers regardingthereliabil-
itycomputational algorithm. Theillustrativeexample
demonstratedtheapplicationof Subset Simulationto
slopestabilityreliabilityanalysisthat involvesalarge
number of randomvariablescharacterizingaspatially
varyingrandomfieldandhighlightedcomputational
efficiencyof Subset Simulationfor theslopestability
problem.
ACKNOWLEDGEMENTS
General ResearchFund[Project No. 9041327(CityU
110108)] andCompetitiveEarmarkedResearchGrant
[Project No. 9041260 (CityU 121307)] from the
Research Grants Council of theHong Kong Special
AdministrativeRegion, China. Thefinancial supports
aregratefullyacknowledged.
REFERENCES
Au, S. K. 2004. Probabilistic failure analysis by impor-
tance sampling Markov chain simulation. Journal of
Au, S. K., and Beck, J. L. 2001. Estimation of small fail-
ureprobabilitiesinhighdimensionsbysubsetsimulation.
Probabilistic Engineering Mechanics, 16(4): 263277.
Au, S. K., and Beck, J. L. 2003. Subset simulation and its
applications to seismic risk based on dynamic analysis.
Journal of Engineering Mechanics, 129(8): 117.
Au,S.K.,Ching,J.andBeck,J.L.2007.Applicationof Subset
Simulation methods to reliability benchmark problems.
Structural Safety, 29(3): 183193.
Au, S. K., Wang, Z.W. and Lo, S. M. 2007. Compart-
mentfireriskanalysisbyadvancedMonteCarlomethod.
Engineering Structures, 29(9): 23812390.
Duncan,J.M.andWright,S.G.2005.Soil Strength and Slope
Stability, J ohnWiley& Sons. Inc. NewJ ersey, 2005.
Low, B. K. 2003. Practical probabilistic slope stability
analysis. Proceedings of Soil and Rock America, MIT,
Cambridge, MA, J une 2003, Verlag Gluckauf GmbH
Essen, Germany, Vol. 2, 27772784.
Low, B. K. 2005a. Reliability-baseddesignappliedtoretain-
ingwalls. Geotechnique, 55(1): 6375.
Low, B. K. 2005b. Probabilistic design of anchored sheet
pilewall. Proceedings of 16th International Conference
on Soil Mechanics and Geotechnical Engineering, 1216
Septermber 2005, Osaka, J apan, Millpress, 28252828.
Low, B. K. 2008. Practical reliabilityapproachusingspread-
sheet.Reliability-basedDesigninGeotechnical Engineer-
ing, EditedbyPhoon, K. K.Taylor&Francis, Londonand
NewYork.
Low, B. K., Gilbert, R. B., andWright, S. G. 1998. Slopereli-
abilityanalysisusinggeneralizedmethodof slices. Jour-
nal of Geotechnical and Geoenvironmental Engineering,
ASCE, 124(4): 350362.
Low, B. K. and Tang, W. H. 1997. Efficient reliability
evaluation using spreadsheet. Journal of Engineering
Mechanics, 127(7): 149152.
Low, B. K. and Tang, W. H. 2007. Efficient spreadsheet
algorithmfor first-order reliability method. Journal of
Metropolis,N.,Rosenbluth,A.,Rosenbluth,M.,andTeller,A.
1953. Equations of statecalculations by fast computing
machines, Journal of Chemical Physics, 21(6): 1087
1092.
Roberts, C. and Casella, G. 1999. Monte Carlo Statistical
Methods, Springer.
Thunnissen, D. P., Au, S. K., and Tsuyuki, G. T. 2007.
Uncertainty quantification in estimating critical space-
craft component temperatures. AIAA Journal of Thermal
Physics and Heat Transfer, 21(2): 422430.
280
Optimal movingwindowwidthinconjunctionwithintraclasscorrelation
coefficient for identificationof soil layer boundaries
J.K. Lim& S.F. Ng
Universiti Teknologi MARA Pulau Pinang, Pulau Pinang, Malaysia
M.R. Selamat & E.K.H. Goh
Universiti Sains Malaysia, Pulau Pinang, Malaysia
ABSTRACT: Theidentificationof layer boundariesanddemarcatingthesoil profileintohomogeneouslayers
areoftenmuchmorecomplicatedthanoneexpectedwhendealingwithhighlyvariablecomplexnatural material.
Thequantitativeapproachesreportedingeotechnical literaturesarelimited, variesandmostlyrestrictedtocase
or project specificbasis. Inthisstudy, theperformanceof intraclasscorrelationcoefficient (RI) inconjunction
withvarioussuggestedwindowwidthsareinvestigatedusingthreefairlydifferentCPT soundingsobtainedfrom
thedatabaseof National Geotechnical Experimental Sites. RI appearstobeapowerful, robustandpersistenttool
andthecorrespondingoptimal windowwidthwasprovenasafunctionof averagedistancebetweenboundaries
whichcouldbedeterminedfromautocorrelationanalysis. Theempirical criterionof 0.7was founduseful in
guidingtheresearcher todecidewhether apeakissignificant enoughtobeconsideredasavalidboundary.
1 INTRODUCTION
1.1 Background of research
A major uncertainty in geotechnical engineering is
the inherent spatial variability of soil properties.
Theimportanceof recognizinguncertaintiesandtak-
ing theminto account in geotechnical design have
been propagated by numerous leaders since 1960s
(Casagrande 1965; Peck 1969; Wu 1974; Leonards
1982; Tang1984; Morgenstern1995; Whitman2000;
Christian 2004). Probability theory and statistics
provide a formal, scientific and quantitative basis
in assessing risk and uncertainties and have been
sprouted in geotechnical engineering research in
recent years. Inlinewiththedevelopment, character-
izationof soil properties has beenadvancedontothe
functions of thedeterministic mean and its stochas-
tic characters, comprises of coefficient of variation
and scale of fluctuation in modeling the inherent
soil variability as a randomfield (Vanmarcke1977;
DeGroot andBaecher 1993; J aksaet al. 1997; Phoon
andKulhawy1999; Fenton1999).
Compliancetostationarityor statistical homogene-
ity criterion is imperative in any soil data analyses.
Arandomfunctionwhichusedtomodeledthevariabil-
ityof soil isconsideredstationary,orweaklystationary
if (Brockwell and Davis 2002): (1) the mean of the
function is constant with distance, i.e. no trend in
thedata, (2) thevarianceof thefunction is constant
withdistance, i.e. homoscedastic, (3) therearenosea-
sonal variations, (4) therearenoapparent changesin
behaviors, and(5) therearenooutlyingobservations.
In other words, a stationary series is essentially a
functionof their separationdistance, rather thantheir
absolutelocations.
In geotechnical characterization undertaking, the
first step usually involves demarcating the soil pro-
fileintolayersor sectionswhicharehomogeneousso
that the result of subsequent analysis is not biased.
A homogeneous layer comprises uniformsoil mate-
rial that has undergone similar geologic history and
possesseswithcertaindistinctivebehaviors.Theiden-
tificationof boundariesandthusdemarcationprocess
areoftenmuchmorecomplicatedthanoneexpected
whendealingwiththishighlyvariablecomplexnatu-
ral material.Thevariabilityexistsnotonlyfromsiteto
siteandstratumtostratum, butevenwithinapparently
homogeneous deposits at a single site (Baecher and
Christian2003).
1.2 Problem statement
It would berather useful to supplement theexisting
procedures with a quantitative systematic approach.
The conventional method, which is based on visual
observation, gives less accuracy andsubstantial sub-
jectivity to the identification of actual boundary of
soil. Existingstatistical toolsarenot widelyexplored,
well-calibrated and properly defined, thus generally
result in unsatisfactory outcome. This paper intends
to resolvetheaboveproblemfor better characteriza-
tionof soil properties. Reporteduseful statistical tools
would be compared in terms of their effectiveness
andtheexistingprocedures arerevampedfor further
improvement.
281
Cone Penetration Test (CPT) is widely used in
soil characterization in view of its ability to pro-
videalmost continuousprofile, widelycorrelatedand
highlyrepeatable(Robertson1986; NCHRP2007). In
this study, theCPT soundings, inparticular thecone
tipresistance, q
c
, wereusedfor detail illustration. The
datawereselectedonthebasisof their spacing, exten-
sivenessanddifferencebetweeneachotherwhichbest
yieldthethoroughexaminationontheperformanceof
statistical toolsinsoil boundarydemarcation.
2 STATISTICAL APPROACHES
2.1 General
Classical andadvancestatistical approachesfortesting
similarityor dissimilarityof univariateor evenmulti-
variate records are believed to be substantial. Some
of theestablishedanalytical toolshavepotential tobe
appliedinthefieldof geotechnical engineeringwith
modificationtosuitthenatureof geotechnical parame-
ters.Nevertheless,thecross-disciplinarycollaboration
undertaking is surprisingly little. Many geotechnical
engineersareunfamiliar withtheunderlyingconcepts
of statisticsandprobability andremainskeptical and
reluctant evento makeanattempt. Inthis paper, two
statistical methods whicharerelatively commonand
simple in identifying the soil layer boundaries are
presented.
2.2 Intraclass correlation coefficient
Intraclass Correlation Coefficient (RI) was reported
byCampanellaandWickreseminghe(1991) asause-
ful statistical method for detecting soil layer bound-
ariesusingCPT soundings. For identificationof layer
boundaries, amovingwindowwidth, W
d
, isfirstspec-
ified and the window is divided into two segments.
TheRI profileis thengeneratedby movingtwocon-
tiguoussegmentsover ameasurement profileandthe
computedindex is plottedcorrespondingto themid-
point of thewindow. RI will always liebetweenzero
andunityandarelativelyhighvalueof RI islikelyto
indicatethepresenceof alayer boundary.
The RI together with its pooled combined vari-
ance (s
2
w
) and the between class variance (s
2
b
) are
definedas:
wheren
1
andn
2
arethesamplesizeof twoequal seg-
ments, aboveandbelowthemiddlelineof thewindow,
s
2
1
ands
2
2
beingthevariancesof thesamplefor thetwo
segments, x ands
2
arethesamplemeanandvariance
withinthedesignatedwindow. Theequationcanalso
bewrittenasfollow(ZhangandTumay 1996) for the
two segments with equal samplesizeof m and their
samplemeanof x
1
and x
2
, respectively.
J udging whether an index value is high enough
to indicate a boundary in a relative sense by visual
observation is fairly subjective and would result in
inconsistency.ZhangandTumay(1996)suggestedthat
thepeak valueof RI whichis equal to or larger than
0.7 can be empirically determined as the boundary
line. Hegazy et al. (1996) proposedthecritical value
asthe(mean+1.65standarddeviation) withalevel of
significanceof 5%.However,Phoonetal.(2003)com-
mentedthattheabovecritical valuesarenotdepending
ontheunderlyingcorrelationstructureof theprofile.
2.3 Window width
As all thesestatistical methods incorporatewith the
conceptof movingwindows, thewidthof thesampling
windowbecomesanimportantparameterwhichcould
have substantial influence on the result of analysis.
Generally, too narrowawindowwill result in unde-
sirableeffect of highnoiselevel withtoomanypeaks
appear.Ontheotherhand,toowideawindowwill over-
smoothen the statistics till missing out the possible
boundariesduetoexcessiveperturbationregion.
Webster (1973) proposed a method to determine
the boundaries on transects automatically and has
foundthat thesuitablewidthfor thecalculationwin-
dowis approximately two thirds of theexpecteddis-
tancebetweenboundarieswherethespacingbetween
boundaries does not differ widely. Theexpecteddis-
tance or average spacing between boundaries could
bedeterminedfroman autocorrelation analysis. The
techniqueissaidreasonably sensitivebut foundlittle
affectedbywindowwidth.
Campanella and Wickremesinghe (1991) elabo-
ratedindetail thestatistical methodsfordetermination
of window width and recommended that it is rather
to adopt an incorrect narrower window width than
wider windowwidthtoavoidmissingoutthepossible
layerboundaries.Twocasestudies, namelyMcDonald
farmSiteandHaneySitehavebeenillustratedandthe
windowwidthsselectedwere1.5mand2.0m, respec-
tively.Totheother extent, windowwidththatlessthan
1.0mshouldnotbeselectedduetonormal distribution
restrictiononthesamples(Wickremesinghe1989).
Zhang andTumay (1996) based on thefinding of
previous research that the standard 10cm
2
electric
cones may require minimumstiff layer thickness of
36cmto 72cmto ensurefull tipresistanceandcon-
cluded that thevalueof thewindow width could be
conservativelytakentobe150cmor 75cmfor half of
thewindow. Nevertheless, they reportedthat primary
282
layeringusually does not providesatisfactory results
duetounevensoil layers. Thebigdifferenceof layer
thicknesseswill resultintoomanylayersinthickzone
and too littlein thethin causing a bias in making a
judgment.
Cafaro and Cherubini (2002) used the same pro-
cedure as proposed by Webster (1973) in analyzing
astiff overconsolidated clay at atest siteinTaranto,
Southern Italy and obtained a fairly wide window
width of 6.8mfor q
c
, f
s
and I
c
profiles. The width
wasreducedto4.8mandthevariationintheRI pro-
filewasfoundnegligibleonboththeposition(depth)
andthevalueof thepeaks.Thegeostatistical boundary
wasfoundnotalwayscorrespondwell tothegeolitho-
logical boundarythussuggestedthepossibleoffset to
beaccountedfor.
Kulatilakeand Um(2003) introduced anew pro-
cedure to detect statistical homogeneous layers in a
soil profile. Inexaminingtheconetipresistancedata
for theclay siteat TexasA&M University, awindow
widthof 0.4mwhichcontains10datapointsineach
sectionhasbeenused.Duetotheshortsectionadopted,
four possiblecombinationsfor themeansoil property
(either constant or withlinear trend) wereconsidered.
The distance between the lower and upper sections
was calculatedandsubsequently generatedalongthe
depthfor evaluationof thestatistical homogeneity at
different levels.
Phoonet al. (2003) adoptedthelower limit of per-
missiblewindowwidth, thatis1.0mingeneratingthe
RI and Bstat profiles. Both profiles manageto cap-
turetheprimary layer boundaries in consistent with
visual inspection, of whichtheBstatpeaksweremuch
moreprominent. Considerablenoises wereobserved
and 3 fault boundaries were identified (no obvious
soil boundaries can be seen in the q
t
record) in the
RI profilewhencomparingtothecritical valueof 0.7.
2.4 Autocorrelation analysis
TheWebstersintuitionthatthesuitablewindowwidth
shouldbeequal toor somewhat lessthantheaverage
distancebetween boundaries has led to theexploita-
tion of autocorrelation analysis (Webster 1973). The
result of his study showedthat theoptimizewidthis
around two thirds of theexpected distancealthough
larger width up to the full expected distance could
still beuseful (mainpeaksappearedinthesameposi-
tionsbutwithdifferentrelativeheights) for thosearea
withmarkedchanges. Theautocorrelationcoefficient
at lagk wasexpressedas
wheren isthenumberof samplingpointsintheseries,
k isthelagandu isthedeviationfromtheseriesmean
at thetth(or t +kth) point.
In the correlogram, i.e. the plot of autocorrela-
tioncoefficient, r
k
against lag, k, theautocorrelation
coefficient will decrease more or less steadily with
increasinglagdistancefromaround1to somemini-
mumvaluenear zeroandfluctuatethereafter. Thelag
distanceover whichthisdecayoccurscanbetakenas
theaveragedistancebetweenboundarieswhichcould
beused as guidancein selection of suitablewindow
width.
3 NUMERICAL EXPERIMENTS
3.1 Selected case studies
For the present study, established database for
National Geotechnical Experimental Sites (NGES)
(http://www.unh.edu/nges/) funded by the Federal
Highway Administration (FHWA) and National Sci-
enceFoundation(NSF)ofAmericawereexplored.The
performanceandusefulnessof several approaches(in
terms of window width and the statistical tool) that
havebeen used by other geotechnical researchers in
identifying layer boundaries were thoroughly exam-
ined. Typical CPT profilesfromthreesitesrepresent-
ing different parts of North America were selected.
These sites have been classified as Level I and II
sitesthat aremost closelyfit thecombinedcriteriaof
research areas as of significant national importance.
TheseCPT soundings wereclosely spaced, extensive
andfairlydifferentbetweeneachother whichbestsuit
for thisexamination. Theselectedsitesare
1) TreasureIslandNaval StationintheSanFrancisco
Bayarea(CATIFS),
2) University of Massachusetts, Amherst campus
(MAUMASSA), and
3) NorthwesternUniversityLakeFill SiteinEvanston
(ILNWULAK).
3.2 Examination of Existing Approaches
Intraclass correlation coefficient (RI) which was
reported to be useful in geotechnical literatures are
examined here. Since the method is used in con-
junctionwithmovingwindowaveragingconcept, the
sensitivityof thewindowwidthingeneratingthemost
optimizeprofileswhichwoulddiscriminatethetrue
boundaries is of great concern. Several criteria in
determiningsuitablewindowwidthdeducedfrompre-
viousresearchers worksareincorporatedinthisstudy
asfollows:
i) Twothirdsof theaveragedistancebetweenbound-
ariesdeterminedbyusingautocorrelationanalysis
(Webster 1973; CampanellaandWickremesinghe
1991),
ii) Theconservativeassumptionforfull tipresistance
of 1.5m(ZhangandTumay1996), and
iii) Theminimumwidthof 1.0mduetonormal dis-
tributionrestrictiononsamples(Wickremesinghe
1989; Phoonet al. 2003).
In reality, perfect result is almost not possible as
theactual soil datacould bereally erratic. However,
the analytical approach (combination of a statistical
283
Figure1. Identificationof soil layer boundariesusingRI of
varyingW
d
for CATIFS site: W
d1
=1.0m, W
d2
=1.5mand
W
d3
=1.9m.
tool withanoptimizewindowwidth) couldbedeemed
satisfactory fromat least two practical aspects. The
approachshouldavoidmissingoutpossibleprominent
layer boundariesandat thesametime, abletocapture
asmanymajorboundariesaspossibleatonetime.And
it is evident frompast literatures that theboundaries
indicated by these statistical tools are often slightly
offsetprobablyduetovariationintheupper andlower
segmentsasmovingthesamplingwindowoverthesoil
profile. Therefore, notethat thetoolscouldserveasa
useful indicator, butyetfinal adjustmentanddecision
haveto bemadewithregards to theoriginal profile,
geological backgroundandnot forgottenengineering
judgment.
3.3 Results of analysis
Theresultsof analysisforCATIFS, MAUMASSA and
ILNWULAK sitesarepresentedinFig. 1, Fig. 2and
Fig. 3, respectively. For eachset of results, 3different
windowwidthsasdelineatedabove(i, ii andiii) have
beenincorporatedwithRI andpresentedside-by-side
for comparison.
varyingW
d
forMAUMASSAsite:W
d1
=1.0m,W
d2
=1.5m
andW
d3
=2.4m.
Case 1: CATIFS Site
FromtheconetipresistanceprofileinFig. 1(CATIFS
site), theheterogeneity of soil from7.0mto9.0mas
comparedtoothersisreadilyobservedthroughvisual
examination. TheRI profilemanagestocapturethese
peaks at both7.0mand9.0mlocations, andanother
one at approximately 2.1m. The above three main
peaks werefound persist for all thetested widths of
1.0m, 1.5mand1.9m(W
d1
toW
d3
) withquiteanum-
ber of noisesappear exceedingtheempirical valueof
0.7 for windowwidths of 1.0mand 1.5m(W
d1
and
W
d2
). Thus, inferencecouldbemadefromtheresults
thattheoptimal windowwidthshouldbearound1.9m
inthiscase.
Limitation of the approach on missing out the
informationat bothends of theprofileis noted. The
apparentboundaryof conetipresistanceat1.26mfor
instanceisbasicallyoutof thecoverageareaof thegen-
eratedoutput asthecomputedindexisplottedagainst
themidpoint of themovingwindow. Inaddition, the
identificationof apotential boundaryaroundthedepth
of 2.1musingRI profileindicatedthat thetool able
284
varyingW
d
for ILNWULAK site: W
d1
=1.0m, W
d2
=1.5m
andW
d3
=1.3m.
to detect aconsiderablesharpchangealongthepro-
filewhich suggested two quasi-linear portions to be
divided.
Case 2: MAUMASSA Site
MAUMASSA site (Fig. 2) was the second case
study where the cone tip resistance profile exhibits
apparent heteroscedasticity characteristic with grad-
ual changeof gradient aroundthepotential boundary.
As shown in Fig. 2, the generated RI profiles for
window widths of 1.0mand 1.5m(W
d1
and W
d2
)
arebasically somenoises inferringthat thewindows
are too narrow. As the window width increased to
2.4m (W
d3
), which is approximately two thirds of
the average distance between boundaries, the true
main peaks appeared, one at approximate depth of
4.4m and another one at 3.0m. Generated profile
at both ends tends to be less reliable as shown by
two erroneous peaks thus should not be considered.
In this case, 2.4m can be considered as the suit-
able width when intraclass correlation coefficient is
used.
Case 3: ILNWULAK Site
Results of the third case study of ILNWULAK site
arepresented in Fig. 3. Theconetip resistancepro-
fileseems to exhibit higher resistancevalues at both
ends, i.e. depths before 1.0mand after 7.0m, and
aninterbeddedheterogeneous layer at approximately
3.3mto 4.3m. RI profiles for all the tested widths
of 1.0m, 1.5mand1.3m(W
d1
to W
d3
) as presented
in Fig. 3 show a very good agreement where the
fourexpectedboundariesweremanagedtodetect.The
mainpeakspersist asthewindowwidthchangesand
more noises can be noticed at smaller widths par-
ticularly for window width of 1.0m(W
d1
). Similar
inference could still be reasonable drawn where the
suitablewidth for this boundaries demarcation exer-
ciserangesat approximately1.3m, obtainedfromthe
autocorrelationanalysis.
3.4 Discussion
Ingeneral, RI appears tobeapowerful tool as it can
capturemostof theprominentmajorboundariesatone
timefairlyaccurately. Besides, it isreasonablyrobust
whichcouldpersistently detect themainpeaksat the
samepositionsevenwithwindowwidthsthatarefairly
different fromtheoptimizeconfiguration. Websters
suggestion (1973) to determinethesuitablewindow
widthasafunctionof averagedistancebetweenbound-
ariesusingautocorrelationanalysiswasvalidated.The
difference for profiles generated using smaller win-
dowwidthsisthatmanyundesiredpeaksornoisesmay
appear,whereaslargerwidthstendtohiddentheneces-
saryboundaries.Theempirical criterionof 0.7(Zhang
and Tumay, 1996) to guide the worker in deciding
whether apeakissignificant enoughtobeconsidered
as avalidboundary is very useful. Thecriterionwas
foundperformingprettywell inmostcircumstancesas
illustratedthroughvarious distinctivecasestudies in
thispaper.
Observingtheresults of analysis, theauthors pre-
sumedthatthesestatistical toolsarelikelytowell per-
forminidentifyingboundariesat whicheachdivided
layer is constituted of linear trend. Nonetheless, the
presumption on this limitation does not restrict the
worker to combine two or more layers in the sub-
sequent analyses as far as they possess very similar
variation characteristics, i.e. scale of variance and
theautocovariancedistance(or scaleof fluctuation).
The modeling which reasonably simplifies the soil
profile into fewer layers within the same geological
formation and at thesametimeremains most of the
importantinformationisalwaysof greatinterestfrom
thepragmatic stand. Notethat statistical tools donot
appreciate explicit soil type classification nor engi-
neeringbehaviors, thus final evaluationanddecision
still liewithsoundengineeringjudgment.
Note that actual soil profile could be extremely
erratic and complex, thus adopting any statistical
method without incorporating engineering judgment
couldbeunsatisfactory. Oneof thedifficultiesasmen-
tionedbyWebster(1973)andZhangandTumay(1996)
285
was solving theprofilewith layer thicknesses differ
widely. To reducethecomplication, theworker must
first be clear on what he expects and properly plan
beforeonestarts theanalysis. For instance, if in the
first place, therough approximateaveragethickness
betweenlayersisbyvisual seemedtobeabout 3.0m,
then the optimize window width would probably be
around that valueor somewhat slightly smaller. Any
attempt that far too small or far too large fromthat
valuewouldbefruitless. Often, onetimedemarcation
might not beadequate; ontheother extent, excessive
subdivisionandmodelingwhichhasnopractical value
shouldbeavoided.
Inthecasewherebyahomogeneouslayerisof clear
evident, forinstancetheclaylayerfromapproximately
12.0mdepthtotheendof explorationof about30.0m
at theCATIFS site, that sectionof profileshouldnot
be mixed together with other relatively thinner lay-
ers at upper depth in theanalysis (notethat theclay
layer from12.0mto30.0mhadbeenexcludedinthe
first casestudyhere). Or else, anypossibleerroneous
peaks appear withinthat sectionshouldbediscarded
after incorporated with visual observation and engi-
neering judgment. For verification, the demarcated
sections shouldbeexaminedusingstationarity tests,
e.g. Kendalls test, runtest, signtest, etc. whichare
not covered in this paper. Every method has its own
limitationsandtheunderlyingconceptsmust bewell
understood in order to fully exploit it appropriately.
Pleasenotethatstatistical tool isbymeantoassistbut
not toconfuse.
4 CONCLUSIONS
Generally, statistical toolscanbeutilizedinidentify-
ing layer boundaries satisfactorily. RI appears to be
apowerful oneas it can capturemost of thepromi-
nent major boundaries at one time fairly accurately.
Also, it is robust and persistent which could detect
themainpeaksat thesamepositionsevenwithfairly
different window widths. Results fromthe analysis
carriedoutshowsthatthelower limitof 1.0mtendsto
createplentyof unnecessarynoiseswhichmightcom-
plicatetheinterpretation.Theresistancezoneof 1.5m
appears to be too restrictive and may only apply to
certainspecific soil profilehorizons. Theconclusion
obtainedfrompreviousresearchusing1.0mand1.5m
windowwidths arelikely to becoincident. Theaver-
agedistancebetweenboundariesfromautocorrelation
analysis seems to bemost relevant, flexibleanduse-
ful.Theempirical criterionof 0.7wasconsistentlywell
performedinguidingtheresearchertodecidewhether
apeakissignificantenoughtobeconsideredasavalid
boundary.
REFERENCES
Baecher,G.B.&Christian,J.T.2003.Reliability and statistics
ingeotechnical engineering.England:J ohnWiley&Sons.
Brockwell, P.J. & Davis, R.A. 2002. Introduction to time
series and forecasting. 2nd Ed., New York: Springer-
VerlagNewYork.
Cafaro, F. & Cherubini, C. 2002. Largesamplespacing in
evaluation of vertical strength variability of clayey soil.
Journal of Geotechnical andGeoenvironmental Engineer-
ing 128(7): 558568.
Casagrande, A. 1965. Role of the calculated risk in
earthwork and foundation engineering. Journal of Soil
Mechanics and Foundation Division 91(SM4): 140.
Christian, J.T. 2004. Geotechnical engineering reliability:
How well do we know what we are doing? Jour-
nal of Geotechnical and Geoenvironmental Engineering
130(10): 9851003.
DeGroot, D.J. &Baecher, G.B. 1993. Estimatingautocovari-
ance of in-situ soil properties. Journal of Geotechnical
Hegazy, Y.A., Mayne, P.W. & Rouhani, S. 1996. Geostatis-
tical assessment of spatial variability inpiezoconetests.
Uncertainty in the geologic environment: from theory to
practice (GSP 58): 254268, NewYork: ASCE.
J aksa, M.B., Brooker, P.I. & Kaggwa, W.S. 1997. Inaccu-
racies associated with estimating randommeasurement
errors. Journal of Geotechnical and Geoenvironmental
Kulatilake, P.H.S.W. &Um, J. 2003. Spatial variationof cone
tip resistance for the clay site at Texas A&M Univer-
sity. InProbabilistic Site Characterization at the National
Geotechnical Experimentation Sites, NewYork: ASCE.
Leonards, G.A. 1982. Investigation of failures. Journal of
Geotechnical Engineering 108(GT2):185246.
Morgenstern,N.R.1995.Managingriskingeotechnical engi-
neering. The3rdCasagrandeLecture. Proceedings of the
10th Pan-American Conference on Soil Mechanics and
Foundation Engineering Vol. 4: 102126, Guadalajara.
NCHRP Synthesis 368 2007. Cone Penetration Testing. A
Synthesis of Highway Practice. National Cooperative
Highway Research Program, Washington D.C: Trans-
portationResearchBoardof theNational Academies.
Peck, R.B. 1969. Advantages and limitations of the obser-
vational methodinappliedsoil mechanics. Geotechnique
19(2): 171187.
Phoon, K.K. & Kulhawy, F.H. 1999. Characterization of
geotechnical variability. Canadian Geotechnical Journal
36: 612624.
Phoon, K.K., Quek, S.T. &An, P. 2003. Identificationof sta-
tisticallyhomogeneoussoil layersusingmodifiedBartlett
statistics. Journal of Geotechnical and Geoenvironmental
Robertson, P.K. 1986. In-situ testing and its application to
foundationengineering. Canadian Geotechnical Journal
23(3): 573587.
Vanmarcke, E.H. 1977. Probabilistic modeling of soil pro-
files. Journal of the Geotechnical Engineering Division
103(GT11): 12271246.
Webster, R. 1973. Automatic soil boundary location from
transect data. Mathematical Geology 5(1): 2737.
Wickremesinghe, D. & Campanella, R.G. 1991. Statistical
methods for soil layer boundary locationusingthecone
penetrationtest. Proc., ICASP6: 636643, MexicoCity.
Wickremesinghe, D.S. 1989. Statistical characterization of
soil profile using in-situ tests. PhDthesis, Univ. of British
Columbia, Vancouver, Canada.
Wu, T.H. 1974. Uncertainty, safety and decision in soil
engineering. Journal of the Geotechnical Engineering
Division 100(GT3): 329348.
Zhang, Z.J. &Tumay, M.T. 1996. Thereliabilityof soil clas-
sificationderivedfromconepenetrationtest. Uncertainty
in the geologicenvironment: Fromtheory to practice (GSP
58): 383408, NewYork: ASCE.
286
Soil variabilitycalculatedfromCPT data
T. Oka& H. Tanaka
Graduate School of Engineering, Hokkaido University, Japan
ABSTRACT: Thedesignmethodbasedonreliabilityhasextensivelyusedforconstructionof civil engineering
structures, eveninfoundationdesign. Sincethegroundis not artificially but naturally created, thereliability
methodfor conventional structuressuchassteel or concretecannot bedirectlytothefoundationdesign. Thisis
becausevariabilityinsoil propertiesmaybedifferentfromthatinothercivil engineeringmaterials. Itisnecessary
toestablishdatabasefor showingvariability invariousgrounds. Inthispresent study, usingdatameasuredby
ConePenetrationtest (CPT), statistical analysesarecarriedout for selected11sites.
1 INTRODUCTION
Thedesignmethodbasedonreliabilitywasintroduced
andhas beengradually usedfor constructionof civil
engineering structures, including design of founda-
tions. However, there are a lot of problems in this
methodwhenapplyingittogeotechnical engineering.
For example, soil properties aredifferent fromthose
of other civil engineeringmaterials, suchasconcrete
and steel. Becausetheground is naturally deposited
except for compacted or improved soils, variability
in soil properties is completely different fromarti-
ficially manufacturedmaterials whoseproperties are
strictly controlled. Inadditionto this inherent differ-
ence, humanerrorsatmeasurementof soil parameters
shouldbetakenaccount. Sampledisturbanceisoneof
keyissuestoobtainreliablesoil parametersat labora-
torytest.Althoughinsitutestsdonotneedtoconsider
thesampledisturbance, measuredvaluesarestrongly
influenced by the drilling method and/or the testing
method. InStandardPenetrationTest(SPT), forexam-
ple, themethodfor droppingthehammer affects the
N value.
Cone Penetration Test (CPT) has been gradually
used even in J apan. The most advantage feature of
theCPT over other insitutestsmaybethat measured
values are nearly free fromhuman factors, because
it does not require a borehole and its testing proce-
dures arerelatively simple. Inaddition, theCPT can
obtain geotechnical information nearly continuously
to depth. In this study, using CPT data measured at
various11sitesinJ apanaswell asoverseas, variability
insoil parametersisexamined. Also, for comparison,
variabilityinN valueforsandygroundandtheuncon-
finedcompressivestrength(q
u
) for clayey groundis
investigated.
2 TESTEDSITESANDMETHODOF CPT
Statistical analyseswerecarriedoutfor11sites, whose
mainfeaturesandlocationareindicatedinTable1and
Fig. 1 (excluding overseas sites), respectively. From
HachirogatatoAmagasaki sitesinTable1, theground
consistsof normallyconsolidatedclayeysoil.Thebur-
denpressureat HachirogataandBusansitesdoesnot
somuchchange(except for fluctuationof theground
water table), whilesitesof SingaporeandAmagasaki
wererecently reclaimed, but consolidationduetothe
fillingisconsideredtobecompleted.
From Kagoshima to Tomakomai, investigated
grounds consist of granular materials. Material at
Kagoshima, NakasibetsuandTomakomai sitesisvol-
canic ash. At Kagoshima site, volcanic ash was
transported and deposited by rivers at Kagoshima
Figure1. TestingSitesinJ apan.
287
Table1. Featuresfor testedsitesandAnalyzedresultsfromCPT data.
Site Soil Type Depth(m) a(MPa/m) b(MPa/m)
1

2
(MPa) Reference
Hatchirogata Clay 1037 0.08 0.03 0.0067 2.89 0.02 Tanaka, 2006
Busan Clay sand 523 0.044 0.022 1.8 10.82 0.028 Tanakaet al., 2001a
nosand 523 0.041 0.022 0.000062 2.63 0.024
Singapore Clay 1830 0.33 0.037 0.00033 2.73 0.024 Tanakaet al., 2001b
Amagasaki1 Clay 1119 0.32 0.091 0.052 3.32 0.04
Amagasaki2 Clay 1119 0.3 0.088 0.066 2.59 0.043
Kagoshima VolcanicAsh 530 9.23 0.0096 0.0986 3.87 1.35
Yodogawa Sand 520 12.1 0.15 0.061 2.93 3.74 Mimura, 2003
Nakashibetsu VolcanicAsh 410 8.61 2.18 0.78 1.98 4.61
Kemigawa Sand 620 8.4 1.41 0.33 3.1 1.94 Tanaka, 1999
Higashiohgishima Filling(sandy) 520 4.4 0.092 2.16 5.18 1.92
Tsuruga Filling 416 3.2 0.3 0.0064 3.26 2.54
(CrushRock)
Tomakomai VolcanicAsh 1320 7.48 0.0037 1.23 4.05 0.85
site (secondary deposition), while at the site of
Nakashibetsu and Tomakomai, the ash was directly
deposited by wind at the eruption (primary deposi-
tion).YodogawaandKemigawasitesarelocatedinthe
riverside(gawa means river inJ apanese), andtheir
ground consists of mainly sandy material. Higash-
iougishimaandTsurugasitesareonareclaimedland
by sandandcrushedgravel, respectively. Moredetail
informationonsoil propertiesfor sitesareavailablein
literaturesindicatedinTable1.
CPT wasconductedfollowingthespecificationof
theinternational referencetestprocedureproposedby
theISSMFE technical committeeonpenetrationtest-
ing(1988): theapex angleof theconeis 10cm
2
(the
diameter is 35.7mm); the apex angle of the cone is
60
; thelocationof thefilter for measuringporewater

pressureistheshoulder behindtheconeandthespeed
of thepenetrationis2cm/s.Thepointresistance(q
t
) is
correctedbytheeffectiveareaandtakesintoaccount
porewater pressureactingonthefilter.
3 ANALYSESOF CPT DATA
3.1 A trend function due to increase in burden
pressure
q
t
measuredbyCPT maybebrokendownintoatrend
function[t(z
1
. . . z
n
)] andset of residualstothetrend
[(z
1
. . . z
n
)]: i.e., q
t
=q
t
(t)+q
t
().Asatypical exam-
ple of CPT result fromclayey soil, Fig. 2 shows q
t
distribution at thesiteof Busan. In this example, q
t
values clearly increases with depth. The q
t
can be
expressedbythefollowingequation:
where, N
kt
, S
u
and
vo
are the cone factor, the
undrainedshearstrengthandthetotal burdenpressure,
respectively.
vo
increases withdepthandfor normally consol-
idated ground, S
u
also increases with depth because
of increaseinconsolidationpressure. Therefore, it is
Figure2. (a) Measuredandtrendq
t
atBusansite. (b) Resid-
ualsat Busansite.
anticipated that q
t
linearly increases with depth (z),
especiallyfor thenormallyconsolidatedsoil layer. On
theother hand, for sandysoil, inanother word, where
the penetration of CPT is performed under drained
condition, itiswell knownthatq
t
doesnotincreaselin-
earlywith
vo
, but with
vo
0.5
. However, inthisstudy,
as a preliminary study, it is assumed that q
t
values
fromCPT haveatrendfunctionof thefollowinglinear
equation:
wherez isdepth.
Constants of a and b are calculated by the least
square regression method. The calculated trend line
at Busan siteis shown in Fig. 2. In this site, depths
for obtainingthetrendlinearerestrictedfrom5mto
23m. Fig. 3 shows observed q
t
values and thetrend
lineatthesiteof Kagoshima, whichiscoveredbythick
volcanicashthat wastransformedbyriver fromShi-
rasu terrace. Constantsof a andb for other areasare
288
Figure 3. (a) Measured and trend q
t
at Kagoshima site.
(b) Residualsat KagoshimaSite.
indicatedinTable1. Forclayeyground, constantof b is
relatedtoincrement of S
u
becauseunit weight of soil
(
t
) does not so much vary in sites. It is shown that
b for Amagasaki is clearly larger than that for other
sites. For granular ground, b constantsarecompletely
different among sites, and those in some sites indi-
catenegativevalues: seefor example, Kagoshima, as
showninTable1andFig. 3.
3.2 Residuals
Residuals (q
t
()) areadifferencebetween measured
q
t
andcalculatedfromthetrendlineq
t
(t) at thesame
depth. Distributionof q
t
() isshowninFigs2(b) and
3(b), for BusanandKagoshimasites, respectively. It
isthoughtthatq
t
() consistsof twocomponents: mea-
surement error andvariationcausedbyheterogeneity
of theobjectivelayer. It isinterestedtonotethat q
t
()
doesnotincreaseindepthbutnearlyconstantfor both
Busan(clay) andKagoshima(volcanicash).Thisindi-
catesthatq
t
() isnotinfluencedbymagnitudeof q
t
. In
another word, q
t
() is not suitabletotreat as normal-
izedsuchascoefficient of variation, q
t
(),q
t
(t), since
q
t
(),q
t
(t) decreasesindepth.
Standarddeviationof q
t
() is indicatedinTable1
(itssymbol is). Asexpected, for clayeygroundis
definitely smaller than that for granular ground, and
itsdifferenceisasmuchas100times.Thismeansthat
variationinq
t
issmallerthanthatforgranulargrounds,
inadditiontosmall q
t
itself for clayeyground.
For studyingpropertiesof q
t
(), itmaybeuseful to
knowwhat shapeof theq
t
() distributionis suitable.
If q
t
() isformedbymeasurementerrors, q
t
() should
followthenormal distribution. Inthis study, Pearson
chart(seeFig. 4) will beusedfor examinationof q
t
()
distribution. Pearsondevelopedanefficientsystemfor
theidentificationof suitableprobability distributions
based on third and forth moment statistics of adata
set, as showninFig. 4(Baecher & Christian). Inhis
chart,
1
forthehorizontal axisand
2
forvertical axis
Figure4. Pearsonchart fromCPT.
areplotted, and
1
and
2
aredefinedbythefollowing
equations.
where, C
sk
andC
ku
areskewnessandkurtosis, respec-
tively, andtheseequationsaregivenasfollows:
where, n,
i
, m
, s
are sample numbers, residuals,

samplemean, samplestandarddeviation, respectively.
m
, s
aregivenasfollows.
Thevaluesof
1
and
2
atvarioussitesareindicated
inTable 1. For clayey ground,
1
is relatively small
comparedwithgranular ground, except for Busan.
1
aswell as
2
attheBusansiteisextremelylarge, even
incomparisonof thosefor granular grounds. Therea-
son for large these values at the Busan site may be
attributedtotheexistenceof sandlayers. Asshownin
Fig. 2, atdepthsof 10and16m, q
t
suddenlyincreases.
Fromthedatashowingtheporewater pressure, it was
revealedthat sandlayers exist inthesedepths. When
q
t
dataatthesesandlayersareomitted, both
1
and
2
aredramatically changed, asshowninTable1. Espe-
cially
1
waschangedfrom1.8to0.000062. However,
the trend function (see a and b in the table) is not
changed (though a little change in a), but is also
slightlychangedonlyfrom0.028to0.024(MPa). The
distributionof q
t
() withsandlayer andwithout sand
289
Figure5. Histogramof residualsat BusanSite.
layerarecomparedinthehistogramof Fig. 5. Itcanbe
seenthattheshapeof thehistogramfor bothcasesare
almostsameeachother, exceptfor afewdataexceed-
ing80kPa, whichcorrespondstoq
t
atsandlayers.The
existenceof theseextremevalues,thoughitsfrequency
isverysmall, significantly
1
and
2
values.
1
and
2
of q
t
() forinvestigatedsitesinthispaper
areplotted in Pearsons chart, as shown in Fig. 4. It
canbeseenthat
1
and
2
relationforclayeyandmost
granular grounds are distributed along log Normal
distributionlineand
1
valuesarenearlyzero.There-
fore, it can bejudged that q
t
() for clayey and most
granular groundsfollownormal distribution. Some
granular grounds (Kemigawa, Higashiougishimaand
Nakashibetsu) mayfollowBetadistribution.
4 ANALYSISOF SPT ANDUCT DATA
AlthoughCPT has gradually usedinJ apan, themost
conventional testingmethodisSPT for granular soils
andtheunconfinedcompressiontest(UCT) forclayey
soil. The same analysis is carried out for data from
SPT and UCT. Figs 6 and 7 show q
u
/2 and N val-
uesfor BusanandKagoshimasites, respectively, with
superimposedtrendlinescalculatedbytheregression
method. It is interestedto notethat thetrendlinefor
the Kagoshima site also has a negative slope, simi-
larlytothat for CPT (seeFig. 4). Analyzedresultsare
indicatedinTable2.
Figure 6. Measured values and trend line from UCT at
BusanSite.
Figure 7. Measured values and trend line from SPT at
KagoshimaSite.
For granular grounds, the slope of the trend line
(i.e., b value) iscomparedfor SPT andCPT inFig. 8.
It can be seen that there is a good relation between
them. Fig. 9showstherelationbetween obtainedby
SPT andCPT. It isalsofoundthat when for CPT is
large, for SPT isalsolarge. Thisfact indicatesthat
thevariationfromthetrendlinemaybecausedmainly
290
Table2. AnalyzedresultsfromUCT & SPT.
Testing Sample
Site Soil Type Method Number a(MPa/m) b(MPa/m)
1

2
(Mpa)
Hatchirogata Clay UCT 28 0.012 0.00072 1.02 4.83 0.0047
Busan Clay UCT 15 0.028 0.0029 9.67 14.04 0.029
Singapore Clay UCT 7 0.016 0.0032 0.46 1.41 0.0019
Kagoshima VolcanicAsh SPT 35 21.46 0.28 0.091 1.95 3.64
Yodogawa Sand SPT 20 21.16 0.17 0.18 4.74 5.55
Nakashibetsu VolcanicAsh SPT 10 11.11 2.96 0.003 3.08 3.59
Kemigawa Sand SPT 22 21.04 3.21 1.54 4.76 4.64
Higashiohgishima Filling SPT 13 1.18 0.75 0.47 3.07 4.19
(sandy)
Tsuruga Filling SPT 24 3.09 0.66 0.19 2.67 4.78
(CrushRock)
Tomakomai VolcanicAsh SPT 26 9.65 0.024 1.21 3.75 1.98
Figure8. ComparsionbetweenSPT andCPT for b value.
Figure9. ComparsionbetweenSPT andCPT for value.
by heterogeneity in theground, not by measurement
error.
Unlike or b,
1
and
2
for Pearsonschart donot
have dimension so that these values are comparable
for grounds measured by SPT and UCT. 1 and
2
arecalculatedfromSPT or UCT dataandplottedin
thePeasonschartof Fig. 10. ComparedwithPeasons
chart fromCPT, pointsfromSPT andUCT aremuch
Figure10. (a) Comparsion between SPT and CPT for
1
value. (b) ComparsionbetweenSPT andCPT for
1
value.
scattered, especially
2
valuesarelarger.
1
and
2
are
comparedinFig. 10(a) and(b). Thoughanymeaning-
ful relation for
1
can behardly identified, it seems
that thereexistsaweakrelationfor
2
.
5 CONCLUSIONS
UsingCPT data, soil variabilitywasexaminedfor 11
differentsites, consistingof varioussoil materials: i.e.,
291
clay, volcanicash, sandandcrushedgravel. Measured
q
t
values are broken down into trend function and
residuals, assumedthatthetrendisexpressedbyalin-
ear linetodepth. Propertiesof residualsareanalyzed
andthemainconclusionsareasfollows:
1) Thedistributionof residualsfromCPTatmostsites
follows the normal distribution. However, some
granular grounds (Kemigawa, Higshiohgishima,
Nakashibetsu) followBetadistribution.
2) Statistical parameterssuchas,
2
calculatedfrom
SPT or UCT haverelativelygoodrelationtothose
fromCPT.
REFERENCES
Baecher, B. &Christain,T. 2003. ReliabilityandStatisticsin
Geotechnical Engineering. England: Wiely
Mimura, M. 2003. Characteristicsof someJ apanesenatural
sands-datafromundisturbedfrozensamples, Character-
istation and Engineering Properties of Natural SoilsTan
et al.(eds), (2): 11491168.
Tanaka, H. 2006. Goetechnical properties of Hachirogata
Clay. Characterization and Engineering Properties of
Natural Soils, Vol. 3: 18311854.
Tanaka, H., Mishima, O. & Tanaka, M. 1999. Applicabil-
ity of CPT and DMT for grounds consisting of large
granualar particles, Journal of Japanese Society for Civil
Engineerings, III49: 273283, (inJ apanese).
Tanaka, H., Mishima, O., Tanaka, M., Park, S. Z,
J eong, G. H. & Locat, J. 2001a. Characterization of
Yangsanclay, Pusan, Korea, Soils and Foundations, Vol.
41(2) : 89104.
Tanaka,H.,Locat,J.,Shibuya,S.,Tan,T.S.&Shiwakoti,R.D.
2001b. Characterization of Singapore, Bangkok and
Ariake clays, Canadian Geotechnical Journal, 38:
378400.
292
Reducinguncertaintiesinundrainedshear strengths
J ianyeChing
Yi-ChuChen
Kok-KwangPhoon
ABSTRACT: Undrainedshear strengths (s
u
) play important roles ingeotechnical designs. Inthecontext of
geotechnical reliability-based design, reducing uncertainties in s
u
can bean important research topic. There
areat least two ways of reducing uncertainties in s
u
: conduct laboratory or in-situ tests to obtain indices or
parameterstocorrelates
u
indirectly. Thewayof reducinguncertaintiesins
u
canbechallenging. Thechallenge
liesinthefact that theso-obtainedindicesandparameters, e.g. CPT value, cannot bedirectlyusedtoestimate
s
u
but canestimates
u
onlythroughcorrelations. Aboveall, thereisachallengeincombininginformation: how
toreduceuncertaintiesins
u
whentherearemultiplesourcesof information? Inthispaper wewill addressthe
aforementionedchallengesandproposeaprobabilisticframeworktohandlethesedifficulties. Setsof simplified
equationswill beobtainedthroughtheprobabilisticanalysisforthepurposeof reducinguncertainties: theinputs
to theequations aretheresults of in-situor laboratory tests andtheoutputs aretheupdatedmeanvalues and
coefficientsof variation(c.o.v.) of thedesirableundrainedshear strengths. Theuncertaintiesins
u
will decrease
when the number of inputs increase, i.e. more information is available. The results of this research may be
beneficial togeotechnical reliability-baseddesign.
1 INTRODUCTION
Uncertaintiesarecommonly encounteredingeotech-
nical engineering. Possible sources of uncertain-
ties include inherent variabilities, measurement
errors, modeling uncertainties, etc. More economi-
cal geotechnical designscanbeachievedbyreducing
the uncertainties in soil shear strengths through site
investigation.
1.1 Reducing uncertainties by correlations
In practice, it is already well-known that field or
laboratory test data, denoted as test indices from
hereon, can becombined to reducetheuncertainties
inundrainedshear strengthsthroughcorrelations.
For instance, giventhefieldSPT-N test data, it is
possibletoinferfirst-orderestimatesof themeanvalue
and coefficient of variation (c.o.v.) of theundrained
shear strength (S
u
) of the clay under consideration.
This process of pairwise correlation is illustrated in
Figure1 for agivenobservedSPT-N value(N), the
correspondingmeanvalueandc.o.v. of theundrained
shear strengthcanbeestimated.
Intheliterature, suchpairwisecorrelationsbetween
varioustestindicesandundrainedshearstrengthshave
beenwidely studied. Table1lists examples of previ-
ousresearchstudyingsuchcorrelations. Inparticular,
Figure1. S
u
versusSPT-Nrelationship.
Kulhawy and Mayne(1990) and Phoon (1995) both
containfairly comprehensivereviews about thepair-
wise correlations between various test indices and
undrainedshear strengths.
1.2 Graphical model for clayey soils
The undrained shear strength (S
u
) considered is
the undrained shear strength determined by CIUC
293
Table1. Previous research studying pairwisecorrelations
betweenvarioustest indicesandundrainedshear strengths.
Correlationpair Examplesof previousresearch
Standardpenetration Terzaghi andPeck(1967);
test (SPT-N) Haraet al. (1974)
Conepenetration KeavenyandMitchell (1986);
test (CPT) KonradandLaw(1987)
Pressuremeter test (PMT) Mair andWood(1987)
Dilatometer test (DMT) LacasseandLumme(1988)
Vaneshear test (VST) Bjerrum(1972); Mersi (1975)
Plasticityindex(PI) Skempton(1957); Chandler
(1988)
Overconsolidation Laddet al. (1977);
ratio(OCR) J amiolkowski et al. (1985)
Figure2. Graphical model for clayeysoils.
(isotropically consolidated undrained compression)
tests.Figure2presentsthegraphical model adoptedfor
claysinthispaper. For themodel of S
u
of clayeysoils,
theadoptedtestindicesarelimitedtothefollowing: (a)
overconsolidation ratio (OCR); (b) energy-ratio cor-
rected SPT-N value(N
60
); (c) adjusted CPT reading
q
T
=q
T

v0
, where
v0
is the total vertical stress,
and q
T
is theCPT reading corrected with respect to
theporepressurebehindthecone.
The underlying assumptions of this model are:
(a) OCR isthemainfactor influencingtheundrained
shear strength, which in turn influences SPT-N and
CPT values, i.e. theundrainedshearstrengthistreated
astheconsequenceof OCR, andSPT-NandCPT val-
uesaretreatedasconsequencesof theundrainedshear
strength; (b) given theundrained shear strength of a
clay, its SPT-N and CPT values are independent of
OCR, i.e. the undrained shear strength serves as a
sufficient statisticsof theSPT-NvalueandCPT read-
ing: oncetheundrainedshear strengthisknown, OCR
does not contain further information of SPT-N and
CPT values; (c) given the undrained shear strength,
its SPT-N valueandCPT readingaremutually inde-
pendent. In other words, the SPT-N value and CPT
readingaretreatedastwopiecesof independentinfor-
mationfor theundrainedshear strength.Thismodel is
deemedto bereasonablefor unstructuredunfissured
inorganicclays.
Thefirst two moments of thefollowingquantities
are needed for the Bayesian analysis: (a) S
u
condi-
tioning on OCR; (b) N
60
conditioning on S
u
; (c) q
T
conditioningonS
u
. Detailsaregivenbelow.
2 PROBABILISTIC MODELSFOR PAIRWISE
CORRELATIONS
2.1 Database
Probabilistic models for the pairwise correlations
betweenundrainedshear strengthsandthechosentest
indicesarenecessaryfortheBayesiananalysis.There-
fore, efforts were made in compiling a correlation
database, e.g. theS
u
vs. SPT-N datapoints shownin
Figure 1, fromthe literature to estimate the proba-
bilistic models for thepairwisecorrelations. For the
database, it is always a concern whether the com-
piled data points are enough to cover a wide range
of possible scenarios. The following guidelines are
followed to mitigate this concern: (a) Unless men-
tioned explicitly, data points in the database do not
include those from special clays, e.g. fissured and
organic clays. Therefore, the corresponding correla-
tions should not beapplied to thosesoils; (b) In the
casethat all datapointsof aparticular correlationare
fromthe same geographical region, that correlation
may beonly applicabletothat region. Ingeneral, the
applicabilitytootherregionsisquestionable. Bayesian
updatingwill notreducethephysical limitationsinher-
entinexistingpairwisecorrelations.Itmerelyprovides
amorerational andsystematicmethodfor combining
informationthat will serveasauseful complement to
engineeringjudgment.
Thefollowingstrategyisadoptedtoobtaintheprob-
abilisticmodelsforthepairwisecorrelationsgiventhe
pairwise-correlationdatapoints:
(a) In the case that the pairwise data points
for a certain correlation are sufficient, the data
quality will be verified through empirical corre-
lations in the literature. Only those data points
that areconsistent withtheliteraturewill belater
used to derive the probabilistic model for that
correlation.
(b) In thecasethat thepairwisedatapoints for a
certain correlation are insufficient, the empirical
correlationprovidedbyliteraturethatbestmatches
thedatapointswill beadoptedtoderivetheprob-
abilisticmodelsfor that correlation. If thec.o.v. of
theadopted empirical correlation is not available
intheliterature, it will beestimatedfromthedata
pointsat hand.
(c) In the case that the pairwise data points
for a certain correlation are absent, the empir-
ical correlation provided by literature will be
adopted.
2.2 Probabilistic models for pairwise
correlations in clays
Recall that the first two moments [mean and coef-
ficient of variation (or standard deviation)] of the
following quantities are needed: (a) S
u
conditioning
onOCR; (b) N
60
conditioningonS
u
; (c) q
T
condition-
ingonS
u
.Thederivationsof thefirsttwomomentsare
presentedindetail below.
294
Figure3. S
u
,
v0
vs. OCR correlation, andthemeanvalue
and95%confidenceinterval proposedbythisresearch.
2.2.1 First two moments of S
u
conditioning on OCR
Overconsolidation ratio OCR is assumed to be the
mainbasicindexaffectingS
u
. Mayne(1988) compiled
aset of S
u
,
v0
vs. OCR data(theS
u
datapointswere
all fromCIUC tests), showninFigure3.
Basedonthedata, aleast-squaremethodistakento
obtainthefollowingcorrelation:
where
Su
is the prediction error term. Its standard
deviationisfoundtobearound0.237. Therefore,
andthestandarddeviationof thiscorrelationis0.237.
Figure3shows themeanvalueand95%confidence
interval of thisequationandthecomparisonwiththe
actual data. InthecasethattheOCR informationisnot
available, theprior c.o.v. for S
u
is takento beavery
largenumber.
2.2.2 First two moments of N
60
conditioning on S
u
The correlation between S
u
of clayey soils and the
SPT-N value is well known. Figure 4 compiles the
pairwisedatafor thiscorrelationsummarizedbyHara
etal. (1974) (alsoseePhoon(1995) andKulhawyand
Mayne(1990)), wheretheundrained shear strengths
aredeterminedby UU tests. Notethat all datapoints
herearefromclaysinJ apan.
Thefollowingequationwasproposed:
whereS
UU
u
denotestheundrainedshearstrengthdeter-
minedbyaUUtest; NisuncorrectedSPT-Nvalue; the
standarddeviationof
1
isroughly0.15(Phoon1995).
TheSPT-Nenergyratioisroughly78%inJ apan(Chen
2004); therefore, theNvaluein(3) isroughly1.3N
60
.
Inother words,
Figure 4. The mean value of 95% confidence interval
proposedbythisresearch.
Figure 5. ln(S
UU
u
,S
u
) vs. ln(S
u
,
v0
) correlation, and the
mean value of 95% confidence interval proposed by this
research.
Equation (4) can be used to obtain the first two
moments of S
UU
u
conditioning on N
60
; however, our
goal here is to derive the first two moments of N
60
conditioningonS
u
. WiththesamedatasetinFigure4,
aleast-squaremethodistakentoobtainthefollowing
equation:
wherethestandard deviation of
2
is 0.407, and the
unit of S
UU
u
isinkPa. The95%confidenceinterval of
thisnewequationisshowninFigure4forcomparison.
Moreover, according to the database presented in
ChenandKulhawy(1993) (seeFigure5), thecorrela-
tionbetweenundrainedshear strengthdeterminedby
UU andCIUC testsareasfollows:
wherethestandard deviation of
3
is roughly 0.167,
and
v0
istheeffectivevertical stress. Themeanvalue
of 95%confidenceinterval together withthedatabase
295
presentedinChenandKulhawy(1993) areplottedin
Figure5. Equations(5) and(6) implythat
where S
u
and
v0
are both in the unit of kPa, and
the standard deviation of
N
is (0.407
2
+1.230
2
0.167
2
)
0.5
=0.456=0.456. Therefore,
2.2.3 First two moments of q
T
conditioning on S
u
Thecorrelationbetweenundrainedshear strengthand
theCPT readinghasbeenstudiedinseveral literature,
e.g. Hanzawa (1992) (direct shear test), Tanaka and
Sakagami (1989) (CIUC test), Fukasawaet al. (2004)
(vane shear and unconfined compression tests) and
Anagnostopoulosetal. (2003) (UUtests).Thefollow-
ing correlation equation is commonly adopted (e.g.
Lunneet al. (2002), KulhawyandMayne(1990)):
whereN
k
iscalledtheconebearingfactor.Theoretical
studies showed that N
k
ranges from7 to 18. Several
experimental studies indicated that themeasured N
k
canvary wildly from4.5-75, probably dueto incon-
sistent referencestrengths, mixing of different types
of cones, needfor correctiononporewater pressure,
etc. (KulhawyandMayne1990).
Ontheother hand, Phoon(1995) reportedthat:
for CIUC test results. ThiscorrespondstoN
k
=12.7,
which agrees well with the theoretical results (N
k
rangingfrom7to18). Phoon(1995) further reported
that the uncertainty of (10.) is roughly 35%. The
probabilisticversionof (10) istherefore
where
4
is zero-meanwithstandarddeviationequal
to0.34, correspondingtothe35%uncertaintyof (10.).
A simpleBayesianargument cantransformtheabove
equationinto:
where
qT
is also zero-meanwithstandarddeviation
equal to0.34. Therefore,
3 BAYESIANINFERENCE WITH
MULTIVARIATETEST DATA
In the case that only a single piece of information
is available, updating the mean value and c.o.v. of
undrainedshear strengthisnot difficult. For instance,
given that SPT-N valueof aclay sampleis 10, from
Figure1itisconcludedthattheupdatedmeanvalueof
S
u
isroughly 160kPa, andc.o.v. isroughly 30%. The
sameprinciplecanbeusedtoupdatemeanandc.o.v.
of undrained shear strength based on another single
pieceof information. However, inthecasethat multi-
variatetest dataisavailable, e.g.: SPT-N andOCR of
aclay sampleareavailable, updatingthemeanvalue
andc.o.v. of theS
u
islessstraightforward.
Bayesiananalysisisanatural wayof handlingmulti-
variateinformation, evenconflictinginformation.The
basicBayes ruleconsistsof thefollowingequation:
wherex andy canbebothvectors; y is theuncertain
variableof interest, whilex is theobservedvariable.
f (y)iscalledthepriorPDFof y, quantifyingtheuncer-
taintiesiny beforeobservationonx ismade,andf (x|y)
iscalledthelikelihoodfunctionof y givenx. f (y|x) is
theupdatedor posterior PDF of y conditioningonthe
informationof x.
As an example, let y be the logarithm of the
undrainedshearstrength.InthecasethatOCRisgiven,
f (ln(S
u
)|ln(OCR)) thenserves as theprior PDF f (y).
If theobservedvariablexisthecorrectedSPT-Nvalue
N
60
, f (ln(N
60
)|ln(S
u
)) servesasf (x|y). Then
represents the updated (posterior) PDF of the
undrainedshear strengthgiventhemultivariateinfor-
mation {N
60
,OCR}. In this paper, we have imple-
mented the aforementioned model assumption that
conditioningonS
u
, N
60
isindependent of OCR.
4 MAINRESULTS
If the variable of interest and the observed variable
arejointly Gaussian, it is possibleto deriveall rele-
vant conditional means and conditional variances in
closed-form. Using these results, the updated mean
andvarianceof thelogarithmof theundrainedshear
strengthsconditioningonvariouscombinationof mul-
tivariate test data are listed in the following with
detailedderivations:
ConditioningonOCR:
296
ConditioningonN
60
:
Conditioningonq
T
:
ConditioningonOCR, N
60
:
ConditioningonOCR, q
T
:
ConditioningonN
60
andq
T
:
ConditioningonOCR, N
60
, q
T
:
The above results provide estimates for the first
two moments of ln(S
u
). Let us denotetheestimated
meanvalueandstandarddeviationof ln(S
u
) bymand
s, respectively, then the mean value and c.o.v. of S
u
areexp(m+s
2
/2)and[exp(s
2
)1]
0.5
, respectively, by
assuminglognormality.
5 CASE STUDIES
5.1 Clays in a deep excavation site of Taipei
A deep excavation siteis extracted fromOu (2006).
SPT-N andCPT testswereconductedat thissite. The
soil profileincludesthreethickclayeylayersandthree
thin sandy layers. The water table is 2 mbelow the
surface.
Conepenetration and vaneshear test results were
takento estimatetheundrainedshear strengthof the
clays. Moreover, several undisturbedclaysamplesare
Table2. In-situtest dataandindicesfor theTaipei case.
Equiv.
Test indices
CIUC
Depth Test S
u
value PI
v0

v0
q
c
q
1
T
(m) type (kPa) OCR (%) N
60
(kPa) (kPa) (kPa) (kPa)
11.3 CK
0
U 43.1 1.6 9.1 4.0 206.4 115.5 835.3 628.9
12.8 UU 76.5 1.4 9.1 3.0 233.4 128.0 810.5 577.1
14.8 VST 82.4 1.2 12.8 3.3 270.2 144.9 730.1 459.9
16.1 UU 88.7 1.2 14.6 4.0 293.6 155.6 713.6 420.0
17.1 CK
0
U 58.4 1.1 17.5 4.0 312.0 164.0 766.8 454.9
17.8 UU 85.8 1.1 18.9 4.0 324.7 169.9 803.7 479.0
18.3 VST 93.8 1.1 18.5 4.0 334.7 174.5 830.0 495.4
20.2 UU 106.2 1.0 17.3 4.0 368.4 190.0 911.9 543.5
20.2 UU 111.2 1.0 17.3 4.0 369.2 190.3 913.7 544.6
20.9 VST 115.7 1.0 16.8 4.0 381.1 195.8 942.6 561.5
22.7 VST 101.9 1.0 16.0 4.8 413.9 210.9 1050.6 636.7
24.0 UU 121.3 1.0 16.2 5.6 437.4 221.7 1210.2 772.8
26.6 UU 139.8 1.0 13.8 7.2 485.4 243.7 1532.6 1047.3
1
Strictly speaking, this is not q
T
becausetheCPT reading was not
correctedagainst theporepressurebehindthecone.
extracted fromthesite, and laboratory tests, includ-
ing UU and CK
0
U tests, were taken to determine
theundrainedshear strengths. Inprinciple, undrained
shear strengths from different test types cannot be
directly compared. Therefore, attempts are made to
convert all undrainedshear strengthsintotheir equiv-
alent CIUC valuesthroughempirical transformations
suggestedbyKulhawyandMayne(1990) for CK
0
UC
toCIUC, suggestedbyChenandKulhawy(1993) for
UU to CIUC and suggested by Ladd et al. (1977)
for VST to field value (closer to DSS value; the
DSS values arefurther convertedto CIUC values by
the transformation equations suggested by Kulhawy
and Mayne (1990)). Table 2 summarizes the tested
undrainedshear strengths andtheir equivalent CIUC
values as well as theother in-situ test indices of the
clayat variousdepths.
Noticethatq
T
dataisnotavailableforthissitesince
thebehindconeporepressurewasnotdocumented.As
aresult, q
c
dataisdirectlytakentocomputeq
T
forthis
case study, i.e. q
T
=q
T

