Geotechnical Earthquake Engineering-Géotechnique Symposium in Print 2015

Geotechnical
Earthquake
Engineering
Géotechnique Symposium in Print 2015
Edited by
Stuart Haigh
Downloaded by [ York University] on [24/09/16]. Copyright © ICE Publishing, all rights reserved.
Géotechnique Advisory Panel Sub-Committee for the Symposium in Print 2015:
Dr Stuart Haigh, University of Cambridge, UK

Professor Angelo Amorosi, Politecnico di Bari, Italy
Professor George Gazetas, National Technical University of Athens, Greece
Dr Erdin Ibraim, Bristol University, UK
Professor Boris Jeremic, University of California, Davis, USA
Dr Jonathan Knappett, University of Dundee, UK
Dr Stavroula Kontoe, Imperial College London, UK
Dr William Murphy, University of Leeds, UK
Professor Ellen Rathje, University of Texas at Austin, USA
Professor Ikuo Towhata, University of Tokyo, Japan
Liam Wotherspoon, The University of Auckland, New Zealand
Related titles from ICE Publishing:

Partial Saturation in Compacted Soils (Géotechnique Symposium in Print 2011).
D. Gallipoli (ed). ISBN 978-0-7277-4175-2.
Bio- and Chemo-mechanical Processes in Geotechnical Engineering

(Géotechnique Symposium in Print 2013).
L. Laloui (ed). ISBN 978-0-7277-6053-1.
Earthquake Design Practice for Buildings, third edition.

E.D. Booth. ISBN 978-0-7277-5794-4.
Designers’ Guide to Eurocode 8: Design of buildings for earthquake resistance.

M.N. Fardis, E. Carvalho, A.S. Elnashai, E. Faccioli, P. Pinto, A. Plumier.
ISBN 978-0-7277-3348-1.
Earthworks: A guide, second edition.

P Nowak, P. Gilbert. ISBN 978-0-7277-5735-7.
ICE Manual of Geotechnical Engineering (2 volumes).

J. Burland, T. Chapman, H. Skinner, M.J. Brown (eds). ISBN 978-0-7277-3652-9.
Core Principles of Soil Mechanics.

S.K. Shukla. ISBN 978-0-7277-5847-7.
ISBN 978-0-7277-6149-1
© Thomas Telford Limited 2016

Papers extracted from Géotechnique © Authors and Institution of Civil Engineers
All rights, including translation, reserved. Except as permitted by the Copyright,

Designs and Patents Act 1988, no part of this publication may be reproduced, stored in
a retrieval system or transmitted in any form or by any means, electronic, mechanical,
photocopying or otherwise, without the prior written permission of the Publishing
Director, ICE Publishing, 1 Great George Street, London SW1P 3AA.
This book is published on the understanding that the authors are solely responsible for
the statements made and the opinions expressed in it and that its publication does not
necessarily imply that such statements and/or opinions are or reflect the views or opinions
of the publishers. While every effort has been made to ensure that the statements made
and the opinions expressed in this publication provide a safe and accurate guide, no
liability or responsibility can be accepted in this respect by the authors or publishers.
Commissioning Editor: Laura Marriott

Production Editor: Rebecca Taylor
Market Development Executive: Elizabeth Hobson
Typeset by Manila Typesetting Company

Printed and bound in Great Britain by TJ International Ltd, Padstow
Preface
Damage to buildings and infrastructure due to geotechnical papers all share the similarity of advancing the current state of
failures has been observed in all the major earthquakes of the the art in geotechnical earthquake engineering research.
last half century, causing many thousands of deaths and many This issue was accompanied by a full-day symposium
billions of dollars of economic damage. Since the 1964 Niigata held on 15th June 2015 at the Institution of Civil Engineers
earthquake, this has resulted in a substantial research effort to (ICE) in London. The symposium included two sessions on
understand soil behaviour under cyclic loading, to predict the field behaviour of soils during earthquakes and model testing
onset of damaging phenomena such as soil liquefaction and of geotechnical systems during earthquakes. The ten papers
to design foundations and geotechnical systems to survive were all presented by their authors on the day, enabling a
earthquake loading. wide-ranging discussion to take place around current issues in
While this research effort has continued for the last fifty geotechnical earthquake engineering. The symposium provided
years, great advances have recently been made using state of an excellent opportunity to discuss the current state of the art in
the art laboratory testing, dynamic centrifuge modelling and geotechnical earthquake engineering and future opportunities
high quality field investigations of system performance during in both research and practice.
earthquakes such as those in Christchurch, New Zealand in This volume also contains three papers published in
2010 and 2011 and the Tohoku earthquake in Japan in 2011. The Géotechnique in 2014 on geotechnical earthquake engineering.
ability of infrastructure to continue to perform post-earthquake These papers, together with those presented during the
is of particular importance in allowing disaster mitigation Symposium in Print are representative of the high quality work
efforts to rapidly relieve suffering in the affected areas. It was on this subject produced by researchers from around the world
also shown to play a major role in resisting multi-hazards and published in Géotechnique.
during the Tohoku earthquake, after which coastal defences It was my great pleasure to act as chair of the Géotechnique
already damaged by earthquake shaking were required to resist advisory panel subcommittee and to have been guest editor of
tsunami loading. this issue of Géotechnique. I wish to thank the other members
The response to this symposium in print was substantial with of the subcommittee who have assisted me over the last 18
abstracts being submitted by authors throughout the world. months for their enormous amount of effort in making this issue
Following the standard review process of Géotechnique, ten and event occur. I would also like to thank all those who have
papers have been accepted for publication in this issue of the reviewed papers for their thorough and timely reviews. Finally,
journal, with several further papers missing the deadline to be I would like to thank Craig Schaper, Journals Editor at ICE
ready for publication according to the planned schedule. These Publishing, for his great efforts in efficiently organising this
remaining papers have been moved to the regular publication issue as well as for his ongoing efforts in making Géotechnique
process of the journal and will appear in future issues. The a success.
papers selected for this symposium cover a wide range of
topics, from liquefaction prediction based on site investigation Stuart Haigh
to the interaction between buildings owing to coupling through University of Cambridge, UK
the soil between them. Despite the wide range of subjects, these August 2015
Contents Preface iii
Session 1: Field behaviour of soils during earthquakes

Assessment of CPT-based methods for liquefaction evaluation in a Liquefaction Potential 3
Index (LPI) framework
B. W. Maurer, R. A. Green, M. Cubrinovski and B. A. Bradley
Correlation between liquefaction resistance and shear wave velocity of granular soils: 13
a micromechanical perspective
X. M. Xu, D. S. Ling, Y. P. Cheng and Y. M. Chen
An interpretation of the seismic behaviour of reinforced-earth retaining structures 25
L. Masini, L. Callisto and S. Rampello
Cyclic and dynamic behaviour of a soft pyroclastic rock 35
L. Verrucci, G. Lanzo, P. Tommasi and T. Rotonda
Empirical predictive relationship for seismic lateral displacement of slopes: 51
models for stable continental and active crustal regions
J. Lee and R. A. Green
Session 2: Model testing of geotechnical systems during earthquakes

Dynamic response of flexible square tunnels: centrifuge testing and validation of existing design 71
methodologies
G. Tsinidis, K. Pitilakis, G. Madabhushi and C. Heron
Influence of initial stress distribution on liquefaction-induced settlement of shallow foundations 89
D. Bertalot and A. J. Brennan
Seismic structure–soil–structure interaction between pairs of adjacent building structures 101
J. A. Knappett, P. Madden and K. Caucis
A new macro-element model encapsulating the dynamic moment–rotation behaviour of 115
raft foundations
C. M. Heron, S. K. Haigh and S. P. G. Madabhushi
Importance of seismic site response and soil–structure interaction in the dynamic behaviour 125
of a tall building founded on piles
E. Bilotta, L. De Sanctis, R. Di Laora, A. D’Onofrio, F. Silvestri
Related content
Some remarks on the seismic behaviour of embedded cantilevered retaining walls 137
R. Conti, G. M. B. Viggiani, and F. Burali D’Arezzo
Revisiting Nigawa landslide of the 1995 Kobe earthquake 149
H. Ling, H. I. Ling and T. Kawabata
Static and cyclic rocking on sand: centrifuge versus reduced-scale 1g experiments 155
P. Kokkali, I. Anastasopoulos, T. Abdoun and G. Gazetas
Maurer, B. W. et al. (2015). Géotechnique 65, No. 5, 328–336 [http://dx.doi.org/10.1680/geot.SIP.15.P.007]
Assessment of CPT-based methods for liquefaction evaluation in a

Liquefaction Potential Index (LPI) framework
B. W. M AU R E R � , R . A . G R E E N �, M . C U B R I N OV S K I † a n d B. A . B R A D L E Y †
In practice, several competing liquefaction evaluation procedures (LEPs) are used to compute factors
of safety against soil liquefaction, often for use within a liquefaction potential index (LPI) framework
to assess liquefaction hazard. At present, the influence of the selected LEP on the accuracy of LPI
hazard assessment is unknown, and the need for LEP-specific calibrations of the LPI hazard scale has
never been thoroughly investigated. Therefore, the aim of this study is to assess the efficacy of three
CPT-based LEPs from the literature, operating within the LPI framework, for predicting the severity
of liquefaction manifestation. Utilising more than 7000 liquefaction case studies from the 2010–2011
Canterbury (NZ) earthquake sequence, this study found that: (a) the relationship between liquefaction
manifestation severity and computed LPI values is LEP-specific; (b) using a calibrated, LEP-specific
hazard scale, the performance of the LPI models is essentially equivalent; and (c) the existing LPI
framework has inherent limitations, resulting in inconsistent severity predictions against field observa-
tions for certain soil profiles, regardless of which LEP is used. It is unlikely that revisions of the
LEPs will completely resolve these erroneous assessments. Rather, a revised index which more
adequately accounts for the mechanics of liquefaction manifestation is needed.
KEYWORDS: earthquakes; liquefaction; sands; seismicity
INTRODUCTION assumed that the severity of liquefaction manifestation is

The objective of this study is to assess the efficacy of three proportional to the thickness of a liquefied layer, the proxi-
common cone penetration test (CPT)-based liquefaction eva- mity of the layer to the ground surface, and the amount by
luation procedures (LEPs), operating within a liquefaction which FSliq is less than 1 .0. Given this definition, LPI can
potential index (LPI) framework, for predicting the severity range from 0 to a maximum of 100 (i.e. where FSliq is zero
of surficial liquefaction manifestation, which is commonly over the entire 20 m depth). Analysing SPT data from 55
used as a proxy for liquefaction damage potential. Utilising sites in Japan, Iwasaki et al. (1978) proposed that severe
data from the 2010–2011 Canterbury earthquakes, this study liquefaction should be expected for sites where LPI . 15 but
investigates the influence of the selected LEP on the accu- not where LPI , 5. This criterion for liquefaction manifesta-
racy of hazard assessments, and assesses the need for LEP- tion, defined by two threshold values of LPI, is subsequently
specific calibrations of the LPI hazard scale. Towards this referred to as the Iwasaki criterion. However, in using the
end, the deterministic LEPs of Robertson & Wride (1998) LPI framework to assess liquefaction hazard in current prac-
(R&W98), Moss et al. (2006) (MEA06), and Idriss & tice, it is not always appreciated that the Iwasaki criterion is
Boulanger (2008) (I&B08) are evaluated. inherently linked to the LEP that was in common use in
While the ‘simplified’ LEP (Seed & Idriss, 1971; Whit- Japan in 1978, which differs significantly from those com-
man, 1971) is central to most liquefaction hazard assess- monly used today. Also, it has been shown that the various
ments, the output from an LEP is not a direct quantification LEPs used in today’s practice can result in different FSliq
of liquefaction damage potential, but rather is the factor of values for the same soil profile and earthquake scenario (e.g.
safety against liquefaction triggering (FSliq) in a soil stratum Green et al., 2014), and thus different LPI values. These
at depth. Iwasaki et al. (1978) proposed the LPI to link differences have led to confusion as to which LEP is the most
liquefaction triggering at depth to damage potential, where accurate, and whether the Iwasaki criterion is equally effec-
LPI is computed as tive for all LEPs.
20ðm The 2010–2011 Canterbury earthquake sequence (CES)
resulted in a liquefaction dataset of unprecedented size and
LPI ¼ Fw(z)dz (1) quality, presenting a unique opportunity to assess the effi-
0 cacy of liquefaction analytics (e.g. Cubrinovski & Green,
2010; Cubrinovski et al., 2011; Bradley & Cubrinovski,
where F ¼ 1 � FSliq for FSliq < 1 and F ¼ 0 for FSliq . 1; 2011). Towards this end, Maurer et al. (2014) evaluated LPI
w(z) is a depth weighting function given by w(z) ¼ 10 � 0 .5z; during the CES at approximately 1200 sites using the
and z is depth in metres below the ground surface. Thus, it is R&W98 CPT-based LEP. Although the Iwasaki criterion was
found to be effective in a general sense, LPI hazard assess-
ments were erroneous for a portion of the study area. In
Manuscript received 31 March 2014; revised manuscript accepted 9 practice, several competing LEPs are used to assess liquefac-
January 2015. Published online ahead of print 23 February 2015. tion hazard in an LPI framework (e.g. Sonmez, 2003; Toprak
Discussion on this paper closes on 1 October 2015, for further details
see p. ii.
& Holzer, 2003; Baise et al., 2006; Holzer et al., 2006a,
� Department of Civil and Environmental Engineering, Virginia Tech, 2006b; Lenz & Baise, 2007; Cramer et al., 2008; Hayati &
Blacksburg, Virginia, USA. Andrus, 2008; Holzer, 2008; Chung & Rogers, 2011; Kang
† Department of Civil and Natural Resources Engineering, University et al., 2014), but the need for LEP-specific calibration of the
of Canterbury, Christchurch, New Zealand. LPI hazard scale has never been thoroughly investigated.
MAURER, GREEN, CUBRINOVSKI AND BRADLEY
Therefore, the objective of this study is to assess the efficacy naissance and using high-resolution aerial and satellite
of the R&W98, MEA06 and I&B08 CPT-based LEPs, imagery (CGD, 2012b) performed in the days immediately
operating within the LPI framework, for predicting the following each of the earthquakes. CPT sites were assigned
severity of surficial liquefaction manifestation. Utilising one of six damage classifications, as described in Table 1,
more than 7000 liquefaction case studies from the CES, this where the classifications describe the predominant damage
study evaluates the influence of the selected LEP on the mechanism and manifestation of liquefaction. For example,
accuracy of hazard assessment, and assesses the need for some ‘severe liquefaction’ sites also had minor lateral
LEP-specific calibrations of the LPI hazard scale. This spreading, and likewise, many ‘lateral spreading’ sites also
evaluation is performed using receiver operating characteris- had some amount of liquefaction ejecta present. Of the more
tic (ROC) analyses, which are commonly used to assess the than 7000 cases compiled, 48% are cases of ‘no manifesta-
performance of medical diagnostics (e.g. Zou, 2007). tion’, and 52% are cases where manifestations were ob-
In the following, the high-quality liquefaction case history served and classified in accordance with Table 1.
dataset resulting from the CES is briefly summarised. This
is followed by a description of how LPI was computed using
three common CPT-based LEPs. An overview of ROC Estimation of amax (peak ground acceleration)
analyses is then presented, which is followed by the analysis To evaluate FSliq using the three LEPs (i.e. R&W98,
of the LPI data. The influence of the LEP on the accuracy MEA06 and I&B08), the peak ground accelerations (PGAs)
of LPI hazard assessment is then discussed. at the ground surface were computed using the robust
procedure discussed in detail by Bradley (2013a) and used
by Green et al. (2011, 2014) and Maurer et al. (2014). The
DATA AND METHODOLOGY Bradley (2013a) procedure combines unconditional PGA
The 2010–2011 CES began with the moment magnitude distributions estimated by the Bradley (2013b) ground
(Mw) 7 .1, 4 September 2010 Darfield earthquake and in- motion prediction equation, recorded PGAs from strong
cludes up to ten events that are known to have induced motion stations, and the spatial correlation of intra-event
liquefaction in the affected region (Quigley et al., 2013). residuals to compute the conditional PGA distribution at
However, most notably, widespread liquefaction was induced sites of interest.
by the Darfield earthquake and the Mw 6 .2, 22 February
2011 Christchurch earthquake (e.g. Green et al., 2014).
Ground motions from these events were recorded by a dense Estimation of ground-water table depth
network of strong motion stations (e.g. Bradley & Cubri- Given the sensitivity of liquefaction hazard and computed
novski, 2011), and due to the extent of liquefaction, the LPI values to GWT depth (e.g. Chung & Rogers, 2011;
New Zealand Earthquake Commission funded an extensive Maurer et al., 2014), accurate measurement of GWT depth
geotechnical reconnaissance and characterisation programme is critical. For this study, GWT depths were sourced from
(Murahidy et al., 2012). The combination of densely re- the robust, event-specific regional ground-water models of
corded ground motions, well-documented liquefaction re- van Ballegooy et al. (2014). These models, which reflect
sponse and detailed subsurface characterisation comprises seasonal and localised fluctuations across the region, were
the high-quality dataset used for this study. To evaluate the derived in part using monitoring data from a network of
influence of the LEP operating in the LPI framework, a large ,1000 piezometers and provide a best estimate of GWT
database of CPT soundings performed across Christchurch depths immediately prior to the Darfield and Christchurch
and its environs (CGD, 2012a) are analysed in conjunction earthquakes.
with liquefaction observations made following the Darfield
and Christchurch events.
Liquefaction evaluation and LPI
The value of FSliq was computed using the deterministic
CPT soundings CPT-based LEPs of R&W98, MEA06 and I&B08, where
This study utilises 3616 CPT soundings performed at sites the soil behaviour type index, Ic, was used to identify
where the severity of liquefaction manifestation was well
documented following both the Darfield and Christchurch Table 1. Liquefaction severity classification criteria (after Green
earthquakes, resulting in more than 7000 liquefaction case et al., 2014)
studies. In the process of compiling these case studies, CPT
soundings were first rejected from the study as follows. First, Classification Criteria
CPTs were rejected if the depth of ‘pre-drill’ significantly
exceeded the estimated depth of the ground-water table No manifestation No surficial liquefaction manifestation or lateral
(GWT), a condition arising at sites where buried utilities spread cracking
needed to be safely bypassed before testing could begin. Marginal Small, isolated liquefaction features; streets had
Second, to identify soundings prematurely terminating on manifestation traces of ejecta or wet patches less than a vehicle
width; ,5% of ground surface covered by ejecta
shallow gravels, termination depths of CPT soundings were Moderate Groups of liquefaction features; streets had ejecta
geo-spatially analysed using an Anselin local Moran’s I manifestation patches greater than a vehicle width but were still
analysis (Anselin, 1995) and soundings with anomalously passable; 5–40% of ground surface covered by
shallow termination depths were removed from the study. ejecta
For a complete discussion of CPT soundings and the geospa- Severe Large masses of adjoining liquefaction features,
tial analysis used herein, see Maurer et al. (2014). manifestation streets impassable owing to liquefaction; .40% of
ground surface covered by ejecta
Lateral spreading Lateral spread cracks were predominant
Liquefaction severity manifestation and damage mechanism, but crack
displacements ,200 mm
Observations of liquefaction and the severity of manifesta-
Severe lateral Extensive lateral spreading and/or large open
tions were made by the authors for each of the CPT spreading cracks extending across the ground surface with
sounding locations following both the Darfield and Christch- .200 mm crack displacement
urch earthquakes. This was accomplished by ground recon-
ASSESSMENT OF CPT-BASED METHODS FOR LIQUEFACTION EVALUATION
non-liquefiable strata; soils having Ic . 2 .6 were considered true positives (i.e. liquefaction is observed, as predicted) and
too plastic to liquefy. Soil unit weights were estimated for false positives (i.e. liquefaction is predicted, but is not
each procedure using the method of Robertson & Cabal observed). Setting the threshold too low will result in
(2010). For the MEA06 procedure, the stress-reduction coef- numerous false positives, which is not without consequences,
ficient, rd, was computed using the Vs-independent equation while setting the threshold unduly high will result in many
given in Moss et al. (2006); in addition, the probability of false negatives (i.e. liquefaction is observed when it is
liquefaction (PL) was set to 0 .15, as proposed by Moss et al. predicted not to occur), which comes with a different set of
(2006) for deterministic assessments of FSliq. For the I&B08 consequences. ROC analyses are particularly valuable for
procedure, fines content (FC) is required to compute normal- evaluating the relative efficacy of competing diagnostic tests,
ised tip resistances (in lieu of FC, R&W98 and MEA06 use independent of the thresholds used, and for selecting an
Ic and CPT friction ratio, Rf, respectively); as such, FC optimal threshold for a given diagnostic test.
values were estimated using both the generic Ic–FC correla- In this study, the competing diagnostic tests are the LEPs,
tion proposed by Robertson & Wride (1998) and a Christch- and the index test results are the computed LPI values.
urch-soil-specific Ic–FC correlation developed by Robinson Accordingly, in analysing the case histories, true and false
et al. (2013). Henceforth herein, I&B081 and I&B082 refer positives are scenarios where surficial liquefaction manifesta-
to the use of the generic and Christchurch-specific Ic–FC tions are predicted, but were and were not observed, respec-
correlations, used in conjunction with the I&B08 procedure. tively. Fig. 2 illustrates the relationship among the positive
The two Ic–FC correlations are shown in Fig. 1; it can be and negative distributions, the selected threshold value and
seen that the Christchurch-specific correlation suggests dif- the corresponding ROC curve, where the ROC curve plots the
ferent Ic–FC trends for Ic , 1 .7 and Ic > 1 .7, where FC is true positive rate (TPR) and false positive rate (FPR) for
estimated to be 10 for all Ic < 1 .7. While thin layer correc- varying threshold values. Fig. 3 illustrates how a ROC curve
tions (i.e. adjustments to CPT data in thin strata to account is used to assess the efficiency of LPI hazard assessment,
for the influence of over- or underlying soils) are applicable
to the LEPs used herein, their use requires judgement, and
an automated implementation of these corrections does not No surficial liquefaction manifestation
yet exist. Given the quantity of case studies analysed, thin
layer corrections were not performed. FSliq was computed at Surficial liquefaction manifestation
1- or 2-cm depth intervals (i.e. the measuring rate of CPT B
A C D
soundings); LPI was then computed with each of the four LPI 5 8·75 14 18
LEPs as per equation (1).
Frequency
OVERVIEW OF ROC ANALYSES