v0
. Because q
T
is always
greater thanq
c
, theactual q
T
values shouldbelarger
than the q
T
value reported in the table to a certain
degree.
Basedonthesedataandtheformulas providedin
the previous section, the updated mean values and
95% confidence intervals (2 standard deviations)
with respect to depth are plotted in the left column
inFigure6for variouscombinationsof test indicesD
(Cases 1-7 in the figure). The mean values E(S
u
|D)
andstandarddeviationsVar(S
u
|D)
0.5
arebothnormal-
ized with respect to the measured equivalent CIUC
valuesS
m
u
, somean=1indicatestheupdatedmeanis
thesameasthemeasuredequivalent CIUC value.
Itisevidentthattheupdatedstandarddeviationof S
u
decreasesasmoreinformationistakenfor theupdat-
ing. Let us takeCases 1, 2and4as examples: When
onlyonepieceof informationisinvolved(Case1for
OCR and Case 2 for N
60
), the confidence intervals
297
Figure6. Updatedmeanvaluesand95%confidenceinter-
valsof S
u
withrespect todepth(normalizedwithrespect to
themeasureS
u
) for theTaipei case.
seem large, manifesting more uncertainties. When
theOCR andN
60
informationis combined(Case4),
the confidence intervals start to shrink; moreover,
although the confidence intervals become smaller,
most of themstill contain1, i.e.: themeasuredequiv-
alent CIUC value S
m
u
still lies within the intervals.
When all information is implemented, i.e. Case 7,
thec.o.v. of theundrainedshear strengthisaslowas
0.16. Comparedtothec.o.v.sforCases1-3thatranges
from0.24to0.34, the0.16c.o.v. isamajor improve-
ment: theuncertainty in S
u
is effectively reduced by
incorporatingmultivariatedata.
Noticethat relativelylarger biasesarefoundinthe
estimatedS
u
valuesfor Cases3, 5and6. Thiscanbe
clearly seenintheplots for Cases 3, 5and6, where
the estimated mean values E(S
u
|D) deviate from1.
These are the cases where q
T
information is incor-
porated. These biases are obviously due to the fact
that the employed q
c
values here are less than the
actual q
T
values, so theestimatedmeanvalues of S
u
aresignificantly less than their actual CIUC values.
However, by also incorporating OCR and N
60
infor-
mation, i.e. Case7, thebias is significantly reduced.
Therefore, incorporating more information not only
reducesuncertaintiesbutalsoreducesbias. Obviously,
if thedatais judged to beincompletesuch as q
T
, it
is preferable not to include these data, even if bias
canbereducedbyincludingmoresourcesof informa-
tion. Inother words, wedonot recommendincluding
all sourcesof informationindiscriminately. Engineer-
ing judgment is still important in this interpretation
process.
Theperformanceof theproposedBayesianmethod
is further compared with the level-1 T.E.A.M.
approach (Technical Expert Averaging Method, Bri-
audet al. (2002)). Thelevel-1TEAM approachis an
intuitiveway of combiningmultivariateinformation.
Taking Case 7 as an example: there are 3 predic-
tionmethods(predictionsbasedonOCR, N
60
andq
T
)
over 13measuredevents{S
m
u
j
, j =1, . . . , 13}, andthe
i-th method gives prediction E(S
uj
|D
i
) for S
m
u
j
. The
ratio r
ij
=E(S
uj
|D
i
),S
m
u
j
is treated as a measure of
performance of the i-th prediction method over the
j-th measured event; it has standard deviation=
i
,
which is estimated as thesamplestandard deviation
of {r
ij
: j =1, . . . , 13}. The TEAM ratio r
TEAM,j
for
thej-thmeasuredevent issimplythearithmeticaver-
age (r
1j
+r
2j
+r
3j
)/3. TheTEAM standard deviation
is simply
TEAM,j
=[(
2
1
+
2
2
+
2
3
)/3]
0.5
it is not
a function of j and hence we drop j fromthe sub-
script from hereon. Note that the TEAM standard
deviation
TEAM
isestimatedbasedontheratios{r
ij
:
j =1, . . . , 13}, which in turn based on themeasured
data{S
m
u
j
, j =1, . . . , 13}.TheTEAM95%confidence
intervalsforthej-thmeasuredeventcanthenbeplotted
asr
TEAM,j
2
TEAM
.
The estimated TEAM ratios and their 95% con-
fidence intervals are plotted in the right column in
Figure6. For thecases not involving q
T
, i.e.: Cases
1, 2 and 4, theperformanceis similar to thosefrom
theBayesianmethodproposedinthis paper (seeleft
column). It is reassuring that theproposed Bayesian
methoddoes agreewiththeempirical rule-of-thumb
observedinnumerousfoundationcapacityprediction
exercises that averaging seems to improvepredic-
tions. It is noteworthy that the proposed Bayesian
method can also be applied to improve estimation
of foundationcapacitiesbycombininglaboratoryand
in-situbasedformulae.TheBayesianmethodisadmit-
tedly moreanalytically involvedthansimpleaverag-
ing, but it does provideamoregeneral andrigorous
framework for derivingpractical approximateresults
suchas Eqs. (16.) to (22.). Inaddition, theBayesian
methodwill servetovalidatethetheoreticallycorrect-
nessof averaging.
Furthermore,itisquitesignificantthattheBayesian
methodgivessimilarstandarddeviationstotheTEAM
methodfor Cases 1, 2and4: thestandarddeviations
givenbytheformer arecompletelyindependentof the
S
u
valuesof the13datapoints(thestandarddeviations
areactual fixednumbersasseeninEqs. (16.) to(22.)),
whilethoseestimatedby thelatter dependon theS
u
values of the13 datapoints. In this sense, thecom-
parisonbetweentheBayesianandTEAM methodsis
alreadyinequitable, becausethelatter usesveryvalu-
ablenewinformation,i.e.:theS
m
u
j
,valuesof the13data
points, that isusuallynot availablebeforesiteinvesti-
gationistaken.TheconsistencybetweentheBayesian
298
andTEAM resultssuggeststhat theBayesianmethod
can effectively predict standard deviations close to
their actual values.
For the cases involving q
T
, i.e.: Cases 3, 5, 6
and 7, theTEAM standard deviations are obviously
smaller thanthoseestimatedby theBayesianmethod
althoughtheTEAM ratios areobviously biased. The
TEAMstandarddeviationissmall becauseitispurely
thesamplestandard deviation of theobserved ratios
{r
ij
: j =1, . . . , 13}: whentheseratiosareconsistently
biased asseeninCases3, 5, 6and7 their sample
standard deviation will besmall. However, thestan-
dard deviation of the Bayesian method is estimated
purelyfromthetrainingdatabase(datafrompast cor-
relations), not fromtheobserved ratios. In fact, one
canseethat thevariancesfromEqns(16.) to(22.) are
independent of theobservedratios. Inthis sense, the
standarddeviationestimatedbytheBayesianmethod
seemstoberobust against thebiasintheratios.
5.2 Clays in various regions of the world
Ten soil profiles are extracted fromRad and Lunne
(1988); they arelocated around theworld, including
Norway, NorthSea, NorwegianSea, England, Brazil
andCanada. Inthesesoil profiles, bothOCR andq
T
dataareavailable; moreover, theactual valuesof S
u
are
knownfromeither CIUC or CK
0
UC tests. Similarly,
the CK
0
UC values are converted to their equivalent
CIUCvaluesbeforecomparisonsaremade.Threesce-
nariosof siteinvestigationinformationareconsidered:
Case1: onlyOCR informationisknown; Case2: only
q
T
informationisknown; Case3: OCRandq
T
areboth
known.
The updated mean values and 95% confidence
intervals, both normalized with respect to the mea-
sured equivalent CIUC S
u
, are shown in the first
column in Figure 7. The analysis results show that
theuncertaintiesareeffectivelyreducedbyincorporat-
ingmoreinformation: whenOCR andq
T
information
is both implemented, i.e. Case 3, the c.o.v. of the
undrainedshearstrengthisaslowas0.195. Compared
tothec.o.v.sfor Cases1and2that rangesfrom0.237
to0.337, the0.195c.o.v. isamajor improvement. Fur-
thermore, althoughtheconfidenceintervalsforCase3
aresmall, they still mostly contain 1, indicatingthat
theanalysisperformssatisfactorily.
For all cases, the performance of the proposed
Bayesianmethodis againcomparedwiththelevel-1
T.E.A.M. approach (the second column). Unlike the
deep-excavation example, they are similar to the
results of the multivariate Bayesian analysis proba-
bly because in the current example, the systematic
bias in theTEAM ratios does not exist. Again, it is
quitesignificantthattheBayesianmethodgivessimi-
lar standard deviations to theTEAM method for all
cases: the Bayesian method can effectively predict
standarddeviationsclosetotheir actual values.
5.3 Discussions for the case studies
For bothcasestudies, theupdatedmeanvaluesand
confidenceintervals (i.e. c.o.v.s) areconsistent to
Figure7. Updatedmeanvaluesand95%confidenceinter-
valsof S
u
(normalizedwithrespecttothemeasureS
u
) for the
tensites.
the actual S
u
data points except the cases in the
first examplewhereCPT informationis involved,
includingCases3, 5and6, wherelargebiasesare
foundintheprediction. However, suchbiases can
beunderstandable: q
c
rather thanq
T
was usedfor
theanalysis. Therefore, it isexpectedthat thepre-
dictedS
u
valueswouldbeconservativesinceq
c
is
alwayslessthanq
T
.
It seems helpful in reducing bias to incorporate
moreinformation.AswehaveseenfromCases3, 5
and 6 in the first example: there are significant
biasesduetotheincorrect useof q
c
. Nonetheless,
when all availableinformation is implemented in
Case7, thebiasisobviouslyreduced.
Incorporatingmoreinformationalsohelptoreduce
uncertainties, i.e. make the confidence intervals
smaller. This is obvious fromtheanalysis results
inFigures6and7.
J udging from the analysis results, Eqs. (16.) to
(22.) seemtoprovideresultsthat areconsistent to
actual S
u
data. Moreover, this consistency seems
to be independent of the location of the site of
interest.
Itispossibletoreducethec.o.v.of S
u
downto0.16if
OCR, q
T
andN
60
informationareall implemented.
This reductioninc.o.v. is significant: thec.o.v. of
the inherent variability of S
u
can be as large as
0.3-0.4. Infact, withonlyinformationof OCR and
q
T
, thec.o.v. isalreadyreducedto0.195.
299
The TEAM approach seems to provide similar
results to the Bayesian method proposed in this
paper whenthereisnosystematic biasintheesti-
mation. This is quitesignificant sinceit indicates
that the Bayesian method can effectively predict
standarddeviationsclosetotheir actual values.
When there is systematic bias, theTEAM approach
maygiveasmall standarddeviation. Nonetheless, the
standarddeviationestimatedbytheBayesianmethod
is based on the training database (data from past
correlations) and is robust against the bias in the
estimation.
6 CONCLUSION
A newframework isproposedtoupdatethefirst two
momentsof undrainedshear strengthsof clayey soils
basedonin-situandlaboratory test dataandindices,
e.g., overconsolidation ratio, SPT-N value and CPT
reading. This new method is based on pairwisecor-
relations developed by literature, but it implements
the Bayesian analysis to accommodate multivariate
correlationstoupdatethemoments.
Themainproductof thispaper isasetof equations
whoseinputsaretheobservedmultivariatetest index
values and outputs aretheupdated mean values and
coefficientsof variation(c.o.v.) of theundrainedshear
strength. Onereal casestudy areemployedto verify
theconsistencyof theproposedframeworkinpredict-
ingshear strengthsof clays. Theresultsshowthat the
proposedframeworkofferssatisfactoryestimationsof
theundrainedshear strengths.
REFERENCES
Anagnostopoulos, A., Koukis, G., Sabatakakis, N., and
Tsiambaos, G. (2003). Empirical correlations of soil
parametersbasedonconepenetrationtestsforGreeksoils.
Geotechnical andGeological Engineering, 21, 377387.
Bjerrum, L. (1972). Embankment onSoft Ground. Proceed-
ings of ASCE Specialty Conference on Performance of
EarthandEarth-SupportedStructures, Lafayette.
Briaud, J.L., Goparaju, K. and Dahm, P.F. (2002). The
T.E.A.M. approach in geotechnical engineering. ASCE
Geotechnical Special Publications No. 116(DeepFoun-
dations2002), Vol. 2, 976992.
Chandler, R.J. (1988). The in-situ measurement of the
undrained shear strength of clays using the field vane.
VaneShearStrengthTestinginSoils: FieldandLaboratory
Studies(DTP1014), ASTM, Philadelphia, 1344.
Chen, J.R. (2004). Axial Behavior of Drilled Shafts in
GravellySoils. Ph.D. Dissertation, Cornell University.
Chen, Y.J. and Kulhawy, F.H. (1993). Undrained strength
interrelationshipsamongCIUC,UU,andUCtests.J ournal
of Geotechnical Engineering, 119(11), 17321750.
Fukasawa,T., Mizukami, J., andKusakabe, O. (2004).Appli-
cability of CPT for construction control of seawall on
soft clay improved by sand drain method. Soils and
Hanzawa, H. (1992). A new approach to determine soil
parameters free fromregional variations in soil behav-
ior and technical quality. Soils and Foundations, 32(1),
7184.
Hara,A.,Ohta,T.,Niwa,M.,Tanaka,S.,andBanno,T.(1974).
Shear modulusandshear strengthof cohesivesoils. Soils
andFoundations, 14(3), 112.
J amiolkowski, M., Ladd, C.C., Germaine, J.T., and
Lancellotta, R. (1985). New developments in field and
laboratorytestingof soils. Proceedingsof the11thInter-
national Conferenceon Soil Mechanics and Foundation
Engineering, SanFrancisco.
Keaveny, J.M. and Mitchell, J.K. (1986). Strength of fine-
grainedsoilsusingthepiezocone. InUseof In-SituTests
inGeotechnical Engineering(GSP-6), Ed. S. P. Clemence,
ASCE, NewYork.
Konrad, J.M. andLaw, K.T. (1987). Undrainedshearstrength
from piezocone tests. Canadian Geotechnical J ournal,
24(3), 392405.
Kulhawy,F.H.andMayne,P.W.(1990).Manual onEstimating
Soil Properties for FoundationDesign, Report EL-6800,
ElectricPower ResearchInstitute, PaloAlto.
Lacasse, S. and Lunne, T. (1988). Calibration of dilatome-
ter correlations. Proceedings of the 1st International
SymposiumonPenetrationTesting(ISOPT-1), Orlando.
Ladd, C.C., Foote, R., Ishihara, K., Schlosser, F. and
PoulosH.G.(1977).Stress-deformationandstrengthchar-
acteristics. Proceedings of 9th International Conference
onSoil MechanicsandFoundationEngineering, Tokyo.
Lunne, T., Robertson, P.K. andPowell, J.J.M. (2002). Cone
PenetrationTestinginGeotechnical Practice. SponPress,
London.
Mair, R.J. and Wood, D.M. (1987). Pressuremeter Testing.
Butterworths, London.
Mayne, P.W. (1988). DeterminingOCR inclayfromlabora-
torystrength.ASCEJ ournal of Geotechnical Engineering,
114(1), 7692.
Mersi, G. (1975). Discussionof Newdesignprocedurefor
stability of soft clays. ASCE J ournal of Geotechnical
Engineering, 101(4), 409412.
Ou, C.Y. (2006). Deepexcavationengineering, Scientific&
Technical PublishingCo. Ltd.
Phoon, K. K. (1995). Reliability-based Design of Founda-
tionsforTransmissionLineStructures.Ph.D.Dissertation,
Cornell University, Ithaca, NY.
Robertson, P.K. andCampanella, R.G. (1989). Guidelinesfor
Geotechnical Design Using theConePenetrometer Test
andCPT withPorePressureMeasurement. Hogentogler
Co., Inc(1989).
Skempton,A. W. (1957). Discussionof Planninganddesign
of newHongKongairport. Proceedingsof Institutionof
Civil Engineers, 7, 305307.
Tanaka, Y. and Sakagami, T. (1989). Piezocone testing in
underconsolidated clay. Canadian Geotechnical J ournal,
26, 563567.
Terzaghi, K. andPeck, R.B. (1967). Soil MechanicsinEngi-
neeringPractice. A WileyInternational Edition, 729p.
300
A casestudyonsettlement predictionbyspatial-temporal
randomprocess
P. Rungbanaphan&Y. Honjo
I.Yoshida
Musashi Institute of Technology, Tokyo, Japan
ABSTRACT: A systematic procedurefor spatial-temporal predictionof settlement is proposed. Themethod
is based on Bayesian estimation by considering both prior information of thesettlement models parameters
and the observed settlement to search for the best estimates of the parameters. By taking into account the
spatial correlationstructure, all observationdatacanbeusedfor rational estimationof themodel parameters
at any location and any time. Thesystemerror can beconsidered by Kalman filter including process noise.
A proceduretoestimateauto-correlationdistanceof theparametersandtheobservation-model error basedon
themaximumlikelihoodmethodisalsoproposed. TheKrigingmethodisconsideredtobeasuitableapproach
for determiningthestatistics of theestimatedmodel parameters at any arbitrary location. A casestudy onthe
secondary compression settlement of thealluvial soil dueto preloading work is carried out. The y log(t)
method is chosen as a basic model for settlement prediction. It is concluded that, while the strong spatial
correlationisrequiredforsignificantimprovementof thesettlementpredictionbytakingintoaccountthespatial
correlationstructure, theproposedapproachgivestherational predictionof thesettlement at anarbitrarypoint
withquantifieduncertainty. Inaddition, includingprocessnoiseinthecalculationcanimprovetheestimation
but careshouldbetakentoassignanappropriatelevel of thissystemerror.
1 INTRODUCTION
So far, all methods of predicting future settlement
usingpast observations arebasedsolely on thetem-
poral dependenceof their quantity. However, thefact
that soil properties tend to exhibit a spatial corre-
lation structure has been clearly shown by several
studies in the past, e.g. Vanmark (1977), DeGroot
& Baecher (1993). It is therefore natural to expect
that theaccuracy of thesettlement predictioncanbe
improvedbytakingintoaccountthespatial correlation
of groundproperties,bywhichtheobservedsettlement
datafromall of thedifferentobservationpointscanbe
simultaneously utilized. Furthermore, by introducing
spatial correlation, it ispossibletoestimatethefuture
settlementof thegroundatanyarbitrarypointbycon-
sidering thespatial-temporal structure. This study is
actuallyanattempt tosearchfor suchanapproach.
2 SPATIAL-TEMPORAL UPDATINGAND
PREDICTINGPROCESS
2.1 Settlement prediction model
Thebasicmodel usedfor settlement predictioninthis
paper is thelinear relationshipbetweenlogarithmof
timeand thesettlement, i.e. y log(t) method. The
model isconsideredtoberational andpractical forpre-
dictionof thesecondarycompression(Bjerrum1967,
Garlanger 1972, Mesri et al. 1997, etc.). Theequation
isgivenby
wherey
k
=settlement at kth step of observation; m
0
andm
1
=constantparameters; t
k
=timeatkthstepof
observation;
k
=observationmodel error.
Thisimpliesthetemporallyindependentcharacteristic
of
k
.
2.2 Bayesian estimation considering spatial
correlation structure
In order to improve the estimation and to enable
local estimation, utilization of Bayesian estimation
considering spatial correlation is proposed in this
paper. This approach uses prior information of the
parametersandtheobservedsettlement datafromall
observation points to search for the best estimates
of the unknown parameters, i.e. model parameters
301
(m
1
and m
0
), auto-correlation distance (), and the
variance of the observation-model error (
2
). The
formulation consists of two statistical components,
namely, theobservationmodel andtheprior informa-
tionmodel. Thesetwomodelswill thenbecombined
byBayes theoremtoobtainthesolution.
2.2.1 Observation model
This model relates theobservationdatato themodel
parameters.Ataspecifictimestepk, letY
k
denotethe
observedsettlementatn observationpoints, x
1
, x
2
, ,
x
n
, where
Thestatevector, , is defined as theestimates of
model parameters(m
1
, m
0
) atthen observationpoints,
asfollows:
Consequently, themodel inEq. (1) canberewritten
inthefollowingmatrixform
where
I
n,n
denotes an n n unit matrix. is a Gaussian
observation-model error vector with E[]=0 and
E[
T
]=V
.V
isacovariancematrix,thecomponents
of whichare
It should be emphasized that
2
is both spatially
temporallyindependent.Furthermore,thiserrorisalso
defined as the combination of the observation error
and themodel error. Thesetwo kinds of errors can-
not beseparatedinpractice, andsoareassumedtobe
integratedinthemodel asshowninEq. (8).
Given and
2
, thepredictedsettlement distribu-
tionat anytimet canberepresentedbythefollowing
multivariatenormal distribution
2.2.2 Prior information model
By assuming two multivariate stochastic Gaussian
fields for m
1
and m
0
, the prior information has the
followingstructure:
where
m
1,0
(x
i
) and m
0,0
(x
i
) denotetheprior mean at obser-
vation point x
i
of m
1
and m
0
, respectively. is the
uncertainty of the prior mean with E[]=0 and
E[
T
]=V
,0
. V
,0
is a prior covariance matrix. By
introducing the spatial correlation structure in the
formulationof V
,0
, wehave
where0
n,n
denotesannxn zeromatrix.
2
m1,0
and
2
m0,0
representthepriorvarianceof m
1
andm
0
, respectively.
(|x
i
x
j
|) denotestheauto-correlationfunction. The
exponential typeauto-correlation function is chosen
for thecurrent study becauseit is commonly usedin
geotechnical applications(e.g. Vanmarcke1977). The
functionisgivenas
wherex
i
, x
j
=spatial vector coordinate, and =auto-
correlationdistance.
It shouldbenotedthat, for thesakeof simplifica-
tion, there are two important assumptions about the
correlationstructurefor formulatingtheabovecovari-
ance matrix. Firstly, m
1
and m
0
are assumed to be
independent of one another. Secondly, the correla-
tionstructures of thesetwo parameters areidentical,
meaning that they share the same auto-correlation
distance.
Given, priormeans, andpriorvariancesof m
1
and
m
0
, theprior distributionof themodel parameters is
alsoamultivariatenormal distributionof thefollowing
form
Itisclear fromthisformulationthatthespatial cor-
relationof soil propertiesisincludedintheformof the
spatial correlationof m
1
andm
0
.Thesettlementsthem-
selvesarenotcorrelatedspatially. Theauthorsbelieve
that this is the most suitable way to introduce the
spatial correlationstructureto thesettlement predic-
tionmodel sinceit issoil propertiesthat arespatially
correlated, not thesettlement.
302
2.2.3 Bayesian estimation
Suppose that the set of observations Y
k
at the time
t
k
for k =0, 1, . . . , K has already been obtained. By
employingBayes theorem, theposterior distribution
of thestatevector canbeformulatedas
where Y denotes the set of all observed data, i.e.
Y =(Y
1
, Y
2
, . . . , Y
K
). BysubstitutingEq. (9) and(15)
into theaboveequation, alikelihoodfunctioncanbe
defined, withthegivenvaluesof
2
and, asfollows:
TheBayesianestimator of , i.e.
, istheonethat
maximizestheabovefunction. Therefore, itisequiva-
lent tominimizingthefollowingobjectivefunction
By differentiatingtheaboveequationwithrespect
tothestatevector, weobtain
Bytrial anderror, thevaluesof
2
, , andthecor-
responding
that give the maximumvalue of the

likelihoodfunction(L) canbeobtained. Thesevalues
are actually the Bayesian estimators for the current
problem.
2.3 Process noise consideration by Kalman
filter method
Intheprevioussection, thebatchprocedurebywhich
all of theobservationsaretreatedequallyfor parame-
ters updating was proposed. In practice, it is natural
to give higher weight to the more recent observa-
tions. Thiscanbedonebyconsideringanuncertainty
parameter, socalledprocessnoise, throughasequen-
tial procedure, Kalmanfilter (Kalman1960, Kalman
& Bucy 1961, J azwinski 1976). The Kalman filter
has two distinct phases: time updating and observa-
tion updating. Thetime updating phaseusesthestate
estimatefromtheprevioustimesteptoproduceanesti-
mateof thestateatthecurrenttimestepbyconsidering
processnoise.Intheobservation updating phase,mea-
surementinformationatthecurrenttimestepisusedto
refinethispredictioninorder toarriveat anewstate
estimate. In fact, it can beproved that, without pro-
cess noise, the Kalman filter gives the same results
as the previously proposed approach, i.e. Bayesian
estimation(Hoshiya&Yoshida1996and1998).
For the estimation of the soil parameters, the
unknown parameters is considered to be stationary,
thentheTime updating processisexpressedby
whereQ
k
denotescovariancematrixof processnoise;
suffixk standsfor thekthstepof processing; k/k 1
represents the kth step estimation conditioned on
processingobservationdatauptothek 1thdatum
Tosystematicallydefinethevalueof processnoise,
we assume that Q
k
are given based on a priori
information, asfollows(Hoshiya&Yoshida, 1998):
wherec isaconstant parameter representingthelevel
of systemerror, i.e. processnoise. From(22) and(23),
wehave
For Observation updating process, theupdatingis
givenby
bydefiningtheKalmangain
wheresuffixk/k, similarly, representsthekthstepesti-
mationconditionedonprocessingobservationdataup
to the kth datum. It should be emphasized that
0,0
and V
,0,0
need to bedefined in thesameway as
0
and V
,0
in Eq. (11) and (13), in order to take into
accountthepriorinformationof themodel parameters
together withthespatial correlationstructure.
Withitsabilitytosequentiallyupdatetheestimation
andsystematicallytakeintoaccountprocessnoise,this
methodis usedinestimationof theunknownparam-
eter, , while
2
and are estimated by Bayesian

estimationdescribedinSection2.2
303
2.4 Local estimation by the Kriging Method
Based on the calculated statistical inferences of
the model parameters at the observation points and
theestimated auto-correlation distance, thestatistics
of the model parameters at any arbitrary locations
can be determined by the ordinary Kriging method
(Krige1966,Matheron1973,Wackernagel 1998).This
methodprovidesanunbiasedandleasterror estimator
builtonthedatafromarandomfield.Itisalsoassumed
thattherandomfieldissecond-orderstationary. Based
ontheestimatedmodel parameters (m
1
, m
0
) at then
observationpointsx
1
, , x
n
, thevalueof m
1
andm
0
, at
anarbitrarypointx
0
canbeestimatedbythefollowing
equations:
where
w
i
(i =1, , n) aretheweightsattachedtothedataat
eachof theobservationpoints. jistheLagrangemul-
tiplier usedfor minimizingtheKrigingerror, andx
0
denotesthespatial vector coordinateat x
0
. |x
i
x
j
|)
representstheauto-correlationfunctionasdefinedin
Eq. (14).
3 CASE STUDY
3.1 Description of the case
The site is a residential land development project
locatedinsuburbareaof Tokyo, J apan. This areais
coveredbyathickalluvial depositwhichcanbeclassi-
fiedasasurfacelayer of peatsfollowedbyaverysoft
clay layer down to thethickness of about 17 meters
(Fig. 1). Below these layers, the layers of medium
densesandandsiltarefound.Inordertoavoidthelarge
amount of settlement duetothethicksoft soil layer at
thesurface, thesoil conditionisimprovedbypreload-
ing priortotheconstruction.AsshowninFigure2, the
preloading surchargewas filled up to themaximum
thickness of about 6mduring thepreloading period
of, approximately, 900days.
Thesettlementobservationswereperformedatboth
duringthepreloadingperiodby thesettlement plates
andafter removal of thesurchargeby measuringset-
tlementof theboundarystonearoundthehousinglots.
Thesettlementafterremoval of thesurcharge, whichis
usedinthisstudy, wasobservedataboutevery600m
2
withthetotal number of observationpointsof 42. The
locationplanof theseobservationpoints is shownin
Figure3, whileall of theobservationdataareshown
Figure1. Soil condition.
Figure2. Surchargethicknessandsettlement vs. time.
Figure 3. Location plan of the observation points and
surchargearea.
as semi-logarithmic plots of settlement and time in
Figure4.
Various techniques have been proposed by sev-
eral authors for predicting the future settlement
using the observed settlement, for example, hyper-
bola method (Sridharan et. al. 1987, Tan 1994),
y log(t) method(Bjerrum1967, Mesri etal. 1997),
and Asaokas method (Asaoka 1978). In this study,
304
Figure 4. Observed settlement vs. time (after surcharge
removal) for all observationpoints.
Figure 5. Observed settlement vs. time (after surcharge
removal) andtrendlineat pointA (seeFigure3).
y log(t) methodis consideredto bethemost suit-
able approach due to the fact that the primary con-
solidation is expected to be completed before the
surchargeremoval, thusthesettlementoccurringafter-
ward should result fromthe secondary compression
process. Figure5showsanexampleof they log(t)
plot at an observation point. It can be seen that,
by excluding a part of data in the early period of
observation, within which the secondary compres-
sion is considered to be influenced by the rebound
effectduetosurchargeremoval, thissemi-logarithmic
relationshipfitsquitewell withtheobservationdata.
3.2 Practical problems and solutions
To deal with thefield observation result, theincom-
pleteness of the data is, of cause, unavoidable. The
example of the observed data shown in Figure 5 is
illustrated arelatively completeset of data, but it is
not always the case. Several observation points suf-
fer themissingof settlementdataatsomeobservation
steps.Theproblemisthat,forbothBayesianestimation
andKamanfilter, thedatafromall of theobservation
pointsarerequiredateverytimesteps. Itisclear form
Eq. (17), (20) and(25) thateverycomponentsof Y
k
is
neededfor every timestepsk =1toK. Tocopewith
thisproblem, thecomponentsof V
whichcorrespond
totheobservationpointswiththemissingdataat that
timesteparereplacedbytheextremelyhighnumbers.
Bythistreat, themissingdatacanbeinitiallyassumed
withinareasonablerangeandit will thenbeignored
fromthecalculationbytheinfluenceof matrixV
.
As can be seen fromFigure 5, the early part of
theobserveddataclearlyshowsdisagreementwiththe
previouslydescribedy log(t) model. Thisisdueto
the fact that the settlement observed in this period,
whichis expectedto result fromthesecondary com-
pression, wasstill stronglyinfluencedbytherebound
effect fromthe removal of preloading surcharge. In
order to implement themodel to thesesets of obser-
vationdata, apart of thedataneedtobeignored. By
investigatingthesettlementdataof all theobservation
point, thedatabeforetheday103tharediscardedfrom
thecalculationbyjudgment.
Choosing appropriate prior statistics of the
unknown parameters (m
1
and m
0
) is also an impor-
tant issue. What hasbeendoneinthecurrent research
is that theprior meanof m
1
andm
0
wereassumedto
beequal tothevalueof slopeandtheintercept of the
trendlineresultingfromthelinear regressionanaly-
sisof settlementvs. timeplottingconsideringthedata
fromall of theobservationpoints. Ontheother hand,
thepriorvariancesareselectedbytryingseveral values
of prior coefficient of variation(COV) andchoosing
theonewhichisrelativelyinsensitivetothechangesof
prior means. Basedonthisapproach, theprior means
of m
1
and m
0
, which are assumed to be identical at
everyobservationpoints, areassignedas109.7cmand
204.1cm, respectively, whiletheprior COV issetas
0.4for calculatingprior varianceof bothparameters.
3.3 Estimation of the auto-correlation distance and
observation-model error
It wasproposedinSection2.2.3that auto-correlation
distance () and the standard deviation of the
observation-model error (
) canbeestimatedby an
optimizationprocedurebasedonBayesianestimation.
Considering the observation data together with the
prior information of themodel parameters, thelike-
lihood values (L) for each pairs of and
can be
determined by Eq. (17). Thevalues of and
that
give the maximumvalue of L will be served as the
Bayesianestimatorsof theseparameters.
Figure6showsthecontour of L in and
space
for the case that all of the settlement data until the
last step of observation, i.e. the day 1017th, is con-
sidered. Inthiscase, theBayesianestimatorsof and
are32mand6.75cm, respectively. Obviously, the

estimatedvaluesof aremorelikelytobechangeable
thanthoseof
.
Inpractice, theobservationdatais collectedstep-
wise for a period of time. Therefore, it is natural
to sequentially update the estimation once the new
sets of observation areprovided. Figure7 illustrates
the plots of the estimated values of and
vs.
observation time, until which the observation data
are used in estimation. It can be observed that the
estimatedvalues of auto-correlation distancetendto
decreasewiththeobservationtime, whilethoseof the
305
Figure6. Contour of likelihoodvalues(L), usingobserva-
tiondatauntil thelast stepof observation(theday1017th).
Figure7. Estimatedvaluesof auto-correlationdistance()
and the standard deviation of the observation-model error
(
) vs. observationtime.
observation-model error tend to increase, depending
on the characteristic of the observed data. Both of
theseestimations seemunstableat theearly stageof
theobservation, indicatinginsufficiencyof theobser-
vationdataforthecalculations. However, theybecome
morestableastheobservationdataaccumulates.
3.4 Settlement prediction and estimation
BasedontheprocedureproposedinSection2.2, the
estimates of model parameters at each observation
Figure 8. mean absolute error of settlement prediction at
thelastobservationtimestep(theday1017th) vs. observation
time.
pointscanbecalculated,consideringpriorinformation
of theparameters and observation data. Using these
parameters, thesettlement at anyspecifictimecanbe
estimatedbyy log(t) model. Thiscanbecompared
with the observed data at this point to discover the
estimationerror.
For quantitatively describing theestimation error,
thetermmeanabsoluteerror isdefinedasfollows:
whereX
est,i
andX
true,i
denotetheestimatedvalueand
true value, respectively, of the parameter to be esti-
matedateachobservationpoint.N
x
representsthetotal
numberof estimatedvalues. Inthiscase, theestimated
valueistheestimatedsettlement, whilethetruevalue
is theobservedsettlement. N
x
is thetotal number of
observationpoints, i.e. N
x
=n.
Figure8showstheplotsof themeanabsoluteerror
forpredictionof settlementatthelastobservationtime
step(theday 1017th) vs. observationtime. For com-
parisonpurpose, thecasethatthespatial correlationis
ignoredisalsopresented.Forthiscase,theobservation
dataateachpointareusedtoupdatethemodel param-
etersof that point itself, i.e. =0. It shouldbenoted
that, for thecasewithconsideringspatial correlation,
the estimated values of auto-correlation distance, as
showninFigure7, areusedinthecalculations.
Asmightbeexpected,thepredictionerrordecreases
with the increase of the available observation data.
However, for thecurrent set of observationdata, con-
sidering the spatial correlation do not significantly
improvetheestimationintermsof themeanerror.This
may bedueto thefact that theauto-correlation dis-
tanceisrelativelyshortincomparisonwiththespacing
betweentheobservationpoints.
Tofurtherinvestigatetheefficiencyof theproposed
method in dealing with the space-time problem, the
observation data at each selected observation point
are removed and the settlement estimations, or pre-
dictions, at this point are performed using the rest
of the observations. Firstly, the estimates of model
306
Figure 9. Comparison between the estimated and the
observed settlement at the day 430th, using data fromthe
day103thto403th.
parameters at eachobservationpoints arecalculated.
Then, theparametersattheremovedobservationpoint
are determined by Kriging method using the esti-
mated auto-correlation distance (see Section 2.4).
Comparisonbetweenthesettlementestimatedbythese
parameters andtheactually observedonereveals the
estimationerror.
Figure 9 shows the comparison between the esti-
matedandtheobservedsettlement of all 42observa-
tionpointsattheday430th.Theobservationdatafrom
theday 103thtotheday 430thareusedinthecalcu-
lation. Obviously, for thecasethat therelativelyweak
spatial correlationisassumed, i.e. assuming =10m
(Fig. 9a), theestimationtendstobeuniformandisnot
likely to beableto represent thevariationof ground
settlement. Ontheotherhand, forthecasethattheesti-
matedvalueof auto-correlationdistance, =52m, is
used (Fig. 9b), the estimation gives relatively more
realistic pattern of settlement in the area. It should
benotedthat theobservationdataat somepoints are
missingduetothecommonimperfectionof thefield
observation. Only theestimated values areshown at
thesepoints. Infact, thisillustratesoneof thepracti-
cal advantagesof theproposedmethod, i.e. theability
toperformestimationat aspecific point without any
observedinformation.
Figure10 shows theplots of mean absoluteerror
(Eq. (30)) of settlement estimation at the removed
observationpointsvs. observationtime. Twocasesof
calculationsarepresented: CaseA andB. CaseA isan
attempttoavoidthetemporal errorresultingfrompre-
diction of futuresettlement; therefore, theestimated
settlement andtheobservedsettlement arecompared
at that observationtime, i.e. thesameway asthecal-
culationsshowninFigure9. CaseBisthecomparison
betweenthepredictedsettlementatthelastobservation
time step (the day 1017th) and the observed settle-
ment at that time. It should be emphasized that the
estimatedvalues of auto-correlationdistanceandthe
observation-model error, which are varied with the
Figure10. Meanabsoluteerror of settlement estimationat
theremoved observation points vs. observation time. Case
A presents the settlement observation at that day. Case B
presentsthesettlement predictionat theday1017th.
observation timeas shown in Figure7, areused for
all calculations.
Itcanbeseenthat, forthebothcases, theestimation
errorsaredecreasedwiththeobservationtime.Inother
words, theestimationcanbeimprovedif moreobser-
vationdataaregiven. However, CaseBgivesthelower
estimationerror, eventhoughthefuturepredictionis
performedinthis case. This may betheresult of the
cancellationof error duringthesettlement prediction.
3.5 Effect of process noise consideration
As mentioned in Section 2.3, higher weight should
begivento themoreresent datathanto theprevious
one. This can bedoneby assigning appropriateval-
uesof processnoise, i.e. systemerror, duringthetime
updatingprocessinKalmanfilter procedure. Accord-
ingto Eq. (23), thelevel of this systemerror canbe
controlled by assigning an appropriate value of the
constant parameter, c.
By performing the same calculation as what is
showninFigure8, but withthedifferent valuesof c,
theeffectof processnoisetothesettlementprediction
canbeinvestigated. Figure11presentstheprediction
error of settlement at the last observation time step
(theday1017th) atdifferentobservationtime. Infact,
for thecasethatc =0, i.e. noprocessnoise, thisisthe
samewiththeoneplottedinFigure8, for considering
spatial correlationcase.
It can beconcluded fromFigure11 that, to some
extend, thesettlement predictioncanbeimprovedby
considering the systemerror. Especially for estima-
tion at last time step (the day 1017th), at which all
observation data until the target day for the predic-
tionareincludeinthecalculations, themeanabsolute
error reducedramatically fromabout 9.8%to 0.3%.
However, this is not always the case. At some stage
of prediction, assuming too high value of process
noisemaymisleadthepredictionandtheerrorbecome
higher instead. This can be seen in Figure 11 when
c =2.0 are assigned. Therefore, the optimization of
this process noise coefficient is required. This is,
however, out of scopeof thecurrent research.
307
Figure 11. Mean absolute error for prediction of settle-
ment at thelast observation timestep (theday 1017th) vs.
observationtimeunder different levelsof processnoise.
4 CONCLUSION
A methodology was presented for observation based
settlementpredictionwithconsiderationof spatial cor-
relationstructure.Thespatial correlationisintroduced
among the model parameters and the settlements at
various points are spatially correlated through these
parameters,whichnaturallydescribethephenomenon.
A casestudyonthesecondarycompressionof allu-
vial depositsduetogroundimprovement by preload-
ing was carried out using the proposed approach. It
wasfoundthat theestimationof auto-correlationdis-
tanceisrelativelyunstableandinsufficientamountsof
observationdatamaymisleadtheestimation. Further-
more, even though it seems that theauto-correlation
distance of the soil parameter is relatively short in
comparisonwiththeobservationpoints spacing, the
proposed method provides therational estimation of
the settlement at any time and any location with
quantifiederror.
Includingsystemerror, i.e. process noise, into the
calculationcangivetheimprovementonthesettlement
estimation to someextend. However, careshould be
taken in assigning appropriatevalueof this parame-
ter toavoidadditional error duetoincludingtoohigh
valueof processnoise.
REFERENCES
Asaoka, A. 1978. Observational procedure of settlement
prediction. Soil and Foundations 18(4): 87101.
Bjerrum, L. 1967. Engineering geology of Norwegian nor-
mally consolidatedmarineclays as relatedto settlement
of buildings. Gotechnique 17(2): 81118.
DeGroot, D. J. & Baecher, G. B. 1993. Estimating autoco-
varianceof in-situsoil properties.Journal of Geotechnical
Garlanger, J.E. 1972. The consolidation of soils exhibiting
creepunderconstanteffectivestress. Gotechnique 22(1):
7178.
Hoshiya, M. & Yoshida, I. 1996. Identification of condi-
tional stochastic gaussian field. Journal of Engineering
Mechanics, ASCE 122(2): 101108.
Hoshiya, M. &Yoshida, I. 1998. Processnoiseandoptimum
observation in conditional stochastic fields. Journal of
Engineering Mechanics, ASCE 124(12): 13251330.
Krige, D. G. 1966. Twodimensional weightedmovingaver-
agingtrendsurfacesfororeevaluation. Proc., of Symp. on
Math., Statistics and Comp. Appl. for Ore Evaluation.
Matheron, G. 1973. Theintrinsicrandomfunctionsandtheir
applications. Adv. in Appl. Probab. 5.
Mesri, G. et al. 1997. Secondary compression of peat with
or without surcharging. Journal of the Geotechnical
Engineering Division, ASCE 123(5): 411421.
Sridharan, A., Murthy, N. S. & Prakash, K. 1987. Rect-
angular hyperbolar method of consolidation analysis.
Gotechnique 37(3): 355368.
Tan, S. A. 1994. Hyperbolicmethodfor settlementsinclays
withvertical drains. Canadian Geotechnical Journal 31:
125131.
Vanmarcke, E. H. 1977. Probabilistic modelingof soil pro-
files. Journal of the Geotechnical Engineering Division,
ASCE 103(GT11): 12271246.
Wackernagel, H. 1998. Multivariate Geostatistics: An Intro-
duction with Applications. 2nd ed. Germany: Springer-
VerlagBerlinHeidelberg.
308
Construction risk management
Reliabilityanalysisof ahydraulicfill slopewithrespect to
liquefactionandbreaching
T. Schweckendiek
Deltares, unit Geo-engineering &TU Delft, Delft, The Netherlands
G.A. vandenHam& M.B. deGroot
Deltares, unit Geo-engineering, Delft, The Netherlands
J.G. deGijt & H. Brassinga
Public Works Rotterdam, Rotterdam, The Netherlands
P. Hudig
Gate Terminal B.V.
ABSTRACT: A recently reclaimedsiteinthePort of Rotterdamwill serveas locationandfoundationof an
LNG terminal. LNG (Liquefied Natural Gas) is recognized as hazardous material and underlies strict safety
requirements. Aspart of thesafetyassessment of theentireinstallation, aspecificanalysishadtobecarriedout
concerningthegeotechnical aspects. Thepaper describes theprobabilistic approachthat was chosentoverify
therequiredlevel of safetyof thehydraulicsandfill regarding(static) liquefaction, slopefailureandbreaching
processes. Several reliability analyses using therespectivephysical process models werecarried out and the
resultscombinedusingafault treeor scenarioapproach, leadingtoupper boundsof thefailureprobability.
1 INTRODUCTION
1.1 Project outline
The paper describes an approach for a geotechnical
reliability analysis problemin a real life project in
2007. For anLNG terminal to bebuilt inthePort of
Rotterdam, hydraulic sandfillingwasusedtoextend
anexistingartificial terraininordertocreatespacefor
4largeLNGtanks(Fig. 1).
Figure1. OverviewLNGterminal.
The original design contained slopes with angles
of 1:2.5, protected mainly against erosion and wave
actionbysteel slagdams. Initially, it wasthought that
compactionwouldnot berequireduntil aroughanal-
ysisof theliquefactionpotential cast thisassumption
into serious doubt. Subsequent more thorough anal-
yses led to several design modifications, the most
important of which were to use a shallower slope
angle of 1:3 and to compact the entire slope itself
upto theheight of thetanks to bebuilt by means of
vibro-flotation. Thefinal representativecrosssection
inFigure2.
Notethat dueto theconstruction process and the
last-minutedesignamendments, someportionsof the
hydraulic fill could not becompacted and remained
in a relatively loose state. These areas in combina-
tionwiththestill relatively steepslopecausedsome
uncertainty about the chance of the occurrence of a
liquefactionflowslidewithsubsequentdamagetothe
foundationof theLNG-tanks.Thisuncertaintywasthe
focusof theanalysisdescribedinthispaper.
1.2 Design requirements
For theLNG installation, as for other activities with
hazardousmaterials, thesafetyrequirementswerefor-
mulated in terms of risk respectively an acceptable
probability of failure, failurebeing defined as some
theoccurrenceof anunwantedevent or accident. The
311
Figure2. Representativecross-section.
safety criterion for the geotechnical aspects treated
in this paper was derived from the overall safety
requirement, being: The probability of a slope fail-
ure, including liquefaction and breaching, affecting
the foundation safety of the LNG-tanks must not exceed
P
f ,adm
=10
6
in the planned life time of the structure
(50 years). Notethat this criterion involves several
potential failuremechanisms.
1.3 Probabilistic approach
Fortheevaluationof theprobabilityof failurestatedin
theprevious section, thecomplex failuremechanism
wassplit intobasically threesub-mechanisms, which
weretractablefor structural reliabilityanalysis.
A choice for a dominant failure scenario did not
seemappropriate, mainlyduetothefact that multiple
failuremechanismswereinvolved. Tothisendseveral
failurescenariosweredefinedinordertoensuretonot
misssignificant contributions.
Theresultsof thesub-mechanismsandthescenarios
werecombinedbymeansof faulttreeanalysisinorder
toobtainthe(upperboundof the)overall probabilityof
failure, whichwasthencomparedtotheacceptability
criterion.
Section2treatsthephysical processmodelsapplied
intheanalysis, whilst section3focuses ontherelia-
bilityanalysisaspects.
2 APPLIEDPHYSICAL PROCESSMODELS
2.1 Liquefaction flow slide and subsequent
breaching
Undercertaincircumstances,loose,saturatedsandele-
ments in aslopemay besensitiveto liquefaction or,
moreprecisely formulated, may beinameta-stable
state, which means that they will liquefy and loose
their strengthunder anyquick loadingif theyarefree
to undergo shear deformation. Incasemost adjacent
sand elements in a slope have a much more stable
state, no liquefaction will occur becausethesemore
stable elements will prevent the shear deformation
of their meta-stable neighbors. However, in a slope
withsufficientlargepocketsof meta-stableelementsa
liquefactionflowslidemay occur. Theconditionsfor
meta-stability mainly concern thesoil statein terms
of density and stresses which will be discussed in
section 2.2. Whether thepockets of meta-stableele-
ments are sufficiently large to enable a liquefaction
flowslidewill bestudiedbyatraditional slopestability
analysisinwhichtheoriginally meta-stableelements
aresupposedtohaveliquefied(section2.3). Thefinal
questioniswhetheraliquefactionflowslidewill result
infailureof thefoundationof thetanks. Incaseof a
relativelyshallowflowslide, thiswill onlybethecase
if abreachintheunprotectedsandcreatedbytheflow
slidewill progressoverasufficientlargedistance.The
breachingprocesswill bediscussedinsection2.4.
2.2 Meta-stability or sensitivity to liquefaction
Themodel thatwasusedinthisstudyfortheundrained
behavior of saturated(loose) sandisbasedonthethe-
orypresentedinStoutjesdijketal (1998), whichisalso
thebasictheoryusedinthesoftwareSLIQ2D, mainly
used by GeoDelft in theNetherlands during thelast
twodecades.
Whilst SLIQ2D only uses an instability or meta-
stability criterion based on material parameters and
the soil state (porosity and stresses) according to
Molenkamp (1989), the approach in this study uses
moreinformationfromthemodeledundrainedbehavi-
ousrespectivelythestresspath.
For agivenin-situstresspoint, theundrainedstress
pathis derivedas afunctionof relativedensity from
extensive laboratory tests. This path allows us two
extracttwotypesof informationthathelpsustojudge
the liquefaction potential and the residual strength
after liquefaction:
1 whether thein-situdensity ishigher or lower than
thewet critical density(WCD, seeFigure5).
If I
D
-WCD, the undrained stress path exhibits
a decreasing deviatoric or shear stress. This is
the most important necessary, however not suffi-
cient, conditionfor meta-stability andthusfor the
occurrenceof instabilityandstaticliquefaction.
2 the maximum generated excess pore pressure
respectivelytheminimumisotropiceffectivestress
312
p
min
, which can be used to estimate the (worst
case) strengthreductionduetoliquefaction.
Both definitions are definitely conservative respec-
tivelywill leadtoupper limitsof failureprobabilities.
Wewill comebacktothisquestioninsection3.
2.3 Slope stability
Theslopestabilitywastreatedbyconventional Bishop
slip circle analyses using the MStab software by
GeoDelft (since 2008 Deltares). Two non-standard
featureshadtobeincluded:
1 Theslopestability analysis had to reflect thesit-
uation, given that liquefaction occurred in the
liquefaction-sensitive parts of the slope. In the
deterministicsetup,thereductioninisotropiceffec-
tivestress was used as measurefor thereduction
of shear capacity, expressedinformof areduced
frictionangle:
2 The Rotterdamarea is not typically earthquake-
prone, however, due to the low required failure
probability, also very low occurrence frequency
seismicloadswereconsidered.AnoptioninMStab
to account for vertical and horizontal peak accel-
erationsintheslopestability analysiswasapplied
(Delft GeoSystems2006).
2.4 Breaching
If slopeinstabilityoccurs,aliquefactionflowslidewill
start, whichmeansthat theinstablesoil massstartsto
slideoverashearsurface. Itwill continuetodosountil
itfindsanewequilibrium.Theflowprocesswill inthis
caseprobablytakenot morethanseveral secondstoa
minute, as follows fromcalculations inwhichinertia
isincorporated. That timeisnot longenoughtocause
significant reduction of the excess pore pressure in
the liquefied sand pockets. Consequently, the shape
of thenewprofilecanbeestimatedby usingBishop
calculationsandthenewslopeprofileischaracterized
byarelativelysteepslopejustabovethesoil mass, that
floweddown. Its locationcanbecharacterizedby L
1
asdefinedinFigure3.
Thissteepslopeconsistsof sandandisnotlikelyto
becoveredbyslagsor other partsof anyslopeprotec-
tion. Part of thesteepslopeissituatedunder water, as
indicatedinFigure3. Thispart of theslopemaystart
breaching.
Breachingisaprocessinwhichasteepunder water
slope, breach, remains temporary stableunder the
influenceof dilationinducednegativeporepressures,
andgraduallymovesbackwardswhilesandgrainsfall
down fromthesurfaceand mix with water to create
aneroding, turbulentsandwater mixture. Theprocess
stops when theheight of theunder water part of the
breach is reduced to zero. The resulting profile is
sketchedinFigure6.
Figure3. Equilibriumprofileafter flowslide.
Figure4. Undrainedstresspathof loosesand.
Figure5. DefinitionWet Critical Density(WCD).
Figure6. Equilibriumprofileafter breachingprocess.
ThebreachingprocessisdescribedbyMastbergen&
van den Berg (2003) and can be modelled by the
computer codeHMBREACH. Givengrainsizedistri-
bution, relativedensityandinitial height of theunder
313
waterpartof thesteepslope, sbh, themodel calculates
thechangeinthisheight asafunctionof thehorizon-
tal distance, fromwhich thetotal distanceof breach
progress L
2
(Fig. 6) canbederived. Theslopeof the
part above the water is determined by the common
shearingprocess andcanbeassumedto equal 1:1.5.
Nowthelength(L
2
L
1
) of thedamagedareafollows.
It isassumed, supportedbyindicativecalculations,
that no significant damage to the foundation of the
tankswill occur aslongas(L
2
L
1
)-22.5m, which
is thedistancebetweenthefoundationandtheslope
crest.
Theprevious section gaveaconciseoverviewof the
concepts and methods used for deterministic evalu-
ation of the sub-mechanisms playing a role in the
present safetyassessment problem. Inthissectionwe
will discusshowanassessment of thecriterionstated
insection1.2wasmadeinaprobabilisticmanner.
First of all, wearedealingwiththeverificationof
adesigncriterion. That implies that it is sufficient to
showthattheupperboundof theestimateof thefailure
probabilityP
f ,sup
fulfillstherequirement:
Thus, we can start with rough, conservative (upper
bound) approaches and apply refinements, if neces-
sary, as illustratedinFigure7. Suchrefinements can
either concern the probabilistic analysis itself (e.g.
treatment of correlations) or more realistic physical
processmodels.
Suchanapproachwasappliedintheproject, though
forsakeof readabilityinthefollowingonlytheanalysis
that ledtothesuccessful outcomeisdescribed.
3.1 System definition
Asdescribedin2.1, theprincipal contemplatedfailure
modeisasequenceof threemechanisms.
To reiterate the sequence shortly, liquefaction of
substantial, uncompacted volumes in the slope part
of thefill may causeaflow sliderespectively slope
failure. The residual profile is common steep in the
upper part and abreaching process may beinitiated
thatcouldendanger thefoundationsof theinstallation
inquestion.
Forthereliabilityanalysis,thissequenceismodeled
byaparallel sub-systeminafaulttree, consequently
combinedbyandAND-gate(Fig. 10).
Given the large uncertainties, it is not trivial to
determineadominantorrepresentativescenarioaswe
areusedtodoindeterministicapproaches. For differ-
entcombinationsof parametersor properties, insome
cases liquefactionandslopefailureintheupper part
may lead to the worst consequences, in other cases
failures in thelower part or deeper sliding surfaces.
Figure 7. Upper and lower bounds of P
f
vs. the design
criterion.
Figure8. Sequenceof mechanismsinFailuremode.
Figure9. Sequenceof mechanismsleadingtotopevent.
Figure10. Sequenceof mechanismsinFailuremode.
Figure11. Schematicrepresentationof twoscenarios.
Oneway to circumvent theproblemof choosingone
scenario, isthedefinitionof several scenarios.
Two examples of such scenarios are presented
schematicallyinFigure11.Themaindifferenceinthis
314
Figure12. Fault tree.
discretedistinction of possibilities is theassumption
of whichof theuncompactedvolumesliquefyandhow
manyat atime, withall thedueconsequences.
All thedefined scenarios areintegrated in afault
tree(Fig. 12). For sakesimplicity, theconservative,
i.e. upperboundassumptionof independence(actually
evenmutuallyexclusivity) ismade(see3.6).
3.2 Parameters and uncertainties
Thein-siturelativedensitiesof thehydraulicfill were
determined by means of theempirical CPT correla-
tionfunctionof Baldi e.a. (1982) whichcorrelatesthe
densityindexI
D
totheconepenetrationvalueq
c
asa
functionof thevertical effectivestress. A total of over
50CPTs wereavailable. Accountingfor bothspatial
variabilityanduncertaintyof thecorrelationfunction
theexpectedvalueof I
D
wasfoundtobe39%witha
standarddeviationof 10%. Thesevalues concernthe
averageof I
D
overapotential liquefiableareaorfailure
surface.
By means of several drained (CD) and dry triax-
ial tests on a number of representative (disturbed)
samples, takenfromthehydraulicfill, theparameters
for theconstitutivemodel (see2.2) weredetermined.
Influenceof soil statewasassessedbyperformingthe
testsat different stressconditionsandporosities. Sta-
tistical analysis of thetest results andconsiderations
on spatial variability, lead to probability distribution
functionsof theimportant material model parameters
for further useintheprobabilisticanalysis.
In order to check the calibrated parameter set, a
number of undrained(CU) triaxial testswasexecuted
on thesamesamples and simulated with themodel.
Measurements and prediction fitted reasonably well
(Fig. 13).
3.3 Meta-stability or sensitivity to liquefaction
Theprobability of meta-stability or thesensitivity to
liquefactionP
liq
of eachareawithnon-compactedsand
was evaluated by determining the probability of the
in-situsandbeinginastatebelowtheWCD(see2.2),
Figure13. Comparisonstress path(CU) betweentest and
calibratedmodel.
givenarepresentativestresspoint intheareaandthe
uncertaintiesinthematerial properties:
withx beingavector containingall randomvariables.
P
liq
was determined by means of Monte-Carlo anal-
ysis. Per scenario, n =10
5
realizations of the state,
material and model parameters were produced and
propagatedthroughthemodel (undrainedstresspath,
Fig. 4). Consequentlytheestimator for P
liq
is:
where x
i
is the ith realization of x and I
C
(x) is the
indicator functionfor conditionC.
Consideringthedefinitionof WCD, beinganeces-
sarynotsufficientconditionforstaticliquefaction,this
isclearlyaconservativeapproachleadingtoanupper
bound estimateof theprobability of liquefaction. In
fact, the results in section 4 show that the estimate
basedonthismethodusually leadtovery highprob-
abilitiesthat intuitivelydonot reflect thejudgment of
most experts. For theassessment of theprobabilityof
sensitivity toliquefaction, it isdefinitely desirableto
315
Figure14. MStabreliabilitymodule.
Table1. Peakaccelerationvalues.
a
max
P{a
max
>
max
} P{a
max
>
max
}
[m/s
2
] [1/year] [1/50year]
0.20 1/475 1-(1-1/475)50=0.1
0.40 1/10000 1-(1-1/10000)50=0.005
useanapproachthatincludesalsothedistancefrom
instabilityor acritical-statemodel. Thiswasnot real-
izedinthecourseof thisproject, butisoneof ourgoals
for thefuture.
Itisalsonotedthatseismicactionwasneglectedin
thisstep. Duetotheverylowintensitythecontribution
wasfoundtobeinsignificant.
3.4 Slope stability, given liquefaction
The second step respectively sub-mechanismin the
contemplated chain of events is slope failure, given
liquefactionhasoccurredinoneor moreof theprob-
lematicuncompactedzones.A total of 6critical failure
modescouldbeidentified.
The slope reliability analysis is carried out using
thereliability moduleof MStab, whichis essentially
FORM appliedto aBishopslipcircleanalysis using
average properties of the soil shear resistance prop-
erties as themain basic randomvariables, thus with
implicit treatment of averaging effects in theproba-
bility distributionsfor theshear resistance(seeJ CSS
2001).
Asmentionedearlier, seismicloadingwasnotcon-
sideredintheinitiationof liquefaction, i.e. theimplicit
assumptionisthatatriggerisalwayspresentwithhigh
probability. However, seismic action was taken into
account in the slope stability analysis. For the con-
sideredarea, twovaluesof peak accelerationa
max
are
givenfor thereturnperiods of 10,000years and475
years(seeTable1). Inordernottousetheheaviestcon-
ditionas deterministic value, aGeneralizedExtreme
Value(GEV) distributioncorrespondingto thegiven
quantileswasusedtointegratetheseismic loadsina
probabilisticmanner.
The resulting GEV-distribution is shown in
Figure15.
Figure15. GEV distributionof a
max
.
Sincetheusedsoftwaredidnot allowustoinclude
theuncertaintyina
max
intheBishop-FORM analysis,
several of theseformanalyseswerecarriedoutforaset
of deterministicvaluesof thepeak acceleration. Sub-
sequently, theresults interms of thereliability index
, conditional ona
max
, canbeintegratednumerically
tosolvethefollowingintegral:
This is practically done by an external FORM-loop
respectively design point search, for details refer to
(Delft GeoSystems2006).
3.5 Breaching, given slope failure
By carryingout anuncertainty analysisontheinitial
breach height sbh and thevalueof L
1
, based on the
uncertainties in thestrength of liquefied sand (
red
)
andthestrengthof thenon-liquefiedand(critical state)
probability distribution functions for these variables
wereestablished. Thebreach length L
2
proved to be
veryinsensitivetoL
1
, reasontogiveit aconservative
deterministicvalue: L
1
=5m(againsimplifiedupper
boundapproach). Theuncertainty insbh, however, is
expressedasalognormal distributionwithanexpected
valueof 1mandastandarddeviationof 1m.
Theresults of alargeseries of HMBREACH cal-
culations could be approximated by the following
equation(responsesurface):
where C1 and C2 are model parameters with log-
normal distributions, expectedvalues 1andstandard
deviations0.1and0.3, respectively.
A reliabilityanalysisonthisresponsesurfaceof the
breachmodel resultedin:
P{(L
2
L
1
) > 22.5m|slopeinstability) = 1.310
7
316
andanexpectedvalueof E(L
2
)=7.8mandastandard
deviation(L
2
)=3.7m.
It should be noticed that the applied models for
thebreachingprocess, givenslopeinstability, arevery
rough. Evenconservativeassumptions, however, make
clearthatnolargedamageistobeexpectedhereinthe
unlikelycasedthatslopeinstabilityoccurs.Thisisdue
to theshallowlocationof theuncompactedareas. In
other cases of liquefaction slopefailures, thelength
L
2
L
1
of thedamagedareamay reachvalues of up
to100morevenmore, asexperienceshows. Research
inthefieldof thebreachingprocess andtheinterac-
tionbetweenliquefactionandbreachingis neededto
improve the models and develop a practical tool to
predict thelengthL
2
L
1
of thedamagedarea.
3.6 Total failure probability
As mentioned earlier, but emphasized again at this
point, theresultspresentedhereintermsof thefailure
probabilityconcernanupperbound. Bydefinition, the
valueof thisprobabilityisexpectedtobelower. Vari-
ousassumptionshaveledtoavalueonthesafeside.
These assumptions can be roughly classified in two
categories:
1. Assumptionsinprobabilisticapproach:
a. Thesoil propertiesintheconstitutivemodelsare
essentiallyindependent andthereforetreatedas
such.
b. For combiningthescenarios, it isassumedthat
theyaremutuallyexclusive, thusthetotal proba-
bilityisthesumof theprobabilitiesof scenarios
i (serial system):
c. The combination of the sub-mechanismprob-
abilities concerns a parallel system. Here the
worstcaseistotal dependencebetweenthesub-
mechanisms. This assumptions is probably not
evenunreasonablesinceinall mechanisms the
samesoil properties play arole. Therefore, the
maximumvalueof thesub-mechanismprobabil-
itiesisusedastheupper boundfor thescenario
probability:
Consequently thetopevent probability isdeter-
minedby:
for n scenariosandm sub-mechanisms.
2. Assumptionsinthephysical-processmodeling:
a. As mentioned in 3.3, the probability of lique-
factionisactuallytheprobabilityof thematerial
beingliquefiable. Moreconditions interms of
stressstateetc. havetobefulfilledfor liquefac-
tiontooccur.
b. In the slope stability analysis, the theoretical
minimumof theshear strengthaccordingtothe
material model isassignedtothezonesthat are
assumed to be liquefied. It is likely that not
the entire affected volumes undergo the total
strengthreductionandthatexcessporepressures
diminish, i.e. thattheshearstrengthisrecovered
at least partially.
At the same time, the assumptions made, indi-
catewherethereis certainly significant potential for
refinementsintheappliedmethod.Moresophisticated
mechanical and constitutive models are in principle
availablefor coupledanalysisinacademia, butnotyet
easilyapplicableinconsultancywork. Thereisachal-
lengefor theapplied sciences community to further
developthesemethodsandtoolscloser toapplication
inpractical problems.
4 RESULTS
For theproject itself, it was shownthat somedesign
amendments werenecessary, suchas thecompaction
of mainly the slope part pf the hydraulic fill and
a slightly shallower slope than initially planned in
order tofulfill thestrict safetyrequirement. Withthis
amendeddesignitwasshownthatthetotal probability
of failure(upper bound, seeprevious section) was in
theorder of P
f ,sup
=10
7
. But rather thanpresenting
morefigures, thetypeof resultsthat canbeproduced
withsuchananalysisareillustratedinthissection:
Theprobabilityof thetopevent inthefault tree, in
thiscasethefoundationsof theinstallationaffected
by slope failure, possibly induced by liquefaction
andbreaching,canbeusedinhigherlevel riskanaly-
sesandreliabilityanalysesof theentireinstallation.
The probabilistic approach therefore provides a
comparability with other elements of the system
that cannot beachievedotherwiseby theclassical
deterministicmethods.
Thefaulttreecontainsprobabilitieson(sub-) mech-
anismlevel. That enables theidentification of the
mostrelevantmechanismsandscenarios.Thisinfor-
mationis extremely useful for optimizationof the
design.
Thereliability analyses on(sub-) mechanismlevel
also produce information on the relative impor-
tance of the variables involved (e.g. FORM gives
influencecoefficients
i
). Someof theseproperties
can either beinfluenced by changes of thedesign
or by acquiring more information and thereby
reducing(epistemic)uncertainty,e.g.additional soil
investigation.
5 CONCLUSIONS
Thework on this paper has lead us to formulatethe
followingthreemainconclusions:
Firstly, thepaper demonstratestheapplicability of
reliability analysis for arather complex geotechnical
probleminareal worlddesignproblem. Inthecourse
317
of thedesignverification, theupper boundof thefail-
ureprobabilityisloweredstep-wisebyrefinementsof
either physical process or probabilistic models until
it is shown that the design fulfills the rather strict
requirements.
Secondly, it should be emphasized that such a
decomposition of theanalyzed failureprocesses can
hardly be done with deterministic approaches. The
commonsafety value, beit afactor, marginor some-
thingelse, wouldbevery difficult tocomposeout of
theresults of theevaluation of thesub-mechanisms.
Onceagain, comparabilityisoneof themajor advan-
tagesusingprobabilisticapproaches.
Finally, of course, a probabilistic approach does
not compensatefor deficiencies inphysical process-
basedmodels, it merelyprovidesaconsistent manner
to deal withtheuncertainties. Intheillustratedcase,
the sometimes quite rough upper bound approaches
led to a satisfactory answer, namely an acceptance
of thedesignby verifyingtherequiredrequirements.
On theother hand, weareconvinced that theuseof
upper boundsledtoarather conservativeassessment.
However, carryingout theindicatedpotential refine-
mentsisnot atrivial task withthecurrently available
methods. Especially for theinitiationof liquefaction,
the currently used models are unsatisfactory. Either
they are of empirical nature and based on a limited
number of (indirect andinterpreted) observations, or
they combineseveral physical-process basedmodels
withrather restrictiveassumptions. Thereisclearly a
needfor better in-depthunderstandingof thephysical
processes and their interaction leading to improved
models.
REFERENCES
J CSS (J oint Committee of Structural Safety) 2001. Prob-
abilistic Model Code, Part 3.07 Soil Properties (last
update08/2006). ISBN978-3-909386-79-6.
Lindenberg, J. &Koning, H.L. 1981. Critical densityof sand.
Geotechnique 31(2): 231245.
Lunne,T. &Christoffersen, H.P. 1983. Interpretationof cone
penetrometer data for offshore sands. In Proceedings
of the Offshore Technology conference, Paper no. 4464.
Richardson, Texas.
Mastbergen, D.R. & Van den Berg, J.H. 2003. Breaching
in fine sands and the generation of sustained turbid-
ity currentsinsubmarinecanyons. InSedimentology 50:
635637.
Molenkamp, F. 1989. Liquefactionas aninstability. InPro-
ceedings Int. Conf. on Soil Mechanics and Foundation
Engineering (ICSMFE): 157163.
Delft GeoSystems 2006. MStab 9.9, User Manual, Delft
GeoSystems, Delft.
Olson,S.M.&Stark,T.D.2003.YieldStrengthRatioandLiq-
uefactionAnalysisof SlopesandEmbankments. In Jour-
nal of Geotechnical and Geoenvironmental Engineering
129 (8): 727737. ASCE.
Sladen, J.A., DHollander, R.D. &Krahn, J. 1985. Thelique-
factionof sands, acollapsesurfaceapproach. InCanadian
Geotechnical Journal 22: 564578.
Stoutjesdijk, T.P., De Groot, M.B. & Lindenberg, J. 1998.
Flowslidepredictionmethod: influenceof slopegeome-
try. InCanadian Geotechnical Journal 35: 3454.
318
A casestudyof thegeological riskmanagement inmountaintunneling
T. Ikuma
Dia Consultants Co., Ltd, Tokyo, Japan
ABSTRACT: Inthistunnelingproject, aresearchwasconductedontheriskmanagementtechniquefor occur-
rence prevention of latent geological risks on a mountain tunnel in J apan. The geological consultant played
important roleabout therevaluationfor unexcavatedsectionandtheproposal of asuitableconstructionmethod
fromgeotechnical viewpoint. The geological risk management research of ground evaluation utilizing this
experienceiscontinuedinother mountaintunnel withthesamekindof geological features.
1 INTRODUCTION
Investigationof thepriorstageof mountaintunnel with
longlengthandhighoverburdenwasconductedforthe
purposeof clarifyingeachfollowingitem.
Theoverall geological structure,distributionof geo-
logical features and characteristics of the tunnel
section
Groundclassificationwithsynthetictechnical con-
siderationbasedontheresultsof investigation
Topography andgeology of portal locations, basic
datafor aproblemanditsmeasuredesign, and
Basicdatafor evaluationof facestability, designof
support, selectionof auxiliarymethod, selectionof
excavationmethodandtunnel drivingmethod
Especially in a tunnel design stage, geographical
andgeological dataof highaccuracyarerequired.
However, itisdifficulttodothehighlyprecisegeo-
logical survey which covers all the extension of the
tunnel as the linear structural object before excava-
tiongiventhepresenttechnical level, andinvestigation
periodandcurrentrequirementof economicefficiency
for geological survey.
Moreover,theJ apaneseIslandsbelongtothemobile
belt through thegeological age, and havevery com-
plicatedgeological structureanddistributionof geo-
logical features. For this reason, recognition of the
uncertainty about thegeological phenomenonintun-
nel constructionandthecorrespondencetothemhave
beenanimportantsubject. Intunnel construction, how
it grasps in a prior stage has linked to the subject
resultingfromthelatentgeological riskbythehetero-
geneityof theground,andtheuncertaintyof geological
informationwithcostandconstructionperioddirectly.
Thegeological riskintunnel constructionexpresses
degreeand gradeof asizeof theuncertainty which
undesirable phenomenon generates for construction
andcontrol of maintenance.
This paper describes an example which reduced
deviationof groundevaluationintheprior stageand
constructionstageinthemountaintunnel built onthe
steep mountains area, and the future view fromthe
viewpoint of geological riskmanagement.
2 GEOMORPHOLOGICAL FEATUREAND
GEOLOGY OF RESEARCHEDTUNNEL
This tunnel is planned by steep mountain area with
12001400m altitude, and some mountain streams
which have channel in the direction of northeast-
southwest crosstunnel alignment (refer toFigure1).
The direction of these mountain streams is in
agreement also with thelineament by an aerial pho-
tograph interpretation. Moreover, it is analyzed as
low-velocity zoneby seismic refraction method, and
isgeomorphological weakzone.
Figure1. Localityandgeological mapof researchedtunnel.
319
Groundconsistsof theNohi Rhyolites(Rhyolitic
Daciticweldedtuff) formedattheCretaceousageand
Graniteporphyry whichintrudethem. Thesewelded
tuffshavereceivedweatheringandhydrothermal alter-
ation, andlithofaciesalsochangesintricately.
3 GEOTECHNICAL PROBLEMS
It becameclear that thefollowinggeotechnical prob-
lems were held from the geological investigation
carriedout intheprior stageinthistunnel
Geological outcropswasdeficientinlandsurfaceas
awhole, anditwasdifficulttograspindetail about
thegeological structureof groundinthepriorstage.
It waspresumedthat weldedtuffsweredistributed
over 96.6%of tunnel length, andgraniteporphyry
was distributed to theremaining 3.4%in thetun-
nel formation level. However, the former has the
development of fracture frequency and the extent
of alteration in various forms, estimated accuracy
for deepbedrockconditionsisconsiderablylow.
It is considered that geological boundary between
welded tuffs and granite porphyry is a alteration
zonewithmany clayey thinlayers, andpossibility
of beingthebedrock whichdeterioratedishighin
thedepths also. Therefore, largeincreaseof earth
pressure caused by tunnel excavation or sudden
water inflowoccurswhenclayeylayer isanimper-
meable wall at section where overburden exceeds
300m. Insuchacase, ithasasignificantimpacton
excavating.
Thegroundclassificationwasperformedinconsid-
erationof thegroundconditionsonthistunnel based
onthetunnel standardforroadtunnel-structure(J apan
RoadAssociation; 1989).
The underground condition of tunnel with large
overburdendependsontheelasticwavevelocityvalue
acquiredmainly by seismic refractionmethod. How-
ever, sincethedepth of not less than about 200mis
alimit of exploration, thereliability of theacquired
elasticwavevelocityvalueisconsiderablylow.There-
fore, accuracyof thepositionof lowvelocityzoneand
boundaryontheclassof thegroundisalsolowinthe
tunnel formationlevel of theselargedepthsections.
Thelow velocity zoneis considered to beabove-
mentionedalterationzoneorshearzoneontheground.
In the design phase, the FEM analysis was con-
ductedin largeoverburden part. As aresult, support
patterncorrespondingtothegroundclassificationwas
ranked higher about the section where overburden
exceeds300m.
Accordingtopredictiontheamountof waterinflow
using hydrogeological conditions and hydraulic for-
mula, itisconsideredthattheamountof steadyinflow
inthistunnel isabout 0.7m
3
/min/km. Theamount of
concentratedwater inflowat stageof constructionis
presumedtobethoseaboutseveral times.Thegenerat-
ingpositionof concentratedwaterinflowcanconsider
attheperipheryof lowvelocityzoneanddirectlyunder
amountainstreaminthetunnel formationlevel.
4 GEOLOGICAL RISK MANAGEMENT
Thistunnel iscrossingsomemountainstreamswiththe
rightangleor thehighangle.Anintersectionpartisin
agreement withthepositionof alow-velocityzonein
manycases. Inthoseparts, althoughtheoverburdenis
large, facefallingandgeneratingwithlargeamountof
water inflowareassumedwithconstruction.
Inrespectof safetyof construction, suchfrequency
andquantityposeaproblemasthegeological risk.The
geological riskmanagementaboutthecountermeasure
of emergencytimeor predictingof thegeological risk
isveryimportant.
The example which was able to be reflected in
construction is reported using the geological risk
management technique.
In order to cope with the geological risks and to
advance construction more smoothly, the tripartite
council which consisted of owner, constructor and
geological consultant were established before con-
struction. Sincegeotechnical informationwasalways
sharedinthetripartitecouncil, quickinvestigationwas
conductedbythegeological consultantwhencollapse
of thefaceandwater inflowoccurredintunnel.
The geological engineer played important role
about the revaluation for unexcavated section and
theproposal of asuitableconstruction method from
geotechnical viewpoint.
4.1 Case example (southern section of this tunnel)
Insouthernsection, after excavationwasstarted, con-
structionwasfavorablycontinuedbyperformingTSP
(Tunnel Seismic Prediction) andhorizontal corebor-
ing as investigation of ahead of the tunnel face.
However,fromthevicinityof STA338,thedeteriorated
bedrockcametoappear frequently, andthesqueezing
fromatunnel sidewall alsobecameremarkable.
When excavation work advanced to STA.333, the
convergencewas increased. A maximumof 500mm
displacementsbysqueezedarefoundatthelateral side
of tunnel at thepoint of STA.331+15, andexcavation
was stopped. The cave of about 3m4msize was
checkedintheupper part of thisface. Theamount of
spring-water fromacavewas1.3m
3
/min(Figure2).
Whenanalterationzoneappears intunnel forma-
tionlevel withdeepoverburden, facefalling, expan-
sionof loosenzoneaccompanyingexcavationandthe
increase in lateral pressure are given. Moreover, it
becameclearthattheconfinedaquiferexistedinahead
of thefacefromresult of horizontal coreboring.
Thesupport patternaccompaniedbythehighrigid
auxiliary method is needed in continuation of exca-
vation judging fromthesesituations. Then, thesteel
pipe fore-piling as the prevention for face falling,
increase of rock bolt as lateral pressure preventive
measures and drainage borings as measure against
water inflow were proposed, respectively. Although
displacement headed for convergence, water inflow
wasnot decreased. Pressureof water inflowalsohad
1.7Mpa.
320
Figure2. Baserockandwater inflowsituationaroundSTA.331.
The geological condition in a tunnel was exam-
inedsynthetically, andanexcavationof drainagewas
proposed as a prevention which breaks through the
confined aquifer with irregular distribution from a
viewpoint of constructionandeconomical efficiency.
AsshowninFigure2, adrainagedrift wasexcavated.
In the point of No.331+16, it encountered the
expectedartesian aquifer by excavatingthedrainage
drift. At that time, the face fell suddenly, the
clayeysandy deteriorated bed rock of 80m
3
col-
lapsed, and the water inflow of 2.5m
3
/min was
occurred.
Thedrainagedrift wasblockadedby thiscollapse.
However, thegenerated water inflowmovefromthe
maintunnel tothedrainagedrift, andthewater level
andtheamount of water inflowhavebeendecreased.
The total amount of water inflow of tunnel was ca.
7500m
3
/day.
Sinceit was still generatedby largeamount water
inflowfromthefaceof this tunnel andthesteel pipe
fore-pilingalso after that, fivemoredrainageboring
wereexecutedwithSTA.331+5.9.
As a result, the amount of water inflow fromthe
facealsodecreasedandcontinuationof excavationwas
attained.
However, since the deep overburden section will
continueandoccurringof suddenwater inflowisalso
predicted, management of water inflowprocessingis
needed.
Forthisreason, drainagecapabilitywasreexamined
andproliferationof thefacilitiesfordrainageinalways
andanemergencywasproposed.
From this and other example (northern section
of this tunnel), the technique of the geological risk
managementtoconstructioninthesectionwheredete-
rioration of the bedrock and generating of a large
amount of water inflowaresummarizedinFigure3.
4.2 Orientation of geological risk
Althoughgeneratingfrequencywaslowaboutthesud-
den water inflow under tunnel construction, it was
predicted that a large amount of water inflow was
expectedat occurredtime.
Themountain streamwhich flows through near a
portal partisaclearstream, andmountaintroutinhab-
its. Moreover, in thedownstreamregion of southern
section, thereisasourceof tapwater usingtheinfil-
tration water of the river, and it is used as drinking
water. Itisnecessarytocareaboutenoughthemeasure
against drainageof thewater bywhichit isgenerated
fromtheinsideof atunnel also fromaviewpoint of
social environment.
Theamountof convergencedisplacementmeasured
byreal constructionissettledintherangepresumedin
thepreliminarysurveystage.Thecracksituationof the
shotcretesurfacewasalsoobservedcarefully, andlin-
ingconcretewas placedafter checkingdisplacement
convergence.
Those positioning is shown in Figure 4 about
theamount of suddenwater inflowandconvergence
among the geological risks in each stage of tunnel
construction.
321
Figure3. Constructionflowchart about thefacefallingsectionandroleof thegeological engineer.
Figure4. Orientationmapof thegeological risks.
In the maintenance stage, these geological risks
are monitor by visual observation and periodical
measurement.
4.3 Quantitative evaluation of ground by
modification index
Next, comparison between initial design stage and
actual results after excavation stage of the ground
classification is done by using the modification
index(i).
The modification index (Inoma: 1984) was used
toanalyzecomparisonof groundevaluationbetween
initial design stage and actual results quantitatively.
Modificationindex(i) isdefinedbyequation1.
322
Figure 5. Contrast of the ground classification between
initial designandexcavationstage.
Where,
R: Differenceof classnumberswithcorresponding
rockmassbetweeneachdesignedandactual case
n: Ratioagainst thetotal lengthwitheachR.
Thisindexisastatistical numerical valueusingroot
meansquareshowingdistributedconditionof avari-
able. Themodificationindexiscalculatedthroughthe
followingprocess;
Procedure 1: First of all, class of ground in original
designstageandactual excavationresult arecom-
pared.Then, ittotalsforeverywidthof aclassabout
changesituationof groundclass betweenthetwo.
Andanaccumulationcurveof thechangedwidthas
variableisdrawn.
Procedure2: Themodificationindex is calculatedby
equation1byusingabove-mentionedR andn.
Theresult of comparing theground classification
at initial designstage, aheadof thefaceinvestigation
stage,andactual resultsafterexcavationstageisshown
inFigure5.
If the modification index is calculated fromthe
ground classification in theahead of faceinvestiga-
tionstageandexcavationstage, it becomes 0.59and
0.80, respectively. Withthegeneral technical level, it
isconsideredi 1.1asastandard, andthequiteeffec-
tivefacedesignwas completedby theaheadof face
investigation.
4.4 Effect of the geological risk management
based on case example in this tunnel
Managementof riskshaverootsingeological features
is predictingappearanceof risk inadvance, andpre-
ventingor avoidingbeforehand. Especially, theuncer-
tainty of geological risk inundergroundconstruction
of our country is highwithcomplexity of geological
Table 1. Brief summary of the risk managementin this
tunnel.
features, anditisnoteasytostraightforwardlyclassify
geological riskmanagement inseveral patterns.
However, geological risk working group of
J GCA researchedgeological riskmanagementpattern
recently,andtheydividedintofollowingthreepatterns.
TypeA: A caseof avoidinggeological risk.
TypeB: A caseonwhichgeological riskis
actualized.
TypeC: A caseof minimizingdamageassociated
withgeological riskwhichwasactualized.
Above-mentionedcaseexamplebelongstoTypeA.
About southern section, deteriorated ground was
expected to some extent from before construction.
However, those scale and extension were uncertain.
Thiscaseconfirmedthesesituationsbyvariousexplo-
rations ahead of face and prevents geological risks
suchasfacecollapseandwater-inflowbeforehand. In
thiscase, sincebedrockdeteriorationwhichrepeated
itself complicatedly, we continued NATM by the
observational method.
Furthermore, suitable correspondence of heaving
and displacement of side wall were completed by
geological risk managements such as convergence
measurement, reset of control criteria value, and
water-inflowmanagement.
Brief summary of thegeological risk management
isshowninTable1.
323
Table2. Groundclassificationof thistunnel.
5 FUTUREVIEW
In the ground where geological and rock mass con-
dition change intricately in tunnel extension and
transversedirectionlikethistunnel, thedifferenceof
groundclassificationbetweeninitial designstageand
constructionstageisremarkableinmanycases.
Abouttheunexcavatedsectionwhichposesaprob-
lemespecially in a construction stage, the ahead of
face investigation result by two or more techniques
is considered synthetically, and grasp of the three-
dimensional ground condition of the tunnel exten-
sion is needed. However, an applicable exploration
methodisdifferentaccordingtooverburden, geologic
structure, andtherockmass conditionof theground.
Therefore, thegeological management which judges
the more effective exploration method appropriately
aboutqualitygraspof theunexcavatedpartof atunnel
isimportant.
Inevaluationaboutdeteriorationpartof theground
consistsof theNohi Rhylolitic rockslikethistunnel,
examinationof theresistivityvaluewaseffective.
However, thegeological problemabout thestruc-
ture of Granite porphyry remained during construc-
tion, and the variation of ground evaluation was
caused. Geological outcrop distribution of Granite
Porphyry was fragmentary and details of those dis-
tribution and geological structure were uncertain in
thepreliminary investigationstage. Intheexcavation
stage, it becameclear that Graniteporphyry hadsev-
eral timesasmanyspreadcomparedwithapreliminary
investigationstage.
Petrographically, theextensional shapeof Granite
porphyry is stock, and widening toward deep. The
depthdistributionshapeof Graniteporphyryremained
asaresidual riskuntil theexcavationstage.
A groundclassificationinthedesignandconstruc-
tionstageof thistunnel isshowninTable2.
Theclass of ground of this tableis conformed to
theground classification of J apan RoadAssociation
(1989). Inaddition, basedontheactual result of the
portal part, the ground classification of the general
Figure6. Groundclassificationof theplannedtunnel with
thesamekindof geological features.
part of this tunnel in a design stage was changed.
Furthermore, theDSC (Different SiteCondition) was
also considered and anewstandard of classification
wasmade.
Moreover, thegroundclassificationschemeinthe
planning stage of A and B tunnel which consists of
samegeological kind around this tunnel is shown in
Figure6 (refer to next page). Thesetwo tunnels are
under excavationnow.
Byrepeatingobservationof thesetunnel facecon-
ditionsandcomparisonof eachgroundcondition, the
morepractical standardof groundwill bebuilt from
nowon.
6 CONCLUSIONS
Itlearnedfromthecaseof facecollapseandsqueezing
of groundwhichoccursatnorthernsectionof thistun-
nel, geological riskmanagement about predictionand
accident preventiontotunnel deformation, andwater
inflowintheunexcavatedpart wascarriedout.
Asaresult, themodificationindex(i) becameabout
0.50.8, concerningthesectionwhereoccurrenceof
tunnel deformationandwater inflowwerepredicted,
anddeviationof thegroundevaluationbeforeandafter
excavationwasabletobemadeconsiderablysmall.
Ingeneral,alargeamountof waterinflowgenerated
inthetunnel insideisthoughttobethegeological risk
onexcavation. However, becauseitwaspreciouswater
resources about the fresh water in particular, water
analysis was used together about this. Furthermore,
separation measures were taken appropriately about
freshandmurky water, andit wasmanagedonwater
inflow.
It was possible to manage appropriately ground
classification and water inflowusing geological risk
management technique, this management effect was
largealsofromastandpoint of theexcavationcost.
324
ACKNOWLEDGMENTS
Theauthor expresses his sinceregratitudeto Profes-
sor Yusuke HONJ O of the Gifu University for his
encouraging advice. Special thanks are extended to
ProfessorTUNEMI WATANABEof theKochi Univer-
sityof TechnologyandMr. YoshihitoSABASE of the
CTI ENGINEERINGCo.,Ltd.forkindadvice.Thanks
areduetoMr.AkiraTAINAKA of theDIA CONSUL-
TANTSCo., Ltd, whoprovidedsuggestionduringthe
preparationof thispaper.
Theauthor also expresses gratitudeto Mrs. Keiko
HORIKAWA of theDIA CONSULTANTS Co., Ltd,
whocooperatedincreationof thefigureandtablefor
thispaper.
REFERENCES
T. Ikuma, K. Hatamoto, K.Yamamoto, T.Shindou, T. Ogawa,
2001, Revaluationof theRock MassbasedonExcavated
result in Mountain Tunnel, The 36th Japan National
Conference on Geotechnical Engineering, TheJ apanese
Geotechnical Societypp.19271928. (inJ apanese)
T. Ikuma2008. A casestudy of thegeological risk manage-
ment usingsuitableinvestigationinmountaintunneling,
Proc. of International Symposium on Society for Social
Management Systems 2008, Kochi. Society for Social
Management System
H. Inoma 1984. Comparison between the Projected Rock
Mass Classification at theInitial Design and theActual
Resultsafter theExcavationunder NATM Method, Jour-
nal of the Japan Society of Engineering Geology, Special
Volume: 6370.(inJ apanesewithEnglishAbstr.)
J apanRoadAssociation, 1989, Technical Standard for Road
Tunnels-Structures, 273p. (inJ apanese)
325
Guidelinefor monitoringandqualitycontrol at deepexcavations
T.J. Bles, A. Verweij, J.W.M. Salemans& M. Korff
Deltares, Delft, The Netherlands
O. Oung& H.E. Brassinga
Public Works Rotterdam, Rotterdam, The Netherlands
T.J.M. deWit
Geomet BV, Alphen a/d Rijn, The Netherlands
ABSTRACT: Geotechnical monitoringandqualitycontrol areoftenusedasastandardapproachof deepexca-
vationriskcontrol.However,oftenthefull potential of monitoringisnotusedandtherelationwithdesignandcon-
structionprocessesisdelimitedtosomegroundwater levelsreadingsandheightmarks. Monitoringoffersmuch
morepossibilitieswhenincorporatedintheprojectsriskmanagementandtheconstructionsqualitycontrol.That
iswhyacommitteeof expertsfromthefieldissettingupaguidelinefor theimplementationof monitoringand
qualitycontrol indeepexcavations.Theguidelinehelpsdesigners, contractorsandclientstodeterminemonitor-
ingbenefitsfortheirprojectandsuggestsopportunitiesforsuccessful embeddingof monitoringindesign, tender,
contract andconstruction. Thispaper dealswiththecontent of theguideline, whichistobereleasedin2009.
1 BENEFITSOF QUALITY CONTROL AND
MONITORING
1.1 Introduction
Failurecosts intheconstructionsector areestimated
to be five to ten percent of the total turnover. A
largeportionof thesefailurecostsisrelatedtounder-
groundconstructionworksincludingdeepexcavations
andpilefoundations. Thefocus of theresearchdone
and of this paper is on deep excavations (i.e. an
excavation for purpose of building an underground
constructionwithdepthsof afewmeterswithamaxi-
mumof approximately30meters) andaccompanying
foundationworks.
Inordertoreducefailurecostsitisnecessarytocon-
trol therisksaccompanyingundergroundconstruction
works. Unwanted and unforeseen events frequently
occur, resulting in constructional and economical
damage and in a negative image of the construc-
tionsector. Examplesaresevereleakages, foundation
piles not reachingtherequireddepth, discomfort for
the public or even damage to deep excavation or
surroundings.
Asaconsequence, groundrelatedriskmanagement
hasrapidlyevolvedinrecentyears. Moreandmoreitis
usedasafruitful methodtoreducegeotechnical risks.
Togivestructureongroundrelatedrisk management
DeltaresdevelopedtheGeoQ-method(VanStaveren,
2006). GeoQ is basedonsix generally acceptedrisk
management steps:
1. Determinationof objectivesanddatacollection;
2. Riskidentification;
3. Riskclassificationandquantification;
4. Riskremediation;
5. Riskevaluation;
6. Transfer of riskinformationtonext project phase
Geotechnical risk management isalready success-
fully usedinmany constructionprojects. Monitoring
and quality control are excellent tools within this
approachfordeepexcavationdesignandconstruction.
1.2 Monitoring and quality control
Manydefinitionsareavailablefor geotechnical mon-
itoringsuchasdescribedby Dunnicliff (1988, 1993).
Themost important aspectsare:
Measurements are performed repeatedly. This is
necessarytogaininsightindeepexcavationbehav-
ior over acertainperiodof time.
Depending on the (failure) mechanism to be
observedmeasurementsareperformedon(elements
of) the construction, in soil and/or on possible
surroundingconstructions.
Measurements canbeexecutedbefore, duringand
after theconstructionperiod.
Measurementsshouldcreatethepossibilitytofore-
see unwanted events and facilitate an incentive
to take appropriate measures in order to prevent
negativeconsequences.
Contractorsalsoperformmeasurementsfor quality
control. These measurements contribute to building
process control. Measurements for quality control
areusually performed only onceand areusually not
327
part of amonitoringplan. Still, many quality control
measurements complement traditional geotechnical
monitoring. Thereforequality control should bepart
of deepexcavationriskmanagement.
1.3 Objectives of measuring
As Marr (2001) states: Geotechnical monitoring
saves money and lives and/or diminishes risks. In
addition, Marr gives fourteen reasons demonstrating
geotechnical monitoringbenefits, varyingfromindi-
cating construction failure to construction process
control and increase of state-of-the-art knowledge.
Ingeneral four types of objectives of measuringand
monitoringareidentified:
1. Operational/qualitative goals: Decision making
with regard to possible occurrence of risks is
improvedby measuringfailuremechanismdevel-
opment. Theprogress of constructionof thedeep
excavationiscontrolledandchecksareperformed
ontheassumptionsmadefor thedesignof thedeep
excavation. Constructivesafetyof deepexcavation
andsurroundingsisalsoguaranteed. Theaimisto
reduceuncertaintyandgainreliability.
Inaddition, qualitycontrol of constructional ele-
ments is an operational goal. Examples are load
tests onanchors or piles, or torquemeasurements
whileinstallingdrilledpiles.
2. Communicativegoals: Deepexcavationsareoften
constructedindensely populatedareas. Therefore
it is very important to get thepublics support in
order topreventcomplaintsthatcanslowdownthe
constructionprocess. Monitoringcanbeefficiently
used to demonstrate the construction process is
under control.
3. Legal goals: Monitoring can be used to answer
questionsaboutliabilityof buildingdamage.Moni-
toringcanalsobearequirementorboundarycondi-
tiontoauthorities permissionfor deepexcavation
construction.
4. Scientificgoals: Monitoringcanprovideexcellent
datafor scientificresearchtoimproveunderstand-
ingof deepexcavation(andsoil) behavior.
2 RESEARCHANDGUIDELINE
2.1 Research
In general practice, monitoring is already used as a
standardpart of deepexcavations risk control. How-
ever, oftenthemonitoringpotential is not fully used
andtherelationwithdesignandconstructionprocesses
islimited. Themainreasonfor thisresearchistoclar-
ifythepossibilitiestousemonitoringmoreefficiently
inpractice.
Weresearchedthewaymonitoringcanbeoptimized
during the entire construction process, providing a
powerful tool withinabroadriskmanagement frame-
workforqualitycontrol andprocessoptimization.The
researchhasbeendonetogether withtheconstruction
industryintheNetherlands(contractors, clients, engi-
neers, monitoring companies and researchers). The
result is aCUR guidelinefor theimplementation of
quality control and monitoring in deep excavations
(CUR isaDutchcivil engineeringresearchinstitute).
Theguidelinewill beavailable(inDutch) mid2009.
2.2 Improvement in monitoring practice
After guideline implementation three types of im-
provements are expected to be achieved in Dutch
constructionpractice. Theseimprovementsare:
1. Increaseof clients awareness of monitoring ben-
efits. Monitoringis oftenmistakenfor atimeand
moneyconsumingactivity, necessaryforsatisfying
theauthorities requirementsandclientsdemands.
Theguidelineunderlines monitoringbenefits and
describeshowtomaximizemonitoringresults.
2. Monitoring will become an integral part of the
building process. This can be separated in the
followingaspects:
Facilitation of explicit responsibility allocation
for different partiesinvolvedintheconstruction
process.
All monitoring activities are based on a risk
management approach and are laid down in a
standardizedmonitoringplan.
Monitoring activities are coordinated by one
party to prevent fragmentation of monitoring
activities.
Measurementsaredirectlyadaptedtothebuild-
ingprocess (frequency, referenceandendmea-
surements, limit values of measurements, com-
municationandinterpretationof data, etc.).
3. Providinganoverviewof all differentmeasurement
techniques applicablefor deepexcavations. Tech-
niquesarecoupledwithspecificrisksincorporated
indifferent constructionmethods inawork break
downstructure.
2.3 Objectives of the guideline
The guidelines overall-goal is to improve the use
of measurements and monitoring, thus improving
quality and risk management. Threeobjectives were
formulated:
1. Describemeasurement techniquesassociatedwith
deep excavation construction, coupled to all rele-
vant constructionrisksandtherespectiveparame-
ters.
2. Present astep-by-stepplanandformat to set upa
solidmonitoringplan.
3. Provideopportunitiestoembedmonitoringinten-
dersandcontractual processes.
3 OVERVIEWOF MEASUREMENT
TECHNIQUES
Many companies deal with monitoring. They all
havetheir own experiencewith different monitoring
328
Figure1. Structuretoidentifymeasurement techniques.
Figure2. (Part of a) WBSof adeepexcavation.
techniques. However, theseexperiencesarenotshared
amongothers.Inthiswaythepractical knowledgelevel
doesnotincreaseandbestpracticeisinfactre-invented
ateachdeepexcavation, especiallyfornewandspecial
techniques.
For a complete outline of all measurement tech-
niques arisk-basedoverviewof adeepexcavationis
developed. This overviewhas thestructureas shown
infigure1.
The elements of a deep excavation are identified
bymeansof awork-break-down-structure(WBS) of a
deepexcavation. TheWBScanbeseeninfigure2.
For every element a list of unwanted events has
been made. Theseunwanted events aredescribed to
show the relevance. In order to keep an overview, a
differentiation has been made for unwanted events
affecting
Table1. Standardfordescribingparameterstobemeasured;
exampledealswithsoil deformations.
Parameter Soil deformation
Measurements(msmt)
of
Deformationinx-, y- and
z-directionwithuseof inclination
or extensoinstruments, leveling
Measurement boundary
values
Measurementsshouldbechecked
duringeachconstructionphase
against designvalues
signal values 80%designvalue
limit values 100%designvalue
Required accuracy
absoluteaccuracy Dependsondesign
frequency A minimumof onetimeper
constructionphase(endmsmt
isstart msmt of next phase).
Duringcritical phasesmore
measurementsarenecessary,
especiallywhentimeisan
important factor
timingof ref. msmt Beforestartingof activities
demandsonreference
measurement
At least twomeasurements
(similar)
timingof end
measurement
Withtimerelatedeffects: 3month
after last activityaffectingthe
deformations. Without
consolidation: 1monthafter last
activityaffectingthe
deformations.
demandsonendmsmt Oneissufficient
Handling of data
processingof data Linesof deformationaccordingto
time. Checkagainst designvalue.
necessaryspeedfor
availabilityof dataat
testingcompany
Speedof processingdepends
onconstructionphaseandrisk;
aimfor maximal threedays.
Communication to all parties
involved.
necessaryspeedfor
decisionmakingwhen
measurement exceeds
boundary
Speedof decisionmakingdepends
onconstructionphaseandrisk;
aimfor maximal threedays.
Communicationtoall parties
involved.
Measurements in
quality-control-plan
or monitoring-plan
Monitoringplan
1. singleelementsof thedeepexcavation
e.g. instabilityof trenchof cementedbentonitewall
2. deepexcavationasawhole, acombinationof those
singleelements
e.g. bursting/heaveof asubmergedconcretefloor
3. surroundingsof thedeepexcavation
e.g. cracks in surrounding buildings caused by
vibratingpiles
For eachunwantedevent theparametershavebeen
identifiedwhichshouldbetakenintoaccountfor pos-
sibledevelopment of theunwantedevent. Withthese
parameters it is possible to find a list of specific
measurement techniques. Each parameter and tech-
niqueisdescribedusingastandard. Examplesof such
standardsareshownintable1andtable2.
329
Table2. Standardfor describingamonitoringtechnique; exampledealswithextensometer.
Monitoringtechnique Extensoinstrument
What is monitored Vertical displacements (swell caused by excavation, settlement caused by
consolidation, relativetostrainsinconcreteconstructions)
Functioning of instrument Extensoinstrumentsareusedtoconstantlymeasuredifferencesindistance
betweentwoor morepointsover theaxisof aborehole. Thismakesit able
todeterminethevertical displacementsof soil layers. Incombinationwith
inclinomeasurementsatotal viewof thedisplacementscanbederived.
Inpractice, anopentubeisplacedinsideaborehole. Fixedpointsareplaced
ondifferent depths. Thedisplacements betweenthesefixedpoints andthe
headof extenso instruments onthetopof thetubearemeasured. E.g. the
instrumentscanbedisplacement recordersor potentioinstruments.
Photo of instrument
Figure of measurements
Accuracy of monitoring technique
a) sensitivitytoinstallationerrors Littlesensitivefor installationerrors
b) sensitivitytoerrorsduringoperation Littlesensitivefor measurement errors
c) vulnerabilityof instruments(solidness) Instrumentsarevulnerable
Explanationandrecommendations Whenusedfor measurementsincenter of deepexcavation, instrumentsare
sensitivefor collisionwithvehicles.
Instruments can be protected by installing a casing till 1 meter below
excavationlevel andattachingthecasingtoastrut.
d) Accuracy(absolute) +/ 0,05mm
e) Measurement range(absolute) 100mm
Relevant influencing factors from
surroundings
Instrumentsshouldbeprotectedincaseof placement inanexcavation.
Long term behavior (calibration, stability) Depends onspecificationof instruments. Referencepoint shouldbemea-
suredregularly.
Procedure of measurements Automatically
Interpretation of data
a) Existingsystemsfor analysisand
interpretation
Absoluteandrelativedisplacementscanbemeasured.
b) Ambiguity Unambiguous(interpretationalwaysobjective)
Maintenance None
Application 1 Swell (vertical soil displacements) at excavation
Suitability for this application Verysuitable
Best practices
a) Number of instruments At minimumonelocationinthecenter of thedeepexcavation
b) Locationof instruments At minimumone measurement anchor at each soil layer. More accurate
resultscanbeobtainedbyusingtwoanchorsper soil layer.
Application 2 Damage to surroundings caused by (densification of sand layers due to)
vibrating or hammering sheet piles
Application 3 Deformation of soil and surrounding buildings caused by bending or
collapse of wall of the pit
For thisexample, app. 2and3arenot workedout indetail
4 DEVELOPINGA MONITORINGPLAN
Theguidelinepresentsastep-by-stepplantoset upa
goodmonitoringplan. Thebasisof thisplanwasfor-
mulatedinHERMES(2003), butadaptedforpractical
useinconstructionworks.
Notall situationsaskforthesamemonitoringinten-
sityandefforts. A largedeepexcavationinabusyand
oldcitycentrewill requireamuchmoreintensemoni-
toringsystemthanastandarddeepexcavationoutside
thecity. Inaddition, thetypeof constructionwill lead
to a different monitoring strategy. With help of the
guideline, the reader can learn which monitoring is
necessary, basedonthesituationandproject risks, to
provideagoodtool for proper riskmanagement.
4.1 Table of contents
A good monitoring plan should include at least the
elementsstatedbelow.Thecapitalsbehindthechapters
refer to thesteps that canbefollowedinorder to get
all necessaryinput for aproper riskbasedmonitoring
plan. Thesestepsaredescribedbelow.
1 Introduction
1.1 Project descriptionandbasic A
assumptions
1.2 Objectivesof monitoring B
2 Resultsfromriskanalysis C
3 Monitoringstrategy DG
4 Operational plan HI
5 Maintenanceplan J
330
6 Measureswhenmeasurement limits K
areexceeded
7 Dismantleplan L
8 Communicationplan M
Thetableof contentsisthesamefor every typeof
project, bigor small, simpleor complex. Thewayitis
elaboratecanhowever bedifferent. Onehastousethe
guidelineinapragmaticway.
4.2 Steps to obtain a risk based monitoring plan
4.2.1 Steps A and B, scope and objectives
Theproject needs demarcation in spaceand timein
order to control thescope. Then, theobjectives (see
different types of objectives in paragraph 1.3 of this
paper) for themonitoringarechosen.
4.2.2 Step C, risk analysis
Riskmanagementisof key-importancetoagoodmon-
itoring plan. Monitoring efforts may be an outcome
of arisk analysis. StepC thereforeincludes ago/no-
go decision. This decision should be based on the
followingquestions:
Istherisktobemonitoredcritical (bigenough)?
Ismonitoringthebestoptioninordertomanagethe
risk?
The risk analysis can be technical on operational
goals, but can also bemoregeneral; for exampleon
communicational goals.A summaryfromthisanalysis
shouldbewritteninthemonitoringplan.
4.2.3 Step D, monitoring strategy
The effort on monitoring needs to be evaluated,
together with theclient, by weighing thebenefits of
measuring(decreaseof risktoloosemoney,time,qual-
ity and/or image of the client) to the costs. In this
waybetter understandingof thenecessityof themea-
surementsiscreated. For elaborationonecanusethe
following steps DG. However, only thedetermined
strategyneedstobereportedandfurther elaboratedin
thefinal monitoringplan.
4.2.4 Step E, parameters
Determinetheparameters to bemeasured. Arethese
parameters sensitiveenoughfor all risks that haveto
be monitored? The scheme in the previous chapter
providesbackgroundfor thisstep.
4.2.5 Step F, demands
Determinethedemandsonthemonitoring:
Signal and limit values of parameters to be mea-
sured, making use of the scheme in the previous
chapter and/or drawings, normsandliterature.
Locationof themeasurements;isthislocationsensi-
tiveenoughinorder tohaveproper measurements?
Sensitivity and range of measurements of each
parameter to be measured; this is defined from
therisks to becontrolled and is ademand for the
instruments.
Frequency of themeasurements; ahigher or lower
frequency can lead to thechoicefor other instru-
ments.
4.2.6 Step G, instruments
Basedontheprevioussteps, typesof instrumentscan
beselected that fit thegiven demands. Each typeof
instrument should haveaspecific goal for measure-
ments.Instrumentswithoutsuchaspecificgoal should
be left out. Afterwards a specific instrument can be
selectedfromaproducer.
4.2.7 Step H, influence from surroundings
Effectsfromactivitiessurroundingtheprojectcandis-
turbthemeasurements. Forexampleheavytrafficona
neighboringroadcausespossiblevibrationsanddaily
temperaturedifferencescauseshrinkageandextension
of constructions. Thiscaninfluenceprocessingof the
data and can put high demands on the maintenance
plan. Sometimes it is necessary to go back to stepF
andchooseadifferent instrument.
4.2.8 Step I, planning of operations
For eachmonitoringinstrument thefollowingshould
beclear:
Location(xy) anddepth(z)
Demandsonreferencemeasurement
Measuringfrequency
Timetablefor obtainingmonitoringdata(relatedto
theconstructionprocess)
Format of data
Demandsontheprocessingof data
Demandsonendmeasurement
4.2.9 Step J, planning maintenance
Planningof necessarycalibrationandmaintenance.
4.2.10 Step K, measures
Animportantstepinthisriskbasedprocessistodecide
about measures to be taken when exceeding signal
andlimitvaluesof measurement. Onlywithpro-active
thinking, measurescanbetakenintimeinordertopre-
vent discussionswhenimmediateactionisnecessary.
However, for eachproject onehas to chooseto what
depthonewantstoelaborateall possiblemeasures.
4.2.11 Step L, dismantling
Short description of when and how dismantling of
the monitoring system will take place and who is
responsible.
4.2.12 Step M, communication
When using monitoring it is very important to have
proper communication between all parties involved.
The processing of the data should be aligned with
the project activities. Especially the maximumtime
span between measurement, processing and taking
measures is of importance. When there is no atten-
tion for communication, monitoring does not make
sense.Afterall, thepurposeof monitoringistoforesee
unwantedeventsandtakemeasuresintime.
331
Therefore, itiscrucial tohaveaneffectivecommu-
nicationplan.Thisplanshouldatleastgiveanswersto
thefollowingquestions:
Whoisresponsiblefor measurementsexecution?
Who is responsiblefor measurement communica-
tion?
Whoisresponsiblefor measurement processing?
Whoisresponsibleformeasurementinterpretation?
Who is responsiblefor takingactionafter exceed-
ingtheboundary valuesof measurements? Howis
ensuredthat theseactionsreallytakeplace?
Whoisendresponsiblefor thetotal program?
5 PRACTICAL USE
In the end, monitoring has to be used in practice.
However, also when monitoring is considered of
much importance, it is often perceived difficult to
divide responsibilities between the different parties
involved in the construction process. The guideline
thereforeprovidessuggestionsonhowmonitoringcan
be embedded in tenders and contracts and how to
spread responsibilities. Distinction is made between
traditional contracts(usingspecifications,designbyor
throughtheclient) andintegratedcontracts(i.e. design
andconstruct).
Threerulesformthebasisfor dividingtherespon-
sibilities:
1. Thepartythat makesacertainchoiceinthedesign
or construction process is responsible for that
choice.
2. This party also takes theconsequences accompa-
nyingthechoice.
3. Accordingly, thisresponsibleparty determinesthe
monitoringwithregardtothischoiceandperforms
themonitoringor assignsathirdparty toperform
themonitoring.
Roughly, thisresultsinthefollowingdistributionof
responsibilitiesfor thedifferent typesof contracts:
Traditional: theclientdeterminestheextentof mon-
itoring and performs it himself, or assigns athird
party to performthe monitoring. The monitoring
canbepart of thespecificationsfor thecontractor.
Integrated: Monitoringis part of thecontract with
thecontractor andthecontractor isresponsiblefor
determinationandperforming.
However, the client can state process require-
ments. For example, theclientcandemandthecon-
tractor toformulatethemonitoringplanaccording
totheguideline.
Duringthetenderphaseitisdifficulttogivespec-
ifications within an integrated contract, becausea
clientcannotbetoospecific, allowingthecontrac-
tor to makethedesign himself. Risk management
is the key to solve this. The client can demand a
contractor to use risk management in his design
approachandtobespecificontherolemonitoring
will playinthetotal projectsriskmanagement.
6 CONCLUSIONS
Theguidelinesoverall-goal wastoimprovetheuseof
measurementsandmonitoring, thusimprovingquality
and risk management. Theguidelineindeed answers
theresearchquestions. Theknowledgeof Dutchmon-
itoringandconstructionindustry withregardtodeep
excavationsisgatheredinorder toget anoverviewof
all differentmeasurementtechniques.Alsoatutorial is
giveninorder toobtainarisk basedmonitoringplan.
Finally, suggestionsareshownfor implementationin
practice.Twocasestudieshavebeenexecutedtocheck
thepractical useof theguidelinewithpositiveresults.
Lastchangesaremadeaccordingtotheresultsof these
cases. Theguidelinewill beavailablefrommid2009.
ACKNOWLEDGEMENTS
This research was only possible with the contri-
butions of all members of the committee (CUR
H416) of experts in the field and Delft Cluster
(www.delftcluster.nl).
REFERENCES
Dunnicliff, J., 1988, 1993, Geotechnical Instrumentationfor
MonitoringFieldPerformance, John Wiley & Sons, Inc.
HERMES, feb 2003, Het Rationale Monitoring Evaluatie
Systeem(TheRational Monitoring Evaluation System),
Delft Cluster
Marr, A.W., 2001, Whymonitor Geotechnical Performance?
49th Geotechnical Conference in Minnesota
Staveren, M. van, 2006, UncertaintyandGroundConditions:
A RiskManagementApproach, Elsevier Ltd.
332
A studyontheempirical determinationprocedureof groundstrengthfor
seismicperformanceevaluationof roadembankments
K. Ichii
Graduate School of Engineering, Hiroshima University, Higashihiroshima, Japan
Y. Hata
R&D Center, Nippon Koei Co., Ltd., Tsukuba, Japan
ABSTRACT: Roadembankmentsshouldhaveacertainlevel of seismic performanceagainst astrongearth-
quake motion. The ground strength parameters (cohesion c and internal friction angle ) are key factors in
theseismic performanceassessment of roadembankments. However, theproceduretodeterminatetheground
strengthparameter isdependentontheexperienceof theengineers, anditisnotwell documented. For example,
thepatterns of in-situ soil test at theembankment (Number of tests and its locations) arenot unique. In this
study, aquestionnairesurvey to76civil engineersinJ apanisconductedtoreveal their empirical procedureof
parameter determination.Theresultof thequestionnaireclarifytheconsiderablevariationof determinedground
strengthparameterswhichdependingontheexperienceof engineers.
1 INTRODUCTION
Roadembankmentsshouldhaveacertainlevel of seis-
mic performanceagainst astrongearthquakemotion
(J apan Society of Civil Engineers (J SCE), 2000).
Therearemany applicableguidelines for theseismic
performanceevaluationof geotechnical worksinclud-
ing road embankments. In the procedure of seismic
performanceassessment, thegroundstrengthparame-
ters(cohesionc andinternal frictionangle) aremost
important.
However, theproceduretodeterminatetheground
strengthparameter isdependent ontheexperienceof
theengineers,anditisnotwell documented.Forexam-
ple, thepatternsof in-situsoil test at theembankment
(Number of testsanditslocations) arenot unique.
Inthisstudy, aquestionnairesurveyto76civil engi-
neers in J apan is conducted to reveal their empirical
procedureof parameter determination. Thequestion-
nairesconsistof 4questions.Thefirstoneisaquestion
for thetypes of geotechnical investigationto becar-
riedout. Thesecondoneisaquestionfor thelocation
and number of in-situ tests. Thethird oneis aques-
tionaskingparameter identificationfor avirtual soil
profile. Thelast oneis aquestionfor theexperience
of therespondents. Inthispaper, theresultof thethird
questionisbrieflyreported.
2 CONTENTSOF QUESTIONNAIRE
2.1 Assumed background for the questions
The following are assumptions for questions. Note
it determined by the viewpoint of simplicity.
Seismic performance assessments of embankments
arerequestedfor thepointA andpoint B, asshownin
theFigure1. ThepointA isastandardtypeembank-
mentonhorizontallylayeredbedrock,butthepointBis
thehalf-bank shapedembankment ontiltedbedrock.
The distance between point A and point B is 5km.
Theseembankments(EmbankmentA &Embankment
B) are both located in Hiroshima Prefecture, J apan,
andfilledbyMasadosand(decomposedgranite). The
shapes of embankments aresummarized in Figure1
andTable1. Theparametersof embankmentsshapes,
such as slope gradient, height and crest width are
alsoshowninTable1. Thebedrocks arevery strong.
Thegroundwater level isvery low, thepossibility of
liquefactionisnonecessarytobeconsidered.
Figure1. Thetargetedembankments.
333
Table1. Parametersof targetedembankments.
Embankment EmbankmentA Embankment B
Parameters (Standard) (Half-bank)
Height H (m) 12 12
Crest widthW (m) 22 22
Gradient of slope 1:1.8(29.1) 1:1.8(29.1)
1:s (deg.)
Gradient of base Horizontal Tiltedbase
1:k (deg.) base 1:3.63(15.4)
Soil material Masado Masado
Figure2. TheNvaluesat thesites.
2.2 Questions
Total 4questions aregiveninthequestionnaire. The
first oneis about thetypeof geotechnical investiga-
tiontobecarriedout. Thesecondoneis thelocation
andnumber of in-situtests.Thethirdoneisaquestion
askingparameter identificationfor avirtual soil pro-
file. Thelast oneis aquestion for theexperienceof
therespondents. Followingisthecontentsof thethird
question.
Standardpenetrationtests(SPT) werecarriedout
at the crest of Embankment A and Embankment B.
TheobtainedN values areshowninFigure2. Based
ontheobtaineddata, pleaseidentifytheshearstrength
parameters (cohesionc andinternal frictionangle)
tocarryouttheslopestabilityassessmentbasedonthe
circular slipmethod.
If possible, pleasementionthemethodtobeused.
Furthermore, inadditionto thefinal estimatevalues
of shear strengthparameters(cohesionc andinternal
frictionangle), pleasementionthepossiblerangeof
parameters.
3 RESULTSOF THE QUESTIONNAIRE
3.1 Characteristics of respondents
Thequestionnairewassenttototal 200engineers.Asa
result, theeffectiveresponsewas76. Thecharacteris-
ticsof respondent areshowninFigure3. Asshownin
Figure3(a), themajorityof respondents(42) arecon-
sultant engineers. The practical working year of the
respondentsissummarizedinFigure3(b). Itisinvery
widerange.
Figure3(c) shows thequalificationof therespon-
dents. Most of therespondent haveat least oneof big
four qualifications(Professional Engineer of J apanin
civil engineeringfield, Professional Engineerof J apan
ingeneral technological project management, Doctor
of Engineering, 1stgradeEngineeringworksmanage-
ment engineer). Therefore, most of the respondents
might havecertain level of technological knowledge
andexperience. Figure3(d) showsthemainbusiness
field of respondents. About one-third of therespon-
dentsaremainlyworkingonthedesignof embankment
or slopestabilityassessment.
3.2 Method for cohesion estimation
Most of theengineersestimatedthecohesionaszero
(c =0), because the Masado soil was used. This is
basedontheengineersexperienceandbytheengineer-
ingjudgmentconsideringthesafetyside.Thecohesion
of asmall quantityisansweredbysomerespondents,
asadummyvaluetoprevent surfacefailureof slopes
in the dynamic FEM analysis. The followings are
empirical equations used in theanswer (J apan Road
Association(J RA), 1996; J RA, 1999).
Where,c (kPa)isthecohesionandh (m)isthedepth
fromthecrest of embankments.
334
Figure3. Thecharacteristicsof respondents.
3.3 Method for internal friction angle estimation
Theresults of answered method for internal friction
angle are summarized in Figure 4. There is no
great difference in the methods for Embankment A
and Embankment B. The answered methods can be
classifiedinto10typesasfollows.
MethodI isthetechniqueproposedbyFukui (Fukui
et al., 2000), anditisonthespecificationsforhighway
bridges(J RA, 2002).
The conversion of N value into N
1
value is as
follows.
Where,
v
is the effective overburden pressure
(kPa).
MethodII isthetechniquewhichisonthespecifi-
cations for highway bridges (J RA, 1980). Theeffect
of the confining pressure is not considered in this
equation.
MethodIII isthetechniquebasedontheactual val-
ues for expressway embankments (e.g. Okubo et al.,
Figure4. Themethodsfor parameter determination.
335
2004). Note a purely empirical decision based on
the experience in expressway embankment is also
classifiedasthismethod.
Method IV is the technique proposed by Osaki
(Osaki, 1959). This method is based on thesoil test
resultsintheKantoregioninJ apan.
MethodVisthetechniquebasedontheeffectof pre-
viousstudiesontheMasado(e.g. J apanGeotechnical
Society(J GS), 2004).
Method VI is the technique based on the guide-
lines for design of expressway by NEXCO (Nippon
ExpresswayCo., Ltd, 2006).
MethodVII isthetechniqueproposedbyHatanaka
(Hatanaka& Uchida, 1996). Thismethodisbasedon
the relationship between CD test results and corre-
spondingNvalue.
Theconversionof NvalueintoN
2
valueconsidering
theeffectof effectiveconfiningpressureisasfollows.
MethodVIII is thetechniquebased on theguide-
linesfor roadearthwork of theslopeengineeringand
slopestabilityengineering(J RA, 1999).
MethodIX isthetechniquebasedontheguidelines
for railways (Railway Technical Research Institute,
2007).
Method X is the technique proposed by Dunham
(Dunham, 1954).
Where, A is the coefficient considering the grain
sizedistributionandthegrainshape.
InFigure4, Others arecommentsthat it isdiffi-
cult to estimatetheinternal friction angleonly from
the N value, and None are non-effective answers.
Aboutahalf respondentadoptedthetechniquesbased
on thespecifications for highway bridges (Method I
andMethodII).
3.4 Estimated result of cohesion
The estimated result of the cohesion is shown in
Figure 5. There is no great difference between the
estimated cohesion for Embankment A and that for
Embankment B. Most of therespondents regard the
cohesionaszero(c =0), andonly15respondentscon-
sideredMasadoasadhesivematerial.Themainreason
of regardingMasadoascohesionlessmaterial (c =0)
istogetevaluationonthesafetyside, thoughacertain
level of cohesioncanbeobservedfor theunsaturated
soil inusual.
3.5 Estimated result of internal friction angle
The estimated result of the internal friction angle is
showninFigure6. As sameas theFigure5, thereis
no great differencebetweentheestimatedvalues for
Embankment A andEmbankment B. Thedegrees of
thedispersionof estimatedinternal frictionangleare
about4degreesinthestandarddeviation, about0.13
inthevariationcoefficient. Evenif thesamemethod
isadopted, theestimatedresult varies. Thisisbecause
of thedifferenceinthewayof dealingwithscattering
N values. Some respondent divide the embankment
intosomelayers. Notethereisauniqueanswer which
regardMasado as purely adhesivematerial (friction-
lessmaterial; =0) becausetheobtainedNvaluesare
not increasedwiththeincreaseof thedepth.
4 DISCUSSION
Astheeffectof variationsinestimatedgroundstrength
parameters (c, ), theMohr-Coulombs failurecrite-
rionis variedas showninFigure7. Theinfluenceof
the determined cohesion is significant in the lower
confining pressure region. However, maybe due to
theeffect of surfacefailurewhichcorrespondstothe
lower confiningpressureregion, itisreportedthatthe
seismic performance of embankments evaluated by
FEM isgreatlydependent onthelevel of theapparent
cohesion(Hataet al., 2009). Therefore, theobtained
variationof theestimatedcohesionimpliesthesignif-
icant variation on evaluated seismic performance of
embankments.
In this study, the degree of the dispersion of the
estimated internal friction angleis almost 0.1 in the
variationcoefficient (refer toFigure6). Ontheother
hand, itisreportedthatthedegreeof theheterogeneity
of the internal friction angle obtained by laboratory
testsfor J apaneseairport embankmentsisalmost 0.1
inthevariationcoefficient(Hataet al., 2008). Inother
words, thedegreeof thedispersionof internal friction
anglebasedonengineeringjudgments is almost ina
samelevel of thedegreeof theheterogeneity of soil
itself.
5 CONCLUSIONS
Inthis study, aquestionnaireto 76civil engineers in
J apantoreveal their empirical procedureof parameter
determinationiscarriedout. Inthispaper, theanswers
for apartial questionarebrieflyreported.
The method for the ground strength parameter
determination fromN values shows awiderangeof
variety. In addition to the difference of the adopted
method, thedifferenceinthewayof dealingwithscat-
teringNvaluesisalsoamajorreasonof thedifference
of theestimatedparameters.
As aresult, acertainlevel of variationintheesti-
matedvaluesonshearstrengthparametersisreported.
336
Figure5. Theestimatedcohesion.
Figure6. Theestimatedinternal frictionangle.
337
Figure 7. The difference of estimated Mohr-Coulombs
failurecriterion.
For example, the degree of the dispersion of inter-
nal frictionanglebasedonengineeringjudgments is
almost inasamelevel of thedegreeof theobserved
heterogeneityof soil strength.Thiskindof knowledge
is quite important to discuss the reliability of seis-
mic performanceevaluation. A detailed examination
of obtained answers and somemoredetailed survey
will bedoneasafuturestudy.
REFERENCES
Dunham, J. W. 1954. Pilefoundationsfor buildings, Proc. of
ASCE, Soil MechanicsandFoundationDivision.
Fukui, J., Shirato, S., Matsui, K. and Okamoto, S. 2002.
Relationship between internal friction angleof thesand
betweenNvalueof thestandardpenetrationtestbasedon
thetriaxial compressiontest(inJ apanese), Technical note
of PWRI, No.3849, 50p.
Hata, Y., Ichii, K., Kano, S. andTsuchida, T. 2008. A fun-
damental survey on the soil properties in the airport
embankments (in J apanese with English abstract), Bul-
letin of the graduate school of engineering, Hiroshima
university, Vol.57, No.1.
Hata, Y., Ichii, K., Tsuchida, T. andKano, S. 2009. A study
ontheseismicresistancereductionof embankmentdueto
rainfall (inJ apanesewithEnglishabstract), Jour. of JSCE
C, Vol.65, No.1. (inprinting)
Hatanaka, M. and Uchida, A. 1996. Empirical correlation
betweenpenetrationresistanceandinternal frictionangle
of sandysoils, Soils and Foundations, J GS, Vol.36, No.4,
19.
J apan Geotechnical Society. 2004. J apanese standards for
geotechnical and geoenvironmental investigation meth-
ods, Standardsandexplanations(inJ apanese), 889p.
J apan Road Association. 1980. Specifications for high-
way bridges, Part IV Substructureedition(inJ apanese),
MaruzenCo., Ltd.
J apan RoadAssociation. 1996. Specifications for highway
bridges, PartIV Substructureedition, 566p, MaruzenCo.,
Ltd.
J apan Road Association. 1999. Guidelines for road earth-
work, Slopeengineering and slopestability engineering
edition, 470p, MaruzenCo., Ltd.
J apan RoadAssociation. 2002. Specifications for highway
bridges, Part V aseismic design edition, 406p, Maruzen
Co., Ltd.
J apanSocietyof Civil Engineers. 2000. Thethirdsuggestion
and commentary about thecivil structure(in J apanese),
Chapter 8, Earthstructuresedition, 2934.
Kitazawa, G., Takeyama, K., Suzuki, K., Okawara, H. and
Osaki, Y. 1959. Tokyogroundmap(inJ apanese), 1819,
GihodoShuppanCo., Ltd.
NipponExpresswayCompanyLimited. 2006. Guidelinefor
design of expressway, earthwork edition (in J apanese),
350p.
Okubo, K., Hamazaki, T., Kitamura, Y., Inagaki, M., Saeki,
M., Hamano, M. and Tatsuoka, F. 2004. A study on
seismic performance of expressway embankments sub-
jectedtoHigh-level seismic load(part 1) Evaluationof
shear strength of embankment materials- (in J apanese),
Proc. of 39th Japan National Conference on Geotechnical
Engineering (CD-ROM), No.881, 17591760.
Railway Technical Research Institute. 2007. Guidelines for
railway, Earth structure edition (in J apanese), 703p.,
MaruzenCo., Ltd.
338
GeoRiskScan Gettinggripsongeotechnical risks
T.J. Bles& M.Th. vanStaveren
Deltares, Delft, The Netherlands
P.P.T. Litjens& P.M.C.B.M. Cools
Rijkswaterstaat Competence Center for Infrastructure, Utrecht, The Netherlands
ABSTRACT: Groundconditionsappear amajor sourceof cost overrunsininfrastructureprojects, whichhas
beenconfirmedby recent Dutchresearch. As ananswer to thesecost andtimeoverruns, groundrelatedrisk
management has rapidly evolved in recent years. Deltares for instance developed the GeoQ-method. In the
Netherlands, Rijkswaterstaat, theCentrefor Public Works of theDutch Ministry of Public Works andWater
Management, isinitiator andowner of all federal infrastructureprojects. TheyaskedDeltarestoperformaGeo
RiskScanonfiveselectedprojectsof thetop20largestDutchinfrastructureprojects.TheGeoRiskScanproved
to bean effectivetool for quickly providing information about thedegreeand thequality of ground related
risk management ininfrastructureprojects. This paper describes theGeoRisk Scan, as well as its application
withinfiveprojects. Theevaluationof thefiveprojectsresultedinsixmainlessonswhicharefurther elaborated
in 20 recommendations. Theselessons may help project owners, engineers, and contractors to managetheir
constructionprojects.
1 INTRODUCTION
1.1 Successful and unsuccessful projects
What makes thedifferencebetweenasuccessful and
anunsuccessful constructionproject?A projectthatis
completed well within budget, time, and its require-
ments, or not? It can not beits sizeand complexity,
because there are successful and unsuccessful ones,
amongst small andlarge, aswell assimpleandcom-
plexprojects. It isalsonot itslocation, becauseevery
country seemstohaveitssuccessful andproblematic
projects. It isevennot becauseof groundconditions,
as weall know examples of successful projects that
havebeencompletedinvery difficult groundcondi-
tions. Theremust beanother reason. Perhapsit isthe
waythemanagementteamof aprojectisabletoman-
agetheinherent presenceof risk, duringall phasesof
realizingtheproject.
1.2 Risk management as an answer
to failure costs
Several studies indicatethat failurecosts inthecon-
struction industry are typically 10 to 30 percent of
thetotal constructioncosts(AvendanoCastilloet al.,
2008). This seems to be a worldwide phenomenon.
Thereis also abundant evidencethat unexpectedand
unfavourablegroundconditions haveaserious stake
in these failure costs (Van Staveren, 2006). In the
Netherlands, Rijkswaterstaat, the Centre for Public
Works of the Dutch Ministry of Public Works and
Water Management, is initiator andowner of all fed-
eral infrastructureprojects.Therefore, Rijkswaterstaat
decidedtopayparticular attentiontothemanagement
of groundrelatedriskswithintheir projects.
1.3 Ground related risk management
Thedevelopment andapplicationof geotechnical risk
managementgetsmoreattentioninrecentyears. More
and more, it is considered an effective and efficient
way of work process for controlling all types of
ground-related risk. For instance Deltares, formerly
knownasGeoDelft,developedtheGeoQ-method(Van
Staveren, 2006), thatisalreadyusedinmanyconstruc-
tionprojectswithgoodresults. TheGeoQapproachis
infactanin-depthapplicationof theRISMANproject
risk management method. GeoQ focuses oncontrol-
lingground-relatedrisks. Themethodisbasedonsix
generallyacceptedriskmanagement steps:
1. Determination of project objectives and datacol-
lection;
2. Riskidentification;
3. Riskclassificationandquantification;
4. Riskremediation;
5. Riskevaluation;
6. Transfer of risk information to the next project
phase.
Theserisk management steps shouldbeexplicitly
takeninall phases of aconstructionproject. Ideally,
the ground related risk management process starts
in the feasibility phase and is continued during the
339
(pre)designphase, thecontractingphase, theconstruc-
tionphaseandtheoperationandmaintenancephase.
Obviouslyforbeingeffectiveandefficient, theground
relatedriskmanagementshouldbealignedwithmore
general projectriskmanagementapproaches. Because
of thesimilarityof riskmanagementsteps, thisshould
benoproblem.
Themaindifferentiatingfeatureof groundrelated
risk management, whencomparedto generic project
riskmanagement, isitsspecificattentiontogeotechni-
cal riskanditsremediation. Therefore, groundrelated
risk management usesconventional risk management
approaches, such as qualitative risk assessments, as
well as specific geotechnical approaches. The latter
includes for example risk-driven site investigations
andmonitoringprogrammes.
2 SET UP OF A GEORISK SCAN
2.1 Introduction and objectives
For gaining insight in the degree and quality of
geotechnical riskmanagementinprojects, Rijkswater-
staataskedDeltarestoperformaGeoRiskScanonfive
selectedprojectsoutof thetop20largestDutchinfras-
tructure projects. The main objectives were gaining
insightinthetypeandcharacteristicsof groundrelated
risks, the possible consequences when these risks
wouldoccur, andthedegreetowhichriskremediation
measuresweretakenwithintheprojects.Moreover,the
results of theGeo Risk Scanwouldgenerateaqual-
ity judgement about the degree of geotechnical risk
management.
In order to achievetheseobjectives theGeo Risk
Scanaimstoquicklyscanbothprocessandcontentof
thegroundrelatedrisk management withinaproject.
Theexecution of awell-structured risk management
process, bytakingthepresentedsixrisk management
steps, is considered as themain boundary condition
forgeneratingeffectiveandefficientgeotechnical risk
management.
If necessary, recommendationstotheprojectorgan-
isations areprovided in order to improvetheproject
performanceand reducing theprobability of ground
relatedfailurecosts.
2.2 Structure of a Geo Risk Scan
Thebasisof theGeoRiskScanistheGeoQapproach
mentioned above. Using this approach theGeo Risk
Scanwas executedfocussingfurthermoreonaspects
suchas:
Deviationbetweenprocessandcontent
Withinthecontextof aproject, thescanisexecuted
fromamoregenericanalysistomoredetailedanal-
yses onspecific points of interest for theprojects
scanned
Anyscanstartswithaqualitativeanalysis; quantita-
tiveanalysesareonly performedwhenconsidered
necessary, basedonthequalitativeanalysis
Thefollowingfour stages areidentifiedregarding
thesebasic assumptions. Thefirst two stages in fact
formtheGeoRisk Scan; thelatter twostages canbe
completedwithinaproject, dependingontheresults
of thefirst twostages.
1. GeoQuickScan qualitativeprocesstest;
2. GeoCheck qualitativecontent andproduct test;
3. GeoRiskAnalysis quantitativecontent analysis;
4. Implementation Geo Risk Management as a
routineworkprocess
3 EXECUTIONOF A GEORISK SCAN
3.1 Stage 1: Geo Quick Scan
3.1.1 Execution of a Geo Quick Scan
Inorder tobeabletoperformthisstage, first onehas
to gain insight in theproject objectives and context.
Therefore, an interview is planned with the project
management team. It is important to haveat least an
interviewwiththetechnical project manager, who is
normallyresponsibleforthetechnical partof aproject.
For larger projects, itcanbeof goodhelptointerview
the risk manager (when present within the project),
project leadersof specificelementsof theproject and
thecontract manager.
The interview is based on a standardized ques-
tionnaireanddeals mainly withtheGeoQ approach.
Examplesof questionsare:
IstheGeoQapproachrecognizableinthescanned
project?
Have all six steps been fully elaborated in each
project phase? Is everything donein order to get
goodresultsfromtheriskmanagement steps?
Haveall sixstepsbeenexplicitly elaboratedineach
project phase?
Itisimportanttoknowwhether astepisperformed
explicitly, following a plan or just as some sort of
unawarecoincidence. Ingeneral, whenastepis only
performedimplicitly, it isnot guaranteedthat innext
projectphasesorinotherprojectsthesameriskdriven
projectmanagementisapplied.Thiscouldcausenega-
tiveconsequences. Further insightisgainedbyasking
for the products available fromthese steps and the
knowledgeandtoolsthathavebeenusedintheproject
toassist intheelaborationof thesteps.
3.1.2 Results of Geo Quick Scan
Elaboration of the interview and studying of the
gathered information make it possible to evaluate
the Geo Quick Scan. Scores are based on table 1
(made for each project phase) and the accompany-
inglegend. Moreover, theapplicationof thesix main
lessonslearned(nextchapterinthispaper) ischecked.
Besidesthisscore, recommendationsareprovidedfor
improvement of thegroundrelatedrisk management
process.
340
Table1. ScoringtheGeoQuickScan.
Degreeof Degreeof
Stepin explicit complete
GeoQ-approach execution execution
1. Settingobjectives
anddatacollection
2. Riskidentification
3. Riskclassification
andquantification
4. Riskremediation
5. Riskevaluation
6. Transfer of risk
information
1point: stepisnt executed
2points: implicitly, but not fullyelaborated
3points: explicitly, but not fullyelaborated
4points: implicitly, andfullyelaborated
5points: explicitly, andfullyelaborated
Thetotal amount of pointsfromtable1givesthejudgement
andareport mark:
-20veryinsufficient 4
2022insufficient 5
2224moderate6
2426sufficient 7
2628good8
>28excellent 9
3.2 Stage 2: Geo Check
3.2.1 Execution of a Geo Check
Thework intheGeoCheck phaseisfocussingonthe
pointsof attentionresultingfromtheGeoQuickScan.
TheGeoCheck dealsparticularly withthecontent or
quality of the projects ground related risk manage-
ment. Analyses andcalculations fromtheproject are
checked qualitatively by making use of experienced
geotechnical engineers. Newcalculationsarenot per-
formed in this stage. Theprimary focus is checking
thework already performed by theproject organisa-
tion. For example, thefollowingquestions shouldbe
answeredduringtheGeoCheck:
Arecorrect startingpointschoseninrelationwith
boundaryconditionsof theproject?
Areall relevant groundrelatedrisksidentified?
Are calculations been performed for the relevant
identifiedrisks?
Areappropriatemodelsandtechniquesapplied?
Areresultsof theperformedanalysesaccordingto
expectations?
3.2.2 Checklists
Despite the experience of geotechnical experts, it is
of major importance to assure all foreseeable risks
are indeed identified. Therefore, using standardized
checklists is very useful. These checklist have been
developedforbuildingpits,roadsanddikesforquickly
gaining insight in thecompleteness of theidentified
groundrelatedrisks. Thesechecklistsprovedtobeof
goodassistanceinall performedGeoRiskScans.
Figure1. Structureof checklists.
Table2. Risktable.
Riskafter
implementation
of measures
Unwanted recommended
event Probability Consequence Risk inscan
All risksinthechecklistsareclassifiedasgeotech-
nical risks, geohydrological risks, geo-ecological
risks,risksrelatedtoobjectsorobstaclesintheground,
risks related to contract requirements or construc-
tion risks. All risks aredescribed in terms of causes