Receiver operating characteristic analyses are used herein
to assess: (a) the efficacy of each LEP for predicting the
severity of liquefaction manifestation within the LPI frame-
work; and (b) the need for LEP-specific calibrations of the
LPI hazard scale. ROC analyses have been extensively used
in assessing medical diagnostic tests in clinical studies (e.g.
imaging tests for identifying abnormalities), as well as in
machine learning and data-mining research (e.g. Swets et
al., 2000; Eng, 2005; Fawcett, 2006; Metz, 2006). In any 0 2·5 5·0 7·5 10·0 12·5 15·0 17·5 20·0 22·5
ROC application, the distributions of ‘positives’ (e.g. lique- Liquefaction potential index, LPI
faction is observed) and ‘negatives’ (e.g. no liquefaction is (a)
observed) overlap when the frequency of the distributions 1·0
are expressed as a function of index test results (e.g. LPI ROC A
values). In such cases, threshold values for the index test 0·9 curve B LPI 5
LPI 8·75
results are selected considering the relative probabilities of
0·8
100 0·7
True positive rate, TPR
Christchurch-specific
90 Increasing LPI
correlation (REA13) 0·6
threshold
Apparent fines content, FC: %
80
Generic correlation 0·5 C
70 (R&W98) LPI 14
60 0·4
50 R&W98
low-plasticity 0·3
40 bound
0·2
30
0·1 D
20
LPI 18
10 R&W98
0
high-plasticity bound 0 0·1 0·2 0·3 0·4 0·5 0·6 0·7 0·8 0·9 1·0
0 False positive rate, FPR
1·25 1·50 1·75 2·00 2·25 2·50 2·75 3·00 3·25 3·50 (b)
Soil behaviour type index, Ic
Fig. 2. ROC analyses: (a) frequency distributions of no surficial
Fig. 1. Correlations between Ic and apparent FC: Christchurch- liquefaction manifestation and surficial liquefaction manifesta-
specific correlation (Robinson et al., 2013) and generic correlation tions as a function of LPI, with four different threshold LPI values
(Robertson & Wride, 1998) shown; (b) corresponding ROC curve
1·0 1·0
Perfect model
See inset
0·9 0·9 figure
OOP
0·8 0·8 R&W98
Perfect model
e MEA06
0·7 ROC 0·7 lin

e I&B081
curve nc
a
0·6 0·6 r m I&B082
rfo
pe
s o- 0·80
0·5 es 0·5 Is 6·0
gu
m 0·75
0·4 ndo 0·4
Ra 4·5
0·3 0·70
0·3 Im
pe p
rfo rov 0·65
rm in 0·2 4·0
0·2 an g
ce 5·5
0·1 0·60
0·1 0·20 0·25 0·30 0·35 0·40
0
0 0 0·1 0·2 0·3 0·4 0·5 0·6 0·7 0·8 0·9 1·0
0 0·1 0·2 0·3 0·4 0·5 0·6 0·7 0·8 0·9 1·0 False positive rate, FPR
False positive rate, FPR
Fig. 4. ROC analysis of LPI model performance in predicting the
Fig. 3. Illustration of how a ROC curve is used to assess the occurrence of surficial liquefaction manifestation. The optimum
efficiency of a diagnostic test. The optimum operating point threshold LPI values (i.e. OOPs) for each LEP are highlighted in
(OOP) indicates the threshold value for which the misprediction the inset figure
rate is minimised, as described in the text
where TPR and FPR are synonymous with ‘true positive ing is discussed later in this paper). It can be seen in Fig. 4
probability’ and ‘false positive probability’, respectively. In that, while the four LPI models perform similarly, MEA06
ROC curve space, random guessing is indicated by a 1:1 line and I&B081 are respectively the least and most efficacious,
through the origin (i.e. equivalent correct and incorrect pre- with AUC ranging from 0 .71 (MEA06) to 0 .78 (I&B081).
dictions), while a perfect model plots as a point at (0,1), To place this performance in context, AUCs of 0 .5 and 1 .0,
indicating the existence of a threshold value which perfectly respectively, indicate random guessing and a perfect model.
segregates the dataset (e.g. all sites with manifestation have Also, as highlighted in Fig. 4, the optimum threshold LPI
LPI above the selected threshold; all sites without manifesta- values for the R&W98, MEA06, I&B081 and I&B082 mod-
tion have LPI below the same selected threshold). While no els are 4 .0, 5 .5, 6 .0 and 4 .5, respectively. Thus, while the
single parameter can fully characterise model performance, lower Iwasaki criterion (i.e. LPI ¼ 5) is generally appropriate
the area under a ROC curve (AUC) is commonly used for this for predicting liquefaction manifestation in Christchurch, the
purpose, where AUC is equivalent to the probability that sites optimum threshold is LEP dependent. The presence of dif-
with manifestation have higher computed LPI than sites with- ferent optimum threshold LPI values for each LEP is not
out manifestation (e.g. Fawcett, 2006). As such, increasing surprising given that different LEPs have been shown to
AUC indicates better model performance. The optimum oper- commonly compute notably different FSliq values for the
ating point (OOP) is defined herein as the threshold LPI value same soil profile (e.g. Green et al., 2014). Although not
which minimises the rate of misprediction (i.e. FPR + (1 � unexpected, these findings may have important implications
TPR), where TPR and FPR are the rates of true and false for liquefaction hazard assessment, as the risks correspond-
positives, respectively). As such, contours of the quantity ing to particular LPI values depend on the LEP used to
[FPR + (1 � TPR)] represent points of equivalent perform- compute LPI.
ance in ROC space. Thus, in plotting the LPI data as ROC Also of interest is the influence of the Ic–FC correlation
curves for each LEP, it is possible to assess both the influence used within the I&B08 LEP. As shown in Fig. 1, the
of LEPs on the accuracy of hazard assessments, and the need Christchurch-specific correlation infers a higher FC than
for LEP-specific calibrations of the LPI hazard scale. does the generic correlation for all values of Ic, resulting in
higher computed FSliq values, and thus lower computed LPI.
As a result, the LPI hazard scale computed using I&B082
RESULTS AND DISCUSSION (i.e. using the Christchurch-specific correlation) is shifted
Utilising more than 7000 combined case studies from the towards lower values relative to the hazard scale computed
Darfield and Christchurch earthquakes, LPI values were using I&B081 (i.e. using the generic correlation) such that
computed using the LEPs of R&W98, MEA06, I&B081 and the median LPI values computed using I&B081 and I&B082
I&B082. are 7 .2 and 4 .1, respectively. In addition to influencing the
LPI hazard scale, the Ic–FC correlation affects model effi-
cacy (i.e. efficiency segregating sites with and without
Prediction of liquefaction occurrence liquefaction manifestations), with I&B081 correctly classify-
In Fig. 4, ROC curves are plotted to evaluate the performing 3% more cases than I&B082 when operating at their
ance of each LPI model in segregating sites with and with- respective OOPs. The slightly weaker performance of
out liquefaction manifestation; this initial analysis assesses I&B082 might be due to the fact that the Robinson et al.
only whether LPI accurately predicts the occurrence of (2013) Christchurch-specific Ic–FC correlation was devel-
manifestations and does not yet consider manifestation se- oped using data from along the Avon River only, while the
verity. Included in Fig. 4 are data from both the Darfield database assessed herein consists of sites distributed through-
and Christchurch earthquakes for all investigation sites, out Christchurch, although further analysis is needed to
except for those where lateral spreading was the predomi- evaluate this hypothesis. As research continues in Christ-
nant manifestation (the separate assessment of lateral spread- church, refined region-specific Ic–FC correlations, which
might improve the efficacy of LPI hazard assessment in result in different risk levels for the same LPI value,
Christchurch, are likely to be developed. particularly with I&B081.
While the preceding ROC analysis showed that optimum
threshold LPI values are LEP dependent, the implications for
liquefaction hazard assessment are not intuitively clear. For Prediction of liquefaction severity
example, it was shown that for the considered dataset the While prediction of the occurrence of surficial manifesta-
R&W98 and I&B081 LPI models have optimum threshold tion is an important component of liquefaction hazard analy-
LPI values of 4 .0 and 6 .0, respectively, but the potential sis, the severity of manifestation is of greater consequence
consequence of failing to account for different optimum to the built environment and is thus of added importance for
thresholds is not easily discerned. To elucidate the signifi- hazard mapping and engineering design. To investigate the
cance of these differences, the probability of surficial lique- capacity of each LPI model for predicting manifestation
faction manifestation is computed herein using the Wilson severity, additional ROC analyses were performed for each
(1927) interval for a binomial proportion. This assessment classification of severity in Table 1; the results are sum-
also allows for application to risk-based frameworks, com- marised in Table 2 in the form of AUC and recommended
plementing the prior evaluation of deterministic threshold threshold LPI values. Where the prior ROC analysis assessed
values. The resulting probabilities are plotted in Fig. 5 where each model’s capacity for predicting any surficial manifesta-
each data point represents one-twentieth of the corresponding tion (i.e. having at least marginal severity), the additional
dataset (,350 case histories) and is plotted as a function of analyses assess their ability to predict that manifestations
the median-percentile for each data bin (i.e. 2 .5th-percentile, will be of a particular severity (e.g. moderate as opposed to
7 .5th-percentile, and so on); also shown are third-order poly- marginal). As mentioned previously, lateral spreading is
nomial regressions for each LPI model. It can be seen from treated separately in this study, and the ‘marginal’, ‘moder-
these regressions that, at an LPI value of 5 .0, the probabil- ate’ and ‘severe’ classifications refer only to sand-blow
ities of liquefaction manifestation corresponding to the manifestations. This distinction is made because lateral
I&B081, MEA06, R&W98 and I&B082 LPI models are 0 .44, spreading is a unique manifestation associated with large
0 .53, 0 .58 and 0 .58, respectively. Conversely, using the permanent ground displacements, and because there are
optimum threshold LPI values found previously, the prob- separate criteria for assessing its severity (e.g. Youd et al.,
abilities corresponding to the respective LPI models are 2002), including the ground slope and height of the nearest
0 .50, 0 .55, 0 .53 and 0 .55. Thus, the optimum thresholds free face (e.g. river bank), among others. Consequently,
correspond to roughly the same probability of manifestation, although site profiles with thin liquefiable layers may have
whereas failing to account for the influence of the LEP could low LPI values, these sites are susceptible to lateral spread-
ing if located on sloping ground or near rivers. Since the
factors pertinent to lateral spreading cases are not considered
in the formulation of LPI, such cases should not be used to
1·0
assess its performance.
I&B082 (R2 0·99)
0·9 From Table 2, the following observations are made.
R&W98 (R2 0·99)
Probability of liquefaction manifestation
0·8 (a) Relative trends in model performance, as suggested by