andconsequences.Theconsequencesarebydefinition
unwantedevents.
By usingthis structureof thechecklists, it is pos-
sible to use themon different scales. If the project
is still inthefeasibility phase, risk identificationcan
bedoneonly onthescaleof unwantedevents. When
moredetail is required, onecanwork fromcauses to
subcausesandestimatetherisksaccordingly.
3.2.3 Results of Geo Check
WhenaGeo Check is performed, theproject organi-
sationgainesinsight inthefollowingquestions:
Areunacceptablegroundrelatedriskspresentinthe
project?
Havealreadyriskremediationmeasuresbeeniden-
tifiedfor theserisks? (basedonexpert judgement
duringtheGeoCheck)
Whichunacceptablerisksarestill present?
Answers on these questions are described as rec-
ommendations for improving thein-depth quality of
ground related risk management. Besides these rec-
ommendationsarisktable(table2) ispresented. Such
risk tables provedto beamorepractical way of dis-
playingtheprojectrisksthanconventional riskgraphs
(figure 2), without loosing insight. Finally, the Geo
Checkisevaluatedbygivingareportmark onascale
of 1to10, basedonexpert judgement.
3.3 Results of Geo Risk Scan
After execution of the Geo Quick Scan (process)
andtheGeo Check (content) atotal overviewof the
degreeof aprojectsgroundrelatedriskmanagementis
available. Rijkswaterstaataskedfor aprojectportfolio
of all scannedprojectsinorder tobeabletocompare
341
Figure2. Riskgraph.
Figure3. Qualitymatrixandproject portfolio
theresultsof thedifferentprojectscans.Figure3shows
the matrix which made this possible. Each scanned
project canbeplacedinthematrix.
Content is evaluated as of more importance than
process. After all, when the results of a project are
good,theprojectobjectiveswill notbeaffected.There-
fore, projects with bad scores for the Geo Quick
Scan still can get a good or moderate overall score.
Nevertheless, these projects should keep focus on
improvingtheprocessof groundrelatedriskmanage-
ment.Maybe,itwasonlyacoincidencethatthecontent
part of groundrelatedriskmanagement of theproject
hadgoodresults!
3.4 Stage 3: Geo Risk Analysis
Theaimof theGeoRiskAnalysisstageistoimprove
projectsgroundrelatedrisk management, either with
focusonprocess, orwithfocusoncontentbyperform-
ing extensiveand if necessary quantitativeanalyses.
ExecutingtheGeoRiskstage, aproject inthebottom
leftcornerof thequalitymatrixinfigure3shouldmove
totheupperrightcornerof thematrixbyexecutingthe
GeoRiskstage.
Analyses are executed on unacceptable risks as
identifiedintheGeoCheck andrecommendationsof
bothGeoQuick ScanandGeoCheck areelaborated.
If necessary, advancedrisk management toolscanbe
used, as well as geotechnical calculations. Examples
are the use of an Electronic Board Roomfor brain-
storm/expert sessions, contractual risk allocation by
theGeotechnical BaselineReport,model experiments,
fieldmonitoring, andsoon. Thismakesit possibleto
analyseandquantifyanyremainingunacceptablerisks
in order to select and execute proper measures. The
project teamitself can executework during theGeo
RiskAnalysisstage.
At the end of the Geo Risk Analysis stage, the
optimal riskassessmentstrategycanbechosen. Possi-
bilitiesareavoidingtherisk, reducingriskprobability
andor consequence, andrisktransfer toathirdparty.
3.5 Stage 4: Implementation
All informationgatheredandrecommendationsfrom
thepreviousstageswill notimprovetheprojectswork,
unless it is implementedintheprojects work. There-
fore, the implementation can be seen as the most
importantstage! Implementationhastobedonebythe
project teamitself. This stageis beyondthescopeof
thispaper andfor instanceelaboratedinVanStaveren
(2009).
4 LESSONSLEARNED
The evaluation of the five scanned projects resulted
in six main lessons which are further elaborated in
20recommendations. All lessonsaredescribedinthis
chapterof thepaper.Thelessonsaresubdividedintwo
maintypes.
The first set of lessons may help to improve the
applicationof well-structuredgroundrelatedriskman-
agementduringprojects.Theselessonsareinteresting
for owners who areresponsiblefor thegeotechnical
conditionsintheir projects, aswell asfor contractors
or engineers who haveto managegroundconditions
towards successful project results. Theselessons are
referredtoaslessonsdealingwithcontent (C).
Thesecondsetof lessonsteacheshowthecoordina-
tionanddelegationof managinggeotechnical risk by
ownerscanbeimproved. Theselessonsseemparticu-
larlyrelevantforownerswhouseinnovativedesignand
buildtypeof contracts andarereferredto as lessons
dealingwithprocess(P).
4.1 Lesson 1 Clear risk management positioning
Lesson 1 concerns thepositioning of ground related
riskmanagement withintheproject.
4.1.1 Ground related risk management should be
an integral part of project risk management,
but with explicit status (P)
Inall of thefivescannedprojects, groundrelatedrisks
were an integral part of the total risk management.
Fromaproject management point of view, thisseems
agoodstrategy,becausemoreaspectsthanonlyground
relatedrisks areof importancefor aproject. Froma
geotechnical specialist point of view, this gives the
342
opportunity to give ground related risks the proper
attention.
However, ground related risks need special
attention, having specialists dealing with them
and executing specific remediation measures. Most
remarkable is that ground related risks have mainly
consequences during the construction and mainte-
nance of the project. Consequently, these risks are
often not given the attention they need, or thought
about as solvable, during the design phase. In each
specificprojectitisrecommendedthatinearlystages
geotechnical experts determine whether or not this
mayresultinunacceptableriskslateronintheproject.
Therefore, groundrelatedrisksneedanexplicitsta-
tus in the total risk management. In two of the five
scanned projects this approach was used with good
results.
4.1.2 All specific ground related risks should be
part of a projects risk register (P&C)
All ground-related risks should not only have an
explicit status, they shouldalsobepart of theproject
risk register. Often, only imprecise ground-related
risks are part of the project risk register. For exam-
ple, phraseslikesoil investigationisinsufficient for
makingagooddesign. Suchfuzzydescriptionsmake
explicit risk management difficult andprobably even
impossible. It is unclear which measures have to be
takenandwhat theanticipatedeffectsare.
Therefore, it is recommended that the ground-
relatedrisk register is part of theoverall project risk
register.
4.2 Lesson 2 Clear risk management
responsibility
Lesson2highlightstheimportancethatanyidentified
ground related risk needs oneor moreowners. Oth-
erwisetherisk will not get therequiredattentionfor
adequateremediation.
4.2.1 Appoint a coordinator who is responsible for
ground related risk management (P)
Scanningthefiveprojects showedtheimportanceof
somebody in the project acting as a coordinator of
all ground related issues. The quality of the project
largely improvedby suchacoordinator. It isnot nec-
essary that this person also is responsible for the
ground related risks themselves. Thetechnical man-
ager of a big infrastructural project is usually too
busytogivegroundrelatedriskstheproper attention.
Thementionedcoordinator shouldassistthetechnical
manager.
4.2.2 All ground related risk should be allocated
contractually to one or more of the parties
within a project (P)
Becauseof theinherentgroundrelateduncertaintyitis
very important tocontractually arrangetheresponsi-
bilities for unwanted events caused by differing soil
conditions. At least, it is important to think about
theconsequences, whenignoringrisks causedby the
uncertaintyaccompanyingthesoil.
Onecouldsimplydivideall riskstooneortheother
party, butoftenpartial riskallocationispreferred. For
instancetheprinciplesandpracticesof thegeotechni-
cal baselinereport (GBR) arerecommended (Essex,
2007). The main principle is to allocate any risk to
thepartyinvolvedthat isbest abletomanagetherisk.
Sometimessharingarisk ispreferred, asbothparties
are(un)ablethemanagetheriskbytheir own.
4.2.3 Ground related risks, completely allocated to
the contractor, needs still evaluation by the
client (C)
Inintegratedcontracts,manyrisksaretransferredfrom
the client to the contractor. However, the client still
bearsconsequenceswhentherisksoccur.Thisisespe-
cially thecasefor immaterial consequences, likeloss
of reputation,safetyorpolitical risks.Theprojectman-
agement teamcan use monitoring and other quality
checksinorder tokeepcontrol over theserisks. These
checksshouldnot onlybeprocesschecks, but should
alsoincludeindepthanalysesof content.
4.3 Lesson 3 Clear risk communication
Lesson 3 stresses theimportanceof transparent risk
communication between all parties involved in the
project, asearlyintheproject asfeasible.
4.3.1 Link explicitly the functional and technical
level of project organisation to each other
(C&P)
All five scanned projects used integrated contracts,
wherethecontractor alsohadtodothedesignor even
financeandmaintenance.Thisimpliesthattheproject
organisation has to pay much attention to the func-
tional description of project specifications. By this
way, all identifiedrisksneedtobetransformedtothe
contractor.
Duringscanningof thefiveprojectstwohandicaps
wereshown:
(Geo)Technical experts have difficulties in trans-
lating their recommendations to this functional
level.
Ontheother hand, project managershavedifficul-
tiesintranslatingthetechnical requirementsof the
expertstofunctional requirements.
Only oneof thefiveprojects excelledin this link
betweenprojectmanagementandgroundrelatedtech-
nical experts. This precious link was formed by one
person who could speak both languages. This is
recommendedfor everyproject.
4.3.2 The risk file of a client should be known by
the contractor and vice versa (P)
Everyprojectorganisationof thefivescannedprojects
had adilemmaabout sharing their risk filewith the
contractor. Many different concepts of sharing this
information(or not) wereencountered.
343
Onemightthinkitisdesirabletoshowthecontrac-
tor all identified(ground-related) risksandviceversa.
By doing so, project organisations however feel like
attracting responsibility to themselves, because the
informationgivento thecontractor cancontainmis-
conceptions. Though, thislast optionisnot esteemed
to becorrect. Risks areonly transferredas points of
attentionsandcannot bewrong. Another rationaleis
thatwithinnovativedesignandbuildtypeof contracts,
onemightbepushingthecontractorinsomedirection,
whenexchangingriskinformation.Afterall, theinten-
tionof anintegratedproject is to usetheknowledge
andexperienceof thecontractorforthedesignandthe
client shouldnot giveimplicit directionstoadesign.
Balancingbetweenthesebothways, frompoint of
view of a professional client, it is recommended to
alwaysexchangerisk informationbetweenparties; at
least after thetender phase.
4.3.3 Tests should be applied on feasibility of
requirements with a geotechnical scope (C)
Requirements with a direct relation to (geo) techni-
cal aspectscanbestricter thanthemaximumaccuracy
of predictions in the design and construction phase.
For example, settlement requirements for roads are
oftenmorestringentthancanpossiblebedesignedand
constructedwithinreasonusingstate-of-the-art tech-
niques and models. Negative consequences are over
dimensioningorrelativelybigeffortsformaintenance.
Itisthereforerecommendedtoverifythefeasibilityof
thosetypesof requirements.
4.4 Lesson 4 A ground related risk register
Lesson4underlinestheimportanceof acorrect, com-
pleteand up-to-dateground related risk register and
givesrecommendationsfor realization.
4.4.1 Description of a risk needs to satisfy basic
demands (C&P)
This is a general recommendation that is applicable
on all types of risks; also ground related risks. It is
importanttodescribeariskintheriskfilewithatleast
thefollowingaspects:
Cause(inwords) of therisk
Consequence(inwords) of therisk
Determinationof probabilityof therisk
Determination of amount of consequences of
therisk (failurecosts, quality loss, delays, loss of
image& publicconfidence, etc.)
Possibleremediationmeasures
Owner of therisk
Responsibilityfor therisk
4.4.2 All GeoQ steps should be explicitly executed
step by step during each project phase (P)
Followingthesix risk management steps will leadto
good risk control of a project and hence a correct
andcompleterisksregister.Thisisgenerallyaccepted.
Therefore, acrucial aspect inall GeoRisk Scanswas
theexaminationof thewaythesestepswerefollowedin
Figure4. Needof earlyrisksessions.
thescannedprojects. However, lookingbackatproject
activitiesit alwaysseemedpossibletodividethemto
thesix risk management steps. Oneneeds to keepin
mindthatreal riskmanagementisonlypossiblewhen
thesestepsaretakenwithanexplicit plan.
Riskmanagementstepsareperformedwithacyclic
approach. Consequently thestepsarenot alwaysper-
formedinsucceedingorder. For examplegatheringof
extrainformation(executingextrasoil investigation)
ispartof step1,butcanbedoneasameasureidentified
instep4. Therefore, whenrisk management stepsare
notperformedinsucceedingorder,thereneedstobean
inducedexplanationby therisk management process
itself (seepreviousexample).
4.4.3 A ground related risk session should be
organized in early project phases (P)
Inearlystagesof aproject littleinvestmentsaremade
andsteeringpossibilitiesarestill high.Thisunderlines
the importance of risk sessions in the early project
phases. However, technical risk sessions are often
ignoredduringthefirstprojectstagesbecauseproject
management conceives technical risks as solvablein
later project stages. Asaconsequenceof thisassump-
tionoptimal technical solutions might beoverlooked
in the early stage of the project. This will require
muchmoreeffortinlaterprojectstages.Thereforealso
technical risk sessionsshouldbeplannedintheearly
stages.
Oneof thescannedprojectsprovedthistobetrue.
Fromanearlyrisksessiononemajor constructionrisk
was identified. Extrasoil investigationwas executed
and the risk got special attention during the tender
phasein which thecontractor had to makeaplan to
showhowthisspecificriskwasmanaged.
4.4.4 Communication should be explicitly risk
based between project and third parties (P)
A projectorganisationcancall inthehelpof thirdpar-
ties (e.g. for soil investigation, monitoring, technical
advice,etc.).Thirdpartiesareoftencalledintomanage
(implicitly or explicitly) identifiedrisks. It is impor-
tanttocommunicatetheserisksasaninstruction.This
ensures that theright analyses areexecuted by third
parties.
Theother way around it is also important for the
project management toask thethirdparties toreport
identifiedriskstothem.Thisensuresamorecompleted
344
risk fileof theproject andno important information
getslost.
4.4.5 Explicit guarantee on completeness and
correctness of ground related risks
and analyses is necessary (P)
Experts are of major importance in soil related risk
management. Caused by theuncertainty of geotech-
nique, almost always different interpretations of the
same risk can be expected. Guarantee on complete-
nessandcorrectnessof groundrelatedriskstherefore
is important. Onecanthink of bigrisk sessions with
manyexperts, colleaguechecks, secondopinionsand
useof checklistsinorder toguaranteethis.
4.4.6 Checklists are recommended as a check on
the completeness of risk files (C)
Accompanying the previous point checklists have
proventobeof goodhelpinorder tocheck thecom-
pletenessof riskfiles. Inparagraph2.4.2isshownhow
agoodchecklist canbeimplementedandused.
4.5 Lesson 5 Risk driven site investigations
4.5.1 Site investigation should be explicitly risk
based (C)
Fromagood(groundrelated) implementedriskman-
agement process, in situ ground investigation and
supportinglaboratory researchcanbeidentifiedas a
good measure. After all, Performing in situ soil and
labresearchwill gaininsight inspecificrisks.
Therisk management that let totheperformingof
theinsitusoil andlabinvestigationshouldbeextended
intheplanfor theinvestigationitself. Six basic steps
canbeused(Staveren, 2006) inorder to besurethat
thecorrect informationisgathered. Inshort:
1. Determinegroundrelatedconstructions;
2. Determinemainmechanismsthataffectthefit-for-
purpose;
3. Determinetherisks if theidentified mechanisms
act adversely;
4. Determinethedesigntechniquesfor theidentified
measurements;
5. Determinethemost critical groundparameters;
6. Determinetheinsitusoil andlabinvestigationcon-
sideringthegroundparametersandthegeological
heterogeneity.
4.5.2 In situ soil and lab research should be
flexible executed (C)
Byusingaflexibleapproachof siteinvestigationit is
possibleto adjust to theobtainedresults duringexe-
cution. On theonehand moredetailed research can
beexecuted if moreheterogeneity is encountered as
expected. Ontheother handtheresearchcanbedone
morebroadif thereisnoreasonfor moredetail.
4.5.3 Quality control of site investigations
is necessary (C)
Labandinsitusoil researchareusedincalculations
in order to make analyses. It is therefore of great
importancethat theparameters derivedfromlaband
insitusoil researcharereliable.Therecanalsobecon-
tractual consequences whensoil investigationresults
weresendtothecontractorbutareproventobewrong.
Questionmarkswerestatedinthreeout of thefive
scannedprojectsregardingthequalityof theinsitusoil
andlabresearch. Especially withspecializedexperi-
ments one should critically apply quality control. A
geotechnical specialistshouldbeabletoperformthese
checks.
4.6 Lesson 6 Risk driven field monitoring
4.6.1 Monitoring should be used as a tool for
guarantee of quality and control of risks (C)
Field monitoring is an excellent tool for controlling
ground-relatedrisksduringtheconstructionandoper-
ationphasesof projects. Obviously, theseprogrammes
needtobedefinedaccordingtotheriskprofileof the
project. With integrated contracts, often monitoring
is coordinated by thecontractor. However, theclient
should always check theresults of theapplied mon-
itoring for the key-risks of the project. Monitoring
shouldnot only becheckedaccordingtotheprocess,
butregularin-depthanalysesof contentshouldalsobe
applied.
4.6.2 Ground related risks should have an explicit
place in a monitoring plan (C)
Monitoring can be broader than only measuring for
controlling ground related risks. However, ground
relatedrisksshouldhaveanexplicitplaceinamonitor-
ingplaninordertomakesurethemeasurementsmake
sensefor thegroundrelatedriskstobemonitored.
5 CONCLUSIONS
Unexpectedgroundconditionsappear amajor source
of cost and time overruns in infrastructure projects,
whichisconfirmedbyrecentDutchresearch.Thepre-
sented Geo Risk Scan proved to bean effectivetool
for quickly providing information about the degree
and the quality of ground-related risk management
in infrastructureprojects. Six main lessons and sup-
portingrecommendations arederivedfromusingthe
GeoRiskScaninfivemajor Dutchprojects.Themain
lessonsare:
1. Clear riskmanagement positioning
2. Clear riskmanagement responsibility
3. Clear riskcommunication
4. A groundrelatedriskregister
5. Riskdrivensiteinvestigations
6. Riskdrivenfieldmonitoring
Theselessonsseemtobegenerically applicablein
construction projects. Ongoing application of these
lessonsinDutchprojectsprovesthisconclusion.
345
ACKNOWLEDGEMENTS
The authors are grateful to all professionals who
havebeeninterviewed, or whootherwisecontributed,
during performing the Geo Risk Scans in the five
infrastructureprojects.
REFERENCES
AvendanoCastillo, J.E., Al-J ibouri, S.H. andHalman, J.I.M.,
2008, Conceptual model for failurecostsmanagement in
construction, Proceedingsof theFifthInternational Con-
ference on Innovation inArchitecture, Engineering and
Construction(ACE), Antalya, Turkey, J uly2325.
Essex, R.J. (ed.), 2007, Geotechnical Baseline Reports for
Construction: SuggestedGuidelines. Danvers: ASCE.
RISMAN, www.risman.nl
Staveren, M. Th. van, 2006, UncertaintyandGroundCondi-
tions: A RiskManagementApproach, Elsevier Ltd.
Staveren, M.Th. van, 2009, Suggestions for implement-
ing geotechnical risk management, Proceedings of the
SecondInternational SymposiumonGeotechnical Safety
andRisk, Gifu, J apan, J une1112(inpress).
346
Reductionof landslideriskinsubstitutingroadof Germi-Chaydam
H. Farshbaf Aghajani
M.Sc in Geotechnical Engineering, Ashenab Consulting Engineers Co., Tabriz, Iran
H. Soltani-J igheh
PhD of Geotechnical Engineering, Technical Faculty, Azerbaijan University of Tarbiyat Moallem, Tabriz, Iran
ABSTRACT: Germi-Chaydamisanearth-fill damwithcentral claycorethat isunderconstructionacrossthe
Germi-Chay River inIran. Duetodamconstruction, part of themainroad, that connects East Azerbaijanand
Ardabil provincestogether, remainswithinthedamreservoir.Therefore, itisnecessarytoconstructasubstituting
road, 5kmlong, outsideof thereservoir. Part of thesubstitutingroadislocatedonthenatural groundslopeis
susceptibletolandslide. Inthispaper, first, thestabilityof theroadslopeisverifiedbyperformingabackanalysis
for theslipsurfaces.Then, anappropriateschemeissuggestedfor constructionof theroadembankmentinorder
toattainpermanentstabilityof theslope. Evaluationof measureddeformationsshowthattheslopedisplacements
havedecreasedconsiderablyandslippingof theslopehavestoppedafter executingthestabilizationtreatment.
1 INTRODUCTION
Landslideisoneof themost important geohazardsin
geotechnical engineering that can beathreat for the
stability of thestructures. Thepotential of landslid-
ing may increases when the slope is consisting of a
weaktextureandalsosubjectedtogroundwater rising
bylocatinginvicinityof awaterway. Constructionof
infrastructures,suchastheroadsandbuildingsoverthe
mentioned slopes may increases the geohazard risk,
andevencanleadtoslideof theslopeandconsequently
failureof theconstructedstructures.
Toavoidfromtheunfavorableincidentsandtoguar-
anteestabilityof structures, thelandslideriskmustbe
managedandreducedbyusingtheefficient andprac-
tical methodsaswell aswithconsideringof thecosts.
Someresearchershaveinvestigatedtheriskassess-
ment andmanagement of thelandslidesbystatistical,
analytical andgeological methods(Lessingetal,1983;
Fell, 1994; Dai etal, 2002). Someof thesemethodsare
appropriate for risk assessment of land slide occur-
rence. However, afewnumbersof methodshavebeen
presented for hazard risk reduction and treatment of
theslippedslopethatareperformableinpracticewith
lowcosts.
In this research, a practical and efficient method
is employedto reducethelandsliderisk of aslipped
slope, as a subgrade of a road, in Iran. The geolog-
ical formations, landslide potential, and stability of
theslopearefirst investigated. Then, an appropriate
methodologyisproposedfor constructionof theroad
embankment with considering the necessity of cur-
rent land slide risk reduction. Also, for examining
the efficiency of proposed method, slope deforma-
tionsaremonitoredatregularintervalsduringandafter
embankment construction.
2 SUBSTITUTINGROADOF GERMI-CHAY
DAM
Germi-Chaydamisanearth-fill damwithcentral clay
core that is being constructed over the Germi-Chay
River, located about 220kmof north-east of Tabriz
city inEast Azerbaijanprovinceof Iran. Theheights
of thedamfrombedrockandriverbedare82and62m,
respectively. Also, lengthandwidthof thedamat the
crest arerespectively 730 and 10 meters. Themaxi-
mumwater level of reservoir is 1460 meters higher
thansealevel. Themainpurposesof theGermi-Chay
damare irrigation of farmlands and supplement of
urbandrinkingwater.
Unfortunately, dueto damconstruction, apart of
themainroad, thatconnectsEastAzerbaijanandArd-
abil provinces together, is located within the dam
reservoir. For thisreason, asubstitutingroadisbeing
constructed, with the length of 5km, outside of the
reservoir. A part of thesubstitutingroadat kilometer
2+065islocatedinthenatural groundslopethat isa
slippedareaandsusceptibletolandslide.
Thegeological investigationof theareashowsthat
theslopeiscomprisedof weak shalematerial located
ontheigneousbedrock. Furthermore, awaterwayhas
beenlocatedinthevicinityof subgradeslope. Sothat
during the rainy seasons, ground water rises within
theslopeandcausestooccur slipsandtensioncracks
in the subgrade slope. Field measurements indicate
that the values of horizontal and vertical displace-
ments during9months are714and346centimeters,
respectively. A view of cracks and slips occurred in
thesubgradeslopearepresentedinFigures 1and2,
respectively.
With respect to the importance of substituting
road fromthe view point of linking two provinces
347
Figure 1. Tention cracks near borehole BH 55 due to
slipping.
Figure2. A viewof slipedinthesubgradeslopeaswell as
substitutingroadaxis.
andits location on theslippedarea, thestabilization
of the subgrade slope and securing of road safety
permanentlyisconcernedinthisresearch.
3 GEOLOGY OF LANDSLIDINGAREA
To investigatethegeological and geotechnical prop-
erties of thestudiedarea, anumber of six boreholes,
denoted by BH-53, BH-54, BH-55, BH-56, BH-57,
andBH-R1, aredrilledinthesubgradeslope.
Theresultsof thesubsurfaceexplorationsandlab-
oratory tests showed that thegeological structureof
the subgrade slope consist of the recent alluvium
(TR), quarternerary traces (TA), crashed red rhyo-
dacite(RhD), quartz diorite& quartz monzo-diorite
(QD& QMD) andalternationof greyshalewithyel-
lowsandstone(Sh& S). Details of theseformations
havebeenillustratedinFigure3(Ashenab, 2005).
Therecent alluviums havelocated at thesides of
waterway and comprised of silt and sand mixtures.
Thethicknessof thisformationislessthan3meters.
Thequarternerary traces comprisemixtures of fine-
grainedsoilsandsomesandandgravel.
Figure3. Geological mapof studyingareaas well as land
slidingmasswithsubstitutingroadaxis.
Sedimentaryformationsconsist of shaleandsand-
stone materials frequently have been repeated with
depthwithintheslippedzone, andexposedatthekilo-
meter 2+065of thesubstitutingroad. Atdeeplevels,
amount of sandstone mass increases and sometimes
thesedimentarylayerisaloneconsistingof sandstone.
Shalemassisintheformsof claystoneandsiltstone.
In order to investigate properties of shale material,
a number of experimental tests were carried out on
the samples obtained fromBH-54 and BH-55 bore-
holes. Theresults indicatedthat thesematerials have
lowshear strengthwithaverageliquidlimit (LL) and
plasticity index (PI) of 42and 17, respectively. Also
accordingto UnifiedSoil ClassificationSystem, the
majority of samples are categorized as CL (ASTM
1997). Because of weak texture of sedimentary for-
mations, preciselydeterminationof layersstrikesand
their dipsisdifficult. However, siteexplorationsshow
thattheslipplaneof slopedoesnotcoincidetothedip
directionof sedimentarylayersof shalematerials.
Attherighthandof thewaterway, geological forma-
tionshavebeenmadefromtherhyodacitemass. Also,
at the upper elevation of subgrade slope, rhyodacite
masswithalmost 50metersdepthislocatedover the
shalelayer.Therhyodaciteformationhasgoodquality
andstrongproperty whichthefragments areusedas
concreteaggregates.
Bedrock formations of quartz diorite & quartz
monzo diorite materials have been embedded under
thementionedlayers. Thebedrockhasrelativelygood
qualityandhighstrengthwithafewjoints. According
tothesiteexplorations, thedepthof bedrockatBH-53,
BH-54andBH-55boreholes are10, 14.8and3.5m,
respectively.
Theslopehas slippedbecauseof seasonal raining
andweakconditionof shalemass. Figure3illustrates
theboundaryanddirectionof slips.Also, ageological
longitudeprofileof theslopealongtheslipdirection
(i.e. G-H cross section in Figure 3) is presented in
Figure4.
Asillustratedinthisfigure, theslippedareacanbe
dividedinto two separateportions. Thefirst portion,
specifiedby S.S.1area, includesthelandslipsinthe
upperelevationsof theslopeoccurredduetothemove-
mentof rhyodacitemassontheshalelayer.Thevalues
348
Figure 4. Geological profile of slope (G-H), locations of
slipsurfacesof 1and2aswell groundwater level.
of displacements at benchmarks installedontherhy-
odacitemassareabout afewcentimeters. Thisfact is
duetoexistenceshallowbedrockinfrontof rhyodacite
mass, whichwithstandagainst themovements.
Thesecondportionof theslipsisgroundmovement
in thelower elevation of theslopethrough theshale
material, i.e. S.S.2areashowninFigures3and4.
Field observations indicated that the amount of
movementsprogressivelyhaveincreasedduringrainy
seasons.Also, subsurfaceexplorationsdeterminedthe
depthof slidingabout10m.Thisrelativelylowdepthis
relatedtothenon-conformingof shalelaminationand
slipdirectionaswell asincreasingof fractionof sand-
stoneat lower depths. Althoughthestrongsandstone
has extended up to considerable depth, the amount
of horizontal movements is much more, about 714
centimetersduringaperiodof ninemonths.
4 BACK ANALYSISOF SLIP
Prior tostabilizethesubgradeslope, it isnecessaryto
knownthemechanical parametersof materialslocated
amongtheslipingsurfaces.Themostefficientmethod
for determinetheseparametersisperformingaback-
stability analysis for slipped slope (Sabatini, 2002).
Withrespect totheprogressivemovement of thesub-
gradeslope, itcanbeconcludedthattheshalematerial
along the slip surface have reached to the resid-
ual condition. In this condition, the soil cohesion is
negligibleandtheeffectivefrictionanglemaybedeter-
mined by performing a stability analysis assigning
safetyfactorof 1.0(USACE, 2003).Theback-analysis
was performedwiththelimit equilibriummethodby
usingSLOPE/Wsoftware. Mohr-Coulombcriterionis
utilizedfor modelingof material.
Sincetheslipsurfaceisknowninthefield, it must
bedefinedcarefullyfor back-analysis. Therefore, the
slip surfaces (S.S.1 and S.S.2) with identified situa-
tion areconsidered as anarrow band madeof weak
material withunknownfrictionangleanddiffer from
theother material of slope. It isnecessarytonotethat
theSLOPE\Wsoftwareisunabletomodel theinter-
faces between different materials in order to define
slipsurfaces.Thus, byapplyingthisapproach, itisnot
necessarytoreducetheparametersof wholeslope.
TheMohr-Coulombparametersof other materials,
suchascohesionandfrictionangle, areobtainedfrom
conductingtriaxial anddirect shear testsonthesam-
ples retrievedfromboreholes. Density, cohesionand
Table1. Strengthparameter of slopematerials.
Friction Cohesion Density
Material type Angle (kPa) (kN/m
3
)
Rhyodacite 30 10 22
Shale 23 100 20
SlipSurface1
15 20 19.5
SlipSurface2
15 15 19.5
*Obtainedfromtheback-analysis
frictionangleof theslopematerials arepresentedin
Table1.
Subsurfaceexplorations determinedgroundwater
level 2mbelowthegroundsurface. Twodimensional
geometryof theslope, materialsof layers, locationsof
slipsurfacesaswell asgroundwater level areshown
inFigure4.
Thevaluesof thefrictionangleandcohesionof slip
surfacematerialsobtainedfromtheback-analysisare
presentedinTable1.Comparisonbetweentheobtained
resultsandrecommendedvaluesinliteratureshowsthe
accuracyof backanalysis(Bowles, 1996).
5 A SCHEME FOR SUBGRADE SLOPE
STABILIZATION
Since the safety of substituting road was threatened
by future probable sliding of subgrade mass totally
duetolargedeformationsexpectedinS.S.2area, it is
necessarytoseek aremedial treatment for stabilizing
of thesubgradeslopewithconsideringpractical and
economical feasibility.
Inlast decades, many approaches andmethodolo-
gieshavebeenintroducedforstabilizingandtreatment
of the slopes. These methods can be categorized in
the following groups: slope geometry modification,
control surfacewaterandinternal seepagecontrol,pro-
vide retention, increase soil strength with injections
and soil reinforcement (Hunt, 2007, Cheng & Lau,
2008). Selection of appropriate method is based on
several factors suchas practical feasibility, economy
andavailablefacilities.
InGermi-Chayproject,utilizingtheslopegeometry
modification method required enormous excavation
andthushighcost. Becauseof slippedandweak tex-
tureof shalemass, theefficiencyof soil reinforcement
alsoisunconfident.
As mentioned in geological description, bedrock
embedded under the slope at S.S.1 area has been
locatedat shallowdepth. Thus this featureis utilized
toconstruct aretainedsystemalongwithappropriate
drainagecapacity.
Therefore, tostabilizetheslope, it wasproposedto
excavatethefoundationof roadembankmentuptothe
bedrock level. Then, thefoundationtrenchwasfilled
withrockfill material toreachtotheidentifiedlevel of
roadembankment and, finally, theroadembankment
was completed. It expected that the road embank-
ment plays roleas abarrier against slopeslidingand
349
Figure5. Geometryof theslopeafterstabilizationandroad
embankment.
Figure6. Potentiallyslipsurfaceafter stabilization.
decreases the movement of the upper mass because
of high strength and stiffness of rockfill material
usedintheembankmentanditsappropriategeometry.
Figure 5 illustrates the stabilization scheme of sub-
gradeslopeafter stabilizationprocessaswell asroad
embankment section.
Theproposedschemeforstabilizationwasmodeled
andanalyzedusingSLOPE/Wsoftwarefor evaluating
safetyof theslopeandroadaswell asfor determining
supposed slip surfaces. Also, the factor of safety of
S.S.1surfacewas determinedby stability analysis of
theproposedgeometry. As indicatedinFigure6, the
resultsof theanalysisshowthat inthepresenceof the
road embankment, thesafety factor of S.S.1 surface
increasesfromnearly1.0(asathresholdconditionin
stability analysis) to2.545. Moreover, thesafety fac-
tor of theembankment constructedonthebedrock is
aboutof 3.85. Sincetheembankmentmaterial iscom-
prisedof rockfill material withhighdrainageproperty,
thephreaticlinelocatesinalower level andthesafety
factor becomesmorethantheoretical calculation.
In addition, a culvert was performed within the
roadembankmenttoconductthesurfaceflowstoward
downstream; hereby the safety increases from this
point of view.
For economic evaluationof proposedstabilization
method, thecosts of this method was calculated and
compared with those of the geometry modification
method.Theresultsindicatedthatinproposedmethod,
the volumes of subgrade excavation and foundation
fillingare21650and16100cubicmeters, respectively.
However, for thesecondmethodrequiredexcavation
is256000cubicmetersaswell as94000cubicmeters
of rockfillingfor embankment constructiontoalevel
higher thanthenormal water level of reservoir. Thus,
theexcavationvolumeof secondmethodisalmost 12
timesgreater thanproposedmethod.
Figure 7. Excavation and filling processes of the strips
duringconstruction.
6 METHODOLOGY OF THE EMBANKMENT
CONSTRUCTION
Inspiteof thecurrent movement of theslopeandlow
strength parameters of shale material, it is expected
that thesubgradeslopeexcavationupto thebedrock
may lead to instability and hazardous in the slope;
particularlythat thetrenchlocatesat thetoeof slope.
As a result, in this research, a comprehensive
method was proposed to excavate the road trench
downtothebedrock level without occurrenceof any
instabilityintheslope.
In the suggested method, first, the material of
embankment basin located in theoutsidepart of the
slippedregion(S.S.1) was excavatedto thebedrock
and then immediately filled with thecoarsegrained
material.
For excavating road embankment foundation
locatedontheslippedarea, itisdividedintoanumber
of narrowstrips(about3meterswidth) perpendiculars
tothelongitudeaxisof road.
Toavoidanyinstabilitywithintheslopeduringcon-
structionprocess, eachstripwasfirst excavatedfrom
downstreamtotheupstreamof theslopeandthenfilled
withrockfill material until reachingtooriginal ground
level. Then, the adjacent strip was performed simi-
larly.Thisoperationwasexecutedforall stripsuntil the
weak shalematerial of theroadsubgradewassubsti-
tutedwiththestrongrockfill material. Sincetheleast
depth of thebedrock located at theeast north of the
embankment, theexcavationandfillingprocesseswas
commencedfromthissituation.
Aftercompletingtheexcavationandfillingof all the
strips, theroadembankment wasconstructedsafely.
Figure7showstheexcavationandfillingprocesses
for one strip of road foundation. Also, the embank-
mentof substitutingroadafter completionisshownin
Figure8.
7 MONITORINGOF SLOPE BEHAVIOR
To control the deformations of slope during con-
struction, displacements was surveyedandevaluated
350
Figure 8. Embankment of substituting road after
completion.
Figure9. Measuredaccumulateddisplacements at bench-
markslocatedontheslopeandconstructionlevel of earthfill.
regularly at the benchmarks. The investigations
showed that the considerable reduction occurred in
thedeformations after constructionof roadembank-
ment. Forexample, variationsof displacementsduring
andafter constructionat thebenchmarkslocatednear
the BH-55 borehole and Z21 are presented in Fig-
ure 9. In this figure, the horizontal displacement of
thebenchmark locatedonBH-55boreholeis721cm
beforetheroadembankmentconstruction(fromOct.-
2006toJ uly-2007). Sotheaveragerateof horizontal
displacement is equal to 2.38cm/day. Although dur-
ing construction of embankment (fromJ uly-2007 to
Oct.-2007), thedisplacementof thisbenchmarkshows
considerable reduction. Unfortunately, during con-
struction operation, this benchmark is destroyed and
inevitablyanother benchmark (i.e. Z21) wasinstalled
in the slip direction and slope behavior was investi-
gatedviamonitoringdisplacementsinthisbenchmark.
The site surveying during road construction indi-
cated that the accumulative value of displacements
in the benchmark Z21 at the similar rainy season
is 16 centimeters (fromOct-2007 to May-2008) and
its average rate is about 0.7mm/day. Comparing
the displacements of the two benchmarks at similar
timeintervals indicates that thedeformation of sub-
grade slope has decreased about 34 times. During
embankment construction, the displacement of Z21
increasesonlyabout4centimeters(fromMay-2008to
Oct-2008). After embankment completionuptonow,
themovement of benhmarkZ21andsubsequentlythe
slipsof slopearealmost stopped.
This consequence indicates the effectiveness of
embankment inreductionof landslideriskandattain-
ment of permanent stabilityof substitutingroad.
8 CONCLUSIONS
In this paper, to reduce the landslide risk and con-
struction of a substituting road, a practical, most
effective and economical method was proposed and
constructed. Also, theenormouscostof slopemodifi-
cationwassavedbyapplyingtheproposedmethodfor
riskmanagement.
Toassurefromstabilityof theslope,asasubgradeof
theroad, duringandafter theroadconstruction, itwas
necessary tomonitoringandevaluatingthedeforma-
tionsof benchmarksinstalledontherhyodacitemass
andshaleslope(upstreamof roadembankment). By
evaluatingthemeasureddata, theslopebehavior can
bemanagedandstudiedcontinuously, andthecritical
deformations, ledtothehazards, canbepredicted.
REFERENCES
Ashenab 2005. Geology Report of Germi-Chay Dam,
AshenabConsults, Tabriz Iran
ASTM 1997. Standard practice for classification of soils for
engineering purposes (Unified Soil Classification Sys-
tem). ASTM D2487. West Conshohocken, Pa. American
SocietyforTestingandMaterials,
Bowles, J. E. 1996. Foundation analysis and design, Fifth
Edition. NewYork, McGraw-Hill
Cheng,Y.M. andLau, C.K. 2008. Slope Stability analysis and
stabilization: new methods and insight. NewYork. Rout
ledgeTaylor & FrancisGroup
Dai, F.C., Lee, C.F. and Ngai, Y.Y. 2002. Landslide risk
assessment and management: an overview, Engineering
Geology, 64(1), 6587
Hunt, R.E. 2007. Geological Hazards: a field guide for
geotechnical engineers. New York. Taylor & Francis
Group,
Fell, R. 1994. Landslideriskassessmentandacceptablerisk,
Canadian Geotechnical Journal, 31(2), 261272. 1994
Lessing, P., Messina, C.P., andFonner, R.F. 1983. Landslide
riskassessment, Environmental Geology, 5(2), 9399
Remondo, J., Bonachea, J. and Cendrero, A. 2005. A sta-
tistical approach to landslide risk modeling at basin
scale: fromlandslide susceptibility to quantitative risk
assessment, Landslide, 2(4), 321328
Sabatini, P.J., Bachus, R.C., Mayne, P.W. Schneider, J.A.
Zettler, T.E. 2002. GEOTECHNICAL ENGINEERING
CIRCULAR NO. 5; Evaluation of Soil and Rock Prop-
erties, FHWA-IF-02-034, Washington, DC,
USACE, 2003, SLOPE STABILITY, EM 1110-2-1902, 31
October 2003, USA
351
Risk assessment
Probabilisticriskestimationfor geohazards: A simulationapproach
M. Uzielli
Georisk Engineering S.r.l., Florence, Italy
S. Lacasse& F. Nadim
International Centre for Geohazards, Norwegian Geotechnical Institute, Oslo, Norway
ABSTRACT: Riskestimationforgeohazardsisinherentlycharacterizedbyconsiderableuncertaintiesininputs
andoutputs. Uncertainty-basedanalysis provides amorerobust andcompleteassessment of risk incompari-
son with deterministic analysis. A probabilistic framework for thequantitative, uncertainty-based estimation
for geohazards is proposed. Theframework relies onMonteCarlo simulationof risk throughthepreliminary
definition of its macro-inputs: hazard, vulnerability and thevalueof elements at risk. Thepaper provides an
operational descriptionof theframeworkaswell asconceptual discussionssupportingandmotivatingeachstep
of theprocedure. A calculationexampleisprovided.
1 INTRODUCTION
Itisgenerallyacceptedthatquantitativeriskestimation
for natural hazards is to be preferred over qualita-
tiveestimation whenever possible, as it allows for a
moreexplicitlyobjectiveoutputandanimprovedbasis
for communication between the various categories
involvedintechnical andpolitical decision-making.
The considerable heterogeneity in conceptual
approachestoriskestimationisawell-knownfact. No
univocal definitionisavailableatpresent, andthecon-
ceptual unificationof risk analysismethodscurrently
appearstobeapracticallyunattainablegoal.
A consistent quantitative risk estimation analysis
must rely on a reference risk framework. UNDRO
(1979), for instance, proposed the following model,
in which risk is calculated as the product of three
macro-factors:
To avoid the undesirable consequences of misinter-
pretations of risk estimates and assessment due to
the aforementioned terminological fragmentation, it
is essential to provide reference definitions explic-
itly. In the ISSMGE Glossary of Risk Assessment
Terms (e.g. http://www.engmath.dal.ca/tc32/), risk is
defined as a measure of the probability and sever-
ityof anadverseeffect tolife, health, property, or the
environment. Hazardistheprobabilitythatapartic-
ular hazardous event occurs withinagivenperiodof
time; vulnerabilityisthedegreeof expectedlossin
anelementor systeminrelationtoaspecifichazard.
Theelements at risk macro-component parameter-
izes thevalueof vulnerablephysical or non-physical
assetsinareferencesystem.Themeasurementunitsof
elementsat risk arenot univocal, anddependat least
onthereferencetimeframeof hazard, thetypologyof
elementsat risk andontheinvestigatorsperspective.
Thevalueof physical assets, for instance, is usually
measuredandexpressedinfinancial units, whilethe
valueof liveshasbeenparameterizedusingfinancial
andnon-financial units(e.g. equivalent fatalities).
Safety and cost-benefit optimization are primary,
essential objectivesof riskmanagement.Thequestfor
optimumdecision-making, which ensures the safety
andperformanceof aphysical systemwhileminimiz-
ingthecost of risk management andmitigation, can
beassociatedtotheconceptof minimizationof excess
conservatism. Conservatismisinitself apositivecon-
cept, asitisfinalizedtotheattainmentof safety, andto
thereductionof thelikelihoodandriskof undesirable
performanceof aphysical systemto alevel whichis
deemedacceptableor tolerable. Excessconservatism,
however, shouldbealso avoidedas it corresponds to
anon-optimal useof resourcestoattaintheaforemen-
tioned goals. While the above qualitative reasoning
is trivial, the borders between under-conservatism,
conservatismand excess conservatismare not eas-
ily assessed in risk management for natural hazards
practicebecauseof therelevanceof uncertainties in
the physical environment. Among the main factors
contributing to uncertainty are: (a) the difficulty in
parameterizingthedestructivenessof geohazards; (b)
theheterogeneity, spatial and temporal variability of
the physical environment; (c) the complexity in the
interactionbetweenthehazardouseventandthephys-
ical system; and(d) theindeterminationinmitigation
costs. Neglecting uncertainties can introduceunnec-
essary conservatismor, on the other extreme, lead
to mitigation countermeasures which areinadequate
in terms of safety andperformance. Although math-
ematical disciplines such as statistical science can
355
effectively contribute to quantifying, modeling and
processing uncertainties, they arenot routinely used
inestimatingriskfor geohazards.
Withreferenceto therisk model inEq. (1), while
hazardisusuallyinvestigatedinanuncertainty-based
perspective, vulnerability and elements at risk are
almostinvariablyestimateddeterministically.Thecon-
ventional quantificationof riskalsoignorestheeffects
of uncertainties, and the term risk is often syn-
onymous with the expected value of risk. A truly
non-deterministic approachto risk estimationshould
acknowledge the existence of uncertainties in all
risk macro-factors, and address such uncertainties
explicitly.
Thispaper illustratesaframework for uncertainty-
based, quantitativerisk estimationfor geohazards by
MonteCarlosimulation. Operationally, theemphasis
is on practical applicability androbustness. As large
uncertaintiesexist intheparametersandmodelsused
for risk estimation, it is not deemed advisable nor
significant, in an operational perspective, to employ
overly refined techniques for uncertainty modeling
anduncertaintypropagationanalysis, assignificantly
morecomplex theoretical frameworks andrestrictive
conditionsonparameters, hardlycompatiblewiththe
typeandquantity of dataavailableinrisk estimation
practice, wouldberequired. Nonetheless, themethods
andcriteriaadoptedintheproposedapproacharethe-
oreticallysoundandwell suitedforsystemswithlarge
uncertainties. Uzielli (2008) extendedthescopeof the
framework to include probabilistic decision-making
criteriafor riskmitigation.
The term Monte Carlo simulation embraces a
wide class of computational algorithms which are
effectively capable of simulating complex physical
and mathematical systems by repeated deterministic
computationof user-definedmodelsusingrandomor
pseudo-random values sampled from user-assigned
distributions. Despite the vast and heterogeneous
character of the algorithms, a common operational
sequence consisting of the following steps can be
identified: (1) probabilisticmodellingof uncertainties
inparameters andmodels; (2) repeateddeterministic
model simulationby computationonsampledmodel
inputs; and(3) aggregationof resultsof theindividual
simulation instances into the analysis output. These
stepsaredevelopedinthefollowingsections.
2 UNCERTAINTY MODELLINGFOR RISK
ESTIMATION
2.1 Uncertainty: definitions
The uncertainty in risk factors can be seen as the
result of the complex aggregation of aleatory and
epistemic uncertainties (see e.g. Phoon & Kulhawy
1999). Aleatory uncertainty stems fromthe tempo-
ral and/or spatial variability of relevant parametric
attributesof boththehazardousevent andvulnerable
elementsinthereferencesystem.Epistemicuncertain-
ties areaconsequenceof thepractical impossibility
to measureprecisely andaccurately thephysical and
non-physical characteristics of the reference system
andthehazardousevent, andtomodel theirinteraction
confidently. Theabsoluteandrelativemagnitudes of
thealeatoryandepistemiccomponentsof total uncer-
tainty are markedly case-specific. The aleatory and
epistemicuncertaintiesaddressedhereinareconscious
uncertainties, in the sense that the analyst is aware
of their existence. Unknown uncertainty refers tothe
knowledgewhichisnotcurrentlyattainablebytheana-
lyst. The reader is referred to Ayyub (2001) for an
interestingdiscussionof uncertaintycategorization.
A number of mathematical techniques for mod-
ellingandprocessinguncertaintiesareavailable. Here,
aprobabilisticapproachischosenfor avarietyof rea-
sons. First, probabilitytheory(mostoftenusedincon-
junctionwithstatistical theory)providesawidelyused,
generallywell understoodandacceptedframeworkfor
whichavast bulkof theoretical andapplicativelitera-
tureisavailable. Second, someof thebasicconceptsin
riskanalysis(e.g. thedefinitionof hazard) areconcep-
tuallylinkedtoprobability. Third, enhancedcomputa-
tional capabilities allow extensive use of techniques
suchasMonteCarlosimulation, whichareof greater
applicability thanother probabilistic techniques such
asFirst-OrderSecond-Momentapproximation,thelat-
ter requiringconstraints onthedegreeof linearity of
thereferencemodelsandonthemagnitudeof uncer-
taintiesininput variables. Fourth, duetothediffusion
of probabilistic concepts inthetechnical disciplines,
probabilisticapproachescanbeimplementedwithrel-
ativeeaseusinggeneral distributionsoftwaresuchas
electronic spreadsheets. Fifth, uncertain parameters
canbemodelledprobabilisticallyusingbothobjective
andsubjectivecriteria, therebyallowinggreater appli-
cabilitytoriskassessment analyses. Thedual nature
of probability, whichincludestheobjectivefrequen-
tist perspectiveandthesubjectivedegreeof belief
perspective, is of great practical significance in the
context of QRE for geohazards as it is most often
necessary to resolveto bothobjectiveandsubjective
modelling. Objectivemodelling can beperformed if
results of descriptive statistical analyses on samples
of the randomvariates of interest are available, for
instance in the formof frequency histograms. Once
a suitabledistribution typehas been selected by the
user, distribution parameters can be retrieved using
appropriateinferential statistical techniquesinvolving
distribution fitting. A purely objective modelling is
seldomfeasibleinthecontextof QRE for geohazards,
wheredataareinvariablylimitedinnumber andqual-
ity and the complexity of the interaction between a
hazardouseventandanyreferencesystemexceedsthe
analystsmodellingandparameterizationcapabilities.
Moreover, site-specific conditions may require sub-
stantial interpretation. A purely subjectivemodelling
relies on theanalysts experience, prior information,
belief, necessity or, more frequently, a combination
thereof. Subjective modelling should not be viewed
asasurrogateof objectivemodelling(Vick 2002). In
practice, probabilistic modelling is invariably hybrid
356
(i.e. both subjective and objective) to some extent.
Well-established frameworks such as Bayesian the-
oryallowrigorousmergingof subjectiveandobjective
probabilisticestimates.
Whether the assignment is objective or subjec-
tive, it is important to recognise that reducing the
magnitudesof thesourcesof uncertaintyrequiresfun-
damentally different actions. Epistemic uncertainty
can be reduced by increasing the amount and qual-
ityof dataandrefiningmodels. Aleatoryuncertainty,
however, may remain unchanged or even increase
with increases in the quality and quantity of data,
because the real degree of scatter in values of rele-
vant parameters may increase if more observations,
better measurement tools and more refined models
become available. To reduce aleatory uncertainty, it
is necessary to increase the resolution of the anal-
ysis (e.g. by defining a greater number of more
specific categories, or subdividingthereferencesys-
teminto geographical sub-units) as to decrease the
intra-categoryheterogeneityof vulnerableelements.
2.2 Uncertainty-based modelling of risk factors
Thecharacter of theinteractionbetweenahazardous
event andavulnerableelement depends onthechar-
acteristics of both the event and the element. In the
technical risk analysis literature, hazard, vulnerabil-
ity and elements at risk are expressed formally in
a variety of ways. Here, risk macro-components are
modelledasfunctionsof intensity.Inqualitativeterms,
intensity parameterizesthedamagingpotential of the
hazardousagent.Theconceptof intensityisnotunivo-
cally establishedinaquantitativesense: thediversity
and heterogeneity of hazardous events are such that
it is difficult to attain a general quantitative defini-
tion. Evenfor anysingletypologyof hazardousevent,
theliteraturerevealsthatdifferentparameterizationsof
intensityhavebeenidentifiedasmostsuitabledepend-
ingontheproblemto beinvestigated. Inearthquake
risk analyses, for instance, commonly adoptedinten-
sityparametersincludepeakgroundacceleration,peak
ground velocity, peak ground displacement, spectral
accelerationandmagnitude.Thesituationisevenmore
complex for landslides. A reference intensity param-
eter should be selected by the user with the aimto
concisely describethemost relevant damagingchar-
acteristic of the event. It depends both on the type
of event and vulnerable element, because different
vulnerableelements may suffer prevalently fromdif-
ferent attributes pertaining to the event. Here, the
aforementioned intensity-dependence is formalized
by identifyingareferenceintensity parameter I
N
for
thehazardous event and by subsequently expressing
hazard, vulnerability and elements at risk quantita-
tively as risk factor functions of I
N
. The framework
proposed herein allows the selection of any scalar-
or vector-valuedintensity parameter fromwhichrisk
macro-factor functionscanbedefinedunivocally.
Three risk factor functions are defined, namely:
thehazard function f
H
(I
N
); thevulnerability function
f
V
(I
N
); andtheelements at risk function f
E
(I
N
). Each
of these functions can be described analytically if a
concisemodel isavailableor devisedbytheuser; oth-
erwise, functionscanbedefinedempiricallybypoints
atrelevantlevelsof nominal intensity.Riskfactorfunc-
tions are defined in a common domain of nominal
intensity values. A nominal valueis arepresentative
deterministicvalueof anon-deterministicparameter.
Inthesubsequent phase, uncertaintymust beasso-
ciated to nominal values. The total uncertainty in
risk factors is, in operational terms, an aggregation
of parameter andtransformationuncertainty. Param-
eter uncertainty in risk factors is due essentially to
thetotal uncertaintyinthereferenceintensityparam-
eter, whichservesasinputtotheriskfactor functions.
Epistemicuncertaintyinintensityresultsatleastfrom
limitedmeasurementcapabilityof dynamiccharacter-
istics (e.g. velocity, momentum, seismic magnitude)
and geometric features (e.g. volume, displacement,
area, depth) andtheuncertainty inthemodel usedto
definethereferenceintensityparameterfromavailable
data. Aleatory uncertainty inintensity is due, among
other things, tothecomplexity of thephysical media
whicharemobilisedinthecourseof ahazardousevent
andtothespatial and/or temporal non-stationarity of
thedynamicandgeometriccharacteristicsof anyhaz-
ardousevent.Aswill beshowninSection3.2,intensity
isdirectlyinvolvedinthesamplingprocessof riskfac-
torsbecausesamplingdistributionsof riskfactorsmust
account for theuncertaintyinintensity.
Transformation uncertainty in risk factors stems
fromtheriskfunctions limitedcapabilityof modeling
andapproximatingthephysical world. Parametrically,
transformationuncertaintycanincludebias andscat-
ter (or dispersion). Bias is related to a functions
precision; scatter isrelatedtoitsaccuracy.
2.3 Uncertainty modeling for simulation
In aprobabilistic simulation perspectivesuch as the
oneadoptedherein,uncertaintymodellingrequiresthe
generationof sampling distributions forintensityI and
for theriskmacro-factorsH, V andE.
Here, two approaches areproposedfor thegener-
ation of sampling distributions: the direct approach
isapplicableif thedistributionparametersareknown
fromobjective analyses or can be assigned subjec-
tively. Suchapproachreliesontheimplementationof
thedefinitionsof theselectedprobabilitydistribution
types. Theindirect approach requiresthepreliminary
generation of distribution parameters fromsecond-
momentstatisticssuchasstandarddeviation, variance
or coefficient of variation.
Experiencehasshownthatarelativelylimitedsetof
distributionsareabletofit satisfactorilyawiderange
of observed phenomena. Selected distribution types
must beconsistent withthedefinitionandproperties
of theparameterswhicharebeingmodelled. Different
distribution types may be used for the same macro-
factor at different intensitylevels.
357
Hazardhasbeendefinedhereinasaprobabilityof
occurrenceof ahazardousevent.Asprobabilityvalues
arebydefinitioninferiorlyandsuperiorlyboundedat0
(nolikelihoodof occurrence)and1(certaintyof occur-
rence) respectively, anydistributionusedinmodelling
hazardmust beboundedintheclosedinterval [0,1].
Vulnerability as defined in Section 1 is also defined
intheclosedinterval [0,1]. Evenat thegreatest level
of generality, it is intuitive that values of Elements
at risk must be both non-negative and non-infinite.
Hence, inferiorlyandsuperiorlyboundeddistributions
areassumedfor all of theriskmacro-components.
Among commonly used probability distributions
which satisfy the conditions of lower- and upper-
boundingaretheuniformdistributionandthePERT
distribution. Thesearetwo special cases of theBeta
distribution. The probability density function of the
Betadistributionfor acontinuous randomvariable
isgivenby
inwhichB(
1
,
2
) isthebetafunctionwithinputs
1
and
2
; and
is the range of , given by the difference of the
twoextremevalues(upper-boundvalue
u
andlower-
boundvalue
l
).
The uniformdistribution is a special case of the
Beta distribution with
1
=
2
=1. The probability
density functionof theuniformdistributioncanalso
beexpressedinsimplifiedform:
Theuniformdistributionshouldbeassumedwhenever
noobjectiveinformationor subjectivemotivationsare
present regarding the existence of values inside the
parameter domainwhicharemorelikelytooccur than
others.
The PERT distribution is a particular case of a
Pearson type-I Betadistribution which requires user
specificationof themodal value
m
of thedistribution
andtheextremesonly. Thecharacteristic parameters
of thePERT distribution
1
,
2
arecalculatedby:
Modal valuescanbetakenascorrespondingtonomi-
nal valueswhentheseareavailable, orcanbeassigned
subjectively. When distribution parameters are not
availableanduncertaintyisparameterizedintermsof
second-moment statistics, the indirect approach can
beusedtoestimatelower- andupper-boundvalues. In
descriptive statistical terminology, the coefficient of
variation (COV) of ageneric parameter isdefined
astheratioof thestandard deviation of adatasettoits
expected (i.e. mean) value:
AstheCOV providesaneffectivemeasureof relative
dispersionof adatasetarounditsmeanvalue, itcanbe
conceptually associatedwithuncertainty, withhigher
values attesting for higher levels of uncertainty. It is
thuspossibletotransposeaqualitativejudgmentonthe
level of uncertaintyinaparameter (or model) quanti-
tativelybyassociatingtoitaCOV. Forinstance, asmall
COV (e.g. COV-0.10) canbeusedto represent the
belief inalowlevel of uncertainty. COVsintherange
0.100.30 can be regarded as intermediate, while
higher valuesattest for highuncertainty.
In the indirect approach, the derivation of lower-
and upper-bound parameters can be achieved using
statistical theory, bywhichrelationsbetweenthestan-
darddeviationandtherangeof arandomvariableare
availabledependingontheknownor assumeddistri-
bution type. Thestandard deviation (), if it is not
availabledirectly, can becalculated by inverting Eq.
(7) usingthemodal valueasexpectedvalue(assuming
thatthedistributionisquasi-symmetricwithrespectto
themodal value). Itisthenusedtocalculatetheupper-
andlower-boundvalues. Thisisachievedbycalculat-
ingtherange. If thedistributionis uniform-type, the
rangeisgivenby:
Incaseof PERT-typedistributions:
Lower- andupper-boundvaluesarethengiven, respec-
tively, bysubtractingandaddingthesemi-rangetothe
modal value.Itmaybenecessarytoimposeconstraints
onlimitingvalues
min
and
max
:
The inherent boundary values
min
and
max
are
parameter-specific: for hazard and vulnerability, for
instance,
min
=0 and
max
=1; for elements at risk,
min
=0and
max
<. Oncecharacteristicparameters
aredefined, uniformor PERT-typeprobability distri-
butions can be generated using the direct approach.
Theindirect approach should beapplied with confi-
denceonly incaseof parameters whichdistributions
canreasonablybeexpectedtobeatleastapproximately
symmetricaroundtheir expectedvalue.
358
3 RISK CALCULATION
Theframeworkillustratedhereinallowstheestimation
of risk by Monte Carlo simulation at Q1 user-
selected target nominal intensity levels, as long as
macro-factor functions are defined at each of such
target levels.
3.1 Preliminary considerations
Inatheoretical perspective, thedefinitionof thesize
of samplingdistributionsisacentral aspect inMonte
Carlo simulation. In most general terms, increasing
thesizeof samplingdistributionsallowsfor reduction
of thebias in thestatistics of thegenerated samples
withreferencetothetrueunknownpopulationstatis-
tics, thusallowingforamorereliablerepresentationof
therandomnessof theparametersunder investigation.
The trade-off for the reduction of estimation uncer-
tainty in samples is alarger computational expense.
In risk estimation for geohazards, distribution types
and parameters are assigned to a large extent sub-
jectively due to the difficulties in obtaining reliable
measurementsandincharacterizingthetypeof uncer-
taintyassociatedwithrelevantparameters. Hence, the
magnitudeof epistemic uncertainty can beexpected
to exceed theestimation uncertainty associated with
excessively small sample size. For this reason, it is
deemednot meaningful topursueanexceedinglyfor-
mal, rigorousestimationof samplingdistributionsize.
The optimumsize can be identified heuristically by
assessing the convergence of outputs for increasing
samplesize, i.e. byadoptingasampledistributionsize
which simulation outputs are sufficiently similar to
thosepertainingto larger samplingdistributionsizes
whileensuringcomputational feasibility.
Investigating correlation among input variates is
another central issueinMonteCarlo simulation, and
theobjectof avastbulkof ongoingresearchinvolving,
for instance, copula theory. In the proposed frame-
work, correlation is implicitly accounted for as the
risk factors areall defined as functions of intensity.
MonteCarlo simulations areperformed at reference
nominal intensity levels. At each referenceintensity
level, risk factors canbeassumedto beuncorrelated
for operational purposes sincethedistribution types
anddistributionparameters usedfor samplingof the
different factorsaremutuallyindependent.
3.2 Operational framework
Operationally, thecalculationof risk at theq-thnom-
inal intensitylevel (q =1,,Q) isachievedthrougha
modular procedure. For each risk factor , the first
stepconsists inthegenerationof aT-sizedsampling
distributionof intensity at theq-thnominal intensity
level I
(q)
N
(whichistakentobethemodal valueof the
distribution) andusingthedirect approach(if lower-
and upper-bound values of intensity I
(q)
l
I
(q)
u
are
available) or theindirect approach(if auser-assigned
COV of parameter uncertaintyof intensityCOV
p
(I
(q)
)
isavailable).
Intensity is inferiorly and superiorly bounded (it
cannot be negative, nor infinite); hence, a PERT
or uniformdistribution type can be assigned to I
(q)
dependingontheinvestigatorsknowledgeor belief.
Thesecondstepconsistsinthecalculationof aT-
sized distribution of the risk factor by applying the
risk factor function deterministically to each of the
T sampledvalues inthedistributionof intensity I
(q)
obtainedinthepreviousstep. Theresultingsampling
distributionisaparametricallyuncertainrepresenta-
tion
(q)
u
of theriskfactor.Theattributeparametrically
uncertain reflects thefact that thegenerated distri-
bution only accounts for the parameter uncertainty
stemming from the propagation of the uncertainty
in themodel inputs, and does not account for trans-
formationuncertainty associatedwiththerisk factor
function.
The third step consists in the characterization of
thetransformationuncertaintyassociatedwiththerisk
factor function at the q-th nominal intensity level
f
+
(I
(q)
N
). Such uncertainty can be parameterized by
adimensionless multiplicativecoefficient k
(q)
+
which
modal value k
(q)
+,m
(representative of the bias in the
risk factor function) is assigned by the user based
onobjectiveor subjectiveassessment. Theassociated
uncertainty can be expressed either in the formof
aCOV of transformation uncertainty COV
t
[f
+
(I
(q)
N
)]
associatedwiththerisk factor function, or by lower-
andupper-boundvaluesk
(q)
+,l
andk
(q)
+,u
. Thefourthstep
consists inthegenerationof aT-sizedsamplingdis-
tribution of the risk factor
(q)
. This is achieved by
scalar-multiplying thesampling distributions of
(q)
p
andk
(q)
+
:
The resulting sampling distribution accounts for
parameter andtransformationuncertainty. Following
completionof theabovestepsforH, V andE, riskcan
becalculatedbyscalar-wiseapplicationof thefunda-
mental equation to thesampling distributions of the
macro-factors:
The procedure described above is implemented at
each of the Q reference nominal intensity levels at
whichrisk estimatesareof interest. Theprocedureis
representedgraphicallyinFigure1.
3.3 Statistical characterization of risk
Thedirect output of theproceduredescribedaboveis
alsoaset of T deterministicriskvalues. Thesecanbe
considered as a sampleof therisk output random
variable. Risk canthenbeinvestigatedquantitatively
throughitssamplestatistics. Statistical moments, for
instance, parameterizepropertiessuchascentral ten-
dency(mean), scatter (variance), degreeof symmetry
359
Figure1. Schemeof procedurefor simulation-based esti-
mationof riskfactors.
(skewness) and shape(kurtosis). Quantiles also pro-
videadetailed statistical description of asampleby
parameterizing its empirical cumulative distribution
function. Thechoiceof thestatistics to beextracted
fromtherisk sampleis relatedto thescopeandgoal
of thespecificriskestimationanalysis.
4 CALCULATIONEXAMPLE
A simplecalculationexampleispresentedhereinafter
for illustrative purposes. The object is the quantita-
tive rainfall-induced landslide risk estimation for a
building.Themaximumrainfall intensityA(inmm/h)
andrainfall durationD (inh) areselectedtoserveas
avector-valuedreferenceintensity parameter. A risk
estimate is desired for D=0.5h and A=20mm/hr.
SampledistributionsizesT =1000000wereadopted
following comparative analyses. The results of the
corresponding deterministic analysis are given for
comparativepurposes.
4.1 Intensity
COVs of parameter uncertainty of 0.03and0.05are
assignedtoD andA, respectively. Theseaccount for
epistemicuncertaintyindurationandamount of rain-
fall. PERT-typesampling distributions of sizeT =1
000000arethusgeneratedforDandAusingtheindi-
rect method. The resulting lower- and upper- bound
valuesareD
l
=0.46h,D
u
=0.54h,A
l
=17mm/hand
A
u
=23mm/h.
4.2 Hazard
Thefollowinghazardfunctionexpressesthenominal
value of the annual probability of occurrence of at
least onerainfall event with agiven intensity A and
durationD:
Parameter uncertaintyinf
H
correspondstotheuncer-
taintyintheintensityparametercomponentsDandA.
Thehazardfunctionisapplieddeterministicallytothe
couplesof sampledvaluesof DandA, therebyobtain-
ingaT-sizeddistributionof theparametricallyuncer-
tain hazard H
p
. Transformation uncertainty in the
hazardfunctionisparameterizedbyCOV
t
(f
H
)=0.30.
A T-sizedPERT-typesamplingdistributionof model
uncertainty is generated by assuming f
H
to beunbi-
ased(i.e. k
H,m
=1) andby calculatingthelower- and
upper-boundvaluesk
H,l
=0.10andk
H,u
=1.90using
theindirectmethod. Thedeterministicoutputvalueof
hazardfor D=0.5handA=20mm/hr is0.32.
4.3 Vulnerability
The following vulnerability function, which reflects
the results of fictitious runout analysis for different
setsof valuesof D andA, isassigned:
A sampling distribution of parametrically uncertain
vulnerability V
p
is obtained by applying Eq. (15)
360
Figure2. Outputsof calculationexample.
deterministically to the sampling distributions of D
and A. Assuming themodel is unbiased, auniform-
type sampling distribution of k
V
is obtained by pre-
liminary calculating lower- and upper-bound values
k
V0,l
=0.57andk
V0,u
=1.43usingtheindirectmethod.
Subsequently, scalarmultiplicationisemployedtocal-
culate the T-sized sampling distribution of V. The
deterministicvalueof V givenbyEq. (15) is0.48.
4.4 Elements at risk
A constant, intensity-invariant value of E (in ) is
assumed:
Lower- and upper-bound values for the multiplica-
tive coefficient k
E
are assigned as k
V,l
=0.75 and
k
Vul
=1.05, respectively. Assumingthemodel isunbi-
ased, a PERT-type sampling distribution of k
E
is
obtainedbythedirectmethod. Subsequently, thesam-
plingdistributionof E isobtainedbymultiplyingeach
sampledvalueof k
E
bythedeterministicvalue.
4.5 Risk
Risk is obtainedby applyingthefundamental model
inEq. (1) tothesamplingdistributionsof H, V andE.
Therelativefrequencyhistogramof theoutputT-sized
distributionis shownalongwiththoseof themacro-
inputs in Figure 2. In the figure, relative frequency
histograms of parametrically uncertain distributions
of H and V are plotted (in grey) along with the
total uncertainty distributions(inblack). Determin-
istic values of risk factors and risk are also plotted
asdashedblack lines. Visual inspectionof theresults
allows appreciation of the capability of simulation
of accommodating heterogeneous distribution types,
andof providingoutputdistributionswhichshapeand
character may bedifficult to establishaprioristically.
Moreover, itmaybeseenthatwhilemodel uncertainty
is not very significant for H, thereis avery relevant
modificationintherelativefrequencyhistogramof V
andE fromtheparametric uncertainty tothetotal
uncertainty case.
Table1 reports thesamplestatistics of theproba-
bilistic simulationas well as theoutput of theequiv-
alent deterministicanalysis. Thetableisstructuredas
Table1. Outputsof probabilisticanddeterministiccalcula-
tionexample(1000000simulations).
H[yr
1
] V[] E[] R[yr
1
]
determ. 0.32 0.48 1500000 224788
mean 0.32 0.47 1450015 218676
std. dev. 0.13 0.12 76185 106489
COV[] 0.41 0.25 0.05 0.49
Q
05
0.16 0.29 1306960 90038
Q
25
0.23 0.37 1400688 142223
Q
50
0.29 0.47 1461618 196418
Q
75
0.38 0.58 1510363 270498
Q
95
0.57 0.66 1553255 425124
follows: thefirst rowcontains theequivalent deter-
ministic outputs of H, V, E and R; thesecond row
reports themeanvalues of theoutput samples of the
sameparametersfromMonteCarlosimulation, while
rows 3 and 4 report the sample standard deviations
andcoefficientsof variation, respectively, of thesame
output samples. Rows 5-9 report the0.05, 0.25 (i.e.
1stquartile), 0.50(i.e. median), 0.75(i.e. 3rdquartile)
and0.95quantilesof thesamples, respectively.
Intheexampleillustratedherein, themagnitudeof
scatterinoutputsamplesof H, V andR aroundsample
means(parameterizedbysampleCOVs)issignificant.
Deterministicvaluesof H, E andR arenotcoincident
withmeanvalues. Thoughthedifferenceisnot large,
statisticsreflecttheskewnessof thedistributions. The
0.50and0.90quantileranges, givenbythedifferences
Q
75
Q
25
andQ
95
Q
05
, respectively, alsoattest for
aconsiderablelevel of uncertainty in risk estimates,
and provide very useful supplementary information
regardingthesensitivityof estimatedrisktoparameter
andtransformationuncertainty.
5 CONCLUSIONS
Thispaperhasillustratedamethodologyfortheproba-
bilisticestimationof riskanditsmacro-factors:hazard,
vulnerabilityandelementsat risk. Whileuncertainty-
based analysis does not imply theelimination of all
of the uncertainties involved in risk estimation for
geohazards, it does allowamorerational assessment
of thelevel of safety, performanceandconservatism
associatedwithriskestimates.
361
MonteCarlosimulationprovidesapowerful, robust
and flexible means of modelling uncertainties and
investigatingtheir propagationinrisk models. More-
over, it allows seamless aggregation of subjective
andobjectiveuncertaintyestimatesinparametersand
models.Anothermainbenefitof simulationisthepos-
sibility to appreciatethefull shapes of uncertainty
distributions in risk factors and risk itself, as well
as to compare the relative magnitudes of epistemic
and aleatory uncertainties. Such an insight into the
characterof theuncertaintyallowsenhanceddecision-
making and the selection of appropriate mitigation
measures to attain target levels of risk. The simple
calculation exampleincluded in thepaper, based on
artificial butrealisticdata, attestsfor theconsiderable
level of uncertaintywhichisinherenttoriskestimation
for geohazards.
ACKNOWLEDGEMENT
theResearchCouncil of NorwaythroughtheInterna-
tional Centrefor Geohazards (ICG). Their support is
gratefullyacknowledged.
REFERENCES
Ayyub, B.M. 2001. Elicitation of expert opinions for uncer-
tainty and risks. BocaRaton: CRC Press.
Phoon, K.K. andKulhawy, F.W. 1999. Characterizationof
geotechnical variability,CanadianGeotechnical Journal
36(4), 612624.
UNDROUnitedNationsDisasterRelief Organization1979.
Natural disasters and vulnerability analysis. Geneva.
Uzielli, M. 2008. Probabilistic risk analysis for geohazards:
a simulation approach. NGI report 20061032-8. Interna-
tional Centrefor Geohazards / Norwegian Geotechnical
Institute, Oslo.
Vick, S.G. 2002. Degrees of belief Subjective probability
and engineering judgment. NewYork: ASCE Press.
362
A researchproject for deterministiclandslideriskassessment inSouthern
Italy: Methodological approachandpreliminaryresults
F. Cotecchia
Department of Civil and Environmental Engineering, Technical University of Bari, Italy
P. Lollino& F. Santaloia
National Research Council IRPI, Bari, Italy
C. Vitone& G. Mitaritonna
Department of Civil and Environmental Engineering, Technical University of Bari, Italy
ABSTRACT: Thepaper presentsamethodologyfor thedeterministicassessmentof thelandslideriskinchain
areasatintermediatescalewhichisbeingdevelopedwithinamulti-disciplinaryresearchproject.Theprocedure,
whichneeds to beimplementedinaGIS system, is beingtestedintheDauniaregion(SouthernItaly), where
tectonisedandfissuredsoilsoutcrop, but it isintendedtobeof referencefor riskanalysisandriskmanagement
withinthoseareas characterizedby similar geological contexts. A first applicationof suchmethodology to a
specificurbancentrefromDauniaisalsopresented.
1 INTRODUCTION
The assessment of landslide risk is a research topic
of increasing interest all over the world due to both
anincreasingawarenessof thedramaticallyimportant
impact of landslides onthesocio-economic environ-
ment andanincreasingrequest for development and
extension of urbanization in areas proneto landslid-
ing. The present paper discusses the formulation of
a methodology that is being developed for regional
landslideriskassessmentwithingeologicallycomplex
areas andsomepreliminary results of its application
attheintermediatescale(i.e. betweentheregional and
theslopescale). Theareas of applicationarelocated
inthesouthernApennines(Italy), wherelandslidingis
widespread and responsiblefor frequent damages to
structuresandinfrastructures.
The methodology is the subject of an on-going
multidisciplinary researchproject, whichaims at the
assessmentof thelandslidehazard, of thecorrespond-
ingvulnerabilityof structuresandof their exposition,
involving different expertises. In particular, both the
landslidehazardandthestructurevulnerabilityassess-
mentsaremeanttobebasedupontheknowledgeof the
failuremechanismsand,assuch,tobenefitfromscien-
tificknowledgeinthefieldsof bothgeotechnical engi-
neeringandstructural mechanics.Atthesametime,the
exposureof theelements at risk is to beinvestigated
accordingtoanalysesof thesocio-economical context
wheretheriskisbeingevaluated. Inthepresent paper
onlytheworkrelatingtolandslidehazardispresented.
Thisworkaimsatthefurtherdevelopmentof Quan-
titativeLandslideHazardAssessment, QHA (Hoetal.
2000), followingadeterministicapproach. Assuch, it
is aimedat exportingthegeo-mechanical interpreta-
tionof slopestabilityandlandslidemechanismsfrom
theslopescale(site-specific) totheregional scale.The
researchworkisdevelopedwithreferencetoatest-site
area, the Daunia region, located at the eastern mar-
ginof thesouthernApennines, whichis aportionof
thechainbelt alongthesubductionzonebetweenthe
african and the euro-asiatic plates, where slopes are
made up of tectonised and fissured soils and rocks.
Here, frequent and intense landsliding involve the
slopes extensively and repeatedly, restraining urban
development. Theresearch is currently investigating
theapplicability of thedeterministic methodology to
thelandslidehazardassessmentof theurbanterritories
intheregion.
In thefollowing, after a brief review of themain
methodologiesfor landslidehazardassessment avail-
ableintheliterature, thepaper focusesonthemethod-
ology being developed in the research. Thereafter,
preliminary results of its application to evaluate the
landslidehazardinthetest-siteregionarediscussed.
2 CURRENT APPROACHESTOLANDSLIDE
HAZARDASSESSMENT
It is generally recognizedthat thecriteriatobesatis-
fiedfor asuccessful landslidehazardassessment are:
the objectivity of the data collection processes, the
detectionof objectiverelationshipsbetweenlandslide
factorsandevents,thepossibilityof on-goingupdating
363
of thehazardassessment procedureandtheadequate
choice of the analysis scale (Aleotti & Chowdhury
1999, Fell et al. 2008). Several methods for land-
slide hazard assessment have been proposed in the
literature, whichsatisfy todifferent extentsthesecri-
teriaandthat canbeclassifiedas either heuristic, or
statistical or deterministic (van Westen et al. 1997,
Aleotti & Chowdhury 1999, Dai et al. 2002, Fall
et al. 2006).
Heuristic methods, also called expert evaluation
models, are the most widespread. They are mainly
basedongeo-structural analyses, geo-morphological
surveys and interpretation of aerial photographs,
which may give indication of: the slope inclination
and vegetation conditions, thegeological set-up, the
existenceof landslidebodies andtheir activity (style
and distribution of movements Cruden & Varnes
1996). Based upon such expert oriented qualitative
interpretations, thehazardassessmentresultsineither
landslideinventorymaps,orlandslidedensitymaps,or
indexmaps. Inparticular, indexmappingimplements
procedures of processing of the factors supposed to
influence the landslide process, which account for
weightingcoefficients(Guzzetti etal.2000)andwhich
resultinhazardcategorymaps.Atpresent, thesemeth-
ods areextensively applied using GIS techniques to
create the factor archives and deduce the category
maps. However, themaindisadvantageof suchmeth-
odsisthesubjectivityinthechoiceof thefactorsand
of the corresponding weighting coefficients, due to
thesubjectivity of theinterpretation of thelandslide
activityandof itsrelationwiththefactors. Indeed, the
choiceof theweightingcoefficientsandof theindexes
arenotrefutedbyanyquantitativeassessment. Inaddi-
tion, thedifficulty of updatingthehazardassessment
duetotheuseof quiteimplicit criteriainitsformula-
tion(Fall et al. 2002, Aleotti &Chowdhury1999, Dai
et al. 2002) representsanother important drawbackof
themethodology.
Statistical methods involve statistical determina-
tions of combinations among the factors that are
supposed to influencelandsliding (Yin & Yan 1988;
Carraraetal. 1991,Aleotti &Chowdury1999). These
methodsstill makeuseof qualitativeinterpretationsof
thelandslidephenomena, but theyrestrainthedegree
of subjectivityof theanalysisbycomparisonwiththe
heuristic methods and allow for the updating of the
hazardmappingwithtime. Theyaregenerallyconsid-
ered to beappropriatefor landslidehazard mapping
at themediumscale(1:10000 1:50000), sincethis
scale is suitable for the acquisition of the landslide
factorvaluestobeprocessedinthestatistical analyses.
However, themaindisadvantageof statistical methods
resultsfromthelackof validationof thestatistical cor-
relationsbetweenthefactorsbeingused, whichresults
fromthe lack of any quantitative assessment of the
slopefailuremechanisms (Hoet al. 2000). Suchdis-
advantagecouldbelimitedbyapplyingsuchmethods
inthelightof anadequatemechanical interpretationof
thelandslideprocesses. However, ingeneral thestatis-
tical methodsarequitestrictlyrelatedtotheareawhere
they arecalibrated and cannot beeasily exported to
areasof different features(Canuti & Casagli 1994).
Deterministic methods assess landslidehazard by
meansof quantitativeevaluationsof thestabilityof the
slopeor, otherwise, of theactivelandslidemechanism,
based upon theknowledgeof thegeo-structural set-
up, themechanical behaviourof thematerialsforming
theslopeandthehydraulicconditions, andthevalues
of the external factors of landsliding (e.g. rainfalls,
seismic loading, humanactions). Landslidehazardis
derived fromthe suscettivity assessment accounting
for the probability of occurrence of given values of
the external factor with time. As such, these meth-
odsrestrainthedegreeof subjectivityof theanalyses.
Due to the necessary use of significant and reliable
databases andof accuratecalculations intheir appli-
cation, so far they have been mainly adopted for
slope-specifichazardassessments.Applicationsatthe
regional scalehavebeencarriedoutmainlyinregions
whereasingle-typefailuremechanismcouldberec-
ognized(Montgomery&Dietrich1994, Cascini et al.
2003, Savage et al. 2004), so that the interpretation
of suchmechanismcouldbeof referenceinthepre-
dictionof failureonmost slopes. Conversely, within
regions of complex geological set-upandwheredif-
ferent landslides appear to occur, landslide hazard
is generally classified by means of index mapping
(Guzzetti et al. 2000). In thefollowing, amethodol-
ogyfor adeterministichazardassessmentatthesmall
scaleisproposed.
3 DETERMINISTIC LANDSLIDE HAZARD
ASSESSMENT AT THE SMALL SCALE
Accordingto theresearchobjectives, thedeterminis-
tic analysis of landslidehazard at regional scalehas
been designed in order to export theuseof thegeo-
mechanical understandingof theslopefailuremecha-
nismfromtheslope scale totheregional scale(small
scale).Thegeo-mechanical understandingof theland-
slide hazard is based upon the recognition of the
relationships betweenthelandslidefactors (Terzaghi
1950, Hutchinson 1988) and the landslide mecha-
nisms (cause-effect relations). Thesecanbededuced
fromanalysesatthreelevels: aI level, or preliminary,
whichaccountsmainlyfor phenomenological studies;
aII level, orintermediate, performedbymeansof limit
equilibriumanalysesof theslopegeotechnical model;
aIII level, oradvanced, whichinvolvesbothnumerical
modellingandfieldmonitoring.Assuch, thelattertwo
analysesresultinquantitativeassessmentsof theslope
stability and of the possible landslide mechanisms.
The I level analysis may represent a first approach
to thequantitativeassessment, if developedaccount-
ingfor: thegeo-hydro-mechanical set-upof theslope,
thehydro-mechanical properties of thematerials and
their influence on the landslide mechanism. There-
fore, amethodologyfordeterministiclandslidehazard
assessment at thesmall scalehas to implement site-
specific studies of the landslide mechanisms which
364
Figure1. Flow-chartof theproposedmethodologyforsmall
scaledeterministiclandslidehazardassessment.
occur intheregionandof their connections withthe
geo-hydro-mechanical factors. These studies should
bewidespreadacrosstheregionandrepresent the1st
phase of themethodology beingproposed. Inpartic-
ular, on one side they are intended to recognize the
setsof geo-hydro-mechanical factorswhichbest rep-
resent theslopesintheregion. Ontheother, they are
aimedat identifyingtheslopefailuremechanisms to
be considered as representative in the region. Such
studiesrequirethecollaborationof differentexpertise,
inthefieldsof topography, geology/geo-morphology
and geotechnics. Their results should bereported in
guidelinesmeanttooutlineaframeworkfor thequan-
titative interpretation of landsliding in the region.
Thereafter, the methodology should implement the
procedures of useof theseguidelines for thehazard
assessment inanyportionor slopeof theregion(2nd
phase of themethodology). Theflow-chart shownin
Figure 1 summarises the main working steps of the
methodology.
The usefulness of the 1st phase studies for the
regional deterministic hazard assessment assumes,
as first, that the representative geo-morpho-hydro-
mechanical set-ups and failure mechanisms in the
regionareof limitednumberand, assecond, thepossi-
bility to identify representative quantitativerelations
between the failure mechanisms and the slope geo-
morpho-hydro-mechanical factors, which would be
thoseof referenceintheregion(Hypotheses 1 and2
in Fig. 1 respectively). A region characterized by a
small number of set-upsandfailuremechanismswill
bedefinedhomogeneous; this is thecasethesmaller
is the extension of the region and the more repet-
itive are the geological features of the soil masses.
As recalledearlier, theliteraturereports thesuccess-
ful useof thedeterministic approachto intermediate
scalelandslidehazardassessmentinchainareaswhere
the geo-mechanical set-ups are so repetitive that a
singlelandslidemechanismisfoundtooccur repeat-
edly, thusinregionswhicharesohomogeneousthat a
singlelandslidemechanisminterpretationisof refer-
enceforthehazardforecastingonanyslope. However,
hypotheses 1 and 2 can be found to be valid also
in regions where the landscape appears to be varie-
gatedandgeologicallycomplexand, assuch, theyhave
beenassumedto support thegeneral applicability of
thedeterministic methodology beingproposed. Such
assumptionhas beenbasedontheawareness that, in
general, thevariationsinmechanical behaviouramong
soils and lithotypes aremorelimited than thevaria-
tionsintheirgeological classificationfeatures. Indeed,
soilsandrockspartof differentgeological formations
are often found to exhibit similar bulk mechanical
response. If this is the case, a classification of the
landscape using, as priority classifying feature, the
mechanical propertiesof thematerialsandtheir com-
binationinthesoil massmaybefarmoresyntheticthan
onebaseduponthegeological features. It wouldfol-
lowthattheuseof mechanicsasameantoclassifythe
set-upsandthephenomenainagivenlandscapewould
simplifyitscharacterization. Thevalidityof Hypothe-
ses 1 and2 inageologically complex regionsuchas
theDauniaregionisdiscussedlater inthepaper.
Figure1reportsaflowchartoutliningthe1st phase
of thedeterministicmethodology.
Step 1 (S1) representsthecreationof ananalytical
databaseof all thefactorsinfluencingtheslopeequi-
librium(Terzaghi 1950, Hutchinson1988). They can
be either internal factors (slope geometry, lithology,
geo-structural set-up, tectonic structures, mechanical
properties of the materials, hydraulic regime in the
slope) or external factors(rainfalls, earthquakes, man
action, natural variationof theslopegeometry). Such
databaseshouldincludebothdataloggedataregional
scaleandmoredetaileddatafromlarger scalestudies.
All thesedifferentsourcedatashouldbeimplemented
inaGISdatabaseafterappropriatescreening(Mancini
et al. 2008). Although large part of such database
is common to heuristic and probabilistic methods,
for the deterministic analyses it must be far richer
ininformationconcerningthehydro-mechanical fea-
turesof thematerials.Thesecond step (S2inFig. 1) of
themethodology concernsthegeo-hydro-mechanical
classificationof thesoil masses(classesGMi inFig. 1)
and theclassification of therepresentative landslide
typologies(Li) intheregion, baseduponthecollected
database and site-specific surveys. Finally, the third
step (S3 in Fig. 1) of the 1st phase is aimed at the
recognition of the connections existing between the
sets of internal factors of landsliding(characterizing
the GMi classes), the external factors and the land-
slidetypologies (Li). It is based both on analyses of
theGIS dataandonsite-specific studies widespread
365
across the region. Such site-specific studies can be
developed at the three aforementioned levels (I, II
andIII). Indetail, theI level analysescouldbebased
on comparisons between the different sliding phe-
nomenaandcorrespondingmechanical andhydraulic
conditions. TheII level analysesconsist inlimit equi-
librium analyses of the sliding process carried out
usingthegeo-hydro-mechanical dataavailableinthe
database. These analyses could be parametric with
respect to the factors for which the evaluation had
beenleastobjective.Theresultsof suchanalyseswould
providequantitatively basedindicationsof thegeom-
etry of thelandslidebody, of thefailuremechanism
and of theaveragestrength mobilized along theslip
surface, aswell aspreliminaryindicationsof thepos-
siblelandslideevolutionandtriggeringfactors. These
resultscouldor notvalidatetheI level interpretations.
The III level analyses consist of both the direct in-
situ monitoring and the numerical modelling of the
landslide processes; as such they provide the most
objectiveinterpretationof thephenomena, but, given
their complexity, theyshouldbeappliedtofewof the
representative casesintheregion.
Theresultsof thethirdstep(S3) studiesshouldout-
line the representative landslide mechanisms in the
region, their causes andpossibleevolution, andmust
be reported into guidelines, which may also imple-
ment either algorithms or modelling procedures and
represent the reference for the susceptibility assess-
mentsintheregion. Suchguidelines, togetherwiththe
regional GIS database, may beupdatedprogressively
withtime. However, theassessmentof theevolutionof
thefailuremechanismandof thetimeof occurrence
of theevent, requiredfor hazardassessments, require
additional information. For example, they may bene-
fit fromtheresult of investigations of multiple-year
topographicmaps, aerial photosandmonitoringdata,
whicharegenerallysubjectsof I level analyses. Alter-
natively, inII andIII level analyses, themodellingmay
giveindicationsof theevolutionof theslopeequilib-
riumand failure conditions with time if accounting
for thevariationof boththeinternal andexternal fac-
tors with time. In addition, III level analyses could
also giveindications about thelandsliderun-out. All
suchindicationswouldbeof usetoadvancefromthe
susceptibilitytothehazardprediction.
Thehazard assessment application within agiven
portion of theregion represents thesecond phase of
the methodology (Fig. 1). In this phase, a specific
databasehastobedevelopedfor theareasubjectedto
assessment.Thisismoredetailedthantheregional one
createdinthefirst phase andimplementsdatalogged
atalargerscale.Thedatatobecollectedconcernall the
factorsof landslidingaswell asall theavailabledata
indicativeof theactivityof movementsontheslopes,
suchas: topographic monitoringdata, interferometer
data, atdepthmonitoringdata(e.g. inclinometerdata),
aswell asdataaboutthedamagesof structuresinteract-
ingwiththemovingslopes. Theprocedureof creation
of suchlocal GIS databaseismeant tobeoutlinedin
the guidelines and is based upon the understanding
of the landslide processes gathered in the 1 phase.
Theinteractiveconsultationof thelocal GISdatabase
and of theresults of the1 phase studies reported in
theguidelinesshouldleadtotheidentificationof the
possiblelandslidemechanismsapplyingtotheareaof
application. In particular, thevalues of thelandslide
factors reported in thelocal GIS databaseshould be
comparedwiththosereportedintheguidelinesfor the
differentgeo-hydro-mechanical set-ups(GMi) incon-
nectionwiththedifferent landslidemechanisms(Li).
This comparison should let the identification of the
landslide mechanismoccurring in slopes character-
izedbyvaluesof thefactorsclosesttothoseof thearea
beingassessed. Theinformationabout suchlandslide
mechanismandaboutitsconnectionwiththelandslide
factorsshouldfinallyprovidesufficientknowledgeto
assesswhat typeof landslidemay occur ontheslope
beingassessedandhowit mayevolve.
4 TEST-SITE OF THE PROJ ECT AND
PRELIMINARY 1ST PHASE RESULTS
Theoutcroppingof stratifiedsedimentary sequences
iswidespreadwithintheDauniaregion.Thesearepart
of several different turbiditic successions (or flysch),
within which rock layers alternate with clay layers,
which are generally highly fissured. The most fre-
quentflyschformationsaretheSanBartolomeoFlysch
(called SBO hereafter) and the Faeto Flysch (FAE),
which both include rock and clay strata: calcareous
strataandclayey marls for theFAE flyschandaren-
ites and clays for the SBO flysch. Conversely, the
RedFlysch(FYR) istheonepredominantlyformedof
highlyfissuredclays.Theheterogeneityandthefissur-
ingcharacterizingtherocks andclays formingthese
formations makes thembe classified as structurally
complex formations (Esu, 1997). In particular, Red
Flyschclaysareclassifiedasvaricolouredscalyclays
andarecharacterisedbytheloweststrengths.Research
studies carriedout by theAuthors (e.g. Cotecchia&
Santaloia2003; Vitoneet al. 2009) inthelast decade
showhowthefissuringpatternof FYR clays (which
issointenseastodefineinter-fissureclayelementsof
millimetre thickness, defined as scales) causes such
verylowstrengths. Cotecchiaetal. (2006) andVitone
et al. (2009) havedemonstratedthat thestatebound-
arysurfaceof thesefissuredclaysisevensmaller than
that of the same material when reconstituted in the
laboratory. Sincethelowstrengthof thefissuredclays
plays a fundamental role in the development of the
slopefailureprocesses in theregion, theknowledge
abouttheirmechanical propertiesiscrucial toeffective
quantitativeanalysesof thefailureprocesses.
Themethodologyproposedabovehasbeenapplied
to the landslide hazard assessment within the urban
areas located in themountainous part of theDaunia
region (scale 1:100001:50000). In the first phase,
thelandslidefactors characterizingtheslopes within
twenty-fiveurbanareashavebeeninvestigated. Inthis
366
Figure2. a) Geological schemeof Italyandlocationof thetest-area; b) Geological SetupGMi andc) LandslideTypologyLi
definedinthestudyarea. Setup(1): rigidcapoverlyingadeformableunit withsub-horizontal contact; setup(2): monoclinal
contact between clayey and rocky units; setup (3): single geological units. Landslide type L1: compound landslide from
mediumto largedepth; L2: mud-slidefromshallowto mediumdepth; L3: rotational landslideevolvinginmud-slidefrom
mediumtolargedepth.
respect, thestudyof thelandslidingoccurringinfree-
fieldareashasbeendisregarded. Thelithological and
geo-structural featuresof theslopesandtheproperties
of thematerials havebeen characterized based upon
theanalysisof geological mapsandreports,geotechni-
cal reports, landslideinventoriesandresultsof in-situ
surveys. All thesedatahavebeenimplementedinGIS
databases, eachrelatingtoasingleurbancentre. The
analysisof thesedatahasbeenaimedattherecognition
of the main geo-mechanical set-ups, representative
for theurbanareas intheregion. Inparticular, it has
been recognized that for most towns, theold part is
foundedonrock strata, that most oftenareeither the
limestones of Faeto Flysch or thesandstones of San
Bartolomeo Flysch. Only the last century urbaniza-
tioncausedtheextensionof theurbanareas towards
theclayoutcroppings, asidetherockoutcroppings.
The rock strata are found either to float above
the clay strata (of either the same flysch or of the
RedFlysch), awareness that theoverall behaviour of
these slopes depends on the features of the succes-
sion and clay strata, thetypeof contact between the
rock and clay strata has been accounted for as the
mainelementcharacterizingtheareasof interest.Thus,
the sub-horizontal contact between an upper stiff
high strength rock slab and a lower stratumformed
of very weak and low stiffness fissured clays, has
been considered to represent a representative geo-
mechanical set-up, classified as GM1 (Fig. 2b). The
geo-mechanical set-up corresponding to the mono-
clinal contact has been instead classified as GM2
(Fig. 2b). Finally, sincesomeurbancentresarefound
tolieinareaswhereasinglelithological unit (rockor
clayunit) outcrops, suchset-uphasbeendistinguished
asGM3,asshowninFigure2b.Forall theGMi set-ups,
thewatertablehasbeenfoundtobeshallow, generally
not deeper than6metresbelowgroundlevel.
The first phase studies have been aimed also at
the recognition and classification of the main land-
slidemechanismsoccurringontheslopesof theurban
areas. For this purpose the GIS database has been
extendedto includeall theavailabledataconcerning
landsliding, suchasthedatareportedinbothnational
andregional landslideinventories, literaturedata, dis-
placement andfailuredatapresentedingeotechnical
reports, informationabout damagesof structuresand
new data resulting fromon-purpose in-situ surveys.
Theanalysesof all thesedatahaveresultedintheiden-
tificationof threemaintypesof landslides, namedas
L1, L2 and L3 in Figure2c. According to theinter-
national landslide classification (Cruden & Varnes
1996),thefirsttypology,L1,isrepresentedbyinterme-
diatedtodeep-seatedcompoundlandslides, of failure
surfacedepthlarger than30mandwidthcomparable
tothelength. Theselandslideshavebeenoftenfound
toberetrogressiveandmultiple.Thesecondtypology,
L2, correspondstomudslides, commonlyeitherlobate
or elongate, of shallowto intermediatedepthsliding
surfaces (30m). Locally, the retrogressive failure
surfacemaybedeeper(3040m).Thethirdtypology,
L3,isrepresentedbydeep-seatedtointermediatedepth
rotational landslidesevolvingintoeither mudslidesor
earthflows downslope. The limited number of land-
slidetypesbeingrecognizedintheregionhasthuscon-
firmedoneof thehypothesisassumedinthedefinition
of themethodology, i.e. that ispossibletorecognizea
limitednumber of representative geo-morpho-hydro-
mechanical set-upsandfailuremechanismsevenina
region of significant geological complexity, such as
theonebeingconsidered.
367
Figure3. CelenzaValfortore: a) geological andgeomorphological map(afterMelidoro1982, modified). Key: 1) urbancentre
area, 2) fillingmaterial, 3) SBOFlysch(a-clayeyunit, b-arenaceousunit, 4) FYR, 5) thrust, 6) strataattitude, 7) deepfailure,
8) landslide(a-crown, b-body, c-depression, d-directionof movement); b) Detail of thesouthernarea.
Thefirst phase studiesof theconnectionsbetween
themainclassesof landslidesinDaunia, Li, themain
geo-hydro-mechanical set-ups, GMi, andtheexternal
landslidefactors (Terzaghi 1950) arestill under way.
Phenomenological (I level) connections have been
envisagedandbothII level (usingMorgenstern&Price
method) and III level analyses (using finiteelement
modellingandfieldmonitoring) arebeingcarriedout
to verify theI level results across theregion. Based
uponI level analyses, mostof theL1andL2landslides
havebeenfoundtobeactivatedintheclaysatthebase
of theslopesandtoretrogressupslope. Conversely, L3
slides areoftenfoundto start at thetopof theslope
andhaveadvancingactivity. Thestrengthof theslope
clays(i.e. thesoil properties) isseentoinfluencethe
depthof theslidingprocess, e.g. L1slidesoccur more
frequentlyinstiffer clays, becomingdeeper thestiffer
istheclay, whereasL2slidesoccur moresofter clays.
Thecontactbetweentherockandclaystrataisseento
influencethetypeof retrogression. For example, the
L2slides haveseveral sourceareas whenretrogress-
inginclays, whereas they tendto haveasingledeep
source area when retrogressing in the top rock stra-
tum. Inaddition, most of thelandslides intheregion
arereactivatedprocesses, whosefirstfailureoccurred
before the beginning of last century. They can be
mainlyclassifiedasveryslowtoslow-ratemovements
(v-510
3
mm/s; Cruden & Varnes 1996), with
possibleaccelerationsafter extremerainfall events.
In thefollowing, preliminary results of theappli-
cation of the methodology for the landslide hazard
assessment inaspecific urbanareaof theregionare
discussed, asresultingfromI andII level analyses.
5 PRELIMINARY 2NDPHASE RESULTS: THE
CASE OF CELENZA VALFORTORE
As for most of the urban centres in the Daunia
Apennines, the promontory of CelenzaValfortore is
borderedbythecrownsof several landslides(Fig. 3a).
Theapplicationof thesecond phase of themethodol-
ogy toCelenzaValfortorehasrequiredthecollection
of all theavailabledataresultingfromprevious site-
specific surveys andinvestigations (e.g. geomorpho-
logical, stratigraphical, inclinometricandpiezometric
data), anddataconcerningdamagestostructurestrig-
geredbylandsliding, together withnewdataresulting
fromon-purposein-situsurveys. All thesedatahave
beenimplementedinaGISdatabase, accordingtothe
guidelinesdevelopedinthefirst phase of thework.
Theconsultationof theGISdatabaseinlightof the
GMi set-ups identified in the region has resulted in
therecognitionof thepromontory of Celenzaas part
of categoryGM2(Fig. 2b). Itisformedbyasequence
of SBO rock strata, SBO clay strata and FYR clay
stratainmonoclinal contact. Theoldpart of thetown
isfoundedontheoutcroppingof thearenaceousSBO
flysch, whereasthemostrecentarea(builtafter 1920)
liesdirectlyontheoutcroppingof eithertheSBOorthe
FYR clay strata(Fig. 3a). Thecomparativeconsulta-
tionof theGISdatabaseof Celenzaandthat resulting
fromthe 1 phase studies across the region, in light
of thelandsidetypologyclassesLi, hasresultedinthe
recognitionof asingletypeof landslidemechanismon
mostof theslopessurroundingthepromontory, thatis
themediumdepthmudslideL2(Fig. 2c). Theseland-
slidesinvolveboththeFYR scaly claysandtheSBO
clays. In most cases, they areretrogressiveand their
rearscarpeitherbordersorinvolvesthefoundationsof
thebuildingsat thetopof theslopes. Whenthemud-
slide develops solely in FYR clays (i.e. the weakest
clays: maximumstrengthc
18kPa
20
) it has
several sourceareas, whereasitisof singlesourcearea
wheninvolveseither thestiffer SBOclays(maximum
strengthc
10kPa
25
) or approachestheSBO
arenitesat thetop.
II level limit equilibriumanalyses havebeen car-
ried out for the susceptibility assessment of slopes
located in the southern portion of the promontory.
368
Figure4. CelenzaValfortore:Landslide1,limitequilibrium
analysis.
In particular, two interactive landslides, respectively
indicated as 1 and 2 in Figure 3, have been anal-
ysed; landslide1isaL2mudslidewithseveral source
areas, whereaslandslide2isaL2withasinglesource
area, which involves a portion of the built area of
thetown(Fig. 3). AccordingtoI level analyses, both
theslidesareretrogressivereactivationprocesses.The
failureprocessof landslide1islikely tohavestarted
at the toe about the river due to erosion processes.
The activation of landslide 2 has probably followed
themovement of landslide1, giventheinteractionof
itstoewithlandslide1.Thelimitequilibriumanalyses
havebeenaimedat identifyingthehierarchyof insta-
bility betweentheslidingbodies (inorder to deduce
thewaytheyinteract)andthemobilizedfrictionangles
along the slip surfaces. The limit equilibriumanal-
yses for landslide1havebeen carriedout in section
A AD, whoseaxisisshowninFigure3andwhich
includes only FYR clays (Fig. 4). The analyses for
landslide2havebeen carried out in asection which
starts at the crown of this landslide (E in Fig. 3),
reachesthemid-slopetoeatpointA (Fig. 3) andthere-
after follows landslide1down to thetoeat theriver
(sectionaxisEAA, Fig. 3). Figure5showsthat
thissectionincludesFYR claysinthelower part and
SBO clays in the top part. As first, the steady-state
pore pressure regime in the slopes has been recon-
structed based upon finiteelement seepageanalyses
(SEEP/WGeostudio2004), providingresults consis-
tentwithfieldmeasurements.Thereafter,suchseepage
regime has been implemented in limit equilibrium
analyses(SLOPE/WGeostudio2004; Morgenstern&
Price1965). For body 1, different slip surfaces have
been analyzed, which haveall been assumed to pass
throughthetopandtoeof thelandsliderecognizedin
thefield(DandA respectively, Fig. 3).
In addition, c
=0kPa has been assumed for the