AUC, are consistent for each classification of manifesta-
0·7
tion severity. While the LPI models perform similarly, the
0·6 I&B081 and MEA06 models are consistently the most
and least efficacious, respectively.
I&B081 (R2 0·99)
0·5 (b) Unsurprisingly, the models are more efficient in predict-
MEA06(R2 0·98)
0·4
ing the incidence of liquefaction manifestation than in
predicting the severity of manifestation (e.g. distinguish-
0·3 ing between marginal and moderate manifestations);
R&W98 nonetheless, the expected severity of manifestation
0·2 MEA06
1σ
increases with increasing LPI.
I&B081 (c) Differences in optimum threshold LPI values extend
0·1
I&B082 throughout the LPI hazard scale, indicating that the utility
0 of the Iwasaki criterion varies among LEPs.
0 5 10 15 20
(d ) Considering the potential for damage to infrastructure,
LPI
lateral spreading manifestations have relatively low
Fig. 5. Probability of liquefaction manifestation optimum threshold LPI values. For example, lateral
Table 2. Summary of receiver operator characteristic (ROC) analyses
LPI Model All Marginal Moderate Severe Lateral Severe lateral

manifestations† manifestation† manifestation† manifestation† spreading spreading
OOP‡ AUC§ OOP‡ AUC§ OOP‡ AUC§ OOP‡ AUC§ OOP‡ AUC§ OOP‡ AUC§
R&W98 4 .0 0 .73 3 .0 0 .68 5 .5 0 .62 10 .5 0 .69 4 .5 0 .83 10 .0 0 .66

MEA06 5 .5 0 .71 5 .0 0 .66 7 .5 0 .60 14 .0 0 .68 5 .0 0 .83 12 .0 0 .64
I&B081 6 .0 0 .78 5 .0 0 .72 9 .0 0 .64 16 .0 0 .69 6 .5 0 .79 8 .0 0 .62
I&B082 4 .5 0 .75 3 .0 0 .70 6 .0 0 .63 11 .0 0 .69 5 .0 0 .86 8 .0 0 .63
Where manifestation severity is characterised as described in Table 1.
†
Excludes sites where lateral spreading was the predominant manifestation, as described in text.
‡
Optimum operating point: recommended optimum threshold LPI value found from ROC analysis.
§
Area under ROC curve: general index of model efficacy, where higher AUC indicates better performance.
spreading and marginal sand-blow manifestations have the LEP-specific calibrations in Table 2. The prediction error
similar OOPs for each respective LPI model (i.e. similar (E) is computed using the thresholds assigned to each
LPI distributions), but the potential for damage to manifestation category, such that E ¼ LPI � (min or max) of
infrastructure is generally much greater with lateral the relevant range. For example, using the Iwasaki criterion,
spreading. This illustrates that while LPI may be useful if the computed LPI is 14 for a site with no manifestation,
for hazard assessment, the influence of local conditions E ¼ 14 � 5 ¼ 9 (where 5 is the maximum of the range of
on the manifestation of liquefaction must also be LPI values for no manifestation), whereas if the computed
considered. As such, the damage potential of lateral LPI is 6 for a site with severe liquefaction, E ¼ 6 � 15 ¼
spreading may not be well estimated by LPI. �9 (where 15 is the minimum of the range of LPI values
for severe liquefaction). Thus, positive errors indicate over-
As was done previously, the probability of manifestation predictions of manifestation severity and, conversely, nega-
is computed to assess the significance of different optimum tive errors indicate under-predictions. While there is no
thresholds, and to allow for application to risk-based frame- precedent for using a ‘moderate manifestation’ threshold
works. Because damage to infrastructure (e.g. settlement of with the Iwasaki criterion, an LPI value of 8 .0 is used herein
structures, failure of lifelines and cracking of pavements) is to facilitate comparisons among the models. Also, in light of
more likely a consequence of moderate or severe liquefac- the separate criteria for assessing lateral spreads, lateral
tion, these cases are used to compute the likelihood of spreading is assigned a wide range of expected LPI values
damaging liquefaction due to sand blows, where marginal consistent with any manifestation, independent of spreading
liquefaction is considered non-damaging. Using the method- severity (i.e. lateral spread sites are only expected to have
ology previously discussed, the probability of moderate or LPI > the threshold for marginal liquefaction).
severe liquefaction is plotted in Fig. 6 along with third-order The distributions of LPI prediction errors are shown for
polynomial regressions for each LPI model. It can be seen each model in Fig. 7 using both the Iwasaki (Fig. 7(a)) and
from these regressions that, at an LPI value of 15 .0 (i.e. the LEP-specific (Fig. 7(b)) hazard scales. It can be seen in Fig.
upper Iwasaki criterion), the probabilities corresponding to 7(a) that the distributions of errors among LEPs vary using
the I&B081, MEA06, R&W98 and I&B082 LPI models are the Iwasaki criterion, as expected. Because the models have
0 .37, 0 .40, 0 .43 and 0 .47, respectively. Conversely, using the different LPI hazard scales, applying the Iwasaki criterion to
threshold LPI values found previously for severe liquefaction each results in dissimilar performance. For example,
(Table 2), the probabilities corresponding to the respective R&W98 and I&B081 under-predict manifestation severity for
LPI models are 0 .39, 0 .39, 0 .38 and 0 .40. Thus, the 38% and 18% of cases, respectively. Conversely, using the
optimum thresholds correspond to roughly the same prob- 50
ability of damaging manifestation, whereas failing to account R&W98
for the influence of the LEP results in different risk levels. 45
Similarly, the optimum thresholds for moderate liquefaction MEA06 45%
Prediction rate (7232 cases): %
40
correspond to the same level of risk (,27%). I&B081
35
I&B082
30 40%
Comparative performance in an applied framework
25
The preceding analyses have suggested the four LPI
models may be capable of assessing liquefaction hazard, but 20
35%
that LEP-specific correlations relating LPI values and sever- 15
0·2 0 0·2
ity of surficial liquefaction manifestations are required. To
compare LEP performance in an applied setting, and to 10
determine whether any LEP is superior for practical intents 5
and purposes, deterministic ‘prediction errors’ are computed
for each case history using both the Iwasaki criterion and 0
25 20 15 10 5 0 5 10 15 20 25
Prediction error: LPI units
0·6
(a)
R&W98 I&B082 (R2 0·99)
50
MEA06
Probability of moderate or severe manifestation
R&W98 (R2 0·98) R&W98

0·5 I&B081 45
MEA06 45%
I&B082 40
Prediction rate (7232 cases): %
I&B081
0·4 35
I&B082
30 40%
25
0·3
20 35%
0·2 0 0·2
15
0·2 2
MEA06 (R 0·97)
10
5
0·1
I&B081 (R2 0·98) 0
25 20 15 10 5 0 5 10 15 20 25
1σ Prediction error: LPI units
0 (b)
0 5 10 15 20
LPI
Fig. 7. Distribution of LPI prediction errors, computed from the
Fig. 6. Probability of moderate or severe liquefaction manifesta- LPI hazard scales defined by: (a) the Iwasaki criterion; (b) LEP-
tion specific calibrations given in Table 2
35
LEP-specific calibrations of the LPI hazard scale (Fig. 7(b)),
the distributions of errors among LEPs are more similar. For |E | 5
MEA06 prediction error: LPI units

example, R&W98 and I&B081 under-predict manifestation 25
severity for 24% and 20% of cases, respectively. In addition,
the rate of accurate prediction (i.e. zero error) is improved
for each LEP; R&W98, MEA06, I&B081 and I&B082 15
accurately predict 44%, 42%, 46% and 44% of cases, re-
spectively. These performance trends mirror those of the
ROC analyses, which indicated that, although the models 5
performed similarly, I&B081 and MEA06 were respectively
the most and least efficacious. However, although accurate 5 Christchurch
predictions of manifestation severity are important, so too is
limiting the rate of highly erroneous predictions, which are Darfield
not necessarily mutually inclusive. While I&B081 has the 15
15 5 5 15 25 35
most zero-error predictions (46%), it also has the most
MEA06 prediction error: LPI units
predictions with |E| . 15 (5%). Conversely, MEA06 has the (a)
least zero-error predictions (42%), but it also has the fewest
predictions with |E| . 15 (2 .5%). Given these inconsisten-
cies, and considering the variety of metrics that might be 35
used to gauge performance, it is difficult to argue that any
one LEP is superior in this applied framework. Thus, using
I&B082 prediction error: LPI units

25
the LEP-specific hazard scales, and based on the prediction
errors computed herein, the performance of the LPI models
is, for practical intents and purposes, equivalent. 15
While minor errors are to be expected in any deterministic
analysis, each model produced significant errors with con-
sequences for hazard assessment. For example, even with 5
calibration, |E| exceeded 10 at 9% of sites, on average, for
each model (e.g. severe manifestation predicted, but no
manifestation observed) and |E| exceeded 5 at 22% of sites, 5 Christchurch
on average, for each model (e.g. no manifestation predicted, Darfield
but moderate manifestation observed). To determine whether 15
certain models perform better in particular locations, predic- 15 5 5 15 25 35
tion errors from the calibrated R&W98, MEA06 and MEA06 prediction error: LPI units
I&B082 models are plotted against one another in Fig. 8. It (b)
can be seen that prediction errors are generally equivalent;
in all, the difference in prediction error between any two of 35
the models exceeds 5 for only 12% of investigation sites.
Thus, locations of under-, over- and accurate prediction are
I&B082 prediction error: LPI units
generally consistent between models. In addition, maps 25