FYR clay. Accountingfor theseconditions, theanal-
yses demonstrate that the slip surface of the most
unstable body in section A AD has maximum
depthof 25m(Fig. 4); F=1applies to this slipsur-
face if
=15
, that is a value consistent with the

strengthpropertiesof FYR clayswhenshearingpost-
peak(at largestrains), but far beforeresidual strength
conditions(
r
6
). For landslide2, thelimitequilib-

riumanalyseshavecomparedthestabilityof landslide
bodies followingtwo cinematic hypotheses. Figure5
shows a sliding body (hypothesis a) that coincides
with landslide 1 in the lower part (portion A A
of the section, Figs. 3, 4) and extends upslope fol-
lowingsectionAE upto thecrown; this hypothesis
Figure 5. Celenza Valfortore: Landslide 2 (A AE
in Fig. 3), limit equilibrium analyses: a) hypothesis a;
b) hypothesisb.
simulates aconnectionof slidingbetweenlandslides
1and2. ThisbodycrossestheFYR claysinthelower
part and theSBO clays in theupper part. Along the
slipsurfacecrossingFYR clays, thesamemobilized
strengthparametersapplyingtolandslide1havebeen
implemented(c
=0kPa,
=15
).Theanalyseshave
lookedforthecritical depthof thislandslideinthepor-
tioncrossingSBOclay; this has resultedto beabout
35m(Fig. 5a), withmobilisedstrengthparametersof
SBOclay:c
=0kPaand
=18
19
(whicharecon-
sistentwiththepost-peakstrengthvaluesof SBOclay).
ThelandslidebodyforHypothesisb(Fig.5b)hastoeat
pointA (Figs3, 5b), crossessolelytheSBOclaysand
reaches starts thecrown E at thetop. Implementing
in theseanalyses thesameSBO strength parameters
asresultingfromtheanalysesfollowinghypothesisa
(i.e. c
=0kPaand
=18
), it results that any land-

slidebodymovinginsectionA A E withtoeatA
ismorestablethanthelonger landslidebody(hypoth-
esis a) moving with toe at the river. Therefore, the
limitequilibriumanalysesappeartoconfirmtheI level
interpretation of thelandslidemechanism, assessing
thattheactivationof landslide2hasfollowedtheslip-
pingof landslide1andthat at present theactivity of
landslide2islikely tobeconnectedtomovementsin
thelower part of theslopes, downtotheriver.
6 RESEARCHPERSPECTIVES
The further development of the research will take
advantage of the results of several on-going in-situ
investigations within different Daunia active land-
slides, of laboratorytestingonthesoilsbeingsampled
aswell asof theresultsof numerical modelling, which
will allowfor theevaluationof thevalidityof boththe
I andII level assessments conductedsofar. Numeri-
cal analysesimplementingstatistical processingof the
external factorswill alsoprovideindicationsaboutthe
timefactor for appropriatehazardassessments.
ACKNOWLEDGMENTS
TheStrategicResearchProject n. 119Landsliderisk
assessment for theplanning of small centres located
369
inchainareas: thecaseof theDauniaregion funded
bytheApulianRegion.
REFERENCES
Aleotti, P. & Chowdhury, R. 1999. Landslidehazardassess-
ment: summary reviewandnewperspectives. Bull. Eng.
Geol. Env., 58, 2144, Springer-Verlag.
Canuti, P & Casagli, N. 1994. Considerazioni sulla valu-
tazionedel rischiodi frana. Proc. Conf. Fenomeni franosi
e Centri Abitati, 27th May 1994. Bologna(Italy).
Carrara, A., Cardinali, M., Detti, R., Guazzetti, F., Pasqui,
V., Reichenbach, P. 1991. GIS techniques and statisti-
cal modelsinevaluatinglandslidehazard. Earth Surface
Processes and Landforms, 16, 427445.
Chowdhury, R. & Flentje, P. 2002. Modern approaches for
assessmentandmanagementof urbanlandslides. 3rd Int.
Conf. Landslides, Slope stability & the safety of infra-
structures. 1112 July 2002. Singapore.
Cotecchia, F., Santaloia, F. 2003. Compressionbehaviour of
structurallycomplexmarineclays.Nakase Memorial Sym-
posium on Soft Ground Engineering in Coastal Areas.
2003: 6372. Nagase,Yokosuka, J apan.
Cotecchia, F., Vitone, C., Cafaro, F., Santaloia, F. 2006. The
mechanical behaviour of intensely fissured high plas-
ticity clays fromDaunia, Characterisation. and engng
properties of Natural Soils, Singapore, 19752003.
Cotecchia, F., Vitone, C. & Santaloia, F 2008. Theinfluence
of intensefissuringonthecompressionbehaviour of two
Italianclays, IV Int. Symp. on Deformation Characteris-
tics of Geomat (IS2008), 2224 September 2008. Atlanta,
Georgia.
Cruden, D.M. & Varnes, D.J. 1996. Landslides Types and
Processes. In: Turner A.K. & Schuster R.L. (eds.) Land-
slides: Investigation and Mitigation. Transp. Research
BoardSpecial Report 247. Nat. Ac. Press, WA, 3675.
Dai, F.C., Lee, C.F. & Ngai, Y.Y. 2002. Landslide risk
assessment and management: an overview. Engineering
geology, 64, 6587.
Esu, F. 1977. Behaviour of slopes in structurally complex
formations. The Geotechnics of Structurally Complex
Formations; Proc. intern. symp.,Capri (Italy),2,292304.
Fall, M, Azzam, R. & Noubactep, C. (2006). A multi-
methodapproach to study thestability of natural slopes
and landslide susceptibility. Engineering geology, 82,
241263.
Fell, R., Corominas, J., Bonnard, C., Cascini, L., Leroi, E.,
Savage, W.Z. et al. 2008. Guidelines for landslide sus-
ceptibility, hazardandrisk zoningfor landuseplanning.
Engineering and Geology, 102, 83111.
GeoStudio2004. Geo-SlopeInt. Ltd., Alberta, Canada.
Guzzetti, F., Cardinali, M., Reichenbach, P & Carrara, A.
2000. Comparing landslide maps: a case study in the
upper Tiber river basin, Central Italy. Environmental
management, 25(3), 247263.
Ho, K., Leroi, E. & Roberts, B. 2000. Keynote Lecture:
quantitativeriskassessment:application,mythsandfuture
direction. Proc., GEOENG 2000, Melbourne, Australia,.
Technomic, Lancaster, 1, 263312.
Hutchinson, J.N. 1988. General Report: Morphological and
geotechnical parametersof landslidesinrelationtogeol-
ogy and hydrogeology. Proc. of the 5th Int. Symp. on
Landslides, Lausanne, 1, 336.
Leroueil, S. &Locat, J. 1998. Slopemovements Geotechni-
cal characterization, risk assessment andmitigation. 8th
Int. IAEG Congress, Balkema, Rotterdam, The Nether-
lands, 933944.
Mancini, F., Ceppi, C., Ritrovato, G. 2008. Analisi del ris-
chiodafranainambienteGIS: Il casodel Sub-Appennino
Dauno (Puglia). Proc. ASITA, LAquila, ISBN 978-88-
903132-1-9. II, 13931398.
Melidoro, G. 1982. Indagini geologiche sulle condizioni
di stabilitdellabitato di CelenzaValfortore. Relazione,
130.
Montgomery, D.R. &Dietrich,W.E. 1994.Aphysicallybased
model for thetopographiccontrol onshallowlandsliding,
Water Resources Research, 30, 11531171.
Morgenstern, N.R. & Price, V.E. 1965. The analysis of
the stability of general slip surface. Gotechnique, 15,
239247.
Savage, W.Z., Godt, J.W., Baum, R.L. 2004. Modelingtime-
dependent areal slopestability, InLacerda, W.A., Erlich
M., Fontoura S.A.B., Sayao A.S.F., (eds.) Landslides -
Evaluation and Stabilization, Proc. of the 9th Int. Symp.
on Landslides, Balkema, 1, 2336.
Terzaghi, K. 1950. Mechanisms of Landslides. Geol. Soc.
Am., BerkleyVolume, 83123.
VanWesten, C.J., Rengers, N., Terlien, M.T.J. & Soeters, R.
1997.Predictionof theoccurrenceof slopeinstabilityphe-
nomenathroughGIS-basedhazardzonation. Geologische
Rundschau, 86, 404414.
Vitone, C., Cotecchia, F., Desrues, J., Viggiani, G., 2009. An
approachtotheinterpretationof themechanical behaviour
of intensely fissuredclays. Soils & Foundation Journal,
inpress.
Yin K.L. & Yan T.Z. 1988. Statistical prediction models
for slope instability of metamorphosed rocks. In: Bon-
nardC. (eds.), Proc. 5th Int. Symp. Landslides, Lausanne,
Switzerland, vol. 2, Balkema, Rotterdam, 12691272.
370
Reliability-basedperformanceevaluationfor reinforcedrailway
embankmentsinthestaticloadingcondition
M. Ishizuka
Integrated Geotechnology Institute Limited, Japan
M. Shinoda
Railway Technical Research Institute, Japan
Y. Miyata
National Defense Academy of Japan, Japan
ABSTRACT: Theshift fromtheallowablestressdesignmethodtothelimit statedesignmethodisadvanced
now. There is a limit when the soil materials and the reinforcements that have potentially variabilities are
quantitativelyevaluatedthoughthelimitstatedesignmethodisaneffectivedesignmethodwhentheperformance
is designed. Inthis research, to evaluatetheperformanceof thestructurethat usedthesoil materials andthe
reinforcementsquantitatively, thereliabilityanalysiswasexecuted.
1 INTRODUCTIONS
When the safety of the structure is evaluated, it is
necessary to handlethesoil modulus reasonably and
quantitatively though it is known well that the soil
modulus vary when various structures are designed
intended for the soil and the bedrock. In the design
that uses the safety factor in a past allowable stress
method, it hascorrespondedby adoptingthevalueof
thesafety sideasavaluefor thedesignwhenuneven
tothegroundconstant etc. However, it wasaproblem
that thetechniquefor evaluatinghowtakingthevalue
of thesafety sidelogically andquantitatively hadnot
beenestablished.
Thereliabilitydesignmethodisatechniquefor can
the evaluation as the safety index or the breakdown
probability(Hereafter, it isrecordedasthelimit state
exceedanceprobability) that is theindex to ruin the
safetyof thestructurebyapplyingaprobabilityanda
statistical theorytotreat theuncertaintylogicallyand
quantitativelytothedesignof thestructureandbecom-
ingit. Quantitativelytreatingthesoil modulusetc. that
haveapotential differencebecomespossiblebyapply-
ingthistechnique. Becausethelimitstateexceedance
probability is an exceedanceprobability of defining
the state of the limit presented in the performance
check typedesignmethod, it isnot necessarilyequiv-
alent to the breakdown probability in case of safety
factor-1.
The reliability design method is divided at three
levels. A so-called limit state design method hits
this in the technique for giving reliability by the
loadcoefficient andtheresistancecoefficient though
the occurrence probability of the mode of break-
ing is not quantitatively appreciableat level I, Level
II is a technique for evaluating reliability request-
ing the limit state exceedance probability fromthe
safety index obtained fromthe mean value and the
standarddeviationof theperformancefunction. Con-
cretely, theFirst- OrderandSecondMoment(FOSM)
method
26
and the First order reliability (FORM)
method
58
,etc.areenumerated.Level III isatechnique
for requesting the limit state exceedance probabil-
ity directly from assumption that characteristics of
an uncertain factor concerning the mode of break-
ing the probability statistical are all already-known.
Concretely, the Monte Carlo method
5,6,911
etc. are
enumerated.
The case where the reliability design method is
applied to the soil structure is enumerated the plu-
ral, andhasresultsespeciallybytheembankment
12,13
andthehavingretainingwall
14,15
, etc. intheexamina-
tionof slopesafety
3,4,14,16,1719
.Ananalytical research
on the reinforcement embankment is actively done
recently
2022
.
Thereinforcement embankment is targetedinthis
research. Level II-reliability analytical methodinthe
loadingconditionwasalwaysappliedandtheheightof
theembankmentor thesafetyindexandthelimitstate
exceedanceprobabilitytotheperformancerank were
calculatedbasedonthedesignbasisof therailway
23
.
And, therelation between thoseindices and conven-
tional safety factor was considered. Refer to As for
the calculation of the cost of the life cycle of the
reinforcement embankment at the earthquake, it is
document
24
.
371
2 ANALYSISMETHOD
2.1 Calculation of safety factor
Safety factor FS of thereinforced embankment was
calculatedby themodifiedFellenius methodlikethe
next expression.
K
h
isthehorizontal seismiccoefficienthere,M
dw
isthe
slidingmoments by its ownweight, M
rw
is theresis-
tancemoment by its ownweight, M
rc
is theresistant
moment by cohesion, M
rt
is theresistant moment by
reinforcements, M
dk
istheincrementof standardslid-
ingmoment by unit seismic inertiaforce, andM
rk
is
thedecreaseof resistantmomentbyunitseismicresis-
tant.Becausetheloadingconditionisalwayslimitedin
staticthisresearch, thetermof thehorizontal seismic
coefficient is not considered. It based on Moreover,
the partial safety coefficient concerning pulling out
reinforcement and thepartial safety coefficient con-
cerningreinforcement breakingstrengthwererailway
standards
23
.
2.2 Calculations of safety index and limit state
exceedance probability
Thereliabilityanalysistechniqueusedbythisexami-
nationisFORM inlevel II-reliability designmethod.
If the design point is one in FORM, and the per-
formance function is linear, the solution with good
accuracy will beobtained in a short time. However,
whentwoor moredesignpointsandtheperformance
functionisnonlinear, abigerror marginiscaused. In
this examination, thedesign point used fromtarget-
ingtheembankmentconstructedonthesteadyground
andtheperformancefunctionusedFORMwithlinear
by oneto beassumable. It easily explains FORM as
follows.
Thevaluefor thedesignisassumed, andthefunc-
tionthat hasX
1
X
n
isassumedtobeg, andthevalue
of theperformancefunctionis assumedto beZ. The
occurrenceof thelimitstateexceedancecanbejudged
asfollows.
Here, performancefunctionZwasset asfollows.
Inthisresearch, FSwasassumedtobeexpression(1).
If thevalueof theperformancefunctionis apositive
value, itcanbesaidexcessivelyof thestateof thelimit
thatthestructureissafeaccordingtoexpression(2). If
theperformancefunctionhas0or anegativevalue, it
meansthestateof thelimit isexceeded, andthereisa
structureatriskaccordingtoexpression(3). Itisdiffi-
cult for performancefunctionZtobecomeacomplex
functionincaseof almostandtorequestthelimitstate
exceedance probability strictly. Then, the limit state
exceedanceprobability is calculatedby performance
functions Z in the Taylor expansion, discontinuing
the series by the first order term, and making it to
linear. InFORM, it expands by thecircumferenceof
becomingof theperformancefunction0.Thatis, when
theTaylor expandsby designpoint x
j
circumference
of abasicrandomvariable, thefollowingexpressions
obtainperformancefunctionZ.
PerformancefunctionZ meanvalue
z
anddispersion
z
arerequestedasfollows.
Safety index canbeobtainedas follows by theuse
of expression(6) andexpression(7).
A basicrandomvariableisandthereismutuallyinde-
pendentlyarelationbetweenthefollowinguptosafety
indexandlimitstateexceedanceprobabilityP
f
(Z0)
accordingto thenormal distributionfunctionincase
of special.
+ is astandardnormal probability distributionfunc-
tion, andthesafetyindex isastandardherebywhich
howmeanvalueZisrelativelyawayfrompoint(Z=0)
dangerouslyisshown.Thatis,thesafetyindex grows
bythemeanvalueof performancefunctionZlargeand
standarddeviationsmall, andthereisroominsafety.
Moreover, if standard deviation is large even if the
meanvalueof performancefunctionZislarge(equiv-
alencetothesafety factor large), it isnot necessarily
safe. Thisrespect isabigdifferencepoint withapast
designmethod.
3 ANALYTICAL MODELS
In this research, the assumption specification
described in P. 384 of document
23
was examined on
thedesignbasisof therailway. Therearetwokindsof
372
Figure 1. Cross sections of embankments (Performance
rankI, gradient of slopes1: 1.8)
examinedstructures, that is, reinforcedembankment
(withlongreinforcement) andnoreinforcedembank-
ment(withoutlongreinforcement). Intheassumption
specification, the reinforced embankment is speci-
fiedfor theperformancerank I andthenoreinforced
embankment isspecifiedfor theperformancerank II
and III. Figure 1 shows the reinforced embankment
model sectionof theperformancerank I, andFigure
2showsthenoreinforcedembankmentmodel section
of theperformancerankII andIII.
Theheight of theembankment wasassumedtobe
four kinds, 3, 4.5, 6, and9m. Theembankment incli-
nationof theperformancerankI wasadjustedto1: 1.8
andtheperformancerank II andIII wereadjustedto
1: 1.5.Theembankmentsectionhasbeendividedinto
two (the surface part and the deep part) according
to P. 58 of document
23
.The range of 2mis called a
embankment surface part (henceforth surface) from
theembankment slope, and other parts areprovided
for embankment deeppart (henceforthdeep)(Fig. 3).
Theapplicationof thevaluefor thedesignof theposi-
tion wherereinforcement is constructed and thesoil
(Table1) isdirectedbasedonthis.
Moreover, therearedivision into two such as the
upper part of embankment and the lower side of
embankment in the soil.
23
. Concretely, the whole is
treatedastheupper partof theembankmentfor 3min
theembankment height onthegroundsidefromtop.
Incaseof 4.5mand6minheight, theupper partof
theembankment fromtopto3m, and3mor lessare
thelower sideof theembankment. Incaseof 9min
height,theupperpartof theembankmentisfromtopto
3m, 3mor lessarethelower sideof theembankment.
Especially, it was assumedananalytical model by
whombermof 2.0minwidthwasinstalledinthecase
Figure 2. Cross sections of embankments (Performance
rankII andIII, gradient of slopes1:1.5)
Table1. Soil material parameters(meanvalue).
Soil Soil Soil
Material Unit 1 2 3
unit weight
t
kN/m
3
18 17 16
Surfacepart Cohesionc kN/m
2
3 3 3
Internal deg. 40 35 30
friction
angle+
Deeppart Cohesionc kN/m
2
6 6 6
Internal deg. 45 40 35
friction
angle+
with9minembankmentheightbecausetherewasreg-
ulationsthatinstalledbermof 1.5minstandardwidth
when the height of the embankment exceeded 6m.
Thatis,theupperpartof theembankmentandthelower
sideusessoil 1forbothandtheperformancerank.The
performancerank II uses soil 1upper part andsoil 2
isusedlower part. Theperformancerank III usessoil
2intheupper part andsoil 3isusedlower part.
Reinforcement wasassumedtobetwokinds(long
andshortreinforcement). Theconstructioninterval of
shortreinforcementis0.3mandtheconstructioninter-
val of longreinforcementis1.5m.Thedesignvalueof
reinforcementisindicatedinTable2. Shortreinforce-
mentwasconstructedintheembankmentsurfacepart
andlongonewasconstructedinthedeeppart(Fig. 3).
373
Table2. Reinforcement material parameters.
Reinforcement Unit Meanvalue
Long kN/m 30
Short kN/m 2
Figure3. Divisionof embankment surfaceandsubsurface.
Table3. Overburdenloads
Performance Track Load
rank structure Unit value
I Concrete kN/m
2
15
IIIII Ballast kN/m
2
10
Table4. Coefficientsof variation.
Unit weight 0.05
Cohesionc 0.10
Internal frictionangle 0.10
Longreinforcement 0.05
short reinforcement 0.05
Thecaseof 9minheight wasexaminedasoneexam-
plethoughit becameacomplex analytical sectionby
theberminstallationfor 6mor moreinheight.
Besidesthis, theappliedloadings
23
wereshownin
Table 3, and the applied coefficient of variation
25,26
isshowninTable4. Here, theprobabilitydistribution
of the soil modulus and the reinforcement constant
wasassumedtobenormal distributionandamutually
independent. Themeanvalueof thesoil modulus set
thevaluebasedonthestandardof therailway.
4 ANALYSISRESULTS
Figure 4 shows the safety index and the limit state
exceedance probability to the embankment height.
Bothmutuallyareanequivalent values, andshowthe
safety side so that the larger the value of the safety
index is, thesmaller thevalueof thesafety sideand
thelimit stateexceedanceprobabilityis.
Asforthesafetyindex,itisunderstoodthatthemax-
imumvaluein4.5mintheembankmentheightistaken
in(a), andthevaluehasdecreasedastheembankment
rises, andeachheight of theembankment isrelatively
highthevalue. It isthought that theeffect of therein-
forcement constructed in embankment deep part is
high fromthis. Moreover, theimpetus by theheight
of theembankment havingincreasedisthought that it
Figure 4. Embankments height versus safety index and
limit stateexceedanceprobability.
is acausethat theresistancepower by thereinforce-
ment construction was surpassing as for the reason
why the safety index takes the maximumvalue for
4.5min embankment height. In the safety index in
(b), the point being seen for the tendency to which
thevaluedecreasesassimilar to(a), theembankment
height rises lowin (b) is an examination problemin
thefuture. Thesafety index in (c) is decreased from
3minembankment height to6m, andhas recovered
by 9m. It is thought that the reason why the safety
index hasrecoveredintheembankment 9mheight is
aneffectof installingberm.Thetendencysimilartothe
safetyindexisshownaboutthelimitstateexceedance
probabilityineachperformancerankof (a)(c).
374
Figure 5. Embankments height versus Safety index and
safetyfactor.
Figure5showstherelationbetweenthesafetyindex
andthesafety factor to theembankment height. The
safety index is thesamedataas Figure4. Thesafety
factor isalmostconstantin(a).Asfor thesafetyfactor
of (b)and(c), thevalueincreasesbothwithanincrease
of theheight of theembankment thoughit decreases
when it reaches 9m. When thesafety index is com-
pared with the safety factor, it has been understood
thattheincreaseanddecreasetendencytosafetyindex
andlimit stateexceedanceprobabilitytotheheight of
theembankment iscorrespondingin(a)(c).
Figure6. Performancerank versus safety index and limit
stateexceedanceprobability.
andthelimit stateexceedanceprobability to theper-
formancerank. Theembankment inclinationof each
performance rank was fixed to a horizontal axis. It
hasbeenunderstoodthatthesafetyindexandthelimit
375
Figure7. Performancerank versussafety index andsafety
factor.
stateexceedanceprobability showasimilar increase
anddecreasetendencyfrom(a) to(d).
and the safety factor to the performance rank. The
safety index is thesamedataas Figure6. As for the
safety factor, thedecreaseof thevalueis seen from
(a) to (d) inorder of performancerank I, II, andIII.
It hasbeenunderstoodthat asimilar tendencyisseen
in therelation between theincreaseand decreaseof
the safety index and the safety index and the safety
factor.
5 CONCLUSION
Theshiftfromtheallowablestressmethodtothelimit
statedesignmethodisadvancednow. Thereisalimit
when the soil materials and the reinforcements that
has potentially variabilities are quantitatively evalu-
atedthoughthelimitstatedesignmethodisaneffective
designmethodwhentheperformanceisdesigned. In
thisresearch, toevaluatetheperformanceof thestruc-
turethatusedthesoil materialsandthereinforcements
quantitatively, thereliabilityanalysiswasexecuted.
Concretely, the reinforced and no reinforced
embankmentsinthestaticloadingconditionof 3, 4.5,
6, 9mheight weremodeledthat usedthedesignbasis
of therailway structurestandard in J apan. Theanal-
ysis that assumed circular arc destruction was done,
thesafety index, thelimit stateexceedanceprobabil-
ity, andthesafetyfactorwerecalculated. FORM(First
order reliability method) was usedabout thecalcula-
tionof thesafetyindexandthelimit stateexceedance
probability, and the modified Fellenius method was
used for the calculation of the safety factor. FORM
hasthefeatureof beingobtainthesolutionwithgood
accuracyinashorttime. Thatisoneof thetechniques
for evaluatingreliability by requestingthelimit state
exceedanceprobabilityfromthesafetyindexobtained
fromthe mean value and the standard deviation of
theperformancefunction. However, it isnecessaryto
meettherequirement, thatthedesignpointistheonly
and theperformancefunction is linear when FORM
isapplied. Accordingly, embankmentsconstructedon
the steady ground where met the requirement was
assumed.Whenmodeling, thepointtohavepaidatten-
tionbesides embankments height weregradeof face
of slopes, arrangementof soil materials, andpresence
of berm.
When the safety index and the limit state
exceedanceprobability that wastheresult of thereli-
ability analysis and the safety factor that was result
of allowable stress method were compared, it was
understood that the tendency to safety was approxi-
matecorresponding. Thefeatureand theproblemin
thefutureof theanalytical result arefrom1) to5).
1) Intherelationbetweentheembankmentheightand
thesafety index, thevaluebecomes small as the
embankment rises andthevalueof thelimit state
exceedanceprobabilityhasgrownastheembank-
ment rises in 3, 4.5, and6mof theperformance
rankII andIII.
2) Therelationbetweentheembankment height and
thesafetyfactorisconstantintheperformancerank
I, and the value of the safety index has become
small as theembankment rises in3, 4.5, and6m
of theperformancerankII andIII.
376
3) Intherelationbetweentheperformancerank and
thesafety index, thesafety index becomes small
as the performance rank rises. Similarly, in the
relation between the performance rank and the
limit state exceedance probability, the limit state
exceedanceprobability growsastheperformance
rankrises.
4) Intherelationbetweentheperformancerank and
thesafetyfactor, thesafetyfactor decreasesasthe
performancerankrises.
5) Asfor thecalculationresult of 9mintheembank-
mentheight, withberm, itwasshownthatstability
wascomparativelyhighbyarrangingtheembank-
mentheight. Itisthoughtthatitisnecessarytotreat
thesafetyindex, thelimit stateexceedanceproba-
bility, andthesafetyfactor besidesother embank-
ment height, and the examination of analytical
model includingbermisfuturetasks.
It was assumed static loading condition in this
research.Inthefuture,trainloadingconditionandseis-
mic loading condition, etc. were examined, and the
reliability of thereinforcement embankment will be
evaluatedoverall.
REFERENCES
[1] Hoshiya, M., Ishii, K. 1986. The reliability design-
ing of structures: Kajima InstitutePublishing (In
J apanese).
[2] Cornell, C. A. 1967. Bounds on thereliability of
structural systems. Journal of Structural Engineer-
ing, 93(1) : 171200: ASCE.
[3] Wu, T. H. andKraft, L. M.1970. Safetyanalysisof
slopes, Journal of Soil Mechanics and Foundation
Division, 96(2): 609630, ASCE.
[4] Tang, W. H., Yucemen, M. S. and Ang, A. H.-
S. 1976. Probability-based short termdesign of
soil slopes, Canadian Geotechnical Journal, 13:
201215
[5] Ayyub, B. M. andHaldar, A.1984. Practical struc-
tural reliability techniques, Journal of Structural
Engineering, 110(8): 17071724, ASCE.
[6] Lian, Y. and Yen, B. C. 2003. Comparison of
risk calculation methods for aculvert, Journal of
Hydraulic Engineering, 129(2): 140152.
[7] Hasofer, A. M. and Lind, N. C. 1974. Exact
and invariant second-moment code format, Jour-
nal of Engineering Mechanics Division, 100(1):
111121. ASCE.
[8] Rackwitz, R. andFiessler, B. 1978. Structural reli-
ability under combined randomload sequences,
Computers and Structures, 9: 489494.
[9] Nguyen, V. U. and Chowdhury, R. N. 1985. Sim-
ulationfor risk analysis withcorrelatedvariables,
Gotechnique, 35(1): 4758.
[10] Gui, S., Zhang, R., Turner, J. P. andXue, X. 2000.
Probabilisticslopestabilityanalysiswithstochastic
soil hydraulic conductivity, Journal of Geotech-
nical and Geoenvironmental Engineering 126(1):
19.
[11] El-Ramly, H., Morgenstern, N. R. andCruden, D.
M. 2002. Probabilistic slopestability analysis for
practice,Canadian Geotechnical Journal,39:665
683.
[12] Matsuo, M. and Kuroda, K. 1974. Probability
approach to design of embankments, Soils and
Foundations, 14(2): 117.
[13] Christian, J. T., Ladd, C. C. and Baecher, G. B.
1994. Reliability applied to slopestability analy-
sis, Journal of Geotechnical Engineering, 120(12):
21802207, ASCE.
[14] Duncan, M. J. 2000. Factorsof safetyandreliability
in geotechnical engineering, Journal of Geotech-
nical and Geoenvironmental Engineering, 126(4):
307593.
[15] Hoeg,K.A.M.andMurarka,R.P.1974.Probabilis-
ticanalysisanddesignof aretainingwall, Journal
of Geotechnical Engineering, 100(3): 349365,
ASCE.
[16] Vanmarcke, E. H. 1977. Reliabilityof earthslopes,
Journal of Geotechnical Engineering, 103(11):
12471265, ASCE.
[17] Alonso, E. E. 1976. Risk analysis of slopeandits
application to slopes in Canadian sensitiveclays,
Gotechnique, 26(3): 453472.
[18] Bergado,D.T.andAnderson,L.R.1985.Stochastic
analysisof porepressureuncertaintyfor theprob-
abilistic assessment of thesafety of earth slopes,
Soils and Foundations, 25(2): 87105.
[19] Li, K. S. andLumb, P. 1987. Probabilistic design
of slopes, Canadian Geotechnical Journal, 24(4):
520535.
[20] Shinoda, M., Horii, K., Yonezawa, T., Tateyama,
M. and Koseki, J. 2006. Reliability-based seis-
micdeformationanalysisof reinforcedsoil slopes,
SoilsandFoundations, 46(4): 477490.
[21] Shinoda, M., Yonezawa, T., Tateyama, M. and
Koseki, J. 2005. Limit state exceedance prob-
abilities of reinforced retaining walls, Journal
of Geotechnical Engineering, 792, III(71): 119
129.(InJ apanese)
[22] Shinoda, M., HaraK., Masuo, T. Koseki, J. 2006.
Reliabilityanalysisonreinforcementsoil structure
that considers statistical character of reinforce-
ment breaking strength, Geosynthetics Engineer-
ing Journal, 21: 8996. (InJ apanese)
[23] RailwayTechnical ResearchInstitute,2007.Design
standard for railway structures., Soil structures,
MaruzenCo., Ltd.. (InJ apanese)
[24] Watanabe, K., Ohki M., Shinoda, M., Kojima, K.
andTateyama, M., 2005. Triaxial tests of thesoil
strengthparameters usedincheckingonembank-
ments stability, Quarterly Report of RTRI, Vol.
19(3), 2934. (InJ apanese)
[25] Hara, K. Masuo, T., 2005. Variability evaluation
of geogrid tension strength in the ISO geotextile
examination standard, Geosynthetics Engineering
J ournal, vol. 20, 287294. (InJ apanese)
377
Maximumlikelihoodanalysisof casehistoriesfor probability
of liquefaction
J ianyeChing
Department of Civil Engineering, National Taiwan University, Taipei
C. HseinJ uang
Department of Civil Engineering, Clemson University, Clemson, USA
Yi-HungHsieh
Department of Construction Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan
ABSTRACT: In this paper, themaximumlikelihood method is used to deriveprobabilistic models for liq-
uefaction potential evaluation using piezoconepenetration test (CPTU). Emphasis of thepaper is placed on
establishingtherelationshipbetweenthefactor of safetycomputedwithanewlydevelopeddeterministicCPTU
model andtheprobabilityof liquefaction. Themodelingerror of thelimit statefunctionischaracterizedwitha
biasfactor andseveral modelsof thebiasfactor areinvestigatedwithinthemaximumlikelihoodanalysisframe-
work. Thebest performingmodel is judgedbasedonthecriteriaof thewell-acceptedInformationTheory, as
well asthroughtheproposedprobability-probability(P-P) plot. Thesignificanceof thedevelopedprobabilistic
model isdiscussedandcomparisonwiththeexistingprobabilisticmodelsispresented.
1 INTRODUCTION
In this paper the maximum likelihood analyses of
a liquefaction database are carried out to develop
probabilistic models for liquefaction potential eval-
uationusingpiezoconepenetrationtest (CPTU). The
liquefaction databaseconsists of casehistories com-
piled by Chen et al. (2008) fromdifferent sources.
Each case, either liquefied or non-liquefied, is char-
acterizedwiththepiezoconepenetrationtest (CPTU)
data, including cone tip resistance (q
t
), sleeve fric-
tion( fs), andpenetrationporewater pressure(u
2
), and
theseismicparameters, includingpeakgroundsurface
acceleration(a
max
) andmomentmagnitude(M
w
). One
objectiveof thestudy is to developamappingfunc-
tionthat relatestheprobabilityof liquefaction(P
L
) to
thefactor of safety(F
S
) against theoccurrenceof liq-
uefactioncomputedwithaCPTU model (Chenet al.
2008; J uang et al. 2008). This mapping function, an
empirical relationship between P
L
and F
S
, is devel-
oped through themaximumlikelihood analysis. The
developedrelationshipallowsfor apractical, one-step
estimateof theliquefaction probability P
L
based on
thecomputedF
S
.
Thesimplifiedprocedurefor evaluatingthelique-
faction potential of a soil was created by Seed and
Idriss(1971) inwhichtheseismicloadingrequiredto
initiateliquefaction was expressed as acyclic stress
ratio (CSR) and the soil resistance against lique-
faction was determined through correlations with in
situ test such as thestandard penetration test (SPT),
conepenetration test (CPT), Becker penetration test
(BPT), and shear wave velocity (V
s
) measurements.
An excellent summary of theSPT-, CPT-, BPT-, and
Vs-based models was documented by Youd et al.
(2001). This paper focuses on the evaluation of liq-
uefactionpotential usingpiezoconepenetrationtests
(CPTU), and thus, only the cone penetration-based
evaluationprocedureisdiscussedhereinafter.
Thereisanincreasingneedtoevaluatetheliquefac-
tionpotential interms of theprobability of liquefac-
tion. Earlier modelsfor theprobabilityof liquefaction
are empirical models established through logistic
regressionanalysisof historical dataorthroughBayes
theorem(for example, J uang et al. 2003). Recently,
Moss et al. (2006) developedasimplifiedmodel for
the probability of liquefaction based on a compre-
hensiveanalysisthat utilizesthemaximumlikelihood
principle. These probabilistic models allow for the
determinationof theprobabilityof liquefactionwith-
out evaluating parameter uncertainty on the part of
theuser. IntheMoss et al. (2006) model, themodel
uncertaintyisfullycharacterizedandclearlystatedby
the model developers. However, it is up to the user
to further incorporate the parameter uncertainty, if
present, inaseparateanalysis(for example, usingthe
MonteCarlosimulation) todeterminetheprobability
of liquefaction. Ontheother hand, J uanget al. (2006)
presentedaspreadsheet solutionthat is basedonthe
first order reliability method (FORM) that explic-
itly considers both model uncertainty and parameter
uncertainty. The user of this method is required to
379
evaluateall input variablesfor their meanvaluesand
standard deviations but the spreadsheet is a stand-
alone tool that already implements the FORM and
thus, no separate analysis on the part of the user is
needed for the determination of the probability of
liquefaction.
TheexistingCPT-basedmethods,eitherdeterminis-
ticor probabilistic, asmentionedpreviously, arequite
well established, andthereis concernof diminishing
returnfor further effort to improveupontheexisting
methods. However, the effect of fine-grained mate-
rial (fines) on liquefaction resistance is still not
well understood, particularly with respect to theliq-
uefactionresistanceof soils that areconsideredtoo
clay rich to liquefy in a CPT-based investigation
(Robertson and Wride 1998). Idriss and Boulanger
(2004 & 2006) questioned the use of soil behavior
typeindex I
c
intheRobertsonandWridemethodas
achangeproxy to theeffect of fines onliquefaction
resistance. ItshouldbenotedthatintheRobertsonand
Wrides formulationfor I
c
, thepenetrationporewater
pressure(u
2
) isnot used; for distinctioninthispaper,
their definitionof I
c
isdenotedasI
c,RW
hereinafter.To
accountfor theeffectof fines, Mossetal. (2006) sug-
gesteduseof frictionratio(R
f
=f
s
/q
t
100) inlieuof
I
c,RW
, whereasLi etal. (2007)andShuttleandCunning
(2007) suggestedusingaversionof I
c
thatincludesthe
penetrationporewater pressureintheformulation.
Based on the work of Li et al. (2007) and
ShuttleandCunning(2007), aCPTU model wasini-
tiated by Chen et al. (2008) and fully developed by
J uanget al. (2008), inwhichthepenetrationporewa-
ter pressure[andthusly, theexcessporepressureratio
(B
q
)] is included in the formulation of I
c
and CRR.
ThisCPTUmodel ismoreapplicabletoawider range
of geomaterials, including soils that wereoncecon-
sideredtooclay-richtoliquefy usingtheRobertson
andWridemethod. TheCPTU model is showntobe
comparableto theexisting CPT-based methods such
as theRobertson andWridemethod in general, and
incases that werejudgedto betoo clay-richto liq-
uefy,itisshowntoyieldimprovedresults.Therefore,
it isconsideredaworthwhileeffort todevelopaprob-
abilisticCPTU-basedmodel for liquefactionpotential
evaluation. Inthis paper, sucheffort is culminatedin
asimplifiedmappingfunction(relationship) between
theprobabilityof liquefactionandthefactor of safety
determinedwiththisCPTU model.
2 DATA SET ANDTHE DETERMINISTIC
CPTU MODEL
A databaseof casehistories, consistingof 190lique-
fiedcasesand123non-liquefiedcases, wascompiled
by Chenet al. (2008) fromfivesources (Moss et al.
2006, Kuet al. 2004, Lai et al. 2004, Brayet al. 2004,
and PEER 2007). Notably, theCPT datareported in
Moss et al. (2006) did not includeB
q
. An examina-
tionof all availableCPTU datarevealedthatfor those
caseswithI
c,RW
eliminate(theI
c
usedintheRobertson
Figure1. Casehistoriesinthedatabaseplottedonthesoil
behavioral classificationchart (after J uanget al. 2008).
andWridemethod) -2.2andfrictionratioR
f
-1.5%,
theabsolutevalues of B
q
arevery small (B
q
-0.05).
Assuming B
q
=0 for thesecases will causea maxi-
mumerrorinthecomputedI
c
of lessthan3%.Thus,the
databasefromMossetal.(2006)wasscreenedwiththe
criteriaof I
c,RW
-2.2andR
f
-1.5%, and116quali-
fied caseswereselected. Thesecaseswereassumed
tohaveB
q
=0andincludedinthedatabase. Therest
of data in the compiled database were derived from
the 1999 Chi-Chi, Taiwan Earthquake and the 1999
Kocaeli, Turkey EarthquakewhereB
q
was available.
This data set provides a balance between cleaner
sands and soils with higher I
c
. Figure 1 as a whole
characterizethesoils in theadopted dataset of case
histories.
ThedeterministicCPTU model (J uanget al. 2008)
was developedthroughlearningof thecasehistories
datawithartificial neural networks(ANN). TheANN
approachhasbeenemployedinpreviousstudieswith
satisfactoryresults(J uangetal.2000&2003; Kimand
Kim2006). InthisCPTUmodel, threederived param-
etersareusedastheinput.Thefirstisanadjustedcone
tip resistance q
t1N
defined by Idriss and Boulanger
2004; 2006):
where
atm
istheatmosphericpressure(1atm=1.013
bars=101.3kPa).
Equation1was proposedby Idriss andBoulanger
(2004; 2006) in conjunction with their CPT-based
simplified method for use in liquefaction potential
evaluation.AccordingtoJ uangetal. (2008), choiceof
380
q
t1N
asdefinedinEquation1, asopposedtoother nor-
malizationmodelssuchasOlsen(1997)andRobertson
andWride(1998), isbasedonthefactthatitproduced
thebest resultsintheANNlearningof casehistories.
ThesecondrequiredparameterfortheCPTUmodel
is I
c
, defined as follows (after J efferies and Davies
1993, J efferiesandBeen2006):
Itshouldbenotedthatthemorewell-knownsoil behav-
iortypeindexdefinedbyRobertsonandWride(1998),
denoted previously as I
c,RW
, is actually a modifica-
tion of Equation 2 (by excluding the B
q
term and
modifyingthecoefficients). AccordingtoShuttleand
Cunning (2007), the dimensionless term {Q
t
(1-
B
q
)+1}inEquation2isfundamental for theevalu-
ationof undrainedresponseduringCPTU sounding,
whichallowsfor greater differentiationbetweensilty
claysandclayeysilts.
Thefirst twoparameters, q
t1N
andI
c
, arerequired
for thecalculationof thecyclicresistanceratio(CRR)
inthisCPTUmodel.TheCRR iscomputedasfollows:
Note that Equation 3 was established through cali-
brationwiththeCPTU casehistorieswheretheCSR
(cyclic stress ratio) was computed as follows (Idriss
andBoulanger 2006):
where
v
and
v
arethetotal stress andtheeffective
stress, respectively, of thesoil of concern at agiven
depth, g is the acceleration of gravity, which is the
unitforpeakgroundsurfaceaccelerationa
max
,r
d
isthe
depth-dependentshear stressreductionfactor, MSF is
themagnitudescalingfactor, andK
istheoverburden
correctionfactor for cyclicstressratio.
SinceEquation3wascreatedonthebasisof Equa-
tions 1, 2, and4, thus, theseequations must beused
together, andcollectively, theyformtheCPTUmodel.
Thevalidity of this CPTU model has beenexamined
by Chenet al. (2008) andJ uanget al. (2008). Figure
2 shows the liquefaction boundary surface that rep-
resents this model along with casehistory data. The
three-dimensional boundarysurfaceisshownat three
different angles as anexampleto illustrateits ability
todelineateliquefiedcasesfromnon-liquefiedcases.
In this paper, this CPTU model is characterized
with rigorous probabilistic analyses, with an objec-
tiveof developinganempirical relationshipbetween
theprobability of liquefaction(P
L
) andthefactor of
safety(F
S
) that iscomputedwiththisCPTU model.
Figure2. Threedimensional boundarysurfacesatthreedif-
ferent angles withcasehistory data(after Chenet al. 2008;
J uanget al. 2008).
3 METHODOLOGY FOR DEVELOPINGP
L
F
S
RELATIONSHIP
To begin with, let us denote the database as D=
{(CSR
i
, q
t1N,i
, I
c,i
, L
i
)|i =1, . . . , 313}. For theithdata
point, L
i
isanindicator of liquefaction: L
i
=1for liq-
uefiedcase; otherwise, L
i
=0.Thesubsetsforthenon-
liquefiedcasesandtheliquefiedcasesarerepresented
byD
1
={(CSR
i
, q
t1N,i
, I
c,i
, L
i
=0)|i =1, . . . , 123} and
D
2
={(CSR
i
, q
t1N,i
, I
c,i
, L
i
=1)|i =1, . . . , 190},respec-
tively. Let usfurther denote ={CSR, q
t1N
, I
c
}.
The goal of the probabilistic analysis is to find
function g(.) so that the probability of liquefaction
P(L =1)=g(). This goal can be re-written as a
mappingfunctionh(.) sothat:
Priortopresentingthemethodology, itisnotedthatthe
boundarysurfaceshowninFigure2canbereducedto
themoretraditional, two-dimensional (2-D) bound-
ary curve if the concept of an equivalent cone tip
resistance is adopted. In this regard, the converting
381
Figure3. CRR computedwithtwodifferent but equivalent
equations(usingq
t1N
versusq
t1N
).
factor K is obtained by numerical solutions of the
CPTU model:
And with this factor, the equivalent cone tip resis-
tanceis computedas: q
t1N
=K q
t1N
. It is notedthis
equivalent tipresistanceisobtainedfor thereference
conditionof I
c
=1.51,thelowestpointonthebound-
arysurfacefor agivenq
t1N
(J uanget al. 2008). Thus,
the CRR calculated with a pair of (q
t1N
, I
c
) using
Equation 3, denoted previously as CRR(q
t1N
, I
c
), is
equal totheCRR computedwiththefollowingequa-
tion(derivedfromEquation3bysettingI
c
=1.51and
replacingq
t1N
withq
t1N
):
Todemonstratetheequivalencyof theCRR computed
fromEquation3withthosecomputedfromEquation
7, the 313 cases in the database are computed with
bothequations, andtheresultsareshowninFigure3.
Asshowninthisfigure, thisequivalencyisvalidated.
It should be noted that although use of Equations 6
and7is not necessarily any easier thanuseof Equa-
tion3, thisequivalencyapproachisneverthelessmore
convenient for subsequent maximumlikelihoodanal-
ysis andis adoptedinthis paper. Figure4shows the
boundarycurve,the2-Dcounterpartof the3-Dbound-
ary surface, alongwiththeequivalent 2-D data. The
CRR definedby this boundary curve(Equation7) is
denotedhereinasCRR(q
t1N
).
Figure4. Two-dimensional boundary curveand datapro-
jected fromthree-dimensional boundary surface shown in
Figure2.
3.1 Limit state function and maximum likelihood
estimate
Let R bethelimit statefunctionso that R>1means
liquefaction:
whereZ characterizesmodelingerrorsof thelimitstate
function. Further let thecumulativedistributionfunc-
tion(CDF) of Z be+
b,
, whereb isthemeanvalueof
Z, serving as thebias factor for theratio CSR/CRR;
=(, a) isthecoefficient of variationof Z, wherea
istheunknownparameterthatcharacterizesthe(, a)
function. The unknown parameters a and b may be
estimatedbasedonthedatabaseD.
Withthelimitstatefunctionbeingdefined, thegoal
expressedinEquation5canberefinedasfollows:
wherea
andb
arethebestestimatesfortheunknown
parameters a andb basedonthedatabaseD; andthe
functionh(F
S
) becomes:
382
Figure5. PDFsandCDFsof four different typesof distri-
butionsexamined.
Todeterminetheparametersa andb withtheprinciple
of the maximumlikelihood, the likelihood function
f [D|(a, b)] isfirst expressedas:
Then, by maximizing theterm, log{f [D|(a, b)]}, the
best estimates, (a
, b
), for theunknownparameters
a andb canbeobtained.
As will beshown later, theshapeof theassumed
probability distribution of +
b,
can affect theresults
of themaximum-likelihoodanalysis. Inthisstudy, the
following cumulative distribution functions (CDFs)
areexamined:
Lognormal:
Gaussian:
MinimumGumbel:
MaximumGumbel:
As an example, theseCDFs and their corresponding
probability density functions (PDFs) are illustrated
in Figure 5 for one set of distribution parameters,
mean=1andcoefficient of variation(COV)=30%.
Of course, this set of distributionparameters is arbi-
trarilyselectedonlyfor illustrationpurpose.
Finally, the formor model of the coefficient of
variation of Z, namely, (, a), is also found to bea
significant factor intheresultsof themaximumlike-
lihoodanalysis. Four simplemodels areexaminedin
thisstudy:
Insummary, thelikelihoodfunctionbasedonall cases
inthedatabaseismaximizedtoyieldthebestestimates
(a
,b
)of theparametersa andb.Thesebestestimates

areaffectedby theassumedtypeof distributionof Z
andtheassumedmodel for theCOV of Z. Thusly, the
problemat handsisnot onlytheneedtoestimate(a
,
b
)butalsotheneedtodeterminethemostappropriate
model totakeregardingthetwoassumptions.
4 SIMULATIONRESULTSANDDISCUSSIONS
Themaximumlikelihood analysis of thedatabaseD
yields an empirical relationship between the proba-
bility of liquefactionandtheinput parameters (q
t1N
,
CSR) asfollows:
whereCRR(q
t1N
) istheCRR computedwithEquation
7that usesq
t1N
asitsinput, asnotedpreviously. This
empirical equationisderivedwiththeassumptionthat
thevariableZ followstheMinimumGumbel distri-
butionandtheassumptionthat theCOV of Z follows
themodel designatedasM
2
(i.e., COV isalinearfunc-
tion of q
t1N
); collectively, thecombinedcondition is
referred to herein as the MinimumGumbel +M
2
model. Use of any other combination (or model) in

themaximumlikelihood analysis yields an equation
similar toEquation20but withdifferent coefficients.
Contoursof liquefactionprobabilitycanbeplotted
easily with Equation 20. Some observations can be
madewithsuchgraphs. Figure6comparestheresults
obtained for two models, Lognormal +M
1
versus
MinimumGumbel +M
1
. Inbothmodels, theCOV
of Z followsM
1
(i.e., constantCOV). However, inFig-
ure6(a), variableZ isassumedtofollowthelognormal
distribution (a right-skewed PDF), and the resulting
contoursareessentiallyequally-spaced.Ontheother
hand,inFigure6(b),variableZ isassumedtofollowthe
MinimumGumbel distribution (aleft-skewed PDF),
and the resulting contours are now unevenly spaced
(thecontour spacingis smaller inthelower right but
largerintheupperleft).Theeffectof theassumedtype
of distributionof Z isevident.
383
Figure6. Contoursof liquefactionprobabilityfortwomod-
els, Lognormal +M
1
versusMinimumGumbel +M
1
.
Figure 7. Contours of liquefaction probability for two
models, Lognormal +M
1
versusLognormal +M
3
.
Similarly, theeffect of theassumedmodel for the
COV of Z can also beobserved. Figure7 compares
the results of two models, Lognormal +M
1
ver-
sus Lognormal +M
3
. Thedifferencein theresult-
ing contours of liquefaction probability can also be
observedclearly. Therefore, the question as to which
model is the most consistent with the data and the
most appropriate to use must be addressed. To the
best knowledgeof thewriters, thisquestionhasnever
beensystematicallyaddressedinthegeotechnical data
analysis.
4.1 Modified P-P Plot as a Means to Justify the
Selected Model
Inthisregard, anewtool isdeveloped, intermsof the
modifiedProbability-ProbabilityPlot(or modifiedP-
P plot for short). Theideais similar to theordinary
P-P plot, whichisveryuseful indeterminingthePDF
typefor theactual distributionof aset of real-valued
samples. However, theordinaryP-P plot isonlygood
for real-valuedobservations. For theanalysisof lique-
factiondata, theobservationsarebinary(either0or1),
andthusly, theordinary P-P plot is not readily appli-
cable. Inthispaper theordinaryP-P plot ismodified.
Recall that in the limit state expressed in Equa-
tion9, R(Z, ) isafunctionof CSR, CRR, andZ. The
liquefactionprobabilitycanbecomputedas:
Figure8. Themodified P-P plots for two models, Mini-
mumGumbel +M
2
versusLognormal +M
1
.
Let us take as an example the choice of lognormal
PDF; if Z is indeed lognormal with mean value=b
and COV =(, a), then the liquefaction probability
will be:
Theprobabilityobtainedfromanassumedtheoretical
PDF model, suchasEquation22, isreferredtoherein
as the theoretical liquefaction probability P
T
L
. This
probabilitywill bethetrue liquefactionprobability
if thelognormal distributionassumptioniscorrect.
Ontheother hand, theempirical liquefactionprob-
ability P
E
L
for agiven valueof P
T
L
can beestimated
fromanadopteddatabase. Forinstance, if weareinter-
ested in knowing P
E
L
corresponding to P
T
L
=0.1, we
can gather all the data points whose P
T
L
values are
close to 0.1 (e.g., within 0.10.05) and determine
theratioof thenumber of actual liquefiedcases (n
L
)
over thetotal number of cases(n) withP
T
L
0.1. This
ratio, denotedby

P
E
L
, is anunbiasedestimator of the
empirical liquefactionprobabilityP
E
L
.
ThemodifiedP-P plot issimply aplot whosehor-
izontal axisisthetheoretical liquefactionprobability
P
T
L
andvertical axisistheestimatedempirical lique-
factionprobability
P
E
L
.If theassumedPDFistheexact,
or closetoexact, distribution, theresultingP
T
L
vs.

P
E
L
plot should roughly follow the 1:1 line. The modi-
fied P-P plot provides a means to visually examine
thevalidityof anassumedtheoretical PDF model.
Basedontheaboveanalysisof simulateddata, the
modifiedP-Pplotwiththeconfidencebandisseenas
aneffectivetool for rejecting anassumedPDF model
givenaset of empirical liquefactionprobability data.
ThemodifiedP-Pplotalsoprovidesavisual meansto
inspect thevalidity of amodel; thecloser theplot of
P
T
L
vs.

P
E
L
is to the1:1 line, thebetter themodel is.
Tothisend, anexampleispresentedbelowtofurther
demonstratethisapproachusingtheadopteddataset
of casehistories, rather thanthesimulateddata.
As an example, two models, the Minimum-
Gumbel +M
2
model and the Lognormal +M
1
model, are compared in terms of the modified P-P

plotusingthedatasetof 313cases. Figure8compares
384
themodifiedP-P plotsobtainedwiththetwomodels.
Theresultsshowthatbothmodelsarenotrejectedat
the95%confidencelevel, asineither case, themod-
ified P-P linefalls within theconfidenceband. It is
interestingtoobservethat bothconfidencebandsare
larger when the liquefaction probability P
T
L
is close
to0.5, andbecomesmaller whentheprobability gets
closeto either 0or 1. This is becausethenumber of
datapointsn issmaller for thebinscloseto0.5andis
larger for thebins closeto 0or 1(seethehistogram
belowtheP-Pplot).Thebinswithlargern havetighter
confidence intervals, and vice versa. Notice that a
wider confidenceinterval (band) does not mean the
assumedmodel isworse: it simplymeansthat thereis
lessevidencetoreject themodel.
AsshowninFigure8, for small liquefactionprob-
abilities (P
T
L
-0.3), the Minimum-Gumbel +M
2
model performs excellently, reflected by the obser-

vation that the P
T
L
vs.