showing the spatial distributions of errors to be very similar
in both earthquakes are provided in an electronic supplement
15
to this paper. Thus, some site profiles have very poor
predictions, irrespective of the LEP used (note that Maurer
et al. (2014) found no correlation between prediction errors 5
and either PGA uncertainty, ground water fluctuation or CPT
termination depth). This suggests that LPI has inherent
limitations in its formulation, such that the variables influ- 5 Christchurch
encing surficial manifestation are not adequately accounted
for. While liquefaction triggering has garnered significant Darfield
research and is a subject of frequent debate, the mechanics 15
15 5 5 15 25 35
of liquefaction manifestation have received less attention. MEA06 prediction error: LPI units
This study highlights that triggering and manifestation are (c)
two distinct phenomena contributing to liquefaction hazard,
and that an improved framework providing clear separation Fig. 8. Comparison of LPI model prediction errors at each
and accounting of the two phenomena is needed. investigation site, as computed by: (a) R&W98 plotted against
Lastly, the 12% of cases with inconsistent prediction MEA06; (b) R&W98 plotted against I&B082; (c) MEA06 plotted
against I&B082
errors between models can be shown to correspond to
‘exceptional’ site profiles. Since the LEP-specific calibrations
are based on the entire dataset (i.e. predominant behaviour LEP-specific calibrations do not account for this divergence
across Christchurch), predictions for site profiles that diverge because the median cumulative thickness of soil strata
from typical conditions may be inconsistent among models. predicted to liquefy below 10 m depth for all the sites in the
As an example, it can be seen in Figs 8(a) and 8(c) that a dataset is only 0 .35 m, according to I&B082. This empha-
number of cases exist where the MEA06 prediction error sises that assessments and/or calibrations of the LPI hazard
significantly differs from that of R&W98 and I&B082. One scale are a function not only of the selected LEP, but also of
common cause of this discrepancy is cases where relatively the chosen dataset, including the geometry and soil charac-
thick, potentially liquefiable layers are present at depths teristics of site profiles, as well as the amplitude and dura-
greater than ,10 m. For such cases the LEPs can yield tion of ground shaking. As such, the applicability of findings
divergent FSliq and hence divergent LPI values. However, the derived herein to other datasets is unknown.
SUMMARY AND CONCLUSIONS Land Information New Zealand (LINZ) for providing the
Utilising high-quality case histories from the CES, this earthquake occurrence data and the Canterbury Geotechnical
study evaluated the performance of the R&W98, MEA06, Database and its sponsor EQC for providing the CPT sound-
I&B081 and I&B082 CPT-based LEPs, operating within the ings, lateral spread observations and aerial imagery used in
LPI framework, for assessing liquefaction hazard. The find- this study. However, any opinions, findings, and conclusions
ings are summarised as follows. or recommendations expressed in this paper are those of the
authors and do not necessarily reflect the views of NSF,
(a) For deterministic analyses, the optimum threshold LPI
ERDC, EQC, NHRP or LINZ.
values for assessing liquefaction hazard were unique to
the LEP used in the LPI framework; suggested optimum
thresholds for the CES dataset are summarised in Table 2.
SUPPLEMENTAL DATA
The use of LPI for assessing lateral spread potential is not
The following are available in an electronic supplement:
recommended.
(a) aerial images representative of the liquefaction manifes-
(b) Taking these LEP-specific threshold values into account,
tation severity classes described in Table 1; and (b) map
receiver-operating-characteristic analyses indicated that,
figures showing the spatial distribution of LPI prediction
while the models performed similarly, the I&B081 and
errors for each LPI model, for both the Darfield and
MEA06 models were respectively the most and least
Christchurch earthquakes.
efficacious.
(c) LPI probability curves were computed to assess the
significance of different optimum thresholds, and to allow
for application in probabilistic frameworks. The optimum NOTE
thresholds were shown to correspond to roughly the same Some of the data used in this study were extracted from
probability of manifestation, whereas failing to account the Canterbury Geotechnical Database (https://canterbury
for the influence of the LEP (i.e. using the Iwasaki geotechnicaldatabase.projectorbit.com), which was prepared
criterion) resulted in different risk levels for the same LPI and/or compiled for the Earthquake Commission (EQC) to
value. assist in assessing insurance claims made under the Earth-
(d ) To compare model performance in a practical setting, quake Commission Act 1993 and/or for the Canterbury
deterministic ‘prediction errors’ were computed for each Geotechnical Database on behalf of the Canterbury Earth-
case history. Using the Iwasaki criterion, the distributions quake Recovery Authority (CERA). The source maps and
of errors among LEPs varied. These distributions became data were not intended for any other purpose. EQC, CERA
more similar using the LEP-specific hazard scales given and their data suppliers, and their engineers, Tonkin &
in Table 2, which also improved the rate of accurate Taylor, have no liability for any use of these maps and data
prediction for all LEPs. or for the consequences of any person relying on them in
(e) Even with calibration, each model had significant any way.
prediction errors (e.g. severe manifestation predicted,
but no manifestation observed). This suggests that LPI
has inherent limitations in its formulation, such that the NOTATION
Ic soil behaviour type index
variables influencing surficial liquefaction manifestation
PL probability of liquefaction
are not adequately accounted for. Rf cone penetration test friction ratio
( f ) The findings presented in this study are based on a dataset rd stress reduction coefficient
from the CES; the applicability of these findings to other Vs shear wave velocity
datasets is unknown. w(z) depth weighting function given by w(z) ¼ 10 � 0 .5z
z depth below ground surface (m)
In conclusion, the following points can be made. standard deviation
(a) The risk levels corresponding to the Iwasaki criterion
varied among LEPs.
(b) Using a calibrated, LEP-specific hazard scale, the REFERENCES
performance of the LPI models was, for practical intents Anselin, L. (1995). Local indicators of spatial association – LISA.
and purposes, equivalent. Geographical Anal. 27, No. 2, 93–115.
(c) The existing LPI framework has inherent limitations such Baise, L. G., Higgins, R. B. & Brankman, C. M. (2006). Liquefac-
tion hazard mapping-statistical and spatial characterization of
that all LEPs have very poor predictions for certain soil susceptible units. J. Geotech. Geoenviron. Engng 132, No. 6,
profiles. It is unlikely that revisions of the LEPs will 705–715.
completely resolve these erroneous assessments. Rather, Bradley, B. A. & Cubrinovski, M. (2011). Near-source strong
a revised index which more adequately accounts for the ground motions observed in the 22 February 2011 Christchurch
mechanics of liquefaction manifestation is needed. earthquake. Seismol. Res. Lett. 82, No. 6, 853–865.
Bradley, B. A. (2013a). Estimation of site-specific and spatially-
distributed ground motion in the Christchurch earthquakes: Ap-
plication to liquefaction evaluation and ground motion selection
ACKNOWLEDGEMENTS for post-event investigation. Proceedings of the 19th New Zeal-
This study is based on work supported by the US National and geotechnical symposium, Queenstown, New Zealand, p. 8.
Science Foundation (NSF) grants CMMI 1030564, CMMI Bradley, B. A. (2013b). A New Zealand-specific pseudo-spectral
1407428 and CMMI 1435494 and US Army Engineer acceleration ground-motion prediction equation for active shal-
Research and Development Center (ERDC) grant W912HZ- low crustal earthquakes based on foreign models. Bull. Seismol.
13-C-0035. The third and fourth authors would like to Soc. Am. 103, No. 3, 1801–1822.
CGD (Canterbury Geotechnical Database) (2012a). Geotechnical
acknowledge the continuous financial support provided by
investigation data. Map layer CGD0010. See https://canterbury
the Earthquake Commission (EQC) and Natural Hazards geotechnicaldatabase.projectorbit.com (accessed 12/01/2012).
Research Platform (NHRP), New Zealand, for the research CGD (2012b). Aerial photography. Map layer CGD0010. See
and investigations related to the 2010–2011 Canterbury https://canterburygeotechnicaldatabase.projectorbit.com (accessed
earthquakes. The authors also acknowledge the New Zealand 12/01/2012).
GeoNet project and its sponsors EQC, GNS Science and Chung, J. & Rogers, J. (2011). Simplified method for spatial
10
evaluation of liquefaction potential in the St. Louis Area. sing liquefaction hazard. J. Geotech. Geoenviron. Engng 140,
J. Geotech. Geoenviron. Engng 137, No. 5, 505–515. No. 7, 04014032.
Cramer, C. H., Rix, G. J. & Tucker, K. (2008). Probabilistic Metz, C. E. (2006). Receiver operating characteristic analysis: a
liquefaction hazard maps for Memphis, Tennessee. Seismol. Res. tool for the quantitative evaluation of observer performance and
Lett. 79, No. 3, 416–423. imaging systems. J. Am. College Radiol. 3, No. 6, 413–422.
Cubrinovski, M. & Green, R. A. (2010). Geotechnical reconnais- Moss, R. E. S., Seed, R. B., Kayen, R. E., Stewart, J. P., Der
sance of the 2010 Darfield (Canterbury) earthquake. Bull. New Kiureghian, A. & Cetin, K. O. (2006). CPT-based probabilistic
Zealand Soc. Earthquake Engng 43, No. 4, 243–320. and deterministic assessment of in situ seismic soil liquefaction
Cubrinovski, M., Bradley, B., Wotherspoon, L., Green, R., Bray, J., potential. J. Geotech. Geoenviron. Engng 132, No. 8, 1032–
Woods, C., Pender, M., Allen, J., Bradshaw, A., Rix, G., Taylor, M., 1051.
Robinson, K., Henderson, D., Giorgini, S., Ma, K., Winkley, A., Murahidy, K. M., Soutar, C. M., Phillips, R. A. & Fairclough, A.
Zupan, J., O’Rourke, T., DePascale, G. & Wells, D. (2011). Geo- (2012). Post earthquake recovery – development of a geotechni-
technical aspects of the 22 February 2011 Christchurch earthquake. cal database for Christchurch central city. Proceedings of the
Bull. New Zealand Soc. Earthquake Engng 43, No. 4, 205–226. 15th world conference on earthquake engineering, Lisbon, Por-
Eng, J. (2005). Receiver operating characteristic analysis. Academic tugal. Tokyo, Japan: International Association of Earthquake
Radiol. 12, No. 12, 909–916. Engineering.
Fawcett, T. (2006). An introduction to ROC analysis. Pattern Papathanassiou, G. (2008). LPI-based approach for calibrating the
Recognition Lett. 27, No. 8, 861–874. severity of liquefaction-induced failures and for assessing the
Green, R. A., Allen, A., Wotherspoon, L., Cubrinovski, M., Bradley, probability of liquefaction surface evidence. Engng Geol. 96,
B., Bradshaw, A., Cox, B. & Algie, T. (2011). Performance of No. 1–2, 94–104.
levees (stopbanks) during the 4 September Mw7 .1 Darfield and Quigley, M., Bastin, S. & Bradley, B. (2013). Recurrent liquefaction
22 February 2011 Mw6 .2 Christchurch, New Zealand, earth- in Christchurch, New Zealand during the Canterbury earthquake
quakes. Seismol. Res. Lett. 82, No. 6, 939–949. sequence. Geology 41, No. 4, 419–422.
Green, R. A., Cubrinovski, M., Cox, B., Wood, C., Wotherspoon, Robertson, P. K. & Cabal, K. L. (2010). Estimating soil unit weight
L., Bradley, B. & Maurer, B. (2014). Select liquefaction case from CPT. Proceedings of the 2nd international symposium on
histories from the 2010–2011 Canterbury earthquake sequence. cone penetration testing, Huntington Beach, CA, USA, paper
Earthquake Spectra 30, No. 1, 131–153. no. 2, p. 40.
Hayati, H. & Andrus, R. D. (2008). Liquefaction potential map of Robertson, P. K. & Wride, C. E. (1998). Evaluating cyclic liquefac-
Charleston, South Carolina based on the 1986 earthquake. tion potential using cone penetration test. Can. Geotech. J. 35,
J. Geotech. Geoenviron. Engng 134, No. 6, 815–828. No. 3, 442–459.
Holzer, T. L. (2008). Probabilistic liquefaction hazard mapping. In Robinson, K., Cubrinovski, M. & Bradley, B. A. (2013). Sensitivity
Geotechnical Earthquake Engineering and Soil Dynamics IV, of predicted liquefaction-induced lateral displacements from the
ASCE Geotechnical Special Publication 181 (eds D. Zeng, 2010 Darfield and 2011 Christchurch Earthquakes. Proceedings
M. T. Manzari and D. R. Hiltunen). Reston, VA, USA: Amer- of the New Zealand Society for Earthquake Engineering Annual
ican Society of Civil Engineers. Conference, Wellington, New Zealand, p. 8.
Holzer, T. L., Bennett, M. J., Noce, T. E., Padovani, A. C. & Seed, H. B. & Idriss, I. M. (1971). Simplified procedure for
Tinsley, J. C. III. (2006a). Liquefaction hazard mapping with evaluating soil liquefaction potential. J. Soil Mech. Found. Div.,
LPI in the greater Oakland, California area. Earthquake Spectra ASCE 97, No. 9, 1249–1273.
22, No. 3, 693–708. Sonmez, H. (2003). Modification of the liquefaction potential index
Holzer, T. L., Blair, J. L., Noce, T. E. & Bennett, M. J. (2006b). and liquefaction susceptibility mapping for a liquefaction-prone
LIQUEMAP: A real-time post-earthquake map of liquefaction area (Inegol, Turkey). Environ. Geol. 44, No. 7, 862–871.
probability. Proceedings of the 8th U. S. national conference on Swets, J. A., Dawes, R. M. & Monahan, J. (2000). Better decisions
earthquake engineering (100th anniversary earthquake confer- through science. Scientific American 283, No. 4, 82–87.
ence), paper no. 89. Oakland, CA, USA: Earthquake Engineer- Toprak, S. & Holzer, T. (2003). Liquefaction potential index: field
ing Research Institute (CD-ROM). assessment. J. Geotech. Geoenviron. Engng 129, No. 4, 315–322.
Idriss, I. M. & Boulanger, R. W. (2008). Soil liquefaction during van Ballegooy, S., Cox, S. C., Thurlow, C., Rutter, H. K., Reynolds,
earthquakes, Monograph MNO-12, p. 261. Oakland, CA, USA: T., Harrington, G., Fraser, J. & Smith, T. (2014). Median water
Earthquake Engineering Research Institute. table elevation in Christchurch and surrounding area after the 4
Iwasaki, T., Tatsuoka, F., Tokida, K. & Yasuda, S. (1978). A September 2010 Darfield earthquake: Version 2, GNS Science
practical method for assessing soil liquefaction potential based Report 2014/18. Lower Hutt, New Zealand: GNS Science.
on case studies at various sites in Japan. Proceedings of the 2nd Whitman, R. V. (1971). Resistance of soil to liquefaction and
international conference on microzonation for safer construction settlement. Soils Found. 11, No. 4, 59–68.
– research and application, vol. II, pp. 885–896. Arlington, VA, Wilson, E. B. (1927). Probable inference, the law of succession,
USA: National Science Foundation. and statistical inference. J. Am. Statistical Assoc. 22, No. 158,
Kang, G. C., Chung, J. W. & Rogers, R. J. (2014). Re-calibrating 209–212.
the thresholds for the classification of liquefaction potential Youd, T. L., Hansen, C. M. & Bartlett, S. F. (2002). Revised
index based on the 2004 Niigata-ken Chuetsu earthquake. Engng multilinear regression equations for prediction of lateral spread
Geol. 169, 30–40. displacement. J. Geotech. Geoenviron. Engng 128, No. 12,
Lenz, A. & Baise, L. G. (2007). Spatial variability of liquefaction 1007–1017.
potential in regional mapping using CPT and SPT data. Soil Zou, K. H. (2007). Receiver operating characteristic (ROC) litera-
Dynam. Earthquake Engng 27, No. 7, 690–702. ture research. On-line bibliography. See http://www.spl.harvard.
Maurer, B. W., Green, R. A., Cubrinovski, M. & Bradley, B. A. edu/archive/spl-pre2007/pages/ppl/zou/roc.html (accessed 15/03/
(2014). Evaluation of the liquefaction potential index for asses- 2014).
11
Xu, X. M. et al. (2015). Géotechnique 65, No. 5, 337–348 [http://dx.doi.org/10.1680/geot.SIP.15.P.022]
Correlation between liquefaction resistance and shear wave velocity of

granular soils: a micromechanical perspective
X . M . X U � † ‡ , D. S . L I N G � , Y. P. C H E N G † a n d Y. M . C H E N �
The shear wave velocity method has become an increasingly popular means to evaluate the
liquefaction potential of granular soils. Understanding the fundamental mechanism underlying existing
empirical or semi-empirical relationships is important for better assessing their reliability. This paper
presents a particle-scale study of the correlation between cyclic resistance ratio (CRR) and the shear
wave velocity corrected for overburden stress (V s1 ). The discrete-element method was used to simulate
a series of undrained stress-controlled cyclic triaxial tests together with shear wave velocity (V s )
measurements. Discrete-element method modelling with various relative densities, confining pressures
and micro-parameters was performed under various cyclic stress ratios (CSRs), and the onset of
liquefaction was illustrated through both macroscopic and microscopic responses, for example,
inferred excess pore-water pressure, mechanical coordination number and redundancy index. The
inter-particle friction was identified as the key micro-parameter that governs the liquefaction resistance
of granular soils. A micro-scale CRR–V s1 correlation considering two independent micro-parameters,
inter-particle friction and particle shear modulus, was then obtained and further validated with the
outcomes from three dynamic centrifuge model tests performed on silica sand no. 8. This study
demonstrates that the CRR–V s1 correlation is particle specific, thus soil specific, and the particle
mechanical properties should be included in the V s -based method for future liquefaction evaluation of
granular soils.
KEYWORDS: centrifuge modelling; discrete-element modelling; particle-scale behaviour; liquefaction
INTRODUCTION geotechnical engineers a promising alternative to evaluate

The simplified method pioneered by Seed & Idriss (1971), the liquefaction resistance of granular soils. This is espe-
based on standard penetration test (SPT) data, has become cially true for sites underlain by soils that are difficult to
the standard practice for evaluating the liquefaction potential penetrate or sample (Andrus & Stokoe, 2000; Kayen et al.,
of granular soils (Youd & Idriss, 2001; Youd et al., 2001). 2013). The V s -based simplified method has attracted an
Earthquake-related soil liquefaction is discussed in relation increasing number of studies, and various ‘boundary curves’
to the cyclic stress ratio (CSR) induced by ground shaking have been relatively well established on the basis of either
at some depth in the ground field data (Robertson et al., 1992; Andrus & Stokoe, 2000;
Juang et al., 2001; Andrus et al., 2004; Juang et al., 2005;
av amax v0
CSR ¼ ¼ 0.65 rd (1) Kayen et al., 2013), or laboratory studies (Dobry et al.,
v90 g v90 1981; de Alba et al., 1984; Tokimatsu & Uchida, 1990;
Chen et al., 2005; Wang et al., 2006; Zhou & Chen, 2007;
where av is the average equivalent uniform cyclic shear Baxter et al., 2008; Zhou et al., 2010; Ahmadi & Paydar,
stress caused by the earthquake and is assumed to be 2014).
0 .65 of the maximum induced stress; g is the acceleration Note that these V s -based correlations were developed from
of gravity; v0 and v90 are total and effective vertical empirical or semi-empirical evaluation of field observations
overburden pressures, respectively; and rd is stress reduc- and laboratory test data following the general format of the
tion coefficient to adjust for the flexibility of the soil simplified procedure by Seed & Idriss (1971). Their funda-
profile. mental mechanisms are still open to question. First, the
The key issue is to characterise the capacity of soil to shear wave velocity measurements are made at small strains,
resist liquefaction based on various routinely used field or whereas pore-water pressure build-up and liquefaction are
laboratory techniques (e.g. SPT, cone penetration test (CPT), medium-to-high-strain phenomena (Roy et al., 1996).
Becker hammer test (BHT) and shear wave velocity (V s ) Whether there is a natural link between these two different
measurement). Among them, the shear wave velocity cor- characterising variables remains unknown. Second, the un-
rected for overburden stress (V s1 ) is considered to offer iqueness of this correlation for all types of granular soils is
also questionable. Although both V s and liquefaction resis-
Manuscript received 4 April 2014; revised manuscript accepted 13 tance were reported to be similarly influenced by many of
January 2015. Published online ahead of print 26 March 2015. the same macroscopic factors (Andrus & Stokoe, 2000),
Discussion on this paper closes on 1 October 2015, for further details their parametric laws may be quite different. Taking the
see p. ii. relative density for an example, previous studies showed that
� Key Laboratory of Soft Soils and Geoenvironmental Engineering of
it has a very strong effect on the liquefaction resistance
Ministry of Education, Department of Civil Engineering, Zhejiang
University, Hangzhou, China.
(Seed & Idriss, 1971), while it has a weak effect on V s : This
† Department of Civil, Environmental and Geomatic Engineering, implies that liquefaction resistance may not uniquely corre-
University College London, London, UK. late with V s for multiple soils (Baxter et al., 2008), as
‡ Department of Engineering, University of Cambridge, Cambridge, indicated by the weakness of the V s -based correlation re-
UK. ported by Liu & Mitchell (2006). More recently, Kayen et
13
XU, LING, CHENG AND CHEN
al. (2013) reported that around 50 non-liquefied sites out of METHODOLOGY AND APPROACH
a global catalogue of 422 case histories were misclassified, Sample preparation
with their data mainly located in the lower intensity zone The DEM simulations were performed using PFC3D (parti-
(i.e. cyclic stress ratio, CSR < 0.20). One possible explana- cle flow code in three dimensions) (Itasca, 2008). In this
tion for these observations is that V s -based correlations are study, around 21 000 polydisperse spherical particles with
soil specific; however, establishing individual relations would diameters ranging from 0 .15 mm to 0 .20 mm were randomly
be rather costly and time consuming (Tokimatsu & Uchida, generated in a cylindrical region (diameter 4 mm 3 height
1990; Zhou & Chen, 2007; Baxter et al., 2008; Zhou et al., 8 mm) with rigid frictionless walls. The particle size distri-
2010; Ahmadi & Paydar, 2014). bution of an assembly is shown in Fig. 1. The radius
The shear wave velocity is well known as a comprehen- expansion method was adopted to facilitate the creation of
sive representative metric for characterising the current state an initially isotropic sample. Each particle was prescribed
of granular soils, which is highly dependent on both macro- with properties including a radius, density, contact stiffness
scopic parameters (e.g. stress state, void ratio or relative and coefficient of contact friction. The Hertz–Mindlin con-
density) and microscopic parameters (e.g. particle proper- tact model was employed in this study as it is suitable for
ties, fabric and coordination number) (Hardin & Richart, simulating pressure-dependent behaviour at small strain
1963; Tatsuoka, 1999; Yimsiri & Soga, 2000; Agnolin & (Sadd et al., 1993; Itasca, 2008; Wang & Mok, 2008). The
Roux, 2007; Clayton, 2011). The liquefaction resistance in gravitational force was neglected in this analysis. Unless
a soil deposit, commonly quantified by the cyclic resistance otherwise stated, the parameters used in the model are listed
ratio (CRR), also depends on its current state (Seed & in Table 1.
Idriss, 1971; Tokimatsu & Uchida, 1990; Chen et al., 2005; Once the DEM assembly has been generated, a numerical
Wang et al., 2006; Baxter et al., 2008; Ahmadi & Paydar, servo-control mechanism specially written for the Hertz–
2014). CRR is generally estimated by performing cyclic Mindlin contact model was implemented to compress the
triaxial tests on reconstituted samples following a given specimen to reach a desired isotropic stress state. The mini-
stress path (e.g. a series of sinusoidal cyclic stress cycles), mum void ratio (emin ) was obtained by setting a low initial
and is defined as the CSR to cause initial liquefaction after value (e.g. e < 0 .333) with the inter-particle friction set to
a certain number of cycles (e.g. 15 cycles corresponding to zero at the assembly generation stage, followed by a number
an earthquake magnitude, M w ¼ 7.5). Exploring the under- of numerical cycles until the isotropic stress reached 90% of
lying fundamental mechanisms of CRR and V s at both the required stress. The inter-particle friction was then
macro and micro level is essential to establish their correla- switched to the required contact friction value and main-
tion for the purposes of liquefaction evaluation of granular tained until the assembly reached an equilibrium state (the
soils. However, it is still a great practical challenge for ratio of the mean unbalanced force to the mean contact
u c
experimenters to examine and accurately quantify the intri- force f =f < 10�3 ) at the desired stress. The maximum
cate characteristics of internal soil structure at the micro void ratio (emax ) was obtained by assigning a large initial
scale. value (e.g. e > 0 .905) with an inter-particle friction of 0 .50,
Numerical simulations using the distinct-element method and by preventing particle spin when the assembly was
(DEM) pioneered by Cundall & Strack (1979) can offer some initially generated. Particle rotation was then permitted when
micromechanical insights to understand the mechanical prop-
erties of granular soils (Thornton, 2000; Cheng et al., 2004;
100
Soga & O’Sullivan, 2010; O’Sullivan, 2011; Zhao & Guo,
2013). Its applicability to the modelling of the cyclically
induced liquefaction behaviour of granular soils shearing at 80
constant volume has been demonstrated (Shafipour &
Percent finer by weight: %
Soroush, 2008). Ng & Dobry (1994) were among the first

researchers who studied the responses of both two-dimesional 60
(2D) disc and three-dimensional (3D) sphere assemblies in
undrained cyclic simple shear loading conditions. Their
results qualitatively agreed with physical tests on sand. 40
Recently, although there have been a number of both 2D
(Sitharam, 2003) and 3D (O’Sullivan et al., 2008; Sitharam
et al., 2009; Soroush & Ferdowsi, 2011) numerical simula- 20
tions that have investigated liquefaction phenomenon based
on strain-controlled loading, the authors are not aware of any
0
DEM studies that have explored the micromechanics both of 0·01 0·1 1
liquefaction and of CRR in a stress-controlled manner, and Diameter: mm
further developed a CRR–V s1 correlation by microscopically Fig. 1. Particle size distribution selection for DEM simulations;
measuring the V s of granular soils. inset diagram is the initially generated sample of spheres
Following similar procedures to laboratory tests, this paper
presents numerical simulations using a particle-scale DEM
investigation of the CRR–V s1 correlation, by conducting a Table 1. Model parameters used in DEM simulations
series of undrained stress-controlled cyclic triaxial test simu-
lations, together with V s measurement. DEM specimens with Parameters Value
various initial relative density, confining pressure and micro-
parameters are tested under various cyclic stress amplitudes. Particle shear modulus, Gg : GPa 1 .0
The onset of liquefaction is illustrated through both macro- Particle friction coefficient, g 0 .50
scopic and microscopic responses. The key micro-parameters Particle Poisson ratio, vg 0 .20
that govern the magnitude of CRR are identified, and a Particle density, rg : kg/m3 2 630
Particle diameter, d: mm 0 .15,0 .20
micro-scale CRR–V s1 correlation is then proposed and Model height, H: mm 8
further validated through the results from three dynamic Model diameter, D: mm 4
centrifuge model tests with silica sand no. 8.
14
LIQUEFACTION RESISTANCE AND SHEAR WAVE VELOCITY OF GRANULAR SOILS
the isotropic stress level reached 90% of the required stress. Sitharam et al., 2009; Yimsiri & Soga, 2010). The excess
For the other relatively loose samples, the inter-particle pore-water pressure ratio is then defined as U ¼ u= m 9 0,
friction coefficients were adjusted by trial and error during where m 9 0 is the initial mean confining pressure. Initial
the assembly generation stage to facilitate the sample reach- liquefaction is said to have occurred when the effective
ing the desired void ratio. Similar procedures for sample stress becomes zero (U ¼ 1.0) owing to the build-up of
preparation can be found in Salot et al. (2009), Thornton & excess pore-water pressure.
Zhang (2010) and Gong et al. (2012). In this study, the emax
and emin at 100 kPa confining pressure are 0 .809 and 0 .585,
respectively. These values are very close to that of the Shear wave velocity measurement
traditional random loose packing (RLP) and random close The measurement of shear wave velocity was implemen-
packing (RCP) for monodisperse spheres, where the emax ted using a pair of virtual ‘disc elements’ fully embedded in
and emin are 0 .818 and 0 .577 (Song et al., 2008; Silbert, the sample, as shown in Fig. 2(a). In the DEM model, a
2010). The relative density (Dr ) of the DEM specimens is group of particles (a circular disc with around 100 particles)
expressed as in a certain region were selected as a transmitter and another
emax � e one as a receiver. The excitation wave was created by
Dr ¼ (2) applying a velocity pulse to the transmitter in the x direc-
emax � emin
tion, which would immediately be transferred to neighbour-
where e is the void ratio after isotropic consolidation. ing particles and subsequently propagated through the whole
sample. The wave propagation could be visualised by way of
the velocity field. Once the disturbance reached the receiver,
Undrained stress-controlled cyclic triaxial test the average velocity in the same direction would be picked
After isotropic consolidation, a series of undrained cyclic up. Fig. 2(b) plots the normalised velocity–time histories of
triaxial tests with various CSRs were simulated in a stress- both the excitation and received waves. The latter is scaled
controlled manner for each numerical specimen, by applying up by five times for better visualisation. The Cartesian
a number of sinusoidal cyclic stress loadings until initial coordinate system oxyz, as shown in Fig. 2(a), is used in this
liquefaction occurred. The velocities of the boundaries were paper for all cases analysed in 3D space. The longitudinal
adjusted in such a way that the cyclic deviator stress direction of the numerical sample is chosen as the z-axis,
followed a sinusoidal cyclic stress history while the specimen and the x-axis and the y-axis directions are the radial
volume remained constant. The input frequency of the cyclic directions.
loading was chosen as 1000 Hz in this study. We performed Note that before performing the shear wave propagation,
a comparison among the cyclic-induced boundary responses it is essential to ensure that the sample has reached a further
u c
and backbone curves of another numerical sample (with equilibrium state (f =f < 10�9 ). To do this, the coefficient
fewer particles) at various frequencies (1 Hz , 5000 Hz). of local damping was set to a relatively high level (e.g. 0 .9)
The results indicated that the cyclic behaviours of the DEM with the purpose of dissipating energy more efficiently and
specimen are essentially not sensitive to the input frequency saving computational time. After more numerical cycles, the
when its value is less than 2000 Hz. This threshold is higher velocities of the particles would eventually become low
than the experimental findings from a cylindrical sand sam- enough compared to the excitation magnitude. The damping
ple 38 mm in diameter and 78 mm high (Bolton & Wilson, factor was then reduced to zero at the end of this stage for
1989) owing to the much smaller size of the DEM specimen the shear wave analysis.
in the present study. These simulations were performed using The V s is calculated using the wave travel time (t) and
four workstations (each with 3 .2 GHz Intel CPU and 8– the distance of the travel path (LTR ), in exactly the same
32 GB memory) and one computer cluster (with eight nodes, way as it is in a laboratory test with bender elements
each with 4 3 2 .26 GHz Intel CPU and 16 GB memory) over (Clayton, 2011).
a few years. LTR
The equivalent excess pore-water pressure (u) is evaluated Vs ¼ (3)
under an assumption of fully saturated conditions, by taking t
the difference between the initial effective confining pressure The travel distance of the shear wave component is
( r90 ) at the beginning of shearing and the current effective generally taken as the tip to tip distance between the
stress ( r9), that is, u ¼ r90 � r9 (e.g. Ng & Dobry, 1994; transmitter and receiver. However, the determination of the
1·3
f 125 kHz
Normalised shear wave velocity, Vs /Vs0
f 200 kHz
1·2
z
1·0
Received wave (5)
0·5 1·1
Excitation wave
Normalised amplitude
Receiver 0
1·0
0·5
1·0
0·9
0·5
0 0·8
Transmitter
x 0·5
o 1·0 0·7
y 0 5 10 15 20 25 30 1 1010 1 108 1 106 1 104 1 102
Time, t: μs 1 109 1 107 1 105 1 103 1 101
(a)
(b)
Ratio of the excitation amplitude to wave length, Adisp /λ
(c)
Fig. 2. Measurement of shear wave velocity: (a) half of the DEM model for wave propagation; (b) transmitted wave and received wave;
(c) choice of velocity amplitude for shear wave excitation
15
travel time is much less straightforward (Jovicic et al., 1996; Cyclic-induced liquefaction: from macro to micro
Viana da Fonseca et al., 2009). A number of methods have The macroscopic liquefaction phenomena are observed for
been commonly used, ranging from the simplest approach all numerical simulations in this study. A typical example is
based on the immediate observation of the wave traces (e.g. presented in Fig. 3, which shows the variation of axial strain
first arrival and peak to peak), to more elaborate signal and excess pore-water pressure ratio U with the number of
processing techniques (e.g. cross-correlation analysis, wave- cycles (N ) for an isotropically consolidated sample at a
let analysis and phase detection analysis). In this study, the confining pressure of 100 kPa (Dr ¼ 50.4%, CSR ¼ 0.125).
cross-correlation analysis was adopted, owing to its super- The inset of this figure illustrates the stress-controlled devia-
iority in both determining the travel time and identifying the tor stress during the cyclic loading. As observed from the
similarities between two signals (Santamarina & Fratta, figure, with the application of constant cyclic stress, both the
1998). cyclic axial strain magnitude and excess pore-water pressure
The velocity amplitude used in this paper for the shear ratio only increase slightly after the first 50 cycles; the axial
wave propagation was selected by comparing the measured strain magnitude then increases dramatically when the nu-
V s from the above-mentioned wave propagation, to the merical specimen approaches initial liquefaction after 64
actual value V s0 interpreted from a boundary p measurement
ffiffiffiffiffiffiffiffiffiffiffi cycles, owing to the development of excess pore-water
of the small strain shear modulus G0 (V s0 ¼ G0 =r). Fig. pressure. In a qualitative sense, this simulation captures the
2(c) plots the variation of the normalised shear wave velo- realistic behaviour of sand liquefaction phenomenon ob-
city V s =V s0 against the ratio of the excitation amplitude to served in laboratory experiments (Seed & Lee, 1966; Zhou
the wave length Adisp =º: When the excitation amplitude is & Chen, 2005).
too high, V s is smaller than V s0 : When Adisp =º reduces to To explore the underlying micro-mechanism of liquefac-
10�5, then shear wave velocity becomes very close to the tion, the mechanical coordination number (C n ) proposed by
boundary measurement (V s =V s0 ¼ 1). In the figure, the value Thornton (2000) is adopted herein. It is calculated as an
indicated by the single star point (Adisp =º ¼ 10�8 ) is used in average coordination number, but excludes particles with
this paper. To ensure a linear elastic wave propagation with- only one or zero contacts that are not contributing to the
out inducing any frictional work, the authors suggest the stable state of stress. It is expressed as
ratio of the displacement amplitude of the excitation wave to 2N c � N b1
the mean particle diameter Adisp =d 50 , 10�4 (Xu et al., Cn ¼ (4)
2012) or Adisp =º , 10�5 , according to Fig. 2(c). N b � N b0 � N b1
where N b and N c are the total number of particles and
contacts respectively; N b1 and N b0 are the number of
RESULTS AND OBSERVATIONS particles with only one or no contacts, respectively. Fig. 4
Over 108 numerical samples were cyclically sheared from plots the evolution of C n with the number of cycles towards
different initial states (i.e. in terms of relative density, initial initial liquefaction. Note that the data are taken at the end of
confining pressure, particle shear modulus, particle Poisson each cycle (see the inset of Fig. 3), except that more data
ratio and inter-particle friction) to initial liquefaction under points are taken in the last incomplete cycle when the
various CSRs, where the excess pore-water pressure ratio U specimen was very close to initial liquefaction. After the
reaches 1 .0. The onset of liquefaction is illustrated through first cycle, C n decreases from 4 .83 to 4 .79, and then remains
both macroscopic and microscopic criteria. The key micro- almost constant until the 50th cycle. After that, the number
parameters that govern the magnitude of CRR are identified, decreases sharply to about 4 .0, which is the minimum
and a micro-scale CRR–V s1 correlation is then proposed and requirement for a stable three-dimensional deposit of fric-
further verified. tional spherical particles (Edwards, 1998).
2·0 1·2
30
Dr 50·4%
Deviator stress, σd: kPa
σm0 100 kPa 20

1·5 1·0
CSR 0·125 10
Nl 64
0
1·0
Excess pore-water pressure ratio, U
10 0·8
20
0·5
Axial strain, εa: %
30
0 1 2 3 4 60 61 62 63 64 65 0·6
Number of cycles, N
0
εa 0·4
0·5
0·2
1·0
U
0
1·5 Axial strain
Excess pore-water pressure ratio
2·0 0·2
0 10 20 30 40 50 60 70
Number of cycles, N
Fig. 3. Variation of axial strain and excess pore-water pressure ratio with number of cycles for a
confining pressure of 100 kPa (Dr 50.4%, CSR ¼ 0.125); inset diagram shows the stress-
controlled cyclic loading
16
5·0 30 1·0
Contact normal anisotropy, ard
Deviator anisotropic coefficients, adr , adn, adt

Mechanical coordination number, Cn
25 Normal contact force anisotropy, and

4·8 0·8
Tangential contact force anistropy, adt
Sliding fraction, Sc: %

20
Mechanical coordination number
4·6 0·6
Sliding fraction
15
4·4 0·4 Dr 50·4%; σm0
100 kPa
Dr 50·4%; σm0
100 kPa 10
CSR 0·125; Nl 64 CSR 0·125; Nl 64 ard 0·239
4·2 0·2
5 and 0·104
4·0 0 0 adt 0·073

0 10 20 30 40 50 60 70
Number of cycles, N 0 10 20 30 40 50 60 70 80
Number of cycles, N
Fig. 4. Evolution of mechanical coordination number and sliding
fraction with the number of cycles for a confining pressure of Fig. 5. Variation of deviator anisotropic coefficients with number
100 kPa (Dr 50.4%, CSR 0.125) of cycles for a confining pressure of 100 kPa (Dr 50.4%,
CSR 0.125)
During the liquefaction process, the number of inter-

particle contacts changes as a result of contact destruction can be observed from the figure that all three coefficients
and contact reorientation. The former can be characterised by ard , and and atd increase (from their initial values of 0 .14, 0 .09
the contact sliding fraction (S c ), which is defined as the ratio and 0 .03, respectively) dramatically (to 0 .239, 0 .104 and
of sliding contact number to the total number of contacts. 0 .073, respectively) when initial liquefaction occurs. The
Fig. 4 also shows the variation of S c with the number of final magnitude of each coefficient indicates that the geome-
cycles. At the beginning of the shearing, the magnitude of S c trical anisotropy (ard ¼ 0.239) at liquefaction dominates the
is only 2 .09%, attributed to the isotropic consolidation, and it mechanical anisotropy (and ¼ 0.104; atd ¼ 0.073), which is
increases to 5 .31% after the first cycle. It then increases the signature of anisotropy at liquefaction (Guo & Zhao,
rapidly from 6 .57% at the 50th cycle up to 29 .5% at the end 2013).
of shearing. At this stage, the majority of the stored energy Figure 6 plots the 2D visualisation (in the xz plane) of the
in the sample is dissipated. Moreover, the redundancy index contact force network changing as initial liquefaction ap-
(I R ) can also be used to illustrate the occurrence of liquefac- proaches. The thickness of the lines is proportional to the
tion. It is defined as the ratio of the number of constraints to magnitude of the contact forces of each contact. As the
the number of degrees of freedom in the system, and ex- number of cycle increases, the contact force network gradu-
pressed as I R ¼ C n (3 � 2S c )=12 by Gong et al. (2012). If ally becomes looser and weaker. The value of average
I R . 1, this indicates a redundant system, otherwise there is normal contact force is 3.49 3 10�3 N at the initial state,
a non-redundant one. In this study, for the sample shown in and reduces to 3.78 3 10�4 N at the liquefaction state, where
Fig. 4, a redundant system is identified at the initial state the contact force network becomes insufficient to sustain any
(I R ¼ 1.190), and an unstable state is recognised when initial more shearing. Fig. 7 illustrates the influence of the CSR on
liquefaction occurs (I R ¼ 0.809). the evolution of the mechanical coordination number to-
To further assess the evolution of contact reorientation wards initial liquefaction. The number of cycles is normal-
during the cyclic loading, three deviator anisotropy coeffi- ised by the number of cycles when initial liquefaction
cients are used in contact normals (ard ), normal contact occurs (N l ). Both cases show that C n remains fairly constant
forces (and ) and tangential contact forces (atd ), which are until N =N l . 0.6:
based on a second-order fabric tensor, a second-order normal
contact force tensor and a second-order tangential contact
force tensor, respectively (Chantawarangul, 1993; Sitharam Liquefaction resistance and shear wave velocity
et al., 2009). For an isotropic assembly, all these three Figure 8 shows the relationship between CSR and N l for
coefficients are zero. The system is anisotropic otherwise. various relative densities at a confining pressure of 100 kPa,
Fig. 5 presents their evolution with the number of cycles. It together with V s measurement at the initial isotropic state.
o
x
(a) (b) (c) (d) (e)
Fig. 6. Contact force network at initial state, 50th, 60th and 64th cycles and initial liquefaction
( f cmax 0.02078N) (Dr 50.4%, CSR ¼ 0.125): (a) N 0; (b) N 50; (c) N 60; (d) N 64;
(e) N 64 .5
17
5·0 0·5
CSR 0·125, Nl 64 σm0 50 kPa
CSR 0·150, Nl 20
Mechanical coordination number, Cn
σm0 100 kPa
4·8 0·4
Cyclic resistance ratio, CRR