P
E
L
line follows the 1:1 line
very well. Since the empirical liquefaction prob-
ability may be viewed as the actual (or close
to actual) probability observed from the data, the
Minimum-Gumbel +M
2
model isseenasanexcel-
lent model in the range of P
T
L
-0.3. On the other
hand, the Lognormal +M
1
model appears to be
conservative in that range. Thus, the Minimum
Gumbel +M
2
model considerably outperforms the
lognormal +M
1
model inthisrange. Inotherranges
of P
T
L
, thetwomodelsperformroughlyequally. Over-
all, theMinimumGumbel +M
2
model outperforms
thelognormal +M
1
model. Thisresult isconsistent
with thehigher AIC and BIC scores obtained previ-
ouslyfortheMinimumGumbel +M
2
model. While
not shown herein, thecomparison of theMinimum
Gumbel +M
2
model witheachof theothermodelsis
performedandsimilarconclusionisreached.Thus, the
modifiedP-Pplotprovidesasimplemeanstovisually
examinethevalidity of anassumedmodel regarding
thedistributionof Z andtheCOV model of Z.
4.2 Summary of the simulation results
Theresultsanddiscussionspresentedpreviouslyhave
establishedthat theMinimumGumbel +M
2
model
isthemost appropriatemodel for themaximumlike-
lihood analysis of the adopted database. Under the
assumption of this model, theempirical relationship
betweentheprobabilityof liquefactionandthevector
of inputparameters isexpressedpreviouslyinEqua-
tion20. SubstitutingthetermCRR(q
t1N
)/CSR withthe
factor of safety F
S
and replacing q
t1N
with K q
t1N
,
thefollowingprobability-safety factor relationshipis
obtained:
Equation23providesameanstoestimatetheliquefac-
tion probability P
L
based on thecalculated factor of
Figure9. Probabilisticanddeterministicboundarycurves.
safetyF
S
. Itshouldbenotedthatunlikethesimplified
Bayesianmodel by J uanget al. (2002), inwhichthe
probabilityP
L
isafunctionof onlyF
S
, theprobability
P
L
isseenhereasafunctionof both F
S
andtheequiv-
alentconetipresistanceq
t1N
(or K q
t1N
).Thecurrent
result appears to be more consistent with the well-
recognizedobservationthattheuncertaintyindifferent
partsof theliquefactionboundarycurve, correspond-
ing to different q
t1N
or CSR levels, is different; for
thosepartsof theboundarycurve, inwhichthecurve
iswell constrainedbyliquefiedandnon-liquefieddata
points, thereis less uncertainty regarding theloca-
tion of thecurve. Thus, for thesameF
S
value, the
probabilityP
L
isexpectedtobedifferentatadifferent
level of q
t1N
.
It is noted that the developed probability model
(either Equation 20 or 23) can be used to construct
thetraditional probabilitycurves(contours), asshown
inFigure9. For comparison, thedeterministic CPTU
model byJ uanget al. (2008) isalsoshowninthisfig-
ure. As shown in Figure 9, the deterministic CPTU
boundary curve is seen to correspond to probabili-
ties in the range of 0.1 to 0.3, which indicates that
thisdeterministicmodel isgenerallyconservativeand
provides a safety level consistent with the gener-
ally acceptablerisk for designof ordinary structures
against liquefaction(J uanget al. 2002).
5 CONCLUSIONS
ThemodifiedP-P plot areshowntobeeffectivetools
for selectingthebest performingmodel inthemaxi-
mumlikelihoodanalysis. Fortheliquefactiondatabase
analyzed, thecombineduseof theMinimumGumbel
model (as the cumulative distribution function) and
theM
2
model (astheCOV model of variableZ) yields
thebestresults.Theresultingmodel (Equation23), an
empirical relationshipthat relatesthefactor of safety
385
(F
S
) totheprobabilityof liquefaction(P
L
), referredto
as theproposedmodel inthis paper, provides asim-
plemeans for evaluatingliquefactionpotential using
piezoconepenetrationtest (CPTU).
The well-recognized observation that the uncer-
taintylevel indifferentpartsof theliquefactionbound-
arycurveisdifferentisconfirmedbytheresultsof this
study. First, different partsof thedeterministicCPTU
boundary curve(implyingthat F
S
=1) arecharacter-
izedwithdifferentprobabilities(Figure9). Sensitivity
analysis within the maximumlikelihood framework
shows that the resulting model for the liquefaction
probability is stable with respect to the adopted
database. The sensitivity analysis also indicates that
theeffect of sampledisparity is quitenegligibleand
nospecial effortonthepartof theuserof theproposed
model inneeded. Overall, thestandarddeviationof the
liquefactionprobability computedwiththeproposed
model isapproximatelyequal to
PL
=0.035.
Theliquefactionprobabilitycomputedwiththepro-
posed model is a nominal probability. The nominal
probabilityimpliesthattheinputsoil (CPTU)andseis-
mic parametershavebeenproperly evaluatedandthe
extent of variation (or uncertainty) in these param-
eters is deemed acceptable. If the uncertainties in
theinput parameters aretoolargetoignore, separate
analysisusingtheMonteCarlosimulationmethodor
thereliability methodthat incorporates explicitly the
parameter uncertaintiesshouldbecarriedout.
ACKNOWLEDGMENTS
The study on which this paper is based was sup-
portedinpart bytheU.S. Geological Survey(USGS),
Departmentof theInteriorunderUSGSawardnumber
07HQGR0053. Theviewsandconclusionscontained
in this paper are those of the writers and therefore
shouldnot beinterpretedas necessarily representing
theofficial policies, expressedor implied, of theU.S.
Government.
REFERENCES
Akaike, H. 1974.A newlookatthestatistical model identifi-
cation. IEEE Transactions on Automatic Control, AC-19:
716723.
Bray, J.D., Sancio, R.B., Durgunolu,T., Onalp,A.,Youd,T.L.,
Stewart, J.P., Seed, R.B., Cetin, O.K., Bol, E., Baturay,
M.B., Christensen, C., andKaradayilar, T. 2004. Subsur-
facecharacterizationat groundfailuresitesinAdapazari,
Turkey. Journal of Geotechnical and Geoenvironmental
Engineering, ASCE, 130: 673685.
Chen, C.H., J uang, C.H., and Schuster, M.J. 2008. Empir-
ical model for liquefaction resistance of soils based
on artificial neural network learning of case histories.
Geotechnical Special Publication, ASCE, Reston, VA,
179: 854861.
Cover,T.M. andThomas, J.A. 1991. Elements of Information
Theory. Wiley, NewYork.
Idriss, I.M. and Boulanger, R.W. 2004. Semi-empirical
procedures for evaluating liquefaction potential during
earthquakes. Proceedings, the 3rd International Conf.
on Earthquake Geotechnical Engineering (ICEGE),
Berkeley, CA. 3256.
Idriss, I.M. andBoulanger, R.W. 2006. Semi-empirical pro-
ceduresforevaluatingliquefactionpotential duringearth-
quakes. Soil Dynamics and Earthquake Engineering. 26:
115130.
J efferies, M.G. andDavies, M.P. 1993. Useof CPTutoEsti-
mateEquivalent SPT N
60
. Geotechnical Testing Journal,
AmericanSocietyforTestingandMaterials, 16: 458468.
J uang, C. H., C.J. Chen,W.H.Tang, andD.V. Rosowsky2000.
CPT-basedliquefactionanalysis, Part1: Determinationof
limit statefunction. Geotechnique. 50: 583592.
J uang, C.H., J iang, T., and Andrus, R.D. 2002. Assessing
probability-based methods for liquefaction evaluation.
J. of Geotech. and Geoenvir. Engrg.,ASCE,128:580589.
J uang, C.H., Yuan, H., Lee, D.H., andLin, P.S. 2003. Sim-
plified CPT-based method for evaluating liquefaction
potential of soils. J. of Geotech. and Geoenvir. Engrg.,
ASCE, 129: 6680.
J uang, C.H., Fang, S.Y., and Khor, E.H. 2006. First order
reliabilitymethodforprobabilisticliquefactiontriggering
analysisusingCPT. J. of Geotech. and Geoenvir. Engrg.,
132: 337350.
Kim, Y.S. andKim, B.T. 2006. Useof artificial neural net-
worksinthepredictionof liquefactionresistanceof sands.
J. of Geotech. and Geoenvir. Engrg., 132: 15021504.
Ku, C.S., Lee. D.H., andWu J.H. 2004. Evaluation of soil
liquefaction in the Chi-Chi, Taiwan earthquake using
CPT. Soil Dynamics and Earthquake Engineering, 24:
659673.
Lai, S.Y., Hsu, S.C., and Hsieh, M.J. 2004. Discriminant
model forevaluatingsoil liquefactionpotential usingcone
penetrationtestdata. J. of Geotech. and Geoenvir. Engrg.,
ASCE, 130: 12711282.
Li, D.K., J uang, C.H., Andrus, R.D., andCamp, W.M. 2007.
Indexproperties-basedcriteriafor liequefactionsuscepti-
bilityof clayeysoils: acritical assessment. J. of Geotech.
and Geoenvir. Engrg., ASCE, 133: 110115.
Moss, R.E.S., Seed, R.B., Kayen, R.E., Stewart, J.P., Der
Kiureghian, A., and Cetin, K.O. 2006. CPT-based prob-
abilistic and deterministic assessment of in situ seismic
soil liquefaction potential. Journal of Geotechnical and
Geoenvironmental Engineering, ASCE, 132: 10321051.
PEER web site 2007. Soils and liquefaction data on 1999
Kocaeli, Turkey Earthquake, http://peer.berkeley.edu/
publications/turkey/adapazari/index.html.
Seed, H.B. and Idriss, I.M. 1971. Simplified procedure
for evaluating soil liquefaction potential. Journal of the
Soil Mechanics and Foundation Div., ASCE, 97(SM9),
12491273.
Schwarz, G. 1978. Estimating the dimension of a model.
Annals of Statistics, 6, 461464.
Shuttle, D.A. and Cunning, J. 2007. Liquefaction potential
of siltsfromCPTu. Canadian Geotechnical Journal, 44:
119.
Youd,T. L., Idriss, I. M.,Andrus, R.D.,Arango, I., Castro, G.,
Christian, J.T., Dobry, R., LiamFinn, W.D., Harder, L.F.,
J r., Hynes, M.E., Ishihara, K., Koester, J.P., Laio, S.S.C.,
Marcuson, W.F., III, Martin, G.R., Mitchell, J.K.,
Moriwaki, Y., Power, M.S., Robertson, P.K., Seed, R.B.,
Stokoe, K.H., II. 2001. Liquefaction resistanceof soils:
summary report from the 1996 NCEER and 1998
NCEER/NSF workshops on evaluation of liquefaction
resistanceof soils. Journal of Geotechnical and Geoenvi-
ronmental Engineering, 127: 817833.
386
Suggestionsfor implementinggeotechnical riskmanagement
M.Th. vanStaveren
Deltares & Delft University of Technology, Delft, The Netherlands
ABSTRACT: Without doubt, geotechnical engineeringisakeysuccessfactor for most constructionprojects.
Currently, riskmanagementgetsmoreandmoreattentionintheseprojects. Nevertheless, withgeotechnical risk
analysis becomingcommonpractice, theroutineapplicationorimplementationof geotechnical riskmanagement
still is anunexploredarea. After numerous debates over thelast years, about why applying geotechnical risk
management inconstructionprojects, anewmajor questionemergeswithinthegeotechnical community: How
implementing geotechnical riskmanagement effectively, efficiently, andpersistentlyinconstructionprojects?
TheDutchDelft Cluster ResearchProgrammeImplementing Risk Management investigatesthehurdlesand
conditionsforsuccessfullyimplementingriskmanagementinorganizations.First,bytheory,dataandinvestigator
triangulation, thesehurdlesandconditionshavebeenidentified, analyzed, clusteredandsynthesizedintoseven
keyhurdlesandtenkeyconditionsfor effectivelyapplying geotechnical riskmanagement. Theresult triggered
another researchquestion: Arethesekeyhurdlesandkeyconditionsalsoappropriatefor actuallyimplementing
geotechnical riskmanagement withinorganizations?
Therefore, thenextresearchphaseconsistedof identifying, selecting, andcombininginnovationmanagement
variablesandtheorieswiththoseof riskmanagement. Thekeyhurdlesandkeyconditionsfor geotechnical risk
management havebeensynthesizedandclassifiedintotwomainimplementationdimensions: (1) geotechnical
riskmanagementmethodologiesand(2) organizationsroutinelyusinggeotechnical riskmanagement.Thispaper
presents thekey conditions for implementinggeotechnical risk management inorganizations inaconceptual
model, together withtwomainimplementationsuggestions.
Inconclusion, theresearchresults demonstratethat theorganizational dimension, whichis usually entirely
neglected, isof key importancefor realizingtheroutineapplicationof geotechnical risk management method-
ologies throughout all of the phases of construction projects. The synthesis of innovation management and
risk management conceptsandvariablesincreasesthechancefor moreeffective, efficient, andpersistent orga-
nizational implementation of geotechnical risk management in construction projects. This may dramatically
strengthenthegeotechnical rolein,andcontributionto,thesuccessful completionof engineeringandconstruction
projectswithinour societies.
1 INTRODUCTION
Thegrowinginterestforriskmanagementinconstruc-
tionprojectsisparamount. Forexample, onNovember
1, 2007, leading Dutch organizations signed a joint
agreement for rigorously implementingrisk manage-
ment within the construction industry. By the year
2012, three ministries, the four largest cities, and
thenational organizationsof contractorsandconsult-
ing engineers aim fully applying risk management
in eighty percent of theprojects in theNetherlands.
Expected benefits are less failure costs, less time
delays, andareductionof thenumber of disputes, by
buildingtrust, increasingtransparency, andimproving
communicationbetweenall constructionproject par-
ties. Therefore, it is promisingthat over theyears, at
leastinthelargeandcomplexprojects, theapplication
of geotechnical riskanalysisseemsbecomingcommon
practice.
However, contrary toother papers, thispaper does
not introduce any new or updated methodology for
applying geotechnical risk analysis or geotechnical
riskmanagement.Whilestill ratherlimited,theamount
of literature about geotechnical risk management is
growing. For example, for generic geotechnical risk
management methodologies reference is made to
Clayton(2001) andVanStaveren(2006).
Moreover, thereis anincreasingamount of litera-
turecovering specific geotechnical risk analysis and
management topics. Examples of these are the dif-
ference between unsafe geotechnical certainty and
safegeotechnical uncertainty(Barends,2005),therole
of the human factor in achieving geotechnical reli-
ability (Bea, 2006), objectiveandsubjectiveways of
geotechnical riskclassification(Altabbaet al., 2004),
andcontractual allocationof geotechnical risk(Essex,
2007).
Unlike most other papers, this paper focuses on
implementing existinggeotechnical risk management
methodologies in organizations. Implementation is
here simply defined as the routinized application
of risk management during the entire design and
387
construction process. Implementation thus differs
highly from often incidentally application of risk
analyseswithinconstructionprojects.
For instance, Halman(2008) addressestheimport-
anceof implementing risk management intheDutch
constructionindustry.Theneedforparticularlyimple-
menting geotechnical risk management in organiza-
tions in the construction industry has been raised
by Smith (2008). A workshop of the US GeoCoun-
cil, in December 2006 with a group of fifty geo-
professionals, revealed that currently themain areas
of attentionintheconstructionindustryareinnovative
contracting, safety, costanalyses, andresearch, devel-
opment, andtraining. Attentiontothesetrendsshould
contribute to providing better, faster, and cheaper
solutions to geotechnical problems in construction
projects. Geotechnical risk management was con-
sideredasthebest chancefor meetingthesedemands
ineach of thetrendareas(Smith, 2008).
Obviously, geotechnical risk management should
be routinely applied, and thus be well implemented
withinorganizations, for materializingbenefits. How-
ever, theauthors experienceteaches that even when
geotechnical professionalsandtheirmanagerssay that
they are applying geotechnical risk management in
engineeringandconstructionprojects, oftenthey are
notactuallydoing itinanexplicitandwell-structured
way. Moreover, evenif they do it inthat explicit and
well structuredway, theyoftenexecutemoreof arisk
analysis, rather than executing thefull risk manage-
ment cyclewithineachandeveryproject phase.
By conventional hit and run risk management
of doing one or two analyses, the potentially large
benefitsof routinelyapplyinggeotechnical risk man-
agement remainhidden. Thisresultsinmissedoppor-
tunities, for instance saving lives of construction
workers by reducing unsafety, increasing profits by
reducingfailurecosts, andspeedinguptheconstruc-
tion process by reducing delays. Similar to quality
management (Imai, 1989), a cyclic approach with
continuous attention to improving little things is
required for effectivegeotechnical risk management
(VanStaveren, 2006). This requires full implementa-
tionwithin(project) organizations.
Therefore, after many debates over thelast years,
about why to apply geotechnical risk management
in construction projects, now a new type of ques-
tionemergeswithinthegeotechnical community: how
to implement geotechnical risk management effec-
tively in construction projects? This how-question
seemsevenmoredifficulttoanswer thantheprevious
why-question. For instance, howto relatediscipline-
based geotechnical risk management to project risk
management inconstructionprojects?
Therefore, thispaper addressesayet highlyunder-
estimated topic: How to realize a routine use of
geotechnical riskmanagement duringplanning, engi-
neering, andconstructionof all sortsof buildingsand
infrastructureprojects?
Todate, thereappearedtobenoliteraturecovering
thistopic, despiteitsutmost relevance. Anyscientific
research and resulting practical guidance about how
to implement risk management in general is very
scarce. Concerninggeotechnical risk management in
particular, research and guidance is entirely lack-
ing. Therefore, thereseems to emergeafreemarket
paradox of high knowledgedemand with no knowl-
edgesupply. Theimplementationof geotechnical risk
management isstill anunexploredareaof research.
The practical research project Implementing Risk
Management of the Dutch Delft Cluster Research
Programme aims to answer the question of how to
implement risk management in organizations in the
construction industry. This research project is per-
formedbyinvolvingresearchersof theunit GeoEngi-
neering of Deltares (formerly known as the Dutch
National Institute for GeoEngineering, GeoDelft),
the unit Innovation and Environment of TNO, the
Technology, Policy, and Management faculty of the
Delft University of Technology, andtheConstruction
Management and Engineering research group of the
University of Twente, Netherlands. Theresearchwill
becompletedby theendof theyear 2009. However,
recent researchresultsfor successfully implementing
geotechnical risk management in organizations are
readily available to be shared with the international
geotechnical community.
One of the innovative research approaches of the
Implementing Risk Management projectisconsidering
the implementation of risk management in organi-
zations a sort of innovation. If new to (part of)
an organization, fostering the routine application of
geotechnical riskmanagement in(part of) theorgani-
zationprovedtohavealotincommonwithimplement-
inginnovationsinorganizations, suchasgeotechnical
quality systems or softwarefor geotechnical design.
These organizations are either a temporary project
organizations for realizing aconstruction project, or
well-establishedfirms.
After this necessarily comprehensiveintroduction,
this paper continues with presenting the research
approach. Then, theresearchresults about risk man-
agement, innovation management and their synthe-
sis for implementing geotechnical risk management
are presented. This generates the suggestions for
implementinggeotechnical riskmanagement, thevery
core of this paper. It ends with the main research
conclusions.
2 RESEARCHAPPROACH
2.1 Introduction
Thischapterbrieflypresentstheresearchapproach for
investigatingtheimplementationof geotechnical risk
management in(project) organizations. Theseorgani-
zationsinvolvepeople,whoworktogetherforrealizing
commongoals. Intheconstructionindustry, thecom-
mon goal is usually realizing projects according to
pre-set quality specifications, andwithinbudget and
planning.
388
Inorganizations ingeneral, andparticularly when
dealingwithrisk, theso-calledhuman factor or peo-
ple factor plays a dominant role (Bea, 2006, Van
Staveren, 2006). Therefore, regarding the nature of
reality(ontology), ahermeneuticworldviewhasbeen
chosenfor thisresearch. Thisconsiderstheworldasa
social construct, withitsinherent subjectivities.
The epistemological point of view concerns
assumptions about the nature of knowledge about
reality. Thedesignscienceparadigmwithapractical
research approach (Van Aken, 2004) has been pur-
posefully selected, for generating solutionoriented
knowledge about implementing risk management in
organizations. Together, the ontological and episte-
mological positions provided the scientific research
framework, whichsynthesizedgeotechnical andorga-
nizational aspects.Thisframeworkconsistedof subse-
quently exploratory andsynthesizingresearchof risk
management and of innovation management. In the
following sections, the four resulting research steps
aredescribed.
2.2 Exploratory risk management research
Theexploratory risk management researchconsisted
of literature surveys and field research. Both aimed
identifyingtherelevantriskmanagementconceptsand
variables. Extensiveliteratureresearchhas beenper-
formedbyusingVanStaveren(2006) andperforming
anadditional survey, whichisreportedinVanStaveren
(2007).
Field research involved in-depth interviewing of
four academic geotechnical and mining experts of
leading universities in the US (Massachusetts Insti-
tute of Technology, University of California, Berke-
ley), theUK (Universityof Southampton), andSouth
Africa(Universityof theWitwatersrand). Inaddition,
three geotechnical and mining consultants fromthe
UK and South Africa were interviewed. Moreover,
Dutch experiences with applying geotechnical risk
management wereretrieved fromaprevious and the
actual Delft Cluster researchproject, as well as from
RISNET. Thelatter istheDutchjointknowledgeplat-
formforapplyingriskmanagementintheconstruction
industry.
2.3 Synthesizing risk management research
Thesynthesizingresearchpart includedanalysis and
classificationof theidentifiedrisk management con-
ceptsandvariables. Provenresearchtactics, including
data and investigator triangulation wereapplied. All
variables were classified into either hurdles, which
obstruct theapplicationof geotechnical risk manage-
ment, or conditions that are required for applying
geotechnical riskmanagement.
An in-depth analysis generated 7 key hurdles and
10 key conditions for applying geotechnical risk
management, which were considered the most rele-
vant variables. This researchresult triggeredanother
research question: Are these key hurdles and key
conditionsappropriateforactuallyimplementing (rou-
tinely applying) geotechnical riskmanagementwithin
organizations?
2.4 Exploratory innovation management research
Theexploratoryinnovationmanagementresearchalso
consisted of literature surveys and field research
for identifying relevant concepts and variables. An
extensiveliteraturesurveyhasbeenperformed, which
includedPh.D. theses, scientifictopjournalsandaddi-
tional literature about innovation management. The
focus was on implementing innovations in organiza-
tions. Fieldresearchincludedin-depthinterviewingof
sevenDutchexperts inimplementinginnovations by
realizingplannedorganizational change. All but one
arewell-knownDutchprofessorsfromuniversitiesof
Amsterdam, Rotterdam, Eindhoven, Groningen, and
Twente, who also performtop management consul-
tancy. Theoneremainingexpert isaprofessional risk
manager involvedinimplementingrisk management
inpublicorganizations.
2.5 Synthesizing innovation management research
Thesynthesizingresearchpart includedanalysis and
classificationof theidentifiedinnovationmanagement
concepts andvariables. Provenresearchtactics, such
asdata, andinvestigator triangulationwereapplied.
In total 55 hurdles and 93 conditions for imple-
mentinginnovationsinorganizationswereidentified.
These variables were compared with those fromthe
risk management research part However, becauseof
the maximumlength of this paper and the focus on
geotechnical risk management, this comparison and
itsconclusionscouldnot bepresentedhere.
Nevertheless,fromthesynthesizingresearchpartof
innovationmanagementparticularlytheresultingcon-
cepts for implementing innovations in organizations
areusedintheremainingpart of thispaper.
3 GEOTECHNICAL RISK MANAGEMENT
This chapter presents the results of the exploratory
and synthesizing research of geotechnical risk man-
agement concepts andvariables.
3.1 Risk management concepts
Analyzing the identified risk management concepts
revealedthreeinterrelatedlevelsforimplementingrisk
management: (1) thediscipline level, (2) theproject
level, and(3) theorganizational level. Figure1sym-
bolizes these three levels as a mountainous area, of
which the risk mountains have steep and slippery
slopes.
Geotechnical risk management represents thedis-
ciplinelevel. Whenreachingthetopof geotechnical
risk management, indicating routinely applied risk
389
Figure1. Threeriskmanagement mountains.
management, thereraisesanother andhigher topthat
represents theproject risk management mountain. If
that top has been reached as well, indicating well-
embedded geotechnical risk management in project
risk management, another top is still there. This lat-
ter topisrepresentingtheorganizational level of risk
management. This level involves managing risks of
entireprojectportfoliosof afirm. For example, acon-
tractor havingtenprojects under constructionshould
compensateoneveryriskyprojectwiththeremaining
nineand less risky projects. This would avoid going
bankrupt, whenall riskswithintheriskyprojectoccur.
In summary, for reasons of acceptability, as well
as for effectiveness, efficiency, and persistence over
time, geotechnical risk management should bewell-
embedded in project risk management. Preferably,
it should furthermore be related to portfolio risk
management. Obviously, realizing this challenge is
more of a management responsibility than that of
a geotechnical engineer. However, the latter engi-
neer may substantially contributetobothproject and
portfolioriskmanagement, byadequatelyperforming
geotechnical risk management duringall engineering
andconstructionproject phases.
3.2 Risk management variables
Asmentionedbefore,all identifiedvariablesforapply-
ing geotechnical risk management in organizations
were classified into either hurdles, obstructing the
applicationof geotechnical risk management, or con-
ditions thatarerequiredforapplyinggeotechnical risk
management.
Fromthe literature survey and field research, in
total 109hurdlesand147conditionsfor successfully
Table1. Numbers of hurdles and conditions for applying
geotechnical riskmanagement, fromseveral datasources.
Hurdles Conditions
Datasource no. no.
VanStaveren(2006) 5 10
VanStaveren(2007) 17 26
Interviewswith7experts 63 73
Delft Cluster andRISNET 24 38
Total numbers 109 147
Table 2. Key hurdles for applying geotechnical risk
management.
No. Category Description
1. Motivation Lackof geotechnical riskmanagement
awareness.
2. Motivation Lackof geotechnical riskmanagement
benefits.
3. Motivation Fear for geotechnical risktransparency.
4. Motivation Difficultyof applyinggeotechnical risk
management.
5. Training Lackof geotechnical riskmanagement
understanding.
6. Tools Lackof geotechnical riskmanagement
methods, protocols, software, guidelines.
7. Tools Lackof geotechnical riskmanagement
benchmarks.
applying geotechnical risk management wereidenti-
fied. Table1 shows thedistribution of thesehurdles
andconditionsover thedifferent datasources.
Despitesomeoverlap of anumber of factors, the
high numbers in Table 1 demonstrate the enormous
complexity of applying geotechnical risk manage-
ment. This raised the following research question:
Which of the unworkable large series of hurdles
and conditions are the most significant hurdles and
conditions?
For answering this question, these hurdles and
conditions have been clustered and synthesized into
seven key hurdles and ten key conditions for effec-
tively applying geotechnical risk management. Three
purposeful selectedmaincategories weremotivation
of engineers for applying geotechnical risk manage-
ment, training required for learning how to operate
geotechnical risk management methodologies, and
tools for facilitatingtheexecutionof geotechnical risk
management.
3.3 Hurdles for geotechnical risk management
Table 2 presents the seven key hurdles or obstruc-
tions that resulted fromthe data analysis, including
thecategory.
Remarkably, four out of the seven key hurdles
aremotivational. Lack of geotechnical risk manage-
ment awareness, the benefits of it, as well as fear
390
for geotechnical risk transparency and difficulty of
applying geotechnical risk management are hurdles
at the level of the individual geotechnical engineer.
Presence of these hurdles will highly restrict his or
her motivation for routinely applying geotechnical
risk management in his or her day-to-day activities.
VanStaveren(2006) present sixsuggestionsfor over-
comingtheseindividual hurdles, includingdeveloping
risk awarenessandtakingsufficient timefor actually
applyingriskmanagement.
Of theremainingthreekeyhurdles, oneistraining-
relatedandtwoconcerntheroleof toolsfor applying
geotechnical riskmanagement.Somesortof education
andtrainingisrequiredforoperatingriskmanagement
tools, suchasriskdatabases. Remarkably, theseventh
keyhurdleexplicitlyaddressesthelack of geotechni-
cal risk benchmarks. This means that alack of clear
levels of acceptablegeotechnical risk, such as max-
imumallowed differential settlements, is also a key
hurdle for applying risk management. This key hur-
dle has been allocated to the tools category, which
includes for example geotechnical design software.
Suchsoftwaremayberequiredforsettinggeotechnical
riskmanagement benchmarks.
3.4 Conditions for geotechnical risk management
Similarly, Table 3 presents the ten key conditions,
whichresultedfromthedataanalysis. Table3present
also thecategory of each key condition. In total, six
outof thetenkeyconditionsareof amotivational type.
Similar to the hurdles, motivational key conditions
dominate.
Obviously, it should be clear to any individual
geotechnical engineer why routinely applying risk
management. Therefore, clear objectives and goals
should be defined, before starting any activities.
Preferably, thesegoalsaremeasurable. Closelyrelated
tothefirst key condition, thereshouldbeanindivid-
ual awareness of the consequences of geotechnical
risk. What are the effects to which parties when a
geotechnical risk occurs? This awareness may raise
thedesireto avoid therisk to occur, and thus grows
riskmanagement motivation.
It may help when geotechnical risk management
responsibilities are clearly defined. A geotechnical
baseline report (GBR) may be useful for allocating
therisk of differingsiteconditions (Essex, 200, Van
Staveren, 2006). By relating, andpreferably incorpo-
rating, geotechnical risk management within project
risk management, economies of efficiency may be
realizedthatcontributestothemotivationof engineers
toapplygeotechnical riskmanagement.
Involving other project stakeholders is also a
motivational factor. Particularly, clients requesting
geotechnical risk management may be helpful for
increasing themotivation of geotechnical engineers.
The last motivational key condition concerning
resources seems obvious. In most largely money-
driven firms in the construction industry, which put
also large time pressure on their projects, resources
Table 3. Key conditions for applying geotechnical risk
management.
No. Category Description
1. Motivation Clear objectivesandgoalsfor applying
geotechnical riskmanagement.
2. Motivation Awarenessof geotechnical risk
consequences.
3. Motivation Contractuallyarrangedresponsibilities
for geotechnical riskanditsallocation.
4. Motivation Clear relationshipof geotechnical risk
management andproject risk
management.
5. Motivation Involvement of all project
stakeholdersinapplyinggeotechnical
riskmanagement.
6. Motivation Availabilityof resources(budget, time)
for applyinggeotechnical risk
management.
7. Training Understandingof geotechnical risk
management bygeotechnical engineers.
8. Training Understandingof riskmanagement in
teamsbygeotechnical professionals.
9. Training Understandingof riskmanagement
andculturebygeotechnical managers.
10. Tools Fit of geotechnical riskmanagement
methodologieswiththeproject
objectives.
suchasbudgetandtimeshouldbemadeavailabletothe
geotechnical engineerswhoshouldapplygeotechnical
risk management. If risk management is effectively
applied, thereturnof investment may beas highas a
factor ten, or more(Smith, 1996, Sperry, 1998).
The training-related key conditions address the
need for understanding geotechnical risk manage-
ment, supplementedbyunderstandingtheapplication
of geotechnical risk management inteams. Thelatter
is of importancefor dealingwiththeinherent differ-
ences insubjectiverisk perception, evenwhenbased
onthesamefactual information, suchas conepene-
trationtest results or results of laboratory index tests
(VanStaveren, 2006). Moreover, particularlygeotech-
nical managers, who are responsible for the use of
geotechnical riskmanagementmethodologiesbytheir
appointedgeotechnical engineers, shouldunderstand
thedominantroleof organizational cultureinroutinely
applyinggeotechnical riskmanagement (or not).
Finally, the selected risk management tools for
applying geotechnical risk management should fit
with the targeted users of those tools (geotechnical
engineers), as well as with the complexity and risk
profileof theproject. Rather sophisticatedtools, such
as MonteCarlo typeof software, may berequiredin
complex projects, while just performing some sen-
sitivity analyses with already available geotechnical
software may be sufficient within the less complex
andsmaller projects. Obviously, thereisnoonerecipe
for whichtoolstoselect. Thisentirelydependsonthe
typeof project, expectedgroundconditions, therisk
propensity of the client and the involved engineers,
andsoon.
391
Table4. Numbers of hurdles and conditions for applying
geotechnical riskmanagement, fromseveral datasources.
Hurdles Conditions
Datasource no. no.
Ph.D. theses 4 14
Scientifictopjournals 6 8
Additional literature 10 22
Interviewswithsevenexperts 35 49
Total numbers 55 93
4 INNOVATIONMANAGEMENT
Thischapterpresentstheresultsof theexploratoryand
synthesizingresearchof innovationmanagementcon-
cepts andvariables.Themainobjectiveof thisresearch
part was providing additional scientific evidencefor
therelevanceof thekeyhurdlesandkeyconditionsfor
implementinggeotechnical riskmanagement inorga-
nizations.Aninnovationmanagementperspectivewas
considereduseful, becauseof theassumedsimilarities
betweenimplementinginnovationsandimplementing
newriskmanagementmethodologiesinorganizations.
4.1 Innovation management concepts
By analyzing the identified innovation management
concepts, the innovation diffusion model by Rogers
(2003) wasconsideredthemostcompleteandrelevant
model. This has been confirmed by comparing this
model withthefewother availablemodels, including
thoseby Klein& Sorra(1996) andLewis & Seibold
(1993). A number of factorsof thedirect andindirect
network externalities adoption model (Song, 2006)
wereaddedtothemodel byRogers(2003). Combining
thesemodelsgeneratedtwomaindimensionsfor hur-
dlesandconditionsfor implementinginnovations: (1)
thoserelatedto theinnovation, and(2) thoserelated
totheorganization, inwhichtheinnovationhastobe
routinelyusedbyitsimplementation.
4.2 Innovation management variables
Similar to theidentified risk management variables,
thevariablesfor implementinginnovationsinorgani-
zationswereclassifiedintoeitherhurdles, obstructing
the routine application of innovations, or conditions
that arerequiredfor implementinginnovations.
Fromthe literature survey and field research, in
total 55 hurdles and 93 conditions for successfully
implementing innovations were identified. Table 4
showsthedistributionof thesehurdlesandconditions
over thedifferent datasources.
Despitesomeoverlapof anumberof factors,similar
toTable1concerningriskmanagement, thehighnum-
bersinTable4demonstratetheenormouscomplexity
of implementing innovations in organizations. This
raisedthefollowingresearchquestion: Whicharethe
maincharacteristicsof thetwopreviously mentioned
Table5. Maininnovationcharacteristics.
Innovation Acknowledgedin
No. characteristics riskmanagement
1. Relativeadvantage yes
2. Compatibility no
3. Complexity yes
4. Triability yes
5. Observability no
6. Indirect networkeffect no
7. Relativeusefulness no
main dimensions for implementing innovations: (1)
theinnovation itself,and(2)theorganization,inwhich
theinnovationwill beimplemented?
For answeringthis question, therisk management
keyhurdlesandkeyconditions(seeTable2andTable
3), as well as all identified hurdles and conditions
for implementing innovations (seeTable 4 for their
numbers) havebeenclassifiedaccordingtothemain
characteristics of innovations and organizations, as
derivedfromcombiningthemodelsof Rogers(2003)
andSong(2006). Next, theresultsof thisexerciseare
presented.
4.3 Innovation management characteristics
Byattributingthe55innovationhurdlesand93inno-
vationconditions tothemaincharacteristics of inno-
vations, it became clear that there are seven main
characteristicsof innovations.However,attributingthe
7keyhurdlesandthe10keyconditionstotheseinnova-
tioncharacteristicsrevealedonlythreeoutof theseven
main characteristics of innovations. In other words,
by considering a risk management perspective only,
the majority of relevant characteristics of risk man-
agement methodologies, 4out of 7characteristics or
57percent, wouldremainhidden.
As Table 5 indicates by the word no in the right
column from an innovation perspective, these hid-
denriskmanagementcharacteristicsarecompatibility,
observability, indirectnetworkeffect, andrelativeuse-
fulness.Therefore, for successfullyimplementingrisk
management methodologies, also these four charac-
teristicsneedtobeaddressed, inadditiontothethree
characteristics that are indicated by yes in the right
columnof Table5. Thelatter originateformsolely a
riskmanagement perspective.
A similar exercisehasbeenperformedfor themain
dimension of the organization in which risk man-
agement has to be implemented. Table 6 shows the
results.
When attributing the 7 key hurdles and the 10
key conditions to the main innovation characteris-
tics, onlytwoof thefour maincharacteristicsof were
acknowledged. Inother words, by consideringarisk
management perspectiveonly, fiftypercent of therel-
evant characteristics of organizations in which risk
392
Table6. Mainorganizationcharacteristics.
Organization Acknowledgedin
No. characteristics riskmanagement
1. Structureof theorganization yes
2. Normswithintheorganization no
3. Organizational implementation no
decision
4. Organizational innovation yes
consequences
Figure2. Conceptual model forimplementinggeotechnical
riskmanagement in(project) organizations.
management methodologies haveto beimplemented
wouldbeentirelyneglected.
As Table 6 indicates by the word no in the right
columnfromaninnovationperspective, thesehidden
organizational characteristicsarethenormswithinthe
organizationandthetypeof decisionmakingwithin
the organization, such as cooperative decision mak-
ingby involvingrisk management users or top-down
decisionmaking. For successfully implementingrisk
management methodologies in organizations, these
latter twocharacteristicsneedalsobeingaddressed.
5 RISK ANDINNOVATIONMANAGEMENT
By using the selected innovation management con-
cepts, as introduced in theprevious chapter, thekey
variables for implementing geotechnical risk man-
agement werefurther synthesizedandclassifiedinto
the two main implementation dimensions. Figure 2
presents the resulting key conditions, which are
presented in a conceptual model for implementing
geotechnical riskmanagement withinorganizations.
Figure2aims todemonstratethat for successfully
implementinggeotechnical riskmanagement inorga-
nizations, fivekeyconditionsfortheriskmanagement
methodologiesandanother fivekeyconditionsfor the
organizationshouldbefulfilled. Moreover, geotechni-
cal riskmanagementneedstoberelatedtoprojectrisk
management.Thelatterneedstoberelatedtoportfolio
risk management, when present in the organization.
The synergies of these three levels of risk manage-
ment would providethemaximumgeotechnical risk
management benefits.
6 MAINIMPLEMENTATIONSUGGESTIONS
Summarizingthisresearchprovidestwomainsugges-
tionsfor implementinggeotechnical riskmanagement
in(project) organizationsintheconstructionindustry:
1. Incorporategeotechnical riskmanagementasmuch
as possiblein project risk management at project
level, andinportfoliorisk management at organi-
zational level;
2. Maximizethepresenceof thefivekey conditions
for implementing geotechnical risk management
methodologies, as well as that of thefiveorgani-
zational keyconditions.
Basedupon thecomprehensiveresearch undertaken,
it is expected that using the two main suggestions
considerably increasethechanceto realizeeffective,
efficient, andpersistentimplementationof geotechni-
cal risk management in(project) organizationsinthe
constructionindustry.
At themoment of writingthispaper, thesesugges-
tionsarebeingempiricallytestedinfour casestudies.
Moreover, a major Dutch public client is using the
suggestionsforreducingthegeotechnical failurecosts
withfiftypercentbytheyear2015(VanStaverenetal.,
2009). Realizing this objective will save this orga-
nization tens of millions Euro per year. Moreover, a
numberof otherDutchorganizationsarefollowingthe
suggestionsaswell.
7 CONCLUSIONS
Thispaperintendstocreateawarenessabouttheimpor-
tance of the routine application, defined implemen-
tation, of geotechnical risk management. This topic
stretches conventional geotechnical engineering, but
fromnowon, itcannotbeneglectedanymore. For tar-
getinggeotechnical engineersandtheir managers, the
mainresultsof anstill ongoingDutchresearchproject
about howto implement risk management inorgani-
zations intheconstructionindustry aresummarized.
This research revealed the enormous complexity of
implementingrisk management methodologies. Syn-
thesizing risk management and innovation manage-
ment concepts considerably structured and reduced
the many implementation variables, while fostering
their completeness.
Ingeneral conclusion, for realizingeffective, effi-
cient, and persistent geotechnical risk management
duringtheentireengineeringandconstructionprocess,
(much)moreattentionshouldbepaidtotheimplemen-
tation of geotechnical risk management methodolo-
giesin(project) organizations. Thisshouldberelated
393
toproject riskmanagement andtoportfolioriskman-
agement, for realizing maximumresults. Moreover,
thepresenceof ten risk management key conditions
should be maximized. Only then, the pursued ben-
efits of geotechnical risk management are expected
to occur. This will give geotechnical engineering,
andparticularlyitsengineers, thecreditsandrewards
they deserve, given the enormous impact of ground
conditions on construction projects. It is up to us,
thefrontrunnersof geotechnical risk management, to
realizethisorganizational-geotechnical challenge.
ACKNOWLEDGEMENTS
The author would like to thank Prof. Bob Bea,
Prof. Herbert Einstein, Prof. Dick Stacey, Prof. Chris
Clayton, Mr. Tim Chapman, Dr. J an Hellings and
Dr. Oscar Steffenfor their interviews.
REFERENCES
Altabba, B., Einstein, H. & Hugh, C. 2004. An economic
approachtoriskmanagementfortunnels.InOzdemir(ed.)
Proc. North American Tunnelling 2004. NewYork: Taylor
& FrancisGroup.
Barends, F.B.J. 2005. Associating with advancing insight:
Terzaghi Oration2005. InProc. 16th International Conf.
on Soil Mechanics and Geotechnical Engineering, 1216
September, Osaka, Japan. Rotterdam: Millpress.
Bea, R.G. 2006. Reliabilityandhumanfactorsingeotechni-
cal engineeringJournal of Geotechnical and Geoenviron-
mental Engineering May: 63144.
Clayton, C.R.I. (ed.) 2001. Managing Geotechnical Risk:
Improving Productivity in UKBuilding and Construction.
London: TheInstitutionof Civil Engineers.
Essex, R.J. (ed.) 2007. Geotechnical Baseline Reports for
Construction: Suggested Guidelines. Danvers: ASCE.
Halman, J.I.M. (ed.). 2008. Risicomanagement in de Bouw:
Nieuwe Ontwikkelingen bij een Aantal Koplopers (in
Dutch). Boxtel: Aeneas.
Imai, M. 1986. Kaizen: The Key to Japans Competitive
Success. NewYork: RandomHouse.
Klein, K.J. & Sorra, J.S. 1996. The challenge of innova-
tionimplementationTheAcademy of Management Review
21(4): 105580.
Lewis, L.K. & Seibold, D.R. 1993. Innovation modifica-
tionduringintra-organizational adoptionThe Academy of
Management Review 18(2): 32254.
Rogers, E.M. 2003. Diffusion of Innovations (5thedn.). New
York: FreePress.
Smith,R.1996.Allocationof risk:Thecaseformanageability
The International Construction Law Review 4: 54969.
Smith, R.E. 2008. An evolving view of geotechnical engi-
neering: A focus on geo-risk management. In Proc.
GeoCongress 2008: The Challenge of Sustainability in
the Geoenvironment, 912 March, New Orleans. Reston:
ASCE.
Song, M. 2006. Research Presentation of the Direct/Indirect
Network ExternalitiesAdoptionModel DINAM.Enschede:
Universityof TwenteConstructionManagement & Engi-
neeringGroup.
Sperry, P.E. 1981. Evaluation of savings for underground
constructionUnderground Space 6: 2942.
Van Aken, J.E. 2004. Management research based on the
paradigmof thedesignscience: Thequestfor field-tested
andgroundedtechnological rules.Journal of Management
Studies, 41(2): 21946.
Van Staveren, M.Th., Bles, T.J., Litjens, P.P.T. & Cools, P.
2009. GeoRiskScan:A successful geomanagementtool.
InProc. 17th Int. Conf. on Soil Mechanics and Geotech-
nical Engineering, 59 October, Alexandria, Egypt (in
prep.).
VanStaveren, M.Th. 2007. Hurdlesandconditionsforimple-
mentingriskmanagementinpublicprojectorganizations.
Delft Cluster Working Paper, Work Package 2, November
2007. Delft: Delft Cluster.
Van Staveren, M.Th. 2006. Uncertainty and Ground
Conditions: A Risk Management Approach. Oxford:
Butterworth-Heinemann.
394
Design method (2)
Frameworkfor evaluationof probabilityof failureof soil nail system
I.S.H. Harahap&W.P. Nanak
Universiti Teknologi PETRONAS, Tronoh, Malaysia
ABSTRACT: This paper proposes conceptual framework to evaluate probability of failure of soil nailing
system.Thepaper startswithareviewoncurrentdesignmethodsof soil nailing. Hazardsrelatedtothestructure
areidentifiedthroughpossibleprevailingfailuremechanisms. Four failuremechanismsof soil nail systemsuch
as pull out failure, nail tendon failure, face failure and general stability failure were identified. Events that
triggeredfailurewereidentifiedthroughfault treeanalysis(FTA). For thispurposeasemi quantitativehazard
and risk analysis framework by Vaunat was adopted. Theframework consisted of four steps: (i) inventory of
factorsthatcouldleadtodanger, (ii) forecastingof danger, (iii) assessmentof hazardand(iv) evaluationof risk.
Frominspectionof fault treediagrams, onlysomeof theeventswerequantifiablethroughstructural reliability
analysis(SRA) method. Weareproposingahybridframeworkthatcombinesstructural reliabilityanalysis(SRA)
withBayesianBelief Network(BBN).Theframeworkcanbeusedaspartof riskmanagementof soil nail system
wherebyrisksassociatedtothehazardscanbequantifiedfor prioritizingrisksmitigationmeasures.
1 INTRODUCTION
1.1 Background
Soil nail systemhas been around over about thirty
yearsasearth-retainingstructure.Thestructuresolves
problemassociated with slope failure during slope
cutting, disturbed terrain and excavation. The basic
concept of soil nailing can bedefined as inserting a
passiveinclusionintothesoil inacloselyspacedpat-
terntoincreasetheoverall shear strengthof thein-situ
soil and restraint its displacement. This systemfun-
damentally improves thestability, in asoil slopefor
example, principally throughthemobilizationof ten-
sion developed in thesoil nails. Generally, thebasic
design concept of soil nailing systemis still based
on the conventional deterministic approaches which
depends upon the transfer of resisting tensile forces
generated in the inclusions into the ground through
frictionat itsinterfaces.
Systemreliability and risk based design concepts
have been increasingly proposed on a number of
papersrecently. However, for applicationinsoil nail-
ing(Yuan, 2003), it islimitedtoevaluationof sliding
mechanismonly, oneof manypossiblefailuremecha-
nismsinsoil nailing.Thepurposeof reliabilityanalysis
istoestimatetheeffectsof surroundinguncertainties
onthesafetyof theapplication. Inviewof that, relia-
bility analysis is ameans of evaluatingtheeffects of
uncertainties(Duncan, 2000). Theprobabilisticmeth-
ods onwhichtheseconcepts arebasedallowfor the
uncertainties in the input parameters and the model
describingpotential failuremodestobeevaluated.
Probability of failureof a systemand its compo-
nents play arolein risk management wheredanger,
hazardandrisk whichpotentiallyoccur inthesystem
canbeevaluatedfor thepurposeof prioritizingmiti-
gationmeasures. Ithasbeendefinedthatdanger isthe
event of failure, hazardis theoccurrenceprobability
of theevent andrisk istheexpectedconsequencesoh
hazards.
In this paper weproposeaframework to evaluate
probability of failure of soil nail system. Using this
wewill beableto assess theeffect of uncertainty to
theoverall performanceof thesystemas well as the
performanceof its components. Theframework con-
sistsof: (i) identificationof faultmechanismsthrough
faulttreeanalysis(FTA), (ii) evaluationof component
probability of failure and (iii) evaluation of system
probability of failure. It would be clear that not all
of the identified failure mechanisms can be evalu-
atedusingstructural reliabilityanalysis(SRA). Forthis
purposeBayesianBelief Network (BBN) wasusedto
evaluate the systemprobability of failure combined
withcomponentprobabilityof failureevaluatedusing
SRA method.
1.2 General concept
The method used within this paper followed a risk-
based approach for soil nail system. To predict the
foldingrisk,Rforexistingsoil nailssystem,itwasnec-
essarytoevaluatetheprobabilityof failure(P
f
) of the
respectiveslopeinstabilityandthepotential damages,
E(D) asresultsfromfailure.Theriskcanbeestimated
astheproduct of theprobability(P
f
) andthedamage
E(D). The concept of determining the overall prob-
ability of failure for a soil nail systemis illustrated
in Fig. 1. The figure shows an eight-step-procedure
whichstarts withdeterministic designandtheselec-
tionof themostrelevantfailuremodesfor whichlimit
397
Figure1. Simplified flowchart for determination of fail-
ureprobabilityof soil nail system(adoptedfromKortenhaus
et al. 2002).
state equations have to be determined. These initial
threestepsarepurelydeterministicandareessentially
basedonfailureanalyses.
Step 3 shows links with theset-up of afault tree
(Step7) whichcanbeusedto schematizethepoten-
tial failuremechanisminsoil nail system. Following
this methodadeterministic fault treecanbederived
whichcanlaterbeusedforprobabilisticdesign.Steps4
and5discusstheuncertaintiesof parametersandmod-
elswhichwill beincludedintheprobabilisticdesign.
Instep6theprobabilityof failureof all limitstateequa-
tionsisperformed. Steps7and8discusstherelation
betweenthefailuremodes andthecalculationof the
overall probabilityof failure(Kortenhausetal. 2002).
2 SOIL NAIL FAILURE MECHANISMS
Generally, the failure mode of soil nail systemcan
beclassified broadly as external and internal failure
mechanisms. External failure refers to the develop-
ment of potential failuresurfaces effectively outside
the soil nail ground mass (active zone). The failure
canbeintheformof sliding, rotation, bearingorother
forms of loss of overall stability. As for internal fail-
ure,itreferstofailureswithinthesoil nail groundmass
(passivezone).
The internal failure modes in the active zones
include failure of the ground mass, bearing failure
underneath soil nail heads, structural failure of the
soil nail under combined actions of tension, shear
and bending, structural failure of the soil nail head
or facingandsurfacefailurebetweensoil nail heads.
While on the passive zones, the modes of failure
are the pullout failure at ground-grout interface or
grout-reinforcement interface(GEO, 2006).
In risk analysis, all possible or potential failure
modesmustbeconsideredtoaddresspotential conse-
quencesof failure.Thisisachievedbyevaluatingavail-
ablenail forcestostabilizetheactiveblockdefinedby
any particular slipsurface. Thefailuremodes of soil
nailscanbecategorizedaspulloutfailure, nail tendon
failure, facefailureandoverall failure(Chow&Tan,
2004). Overall, the potential failure modes must be
consideredinevaluatingtheavailablenail forcetosta-
bilizetheactiveblock definedby any particular slip
Figure2. Pulloutfailuremode(a)beforeand(b)afterfailure
(fromFHWA, 1998).
Figure3. Nail tendonfailuremode: (a) beforeand(b) after
failure(fromFHWA, 1998).
surface. The following are general failure modes of
soil nail:
2.1 Pullout failure
Thisfailureresultsfrominsufficientembeddedlength
intotheresistantzonetoresistthedestabilizingforce.
Thepullout capacity of thesoil nails is governed by
thefollowingfactors(Tan& Chow, 2004):
Thelocationof thecritical slipplaneof theslope
Thesize(diameter) of thegroutedholefor soil nail
Theground-grout bondstress(soil skinfriction)
Nail tendon failure Thisfailureresultsfrominad-
equate tensile strength of the nails to provide the
resistantforcetostabilizetheslope. Itisprimarilygov-
ernedby thegradeof steel usedandthediameter of
thesteel (FHWA, 1998).
Face failure This aspect of failure mode for
soil nailing is sometimes overlooked as it is gener-
ally wrongly assumed that thefacedoes not resist
any earth pressure(Chow & Tan, 2004). Thesefail-
urestendtobefail ineither facingfailureor thefront
if nailszoneslidingoff (FHWA, 1998).
Overall Failure This aspect of failure mode is
commonlyanalyzedbasedonlimit equilibrium meth-
ods. Theanalyses arecarriedout iteratively until the
nail resistantforcecorrespondstothecritical slipplane
fromthelimit equilibriumanalysis. Tocarryout such
iterative analysis, it is important that the nail load
diagram(Fig. 5) isestablished(Tan& Chow, 2004).
FromFig. 5, it can beseen that thenail load dia-
gramconsist of threezones, A, B and C. ZoneA is
398
Figure4. Facefailuremode(fromFHWA, 1998).
Figure5. Nail loaddiagram(fromFHWA 1998).
governed by the strength of the facing, T
F
and also
theground-grout bondstress Q. If thefacingof soil
nails is designed to take full tensile capacity of the
nail, T
N
, thenthefull tensilecapacityof thenail canbe
mobilizedevenif thecritical slipcirclepassesthrough
ZoneA. However, to designthefacingwithfull ten-
silecapacity of nails, T
N
, instead of lower T
F
is not
economical for highslope(e.g. morethan15m).
Zone B is governed by the nail tendon tensile
strength. The ultimate tendon capacity is calculate
usingthefollowingequation:
whereA
N
=nominal diameter of nail, andF
Y
=yield
strengthof tendonmaterial.
Zone C is governed by the ground-grout bond
strength. Theultimateground-grout bondloadQ can
bedeterminedusingthefollowingequations:
where,
n
=average radial effective stress,
des
&
c
des
=designvaluesfor thesoil shearingresistance.
Theaverageradial effectivestress
n
actingalong
thepull-out lengthof asoil nail maybederivedfrom:
where,
v
=average vertical effective stress, cal-
culated mid-way along nail pull-out length and
K
L
=coefficient of lateral earth pressure parallel to
slope.
If active conditions (i.e.
v
=K
a
v
) assumed to
developperpendicularlytotheslope
wherethevalueof K
a
maybetakenas:
Critical nominal nail headstrength, T
FN
iscalculated
asfollow:
whereT
FN
=critical nail headstrength, C
F
=Flexure
pressurefactor, m
V,NEG
& m
V,POS
=vertical nominal
unit moment resistanceat thenail headandmidspan,
andS
H
& S
V
=respectively arehorizontal & vertical
nail spacing.
Facefailure: Vertical nominal unit moment
where, A
s
=area of tension reinforcement in facing
panel widthb, b =widthof unit facingpanel (equal
to S
v
), d =distancefromextremecompressivefiber
tocentroidof tensionreinforcement, f