Fitting curve
4·6 0·3
Liquefaction
4·4 0·2
4·2 0·1 Non-liquefaction
Dr 50·4%; σm0 100 kPa

4·0 0
0 0·2 0·4 0·6 0·8 1·0 0 30 60 90 120 150
Normalised number of cycles, N/Nl Stress-corrected shear wave velocity, Vs1: m/s
Fig. 7. Influence of cyclic stress ratio on the evolution of mech- Fig. 9. Correlation between liquefaction resistance and shear
anical coordination number for a confining pressure of 100 kPa wave velocity
(Dr 50.4%)
spheric pressure (typically 100 kPa); and V s1 is the over-
0·5 burden stress-corrected shear wave velocity, which is
Dr 61·2% expressed as (Robertson et al., 1992)
Dr 55·8% � �n=2
0·4 Vs 109·6 m/s Dr 52·7% Pa
Nl 15 V s1 ¼ V s (6)
� v9
Cyclic stress ratio, CSR
Dr 50·4%
Dr 50·0%
0·3
Vs 104·8 m/s Dr 39·3% where � v9 is the overburden stress and n is the power
Dr 17·4% exponent. The value of n is 0 .386 for the present DEM
Vs 99·1 m/s simulations (Xu et al., 2013), which is determined by
0·2
Vs 96·1 m/s σm0 100 kPa performing the shear wave velocity measurement on a num-
Vs 91·1 m/s ber of samples at different initial states, such as void ratio
Vs 93·0 m/s and confining pressure. It is lower than that of real sand
0·1
(typically n¼0.5), owing to the idealised sphere–sphere
Vs 82·4 m/s Hertzian contact characteristics in this study, rather than the
0 non-spherical contacts in real sand particles (Santamarina &
1 10 100 1000
Number of cycles to liquefaction, Nl Cascante, 1996).
Fig. 8. Curves of CSR against number of cycles to liquefaction for

a confining pressure of 100 kPa Particle-specific CRR–Vs1 correlation
In order to further explore the fundamental micro-mechan-
ism of the CRR–V s1 correlation, the influence of the particle
From the figure, it is evident that the CSR curves move mechanical properties on both liquefaction resistance and
downwards significantly as Dr reduces, which means that a shear wave velocity were examined. The benchmark speci-
much smaller CSR is required to liquefy a looser soil men was selected as the one isotropically consolidated to a
sample. This suggests that the DEM simulations have cap- confining pressure of 100 kPa, with relative density of
tured quantitatively the effect of Dr on liquefaction potential 50 .4%. Particle properties listed in Table 1, such as particle
observed in laboratory experiments (e.g. Zhou et al., 2010). shear modulus, particle Poisson ratio or friction coefficient,
The magnitude of CRR is taken as the value of CSR when were replaced by various values at the end of the isotropic
N l equals 15 for each specimen. As expected, CRR also consolidation stage, and then numerically cycled until the
decreases as Dr decreases. It is worth noting that, in Fig. 8, new specimen reach its equilibrium state. During this pro-
the single point specified by a dotted circle corresponds to cess, no significant changes in the void ratio, mechanical
the results previously presented in Figs 3–6. coordination number and contact normal anisotropy were
To develop the liquefaction evaluation chart, the CRRs found. After that, these samples were brought to initial
obtained from Fig. 8 are plotted against the converted stress- liquefaction as before.
corrected shear wave velocity measured at the initial state of Figure 10 shows the effects of particle Poisson ratio vg ,
each sample in Fig. 9. The data from tests under a different shear modulus Gg and friction coefficient �g on CSR–Nl
confining pressure (� m 9 0 ¼ 50 kPa) are also added to the curves. It is evident from Fig. 10(a) and Fig. 10(b) that both
figure. All the data fall into a very narrow band. This Gg and vg have negligible influence on CRR, as all these
demonstrates that there exists a unique CRR–V s1 correlation curves almost overlap. For instance, the values of CRR are
for a granular soil at the macro scale. The data are then 0 .159, 0 .167 and 0 .162 when the magnitudes of Gg are set
fitted using the following equation as 1 GPa, 5 GPa and 10 GPa, respectively. However, a sig-
!�=2 nificant effect of �g on CRR is found in Fig. 10(c). With the
rg (V s1 )2 increase of particle friction coefficient from 0 .3 to 0 .9, the
CRR ¼ Æ (5) magnitude of CRR increases 3 .65 times, from 0 .074 to
Pa
0 .270. The above observations imply that the inter-particle
where Æ ¼ 7.218 3 10�8 and � ¼ 5.339 (with the fitting friction is the governing micro-parameter for liquefaction
correlation coefficient R2 ¼ 0.98) are material constants resistance. This is understandable as soil liquefaction is a
related to the particle mechanical properties; Pa is the atmo- large deformation process with energy dissipation (e.g.
18
0·5 60
Mobilised angle of internal friction, φ: degrees

Dr 50·4%; σm0
100 kPa Thornton (2000) (φcv) DEM results (φmax)
Gg 1 GPa; μg 0·50 Thornton (2000) (φmax) DEM results (φcv)
50
0·4 υg 0·10 Yimsiri & Soga (2010) (dense, φmax)
υg 0·20 Yimsiri & Soga (2010) (loose, φmax)

υg 0·30 40
0·3
φmax 34·4°
30
0·2
20
0·1
Nl 15 10 Silica sand no. 8
φg 39·9°
0 0
1 10 100 1000 0 10 20 30 40 50 60
Number of cycles to liquefaction, Nl Inter-particle friction angle, φg: degrees
(a)
0·5
Fig. 11. Relationship between mobilised angle of internal friction
Dr 50·4%; σm0
100 kPa
and inter-particle friction angle
υg 0·20; μg 0·50
0·4 Gg 1 GPa
Gg 5 GPa
Gg 10 GPa clearly shown that both �max and �cv from the current DEM
0·3 simulations are in good agreement with those of the pre-
vious research, and the �max –�g relationship for the present
DEM assemblies with spherical particles can be expressed
0·2
as
�max ¼ a(�g )b (7)
0·1
Nl 15
where a ¼ 10.121 and b ¼ 0.332 (for �g ranging from 0 .1
0 to 0 .9).
1 10 100 1000
Number of cycles to liquefaction, Nl
For the shear wave velocity of granular soils in the initial
(b) state, although not presented here, the current DEM simula-
0·5 tions with particle shear modulus ranging from 1 .0 GPa to
Dr 50·4%; σm0
100 kPa 50 GPa indicate that the relationship between V s and Gg
Gg 1 GPa; υg 0·20 follows a power law, with a power index of m ¼ 0.320 (Xu,
0·4
μg 0·30 2012). The influence of particle Poisson ratio vg on V s is
μg 0·50 found to be negligible (e.g. V s only increases from 92 .17 m/s

0·3
μg 0·70 to 94 .68 m/s when vg varies from 0 .1 to 0 .4 with
μg 0·90 Dr ¼ 50.4% and � m 9 0 ¼ 100 kPa). No effect of inter-particle
friction �g on V s was discovered, as the V s measurement of
0·2 the present study is within the elastic range. All these
observations are consistent with the previous theoretical re-
search based on various hypotheses (Chang et al., 1991;
0·1 Santamarina & Cascante, 1996; Yimsiri & Soga, 2000).
It can be concluded that two independent micro-param-
Nl 15
eters, the inter-particle friction and the particle shear mod-
0
1 10 100 1000 ulus, govern the liquefaction resistance and the shear wave
Number of cycles to liquefaction, Nl velocity of granular soils, respectively. This implies that the
(c) previous CRR–V s1 correlation shown in Fig. 9 should also
Fig. 10. Influence of particle mechanical properties on liquefac- be dependent on particle properties, rather than being a
tion resistance: (a) particle Poisson ratio; (b) particle shear unique relationship for all types of granular soils.
modulus; (c) inter-particle friction Hence, the present authors carried out some extra DEM
simulations to demonstrate the existence of particle-specific
CRR–V s1 correlations, by varying the particle shear modulus
through contact sliding, as shown in Fig. 4), which is highly and friction coefficient. Fig. 12 presents the influence of
related to the contact frictional property. The higher the particle shear modulus on the CRR–V s1 correlation. The
value of �g , the more difficult it is to induce contact sliding curves are the predictions from equation (5), except that the
(Thornton, 2000), resulting in a higher CRR. These findings values of Æ are reduced based on the V s –Gg power law.
suggest that CRR is highly dependent on the inter-particle Interestingly, this indicates that they are in good agreement
mobility (i.e. void ratio, coordination number, confining with the results from DEM simulations at various Gg levels.
pressure and inter-particle friction), rather than the deform- For the effect of inter-particle friction, the obtained DEM
ability of the soil particle itself. data are plotted in Fig. 13, and best fitted using equation (5)
Figure 11 presents the correlation between the mobilised by keeping Æ constant. The fitting correlation coefficients
angle of internal friction � (both at peak �max and critical for both cases (�g ¼ 0.3 and �g ¼ 0.7) are higher than 0 .90,
state �cv ) and inter-particle friction angle �g obtained from with � equal to 5 .135 and 5 .440, respectively. Therefore,
a series of drained DEM simulations at a confining pressure from the present DEM simulations, the CRR–V s1 correlation
of 100 kPa, together with some other results from the can be reasonably characterised by equation (5), where the
literature (Thornton, 2000; Yimsiri & Soga, 2010). It is coefficient Æ is dependent on Gg for a given �g , and the
19
0·5
Prediction from equation (5) (Gg 1 GPa)

DEM results (Gg 2 GPa)
0·4 DEM results (Gg 5 GPa)
DEM results (Gg 10 GPa)
� 5·339
0·3
0·2
α 2·205 108 α 4·600 109 α 1·406 109

α 7·218 108
0·1
Vg 0·20; μg 0·50
0
50 100 150 200 250
Stress-corrected shear wave velocity, Vs1: m/s
Fig. 12. Influence of particle shear modulus on CRR–V s1 correlation
0·5
DEM results (μg 0·3) and the results from three dynamic centrifuge model tests
DEM results (μg 0·5) � 5·440
(cases A, B and C) performed on silica sand no. 8 (Zhou et
0·4
al., 2010). These tests were conducted in a laminar box,
DEM results (μg 0·7)
� 5·339 together with three pairs of bender elements, two pairs of

Fitting by equation (5) (μg 0·3)
earth pressure meters, eight miniature accelerometers, six
Fitting by equation (5) (μg 0·5) miniature pore pressure transducers and three laser displace-
0·3 � 5·135
Fitting by equation (5) (μg 0·7) ment transducers. The silica sand no. 8 is poorly graded,
α 7·218 108 with a mean diameter of 0 .084 mm (emax ¼ 1.381,
0·2 emin ¼ 0.721). The peak internal friction angle �max is 34 .48.
Liquefaction
The relative density of the soil used in the dynamic centri-
fuge model varies from 61 .7% to 89 .8%. From these tests,
0·1 Non-liquefaction
84 liquefied data sets and 120 non-liquefied data sets were
Gg 1 GPa; υg 0·20
produced.
0
As stated previously, the use of equation (8) to character-
0 30 60 90 120 150 ise the CRR–V s1 correlation requires two micro-parameters:
Stress-corrected shear wave velocity, Vs1: m/s inter-particle friction and particle shear modulus. Direct
measurement of these particle properties is not easy to
Fig. 13. Influence of inter-particle friction on CRR–V s1 correla-
achieve. Instead, one can calibrate them through establishing
tion
the micro–macro relationship, as shown in Fig. 11. Based on
this �max –�g relationship, equation (7), the value of �g
power index � is determined by �g : This can be expressed suitable for the silica sand no. 8 (�max ¼ 34 .48) is obtained
explicitly as as 39 .98 (�g ¼ 0.836), leading to � ¼ 5.483 by way of
" #�=2 equation 8(b). For the particle shear modulus, it is calculated
Æ0 rg (V s1 )2 by comparing the V s obtained from the lab testing using
CRR ¼ (8a) bender elements and the DEM V s measurements from the
(Gg =Gg0 )m� Pa
present study at a similar relative density range, and by
� ¼ B arctan (A�g ) þ C (8b) assuming that the real sand follows the same V s –Gg power
law. This is reinforced by the fact that both sets of the V s
where Gg0 ¼ 1 GPa; Æ0 ¼ 7.218 3 10�8 ; A ¼ 4.890; data were found to be linearly related to Dr with a similar
B ¼ 0.971; and C ¼ 4.191: These two parameters may be slope. Using this method, the value of Gg was calibrated as
dependent on particle shape and particle size distribution. 6 .649 GPa for the silica sand no. 8 (Xu, 2012). This meas-
This requires further investigation. ured value is around four times lower than most published
DEM work (Thornton, 2000; Thornton & Zhang, 2010;
Gong et al., 2012; Huang et al., 2014). This may be due to
VALIDATION AND DISCUSSION particle roughness and the imperfect shape of real sand
The present authors attempted to validate the above particles (Chang et al., 1991).
observation by comparing the prediction from equation (8) Figure 14 presents the predicted CRR–V s1 curve from
20
0·9
(1) Andrus & Stokoe (2000) (1) (5)
(4)
(2) Zhou et al. (2010)
(3) Ahmadi & Paydar (2014) (Babolsar sand)

(2)
(4) Ahmadi & Paydar (2014) (Firoozkooh sand)
(5) Present study

0·6
(3)
Equation (8)
CSR or CRR
Gg 6·649 GPa; μg 0·836
Case A (Liq)
Case A (Non-liq)
0·3
Case B (Liq)
Case B (Non-liq)
Case C (Liq)
Case C (Non-liq)
0
50 100 150 200 250
Corrected shear wave velocity, Vs1: m/s
Fig. 14. Validation on the CRR–V s1 correlation (after Zhou et al., 2010)
equation (8) using both the calibrated parameters described CONCLUSIONS

above and the 204 CSR data sets from the dynamic centri- This study has explored the fundamental micromechanics
fuge tests (Zhou et al., 2010). The latter were converted of liquefaction resistance and its correlation with the shear
from a series of irregular time histories recorded in centri- wave velocity of granular soils, by performing a series of
fuge tests according to equation (1). Liquefied data are undrained, stress-controlled, cyclic triaxial tests on DEM
represented by solid symbols in the figure. Non-liquefied specimens with various relative density, confining pressure
data are represented by hollow symbols. As shown in Fig. and particle mechanical properties, together with V s meas-
14, the predicted curve separates almost all (about 92%) of urement at the initial state. A number of novel findings were
the liquefied data properly, and even for the non-liquefied obtained, and these are summarised as follows.
data more than 78% of them are correctly classified. This is
very similar to the previous evaluation done by Zhou et al. (a) The CRR–V s1 correlation of granular soils is particle
(2010). For the misclassification of the non-liquefied data, specific, and thus soil specific, and is governed by two
Zhou et al. (2010) had clarified that the initial liquefaction independent micro-parameters, inter-particle friction and
criterion is not always applicable to dense sand. They did particle shear modulus. The proposed micro-scale CRR–
observe some considerable settlement after dissipation of V s1 correlation, with two independent coefficients Æ and
pore-water pressure, although these data could not be classi- � explicitly relating to the particle shear modulus and
fied as liquefied according to the initial liquefaction criter- inter-particle friction, respectively, was validated by the
ion. While the correlation proposed by Andrus & Stokoe outcomes from three dynamic centrifuge model test
(2000) considerably underestimates the liquefaction resis- performed on silica sand no. 8 (Zhou et al., 2010). This
tance of silica sand no. 8 within the low-velocity range, it study provides clear, yet simple, micromechanical
overestimates the resistance in the high-velocity range, as evidence that current liquefaction evaluation with Vs
shown in Fig. 14. The more recent relationship established alone is not sufficient, and could lead to overestimation
by Ahmadi & Paydar (2014) from two types of clean sands, of CRR. The effect of inter-particle friction should also
namely the Firoozkooh and the Babolsar sands, are also be taken into account to consider the large deformation
shown in Fig. 14, and the one based on the Babolsar sand process of liquefaction in granular soils.
gives much less satisfactory evaluation than that of the (b) The key micro-parameters that govern the liquefaction
Firoozkooh sand. These observations strongly support the resistance of granular soils were identified. A systematic
fact that the proposed CRR–V s1 , which separates the lique- parametric study on DEM samples with various particle
fied and the non-liquefied data, is soil type specific. mechanical properties indicated that it is the inter-particle
This DEM study supports the use of equation (8) and mobility (e.g. inter-particle friction), rather than the
suggests that the peak internal friction angle at the dense deformability of the soil particle itself (e.g. particle shear
state can potentially be used to obtain an appropriate � modulus and Poisson ratio) that determines the magni-
value, where the Æ value can be obtained from the shear tude of CRR of granular soils. The higher the value of
wave velocity measurement. A satisfactory evaluation of the inter-particle friction, the more difficult it was to induce
liquefaction potential of granular soil can then be obtained. contact sliding, giving higher CRR. A micro–macro
21
relationship between the inter-particle friction and the Nl number of cycles to cause initial liquefaction
internal friction angle at peak was also developed. n power exponent in equation (6)
(c) The micromechanics of the initial liquefaction process Pa atmospheric pressure
based on stress-controlled loading were investigated. The rd stress reduction coefficient
numerical data showed that all of the relevant Sc contact sliding fraction
t wave travel time
micromechanical parameters, such as the mechanical U excess pore water pressure ratio
coordination number, the sliding fraction, the deviator u excess pore water pressure
anisotropic coefficients and even the force network, Vs shear wave velocity;
remained almost constant until the number of cycles N V s0 shear wave velocity interpreted from a boundary
reached a relatively high portion (. 0 .6) of the total measurement
number of cycles at which initial liquefaction occurred V s1 overburden stress-corrected shear wave velocity;
N l : The onset of liquefaction is fairly sudden, leading to a Æ material constant in equation (5)
dramatic change of the micro-parameters. When initial Æ0 material constant in equation (8a)
liquefaction occurs, the mechanical coordination number � material constant in equation (5)
�a axial strain
reduces to 4 .0, redundancy index becomes smaller than
º wave length in Fig. 2(c)
1 .0, and geometrical anisotropy dominates mechanical �g inter-particle friction coefficient
anisotropy. ıg particle Poisson’s ratio
r sample density
rg particle density
ACKNOWLEDGEMENTS �m9 0 mean effective stress
Much of the work described in this paper was supported � r9 radial effective stress
by the National Basic Research Program of China (grant no. � r90 radial effective stress after isotropic consolidation
2014CB047005), the National Natural Science Foundation of � v9 overburden stress
� v0 total vertical overburden stress
China (grant no. 51278451), the Zhejiang Provincial Natural
� v90 effective vertical overburden stress
Science Foundation of China (grant nos. LZ12E09001 and �av average equivalent uniform cyclic shear stress
LY12E08011) and Erasmus Mundus External Cooperation � internal friction angle
Window (EM ECW) scholarship. These financial supports �cv internal friction angle at critical state
are gratefully acknowledged. The authors thank Professor �g inter-particle friction angle
Kenichi Soga and Professor Malcolm Bolton at University of �max peak internal friction angle
Cambridge, Professor J. Carlos Santamarina at Georgia
Institute of Technology and Professor M A Curt Koenders at
Surrey University for their valuable exchanges of ideas and
help with this study.
REFERENCES
Agnolin, I. & Roux, J.-N. (2007). Internal states of model isotropic
granular packings. III. Elastic properties. Phys. Rev. E 76, No.
NOTATION 6, 061304.
A coefficient in equation (8b) Ahmadi, M. M. & Paydar, N. A. (2014). Requirements for soil-
Adisp displacement of the excitation wave specific correlation between shear wave velocity and liquefaction
a coefficient in equation (7) resistance of sands. Soil Dynam. Earthquake Engng 57, 152–
and normal contact force anisotropy coefficient 163.
ard contact normal anisotropy coefficient Andrus, R. D. & Stokoe, K. H. I. (2000). Liquefaction resistance of
atd tangential contact force anisotropy coefficient soils from shear-wave velocity. J. Geotech. Geoenviron. Engng
amax peak horizontal ground surface acceleration 126, No. 11, 1015–1025.
B coefficient in equation (8b) Andrus, R. D., Stokoe, K. H. I. & Juang, C. H. (2004). Guide for
b power index in equation (7) shear wave-based liquefaction potential evaluation. Earthquake
C coefficient in equation (8b) Spectra 20, No. 2, 285–308.
Cn mechanical coordination number Baxter, C. D. P., Bradshaw, A. S., Green, R. A. & Wang, J. H.
D model diameter (2008). Correlation between cyclic resistance and shear-wave
Dr relative density velocity for providence silts. J. Geotech. Geoenviron. Engng
d particle diameter 134, No. 1, 37–46.
d 50 mean diameter of particle Bolton, M. D. & Wilson, J. M. R. (1989). An experimental and
e void ratio after isotropic consolidation theoretical comparison between static and dynamic torsional soil
emax maximum void ratio tests. Géotechnique 39, No. 4, 585–599, http://dx.doi.org/
emin minimum void ratio 10.1680/geot.1989.39.4.585.
f excitation frequency in Fig. 2(c) Chang, C. S., Misra, A. & Sundaram, S. S. (1991). Properties of
f c mean contact force granular packings under low amplitude cyclic loading. Soil
f cmax maximum contact force Dynam. Earthquake Engng 10, No. 4, 201–211.
f u mean unbalanced force Chantawarangul, K. (1993). Numerical simulations of three-dimen-
G0 small strain shear modulus sional granular assemblies. PhD thesis, University of Waterloo,
Gg particle shear modulus Canada.
Gg0 material constant in equation (8a) Chen, Y. M., Ke, H. & Chen, R. P. (2005). Correlation of shear
g acceleration of gravity wave velocity with liquefaction resistance based on laboratory
H model height tests. Soil Dynam. Earthquake Engng 25, No. 6, 461–469.
IR redundancy index Cheng, Y. P., Bolton, M. D. & Nakata, Y. (2004). Crushing and plastic
LTR distance between the transmitter and receiver deformation of soils simulated using DEM. Géotechnique 54, No.
m power index in equation (8a) 2, 131–141, http://dx.doi.org/10.1680/geot.2004.54.2.131.
Mw earthquake magnitude Clayton, C. R. I. (2011). Stiffness at small strain: research and
N number of cycles practice. Géotechnique 61, No. 1, 5–37, http://dx.doi.org/
Nb total number of particles 10.1680/geot.2011.61.1.5.
N b0 number of particles with no contact Cundall, P. A. & Strack, O. D. L. (1979). A discrete numerical
N b1 number of particles with only one contact model for granular assemblies. Géotechnique 29, No. 1, 47–65,
Nc total number of contacts http://dx.doi.org/10.1680/geot.1979.29.1.47.
22
de Alba, P., Baldwin, K., Janoo, V., Roe, G. & Celikkol, B. (1984). Shafipour, R. & Soroush, A. (2008). Fluid coupled-DEM modelling
Elastic-wave velocities and liquefaction potential. Geotech. Test- of undrained behavior of granular media. Comput. Geotechnics
ing J. 7, No. 2, 77–87. 35, No. 5, 673–685.
Dobry, R., Stokoe, K. H., Ladd, R. S. & Youd, T. L. (1981). Silbert, L. E. (2010). Jamming of frictional spheres and random
Liquefaction susceptibility from S-wave velocity. Proceedings of loose packing. Soft Matter 6, No. 13, 2918.
the ASCE national convention: In situ tests to evaluate liquefac- Sitharam, T. G. (2003). Discrete element modelling of cyclic behav-
tion susceptibility, New York. New York, NY, USA: ASCE. iour of granular materials. Geotech. Geol. Engng 21, No. 4,
Edwards, S. (1998). The equations of stress in a granular material. 297–329.
Physica A: Statist. Mech. Applic. 249, No. 1–4, 226–231. Sitharam, T. G., Vinod, J. S. & Ravishankar, B. V. (2009). Post-
Gong, G., Thornton, C. & Chan, A. H. C. (2012). DEM simulations liquefaction undrained monotonic behaviour of sands: experi-
of undrained triaxial behavior of granular material. J. Engng ments and DEM simulations. Géotechnique 59, No. 9, 739–749,
Mech. 138, No. 6, 560–566. http://dx.doi.org/10.1680/geot.7.00040.
Guo, N. & Zhao, J. (2013). The signature of shear-induced aniso- Soga, K. & O’Sullivan, C. (2010). Modeling of geomaterials behav-
tropy in granular media. Comput. Geotechnics 47, 1–15. ior. Soils Found. 50, No. 6, 861–875.
Hardin, B. O. & Richart, J. F. (1963). Elastic wave velocities in Song, C., Wang, P. & Makse, H. A. (2008). A phase diagram for
granular soils. J. Soil Mech. Found. Div., ASCE 89, No. 1, 33–65. jammed matter. Nature 453, No. 7195, 629–632.
Huang, X., Hanley, K. J., O’Sullivan, C. & Kwok, C. Y. (2014). Soroush, A. & Ferdowsi, B. (2011). Three dimensional discrete
Exploring the influence of interparticle friction on critical state element modeling of granular media under cyclic constant
behaviour using DEM. Int. J. Numer. Analyt. Methods Geomech. volume loading: a micromechanical perspective. Powder Tech-
38, No. 12, 1276–1297. nol. 212, No. 1, 1–16.
Itasca (2008). Manual of particle flow code in 3-dimension, version Tatsuoka, F. (1999). Small strain behavior of granular materials. In
4. 0. Minneapolis, MN, USA: Itasca Consulting Group, Inc. Mechanics of granular materials (eds M. Oda and K. Iwashita),
Jovicic, V., Coop, M. R. & Simic, M. (1996). Objective criteria for pp. 299–308. Rotterdam, the Netherlands: Balkema.
determining G max from bender element tests. Géotechnique 46, Thornton, C. (2000). Numerical simulations of deviatoric shear
No. 2, 357–362, http://dx.doi.org/10.1680/geot.1996.46.2.357. deformation of granular media. Géotechnique 50, No. 1, 43–53,
Juang, C. H., Chen, C. J. & Jiang, T. (2001). Probabilistic frame- http://dx.doi.org/10.1680/geot.2000.50.1.43.
work for liquefaction potential by shear wave velocity. J. Geo- Thornton, C. & Zhang, L. (2010). On the evolution of stress and
tech. Geoenviron. Engng 127, No. 8, 670–678. microstructure during general 3D deviatoric straining of granular
Juang, C. H., Yang, S. H. & Yuan, H. (2005). Model uncertainty of media. Géotechnique 60, No. 5, 333–341, http://dx.doi.org/
shear wave velocity-based method for liquefaction potential 10.1680/geot.2010.60.5.333.
evaluation. J. Geotech. Geoenviron. Engng 131, No. 10, 1274– Tokimatsu, K. & Uchida, A. (1990). Correlation between liquefac-
1282. tion resistance and shear wave velocity. Soils Found. 30, No. 2,
Kayen, R., Moss, R. E. S., Thompson, E. M., Seed, R. B., Cetin, K. 33–42.
O., Kiureghian, A. D., Tanaka, Y. & Tokimatsu, K. (2013). Viana da Fonseca, A., Ferreira, C. & Fahey, M. (2009). A frame-
Shear-wave velocity-based probabilistic and deterministic assess- work interpreting bender element tests, combining time-domain
ment of seismic soil liquefaction potential. J. Geotech. Geoen- and frequency-domain methods. Geotech. Testing J. 32, No. 2,
viron. Engng 139, No. 3, 407–419. 1–17.
Liu, N. & Mitchell, J. K. (2006). Influence of nonplastic fines on Wang, J., Moran, K. & Baxter, D. P. (2006). Correlation between
shear wave velocity-based assessment of liquefaction. J. Geo- cyclic resistance ratios of intact and reconstituted offshore
tech. Geoenviron. Engng 132, No. 8, 1091–1097. saturated sands and silts with the same shear wave velocity. J.
Ng, T.-T. & Dobry, R. (1994). Numerical simulations of monotonic Geotech. Geoenviron. Engng 132, No. 12, 1574–1580.
and cyclic loading of granular soil. J. Geotech. Engng 120, No. Wang, Y. H. & Mok, C. M. B. (2008). Mechanisms of small-strain
2, 388–403. shear-modulus anisotropy in soils. J. Geotech. Geoenviron.
O’Sullivan, C. (2011). Particulate discrete element modelling: A Engng 134, No. 10, 1516–1530.
geomechanics perspective. London, UK: Taylor and Francis Xu, X. (2012). Study on the micromechanism of sand liquefaction
Group. and its evaluation. PhD thesis, Zhejiang University, China (in
O’Sullivan, C., Cui, L. & O’Neill, S. C. (2008). Discrete element Chinese).
analysis of the response of granular materials during cyclic Xu, X., Ling, D., Cheng, Y. & Chen, Y. (2012). Micromechanical
loading. Soils Found. 48, No. 4, 511–530. study on shear wave velocity of granular materials using discrete
Robertson, P. K., Woeller, D. J. & Finn, W. D. L. (1992). Seismic element methods. In Discrete element modelling of particulate
cone penetration test for evaluating liquefaction potential under media (ed. C.Y. Wu), Special Publication 339, pp. 264–270.
cyclic loading. Can. Geotech. J. 29, No. 4, 686–695. Cambridge, UK: Royal Society of Chemistry.
Roy, D., Campanella, R. G., Byrne, P. M. & Hughes, J. M. O. Xu, X., Cheng, Y. & Ling, D. (2013). The influence of void ratio
(1996). Strain level and uncertainty of liquefaction related index on small strain shear modulus of granular materials: a
tests. In Uncertainty in the geologic environment: from theory to micromechanical perspective. In Powders and grains, Proceed-
practice (eds C. Shackelford, P. Nelson and M. Roth), GSP No. ings of the 7th international conference on micromechanics of
58, pp. 1149–1162. Reston, VA, USA: American Society of granular media, Sydney, Australia (eds A. Yu, K. Dong, R.
Civil Engineers. Yang and S. Luding), pp. 201–204. Sydney, Australia: AIP.
Sadd, M. H., Tai, Q. M. & Shukla, A. (1993). Contact law effects Yimsiri, S. & Soga, K. (2000). Micromechanics-based stress–strain
on wave propagation in particulate materials using distinct behaviour of soils at small strains. Géotechnique 50, No. 5,
element modeling. Int. J. Non-linear Mech. 28, No. 2, 251–265. 559–571, http://dx.doi.org/10.1680/geot.2000.50.5.559.
Salot, C., Gotteland, P. & Villard, P. (2009). Influence of relative Yimsiri, S. & Soga, K. (2010). DEM analysis of soil fabric effects
density on granular materials behavior: DEM simulations of on behaviour of sand. Géotechnique 60, No. 6, 483–495, http://
triaxial tests. Granular Matter 11, No. 4, 221–236. dx.doi.org/10.1680/geot.2010.60.6.483.
Santamarina, J. C. & Cascante, G. (1996). Stress anisotropy and Youd, T. L. & Idriss, I. M. (2001). Liquefaction resistance of soils:
wave propagation: a micromechanical view. Can. Geotech. J. 33, Summary report from the 1996 NCEER and 1998 NCEER/NSF
No. 5, 770–782. workshops on evaluation of liquefaction resistance of soils. J.
Santamarina, J. C. & Fratta, D. (1998). Introduction to discrete Geotech. Geoenviron. Engng 127, No. 4, 297–313.
signals and inverse problems in civil engineering. Reston, VA, Youd, T. L., Idriss, I. M., Andrus, R. D., et al. (2001). Liquefaction
USA: ASCE Press. resistance of soils: Summary report from the 1996 NCEER and
Seed, H. B. & Idriss, I. M. (1971). Simplified procedure for 1998 NCEER/NSF workshops on evaluation of liquefaction
evaluating soil liquefaction potential. J. Soil Mech. Found. Div., resistance of soils. J. Geotech. Geoenviron. Engng 127, No. 10,
ASCE 97, No. SM9, 1–15. 817–833.
Seed, H. B. & Lee, K. L. (1966). Liquefaction of saturated sands Zhao, J. & Guo, N. (2013). Unique critical state characteristics in
during cyclic loading. J. Soil Mech. Found. Div., ASCE 92, No. granular media considering fabric anisotropy. Géotechnique 63,
6, 105–134. No. 8, 695–704, http://dx.doi.org/10.1680/geot.12.P.040.
23
Zhou, Y. & Chen, Y. (2005). Influence of seismic cyclic loading Geotech. Geoenviron. Engng 133, No. 8, 959–972.
history on small strain shear modulus of saturated sands. Soil Zhou, Y., Chen, Y. & Shamoto, Y. (2010). Verification of the soil-
Dynam. Earthquake Engng 25, No. 5, 341–353. type specific correlation between liquefaction resistance and
Zhou, Y. & Chen, Y. (2007). Laboratory investigation on assessing shear-wave velocity of sand by dynamic centrifuge test. J.
liquefaction resistance of sandy soils by shear wave velocity. J. Geotech. Geoenviron. Engng 136, No. 1, 165–177.
24
Masini, L. et al. (2015). Géotechnique 65, No. 5, 349–358 [http://dx.doi.org/10.1680/geot.SIP.15.P.001]
An interpretation of the seismic behaviour of reinforced-earth retaining

structures
L . M A S I N I , L . C A L L I S TO a n d S . R A M P E L L O
This paper provides an interpretation of the behaviour of retaining structures made of geosynthetic-
reinforced earth, subjected to a severe seismic loading. It is seen that during strong ground motion the
main source of energy dissipation derives from the transient activation of plastic mechanisms within
the soil mass: these mechanisms can be global, local, or a combination of the two. Using numerical
pseudo-static analyses and limit analysis methods it is shown that three retaining structures having a
similar overall seismic resistance, expressed by their critical seismic coefficient, activate different –
global, local, or combined – plastic mechanisms. The seismic performance of the different retaining
structures is then evaluated through a series of dynamic analyses in which acceleration–time histories
are imposed to the bottom boundary of the same numerical models used for the pseudo-static
analyses. The results of the dynamic analyses are interpreted in the light of the plastic mechanisms
evaluated with the pseudo-static procedure. They show that for the reinforced-earth structures there is
always a local contribution to the dissipation of energy during strong motion, evidenced by the
attainment of the available strength in different portions of the soil-reinforcement system, and that this
energy dissipation has a substantial influence on the seismic performance of the system. These results
extend the current understanding on the seismic behaviour of reinforced-earth retaining structures and
can be used to provide some guidance for design.
KEYWORDS: design; earthquakes; reinforced soils
INTRODUCTION soil and the reinforcement: in fact, the design is typically