c
=compressive
strengthof theconcrete
Facefailure: Punchingshear strength
Nominal internal punching shear strength of the
facing, V
N
D
c
= b
PL
+h
c
Nominal nail headstrength, T
FN
where C
S
=pressure factor for punching shear, and
A
C
& A
GC
bearing plate connection details given in
FHWA (1998).
It isclear that themobilizednail resistanceshould
not exceedthenail loadenvelopedevelopedfromthe
threefailurecriteriadiscussedearlier. Therefore, the
nail resistancetobeinput intoslopestabilityanalysis
should refer to thenail load diagram(Fig. 5) corre-
spondingto theavailablebondlengthfor thecritical
slipplane(Fig. 6).
A simple limit state equilibriumusing method of
slicecanbederivedasfollow:
where, The FoS stands for factor of safety,.
Q=capacity, momentduetoresistanceforces(capac-
ity) andL =load, moment duetodrivingforces,
399
Figure 6. (a) Available bond length for slope stability
analysis, (b) Forcediagram.
and
where n =number of slices, m=number of nail,
T
NJ
=ultimateground-groutcapacityatnail number j
establishedwithconsiderationtonail-soil interaction
model giveninFig. 5. Other parametersareshownin
Fig. 6.
2.2 Failure analysis
Through assessment of the potential failures which
beendescribed, soil nail failuremechanismshavebeen
identifiedwhichmayeventuallyleadtofailingof soil
nail system. The summarized procedures are shown
inFig. 7for theinternal, external andglobal stability
failuresof thesoil nail system.
All relevantprocessesasshowninFig. 7needtobe
describedandsimplifiedbylimitstateequation(LSE).
Theprocesses on theglobal stability of thesoil nail
Figure7. Failuremodesof soil nailssystem.
systemareslightlymorecomplexbutcanbesimplified
inthesameway.
3 EVALUATIONOF PROBABILITY OF
FAILURE
In the reliability theory context, the word failure
doesnotnecessarilyimplycatastrophicfailure.Todis-
tinguishthecontextual of reliabilitytheory,slidingof a
retainingwall for example wouldbemoreappropri-
ately describedasunsatisfactory performance than
asfailure. Whilefor another example, slopefailures
which involve large displacements are appropriately
describedbythewordfailure. Itisimportanttobear
inmindthelikelyconsequencesof themodeof fail-
urebeinganalyzed, andwhethertheyarecatastrophic
or morebenign(Duncan, 2000).
The tools in performing a reliability analysis fall
intothreebroadcategories. Thefirst arethemethods
of reliabilityanalysis, secondincludestheeventtrees,
fault treesandinfluencediagramswhichdescribethe
interaction among events and conditions in an engi-
neeringsystem, andthethirdincludesother statistical
techniques. Inparticular, someproblemsaresopoorly
definedthat it is useless to try to formulatemechan-
ical models thus causing engineer to rely on simple
statistics(Christian, 2004).
Direct reliability analysis procedure is borrowed
from practice in structural analysis, which prop-
agates the uncertainties in properties, geometries,
loads, water levels, etc., through analytical models,
toobtainprobabilisticdescriptionsof thebehavior of
a structure or system. The common approach is the
loadresistanceor capacitydemandmodel.
3.1 Structural reliability analysis
Uncertainty related to design variables such as load
(F) andcapacity (Q) is modeledas randomvariable,
anddesignrisk isgivenasprobabilityof failure(P
f
).
The basic of reliability problemis to evaluate prob-
ability of failure P
f
from pertinent statistical data
of Q and F, usually in the formof mean (m
F
and
m
Q
) and standard deviation (S
F
and S
Q
). If each of
Q and Fhas normal distribution, then the range of
400
safety(M =QF), alsohavenormal distribution, as
such mean and standard deviation has thefollowing
relationship:
If probability distribution of M is known, the
probabilityof failure(P
f
) canbeevaluatedasfollow:
whereProb(.) isprobabilityof aneventand+(.) stan-
dardnormal cumulativedensity function. Becauseof
difficulty in using probability of failure particularly
and small probability, and the connotation of fail-
ure, thereforereliability index has beenusedas a
measureof failure, where isdefinedas:
where +
1
(.)=inverse of standard normal cumula-
tivefunction, =theother formof P
f
whichismore
meaningful because of its scale of magnitude. The
rangeof reliabilityindexcommonlyusedingeotech-
nical andstructurearebetween1and4, correspondto
probabilityof failurerangebetween16and0,003%.
3.2 System reliability evaluation
Theconcept usedto consider multiplefailuremodes
and/or multiplecomponentfailureisknownassystem
reliability evaluation. A complete reliability analy-
sis includes both component-level and system-level.
Therearethreetypesof systems:
Component system A system with only one
performancefunction.
Series system or weakest link system A system
withmultipleperformancecriteriaandsystemfailure
may bedefinedasoccurringwhenany of thecriteria
isviolated.Thussystemfailureisdefinedbytheunion
of theindividual performancefailures. Thefirst-order
boundsfor asystemwithn failureeventsE
i
are:
where P
S
=probability of failure of the systemand
P(E
i
)=probabilityof failureof thei failureevent.
The lower bound is when all events are perfectly
dependent.Theupperboundiswhentheyaremutually
exclusive. If they are independent, the upper bound
becomes:
Parallel or redundant system This is thecaseif
failureoccurs when all thecomponents fail. System
Figure8. Safety, reliabilityandcapacity-demandmodel.
failureisdefinedastheintersectionof theindividual
(component)failureevents.Thefirstorderboundsare:
Inmostcasesthelower boundwill beclosetozero.
Theupper bound is exact if theevents areperfectly
correlated. If they areindependent, thelower bound
becomes:
Asshown, toanalyzeaseriesoraparallel systemwe
havefirsttocalculateindividual componentsprobabil-
ityof failures; thentherest isonlysubstitutioninthe
aboveequations to get systemprobability of failure.
So, inthisstudywehaveconcentratedontheanalysis
of component systemwithoneperformancefunction.
3.3 Bayesian belief network
Belief networksaregraphical representationsof mod-
els that capture relationship between model param-
eters. The network identifies variables that interact
directly and simplify belief updating by limiting
such variables processing to its local neighborhood.
401
Figure 9. Propagation of uncertainty in Bayesian Belief
Network.
Bayesianbelief network (BNN) isacausal belief net-
work where the nodes represent stochastic variables
and thearcs specify direct causal influencebetween
the linked variables (Suermondt, 1992). According
to Bayes ruletheposterior probability, or belief, of
hypothesisH giventheevidencee is
wherep(e|H)=likelihood of theevidencee,\ if the
hypothesesH istrueandp(H)=prior probabilitythat
Histhecorrect hypotheses. Thepropagationof belief
fromparent to its children for case of three nodes
networkisobtainedfromchainrule:
where (Z) is evaluated from p(Y|Z) and (X) is
evaluatedfromknownevidence.
4 RELIABILITY ANALYSISOF SOIL NAIL
SYSTEM
Basedonthefailureanalysisdescribedintheprevious
section, faulttreeanalysisisappliedfor thereliability
deterministicof eachfailuremechanismsof soil nails
system.Theresultsaresetupindetailedfaulttreecom-
prisingthedifferentfailuremechanismsforthepullout
failure, nail tendonfailure, facingfailureandoverall
instability.Thementionedfour(4) failuremechanisms
identifiedinthesoil nailssystemwasdefinedtobethe
top-eventof thefaulttree.Theindependenceof failure
modes was carefully checked and adjusted. Thefol-
lowingshowstheresultsof theanalysesonthefailure
mechanismsof soil nailssystem. Thetoplevel events
andsub-toplevel eventsthat causefailureof soil nail
systemareshowninFig. 10. Thetopeventsthatcause
failureinsoil nail systemareidentifiedaspull outfail-
ure, nail tendonfailure, punchingfailureandgeneral
stability failure. The sub-top event are common for
all toplevel events, theyarefault duringdesign, fault
duringconstruction,faultduetopoormaintenanceand
Figure10. Topeventandsub-topeventthatcausefailureto
soil nail system.
Figure11. Failureeventthatleadtotendonpull outfailure:
FaultTreeAnalysis.
fault duetoexternal factorssuchasunexpectedload,
changeintheset upof surroundingarea, etc. Thelast
threesub-topeventsareundevelopedeventsinasense
that qualitativeprobability evaluationaredifficult to
obtainandamendabletoothernonqualitativemethod.
Fault treediagramof pull out failuretogether with
itsconsolidatedformisshowninFig. 11. Examination
shows that events that leadtofailurearedistinguish-
ablebetweeneventsanditsprobability of failurecan
402
Figure12. Datasheet for soil nail example.
be quantified through structural reliability analysis
(SRA), i.e. event E5. Event E5 is read as event that
over estimate of tendon capacity due to uncertainty in
parameters to determine ground-grout bond ultimate
strength.
The event of over estimation of tendon pull out
capacity (E2) is related to events E3, E4, B6 and
B7. In turn E3 is dependent on B1, B2 or B3. The
effect of constructionmethodis difficult to quantify
usingstandardSRA method, however, it issuitablefor
analysisusingBBN. EventE6isdependentoeitherB4
or B5. Fault treediagramsanditsconsolidatedforms
for tendon failure, face failure and general stability
failurecanbeestablishedsimilarly.
A moresimplecausal relationshipcanbeobtained
usingBNN asshowninFig. 11. InspectiononFig. 11
indicatesthat:
1. Causal relationshipof somenodescannotbecasted
into capacity demandmodel andthereforecannot
beevaluatedusingSRA.
2. Propagation of uncertainty not necessarily in
True-Fault relationship, thereforesystemfailure
evaluation using Eqs. 15 to 18 provide only the
upper andlower bounds.
5 EXAMPLE
Thefollowingexampleis takenfromFHWA manual
as anexamplefor LRFD designof soil nailing. Data
geometry of thesoil nail isgiveninFig. 12andlevel
of detail forBNN isgiveninFig. 13.Thegreynodesin
Fig. 13represent stochastic variables as knownvari-
ables. Since the face failure does not lead to total
systemfailureitisnotincludedintheevaluationof sys-
temfailure, however thefailuremodeisstill relevant
for riskevaluation.
Thesoil nail systemprobability of failureis eval-
uatedbasedontheprobability of failureof eachnail
whichisinadifferentstageof failuregiventhefailure
surface. Thepropagationof different stageof failure
is propagated into failure of the systemusing BNN.
Figure13. Bayesianbelief networkfor soil nail system.
Giventhegeometryandstochasticvariablesthesteps
areasfollows:
1. Establish trial failurecircleand calculatetension
at eachtendonusingSRA withlimit statefunction
given in Eq. (1) considering soil-nail interaction
diagramshownFig. 5.
2. Evaluate system failure probability using BNN
adopting P
f
of each tendon established in Step 1
asnodesinBNN. Dependencyof failureof thesys-
temoneachnail isfoundusingSRA withlimitstate
functiongiveninEqs. (10) and(11)
3. Repeat Step1withnewtrial circleuntil optimum
P
f
canbeestablished.
4. Exampleontheapplicationof theframework and
results of parametric studies will be presented
duringtheconference.
6 CONCLUDINGREMARKS
Based on the preceding discussion the following
conclusionscanbeadvanced:
1. Wehavereviewedcurrent designapproachfor soil
nail system. Failuremechanismof thesoil nail sys-
temconsistsof pull outfailure, tendonfailure, face
failureandgeneral stabilityfailure. Faulttreeanal-
ysisrevealsthatfailureeventsaresolvabletoeither
quantitativeor qualitativemethod.
2. We are proposing a hybrid approach to evaluate
probability of failure of soil nail system. Events
thattriggerfailureof thesystemareevaluatedusing
quantitativeandqualitativemethod. Thestructural
reliability analysismethodisusedfor quantitative
methodandBNN isusedtopropagatebelief among
thenodesformingthesystem.
3. Exampleontheapplicationof theframework and
results of parametric studies will be presented
duringtheconference.
403
ACKNOWLEDGEMENT
Theauthorsacknowledgethemanagement of Univer-
sity Teknologi PETRONAS for their encouragement
andsupportstoattendtheconference
REFERENCES
Chow, C. M. & Tan, Y. C. (2006), Soil Nail Design:
A Malaysian Perspective, International Conference on
Slopes 78 August 2006, KualaLumpur, Malaysia.
Christian, J.T., Geotechnical EngineeringReliability: How
Well Do We Know What We Are Doing? Journal of
Geotechnical and Geo-Environmental Engineering,Octo-
ber 2004.
Duncan, J.M., Factor of Safety andReliability inGeotech-
nical Engineering, Journal of Geotechnical and Geo-
Environmental Engineering, April 2000.
FHWA, 1998, Manual for Design&ConstructionMonitor-
ingof Soil Nail Walls, FHWA-SA-96-69R, U.S. Depart-
mentof Transportation, Federal HighwayAdministration,
WashingtonD.C.
Kortenhaus, H. Oumeraci, R. Weissmann, & W. Richwien,
(2002), Failure Mode and Fault TreeAnalysis for Sea
andEstuary Dikes, researchincoastal engineeringand
funded by the Federal Ministry for Science, Education,
ResearchandTechnology(BMBF), Germany.
Suermondt, H.J. (1992) ExplanationinBayesianbelief net-
works, thesis submitted as partial requirement for the
degreeof Doctor of Philosophy, StanfordUniversity.
Vaunat, J. & Leroueil, S. (2002), Analysis of Post-Failure
SlopeMovements within theFramework of Hazard and
Risk Analysis, Natural Hazards, 26: 83109, Kluwer
AcademicPublishers, Netherlands.
Yuan, J.-X., Yang, Y., Tham, L.G., Lee, P.K.K. & Tsui, Y.,
New Approach to Limit Equilibrium and Reliability
Analysis of Soil NailedWalls, International Journal of
Geomechanics, 3(2):1345151, Dec. 2003.
404
Reliabilityanalysisof embankment damsusingBayesiannetwork
DianqingLi, HaihaoLiu, ShuaibingWu
State Key Laboratory of Water Resources and Hydropower Engineering Science,
Wuhan University, Wuhan, China
ABSTRACT: ThispaperaimstoproposeaBayesiannetworkapproachappliedforreliabilityanalysisof dams.
Thebasic principleof theBayesiannetwork is briefly introduced. Theimportancecoefficient of nodes inthe
Bayesian network is defined and the formula of the importance coefficient is presented. Then the Bayesian
network is employed to evaluatethereliability of embankment dams. Theresults indicatethat thereliability
of embankment dams can beevaluated using theBayesian network in amorerational way. Therelationship
betweenthecauses andconsequences associatedwithdamsafety evaluationcanbeexpressedinanintuitive
formatthroughtheBayesiannetwork. Inaddition, thecommoncausefailureassociatedwithreliabilityevaluation
of embankment damscanbeeasily solvedusingtheBayesiannetwork. Theresultsof importanceanalysiscan
aidintheidentificationof whichfactorsarecandidatesfor mitigationactions.
1 INTRODUCTION
It is increasing recognized by decision makers that
efficientmanagementandconsistentquantificationof
risks associated with largedams is an issueof soci-
etal concern. Thebook entitledRisk and Uncertainty
in Dam Safety provides details of theanalysis meth-
ods available to characterize and quantify the risks
anduncertaintiesindamsafety(HartfordandBaecher
2004).Thereliabilityandriskof damshasbeenexten-
sivelystudiedintheliterature. For example, Benjamin
(1983) exploredtheapplicationof statistical decision
theory to the dams and levees. Lave and Balvanyos
(1998) studiedtherisk analysisandrisk management
usingbenefit-cost analysis. Inrecent years, Bayesian
network(BN)(J ensen1996;J ensen2000)hasreceived
considerableinterestforriskassessmentsandasadeci-
sionsupport tool for technical systems. For instance,
Friis-Hansen(2001)exploredtheapplicationof BNon
marinestructures. Faber et al. (2002) studiedtherisk
assessment of decommissioning options using BN.
Bayraktarli etal. (2005) proposedaframeworkfor the
assessmentandmanagementof earthquakerisksbased
onBN, whereoptimal riskmitigationactionsareiden-
tifiedbasedonindicators. Holichy (2005) developed
aBN for firesafety designsituations. Straub(2005)
exploredtheapplicationof BNfornatural hazardsrisk
assessment.
The objective of this paper is to demonstrate the
potential and advantages of BN for the application
on reliability analysis of embankment dams. Firstly,
a brief introduction to the BN is presented. There-
after the importance coefficient for nodes in BN is
defined and the formula for calculating the impor-
tance coefficient is presented. Finally, examples are
investigatedtoillustratethepotential andadvantages
of BN for the application on reliability analysis of
embankment dams.
2 BAYESIANNETWORK
A Bayesiannetwork, alsoknownasaBayesianbelief
network, is a directed acyclic graph formed by the
variables (nodes) together with the directed edges,
attachedby tableof conditional probabilities of each
variableonall itsparents(J ensen1996). Thereforeit
is a graphical representation of uncertain quantities
anddecisions that explicitly reveals theprobabilistic
causal dependence between the variables as well as
the flow of information in the model. The Bayesian
networkmethodologyhasbeendevelopedandapplied
mostlyinthefieldof artificial intelligence. Moststud-
iesmainlyfocusedondevelopingefficientalgorithms
forlargeBayesiannetworkconstructionandcomputa-
tion. A brief andconciseoverviewonBN isprovided
byPearl andRussell (2000), moreextensivetextbooks
onBNincludeJ ensen(1996) andJ ensen(2001). Fur-
thermore, many softwarepackages, bothcommercial
andfreeware, areavailablefor thecomputationof BN.
Therefore, onlyahighlycondensedintroductiontoBN
ispresentedinthefollowing.
BN is a probabilistic model based on directed
acyclicgraphs. InaBayesiannetwork, thenodeswith-
outanyarrowsdirectedintothemarecalledrootnodes
that haveprior probability tables (discretenodes) or
functions (continuous nodes) associated with them.
The nodes that have arrows directed into themare
called child nodes and the nodes that have arrows
directedfromthemarecalledparentnodes. Eachchild
hasaconditional probabilitytable(or function) asso-
ciatedwithit, giventhestate(or value) of theparent
405
Figure1. A simpleBayesiannetwork.
nodes. TheBN representsthejoint probability distri-
butionP(X) of aset of variables X=X
1
, X
2
, , X
n
.
Thejoint probability distribution can beeasily com-
putedusingtheBN. However, inorder toobtainsuch
joint probability distribution, thenumber of required
probabilitiesincreasesexponentiallywiththenumber
of variables in the model. In the last decade, some
algorithms inorder to facilitatecalculatingingraph-
ical models havebeen developed (e.g., Celeux et al.
2005). BNenablesanefficient modelingbyfactoring
of the joint probability distribution into conditional
distributionsforeachvariablegivenitsparents.A sim-
pleBNisillustratedinFig. 1(Straub2005). Itconsists
of threevariablesX
1
toX
3
. X
1
isaparentof X
2
andX
3
,
whicharechildrenof theformer.
The joint probability distribution of this network
showninFig. 1isgivenby(Straub2005)
According to the assumptions, namely, d-separation
and conditional independence, the joint probability
distributionfor anyBNcanbeexpressedas
For computational reasons, BN is restricted to vari-
ables with discrete states. In some cases, Gaussian
variables can be used. However, it is required to
discretize all continuous randomvariables. The BN
allowsenteringevidence, theprobabilitiesinthenet-
work canbeupdatedwhennewinformationbecomes
available. For example, when the state of X
2
in the
network as shown in Fig. 1 is observed to be e,this
information will propagatethrough thenetwork and
thejoint prior probabilities of X
1
andX
3
will change
to thejoint posterior probabilities as follows (Straub
2005):
Notethat thecommoninfluencingvariableX
1
intro-
duces a dependency between X
2
and X
3
.This is a
typical situationindamriskassessment. For instance,
X
1
may represent the load conditions, and X
2
and
X
3
are damsites at two different locations. The BN
canbeextendedtodecisiongraphsbyincludingdeci-
sion nodes and utility nodes in the network, which
enables theassessment and theoptimization of pos-
sible actions in the framework of decision theory.
The optimal action is the one yielding the maximal
expectedutility. Thedecisionproblemondamsafety
isbeyondthescopeof thispaper.
Onemainpurposeintheprobabilityof damfailure
is to findout themost important causeso that more
attentionshouldbepaidtoit toreducetheprobability
of damfailure. Theimportanceanalysiscanbeeasily
conducted in theBayesian network. Theimportance
coefficient associatedwiththenodesrepresentingthe
basicvariablescanbedefinedas
where
i
istheimportancecoefficient of theithnode
intheBayesiannetwork; p
0
is thefailureprobability
of parent node; p
i
is thefailureprobability of parent
nodewhenthestateof theithnodeissafe.
3 ILLUSTRATIVE EXAMPLES
Reliability analysis of embankment dams is investi-
gatedtoillustratetheapplicationof Bayesiannetwork.
Thefirstexampleisthereliabilityanalysisof embank-
ment dams with common cause failure factors. The
secondexampleisreliabilityanalysisof embankment
damsbasedondatainChinausingBayesiannetwork.
Inthefirstexample, theconsideredcausesof damfail-
ureareovertoppingandslopefailure.Boththespillway
failureandtheunexpectedfloodcanleadtoovertop-
ping. Theunexpected flood and cracks in dambody
canleadtoslopefailure. It canbeseenthat theunex-
pectedfloodcanleadtonotonlyovertoppingbutalso
slopefailure, whichisthecommoncausefailurefac-
tor in the reliability analysis of dam. Therefore, the
reliabilityanalysisof damshouldaccount for thecor-
relationbetweentheprobabilitiesof overtoppingand
slopefailure.TheBayesiannetworkfortheaboveprob-
lemis shown in Fig. 2. It can be seen fromFig. 2
thatthenoderepresentingcrackindambodyincludes
threestates, namely, normal crackthatwill notleadto
slopefailure, serious crack that will possibly leadto
slopefailure,andcatastrophiccrackthatwill definitely
lead to slope failure. Assuming that the occurrence
probabilitiesof normal crack, seriouscrack, andcatas-
trophiccrackindambodyare0.949, 0.05, and0.001,
respectively. All theremainingnodeshavetwostates,
namely, trueand falserepresenting safeand failure,
respectively. Thecorrespondingprobabilitiesarealso
shown in Fig. 2. The cause and effect relationships
defining these mechanisms are strongly interrelated
andcannot representedby conventional event treeor
fault tree, whichconsiderseachmechanismindepen-
dently. However, suchproblemcanbeeasilysolvedby
theBayesiannetworkasdiscussedabove.
Basedonthedatareportedby Li et al. (2006), the
main causes of damfailure and the corresponding
annual probability of failureinChinaarereproduced
inTable1. It isclear that theovertoppingisthemajor
cause of damfailures, which results in 50.2% dam
failures among the total damaccidents. 34.8% dam
406
Figure2. Bayesiannetworkincludingcommoncausefailurefor reliabilityevaluationof embankment dam.
Table1. Causesof damfailureandannual probabilityof damfailurebasedonLi et al. (2006).
Causesdamfailure No. Percentage(%) Annual damfailure(10
4
)
Overtopping Unexpectedflood 435 12.6 1.01
Insufficient dischargecapacity 1032 37.6 3.29
Qualityof dambody Seepageof damfoundation 701 20.2 1.77
Slopestabilityof dambody 110 3.2 0.28
Spillway 208 6.0 0.53
Flood-releasingtunnel 5 0.1 0.013
Culvert 168 4.9 0.42
Collapseof dambody 13 0.4 0.033
Humanerrors 185 5.3 0.47
Others 212 6.1 0.54
Total 3339 8.75
failures arecausedby thepoor quality of dambody.
Human errors and theothers only occupy 15%dam
failures.BasedonTable1andtheprincipleof Bayesian
network, thecorrespondingBayesiannetwork shown
inFig. 3for reliabilityanalysisof embankment dams
is constructed with the help of the Hugin software.
For theconsideredproblem, thenodehas two states,
namely, true and false. Based on the data shown in
Table 1, the values of node can be determined as
8.44E-04, whichisthesamemagnitudeastheglobal
probabilityof failureintheworld.
To identify the key factors contributing to the
damfailure, theimportancecoefficients for thecon-
sidered factors should be evaluated. For illustrative
purposes,thenoderepresentingtheseepageisselected
for importanceanalyses. TaketheBayesian network
showninFig. 3asanexampleagain. Theimportance
analysiscanbeconductedbysettingtheprobabilityfor
thestateof falseaszerointheBayesiannetwork,which
impliesthattheseepagefailurecannotoccur. Thecor-
respondingBayesiannetwork isshowninFig. 4. The
annual probability of failurefor theconsidered dam
can be calculated with the help of the Hugin soft-
ware. The probability for quality of dambody with
problemsdecreasesfrom3.05E-04to1.27E-04when
somemeasuresaretakentoavoidtheseepagefailure.
Accordingly, thecorrespondingannual probability of
failuredecreases from8.44E-04to 6.67E-04. Apply-
ingEq.(4).theimportancecoefficientsforseepagecan
beobtainedas0.21. Similarly, theimportancecoeffi-
cients for theother variables canbecalculatedusing
Eq. (4) andareshowninTable2. Ascanbeseenfrom
Table2, insufficientdischargecapacityandseepageof
damfoundationaresignificantrandomvariableswith
higherimportancecoefficients.Thatis, changesinthe
dischargecapacityandtheseepageof damfoundation
will significantlyinfluencetheprobabilityof damfail-
ure.Therefore, toreducetheprobabilityof damfailure
407
Figure3. Bayesiannetworkappliedfor reliabilityevaluationof embankment dam.
Figure4. Bayesiannetworkfor reliabilityevaluationof embankment damwithout seepagedamage.
effectively, somemeasuressuchasenlargingspillway
should be taken to increase the discharge capacity.
Among the remaining randomvariables, the proba-
bilityof damfailureismostsensitivetochangesinthe
unexpectedflood, whilechangesintheflood-releasing
tunnel appear tobetheleast important factor.
4 CONCLUSIONS
The application of BN on reliability analysis of
embankment dams is demonstrated in this paper.
Anapproachfor importanceanalysis associatedwith
nodesinBNisproposed. Twoexamplesonreliability
408
Table 2. Importance coefficients of factors causing dam
failure.
Importancecoefficients
Factorscausingdamfailure (%)
Unexpectedflood 13.0
Insufficient dischargecapacity 39.0
Seepageof damfoundation 21.0
Slopestabilityof dambody 3.3
Spillway 6.2
Flood-releasingtunnel 0.1
Culvert 5.1
Collapseof dambody 0.5
Humanerrors 5.5
Others 6.3
analysisof embankmentdamsarepresentedtodemon-
strate the validity and capability of the proposed
approach. The results indicate that the reliability
of embankment dams can be evaluated using the
Bayesiannetworkinamorerational way.
The relationship between the causes and conse-
quences for damsafety evaluation can beexpressed
inanintuitiveformat throughtheBayesiannetwork.
In addition, the common cause failure problemfor
reliabilityevaluationof embankmentdamscanbeeas-
ily solvedusingtheBayesiannetwork. Theresultsof
importance analysis can aid in the identification of
whichfactorsarecandidatesformitigationactions.For
theconsideredexample, insufficientdischargecapac-
ity and seepage of dam foundation are significant
randomvariableswithhigherimportancecoefficients.
Accordingly, toreducetheprobability of damfailure
effectively, somemeasuressuchasenlargingspillway
shouldbetakentoincreasethedischargecapacity.
ACKNOWLEDGMENT
This research was substantially supported by grants
from the National Natural Science Foundation of
China(Project No. NSFC50879064) andtheNatural
ScienceFoundationof Hubei Province, China(Project
No. 2008CDA091).
REFERENCES
Bayraktarli, Y.Y., Ulfkjaer, J., Yazgan, U., et al. 2005. On
the application of Bayesian probabilistic networks for
earthquake risk management. Proceedings of the 9th
International Conferenceon Structural Safety and Reli-
ability. EditedbyAugusti G, Schuller GI, Ciampoli M,
Millpress, Rotterdam, 35053512.
Benjamin, J.R. 1983. Risk anddecisionanalyses appliedto
damsandlevees. Structural Safety1(1): 257268.
Celeux, G.C., Corset, F., Lannoy, A. et al. 2006. Design-
ingaBayesiannetwork for preventivemaintenanceform
expert opinions in a rapid and reliable way. Reliability
EngineeringandSystemSafety91(7): 849856.
Faber, M.H., Kroon, I.B., Kragh, E. et al. 2002. Risk
assessment of decommissioning options using Bayesian
networks. J ournal of Offshore Mechanics and Arctic
Engineering124(4): 231238.
Friis-Hansen,A.2001.BayesianNetworksasaDecisionSup-
port Tool in MarineApplications. PhD thesis, Technical
Universityof Denmark.
Hartford,D.N.D.&Baecher,G.B.2004.RiskandUncertainty
inDamSafety. ThomasTelford.
Holick, M. 2005. Reliability and risk assessment of build-
ings under firedesign situation. Proceedings of the9th
International ConferenceonStructural SafetyandRelia-
bility. EditedbyAugusti G, Schuller G I, Ciampoli M,
Millpress, Rotterdam. 32373242.
J ensen, F.V. 1996. An Introduction to Bayesian Networks.
NewYork: Springer.
J ensen, F.V. 2001. BayesianNetworksandDecisionGraphs.
Springer-Verlag, Newyork.
Lavel, B. & Balvanyos, T. 1998. Risk analysisandmanage-
ment of damsafety. RiskAnalysis18(4): 455462.
Pearl, J. & Russell, S. 2000. Bayesian Networks. UCLA
CognitiveSystemsLaboratory,Technical Report(R-277).
Straub, D. 2005. Natural hazards risk assessment using
Bayesiannetworks. Proceedings of the9thInternational
Conference on Structural Safety and Reliability. Edited
by Augusti G, Schuller G I, Ciampoli M, Millpress,
Rotterdam. 25092516.
VonThun, L.J. 1987. Useof risk based analysis in making
decisionsondamsafety. EngineeringReliabilityandRisk
inWater Resources, Editedby Duckstein, L, andPlateE
J. MartinusNijhoff Publishers. 135146.
409
Identificationandcharacterizationof liquefactionrisksfor high-speed
railwaysinPortugal
P.A.L.F. Coelho&A.L.D. Costa
Department of Civil Engineering, Faculty of Science and Technology, University of Coimbra, Portugal
ABSTRACT: ThePortugueseGovernmentdecidedtodevelopaHigh-SpeedRailway(HSR)networklinkingits
majorcitiesandotherEuropeancountriesviatheextensiveSpanishnetwork.Thisencouragedthedevelopmentof
aninternational researchprojectaimingatassessing, analyzingandmitigatingrisksaffectingthistransportation
infrastructure. Risk analyses undoubtedly depend on the ability to identify and characterize the risk factors
affecting theHSR construction and operation. Theinformation availableto performthesetasks is generally
scarce and ambiguous, which limits the use of ordinary methods. This paper considers the use of experts
opinionstocarryout identificationandquantitativeassessment of earthquake-inducedliquefactionrisks. Both
theprobability of occurrenceandtheconsequencesresultingfromtherisksareconsidered. Liquefactionrisks
areselected becauseof howseriously they can affect theHSR. Theresults arecompared with respect to the
level andtypeof experienceof theexperts. It is shownthat experts opinions areauseful tool to characterize
liquefactionrisks, evenif someissuesneedtobeconsidered.
1 INTRODUCTION
The Portuguese Government decided to develop the
firstHigh-SpeedRailway(HSR) networkinthecoun-
try. This network will significantly reduce journey
durations between major Portuguese cities located
along the coast, where most part of the population
lives. It will also radically improve the quality and
speed of international railway connections, via the
plannedconnectionstothevast Spanishnetwork.
Thepolitical supporttotheconstructionof theHSR
andto theestablishment of joint researchprojects in
thefieldof transportationinfrastructuresmotivatedthe
creationof aninternational researchproject that aims
atidentifying, assessing, analyzingandmitigatingthe
risksthatmayaffecttheconstructionandtheoperation
of a HSR, which involves uncommonly high finan-
cial and technical complexity. Hence, Risk emerged
as acollaborativeresearch project involving various
Portuguese Universities and Research Centers and
MIT, which has a long and fruitful experience with
researchonriskanalysesinengineering(e.g. Einstein
et al. 1995). This experience includes risk analyses
of railwayinfrastructuresbuilt incountrieslikeJ apan
that, as Portugal, are seismically active and present
complex geotechnical conditions (Sussmnan &
Shimamura2007).Theaimof theprojectistodevelop
tools that allow the development of comprehensive
risk analyses to support political decisions, namely
inHSR.
A component of Risk Project considers the seis-
mic and geotechnical risks affecting the Portuguese
HSR. This is justified by the fact that seismic and
geotechnical riskscanhaveveryseriousimpactonthe
construction and operation of the HSR. In addition,
thereisconsiderabledifficultyinidentifyingandchar-
acterizing, interms of probability of occurrenceand
magnitudeof theconsequences, thedifferent hazards
of seismic and geotechnical nature that may affect
the HSR. Therefore, the consideration of the seis-
micandgeotechnical risksisafundamental aspect of
risk analyses whenappliedto theHSR to bebuilt in
Portugal.
When performing risk analyses, experts opinions
areoftentheonly practical alternativeto accomplish
the task of identifying and assessing seismic and
geotechnical risks. Basedontheir experience, experts
identify therisksthat may possibly berelevant tothe
problemunder consideration. They also quantify the
probability that each risk will effectively material-
izeandtheeffectscausedby their occurrence, which
dependsalot onthelocal conditions.
Despitetheunderstandableneedtouseexpert pan-
elsasatool toappraisethedataallowingseismicand
geotechnical risks to be incorporated in risk analy-
ses, the soundness of this tool is seldomdiscussed.
This paper presents a preliminary assessment of the
validityof usingexperts opinionstocharacterizethe
risksrelatedtotheoccurrenceof earthquake-induced
(cyclic) liquefactioninHSR.
2 THE USE OF EXPERTS OPINIONSINRISK
ANALYSIS
Inordertocarryoutriskanalysis, differentprocedures
can be used to characterize the risk factors relevant
411
to the problem under consideration (Scawthorn
2008):
Analytical;
Empirical;
Expert opinion.
In many circumstances, it is practically impossi-
bletoaccomplishanalytical solutionsabletoquantify
acertain risk. This is particularly truein thecaseof
geotechnical and seismic risks, whereeven approxi-
matenumerical solutionscanrarelypresent afeasible
alternative.Empirical solutionsalsohavelimitedvalue
inthecaseof seismichazards, sincethereturnperiod
tendstobeconsiderablylarge. Therefore, itisimprac-
ticabletocollectenoughgoodqualitydatatoperform
statistical analysis.Withrespecttothisissue, Christian
(2004) expresseshisconcernthatpropertiesestimated
fromsmall samplesmay beseriously inerror. There-
fore, expert opinion is often the ultimate source of
informationthat canbeused.
Hartford(2008)considersthatthetermexpertjudg-
ment involves enumerating subjective probabilities
thatreflectanexpertsdegreesof belief. Healsostates
that an expert must possess two fundamental capa-
bilities: the substantive expertise and the normative
expertise. Thefirst onedetermines howwell aset of
assessmentspredictstheactual outcome, highproba-
bilitiesbeingonaverageassignedtothoseeventsthat
turnout tooccur. Thesecondonerelatestotheability
of assigningprobabilitiestotheeventsthatcorrespond
totheir empirical frequencyof occurrence.
Expert opinions can be used at different phases
in risk analysis involving seismic and geotechnical
risks, namely duringtheearly planninganddecision
stages, duringthecomparisonof technical alternatives
and during the final design and construction stages
(Einsteinet al. 1992). Theusefulness of this tool has
beendemonstratedinpractice, namelyincaseswhere
thegeotechnical part wasasignificant element of the
risk analysis (Einstein & Haas 1984, Einstein et al.
1992).
Christian(2004)reckonsthatexpertsgenerallytend
toestimatemeantrendswell buttendtounderestimate
uncertainty. Inaddition, heconsidersthatexpertstend
Figure1. Hydrological, geological andseismiccharacteristicsof thePortugueseterritory.
tobeoverconfident intheir estimates. However, these
conclusions aredebatable, mostly dueto thereasons
whyexperts opinionsbecamepopular: theresultsare
typicallydifficult toverify(Scawthorn2008).
Taking into account all the issues regarding the
useof this tool inrisk characterization, Faber (2008)
considers that the use of expert panels still lacks a
generallyapplicableandconsistent philosophy.
3 LIQUEFACTIONINPORTUGAL
3.1 Characteristics of the territory
Earthquake-inducedliquefactionisaserioushazardto
aHSR, whichisextremelysusceptibletosomeeffects
causedbytheoccurrenceof thephenomenon:
thesettlement or failureof thegroundfoundations,
whichcancausederailment of thehigh-speed-train
and/or long-lasting disruption of the operation or
reductionof itsquality;
thevibrationinducedat thegroundsurface, which
cancausederailment of thehigh-speedtrain, even
if the earthquake causes no permanent deforma-
tionof therailway, asobservedinNiigataChuetsu
Earthquake(Watanabeet al. 2004).
ThePortugueseHSR will alsobevulnerabletothe
effectsof earthquake-inducedliquefaction. Infact, the
geological andhydrological conditionsof asignificant
partof theterritorywhich,alongwithitsseismicity,are
showninFigure1, favortheoccurrenceof earthquake-
inducedliquefaction.
First of all, various regions locatednear thecoast
andalongthealluvial plainsof themajor Portuguese
rivers (Figure 1a), especially South of the city of
Porto, haverelativelyshallowwatertables. Inaddition,
manyof depositsexistinginthoseregionsaremostly
formed by recent sedimentary soils, which areillus-
tratedasdarkareasinthegeological mapof Figure1b.
Becausethesedeposits commonly containcohesion-
lessmaterialsinaloosecondition, theyareextremely
pronetoearthquake-inducedliquefaction. Finally, the
Portugueseterritory is aseismically activeregion, as
412
boldly demonstrated by the catastrophic event that
struckLisbonin1755.Theseismiczoningproposedin
EC8(Figure1c)showsthatthesouthernpartof theter-
ritoryismorepronetolargeseismicevents,namelythe
areasaroundandsouthof Lisboninthevicinityof the
westernandsoutherncoasts. Theseismicity depends
onthemechanismscausingtheearthquake, thecoun-
trybeingaffectedbybothinterplateandintraplatetype
of events.
3.2 Historical liquefaction events and implications
to the Portuguese HSR
Figure2undoubtedlyunderlinestheargumentthatliq-
uefaction can be particularly damaging to the HSR
inPortugal. Infact, analysesof historical testimonies
describingtheeffectsof pastearthquakessuggestthat
liquefactionhaseffectivelyoccurredrepeatedlyinthe
pastintheterritory(Figure2a). Theoccurrenceswere
instigated by earthquake events of different magni-
tudeand having distinct mechanisms in their origin.
Theresults compiled by J orge& Vieira(1997) sug-
gestthatthe1755LisbonEarthquakecertainlycaused
liquefaction episodes in all theregions identified as
potentiallyliquefiable. However, nosureevidencewas
foundregardingliquefactionoccurrencesduetoother
earthquakesoutsidethealluvial plainof Tagus(Tejo)
river.
This implies that the Portuguese HSR may be
affectedmoreorlessfrequentlybyliquefactioneffects
inasignificant part of itsalignment (Figure2b). This
alignment is fairly rigid, dueto theneedto servethe
population, whichlives predominantly intheregions
whereliquefiabledepositsarepresent (Figure2c).
3.3 Empirical assessment of liquefaction risks
In order to qualitatively judge the value of experts
opinions inthecharacterizationof liquefactionrisks,
Figure2. Alignmentsof theHSRinrelationtothedemographicdistributionandthehistorical liquefactioneventsinPortugal.
anempirical assessmentof thoseriskswascarriedout
for theHSR sectionbetweenLisbonandPortousing
theinformationavailable.
Thefirst principleacceptedwas thehistorical cri-
terion, whichstatesthat liquefactiontendstorecur at
locations whereit occurred in previous earthquakes,
asobservedinthefield(Youd1984) andincentrifuge
models(Coelhoetal. 2006). Ontheotherhand, evenif
thescaleof theeffectswill dependonthelocal ground
conditionsandmagnitudeof theseismicmotion, it is
reasonabletoexpect that somedegreeof damagewill
alwaystakeplaceduetoliquefaction. Thus, theprob-
ability of occurrenceof aliquefactionrisk equalsthe
probability of exceedanceof anearthquakethat once
liquefied the ground in that location. Consequently,
the annual probability of occurrence of a liquefac-
tion risk is given by theinverseof thereturn period
of agroundmotionthat inducedlocal liquefactionin
previousoccasions.
Thereturnperiodof interplateearthquakes gener-
atedintheseaandaffectingthePortugueseterritory,
such as the1755 and 1761 events, is in theorder of
athousand years. Intraplateearthquakes occur more
often, the return period for those generated in the
Lower Tagus Valley region being in the order of a
fewhundredyears. Table1relatesreturnperiodsand
earthquakemagnitudesfor theLisbonarea.
Table 2 lists the magnitudes and origins of the
earthquakes supposed to have caused liquefaction
Table1. Returnperiodsandmagnitudesof earthquakesin
theLisbonarea(CamposCosta2008).
Returnperiod
(years) 95 200 475 700 975 2000 5000
Magnitude 7.2 7.6 7.9 8.1 8.2 8.4 8.5
(M)
413
Table 2. Magnitudes and origins of the earthquakes
considered when identifying liquefaction occurrences in
Figure2a.
Earthquakecharacteristics
Year Origin Magnitude
1531 LowerTagusValleyregion 7
1755 Interplate(exact locationunknown) 8.59
(1,2)
1761 Interplate(Gorringezone) 8
1856 LoulFault
1858 SadoSubmarineValleyregion 7
1909 LowerTagusValleyregion 5.86.3
(3,4)
(1)
Gutscher (2006);
(2)
Chester (2008),
(3)
Teves-Costaet al.
(1999);
(4)
SousaOliveiraet al. (2006)
in Portugal in the past, which are plotted in
Figure2a.
Largerinterplateearthquakes, generatedinthecon-
tact between the Eurasian and African plates, have
magnitudes of 8 or higher. These events, which can
causeliquefactioninall themajorliquefiabledeposits
existing in thePortugueseterritory, may havereturn
periods as high as 5000 years, according toTable1.
However, more comprehensive studies suggest that
theseeventshavelower returnperiods. Therearesug-
gestionsthatthesevarybetweenanoverlypessimistic
estimateof 614105years(Chester2008)andamore
realistic predictionrangingfrom1500to 2000years
(Gutscher 2006). Consideringthemiddlepointof this
last rangeasthemost suitablevalue, theannual prob-
ability of exceedance of an interplate seismic event
capable of liquefying deposits in different regions
in Portugal is around 0.06%. Therefore the annual
probability of liquefaction occurring in a particular
liquefiabledepositinPortugal, unlessitislocatednear
Tagusriver, maybeintheorder of 10
4
to10
3
.
Withrespecttointraplateearthquakes, whosemag-
nitudecanreachvaluesashighas6to7,theprobability
of occurrencetendstobelarger.AccordingtoTable1,
thereturnperiodfor that typeof events is under one
hundred years, which suggests that theannual prob-
ability of exceedanceis closeto 1%. Based on this,
andconsideringthattheseearthquakeshavedefinitely
liquefieddepositsintheTagusriver valley, theannual
probability of liquefaction occurring in this location
isintheorder of 10
2
. Becausenofirmevidencewas
foundthattheseseismiceventscaninstigateliquefac-
tioninother deposits, this probability may not apply
to liquefiable deposits located outside the region of
Tagusriver valley.
Insummary, theannual probabilitiesestimatedfor
the interplate (10
4
to 10
3
) and intraplate (10
2
)
earthquakes approximatetheprobability of liquefac-
tiontooccurinliquefiabledepositsoutsideandwithin
theregionof Tagusriver valley.Theprobabilityof liq-
uefactionriskstomaterializewill obviously beequal
toor lessthantheprobabilityof liquefactiontooccur,
each risk having its own probability of occurrence.
For example, theprobability of liquefaction-induced
settlementstoaffecttheoperationof theHSRiscloser
totheprobability, P
L
, of liquefactiontooccur thanis
theprobabilityof trainderailmentduetoliquefaction-
inducedfoundationsettlementsor failure, bothbeing
belowP
L
.
4 USE OF EXPERTS OPINIONSINRISK
CHARACTERIZATION: A CASE STUDY
Inconnectionwiththeresearchperformedunder Risk
Project, asurveywasconductedtoevaluatethevalue
of experts opinionsinassessingdifferent riskfactors
for theHSR. Usingasetof riskfactorsidentifiedbya
restrictednumber of experts involvedintheresearch
project, other experts were invited to check if the
risks listed were adequate and to quantify the prob-
ability of occurrenceandthepotential consequences
for eachrisk.Thispaper discussestheresultsobtained
regardingliquefactionrisks.
4.1 Risk factors identified
This paper considers the risks related to the lique-
faction phenomenon and affecting theembankments
of the HSR. The risks taken into account, which
concernthedeformation, stability andsoil-structure-
interactionaspectsof theproblem, are:
Risk L1: settlement of the embankment due to
liquefactionof thegroundfoundation;
Risk L2: failureof theembankment dueto lique-
factionof thegroundfoundation;
RiskL3: trainderailmentduetoliquefactioneffects
on the ground motion propagation through the
deposit.
4.2 Groups of experts consulted
To assess theimportanceof theprofessional experi-
ence and occupation of the experts, three groups of
experts wereinvitedto participateinthesurvey. The
groups includeexperts that arerepresentativeof the
followingprofessional categoriesinPortugal:
G1: industry/practitioners(8experts);
G2: junior academics(7experts);
G3: senior academics(9experts).
Thus, theresults canbecomparedto perceivethe
influenceof theexperts previouslyaccumulatedlevel
andtypeof experience. Thesoundnessof theexperts
opinionsmayalsobeexaminedonaqualitativebasis.
4.3 Regions tested
Two regions were considered in the present study
(Figure 3). The northern region is located on the
alluvial plain of Mondego river, in potentially lique-
fiablesoils, inanareaof moderatetolowseismicity.
Thesouthernareais locatedinthehigher-seismicity
regionsurroundingLisbonontherelativelyhardrock
formationsof theJ urassicandCretaceous.
414
Figure3. Locationof theregionstestedinhydrological and
geological mapsof thePortugueseterritory.
4.4 Information requested
Eachexpert invitedtoparticipateintheinvestigation
wasaskedtoclassifytheprobabilityof occurrenceand
themagnitudeof theeffectsfor eachriskfactor, using
a semi-quantitative scale comprising 5 levels. The
annual probability of occurrencecouldvary between
theextremevaluesof 10
4
and1, with3intermediate
values varying by a factor of 10. The consequences
of eachrisk occurrencecouldbequalitatively graded
as: verylittlesignificant, littlesignificant, significant,
seriousandveryserious.
Thestudy was carriedout throughablindwritten
survey, no supplementary explanations being given
on the risks proposed. The surveys were completed
individually and no attempt was made to promote a
consensus between the experts with respect to the
valuesassignedtoeachrisk.
4.5 Results obtained
Almost every expert agreed with theset of risk fac-
torslisted. Noneof thefewexpertsdescribingthelist
as incompletesuggests additional risks relatedtothe
occurrenceof thephenomenonof earthquake-induced
liquefaction.
4.5.1 Probability of occurrence
The experts consulted assessed the probability of
occurrenceof therisk factors proposedinthesurvey
for thetworegions, asshowninFigure4. Except for
riskL3, all theexpertsmanagedtoquantifytheproba-
bilityof occurrencefor eachriskfactor. Thissuggests
that risk L3is poorly understoodor evenignoredby
many engineers. Thefact that thenumber of experts
declining to estimate the probability of occurrence
for risk L3 is considerably larger within the least-
experienced group of junior academics corroborates
thishypothesis.
Figure 4. Annual probability of the occurrence of lique-
factionrisks inthetworegions consideredas quantitatively
assessedbydifferent categoriesof experts.
Figure 4 highlights the considerable disparity of
values proposed by the experts within any of the
groups. Maximumand minimumpredictions for the
probability of occurrence often differ by two orders
of magnitude, but in some cases the extreme val-
ues diverge by three orders of magnitude. Industry
and senior academic experts tend to agree more on
theprobabilityof occurrence, yieldingsimilar relative
discrepanciesandaveragevaluesfor eachrisk.
Table 3 presents the averages of the probabilities
of occurrencefor eachrisk calculatedper regionand
415
Table 3. Averages
(1)
of the annual probabilities of
occurrenceof liquefactionriskspredictedbytheexperts.
Averageannual probabilityof occurrence
J unior Senior
Risk Region Industry academic academic Global
L1 1 5.610
3
1.410
3
1.710
3
2.410
3
2 3.210
3
5.210
3
1.310
3
2.810
3
L2 1 5.610
3
0.510
3
2.210
3
1.810
3
2 3.210
3
3.710
3
1.710
3
2.710
3
L3 1 3.710
3
4.010
3
3.210
3
3.610
3
2 3.710
3
1.810
3
2.410
3
2.510
3
(1)
computed from 10(average of log of probabilities), ignoring
the no answer replies.
per groupof experts, inthemanner stated. It canbe
inferredthat:
theannual probability of occurrenceof therisksis
estimatedbetween0.510
3
and5.610
3
;
industryandsenioracademicexpertstendtopredict
that liquefaction risks are more likely to occur in
region 1 than in region 2, while junior academic
memberstendtoexpressacontraryview;
in region 1, senior and junior academics believe
that risk L3 has the largest probability of occur-
rence, whereas practitioners consider that this risk
isthelesslikelytooccur, whichsuggeststhatinfor-
mationonoccurrencesof earthquake-inducedtrain
derailment isbetter knownamongst academics;
in region 2, senior academic and industry experts
consider L3asthemost likely risk tooccur, junior
academicsconsideringthatL1hasthelargestprob-
abilityof occurrence;
industryexpertsalwayspredictahigher probability
of ariskoccurringthansenior academics;
industry expertsandsenior academicstendtopro-
duceestimatescloser totheglobal averagevalues,
whilejunior academicsassignvaluesthat arearbi-
trarilyaboveor belowthat averageandthat tendto
havesignificant impact onit.
Theresultssuggestthatindustryexpertsarealways
morepessimistic when evaluating therisk of occur-
rence in comparison to senior academic. This may
reflect the need of industry experts to incorporate a
factor of safety indesign, whichapparently does not
influenceasmuchtheassessment madeby academic
experts.Thevaluesproposedbyjunioracademicsvary
randomly andoftensignificantly aroundtheaverage,
whichmayreflecttheir inexperiencewithsomeof the
risks. Evenif industryandsenioracademicpredictthat
therisksaremorelikelytooccurinregion1, theglobal
averageiscontrolledbytheopinionof thejunior aca-
demics. In fact, this group predicts, on average, that
liquefactionriskshaveaprobabilityof occurrencethat
is about 4 to 7 times larger in region 2, for risks L1
andL2, respectively.
Figure5. Consequences of theoccurrenceof liquefaction
risksinthetworegionsconsideredasqualitatively assessed
bydifferent categoriesof experts.
4.5.2 Consequences of occurrence
Theexpertsconsultedalsoassessedtheprobabilityof
occurrencerelativetoeachriskfactor for bothregions
(Figure5). Thevariability of thevalues proposedfor
themagnitudeof theeffectsinducedbytheoccurrence
of each risk factor, within each group, is very sig-
nificant. Infact, 3or 4different levels of magnitude
arenormallyusedbythedifferent expertstoquantify
theconsequences of arisk factor inacertainregion.
Moreover, senior academicsneedtoemploythe5lev-
elsof magnitudeproposedinthescalewhentryingto
assesstheconsequencesof liquefactionriskL2inboth
regions.
416
Table 4. Averages of the consequences
(1)
of liquefaction
riskspredictedbytheexperts.
Averageconsequencesof occurrence
J unior Senior
Risk Region Industry academic academic Global
L1 1 3.0(s) 3.4(s/se) 3.0(s) 3.1(s)
2 3.5(s/se) 3.6(s/se) 3.3(s/se) 3.5(s/se)
L2 1 4.0(se) 3.1(s) 3.3(s/se) 3.5(s/se)
2 3.9(se) 3.9(se) 3.4(s/se) 3.7(s/se)
L3 1 4.0(se) 3.4(s/se) 4.0(se) 3.8(se)
2 4.3(se/vse) 3.2(s) 4.1(se) 3.9(se)
(1)
measured in the following scale: 1- very little significant
(vls); 2- little significant (ls); 3- significant (s); 4- serious
(se); 5- very serious (vse).
Morethan25%of thejuniorexpertsandabout10%
of theotherexpertswereunabletoquantifytheeffects
of theoccurrenceof liquefactionrisk L3. Becauseall
theexpertsmanagedtoquantifytheconsequencesrel-
ative to the other risks, this suggests that risk L3 is
worseunderstoodthantheothers, especiallybyjunior
academics.
Table4presents theaverages of theconsequences
of theoccurrencefor eachrisk predictedby thethree
groupsof expertsfor eachregion. Inviewof thescale
usedtoassesstheconsequences, rangingfrom1(very
littlesignificant) to5(veryserious), itcanbeobserved
that:
the consequences predicted for each risk vary
betweenaminimumvalueof 3.0, indicatingsignif-
icanteffects, andamaximumvalueof 4.3, whichis
indicativeof serioustoveryseriouseffects;
normally, theconsequences of thethreerisks con-
sideredareexpectedtobemoresevereinregion2
than in region 1, which suggests that the experts
focusedmoreontheseismicity of theregionthan
onthegeological andgeotechnical conditionsof the
foundation;
in region 1, all the groups believe that risk L3
cancausesomeof themost serious consequences,
althoughindustryandjunior academicexpertsalso
believe the effects of L3 are comparable to those
duetorisksL2andL1, respectively;
in region 2, industry and senior academic experts
identifyriskL3asthemostseriousintermsof con-
sequences, whereasjunioracademicexpertspredict
thisriskswouldhavetheleast consequences;
industry experts tendto predict moreserious con-
sequencesinmostpartof thecasesthansenior aca-
demics, althoughthedifferenceismoresignificant
whentheprobabilityof occurrenceisevaluated;
industry expertsandsenior academicstendtopro-
duceaveragepredictions of consequences that are
closertotheglobal averagevalues,whiletheaverage
estimatesof junioracademicsvarymorenotablyand
randomly aroundtheaverage, thoughnot as much
asintermsof theprobabilityof occurrence.
Theresultssuggest that, alikethepredictionof the
probabilityof occurrence, industryexpertstendtobe
morepessimisticthansenior academicwhenevaluat-
ing theconsequences of theoccurrenceof therisks.
Also, the larger variation of the values proposed by
junior academicsmayindicateapoorer understanding
of someof therisks.Thefactthatlarger consequences
areexpectedtooccur inregion2suggest that experts
believethat, if therisks occur, themagnitudeof the
effects will be governed by the scale of the seismic
motion and not by the geological and geotechnical
conditions of thesite. Experts also seemto fear the
occurrence of derailments, namely those caused by
thevibrationof thegroundsurface.
4.6 Soundness of the experts opinions
Oneof themaindifficultieswhenassessingthesound-
nessof theopinionsexpressedbyexpertsinrelationto
riskanalysis, namelywheninvolvinggeotechnical and
seismichazards, istheveryfewtruthful dataavailable.
Still, qualitativeassessmentof someof thevaluespro-
ducedby theexperts canbeperformedbasedonthe
resultspresented.
Comparingtheestimatesproducedbetweendiffer-
ent groups, it isapparent that:
estimates produced by senior experts exhibit less
variation than thoseof junior experts, which sug-
geststhat thelevel of experienceof theexpert isan
important factor;
predictionsof industryexpertsaremorepessimistic
thanthoseof academicexpertswiththesamelevel
of experience, bothwithrespecttotheprobabilityof
occurrenceandtotheconsequences, whichproves
thatthetypeof experienceof theexpertsinfluences
their opinions;
becausemanyearthquakesmaycausegrounddefor-
mations without causing its failure, the fact that
senior experts suggest that L2 is at least as likely
tooccur asL1seemsincongruent;
thereasonwhytheconsequencesof theriskstendto
belarger inregion2, wherethegroundfoundation
islesssusceptibletoliquefaction, isunclear;
thefactthattheconsequencesof riskL3aredifferent
inregions1and2, asproposedbyall thegroupsof
experts, isnot straightforwardlyexplicable.
Similar analyses can also be performed based on
the absolute values proposed by the experts for the
probability of occurrenceof each risk. As discussed
in section 3.3., the annual probability of a liquefac-
tionrisk tomaterializeinaliquefiabledeposit varies
fromabout 10
4
to 10
3
for region 1 and is about
10
2
for region2.Thus, therangeof annual probabili-
tiesof occurrenceobtainedfor region2, rangingfrom
1.310
3
to5.210
3
, seemsqualitatively reason-
able, sincethelocal deposits arefairly unsusceptible
to liquefaction. For region 1, the annual probabil-
ity of occurrence, varying between 0.510
3
and
5.610
3
, seems overestimated, as it exceeds the
expectedprobabilityof liquefactionoccurrenceat the
site.
417
5 CONCLUSIONS
Risk analysesconsideringtheeffectsof geotechnical
andseismic hazards areextremely demanding, espe-
cially with respect to thecharacterization of therisk
factors interms of theprobability of occurrenceand
magnitude of the effects. Suitable risk analyses are
required to justify public investment in large trans-
portation infrastructures, such as theoneplanned in
Portugal for theHSR. Duetothecharacteristicsof the
territory, thegeotechnical andseismic aspectsof risk
havecrucial importance, whichmotivatedaresearch
onthissubject.
Thestudypresentedinthispaper considerstheuse
of experts opinions as atool to providethedataon
theprobabilityof occurrenceandonthemagnitudeof
theconsequencestotheHSRduetoliquefactionrisks.
Basedonablindwrittensurveytothreedifferenttypes
of experts (junior andsenior academics andindustry
experts), itcanbeconcludedthatexperts opinionsare
auseful tool tocharacterizeseismicandgeotechnical
risks. However, someissuesmaynot beignored.
The level and type of experience of the experts
may have significant impact on the data obtained:
experiencedexpertsyieldlessvariabilityandindustry
members are more pessimist than academics. Also,
some values proposed are incongruent. In order to
overcometheseproblems, getmoreunbiaseddataand
tohelpclarifyingpossibledoubtsof theexperts, joint
meetingsmaycontributetoimprovetheuseof experts
opinionsinriskanalyses.
ACKNOWLEDGEMENTS
This publication was madepossibleby thegenerous
support of the Government of Portugal through the
PortugueseFoundationfor International Cooperation
inScience,TechnologyandHigherEducationandwas
undertakenintheMIT-Portugal Program.
REFERENCES
Campos Costa, A. 2008. Fundaments of probabilistic risk
analysis Methods and applications, Advanced Course
onRiskmanagementincivil engineering, LNEC, Lisbon.
Chester,D.K.2008.The effects of the 1755 Lisbon earthquake
and tsunami on the Algarve region, southern Portugal,
GeographyGeographical, 93(2): 7890.
Christian, J.T.C. 2004. Geotechnical Engineering Reliabil-
ity: How Well Do We Know What We Are Doing?- Karl
Terzaghi Lecture, ASCE-J ournal of Geotechnical and
Geoenvironmental Engineering, 130(10): 9851003.
Coelho, P.A.L.F 2007. In situ densification as a liquefaction
resistance measure for bridge foundations, PhD Thesis,
CambridgeUniversity, UK.
Coelho, P.A.L.F., Haigh, S.K. & Madabhushi, S.P.G. 2006.
Effects of successive earthquakes on saturated deposits of
sand, International ConferenceonPhysical Modellingin
Geotechnical Engineering, Hong-Kong.
EinsteinH.; ChiaverioF.; Koppel U. 1995. Risk analysis for
the Alder tunnel, Int. J ournal of Rock Mechanics and
MiningSciencesandGeomechanicsAbstracts, 32(4)
Einstein, H.H, Dudt, J.-P., Halabe, V.B. & Descoeudres, F.
1992. Decision Aids in Tunneling Principle and Prac-
tical Aplication, Swiss Federal Officeof Transportation,
ProjectAlpTransit.
Faber, M. 2008. Management of structural risks, Advanced
CourseonRiskmanagementincivil engineering, LNEC,
Lisbon.
Gutscher, M.-A. 2006. The great Lisbon earthquake and
tsunami of 1755: lessons from the recent Sumatra earth-
quakes and possible link to Platos Atlantis, European
Review, 14(2): 181191, CambridgeUniversityPress.
Hartford, D. 2008. Science-based management of civil asset
risk- Objectives, Principles, Process and Analytical Tech-
niques, Advanced Course on Risk management in civil
engineering, LNEC, Lisbon.
INE 1991. XIIth General Census of the population, National
Instituteof Statistics, Portugal.
J orge, C. & Vieira, A.M. 1997. Liquefaction potential
assessment- application to the Portuguese territory and
to the town of Setbal, Seismic behaviour of ground
and geotechnical structures, ScoePinto(ed), Balkema:
3343.
LNEC2008Study of alternative locations for the NewLisbon
Airport, Report LNEC2/2008-DT,LNEC(in Portuguese).
Pardal,L.F.2005The Portuguese Development of High-Speed
Railway, inwww.rave.pt (in Portuguese).
Scawthorn, C. 2008. Earthquake risks, AdvancedCourseon
Riskmanagement incivil engineering, LNEC, Lisbon.
Sousa Oliveira, C., Roca, A. & Goula, X. 2006. Assess-
ing and Managing Earthquake Risk: Geo-scientific and
Engineering Knowledge for Earthquake Risk Mitigation:
Developments, Tools, Techniques, Springer Ed.
Sussmnan, J.M. & Shimamura, M 2007. R&D Symposium
roundtable- Cooperation with MIT, J R EAST Technical
ReviewNo.10, J apan.
Teves-Costa, P., Borges, J.F., Rio, I., Ribeiro, R. &Marreiros,
C. 1999. Source Parameters of Old Earthquakes: Semi-
Automatic Digitization of Analog Records and Seismic
Moment Assessment, Natural Hazards J ournal, 19 (2-3):
205220.
Watanabe, G., Lee, T.Y, Nagata, S., Sakellarai, D. &
Kawashima, K. 2004. Preliminary investigation on the
damage of transportation facilities in the 2004 Niigata
Chuetsu Earthquake, Technical Report, Kawashima
ResearchGroup, J apan
Youd, T.L. 1984. Recurrence of liquefaction at the same site,
8thWCEE, USA, Vl.3:231238.
418
Fieldcharacterizationof patternsof randomcracknetworks
onvertical andhorizontal soil surfaces
J.H. Li & L.M. Zhang
Department of Civil and Environmental Engineering, The Hong Kong University of
Science and Technology, Clear Water Bay, Hong Kong
ABSTRACT: Cracksarecommoninnatural andengineeredsoilsandprovidepreferential pathwaysfor water
infiltrationinto soils. Randomly distributedcracks caninducerandomness andanisotropy inpermeability of
soils.Theobjectivesof thispaper aretostudythecrackpatternandtheprobabilitydistributionof theorientation
of cracks ontwo soil surfaces inthenatural environment. Fieldtests wereconductedto study thedesiccation
crack pattern by fully exposing the soils to rainfall and evaporation. A digital imaging method was used to
measurethecracks. Thepatterns of thecracks on avertical soil surfaceand on ahorizontal soil surfaceare
investigatedseparately. Thecracksat thesurveysiteformaninter-connectedcolumnar structure. Thecrackson
thevertical soil surfacearemainlyalongthevertical directionandthehorizontal direction.Therefore, thecracks
aregroupedintotwocracksets: avertical cracksetandahorizontal crackset.Theprobabilitydistributionsof the
orientationsof thetwocrack setsfollownormal distributions. Ontheother hand, theorientationsof thecracks
onthehorizontal soil surfacearerelativelyuniformineachdirectionandfollowauniformdistribution.
1 INTRODUCTION
Cracks arecommon in natural and engineered soils.
The presence of cracks may decrease the stability
of slopes through threeeffects. First, cracks provide
preferential pathways for water flow, whichwill sig-
nificantlyincreasethehydraulicconductivityof slope
soilsand, inturn, porewater pressuresinslopesoils.
As aresult, theshear strength of theslopesoils will
bereduced. Second, water-filledcracksexert anaddi-
tional driving force on the slope. Third, cracks can
formapartof thecritical slipsurfacethathasnoshear
strength, whichleadstoreducedsafetymargin.
Cracksinsoilscanbeinducedbydesiccation, tem-
peraturechanges, unevensettlement, constructionetc.
Crackpatternsmaybedifferentfor cracksof different
origins. Itisessentiallyimpossibletodeterministically
characterize the pattern and geometry of a complex
network of cracks. Therefore it has become more
commontodescribethecrack network stochastically
(Longetal. 1982).Theflowof waterthroughacracked
soil isstronglyinfluencedbytherandomlydistributed
cracks (Morris et al. 1992; Konrad & Ayad 1997;
Perret et al. 1999). The randomness in the geomet-
rical parametersof crackscaninducerandomnessand
anisotropyinthepermeabilitytensor(Li 2008; Li etal.
2009). Particularly, theanisotropy of permeability is
significantly affectedby theorientationof cracks(Li
& Zhang 2007). Therefore, the study on the crack
patternandtheprobabilitydistributionof crackorien-
tationwill provideabasistoanalyzetherandomness
andtheanisotropyof permeabilityinducedbyrandom
crack networks. The objectives of this paper are to
characterizethepatternsof desiccationcracksandto
studytheprobabilitydistributionsof theorientationof
thecracksonavertical soil surfaceandonahorizontal
soil surfacethroughafieldstudy.
Thesurveysiteislocatedat thefoot of Huangshan
slope, which is beside Nanxu Avenue in Zhenjiang,
downstream of Yangtze River, China. The average
annual rainfall atthesurveysiteis1088mm.Theaver-
age annual duration of sunshine is 2001 hours. The
survey sitewas excavatedandbackfilledin2002. In
2003, a landslide occurred after a heavy rainstorm,
which led to serious traffic congestion. Thedensely
distributed cracks on the slope are believed to have
playedacrucial roleinthelandslide(Xiaoetal. 2004).
Crack survey, doubleringinfiltrationtests, andtwo-
stage infiltration tests were conducted to study the
geometry of cracks and the hydraulic properties of
the cracked soil. A vertical soil surface and a hori-
zontal soil surface(asshowninFig. 1) wereselected
toconductfieldcharacterizationof cracks. Fig. 1also
showsthegroundconditionsof thetest site. Thepar-
ticlesizedistributionof theshallowsoil is shownin
Fig. 2. Thepredominant particles aresilt (34%) and
clay (35%), withsomesand(23%) andgravel (8%).
Typical geotechnical properties of thesoil, including
dry unit weight, saturatedwater content, liquidlimit,
plasticlimit, shrinkagelimit, andexpansiveindex, are
given inTable1. Thewater content in thefield was
in the range of 30% to 13%. The soil is classified
as a silty clay with a mediumexpansivity in accor-
dancewiththecriteriainASTM standardD4829-03
(ASTM 2003). An expansive soil exhibits signifi-
cant swellingandshrinkinguponwettinganddrying.
419
Figure1. Groundconditionsat thesurveysite.
Figure2. Particlesizedistributionof thesiltyclay.
Table1. Propertiesof thesiltyclayat thesurveysite.
Soil property Values
Dryunit weight (kN/m
3
) 16.7
Saturatedwater content (%) 30.0
Liquidlimit (%) 40.4
Plasticlimit (%) 15.6
Shrinkagelimit (%) 13
Expansiveindex 56
Consequently, densely populated cracks develop at
shallowdepths.
2 FIELDCHARACTERIZATIONOF RANDOM
CRACK NETWORKS
A fieldsurveywasconductedtoidentifythecrackson
avertical soil surfaceandonahorizontal soil surface
thatwereexposedtorainfall andevaporation.A digital
imagingmethodwasusedtologthecracks.Alargesur-
veyareawaspartitionedintomanysmaller plots. The
sizeof eachsmall plotwasabout180mmby210mm.
Snapshots weretakenof eachplot periodically. Each
snapshot wastakenwithminimizeddistortionbyfix-
ingthecameraonatripodandusingtwoscalesplaced
around the plot as a reference. The camera had a
5-Megapixel capacity and the effective pixels were
about 2560by 1920. Theresolutionof eachsnapshot
was theratio between thepicturesizeand theeffec-
tivepixels, whichwasabout 0.09mm. Eachsnapshot
wasimportedintoAutoCADandscaledtoitsfull size
basedonthereferencescalesaroundthepicture. The
snapshots of anareaof interest wereassembled. The
combinationof thesnapshotsgaveacompletepicture
of thecracksinthelargesurveyarea.
Figure3shows apictureof therandomcrack net-
work on the vertical soil surface. Fig. 4 shows a
picture of the random crack network on the hori-
zontal soil surface. Each picture covers a 530mm
by 820mmareaandcomprises of 12snapshots. The
cracksformedacomplex,three-dimensional columnar
structurealthoughonly theparametersof thesurface
cracks weremeasuredinthis study. Mizuguchi et al.
(2005) observedasimilar columnar crackstructurein
an 11.6mmthick starch-water mixtureusing aresin
solidificationmethod.
A continuous openingalongacertaindirectionin
thesoil isdefinedasacrack.Thecrackisapproximated
byastraight lineasshowninFig. 5. Thegeometryof
thecrack canbedescribedby its length, orientation,
location, andaperture(asshowninFig. 5). Thelength
of the crack is defined as the length of the straight
line that approximates the crack. The orientation of
the crack is defined as the angle between the crack
andtheglobal coordinates.Thecracklengthandcrack
orientation weredeterminedby theaidof AutoCAD
inthisstudy.
420
Figure 3. A digital image of the crack network on the
vertical soil surface.
3 PATTERNOF CRACKSONTHEVERTICAL
ANDHORIZONTAL SOIL SURFACES
Thecracksonthetwosoil surfacesareapproximated
bystraightlinestostudythecrackpattern.Fig.6shows
thecrackstructureonthevertical soil surfaceobtained
fromthesnapshotsshowninFig. 3. Fig. 7showsthe
crackstructureonthehorizontal soil surfaceobtained
fromthesnapshotsshowninFig. 4. Mostcracksinthe
sampleareatendtobeconnectedandformtriangles,
quadrangles, pentagonsor hexagonsonboththeverti-
cal soil surfaceandthehorizontal soil surface. These
polygonsareinterconnectedandformarandomcrack
network. Thecracksformapolygonal networkdueto
theinfluencesfromboththestressfieldandthemois-
turefield in thesoils. Becausetheevaporation from
crack walls is more rapidly than that fromthe bulk
soil, the water content gradient is large at the ends
andturningpointsof theexistingcracks. Asaresult,
the stresses at the ends and turning points are large
andthebranchingof cracksismost likely tooccur at
thesepoints.Thenewcracksconnectwiththeexisting
cracksandformapolygonal network.
Thereare106cracksinthesurveyareaonthever-
tical soil surfaceshowninFig. 6. Thesecracks form
one triangle, twelve quadrangles, twelve pentagons,
Figure 4. A digital image of the crack network on the
horizontal soil surface.
Figure5. Approximationof acrack.
andthreehexagons. Thereare903cracks inthesur-
vey areaonthehorizontal soil surfaceshowninFig.
7. Thesecracks formninetriangles, 74quadrangles,
96pentagons, and32hexagons. Thedistributions of
thepolygonsalongthex andy directionsarerelatively
uniform. The number of polygons in the middle of
421
Figure 6. Crack structure on the vertical soil surface.
(unit: m).
Figure 7. Crack structure on the horizontal soil surface.
(unit: m).
theplot is larger thanthat near theedges of theplot,
becausemanycracksnear theedgedonot formcom-
plete polygons. The interconnected polygonal struc-
tureof cracks offers preferential pathways for water
flowthroughthecrackedsoils.
4 ORIENTATIONOF CRACKSONTHE
VERTICAL ANDHORIZONTAL SOIL
SURFACES
Cracks on the horizontal soil surface distribute uni-
formlyalongthex andy directionsasshowninFig.7.
These cracks developed mainly due to desiccation
of the soil in the relatively homogenous horizontal
ground. Thesoil water evaporatesuniformlyfromthe
soil surface. Intheprocess, thesoil suctionincreases
and thesoil volumedecreases, which inducetensile
stresses in the surface soils (Fredlund & Rahardjo
1993). When the soil properties are isotropic in all
horizontal directions, the tensile stresses are identi-
cal inall horizontal directions. Asthetensilestresses
become larger than the tensile strength of the soil,
cracks will develop. Theisotropic stress fieldresults
intheuniformly distributedcracks onthehorizontal
soil ground.
Cracks onthevertical soil surfacearemainly dis-
tributedintwodirections: thehorizontal directionand
thevertical direction(as showninFig. 6). Cracks in
different directionsresult fromdifferent stressstates.
Anisotropicincremental tensilestressfieldispresent
onthevertical soil surfaceduetoisotropic dryingin
thevertical plane, whichis similar to thestress field
inthehorizontal soil ground. However, thegravityof
soil exertsanadditional anisotropicstressfieldinthe
soil withthemajorprincipal stressinthevertical direc-
tion.Thesuperimpositionof theisotropictensilestress
fieldandtheanisotropic gravity stress fieldgives an
anisotropic resultant stress field prior to crack for-
mation. Thedirectionof themajor principal stress is
consistent withthedirectionof thegravity, that is, in
thevertical direction. Asaresult, thehorizontal stress
must betheminor principal stress and tends to bea
tensilestress. Whenthetensilestressislarger thanthe
tensilestrengthof soil, cracks will appear first along
thevertical direction. Hence, thecracksinthevertical
directionformmainly dueto thecombinedactionof
thesetwostressfieldsresultingfromdesiccationand
thegravityof soil.
The cracks in the horizontal direction develop
mainlyduetodesiccationof thesoil. Whenthedesic-
cationinducedtensilestress is larger thanthetensile
strength of thesoil, newcracks formand tend to be
perpendicular to the existing cracks. The phenom-
enacanbeinterpretedusingaspringnetwork model
(Vogel et al. 2005). Inthis model, soil is represented
by a two-dimensional triangular network of simple
Hookean springs. When the springs break, a crack
will form. Perpendicular to a crack the neighboring
springs are relaxed, while parallel to a crack, the
tension is slightly higher compared to thefully con-
nectedsprings. Consequently, if acracktipapproaches
422
Figure8. Histogramand assumed normal distribution for
the orientation of the cracks on the vertical soil surface:
(a) thevertical crackset; (b) thehorizontal crackset.
another existingcrack, thejoint of thetwocrackswill
bemostlikelyat90
.Therefore, thenewcracksetwill
be perpendicular to the existing vertical cracks and
mainlyinthehorizontal direction.
The cracks on the vertical soil surface can be
groupedinto two crack sets: avertical crack set and
ahorizontal crack set. Theaverageorientationof the
vertical crack set is about 87
; the average orienta-

tionof thehorizontal crack set isabout 4
. Mahtab
et al. (1972) developedanumerical methodfor ana-
lyzing clusters of orientation data. It was found that
rock fractureorientationsfollowedanormal distribu-
tionwithameanvalueintheprevailingdirection. In
thisstudy, thenormal distributionisassumedtofitthe
observeddataof crack orientationonthevertical soil
surface.
Figure8showsthehistogramsof crackorientation
forthetwocracksetsonthevertical soil surface.Achi-
squaretest isusedtocheck thegoodnessof fit of the
assumednormal distributions for thetwo crack sets.
Thechi-squaretest is a widely used goodness-of-fit
test (AngandTang2007). Thevertical crack set has
19intervals; thereforethedegreeof freedom(d.o.f.) is
f =192=17. Thecritical value, c
1
,
f
, is40.8for
thenormal distributionatasignificancelevel of 0.001,
andtheobserveddifferencec
s
is8.1.Theobserveddif-
ferenceissmallerthanthecritical value.Therefore, the
normal distributioncanbeusedtomodel thevertical
cracksetonthevertical soil surface. Similarly, for the
horizontal crack set, theobserveddifference, 11.2, is
Figure9. Histogramandassumeddistributionsfor orienta-
tionof thecracksonthehorizontal soil surface.
muchsmaller thanthecritical value, 40.8, atasignifi-
cancelevel of 0.001.Accordingtothetest, thenormal
distributionisvalidat the0.001significancelevel for
modelingtheprobability distributionsof orientations
for thecracks on thevertical soil surface. Normally
distributedorientationisalsoobservedforfracturesin
rocks(National ResearchCouncil 1996).
Figure9 shows thehistogramfor theorientations
of thecracksonthehorizontal soil surface. Bothnor-
mal distributionanduniformdistributionareassumed
tomodel theorientationof cracks. Theobserveddif-
ference for the normal distribution is 155, which is
larger than the critical value, 66.6, at a significance
level of 0.001. The result indicates that the normal
distribution cannot beused to model theorientation
of thecracks onthehorizontal soil surface. Thechi-
squaretestresultsforuniformdistributionarelistedin
Table2. Theresultsindicatethattheuniformdistribu-
tionisappropriateatthe0.001significancelevel. The
averageorientationforthecracksisabout90
.Therel-
ativelyuniformlydistributedcrack orientationresults
fromthe isotropic drying of the relatively homoge-
neoussoilsintwohorizontal directions. Theuniform
distributionof crackorientationgivesrisetouniformly
distributedpolygons.
Thedeviationof theprobabilitydistributionfor the
orientationof thecracksonthetwosoil surfacesmay
result fromdifferent stress fields prior to crack for-
mation, as discussedintheprevious sections. Onthe
vertical soil surface, anumber of approximately par-
allel cracksarepresentandcharacterizedastwocrack
sets. Each set of cracks has a dominant orientation
and thedeviations fromthedominant direction may
benormallydistributed. Insteadof obviouscracksets,
uniformlydistributedcrackpolygonsarepresentinthe
horizontal ground. Desiccationcracksdisperseevenly
andresult inuniformly distributedcrack orientations
onthehorizontal soil surface.
5 CONCLUSIONS
A field survey was conducted and adigital imaging
methodwasusedtodocument andanalyzethecracks
423
Table2. Resultsof chi-squaretestfor orientationof cracksonthevertical soil surfaceand
onthehorizontal soil surface.
Assumed d.o.f. Critical value Observed
Location distribution f c
10.001,f
differencec
s
Vertical Vertical set Normal 17 40.8 11.2
surface Horizontal set Normal 17 40.8 8.1
Horizontal surface Uniform 35 66.6 40.8
Normal 35 66.6 155.0
onavertical soil surfaceandahorizontal soil surface.
Thecrack patternandtheprobability distributionsof
theorientationof thecracksonthesetwosoil surfaces
werestudied.
Most cracks tend to be connected and form
triangles, quadrangles, pentagonsorhexagonsonboth
the vertical soil surface and the horizontal soil sur-
face. Thesepolygons areinterconnected and forma
randomcrack network. Theinterconnectedpolygonal
structureof cracksprovidespreferential pathwaysfor
water flow. Theisotropicdryingonthehorizontal soil
groundresults inanisotropic stress fieldinthehor-
izontal directions. The uniformstress field leads to
uniformlydistributedcracksandcrackpolygonsonthe
horizontal soil surface. Hence, theorientation of the
cracksonthehorizontal soil surfacefollowsauniform
distribution. Thereis anadditional anisotropic stress
fieldinducedbythegravityinthesoil onthevertical
surfacebesidestheisotropicstressfieldduetodesicca-
tion. Cracksonthevertical soil grounddevelopunder
theeffectsof theresultantanisotropicstressfield, with
themajor andminor principal stressesalongtheverti-
cal andhorizontal directions, respectively. Therefore,
thecracksonthevertical soil surfacearecharacterized
bytwocracksets: avertical cracksetandahorizontal
crack set. Theorientations of thetwo sets of cracks
follownormal distributions.
ACKNOWLEDGEMENTS
This research was substantially supported by the
Research Grants Council of the Hong Kong Spe-
cial AdministrativeRegion(Project No. 622206). The
authors would liketo thank ProfessorsY. Wang and
J. Wei, Mr. C. P. Kwong, Mr. D. Feng, andMr. H. H.
Chenfor their kindhelpduringthefieldsurvey.
REFERENCES
Ang, A.H.S., andTang, W.H. 2007. Probability concepts in
engineering: emphasis on applications in civil & environ-
mental engineering. J ohnWiley& Sons, NewYork.
ASTM 2003. DesignationD4829-03: Standardtest method
for expansion index of soils. ASTM Book of Standards,
Vol. 04.08, West Conshohocken, PA.
Fredlund, D.G., andRahardjo, H. 1993. Soil mechanics for
unsaturated soils. J ohnWiley& Sons, NewYork.
Konrad, J.M., andAyad, R. 1997. Desiccationof asensitive
clay: field experimental observations. Can. Geotech. J.,
34: 929942.
Li, J.H. 2009. Field experimental study and numerical sim-
ulation of seepage in saturated/unsaturated cracked soil.
PhD thesis, The Hong Kong University of Science and
Technology, HongKong.
Li, J.H., and Zhang, L.M. 2007. Water flow through ran-
domcracknetworkinsaturatedsoil. Geotechnical Special
Publications, 170: 205214, ASCE, Reston, Va.
Li, J.H., Zhang, L.M., Wang, Y., and Fredlund, D.G. 2009.
Permeability tensor and REV of saturated cracked soil.
Canadian Geotechnical Journal, inpress.
Long, J. C. S., Remer, J. S., Wilson, C. R., andWitherspoon,
P. A. 1982. Porousmediaequivalentsfor networksof dis-
continuous fractures. Water Resources Research, 18(3),
645658.
Mahtab, M.A., Bolstad, D.D., Alldredge, J.R., andShanley,
R.J. 1972. Analysis of fracture orientations for input to
structural modelsof discontinuousrock. U.S. Bur. Mines,
Rep. Invest. RI (7669).
Mizuguchi, T., Nishimoto, A., Kitsunezaki, S.,Yamazaki,Y.,
andAoki, I. 2005. Directional crackpropagationof gran-
ular water systems. Physical Review E, 71(5): 056122.
Morris, P.H., Graham, J., andWilliams, D.J. 1992. Cracking
indryingsoils. Can. Geotech. J., 29(2): 263277.
National Research Council 1996. Rock fractures and fluid
flow: contemporary understanding and applications.
Washington, D.C.
Perret, J., Prasher, S.O., Kantzas, A., andLangford, C. 1999.
Three-dimensional quantificationof macroporenetworks
in undisturbed soil cores. Soil Sci. Soc. Am. J., 63(6):
15301543.
Vogel, H.J., Hoffmann, H., Leopold, A., andRoth, K. 2005.
Studiesof crackdynamicsinclaysoil:A physicallybased
model for crackformation. Geoderma, 125: 213223.
Xiao, G.F., Chen, C.X., Lin, T., andLiu, C.H. 2004. Stability
analysis of clayey gentleslopes consideringvariationof
water level. Rock and Soil Mech., 125(11): 17541760.
424
Stochasticmethodsfor safetyassessment of aEuropeanpilot
site: Scheldt
M. Rajabalinejad, P.H.A.J.M. vanGelder & J.K. Vrijling
Hydraulic Engineering Department, Faculty of Civil Engineering, TUDelft, the Netherlands
ABSTRACT: In this study, weprovideacomparison to theapplied methods for thesliding process in the
reliabilityanalysisof thepilotsiteScheldt.ThesemethodsarebasedonBishoptechniquebenefitingtherandom
fieldtheory(Vanmarcke1983).Thefiniteelementmethodisusedasabenchmarktoevaluatetheoutcomeof the
Bishopmethodintegratedwiththerandomfieldtheory whichtakesintoaccount variationof soil parameters.
Sincetheslidingprocessisakey issueinthesafety assessment of dikerings, thiscomparisonisanimportant
issue.
1 INTRODUCTION
1.1 Pilot site Scheldt
Flood defence systems are becoming more impor-
tant structurestoengineers, inhabitants, anddecision
makers. FloodSite project
1
isanevidenceof acollec-
tive attempt of 37 European countries to reduce the
risk of floodingintheir homelands. Thereareseveral
pilotsitesdefinedinthisprojecttoapplyandevaluate
newtechniques: River ElbeBasininGermany, River
ThamesEstuaryinEngland, andRiverScheldtEstuary
intheNetherlands.
TheScheldt pilot siteis atypical NorthSeaarea
protected against coastal flooding by means of dif-
ferent flooddefencestructuressuchasforelands, sea
dikes, dunes and other constructions. The Western
Scheldt forms theentranceto theharbor of Antwerp
(Belgium). Water levels are influenced by the wind
surgesontheNorthSea, aswell astheriverdischarges
fromtheScheldt.Therearefoursurroundingdikering
areas longtheWesternScheldt. Thesedikerings are
numbered from29 to 32 as shown in Figure 1. The
planviewof theareaisalsopresentedinFigure2.
Thereliability assessment of this siteis animpor-
tant task which is to be applied by utilizing of the
limit state equations (LSE). A considerable number
of LSEs areaddressed in thereport of thetask 4 in
FloodSite (Allsop2007). Asamatter of fact, theout-
comes of the task 4 and task 7 (Van Gelder 2008)
present thebuildingblocksof thisenormousresearch
project. In these studies, an attempt has been made
to defineandapply LSEs for different failuremodes
of a usual flood defence structure, and most of the
presentedLSEscanbeeasilyutilizedforsafetyassess-
ment. Among different failuremodes, sliding is one
of the important and influential modes which can
1
www.floodsite.net (20042008). Integrated Flood Risk
AnalysisandManagement Methodologies.
Figure1. Dikeringareasinthesouthernpartof theprovince
Zeeland are presented, this area is counted as number 32
(VanGelder et al., 2008).
be modeled in two classes of analytical approaches
(AA) and finite elements (FE). Since the reliability
assessment of the FE models is a complicated and
timeconsumingprocess(Rajabalinejad, Kanningetal.
2007), a few analytical methods areprogrammed in
Mprostab. Thisprogramisabletocommunicatewith
PCRing. PCRingisalsoareliabilitytool whichtakes
intoaccountdifferentparameterstoassesstheoverall
safety of a dike ring. Therefore, sliding which is an
important failuremechanisminthesafetyassessment
of dikes,istobecarefullywatched.Thispaperpresents
apartof theobservationsforthecorrespondenceof the
currently implementedmethodwith afiniteelement
model usingPlaxiscode.
2 THE MODEL
The dike ring area Zeeuws-Vlaanderen, counted as
Dike Ring 32, was initially divided into 287 dike
425
Figure2. Planviewof theScheldt estuary, thewhiteareaiscalledZeeuwschVlaannderen whichreferstoDikeRing32,
Scheldt estuary, theNetherlands(Googleearth, 2007).
Figure3. Theselectedcrosssectionsof DikeRing32inZeeuws-Vlaanderen, Scheldt, theNetherlands.
sections. Then, a selection was made and totally 33
dikeand4dunesections aredeemedto berepresen-
tative of the total dike ring. This number is without
the water retaining structures (14 structures). The
lo-cation of the selected dike sections is shown in
Figure3. Becausecalculatingtheslidingmechanism
is anelaborateprocess, this calculationwas not per-
formed for all sections. Thedistrict water board has
madeaselectionof 7cross-sectionprofiles (out of a
seriesof 40that wereusedfor thetesting) duringthe
processof schematization(VanGelder et al., 2008).
Asaresult, 7Profileshavebeenselectedtocalcu-
latetheprobabilitiesof failureconcerningthesliding
mechanismwithMproStab. Inthisstudy, twoof these
sectionsarebrieflypresented, morecompletereportis
presentedin(RajabalinejadandvanGelder 2007).
2.1 Section ALS166B
A typical cross sectionof thedikeringinZeelandis
presented in Figure 4. This section is located in the
NorthEast part of Ring32, as indicatedinFigure3.
The available data for this area can be divided in
twocategoriesof geometryandgeotechnics. Figure4
showsthegeometry of thiscrosssectioninwhichall
thesoil layersarehorizontal.
Thematerial properties of this dikearepresented
inTable42of thereport of (Rajabalinejad and van
Gelder2007), wherethematerial numbersinthistable
arecorrelatedtothenumbersof soil layersinthefig-
ure. Themain portion of this dikeis sand on top of
thefigure, anditstoeismadeof clay. Themainmate-
rial of foundation, also, is sand. However, there are
somelayers of peat (Veen), andlooseclay (Slapklei)
426
Figure4. A typical crosssectionof theDikeRing32inthe
Scheldt pilot site, namedasALS166B depictedbyMStab.
Figure5. A typical crosssectionof theDikeRing32inthe
Scheldt pilot site, namedasEMMA118depictedbyMStab.
under thebody of dike. Thevariationof soil proper-
ties is consideredby meanvalue, standarddeviation,
and a distribution type which are presented in the
aforementionedreport.
2.2 Section EMMA118
Figure5showsthecrosssectionEMMA118.Theloca-
tion of this section is illustrated in Figure 3. This
sectionisalsoasandydikelocatedonthesand, peat,
and loose clay. The downstreamtoe of the dike is
made of clay to prevent erosion and make the dike
less permeable. The detailed properties of materials
arepresentedinTable52in(Rajabalinejadandvan
Gelder2007).Themeanvalue, standarddeviation, and
thedistributiontypeof eachvariablearealsopresented
inthistable.
3 THETHEORY
ApartfromSection3.1,whichpresentsanintroduction
tothesoil variation, therestof thispaper isconsistent
withtheuser manual of MProstab(Calle1994).
3.1 Lognormal distribution
Here we apply a lognormal probability distribution
function(PDF) incombinationwithanauto correla-
tionfunctiontodefinethestochasticmodel forboththe
Figure 6. A comparison between normal and lognormal
PDF, wherej
X
=100and
X
=10.
cohesion(c) andtangent of thefrictionangle(tan).
Thereisalsoapossibilityof assigningaNormal PDF
to the soil parameters, if there is a low probability
of generating values less than zero. However, given
anormal PDF thereis always apossibility of getting
negativevalues. A Normal PDF of arandomvariable,
X, isdefinedas
where j and are respectively mean and standard
deviation of its PDF, and x is a scalar value called
randomsample.
If therandomsamplex fromanormal randomvari-
ableX is replacedby ln(x), alognormal distribution
comesout.Theparametersof alognormal distribution
canbeobtainedfromthefollowingequation:
Where
and
A lognormal distribution yields always positive val-
ues. ThePDF of normal andlognormal distributions
aredepictedinFigure6for themeanvalueof 100and
standard deviation of 10. In fact, a lognormal PDF,
shownbydashedline, isanasymmetrical distribution,
andgeneratedrandomnumbersaredifferent withthe
Normal distribution. However, thereisnot alittledif-
ferenceforsmall ratiosof themeanvalueandstandard
deviation, j[x],[x].A goodcomparisonbetweenthe
normal andlognormal PDF isgivenin(Griffithsand
Fenton2007).
3.2 Autocorrelation function
MProstab describes thepattern of fluctuation within
a soil layer as a weak stationary. Weak stationary
means that for any two spatial points (x
1
, y
1
, z
1
) and
(x
2
, y
2
, z
2
), wherex andz aretwohorizontal andythe
427
vertical spatial coordinates, the(marginal) probabil-
ity distributions areidentical andthat thecorrelation
amongtherandomvariablesisonly afunctionof the
distancebetweenthesepoints.Theselectedautocorre-
lation function, whichexpressesthecorrelationamong
anytwopointsasafunctionof thedistancelags, isof
amodifiedGaussiantype:
Theautocorrelationfunctionis afunctionof thedis-
tance between two points in different directions of
x, y, andz as
D
h
andD
v
aretheso-calledcorrelation lengths, which
arerelated to thescales of fluctuation as introduced
by (Vanmarcke, 1977). Typical values of D
h
range
between25mand100mandvaluesof D
v
mayrange
between0.1mand3m(Calle1994). Theparameter
isavarianceparameter, whichequalstheratioof local
variance, i.e. thevarianceof fluctuationsrelativetothe
meanvaluealongavertical line, andthetotal variance,
whichisthevariancerelativetothemeanvaluetaken
over thewholedeposit space.
For =1 the autocorrelation function takes on the
classical Gaussian form, which is often suggested
in literature. It was found, however, that such type
may be inconsistent with actual measurements. The
parameter enables consistent modeling of thepres-
ence of overall weak and strong locations within a
layer. For fluctuationsof coneresistanceinclayeylay-
ers, values ranging between 0.5 and 1.0 havebeen
found(Calle1994).
Parameters of the probability distributions, i.e.
expected mean values and standard deviations, are
usually estimatedonthebasis of series of laboratory
or in-situtest results. Sincethenumber of samplesis
limited, theestimators arestatistically biased. A bias
estimatorisasourceof uncertainty, andMprostabsug-
geststotakethisuncertaintyintoaccountbyadjusting
thefieldvariancewithafactor (n+1)/nandmodifying
theautocorrelationfunctioninto:
If bothc
andtan areestimatedfromtest resultsof

thesameexperiment,e.g.triaxial test,theestimatesare
(negatively) correlated. Inthestochasticfieldmodel it
isassumedthatthefluctuationfieldsof thetwoparam-
etersarenot correlated. Correlationamongestimates
c
and tan fromthetest result of asample, results

incorrelatedestimatesof theexpectedmeanvaluesof
theseparameters. Thelarger thenumber of tests, the
smaller thecoefficientof correlationamongestimates
of expectedmeanvaluesof theparameters; correlation
canbetakenintoaccountinMProstab.Theassumption
of zero correlation, however, is slightly conservative
(Calle, 1990).
3.3 Bishop technique
Most of theproblems of slopestability aremodeled
on thebaseof physical equilibriumbetween param-
eters; yet they are statistically undetermined. As a
result, some simplifying assumptions are necessary,
and under various assumptions, different methods
have been developed. Some of the popular methods
areFellenius, Bishop, J anbu, andSpencer (Malkawi,
Hassanet al. 2000). Hereweconsider thesimplified
Bishopmodel whichsatisfiesonlytheoverall moment
and is applicable to a circular slip surface. For the
Bishopmodel, thefactorof safetyisdirectlyobtained;
thismethodassumesthattheinter sliceforcesarepar-
allel tothebaseof eachslice,thustheycanbeneglected
(Figure7).
On the base of the Bishop Method, a limit state
equation can bedefined as adifferencebetween the
total resistanceanddrivingforcesas
wheretheresistanceanddrivingmomentsarerespec-
tively called as M
r
and M
0
. Another stochastic
Figure7. This figureis aschematic representation of the
BishopMethod.Also, theconsideredforcesonasampleslice
areshown.
428
parameter which is implemented into the model is
called themodel factor, q which has expected value
j
q
; andthestandarddeviation
q
.Asaresult, thelimit
stateequationturnsto
Theresistancemoment isobtainedby
whereF is thesafety factor; u is thewater pressure;
cand arethesoil parameters; Wistheweightof the
slice; andtheother parametersareshowninFigure7.
Thedrivingmomentcomesfromthetotal weightof
eachsliceas
Consideringthestochasticvariablesasc, , andvari-
ation of the water level u and the model factor, the
first order estimationof thelimit stateequationat the
expectedpoint isobtainedas
whereZ
l
isarandomvariablewhichmaybeestimated
by its first and second moment which respectively
areE[Z
l
] and
2
[Z
l
]; and is thegradient, andT is
the transpose sign. Then, given the first and second
moment of thefirst order extensionof thelimit state
equation, thereliabilityindexcanbeobtainedas
whichleadstotheestimatedprobabilityof failureby
where + is the probability that the randomvariable
doesnotexceedadesignvalue, assumingthestandard
normal distribution,
N
and
3.4 Finite element approach
Finite Element analysis is a technique for solving
partial differential equations by discretizing of the
equation in the space dimension. This discretization
canbedonebydifferentelementshapes; but, withthe
finitenumber of elements. Then, amatrixcanbecon-
structed to relate the inputs to the outputs on some
specific pointswhicharecallednodes. Thenthesys-
temof equations can be solved and the inputs and
outputsarerelatedonthebaseof apartial differential
equation.
The finite element models of the cross sections
ALS166B and EMMA118 are made by Plaxis
2
and
presentedinFigures8to10.Theirgeometryisaccord-
ing to the Mprostab model presented in Figure 4,
andfurther informationof themodel is presentedin
Table1.Thegroundwaterflow,also,isassumedhydro-
statictoprovideacloser comparisonwiththeBishop
method.
Table1. Numbers, typeof elements, andintegrationspoints
usedinthefiniteelement analysis.
Type Typeof element Typeof integration Total no.
Soil 6-nodetriangle 3-point Gauss 410
Figure 8a. The finite element model of cross Section
ALS166B, depictedbyPlaxis.
Figure8b. Thefiniteelementmodel of SectionEMMA118,
depictedbyPlaxis.
Figure8c. Thefiniteelement model of SectionALS166B,
depictedbyPlaxis.
2
Plaxis, B. V. (2002). Plaxis, finiteelementcodefor soil and
rockanalyses. R. B. J. Brinkgreve. Delft, Netherlands.
429
Figure9. Threedifferent resultsfor thefailureof thecross
sectionALS166B. (a)ShowstheSlipCirclebyBishopModel
andandmodel factor 0.9. (b) Showstheslipcirclecalculated
byBishopModel andmodel factor1.00. (c)Showsthefailure
shapewhichiscalculatedbyfiniteelement analysis, Plaxis.
4 RESULTSANDDISCUSSION
Resultsarepresentedfor twotypical crosssectionsof
theDikeRing32whichwereaddressedinSection2.
Some differences are shown between the reliability
assessment of theBishopandfinit element method.
Thiscomparisonisimportantfromthepointof view
that many reliability programs which are related to
dikes and coastal structures use this technique. For
instance, PC_Ring which is developed to assess the
total safety of dikes intheNetherlands is fedby this
techniquefor theslidingmechanism(seeSection1).
Thefiniteelement methodpresentsmoreaccurate
results. Thetan c reduction techniqueis imple-
mentedintothefiniteelement analysis. Applyingthis
technique, thestrengthparameters of soil aregradu-
allyreduceduntill failureof themodel.Thisprovidesa
relativelyaccurateandmeaningful approachtofailure.
To compare different failure shapes in section
ALS166B, threeillustrations areshown in Figure9.
They present the Bishop results with and without
model factor together with the finite element out-
put. Thefigures show influenceof themodel factor
on the failure shape. There is a considerable differ-
encebetweenFigure9(a) andFigure9(b) as aresult
of the different model factors. As a matter of fact,
Figure9(a) doesnotcollapse, butithaspartial failure.
Therefore, it is possible to have a big difference in
Figure10. Threedifferentresultsforthefailureof thecross
section EMMA118. (a) Shows the Slip Circle by Bishop
Model andandmodel factor0.9. (b) Showstheslipcirclecal-
culatedbyBishopModel andmodel factor1.00.(c)Showsthe
failureshapewhichiscalculatedby finiteelement analysis,
Plaxis.
failure modes (slip circle) using this Method. This
fact needsmoreattentionwheretheloosefoundation
preventsthecircular failureshape.
However, when theshapeof failureis almost cir-
cular theresultsof finiteelementsandBishopModel
seemtobeinagoodcorrespondenceaspresentedin
Figure 10(a) and Figure 10(b). This kind of failure
isusually expectedwhentherearenot quitedifferent
materialswithquitedifferentstrengthpropertiesinthe
bodyandfoundationof dike.
5 CONCLUSION
Thefiniteelementmethodisanaccuratetechniquefor
failureassessmentof dikes.Thefiniteelementmethod
determines theactual water tableand water head of
different spotsinaflooddefencestructure.
Ontheotherhand,theBishopmethodisasimplified
methodwhichis widely usedinslopestability prob-
lems. It is less time consuming and easier to apply.
It needs, also, less parameter incomparisonwiththe
finiteelementmethod. It, therefore, canbeconsidered
for the first estimation of the probability of failure.
Nevertheless, it isshownthat theresult of thesimpli-
fied Bishop method in somecases may befar from
thereal failuremode, asshowninFigure9, or ingood
correspondencewithit, asshowninFigure10.
Theinfluenceof themodel factor onthesafety is
morecoherent to themodel itself if thewholelimit
statefunctionis multipliedby amodel factor. Other-
wise, if themodel factor ispartiallyimplementedinto
430
thelimitstateequationorwhenthelimitstateequation
needstobediscretized, theoutputsmaynotalwaysbe
closetothefiniteelement analysis. Figure9showsa
strongeffect of themodel factor ontheresults.
As a result, before assigning a model factor to a
model, we should pay attention to the failure shape
(slipcircle) whichneedstobeclosetoacircle, other-
wiseimplementationof amodel factor might leadto
unexpectedresults.
6 SYMBOLSANDABBREVIATIONS
j
X
Meanvalueof thevariableX witha
Normal PDF
X
Standarddeviationof theNor variableX
withaNormal PDF
j
lnX
Meanvalueof thevariableX with
aLognormal PDF
lnX
Standarddeviationof theNor variableX
withaLognormal PDF
f
X
PDF of thevariableX
+ Theprobabilityof exceedingthedesign
point inastandardnormal PDF
B Thereliabilityindex
q Model Factor
Frictionangle
D
h
Horizontal correlationlength
D
v
Vertical correlationlength
LSE Limit stateequation
PDF ProbabilityDistributionFunction
(
x
,
y
,
z
) Autocorrelationfunction
r(
x
,
y
,
z
) Modifiedautocorrelationfunction
REFERENCES
Allsop, W. (2007). Failure Mechanisms for Flood Defence
Assets. Floodsite report tasks. Floodsite.
Calle, E. O. F. (1994). MPROSTABusersguide.Geodelft,
Delft.
Griffiths, D. V. andG. A. Fenton(2007). Probabilistic meth-
ods in geotechnical engineering, Springer, Wien, New
York.
Malkawi, A. I. H., W. F. Hassan, et al. (2000). Uncer-
tainty and reliability analysis applied to slopestability.
Structural Safety 22(2): 161187.
Plaxis, B. V. (2002). Plaxis, finiteelement codefor soil and
rockanalyses. R. B. J. Brinkgreve. Delft, Netherlands.
Rajabalinejad, M., W. Kanning, et al. (2007). Probabilistic
assessment of the flood wall at 17th street Canal, New
Orleans. Risk, Reliability and Societal Safety, Vols 13:
22272234.
Rajabalinejad, M. and P. H. A. J. M. van Gelder (2007).
Geotechnical applications of the probabilistic methods
FloodSite, Task 7, TUDelft.
Rajabalinejad, M. (2009). Reliability Methods for Finite
Element Models.Amsterdam, theNetherlands, IOSPress.
Rajabalinejad, M., P. van Gelder, et al. (2008). Improved
Dynamic Limit Bounds in Monte Carlo Simulations. 49th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural
Dynamics, and Materials Conference, Schaumburg, IL,
USA,AmericanInstituteofAeronauticsandAstronautics.
Rajabalinejad, M., P. H. A. J. M. van Gelder, et al.
(2008). Probabilistic Finite Elements with Dynamic Limit
Bounds: A Case Study: 17th Street Flood Wall, New
Orleans. 6thInternational ConferenceonCaseHistories
in Geotechnical Engineering and Symposiumin Honor
of Professor J ames K. Mitchell, Rolla, Missouri USA,
Missouri Universityof ScienceandTechnology.
Rajabalinejad, M., P. vanGelder, etal. (2008).Application of
BayesianInterpolationinMonte CarloSimulation.Safety,
Reliability andRiskAnalysis(ESREL), Valencia, Spain,
Taylor andFrancisGroup, London, UK.
Van Gelder, P. (2008). Reliability Analysis of Flood Sea
DefenceStructures andSystems. Floodsite task reports.
Floodsite, Floodsite.
Vanmarcke,E.(1983).Randomfields, analysis and synthesis,
MIT PressCambridge, Mass.
www.floodsite.net(2004-2008). IntegratedFloodRiskAnal-
ysisandManagement Methodologies.
431
A report by JGS chapter of TC23 limit state design in
geotechnical engineering practice
Codecalibrationinreliabilitybaseddesignlevel I verification
format for geotechnical structures
Y. Honjo, T.C. KieuLe&T. Hara
M. Shirato
Public Works Research Institute, Tsukuba, Japan
M. Suzuki
Research Institute, Shimizu Corporation, Tokyo, Japan
Y. Kikuchi
Port and Airport Technical Research Institute, Yokosuka, Japan
ABSTRACT: This report is prepared by a working group in J GS chapter of TC23 Limit State Design in
Geotechnical Engineering Practice andsummarizes theimportant points to noteandrecommendations when
newdesignverificationformulasaredevelopedbasedonLevel I reliabilitybaseddesign(RBD) format.
Thereareseveral differenttypesof verificationformatinRBDLevel I suchaspartial factormethod(PFM),load
andresistancefactor design(LRFD), etc. LRFDformatisrecommendedinthisreportasthemostsuitableveri-
ficationformatatpresentbasedonthefollowingreasons: (1) a designer can trace the most likely behavior of the
structure to the very last stage of design if thecharacteristicvaluesof theresistancebasicvariablesareappropri-
atelychosen,(2)theformatcanaccommodateaperformancefunctionwithhighnon-linearityandwithsomebasic
variablestobothforceandresistanceside, and(3) codecalibrationispossiblefor thecaseonlytotal uncertainty
of thedesignmethodisknownandthetotal uncertaintycannot bedecomposedtoeachsourceof uncertainty.
It is recommended that thedesign valuemethod (DVM) should beadopted as thebasic approach for the
codecalibration. Thereasons for this recommendationareas follows: (1) LoadandResistancefactors canbe
determinedbasedonasoundtheoretical framework, suchastarget reliabilitylevel, COV of basicvariablesand
sensitivityfactors, (2) redesignof astructureisnotrequiredif theinitiallydesignedstructurehasreliabilitylevel
thatisnotveryfarfromthetarget, (3)byreferringtothesensitivityfactors, onecanevaluatecontributionsof each
loadandresistancecomponent. Ontheother hand, DVMbasedonFORMoftenhasproblemssuchasinstability
andnon-uniquenessof convergencewhentheperformancefunctionsbecomecomplexandhighlynon-linear. Itis
recommendedinthisreporttoemployMonteCarlosimulation(MCS) methodwhentheperformancefunctions
arecomplex andnon-linear. MCS iseasy toimplement for any complex performancefunction, andasolution
canalwaysbeobtainedif sufficientcomputational effortisput.A methodisproposedinthisreporttodetermine
factors based on DVM approach by MCS. Furthermore, amethod that applies thesubset MCMC method is
suggestedtoimprovetheefficiencyof calculationincarryingout thisprocedure.
Somerelatedissuesconcerningcodecalibrationarereferredto. Thesearedefinitionof characteristicvalues
of basicvariables, thebaselinetechniqueanddeterminationof appropriatereliabilitylevels.
Itisemphasizedthatameanvaluethatmayhavetakenintoaccountthestatistical uncertaintyshouldbeused
asacharacteristicvaluefor resistancesidebasicvariables. It isrelatedtothephilosophythat a designer should
trace the most likely behavior of the structure to the very last stage of design as much as possible.
Thebaselinetechniqueisexaminedby somesimplecalculationtoshowthat takinghigh/lowfractilevalues
havesomemerit for loadsidebut themerit is not so obvious for theresistanceside, thus takinglowfractile
valuesfor thecharacteristicvalueof theresistancesidebasicvariablesisnot necessary.
Someinformationontargetreliabilityindexandbackgroundrisksareprovidedforconvenienceof thereaders.
1 INTRODUCTION
1.1 Background and purpose of the report
Manydesigncodesintheworldhavenowbeenrevised
toRBDLevel I format. Thelimit statedesign(LSD),
PFM and LRFD are all included in this category
of design verification format. The factors such as
WTO/TBTagreement,nearfutureestablishmentof the
Structural Eurocodes, ISO documents (e.g. ISO2394
435
(ISO, 1998)) are all taking part in this movement
(Honjo, 2006).
Some of the examples of such design codes in
geotechnical engineering are Eurocodes 7 (CEN,
2004), Ontario Highway Bridge Code (MOT, 1979,
1983, 1991), theCanadianBridgeDesignCode(CSA,
2000; Becker, 2006), AASHTOLRFDBridgeDesign
Specifications (AASHTO, 1994, 1998, 2004; Allen,
2005; Paikowsky, 2004) etc. In J apan, AIJ guideline
for loads on buildings (AIJ, 1993, 2004), AIJ guide-
line for limit state design of buildings (AIJ, 2002),
J GSGeocode21(J GS, 2006) andTechnical standards
for port andharbor facilities(J PHA, 2007) havebeen
developedbasedonthisapproach.
Thisreport ispreparedbyaworkinggroupinJ GS
chapter of TC23 Limit State Design in Geotechnical
Engineering Practice and summarizes theimportant
pointstonoteandrecommendationswhennewdesign
verificationformulasaredevelopedbasedonLevel I
reliabilitybaseddesign(RBD)formatforgeotechnical
structures.
1.2 Reliability based design
1.2.1 Design methods
A designmethodisdefinedasa method or procedure
to make decision under uncertainties encountered in
a structural design process (Matsuo, 1984). Themost
traditional, popular and widely used design method
isthesafetyfactor method. Thismethodintroducesa
safetyfactor whichisdefinedasaratiobetweentotal
resistanceandtotal force, andshouldusually bekept
avaluemorethanonegivenbyexperience.
The safety factor method is a superior design
methodwhichissupportedbyavast amount of expe-
riences. However thismethodhassomedrawbacksas
follows:
(1) Thesafetyfactorisnotanabsolutemeasureof the
safetyof astructure.A safetyfactorisdetermined
basedonthepastexperiences, thusonecannotsay
astructurebuiltonsafetyfactortwiceasbigasthe
oneusuallyemployeddoesnot necessaryimplies
twiceassafe.
(2) Oneof thereasonsthesafetyfactorisnotanabso-
lutemeasureof safety comes fromthefact that
thismethodhasbeendevelopedhandinhandwith
theallowablestress design(ASD) method. ASD
methoddoesnot necessarydesignstructurefor a
limit state.
As the progress of plastic theory in mechanics,
which directly takes into account the failure of a
structureor amember, thelimit statedesign treated
resistance and forces as randomvariables fromthe
beginning. This methodhas beenestablishedas reli-
ability based design (RBD) method today. A brief
summary of development of RBD can be described
asbelow.
So called the classic RBD method, which is the
directoriginof theRBDtodayandisbaseddirectlyon
theprobability theory, startedin1940sby Fredenthal
Table1. Classificationof RBD.
Reliability
Level Basicvariables evaluation Verification
Random Failure Acceptable
variable probability level of
withfull reliability
distribution. Economical
optimization
etc.
L
e
v
e
l
I
I
I
Random Reliability Target
variables index reliability
withmean, index
varianceand
covariance
L
e
v
e
l
I
I
Deterministic Partial factors, Verification
variables Load, resistance formula
factors
L
e
v
e
l
I
I
andothersinthewesternworld(Takaoka,1988).How-
ever, thisclassictheoryhadthedrawbacksasfollows
(Matsuo, 1984):
(1) Inmodelingtheuncertaintiesof forcesandresis-
tances, it isthetail part of thedistributionthat is
reallyaffectingthereliabilityof astructure. How-
ever, it is very difficult to accurately model the
tail part fromlimiteddata.
(2) The calculation of failure probability by proba-
bility theory requires multipleintegrations. This
calculationbecomesactuallyimpossiblewhenthe
number of randomvariablesincreasestoseveral,
whichisverytypical inactual designcalculation.
Thusthesolutioncannot beobtained.
(3) The design calculation method is a simplifica-
tion and idealization of real phenomena. It is
rarely possibletoevaluatethemodel uncertainty
involved in this kind of method. Thus, the cal-
culatedfailureprobability cannot beanaccurate
indicator of thefailureevent.
One of the proposals to overcome the drawback
(2) wasFOSM (First Order SecondMoment) method
by Cornell (1969). In this method, the uncertainties
areonlytreatedparametricallybymean, varianceand
covarianceof randomvariables. Thenon-linear equa-
tionsarelinearizedbyusingTaylorexpansion, Cornell
(1969) alsointroducedthereliabilityindex, , inplace
of failureprobabilityP
f
.
Theconcept proposedby Cornell (1969) wascon-
tinuously developedandrevisedby other researchers
suchasDitlevesen,Rackwitz,LindandHasofer.These
effortsleadtoFORM(FirstOrderReliabilityMethod)
whichisthestandardmethodinRBDtoday.
RBD is usually classified into three levels as
presentedinTable1:
Level III: Basic variables are treated as random
variables with full distributions. The failure proba-
bility is evaluated based on a performance function
definedinthebasicvariablespaceandseparatestable
regionfromfailureregion.
436
Level II: A simplified version of Level III. Basic
variablesareparametrically describedby mean, vari-
anceandcovariance.A methodsuchasFORMisused
toevaluatethereliabilityindex.
Level I: Necessary safety margin is preserved by
applyingpartial factorstocharacteristicvaluesof basic
variables directly (i.e. PFM) or to calculated forces
and resistances (i.e. LRFD). All thecalculations are
done deterministically, and a designer needs not to
haveknowledgein reliability theory. However, code
writers needto determinethefactors basedonrelia-
bility analysis. Themethodis sometimes calledlimit
state design (LSD) and many design codes are now
beingrevisedtothisformat.
1.2.2 Design methods and design codes
A designcodeisaguidelinedocument that describes
procedureof designforgeneral androutinelydesigned
structures so that certain level of technical quality
is assured (Ovesen, 1989, 1993). Therefore, most of
the structures we see daily are designed and built
accordingtodesigncodes, andthisfactreflectsusthe
importanceof designcodes.
Asexplainedintheprevioussection, many design
codesintheworldtodayareintheprocessof revision
fromtraditional ASDtoLSDorLevel I RBD. Someof
thefeaturesof Level I RBDcanbelistedasfollows:
(1) Reliability of astructureis evaluatedbasedonso
calledalimitstateof astructure,acriterionthatsep-
arates desirablestate(e.g. stablecondition) from
undesirable state (e.g. failure). The ultimate and
theserviceability limit states areexamples of the
typical limit statesemployedinRBD.
(2) Theuncertainties involvedinbasic variables con-
cerningactions,material properties,shapeandsize,
etc. aretakencarebyfactorswhicharedetermined
based on reliability analyses. However, general
usersof thesefactorsarenotrequiredtounderstand
thedetailsof thereliabilityanalyses.
(3) The harmonization of design codes is possible
underidentical concept.Traditionally, designcodes
are developed in different ways among different
structures (e.g. buildings vs. bridges), and differ-
entmaterials(e.g. steel, concrete, composite, wood
andgeotechnical).
ISO documents such as ISO2394 General prin-
ciples on reliability of structures (ISO, 1998) plays
important role in the code development because
WTO/TBT agreement requiresfor themember coun-
triestorespect theinternationallyagreedstandards.
Althoughmost of theimportant partsof RBDhave
been developed by structural engineers as explained
in the previous section, geotechnical RBD also has
some original developments (Meyerhof, 1993). It is
natural for engineers to invent methods to encounter
uncertaintieswhiletheycarryout designintheir own
ways.
Themost significant contribution of geotechnical
RBD camefromBrinchHansen(1953, 1956, 1967).
He was the first person who used limit states in
geotechnical engineering and proposed to introduce
partial safety factors into design codes. Under his
influence, a geotechnical design code using partial
safety factors has beenestablishedinDenmark since
1965. Thus, LSD became much popular in Scandi-
navian countries in early years, which added some
differentbackgroundtoEurocode7comparedtoother
partsof theStructural Eurocodes.
One of the most well known structural reliabil-
ity scholars, Ove Ditlevsen, states in his structural
reliabilitytextbookthat
A consistentcodeformulationof adetailedpar-
tial safety factor principle was started in the
1950s in Denmark beforeother places in the
world. Thisdevelopmentgotparticular support
fromthe considerations of J. Brinch Hansen
who applied theprinciples in thefield of soil
mechanics. (DitlevsenandMadsen, 1996, p.31)
Itisunderstoodthatcontributionof BrinchHansen
hasbeenrecognizedwidely.
Oneof newtrendsconcerningdesigncodesisuprise
of PerformancebasedDesign(PBD) concept, which
requires transparency and accountability in design
(Honjo& Kusakabe, 2002; Honjoet al., 2005). RBD
(or LSD) seemstobetheonlyrational tool toprovide
adesign verification procedurethat designs astruc-
turefor clearlydefinedlimit states(i.e. performances
of structures andmembers) andintroduces sufficient
safety margin. It is concludedthat RBD will beused
asatool todevelopdesigncodesat least for thenext
several decades.
2 FORMAT OF VERIFICATIONFORMULA
2.1 PFM vs. LRFD
The definitions of partial factor method (PFM) and
loadandresistancefactordesign(LRFD) inthisreport
aregiveninthefollowingsubsections:
2.1.1 PFM
Thebasicphilosophyof thisformatistotreattheuncer-
tainties at their origin. Thus many partial factors are
introducedinthisformat, whichcanbewrittenas:
where,
ri
: partial factormultiplyingtobasicvariables
x
ri
inresistancesideof theequation;
si
: partial fac-
tor multiplyingto basic variablex
si
inloadside.
Ri
and
Si
, another important group of partial factors
needtobeintroducedtocover designmodelingerror,
redundancy, brittleness, andstatistical uncertainties.
Itshouldbenoticedthat, partial factorsformaterial
propertiesareintroducedtoencounteragainstmaterial
uncertainty, andacharacteristicvalueof amaterial is
usually discountedby amaterial partial factor before
it is inserted to a design equation to calculate the
resistance.
437
2.1.2 LRFD
Theresistanceandtheload(external action) arefirst
calculated based on the characteristic values of the
basic variables, then load and resistance factors are
appliedtoresultingresistanceandloadcomponentsto
assureappropriatesafetymargin.
It may be convenient to list two types of LRFD
format asbelow:
where,
R
: aresistancefactor and
Sj
: aloadfactor for
aloadcomponent S
j
where,
Ri
: resistance factors for resistance compo-
nentsR
i
and
Sj
: loadfactorsfor loadcomponentsS
j
.
InEq. (3), theresistancesidemaybedecomposed
into several lump sums that can besuperimposed to
obtain thefinal resistance(for example, piletip and
sideresistances). Different resistancefactors may be
appliedtothedifferent terms. Thisapproachissome-
timescalledMLRFD MultipleLoadandResistance
Factor Design (Phoon et al. 1995, Kulhawy and
Phoon, 2002).
Traditionally, PFM is developed mainly in
Europe(Gulvanessian and Holicky, 1996), whereas
LRFDintheNorthAmerica(Ellingwoodet al., 1980,
1982). However, they are mixedly used world wide
now. In Eurocode 7 (CEN, 2004) and Geocode 21
(J GS, 2006), Material Factor Approach (MFA) and
Resistance Factor Approach (RFA) are respectively
usedinalmost thesamewayasPFM andLRFDhere.
2.2 Pros and cones of PFM
PFM has advantages over LRFD on the following
aspects:
It isintuitively felt reasonabletoencounter uncer-
taintiesat their sources.
It is easier to accommodate the development of
construction technique and design method when
the uncertainties are treated at their origin: one
cansimplychangethepartial factorsrelatedtothe
improvement bythenewdevelopments.
However, it is very difficult to evaluate the reli-
ability of structure by reliability analysis based on
superpositionof all uncertainsourcesforthefollowing
reasons.
Not all sources of uncertainties areknown quan-
titatively. Especially the model uncertainties are
difficult toevaluate.
Someof thesourcesarecorrelatedanddifficult to
consider correlationinthereliabilityanalysis.
Sometimesoverall uncertaintyof thestructurecan
beestimated. However, itisdifficulttobreakdown
theresult intoeachsource.
Many of thefactorshavenon-linear effectsonthe
resulting uncertainty of the structural reliability.
Thus it is difficult to control thereliability of the
structureat thesourcesof uncertainties.
Since PFM modifies the material property that
needs to beinput to thedesigncalculationbefore
the calculation, the actual behavior of structure
based on this calculation can be far from real-
ity (or themost likely behavior). It is considered
that, especially in geotechnical design where the
engineering judgment plays very important role,
suchunrealistic calculationmay not befavorable.
Ingeotechnical design, thereshouldbeaphiloso-
phy that a designer should keep track of the most
likely behavior of the structure toward the end of
the design calculation as much as possible.
As could be suggested by the above discussion,
althoughPFMlookstheoreticallysound, itencounters
manydifficultiesinpractical situations. Itisverydiffi-
cult, if not impossible, torigorouslydeterminepartial
factorsinpractice.
2.3 Pros and cons of LRFD
Pros of PFM arecons of LRFD. However, LRFD is
superior toPFM inthefollowingpoints:
Especially for theresistanceside, thedesign cal-
culation based on LRFD predicts themost likely
behavior of thestructuretoitslast stageof design,
andit isincoincidencewiththephilosophy that a
designer should keep track of the most likely behav-
ior of the structure to the last stage of the design as
much as possible.
In geotechnical design, where the interaction
between a structure and ground is so high, it is
not possibleto know whether thediscount of the
material parameter valuemayresult safedesignof
astructure. It isespecially truefor asophisticated
designcalculationmethodlikeFEM.
In many codecalibration situations, uncertainties
of adesignmethodareprovidedasresultsof many
loadingtests, for exampleintheformof database.
Thegivenuncertaintiesincludeall aspectsof uncer-
tainties, and it is impossible to decompose to
various sources. Therefore, in practice, it is pos-
sibletocalibratecodesbasedonLRFDbut not on
PFM.
LRFD has a design verification formula that is
morecloseto thetraditional safety factor method
thanPFM. It maybeeasier for practical engineers
tobecomefamiliar withLRFDthanPFM.
For thesereasons, LRFDisconsideredtobebetter
format thanPFM ingeotechnical engineeringdesign
codeat least inthepresent situation.
3 DESIGNVALUE METHOD
3.1 General consideration on code calibrations
An engineer can design structures by using Level I
RBDformatwithoutknowingthedetailsof probability
438
Figure1. Illustrationof designvaluemethod.
andstatistical knowledge. Ontheother hand, acode
writer hastodeterminethefactorstointroducesuffi-
cientsafetymargintothedesignedstructuresbasedon
detail knowledgeontreatmentof uncertainties, which
isknownascodecalibration.
Thebasicissuesof codecalibrationcanbelistedas
follows:
Thereareinfinitenumbersof combinationsof fac-
tor values in design verification to give required
reliabilitylevel toastructure. Mathematically, this
problemis ill-posed(meaningno uniquesolution
exists).
Even for same category of structure, the design
conditionsvaryfromonestructuretoanother. The
factor valuesmay dependontheseconditionsand
maynot besamefromonetoanother.
For thesereasons, Level I RBDmaynot belooked
atasanultimatedesignmethodtointroducesufficient
reliabilitytoastructure.Thereshouldbealwaysaway
open to carry out reliability design of astructureby
moredirect methodsuchas Levels II or III RBD. In
fact, itisrecommendedtodesignimportantstructures
bydirectlyemployinghigher level RBDmethods.
3.2 Design value method
DesignValueMethod(DVM) isoneof theconvenient
methodstodeterminethefactorsfor Level I Reliabil-
ity Based Design Method to overcome the ill-posed
problempointedout above. Thebasicideaof DVM is
that all thefactoreditems, i.e. basicvariables, will be
assigned thevalues at thedesign point based on the
relationship with thecharacteristic values. Here, the
design point is defined as a point on the limit state
line which has the maximum likelihood value in the
basic variable space (Figure1).
Hasofer andLind(1974) hasproposedtocalculate
reliability index asaninvariant, wherethebasicvari-
ablespaceis transformedto astandardizedspace. In
this standardizedspace, thedesignpoint canbepre-
sentedasaclosetpointfromtheoriginonthelimitstate
line(Figure2). For thisgraphical representationof ,
Figure2. Definitionof designpointandsensitivityfactors.
their reliability index is sometimes called geometric
reliabilityindex(Ditlevsen& Madsen, 1996).
The DVM can be most visually and intuitively
explainedwhentheperformancefunctionconsistsof
alinear combinationof twoindependent normal ran-
domvariables,namelytheresistanceRandtheexternal
forceS. Theother typical caseisthattheperformance
functionconsists of alinear combinationof lnR and
lnS, whereR and S independently followlognormal
distributions. These two cases will be discussed in
detail inthefollowingsubsections.
3.2.1 Linear performance function with Two
Independently Normally Distributed Basic
Variables R and S
Mean and standard deviation of safety margin M,
are j
M
=j
R
j
S
and
M
=
2
R
+
2
S
. Thus, the
reliabilityindex isgivenby
Itisrequiredindesignthatthereliabilityindex must
bealwayslargerthanatargetreliabilityindex,
T
, that
ischosenbeforehand, i.e.,
T
.
Also, it isdefinedthat,
439
where,
R
=

R
2
R
+
2
S
and
S
=

S
2
R
+
2
S
aretermed
sensitivityfactorsof R andS.Then, thefollowingscan
beobtained:
It is obvious that (j
R
+
T
R
) and(j
S
+
T
S
)
aredesign values. Let characteristic values of R and
S beR andS, respectively. Thepartial factors which
can fulfill the target reliability index,
T
, can be
determinedasfollows:
where,
R
and
S
satisfy:
Thedefinitionof designpoint andsensitivity factors
areillustratedinFigure2.
3.2.2 Linear performance function, with
independently lognormally distributed basic
variables
If X isanindependentlylognormallydistributedvari-
ableswithmeanE[X] andvarianceVar[X], thenlnX
will be an independently normally distributed vari-
ableswithmeanE[lnX] andvarianceVar[lnX]. Mean
andvarianceof M aregivenby:
Ontheother hand,
Inaddition,
wherethesensitivityfactors
R
and
S
aredefinedas:
For =
j
M
M

T
, thefollowingcanbeobtained:
SubstituteEq.(11) intoEq.(12), therefore:
Finally, thepartial factorscanbedefinedby:
Thetarget reliabilitylevel canbeguaranteedby:
R
R
S
S
3.2.3 General solution of design value method
General solutionof theDesignValueMethodincludes
many basic variables (multivariable) with a nonlin-
ear limit state function. Let X =X
i
, i =1, . . . , n be
avector of n basic randomvariables. Thebasic ran-
domvariables X
i
, (i =1, . . . , n) followadistribution
functionf
X
(X).
ThesafetymarginM of n randomvariablesinthis
case can be expressed as M =g(X)=R(X)S(X).
Thesolution of theDVM is obtained by solving the
equations:
A general codecalibrationprocedurebasedonDVM
using Monte Carlo Simulation will be proposed in
section5.
4 SOME ISSUESRELATEDTOCODE
CALIBRATIONS
4.1 Characteristic Values
4.1.1 Introduction
When using PFM and LRFD of the Level I RBD
verificationfor designacertainstructure, oneof the
important problemsthat needstobecarefullyconsid-
ered is howthecharacteristic values of material and
loadshouldbedetermined. Ingeneral, thecharacter-
istic valueof thematerialssuchassteel andconcrete
whichhavealowdispersioninphysical propertiesis
usuallychosenbyfractilevalues(usually1%or 5%).
If afractilevalueisalsousedascharacteristicvalueof
soil material, it will beconsiderablysmall comparing
tothemeanvalue.
440
Inadesigncalculation, if theoutputvaluespropor-
tionallychangewiththereductionof theinput values
(i.e. a linear model), whatever safety margin intro-
ducedintheinput valueisproportionally propagated
totheoutput. Indesignof foundationstructures, how-
ever, calculation of soil stress, bearing capacity etc.,
followhighlynon-linear models.
Furthermore, Finite Element Method and other
numerical calculationmethodshavebeenusedbroadly
anddirectly ingeotechnical design. Inthesecases, if
thecharacteristicvalueof amaterial usedfor designis
considerablydifferentfromitsmeanvalue, thebehav-
ior of thedesignedstructurewill befar fromthemost
likely behavior of the real structure.Also, considering
the complex interaction among basic variables, it is
impossible to judge whether reduction of the mate-
rial parameters would preserve safety of the whole
structure.
Simpson and Driscoll (1998) mentioned in the
Commentary of Eurocode 7 that characteristic val-
uesof geotechnical parametersarefundamental toall
calculations carriedout inaccordancewiththecode.
Their definitionhasbeenthemost controversial topic
inthewholeprocessof draftingEurocode7.
Thecharacteristic valueis usually equal to acer-
tainpercentileof thedistributionof thatquantity(e.g.,
mean, mode, median, meanminusonestandarddevia-
tion, fractilesetc.). Whitman(1984) showedthatsome
engineers usethemean value, whileothers used the
most conservative of the measure strengths for the
characteristicvalue.
According to theEurocode0, thenominal values
for resistancegiven by producers should correspond
(atleastapproximately) tocertainfractiles. For exam-
ple, for steel structures, thevaluesfor resistanceside
should correspond to the5%-fractile. In thecaseof
action side, the characteristic values are defined as
the50%-fractile(meanvalue) for permanent actions;
98%-fractileof thedistributionof theannual extremes,
whichmeansanaveragereturnperiodof 50yearsfor
variableactions(e.g. climaticactions).
4.1.2 Characteristic value in Eurocode 7
Numerousdebatesonchoiceof theappropriatechar-
acteristic value for geotechnical design have been
donerelatedtothedevelopmentof Eurocode7. These
methods can be listed such as statistical method in
Eurocode 7, method proposed by Schneider (1997),
Bayesianapproach, etc. However, theselectionof the
characteristic value is still open for debate between
engineers. InEurocode7, thecharacteristic valuesof
loadsandresistancearechosenasexplainedindetailed
byOrr (2000).
(1) Characteristic value of loads
Thecharacteristic values of permanent loads derived
fromtheweights of materials, including water pres-
sures, arenormallyselectedusingaverageor nominal
unit weightsfor thematerials, withnoaccount being
taken of the variability in unit weight. Characteris-
tic earth pressures are obtained using characteristic
Figure 3. Process for obtaining design values from test
results, Orr & Farrell (1999).
groundpropertiesandsurfaceloadsandincludechar-
acteristicwater pressures. Thecharacteristicvaluesof
variable actions, for example wind and snow loads,
are either specified values or values obtained from
meteorological recordsfor theareaconcerned.
(2) Characteristic values of geotechnical parameters
Inarticle2.4.3(2,3,4,5)of Eurocode7,itissaidthatthe
characteristic value of geotechnical material parame-
tersarebasedonanassessmentof thematerial actually
inthegroundandthewaythat material will affect the
performanceof thegroundandstructureinrelationto
aparticular limit state. Thecharacteristic valueof a
soil androckparameter shall beselectedasacautious
estimateof thevalueaffectingtheoccurrenceof the
limit state.
Eurocode 7 proposed the process to obtain the
characteristicvaluesof geotechnical parametersfrom
theresults of fieldandlaboratory tests. This process
involves thethreestages summarized by Orr & Far-
rell (1999) as shown in Figure 3, during which the
followingvaluesareobtained:
1. Measuredvalues
2. Derivedvalueof aparameter
3. Characteristicvalueof aparameter.
A measuredvalueissimplydefinedinEurocode7,
Part 2as thevaluemeasuredinafieldor laboratory
test.A derivedvalueisthevalueof agroundparameter
at oneparticular locationinthegroundwithout con-
siderationof thenatureof thestructureandobtainedby
theory, correlationor empiricismfrommeasuredtest
results. Orr (2000) also emphasized that thederived
valuesof thesameparametermightvarywithdifferent
typesof testsused.
(3) Statistical methods for determining the
characteristic value
The characteristic value of ground property in
Eurocode7isdefinedasthevaluesuchthat theprob-
abilityof aworst valuegoverningtheoccurrenceof a
441
limitstateisnotgreater than5%, i.e. 95%confidence
level that the actual mean value is greater than the
selectedcharacteristicvalue. Eurocode7alsoempha-
size that the statistical method should be compared
totheresultsobtainedbyBayesianmethodinsteadof
using only pure statistical approach without consid-
erationontheactual designsituationandcomparable
experience.
The characteristic value, X of a soil property in
statistical methodisgivenby:
where j
X
is mean value, kn is a factor, depending
onthetypeof statistical distributionandthenumber
of test results, andV
X
is thecoefficient of variation.
Sincetheactual mean value, j
X
of asoil parameter
cannot normally bedeterminedstatistically by asuf-
ficient number of tests, it must beassessedfromthe
averagevalue,

j
X
of thetest results.
Onewaytoobtainthefactork
n
intheaboveequation
is usingpseudonymof Student proposedby Gossett.
Orr (2000) also noticed that the characteristic value
obtained by statistical method using Student t value
will becometoo cautious anduneconomic incaseof
limited number of test results which is common in
practice.
(4) Schneiders method for determining the
Basedoncomparativecalculations, Schneider (1997)
hasshownthatagoodapproximationtoX isobtained
whenk
n
=0.5, i.e. if thecharacteristicvalueischosen
asonehalf astandarddeviationbelowthemeanvalue
asinthefollowingequation:
wherethemean, standarddeviationandcoefficientof
variationareobtainedfromtest results.
(5) Bayesian approach for determining the
This approach may be used to determine the char-
acteristic valuewhensomecomparableexperienceis
availablethatenablesthemeanandstandarddeviation
valuestobeestimated. Inthissituation, itispossibleto
combinethetestresultswiththeestimatedvalues, i.e.
apriori values, usingBayes theoremso as to obtain
morereliablemeanandstandarddeviationvaluesand
henceamorereliablecharacteristicvalue.Therelevant
equations for themeanandstandarddeviationusing
theBayesianapproachare(Tang, 1971):
Figure4. Flowchart for determiningthedesignvalues of
thegeotechnical parameters.
whereX
1
,
1
aretheestimatedmeanandstandarddevi-
ationvaluesbasedonexperience; X
2
,
2
arethemean
and standard deviation values obtained fromthetest
results; andX
3
,
3
aretheupdatedmeanandstandard
deviationvalues.
4.1.3 Characteristic value in Geocode 21
InthePrinciples for Foundation Design Grounded on
Performance Based Design Concept (nicknameGeo-
code 21 )(J GS4001,2004),thecharacteristicvaluesof
soil parametersarerecommendedtobedefinedbased
onthephilosophyexplainedbelow.
In Geo-code 21, the definitions of characteristic
valueanddesignvalueareshowninFigure4andstated
asbelow:
Measured Value isthevalueobtainedfromvariety
of investigationortesting. Forexample: groundwa-
ter level, SPT N value, stressandstrainintriaxial
test.
Derived Value is estimated fromMeasuredValue
after considerationonsomecertaintheories, expe-
riences, andcorrelation. For example: thecohesion
andinternal frictionangleobtainedfromMohr cir-
cleof triaxial test,Youngsmodulusobtainedfrom
SPT Nvalue, andsoon.
Characteristic Value is arepresentativevalueof a
soil parameter that ismost appropriatelyestimated
forpredictingthelimitstateof afoundation/ground
model in design. Deciding a characteristic value
must be based on theories and experiences, as
442
well as sufficiently consider thedispersionof soil
parameters andtheapplicability of thesimplified
model.
Geo-code21definedthecharacteristic valueas
follow:
The characteristic value of a geotechnical
parameter is principally thought of as the aver-
age (expected value) of the derived values. It is not
a mere mathematical average, but also accounts
for the estimation errors associated with statistical
averaging. Moreover, the average value shall be
carefully and comprehensively chosen considering
data from past geologic/geotechnical engineering,
experiences with similar projects, and relation-
ships and consistencies among the results of differ-
ent geotechnical investigations and tests. (Design
principle2.4.3(c))
Design Value isthevalueof foundationparameter
that isusedindesigncalculationmodel incaseof
material factor approach and can be obtained by
applyingpartial factor tocharacteristicvalue.
4.1.4 Various discussion on characteristic value
In Phoon et al. (2003), it is stated that froma reli-
ability perspective, the key consideration is that the
engineershouldnotbeallowedtointroduceadditional
conservatisminto the design by using, for example,
somelower bound value, becausetheuncertainty in
thedesign parameters already is built rationally into
theRBDequations. Phoonet al. (2003) alsorealized
thatitisveryconventional tochoosecharacteristicval-
uesof loadsandresistancesfor veryhigh/lowfractile
values, suchas95%, 5%andthereasonfortakingsuch
values arebasedonso calledtheBaseline Technique
that thevaluesdeterminedbasedonthesefractileval-
uesarelesssensitivetothechangeof uncertaintiesof
basic variableusedinthecodecalibration. However,
thisisnolongertruewhenCOV exceeds0.3.Thiswill
beexplainedindetail inthenext section.
4.2 Baseline Technique
It is very conventional to choose characteristic val-
ues of loads, S, andresistances, R, for very high/low
fractile values, such as 95%/5%. It is said that one
of thereasonsfor takingsuchvaluesarebasedonso
calledBaseline Technique (e.g., Phoonet al., 2003).
TheBaseline Technique says that if S andR arecho-
senashigher andlower fractilevaluesrespectivelyto
defineloadandresistancefactors, thenloadandresis-
tancefactors, i.e.
S
and
R
, arerelativelyinsensitive
(or robust) to changeof coefficient of variationof S
andR, i.e. V
S
andV
R
.
Inthis section, theeffectiveness of Baseline Tech-
nique inchoosingtheloadandresistantfractilevalues
whichareusedfordeterminationof loadandresistance
factors will be discussed. Namely, for each fractile
valueof Sand R, load and resistancefactors will be
calculatedwhilechangingcoefficientof variationand
fractilevaluesof S andR.
Suppose a limit state function is given based on
two statistically independent basic variables R andS
asfollows:
whereR is resistancesidewhereas S is loadingside.
RandS followNormal distributionfunctionwithmean
andstandarddeviationof j
R
and
R
for R, andj
S
and
S
for S. M is arandomvariabledenotingthesafety
margin.
The following procedure is introduced in subsec-
tion3.2.1whereDVM was introduced. Theloadand
resistancefactorscanbeobtainedasfollows:
where, characteristic values R andS canbeobtained
bythefollowingequationsbychoosingproper fractile
values, w
R
andw
S
, respectively.
where, +
1
(): inverse standard normal probability
distributionfunction.
Tostudychoicesof R andS onsensivitiesof
R
and
S
, whencoefficient of variationV
R
andV
S
changes,
wecarried out aparametric study whereparameters
areset asfollows:
Target reliabilityindex:,
T
=2, 3
Meanvaluesof R andS: j
R
=7.0, j
S
=3.0
V
R
, V
S
: 0.10, 0.20, 0.30
Percentageof fractile: w
R
=1%50%, w
S
=99%
50%.
Sensitivityfactorsinthecaseof linearperformance
function with two statistically independent basic
variablescanbecalculatedasfollows:
Someof theimportant figures usedintheconsid-
erationaspresentedinFigures58.
Someremarkscanbemadeasthefollowings:
a. According to Figures 5a, b & c, as far as
S
is
concerned, it changes with changeof V
S
, but the
change is less when higher fractile value of S is
taken. Therefore, the Baseline Technique is valid
that thecharacteristicvalueS shouldbechosenas
443
Figure5(a). V
S
vs.
S
, Normal,
T
=3, V
R
=0.1.
Figure5(b). V
S
vs.
S
, Normal,
T
=3, V
R
=0.2.
Figure5(c). V
S
vs.
S
, Normal,
T
=3, V
R
=0.3.
ahigher fractilevalue, for example95%or 99%.
It should benoticed, however, that if onetakes a
very high fractile value, one may get a load fac-
tor,
S
, that issmaller thanunity(Figures6a, b, &
c). Thismay not bepreferablefor practical design
engineers.
b. However, for
R
, theadvantageof Baseline Tech-
nique is not as obvious as that for
S
. As can be
seen fromFigures 7a, b, & c, stability of
R
for
lower fractile values is not so obvious as that of
valuescloser tomeanvalue.
Furthermore,
R
is very unstableto changeof V
R
whenavery lowfractilevalueis taken. Thesame
statement canbereadfromFigures8a, b, & cthat
R
sfor differentV
R
arenotgettingcloser for small
fractilevalueof R.
c. The statements (a) and (b) are more pronounced
when
T
is larger, although the results are not
attachedinthispaper.
Therefore, for thecasewithtwoindependentlysta-
tisticallynormally-distributedrandomvariablesSand
RinthelinearperformancefunctionM =RS,itisrec-
ommendedtotakehighfractilevaluefor S, however,
Figure6(a). w
S
vs.
S
, Normal,
T
=3, V
R
=0.1.
Figure6(b). w
S
vs.
S
, Normal,
T
=3, V
R
=0.2.
Figure6(c). w
S
vs.
S
, Normal,
T
=3, V
R
=0.3.
usingthelowfractilevalueforRmaynotbeaseffective
asincaseof S.
The same procedure is also done for the case
that Sand Rfollow lognormal distributions. It is also
observed that the Baseline Technique is valid for S
than for R when S and R followlognormal distribu-
tions. Furthermore, comparedto thenormal case, all
thestatementsaremorepronouncedinlognormal case.
Inconclusion, theadvantageof Baseline Technique
ismuchmoreobviousindeterminingtheloadfactor,
S
, comparingtothatof
R
. Itisrecommendedtotake
highfractilevalueforS, howeverthelowfractilevalue
neednot tobechosenfor R.
4.3 Determination of a target reliability
Asintroducedinmanyreferencesandtextbooks, there
aremainlythreewaystodetermineatarget reliability
level:
(1) Adopt reliability level that preservedintheexist-
ingstructures.Thisimpliespreservethereliability
level of theexistingdesigncodes.
444
Figure7(a). V
R
vs.
R
, Normal,
T
=3, V
S
=0.1.
Figure7(b). V
R
vs.
R
, Normal,
T
=3, V
S
=0.2.
Figure7(c). V
R
vs.
R
, Normal,
T
=3, V
S
=0.3.
Figure8(a). w
R
vs.
R
, Normal,
T
=3, V
S
=0.1.
(2) Determinethereliabilitylevel basedoncompari-
sonwithotherexistingindividual andsocial risks,
i.e. thebackgroundrisks.
(3) Useoptimizationschemebasedonsomeeconomic
or other criteria(e.g. LQI).
Themethodsareorderedfromsimpleandpractical
onestomoretheoretical andideal ones. Intheactual
codecalibration, thefirstmethodismostlyemployed.
Theinformationobtainedinthesecondisuseful when
Figure8(b). w
R
vs.
R
, Normal,
T
=3, V
S
=0.2.
Figure8(c). w
R
vs.
R
, Normal,
T
=3, V
S
=0.3.
oneneedstocomparethedesignreliabilityrisktoother
social risks. This situationis comingto beimportant
whenoneneedstomakedecisionbetweeninfrastruc-
turebuilding and other soft alternatives. For thelast
method, it may be important to understand the gen-
eral frameworkof themethod. However, thismethodis
rarelyusedinpractical codecalibration, anddiscarded
fromthisreport.
4.3.1 Target reliability by existing design codes
Duetovariousdifficultiesinevaluatingabsolutelevel
of reliability of designedstructure, most of practical
codecalibrationsaredonebysettingtarget reliability
level based on existing structures assuming that the
existing structures are satisfying the reliability level
requiredbythesociety.
Commonpracticeistoevaluatethereliabilitylevel
of theexistingstructurebeforesettingthetarget reli-
ability level. This implies the reliability analysis of
structuresdesignedbytheexistingdesigncodes.
For theconvenienceof codewriters, sometypical
target reliability indices,
T
, arelistedinadocument
like ISO2394 (ISO, 1998). ISO2394 first gives the
relationshipbetween andfailureprobabilityassum-
ing followsthenormal distributionasshowninTable
2.Thetypical targetreliabilityindicesarepresentedin
Table3.
AISC LRFD(Elingwoodet al., 1980, 1982) which
isLRFDbasedcodeadoptedtarget tobe2.53.5for
steel beams andcolumns, and3.03.5for RC beams
and columns. This was one of the first attempts to
calibrateload and resistancefactors in LRFD based
designcode.
445
Table5. Relationshipbetween andP
f
.
P
f
10
1
10
2
10
3
10
4
10
5
10
6
10
7
1.3 2.3 3.1 3.7 4.2 4.7 5.2
Table6. Taget -values(life-time, examples).
Relativecostsof Consequencesof failure
safetymeasures
small some moderate great
High 0 A 1.5 2.3 B 3.1
Moderate 1.3 2.3 3.1 C 3.8
Low 2.3 3.1 3.8 4.3
Somesuggestionsare:
A: for serviceabilitylimit sates, use =0for reversible
and =1.5for irreversiblelimit sate.
B: for fatiguelimit states, use =2.3to3.1,
dependingonthepossibilityof inspections.
C: for ultimatelimit statesdesign, usethesafety
classes =3.1, 3.8and4.3.
Thetarget s that havebeenusedingeotechnical
codecalibrationshavebeensummarizedinPaikowsky
(2004). Basedonthereview, they adoptedthetarget
?for asinglepiletobe3.0, whereas2.33isusedfor
grouppilessupportingafoundationwithmorethan5
piles. Thereduced was adopted in thegroup piles
becauseof theredundancyof thefoundationsystem.
There are many proposed target in the various
reports andpapers. It is, however, important to care-
fully reviewthepurposefor whichthetarget isset.
They may bedifferent for material, members, failure
modes, limit statesandcalculationmethods. Further-
more, can be defined either for annual or for life
timeof astructure, anddistinctionbetweenthetwois
sometimesoverlooked.
4.3.2 Criteria on allowable risk
ICG (2003) defines risk as probability of an event
times consequence if the event occurs. Similar defi-
nitionsof risk canbefoundinmany other literatures
(e.g. Beacher & Christian, 2003). Therefore it is
importanttoconsiderbothprobability(i.e. frequency)
of theoccurrenceof event anditsconsequenceat the
same time. Furthermore, in preparing countermea-
sures for reducing risk, both reduction of frequency
andconsequenceneedtobeconsidered.
It is not appropriate to think there is an absolute
threshold between acceptable risk and unacceptable
risk. Every individual has his/her own preference
for risk, and therefore acceptable risk level is dif-
ferent from a person to another depending on the
riskunder consideration. Starr (1969) haspointedout
that thereis quitedifferencebetweenacceptablelev-
elsfor involuntary andvoluntary risks. Slovic (1987)
found that factors like dread or uncontrollable and
unknown or unobservable influence considerably in
riskrecognitionof general individuals.
Inordertoaccommodatesuchambiguityinaccept-
ablerisklevel recognitionintoaccount, HSE(1999) of
UK proposedtodividerisklevelsintothreeregions:
(1) Unacceptable region: risk cannot be justified
except inextraordinarycircumstances.
(2) ALARP (as low as reasonably possible) region:
tolerableonlyif thecostof riskreductionisgrossly
disproportionatetothereductioninriskachieved.
(3) Risknegligibleregion:furtherefforttoreducerisk
isnot normallynecessary.
ALARP region is well established in case law in
UK, and has been adopted in UK health and safety
Act (GEO, 2007; Diamantidis, 2008).
Diamantidis (2008) has proposed to make dis-
tinction between individual risk and societal risk in
developingcriteriafor acceptablerisk level, whichis
alsofollowedinthereport.
(1) Individual risk
Individual risk isdefinedastheannual probability
of being harmed by ahazardous situation. A typical
individual risk isfatality risk whichcanbemeasured
bytheannual probabilityof beingkilledbysomeharm.
Table7presentsannual fatal accidentrateinthedevel-
oped countries by Diamantidis (2008). In addition,
annual fatal accidentrateinJ apanderivedfromofficial
statistics in 2004 in J apan by Honjo (2006) arepre-
sentedinTable8. Someof thefiguresarealsocopied
toTable7forcomparison. Itshouldbenoticedthatthe
methodsusedtoobtaintheseratesinthesetwotables
aredifferent, thuscomparisonmaybemadewithsome
care.
It maybepossibletosayfromtheseresultsthat the
risk level for eachevent is different whether they are
voluntaryor involuntary, andalsobenefit onecanget
fromeachevent.
(2) Societal risk
One of the most popular ways to describe the
societal risk is by an FN curve. F stands for fre-
quency and N for number of fatalities. An FN
curvepresents probability of occurrenceof an acci-
dent which involves more than n fatalities, and has
relationshipwiththecumulativedistributionfunction
of N asfollows:
FN curvewaswidelyknownbyareportpreparedby
USNRC(1975), knownasWash1400, toshowtherisk
of buildingnuclear power plants inUS. FN curves
showing various societal risks with therisk of oper-
atingnuclear power plantswerepresentedinorder to
justify theconstruction of newnuclear power plants
intheUS. Inthefieldof geotechnical engineering, an
FN curvedrawnby Baecher in1982is well known
(Baecher & Christian, 2003).
Someof relativelynewFN curvesproposedinthe
variouspartsof theworldhavebeenplottedinFigure9.
ThesearetheDutchgovernmentpublishedriskguide-
lines (Versteeg, 1987), the Hong Kong government
planning department (HKPD, 1994) and ANCOLD
446
Table7. Fatal accident rateindevelopedcountries(modifiedfromDiamantidis, 2008).
Duringactivity Proportionof Annual Annual fatal ratein
Causeof death (/108hrs) time(average) probability J apan(Honjo, 2006)
Rockclimbing 4000 0.005 2.0010
3
Motorcycleaccident 300 0.01 3.3310
4
Skiing 130 0.01 1.2510
4
Workersinhighrisebuildingindustry 70 0.2 1.4310
4
Deepseafishing 50 0.2 1.0010
3
Workersonoffshoreoil andgasrigs 20 0.2 4.0010
4
Diseaseaveragefor 4044agegroup 17 1 1.6710
3
Travel byair 15 0.01 1.410
5
Travel bycar 15 0.05 7.710
5
5.810
5
Diseaseaveragefor 3040agegroup 8 1 8.310
4
Coal mining 8 0.2 1.710
4
Travel bytrain 5 0.05 2.510
5
Constructionindustry 5 0.2 1.010
4
9.810
4
Agriculture(employees) 4 0.2 8.310
5
2.210
4
Accidentsinthehome 1.5 0.8 1.110
4
Travel bylocal bus 1 0.05 510
6
Chemical industry 1 0.2 210
5
2.310
4
Californiaearthquake 0.2 1 210
5
3.410
6
(averageover
1960-2004?
Table8. Annual fatal rateinJ apanfromvarious causes in
2004(Honjo, 2006).
Causeof death Annual fatal rate
All causes 0.8010
2
Cancer 0.2510
2
Heart disease 0.1210
2
Cerebrovascular disease 0.1010
2
pneumonia 0.7510
3
Workrelatedaccident 0.2610
4
Natural hazard 0.2410
5
Trafficaccident 0.5810
4
Suicide(2003) 0.2510
3
(1994). All these FN curves divide the space into
threeregions, namely unacceptable, ALARP andrisk
negligible. Furthermore, it is observed that all three
proposalshaveverysimilarborderlinesinthelowcon-
sequencearea(i.e. thenumber of fatalitiesbetween1
to10), butquitedifferentbordersinhighconsequence
region. Itissuggestingdifficultiesindeterminingrisk
acceptance for low frequency high consequence
events.
5 CODE CALIBRATIONBY MCS
A methodisproposedinthisreport todetermineload
and resistance factors based on DVM approach by
MCS. OnlyamethodthatusesordinaryMCSisintro-
ducedhere. A methodthat appliesthesubset MCMC
method to improve the efficiency of calculation is
proposedinKieuLeandHonjo(2009).
5.1 Background of the proposal
5.1.1 LRFD format
The reasons for recommending LRFD format for
design verification have been explained in detail in
section2. TheLRFDformatthathasemployedinthis
reportisof multipleresistancefactorsformatwhichis
givenasbelow:
Both force and resistance are lumped into several
terms. S
j
s and R
i
s are functions of basic variables,
andsomeof basic variables areincludedintheboth
terms.
5.1.2 Merits of MCS
Theperformancefunctionsusedinreliabilityanalysis
becomemorecomplex andnon-linear. This situation
makes it difficult to use FORM in reliability analy-
sisduetotimeconsumingprogrammingworks, slow
convergenceandexistenceof thelocal optimumsolu-
tions. Ontheother hand, MCSiseasytoimplementto
anycomplexperformancefunctionsandastablesolu-
tioncanbeobtainedif enoughcomputational timeis
provided.
Theproblemof computational timeis gettingless
dueto rapiddevelopment of computers. Theauthors
feel problemsof MCSincodecalibrationmaybethe
followings:
In code calibration using solely MCS, a struc-
tureneedstoberedesignedevery timefactorsare
modified.
It is difficult to obtain overall view of the code
calibrationprocesswhenMCSisused.
447
Figure9. RelativelynewFN curvesproposedinthevariouspartsof theworld.
5.1.3 Merits of DVM
DVM is developed in the framework of FORM to
determine factors to ensure sufficient safety margin
to be secured in design. This method has following
advantages:
Loadandresistancefactorscanbecalculatedbased
on target reliability index, sensitivity factors and
COV (coefficient of variation) of S andR.
Redesignof astructureisnot requiredinthecode
calibrationprocessaslongasinitiallychosenstruc-
tural dimensions have reliability level that is not
veryfar fromthetarget.
Contributionof eachloadandresistancetermcan
beevaluatedbythesensitivityfactor.
5.2 Code calibration by ordinary monte carlo
simulation
In complex cases where a performance function
contains basic variables in both load and resis-
tancefunctions, i.e. g(X)=R(X)S(X), whereX=X
i
,
(i =1, . . . , n), and shows high non-linearity for the
basic variables, it is difficult, if not impossible, to
determineloadandresistancefactorsbyusingDVM.
A simplemethodwhichisbasedonDVMconceptand
yettakesadvantagesof OrdinaryMonteCarlosimula-
tion(OMCS) toobtaintheloadandresistancefactors
isproposed:
Step1: Define basic variables X
i
(i =1,. . ., n) and
their PDFs.
Step 2: Carry out OMCS. TheOMCS is carried
out based on the probability distributions of basic
randomvariablesX
i
(i =1, . . . , n).N
t
samplesaregen-
erated. P
f
maybeestimatedastheratioof N
f
(number
of samplesfall intothefailureregion) andN
t
. Inaddi-
tion, by usingthegeneratedsamples, it ispossibleto
estimatethejoint densityfunctionf
R,S
(R, S).
Step 3: Define an approximate design point
(R*,S*). Amongthegeneratedpairsof (R,S), choose
theones closeto thelimit stateline, andchooseone
thathasthemaximumlikelihood. Theonechosencan
beusedas anapproximationof thedesignpoint and
makeit (R*, S*).
Step 4: Calculate
R
,
S
and
R
,
S
. Thesensi-
tivity factors canbeobtainedby either of thefollow
twomethods:
Method 1:
where, (Z
R
, Z
S
the standardized space, i.e.Z
R
=(R
j
R
),
R
and
Z
S
=(S
j
S
),
S
respectively.
Method 2:
Eqs. (8) and (9) should beused to estimatetheload
andresistancefactors.
448
Table 9. Sensitivity factors, load and resistance factors
obtainedfor simplelinear performancefunctionexample.
R

S

R

S
OMCS 0.72 0.69 0.67 1.12
Subset MCMC 0.69 0.73 0.69 1.15
Trueresults 0.71 0.71 0.68 1.13
If it is judged more appropriate to fit lognormal
distribution, the sensitivity factors can be estimated
byeither of thetwomethodsasfollows:
Method 1:
where, (Z
lnR
, Z
lnS
standardized space, i.e. Z
lnR
=(lnR
j
lnR
),
lnR
andZ
lnS
=(lnS
lnj
S
), ln
S
.
Method 2:
Eqs. (13) and(14) shouldbeusedto estimatethe
loadandresistancefactors.
5.3 Examples
5.3.1 Linear performance function Z =RS
R and S areindependently normally distributed ran-
domvariables, i.e. R N(7.0,1.0) andS N(3.0,1.0)
(KieuLe& Honjo, 2009).
In this example, the true results are known as
follows:
Truefailureprobability: P
f
=0.2310
2
Truereliabilityindex: =2.83
Truedesignpoint: (R
, S
)=(5.0,5.0)
In this example sensitivity factors obtained by
Method 1 (Eq.(29)) aretruevalues. Load and resis-
tance factors are then obtained for a certain target
reliability index, for example,
T
=3.2. The results
arealsoshowninTable9.
The results obtained by OMCS after 50,000 runs
areasfollows:
Probabilityof failure: P
f
=0.2510
2
Designpoint: D(R
, S
)=(5.1,4. 8)
SensitivityfactorsarecalculatedbyusingMethod2
showninEq.(30), whereastheloadandresistancefac-
torsareobtainedfor
T
=3.2,andtheresultsareshown
inTable9.
A similar procedure based on DVM and Subset
MCMCbyKieuLe&Honjo(2009) isalsoattempted
withNt =100samples. Anexampleof thegenerated
samplesisshowninFigure10.
Figure 10. Generated samples for simple linear perfor-
mancefunction.
Probabilityof failure: P
f
=0.2610
2
Thesamplesof thelaststepareusedtoestimatethe
designpoint.
Designpoint: D(R
, S
)=(5.1,5.0)
SensitivityfactorsarecalculatedbyusingMethod2
showninEq.(30). Finally, loadandresistancefactors
areobtainedfor
T
=3.2. Theresults areasloshown
inTable9.
Itmayberecognizedthattheresultsobtainedbythe
proposedmethodbasedonOMCS(orSubsetMCMC)
andDVMareveryclosetothetrueresults. Therefore,
themethodseemstoworkwell inthissimpleexample.
5.3.2 A Retaining wall example
Determination of load and resistance factors
Theproposed method is now applied for reliabil-
ity analysis of a gravity retaining wall (Figure 11)
under slidingfailuremode(KieuLe& Honjo, 2009).
ThisexampleistakenfromOrr (2005). Thenecessary
parameters of the retaining wall, soil below and fill
abovetheretainingwall aredescribedbelow.
Propertiesof sandbeneathwall: c
s
=0,
s
,
s
Propertiesof fill behindwall: c
f
=0,
f
,
f
Groundwater level isatdepthbelowthebaseof the
wall
Friction angle between base wall and underlain
sand:
bs
Thicknessof retainingwall: w=0.4m
Thelimit statefunctionof theproblemisgivenas:
whereR: total horizontal forcethat resists horizontal
movement of the wall, S: total horizontal force that
movesthewall.
449
Figure11. Descriptionof retainingwall.
Theactual formof R andS aregivenasfollows:
where, tan(45
f
,2)=
1+(tan
f
)
2
) tan
f
tan(45
s
,2)=
1+(tan
s
) +tan
2
Thus,
s
,
f
,
c
, tan
s
, tan
f
, tan
bs
andq are
consideredasbasicvariablesof thelimitstatefunction
andsomeof themareincludedbothinR andS. The
probabilistic parametersof basic variablesareshown
inTable10.
OMCS has beencarriedout for 100,000samples.
Theestimatedfailureprobabilityandreliabilityindex
areobtainedasbelow:
P
f
= 0.2910
2
(i.e. = 2.76)
Thisapproximateddesignpoint:
(R = 168.9, S = 168.8)
Basedon100,000generatedsamples(R, S), meanand
standarddeviationof R andS canbeestimated.
j
R
=204.5
R
=23.6
j
S
=122.5
S
=21.1
In addition, based on fitting test on the gener-
ated points, the density function f
R,S
(R, S) seems to
bebest-fittedtoboththejoint normal andthelognor-
mal distributionfunctions.Therefore, thejointdensity
functionf
R,S
(R, S)maybeassumedtobeeitherthejoint
normal or thelognormal distribution.
Thesensitivityfactorsareobtainedbytwomethods.
Finally, loadandresistancefactorsareobtainedfor a
Table10. Probabilisticparametersof basicvariablesinlimit
statefunctionof retainingwall under slidingfailuremode
Variables Distribution j COV
X
1

f
Lognormal 20 1.000 0.05 2.994 0.050
X
2

s
Lognormal 19 0.950 0.05 2.943 0.050
X
3

c
Lognormal 25 1.250 0.05 3.218 0.050
X
4
tan
f
Lognormal 0.781 0.107 0.14 0.256 0.136
X
5
tan
s
Lognormal 0.675 0.086 0.13 0.402 0.127
X
6
tan
bs
Lognormal 0.577 0.070 0.12 0.557 0.120
X
7
q Lognormal 15 1.500 0.10 2.703 0.100
Table11. Resultsobtainedfor
T
=3.1, sensitivityfactors
areobtainedbasedonmethod1& 1
ByOMCS BySubset MCMC

R S R S R S R S
0.57 0.82 0.64 0.77 0.54 0.84 0.61 0.79
0.61 1.79 0.61 1.85 0.62 1.76 0.61 1.83
Table12. Resultsobtainedfor
T
=3.1, sensitivityfactors
areobtainedbasedonmethod2& 2
ByOMCS BySubset MCMC

R S R S R S R S
0.75 0.67 0.56 0.83 0.73 0.68 0.54 0.84
0.57 1.69 0.63 1.91 0.57 1.66 0.63 1.87
targetreliabilityindexassumedtobe
T
=3.1andare
showninTable11and12.
Notethatthecharacteristicvalueof RandS adopted
inthisexampleiscalculatedfromcharacteristicvalues
of thebasic variables which arechosen as themean
valuefor X
1
toX
6
, while95%of fractilevaluefor X
7
.
ItseemsthatOMCSmethodinthiscaserequiresvery
larger calculationeffort.
Subset MCMC simulationisthencarriedout 1000
runs to obtainP
f
of thegravity retainingwall under
slidingfailuremode.Thenumberof samplesgenerated
foreachvariableinastepisN
t
=100.Inaddition,from
thegeneratedsamplesof basicvariablesduringsubset
MCMC simulation, seriesof numbersof R andS are
also calculated. Thereafter, theseR and S values are
utilizedfor determinationof partial safetyfactors.
The estimated P
f
by 1000 simulation resulted
the mean value of 0.2710
2
(i.e. =2.70), and
COV(P
f
)=1.40.
Designpoint:
f
=20.2
s
=19.1
c
=24.1
tan
f
=0.588
tan
s
=0.624
tan
bs
=0.472
q =14.9
450
Thisapproximateddesignpoint isequivalent topoint
(R=169.1, S =166.8).
N
t
pairs of (R, S) calculated from N
t
generated
pointsX
i
(i =1, . . . , n),thataregeneratedinthesubset
F
0
, areutilizedtoestimatemeanandstandarddevia-
tionof R andS. Thejoint density functionf
R,S
(R, S)
is also assumed to be either the joint normal or the
lognormal distribution.
j
R
=201.1
R
=21.8
j
S
=120.8
S
=20.2
Thesensitivityfactorsareobtainedbytwomethods.
Loadandresistancefactors(showninTable11and12.)
are obtained for the same
T
and the characteristic
valueof R andS asinthecaseof OMCS.
Followingsaresomeremarksontheresults:
LoadandResistanceFactorsobtainedall thecases,
i.e. different joint density functions, different way
tocalculatesensitivityfactors, areverysimilar.
Theresultsobtainedbasedontheestimateddesign
point inTable11(bymethod1& 1) isbetter than
thoseobtainedbasedononlythesamplesgenerated
inthefirststepof subsetMCMCprocedureinTable
12(method2& 2), asexpected.
Theresults obtainedby OMCS (for 100,000sam-
ples) andSubset MCMC arequitesimilar.
This report is prepared by a working group in J GS
chapter of TC23 Limit State Design in Geotechni-
cal Engineering Practice andsummarizestheimpor-
tant points to note and recommendations when new
designverificationformulas aredevelopedbasedon
Level I RBD format. The followings are the major
recommendationsof thisreport.
(1) Several RBDLevel I verificationformatssuchas
PFMandLRFDhavebeenproposed. Itisrecom-
mendedtouseLRFDformatatpresentduetothe
reasonsbelow:
-A designer can trace the most likely behavior of
the structure to the very last stage of design if
the characteristic values of the resistance basic
variablesareappropriatelychosen.
-The format can accommodate a performance
function with high non-linearity and with some
basicvariablestobothforceandresistancesides.
-Code calibration is possible for the case only
total uncertainty of thedesignmethodisknown,
wherethetotal uncertaintycannotbedecomposed
toeachsourceof uncertainty.
(2) The DVM should be adopted as the principal
approachincodecalibration. MCS methodmay
beusedforreliabilityanalysis.Aprocedureispro-
posed to use MCS method to obtain sensitivity
factorsanddesignpoints.
(3) It is recommendedto useameanvalueas char-
acteristicvalueof basicvariablesrelatedtoresis-
tance. Statistical uncertainties in estimating the
meanvaluemaybeconsidered.Thisisagainbased
on the philosophy that a designer should trace
the most likely behavior of the structure to the
very last stage of design. Thisisveryimportantin
geotechnical designwhereengineeringjudgment
playsimportant roleindesign.
(4) Baseline technique which states a fractile value
shouldbeusedfor acharacteristicvaluebecause
of thestabilityof partial factorsagainstchangeof
variation of abasic variableis criticized. It was
shownbysomesimplecalculationsthatthismerit
doesnot exist for resistancesidebasicvariables.
REFERENCES
AASHTO. 1994. LRFD BridgeDesignSpecifications, 1
Ed.
,
AmericanAssociationfor StateHighwayandTransporta-
tionOfficials, Washington, D.C.
Ed.
,
Ed.
,
AIJ. 1993. Recommendationsfor LoadsonBuildings,Archi-
tectural Instituteof J apan
AIJ.2002.RecommendationsforLimitStateDesignof Build-
ings, Architectural Instituteof J apan, J apaneseversion
AIJ. 2004. AIJ Recommendations for Loads on Buildings
(2004), Architectural Instituteof J apan
ALCOLD. 1994. Guidelines on risk assessment, Australian
National CommitteeonLargeDams
Allen, T.M. 2005. Development of Geotechnical Resistance
FactorsandDowndragLoadFactorsfor LRFD Founda-
tionStrengthLimit StateDesign. ReferenceManual, No.
FHWA-NHI-05-052.
Au, S.K., &Beck, J.L. 2003. Subsetsimulationanditsappli-
cationtoseismicriskbasedondynamicanalysis. J ournal
of engineeringmechanics, Vol. 129(8): 901917.
BaecherG.B, &Christian, J.T. 2003. Reliability and Statistics
in Geotechnical Engineering.J ohnWiley&Son,England.
Becker, D.E. 2006. Limit states design based codes
for geotechnical aspects of foundations in Canada.
TAIPEI2006 International Symposium on New Genera-
tion Design Codes for Geotechnical Engieering Practice,
Taiwai.
BrinchHansen, J. 1953. Earthpressurecalculation. Copen-
hagen, Denmark: TheDanishTechnical Press
BrinchHansen, J. 1956. Limit state and safety factos in soil
mechanics. Danish Geotechnical Institute, Copenhagen,
BulletinNo. 1.
BrinchHansen,J.1967.Thephilosophyof foundationdesign,
designcriteria, safetyfactor andsettlementlimit, Bearing
Capacity and Settlement of Foundations (ed. A.S. Vesic),
pp. 913, DukeUniversity.
Cornell, C.A. 1969. A Probability Based Structural Code,
Journal of ACI, Vol. 66, No. 12, pp. 974985
CSA. 2000. Canadian Highway Bridge Design Code, CSA
S6-2000. Rexdale, Ontario: CanadianStandardsAssocia-
tion
Ditlevsen, O., & Madsen, H.O. 1996. Structural Reliability
Methods. J ohnWiley& Sons, England.
Diamantidis, D. 2008. Backgrounddocumentonriskassess-
ment in Engineering: Risk Acceptance Criteria (Docu-
ment #3), J CSS(J oint CommitteeonStructural Safety).
CEN, 2004. EN 1997-1Eurocode7: Geotechnical design
Part 1: General rules.
451
Ellingwood, B., Galambos, T. V., MacGregor, J.G., & Cor-
nell, C.A. 1980. Development of a Probability Based Load
Criterion for American National Standard A58: Build-
ing Code Requirements for Minimum Design Loads in
Buildings and Other Structures.NBSSpecical publication
577.
Ellingwood, B., MacGregor, J.G., Galambos,T.V., &Cornell,
C.A. 1982. Probability Based Load Criteria: Load Fac-
tors andLoadCombinations. ASCE, Vol. 108, No. ST5:
978997.
Gulvanessian, H., &Holicky, M. 1996. DesignersHandbook
to Eurocode 1 Part 1: Basis of design. ThomasTelford,
London.
GEO. 2007. Landslide Risk Management and the Role of
Quantitative Risk Assessment Techniques (Information
Note12/2007).
Hasofer, A.M., & Lind, N.C. 1974. Exact and Invariant
Cecond-MomentCodeFormat.ASCE,Vol.100,No.EM1:
111121.
HKGPD. 1994. HongKongPlanningStandardsandGuide-
lines, Chapter 11, Potential Hazard Installations, Hong
Kong.
HSE. 1999. Reducing risks, Protecting people, Health and
SafetyExecutive, London.
Honjo,Y. &Kusakabe, O. 2002.A keynotelectureProposal
of a comprehensive foundation design code: Geo-code
21 ver. 2, Foundation design codes and soil investiga-
tion in view of international harmonization and perfor-
mance based design, Balkema(Proc. of IWSKamakura),
p. 95103.
Honjo, Y., Y. Kikuchi, M. Suzuki, K. Tani and M. Shirato,
2005.J GSComprehensiveFoundationDesignCode:Geo-
code21, Proc. 16thICSMGE, pp. 28132816, Osaka.
Honjo, Y. 2006. Somemovementstowardestablishingcom-
prehensivestructural designcodesbasedonperformance-
basedspecificationconcept inJ apan. TAIPEI2006 Inter-
national Symposium on New Generation Design Codes
for Geotechnical Engieering Practice, Taiwai.
ICG. 2003. Glossary of Risk Assessment Terms for ICG,
based on recommendation of ICOLD, IUGS working
group on landslides and ISDR/UN, edited by F. Nadim,
S. LacasseandH. Einstein.
ISO 1998. ISO 2394 General principles on reliability for
structures.
J GS. 2006. J GS4001-2004Principles for foundation designs
grounded on a performance-based design concept
(Geocode-21), J apaneseGeotechnical Society.
J PHA. 2007. Technical standardsfor port andharbor facili-
ties, inJ apanise.
KieuLe, T.C. & Honjo, Y. 2009. Study ondeterminationof
partial factorsfor geotechnical structuredesign, Proc. IS
Gifu(under printing).
Kulhawy, F.H., & Phoon, K.K. 2002. Observations on
geotechnical reliability-based design development in
NorthAmerica. Foundation Design Codes and Soil Inves-
tigation in view of International Harmonization and
Performance: 3148.
Matsuo, M. 1984. Geotechnical Engineering: Concept and
practice of reliability based design, Gihodo-suppan,
Tokyo, inJ apanese.
Meyerhof, G.G. 1993. Development of geotechnical limit
state design. Proceedings of the International Sympo-
sium on Limit State Design In Geotechnical Engineering,
Copenhagen: DanishGeotechnical Society, 1: 112.
MOT. 1979. Ontario Highway Bridge Design Code, Ministry
of Transportation, Canada
MOT. 1983. Ontario Highway Bridge Design Code, Ministry
of Transportation, Canada
MOT. 1991. Ontario Highway Bridge Design Code 3
Ed.
,
Ministryof Transportation, Canada
Moan T. 1998. Target levels of reliability-based reassess-
mentof offshorestructures. In: Shiraishi, Shinozika,Wen,
editors. Structural safety and reliability (proceedings of
ICOSSAR97). Rotterdam, Balkema: 204956.
Orr, T.L.L., & Farrell, E.R. 1999. Geotechnical design to
Eurocode 7. SpringerVerlag, London.
OrrT.L.L. 2000. Selectionof characteristicvaluesandpartial
factorsingeotechnical designstoEurocode7. Computers
and Geotechnics, 26: 263279.
Orr, T.L.L. 2005. Proceedings of the International Workshop
on the Evaluation of Eurocode 7. Ireland.
Ovesen, N.K. 1989. General report, Session30, Codes and
Standards, Proc. ICSMGE, RiodeJ aneiro.
Ovesen, N.K. 1993. Eurocode7:A Europeancodeof practice
for geotechnical design, Proceedings of the International
Symposium on Limit State Design in Geotechnical Engi-
neering, Copenhagen: Danish Geotechnical Society, 3:
691710.
Paikowsky, S.G. 2004. Load and resistance factor design
(LRFD) for deep foundations. NCHRP Report 507.
Phoon, K.K., Kulhawy, F.H., & Grigoriu, M.D. 1995.
Reliability-based design of foundations for transmis-
sion line structures. Report TR-105000, Electric Power
ResearchInstitute, PaloAlto, CA.
Phoon, K.K., Kulhawy, F.H., & Grigoriu, M.D. 2000.
Reliability-based design for transmission line structure
foundations. Computers and Geotechnics 26: 169185.
Phoon,K.K.,Becker,D.E.,Kulhawy,F.H.,Honjo,Y.,Ovesen,
N.K., &Lo, S.R. 2003. WhyconsiderReliabilityAnalysis
for Geotechnical Limit StateDesign?. LSD2003 Interna-
tional Workshop on Limit State Design in Geotechnical
Engineering Practice, USA.
Schneider, H.R., 1997. Definition and determination of
characteristic soil properties, ContributiontoDiscussion
Session2.3, Proc. ICSMGE, Hamburg, Balkema.
Simpson,B.&Driscoll,R.1998.Eurocode 7acommentary.
CRC Ltd., London.
Simpson, B. 2000. Partial factors: where to apply them?.
LSD2000: International Workshop on Limit State Design
in Geotechnical Engineering, Melbourne, Australia.
Slovic, P. 1987. Perceptionof risk. Science, 236, p. 280285.
Starr, C. 1969. Social benefit versus technological risk,
Science, 165, 13321238.
Takaoka, N. 1988. History of reliability theory. J ournal of
Structural Engineering Series 2: Evaluation of lifetime
risk of structures, J apaneseSociety of Civel Engieers, p.
294300, inJ apanese.
Tang, W. 1971. A Bayesian evaluation of information for
foundationengineeringdesign. International Conference
on Applications of Statistics and Probability to Soil and
Structural Engineering.
USNRC. 1975. Reactor Safety Study, WASH-1400, NUREG
75/014.
Versteeg. 1987. External safety policy in the Netherlands:
Anapproachto risk management, Journal of Hazardous
Materials, 17, p. 215221
Whitman, R.V. (1984). Evaluatingcalculatedriskingeotech-
nical engineering. ASCE, J. Geotech. Eng., 110, 2:
145186.
452
Author index
Aghajani, H.F. 347
Allen, T.M. 103
Amatya, S. 97
Au, S.K. 89, 275
Bathurst, R.J. 103
Bles, T.J. 327, 339
Bond, A.J. 111
Brassinga, H. 311
Brassinga, H.E. 327
Cao, Z.J. 89, 275
Chen,Y. 251
Chen,Y.-C. 293
Cherubini, C. 83, 165
Ching, J.-Y. 193
Ching, J.Y. 69, 127, 293, 379
Chung, M. 173
Coelho, P.A.L.F. 411
Cools, P.M.C.B.M. 339
Costa, A.L.D. 411
Cotecchia, F. 363
deGijt, J.G. 311
deGroot, M.B. 311
deWit, T.J.M. 327
Frank, R. 111
Fujimura, T. 147
Fukui, J. 185
Goh, E.K.H. 281
Hagiwara, M. 185
Hao, S.-Q. 211
Hara, T. 435
Harahap, I.S.H. 397
Hata, T. 217
Hata,Y. 333
Hayakawa, K. 257
Hira, M. 265
Hirose, T. 185
Honda, M. 155
Honjo,Y. 39, 75, 155, 301, 435
HseinJ uang, C. 379
Hsieh,Y.-H. 69, 379
Hu,Y.-G. 193
Huang, B. 103
Huang, H.-W. 211
Huang, H.W. 27, 229
Hudig, P. 311
Huh, J. 173
Ichii, K. 333
Ichikawa, H. 141
Ikuma, T. 319
Ishizuka, M. 371
Kani,Y. 257
KieuLe, T.C. 75, 435
Kikuchi,Y. 39, 155, 201, 435
Kim, H.Y. 135
Kisse, A. 97
Kohno, T. 119, 177
Korff, M. 327
Koseki, J. 185
Kurata, T. 147
Kwak, K. 173
Lacasse, S. 355
Lee, J.H. 173
Lee, K.H. 135
Lesny, K. 97
Li, D.Q. 405
Li, J.H. 419
Lim, J.K. 281
Lin, H.-D. 127
Lin, W.K. 159
Litjens, P.P.T. 339
Liu, H.H. 405
Liu, X. 251
Lollino, P. 363
Maeda, K. 221
Maeda,Y. 185
Mitaritonna, G. 363
Miyata,Y. 217, 371
Mori, M. 147
Mukaitani, M. 247
Murakami, A. 147
Nadim, F. 13, 355
Nagao, T. 39
Nakane,Y. 257
Nakatani, S. 119, 177
Nakaura, T. 177
Nakayama, T. 141
Nanak, W.P. 397
Ng, S.F. 281
Ning, Z.W. 229
Nishimura, S. 147
Oishi, T. 185
Oka, T. 287
Okazaki,Y. 247
Okuda, M. 257
Orr, T.L.L. 111
Oung, O. 327
Paikowsky, S.G. 97
Park, J.H. 173
Phoon, K.-K. 69, 193, 293
Rajabalinejad, M. 425
Rungbanaphan, P. 301
Salemans, J.W.M. 327
Santaloia, F. 363
Schuppener, B. 111
Schweckendiek, T. 311
Selamat, M.R. 281
ScoePinto, P.S. 51
Shimizu,Y. 141
Shinoda, M. 371
Shirato, M. 119, 177, 435
Simpson, B. 111
Soltani-J igheh, H. 347
Suzuki, M. 141, 147, 201, 435
Tanaka, H. 287
Tanaka, K. 247
Uzielli, M. 355
vandenHam, G.A. 311
vanGelder, P.H.A.J.M. 425
vanStaveren, M.Th. 339, 387
Verweij, A. 327
Vessia, G. 83, 165
Vitone, C. 363
Vrijling, J.K. 425
Waku, A. 221
Wang, Q. 89
Wang,Y. 89, 275
Watabe,Y. 39
Wu, S.B. 405
Wu, T.H. 1
Xie, X.Y. 229
Xue,Y.D. 27
Yamamoto, K. 265
Yamamoto, R. 247
Yamamoto,Y. 185
Yang,Y.Y. 27
Yen, M.-T. 127
Yoon, G.L. 135
Yoon,Y.W. 135
Yoshida, I. 301
Yoshinami,Y. 141
Yuan,Y. 211
Yuasa, T. 221
Zhang, L.M. 159, 237, 419
453

Geotechnical Risk and Safety

Uploaded by

Copyright:

Available Formats

Geotechnical Risk and Safety

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Geotechnical Risk and Safety

Uploaded by

Copyright:

Available Formats

GEOTECHNICAL RISK ANDSAFETY

PROCEEDINGS OF THE 2ND INTERNATIONAL SYMPOSIUM ON GEOTECHNICAL

deGroot andBaecher 1993

0.14 24 0.063 Christianet al. 1994

For circular arcslipsurfaceandH =12m.

1.1,0.16 X X 1.1,0.16 Gilbert et al. 1998

fromaproposal distribution q(x

)=(5.0,5.0), which is very closeto thevalue

is considered acoefficient of variation equal

, shows a parabolic trend with a rapid increase

wont bediscussedfurther on.

=30, 40e50%: H=5m(continuous

variation. In this case,

do not affect the slopes

valuesfor thecaseof 5mslopeheight.

but alsotoslopeheight variation.

Cov standsfor coefficient of variation.

, basedonproposalsby Reissner (1924) andVesi c

MCS FOSM Recommended

=coefficient of variationof thebiasMCS=MonteCarloSimulationusing

suggeststhatasimilar biasexistsinthisfactor assum-

was found to be related to the soil inter-

may explaininalargepart theuncertainty

is aconstant. For agiven valueof n

and, for the calculation of effec-

the calculation is performed with characteris-

Non- Drivenpile 21 0.0599 0.39

=1.89 and 1.2 to the elastic limit

isthedesignloadof thepile; theC

) aretakentobeflatPDFs, i.e.: noprior

}. Thesesampleswill beusedin(4) and

relationshipsfor all scenar-

=3andis less than

=4. For thecoderegu-

=4. For the code regulation of

relationswill fall intothe

to the horizontal. The groundwater level is

, theslopesurfacewill containathinlayerof fill,

=footing base area and is equivalent to A of

=modifiers for the scale effect

Site-1aand Site-1b tests wereconducted at thesamesite,

denote the event

(Z)-1, i.e.: thefailureevent of thetrial surface

(the actual overall probability) is always

can be found by iteratively

is highly non-trivial because most slope

in general is not the actual

even when the two assumptions are violated in the

is the center of the j-th Gaussian PDF, taken to be

of the j-th rep-

||). In this paper, aglobal

, sincemax (=) isbeingcalcu-

NonCarcinogens Referential dose(RfD0) 0.006 mg/(kgday): TCE U.S.EPA (1996)

-LCC, it will bejudgedthat It isnecessaryto

500 350 375 350

and areequal tozerobecauseof the

. Intheanalysis, therangesof the

showninFig. 4(d), it isobservedthat theareaof con-

, thecontact shear stressesis

. In this figure, the solu-