The seismic behaviour of geosynthetic-reinforced earth re- carried out using solutions based on either limit equilibrium
taining structures has been the subject of a large number of (Bathurst & Cai, 1995; Cascone et al., 1995; Motta, 1996;
research activities, including field observations (Kramer & Ling et al., 1997; Biondi et al., 2008; Basha & Basudhar,
Paulsen, 2001; Wartman et al., 2006; Shinoda et al., 2007; 2010) or limit analysis (Michalowski, 1998; Ausilio et al.,
Maugeri & Biondi, 2008; Tatsuoka, 2008; Koseki et al., 2000; Leshchinsky, 2001). A direct evaluation of the seismic
2009), shaking-table experiments on small-scale models performance of these structures can be obtained through
(Watanabe et al., 2003; El-Eman & Bathurst, 2004, 2005, the sliding-block method of Newmark (1965), in which the
2007) and full-scale models (Ling et al., 1997, 2005; Ling critical seismic coefficient is calculated with one of the
& Leshchinsky, 1998), and centrifuge tests (Izawa et al., available pseudo-static solutions (Lin & Whitman, 1986;
2004; Kramer & Paulsen, 2004). Remarkably, a number of Ambraseys & Menu, 1988; Yegian et al., 1991; Conte &
field observations have shown a generally good performance Rizzo, 1996; Rampello et al., 2010; Biondi et al., 2011;
of geosynthetic-reinforced earth retaining structures sub- Callisto & Rampello, 2013). A review of existing sliding-
jected to severe seismic loading (e.g. Koseki et al., 2009), block methods was presented by Cai & Bathurst (1996).
and this finding is consistent with observations resulting It is important to emphasise that, although the under-
from shaking table experiments on model reinforced-earth standing of the fundamental mechanisms controlling the
structures (Ling et al., 2005). Intuitively, this satisfactory dissipation capabilities and the ductility of these structures is
behaviour can be ascribed to the possibility that these still incomplete, many design procedures implicitly assume
structures contribute to energy dissipation through the an indefinitely ductile behaviour using, for instance, kine-
development of internal plastic mechanisms, and possess an matic solutions to evaluate the resultant force in the whole
overall ductile behaviour deriving from the large deformation reinforcement system, and then redistributing quite arbitra-
that can be accommodated by the soil-reinforcement system. rily this total force among the different reinforcement layers
It is clear that such structures respond to a severe seismic (e.g. Ling et al., 1997; Michalowski & You, 2000).
loading by mobilising the strength of different portions of The present paper contributes to the understanding of the
the system, including the reinforcing elements, the soil– main factors controlling the seismic performance of geosyn-
reinforcement interfaces, the retained soil and possibly the thetic-reinforced earth retaining structures, by examining the
foundation soil. It may then be anticipated that the seismic behaviour of three different idealised structures retaining the
behaviour of these structures is controlled by the strength same backfill and resting on the same foundation soil.
properties, rather than by the pre-failure behaviour of the Specifically, a reference structure is conceived in such a way
that it has a relatively low resistance, under both static and
seismic conditions, but is characterised by a potentially large
capability of dissipating energy, thanks to a dense and
Manuscript received 26 March 2014; revised manuscript accepted 29
January 2015. Published online ahead of print 23 March 2015.
ductile reinforcement layout. The seismic behaviour of this
Discussion on this paper closes on 1 October 2015, for further details reference structure is compared to that of two additional
see p. ii. structures that have the same overall resistance, but provide
Department of Structural and Geotechnical Engineering, Sapienza progressively less opportunity of an internal dissipation of
Università di Roma, Rome, Italy. energy. The first two structures are made of geosynthetic-
25
MASINI, CALLISTO AND RAMPELLO
45 m B
reinforced earth, but the first one develops large plastic
deformations in the reinforced zone, while the second one,
having stronger reinforcements, activates a plastic mechan-
ism that includes a significant portion of the backfill. The
third structure is a non-dissipative one and can activate only
15 m
external mechanisms. The seismic behaviour of these idea-
lised structures is examined employing a combination of
pseudo-static analyses, based on kinematic solutions and on
finite-difference calculations, and dynamic non-linear time-
domain analyses.
While these are only three specific idealised cases, the
55 m
conclusions deriving from the interpretation of their seismic
response in terms of predominant plastic mechanisms are
indeed quite general, and shed some light on the factors
determining a good seismic performance for this type of
40 m
retaining structures.
PROBLEM LAYOUT
In the reference scheme A of Fig. 1, a fill of height
H ¼ 15 m is retained by an earth structure with a batter
¼ 108, reinforced with geo-grids of uniform length
B ¼ 11 .25 m ¼ 0 .75H having a spacing, s ¼ 0 .6 m and a 101·25 m
tensile strength, TT ¼ 25 kN/m. The fill is made of a coarse-
grained cohesionless material with an angle of shearing Fig. 2. Detail of the finite-difference grid adopted for case A,
showing the boundary conditions adopted in the pseudo-static
resistance 9 ¼ 358, while the foundation soil has 9 ¼ 288 analyses
and a cohesion c9 ¼ 10 kPa. The resistance at the soil-
reinforcement contact is purely frictional with an angle of
shearing resistance 9s equal to that of the parent soil. A while the horizontal displacements were inhibited at the
study of the effect of the strength at the soil–reinforcement lateral boundaries. Also, in these analyses the soil was
interface was also carried out, using an interface strength modelled as an elastic–perfectly plastic material with a
factor of 0 .7; it was found that this reduction in the contact Mohr–Coulomb plasticity criterion, zero dilatancy, a Poisson
strength has a small influence on the pattern of behaviour ratio of 0 .3 and a shear modulus equal to 20 to 50% of the
found in the reference analyses. All the materials are dry small strain shear stiffness (discussed in the next section),
and have a unit weight of 20 kN/m3. Schemes B and C were depending on the computed average strain level.
devised to have the same resistance of case A, but with a The reinforcing levels were modelled as Flac strip ele-
smaller capability of internal energy dissipation: in the case ments, reacting only to axial tension. The constitutive model
B the reinforcements are shorter (B ¼ 7 .9 m, B/H ¼ 0 .53) for these elements is shown in Fig. 3(a): the relationship
and stronger (TT ¼ 35 kN/m), while case C is that of a between the axial force T and the axial strain in the strip,
retaining structure with an infinite internal resistance (mod- simulating that of a polyester (PET) geo-grid, is assumed to
elled as an elastic material) with B ¼ 5 .6 m. be elastic–perfectly plastic, with an axial stiffness EA, a
yield strain y and an ultimate strain u; for strains larger
than u the strip element loses entirely its tensile strength TT,
ANALYSIS OF PLASTIC MECHANISMS
The fundamental analysis tool to study the plastic me-
chanisms for the above schemes is a pseudo-static numerical T
finite-difference analysis, carried out in plane strain condi- εpl
lim
tions using the computer code Flac v.5 (Itasca, 2005), in TT

which the seismic forces are increased progressively until
the strength of the system is fully mobilised. Fig. 2 shows a
detail of the calculation grid employed for the case A: it has
a total width of about 100 m and extends 40 m below the
earth fill; it includes 19 800 quadrilateral soil zones having a EA
size as small as 0 .2 m in the vicinity of the reinforced area.
In the static and pseudo-static analyses, both horizontal and εy εu ΔL
vertical displacements were restrained at the base of the grid, ε
L
(a)
Flexible wall facing
Case A Case B Case C
Geogrid layer Backfill soil
� s
H
� Connected node
Free node
(b)
B 0·75H 0·53H 0·38H Fig. 3. (a) Constitutive behaviour assumed for the reinforcing
layers; (b) detail of the connection of a reinforcing layer to the
Fig. 1. Layouts of the three idealised earth-retaining structures wall facing
26
SEISMIC behaviour OF REINFORCED-EARTH RETAINING STRUCTURES
displaying a perfectly brittle behaviour. When this occurs, Starting from the end of construction, a pseudo-static
the calculation algorithm sub-divides the strip into two analysis was carried out, applying a uniform horizontal body
separate elements that cannot interact with each other. The force expressed as a fraction kh of gravity. The value of the
contact at the soil–strip interface was simulated as elastic– seismic coefficient kh was increased progressively until con-
perfectly plastic with a very large stiffness Ks and a purely vergence, evidenced by a steady reduction of the unbalanced
frictional strength �9s ¼ �9 ¼ 358. forces, became no longer possible. Under this circumstance,
The mechanical properties of the reinforcing layers are the numerical model exhibited a well-defined mechanism,
typical of a medium-strength PET geo-grid and reported in associated with a plastic flow of the soil. The seismic coef-
Table 1; it was assumed that the axial stiffness of the ficient kh that activates the mechanism is termed ‘critical’
reinforcements increases proportionally to their strength – and is indicated as kc. Parametric studies were carried out to
that is, that the yield strain is the same for both cases A and check that the finite-difference grid was sufficiently fine and
B. A wall facing was modelled by assigning a purely elastic adequately extended away from the excavation. It was also
behaviour to the couple of soil zones closest to the lateral found that the solution does not depend on the stiffness of
surface of the reinforced structure. The external end of each the materials and therefore can be assumed to be a result of
strip element was connected rigidly to the wall facing, while the strength properties only. Since the pseudo-static analyses
the internal end was left unconstrained (Fig. 3(b)). were carried out up to critical conditions, the values of the
After initialising the effective stresses in the foundation computed displacements are only conventional, since they
soil, the construction of the reinforced soil structure and the refer to a system that is accelerating due to the static
fill was simulated in 25 steps; each step included the activation of a plastic mechanism. Therefore, for the pseudo-
activation of a reinforcement element, the corresponding static analysis it was assumed that the strip elements are
portion of wall facing, and a 0 .6 m-thick soil layer. infinitely ductile, that is, their ultimate strain �u is infinite.
At the end of construction, a maximum horizontal The critical seismic coefficient kc obtained for the three
displacement equal to 1 .1 % and to 1 .0 % of the wall height schemes of Fig. 1 are very similar, varying from 0 .060 to
was computed for cases A and B, respectively, while a value 0 .066. Fig. 4 shows the contours of shear strains ª obtained
of 0 .5% H was obtained for the retaining wall of infinite in critical conditions for the three different structures, with
internal resistance (case C). Starting from this stage, an an indication of the corresponding critical seismic coefficient
evaluation of the safety with respect to a static collapse was kc. It can be seen that, although for the three cases the
carried out iteratively (e.g. Callisto, 2010) reducing progres- critical coefficient is almost identical, the deformation pat-
sively the strength parameters of the soil and the soil- tern is very different. For case A, two concurrent plastic
reinforcement contact; these analyses yielded a strength mechanisms seem to emerge: a first one is confined within
factor slightly lower than 1 .25. the reinforced zone, and does not seem completely devel-
oped, in the sense that the strain contours show smaller
Table 1. Mechanical parameters adopted for the geosynthetic values near the top of the structure; a second one involves
reinforcements both the reinforced area and the upper part of the backfill,
evidencing also a large strain gradient at the contact between
Case A Case B the backfill and the reinforced structure. The two mechan-
isms converge towards the toe of the structure, marginally
TT: kN/m 25 35 involving the foundation soil. For case B, there seems to be
EA: kN/m 1250 1750 only one, more definite plastic mechanism, going through
�y 0 .02 0 .02
the bottom part of the reinforced structure, and mobilising
�u /�y 10 10
�9s : degrees 35 35 the shear strength in most of the backfill; the foundation soil
Ks: kN/m per m 106 106 is still involved only marginally, but to a slightly larger
extent than for case A. In case C, any internal failure is
Case A Case B Case C
kc 0·060 kc 0·066 kc 0·060
Limit analysis Limit analysis

Trasl. – kc 0·101 Trasl. – kc 0·086
Rot. – kc 0·115 Rot. – kc 0·106
Rot. – kc 0·132
(a) (b) (c)
Fig. 4. Results of the pseudo-static analyses carried out up to critical conditions: contours of shear strains (the actual values are not
relevant in critical conditions); and interpretation using kinematic limit analysis solutions
27
Case A Case B
inhibited, and the mechanism can only develop within the
external soil: it involves the development of the active limit 90
conditions in the backfill and the mobilisation of the strength
of the foundation soil. The effect of a uniform vertical
h
θ
0·1
0·115
h

θ
80

acceleration was also studied, by assuming either upward or
5
0
θ
0
θ
downward inertial forces equal to 50% of the horizontal
0· ·2
0·106
15
5
0
ones. These vertical forces caused for all the schemes a 10%
0·1
70
increase or decrease of kc, respectively, but had no appreci-
θ0: degrees
able effect on the plastic mechanisms.
0·3
The above mechanisms were back-interpreted implement- 60
ing two different kinematic approaches based on limit analy- 0·132
sis, as illustrated schematically in Fig. 5. The first solution
0·4
50
was proposed by Michalowski & You (2000) and assumes
0·2
that a portion of reinforced soil slides along a log-spiral
40
surface that passes through the toe of the structure; the
0·5
0·15
sliding surface can either include only the reinforced zone,
or extend out of the reinforced zone to the upper part of the 30
backfill. For a given value of the angle of shearing resistance 70 80 90 100 70 80 90 100
�9, a log-spiral sliding surface is fully specified by the θh: degrees θh: degrees
angles Ł0 and Łh of Fig. 5(a). The second solution is relative (a) (b)
to a two-block mechanism, in which the first soil block is Fig. 6. Contours of angles Ł0 and Łh of Fig. 5(a) corresponding to
triangular and represents the upper part of the backfill, while different critical seismic coefficients for layouts A and B
the second block is a quadrilateral representing the portion
of the reinforced soil involved in the mechanism. Following
the upper bound theorem of perfect plasticity, the solution 4(a) derives from the analysis of the two-block mechanism
for both cases is obtained by equating the rate of work done of Fig. 5(b); it describes effectively the critical mechanism
by the external force to that dissipated along the sliding obtained from the numerical analysis and corresponds to a
surfaces and within the reinforcing elements that are inter- seismic critical coefficient kc ¼ 0 .10 that, although larger
sected by the surfaces. Using these two solutions, an itera- than the numerical one, is the smallest among those obtained
tive search was carried out to define the mechanism with the kinematic solutions.
corresponding to the minimum critical seismic coefficient. For case B, the contour plot of Fig. 6 shows that there is
Figure 6 shows, for cases A and B, the contours of the a single minimum for kc ¼ 0 .106, corresponding to a log-
critical seismic coefficient kc obtained varying the angles Ł0 spiral surface extending to the backfill (the irregularities of
and Łh. As a general result, the values of kc obtained from the contour are due to the mobilisation of the reinforcement
limit analysis are somewhat larger than those resulting from strength and indicate that the sliding surface is partly
the numerical pseudo-static analyses. This may be due both external). This surface is shown in Fig. 4(b) with a dashed
to the upper bound nature of the kinematic approach, and to line, and is seen to be roughly consistent with the numerical
the effect of the associated flow rule assumed in limit results. The two-block mechanism of Fig. 5(b) yields a very
analysis, in contrast with the assumption of zero dilatancy similar mechanism, shown with a continuous line in Fig.
made in the numerical computations. For case A, the mini- 4(b), but a smaller critical seismic coefficient kc ¼ 0 .086.
mum kc is equal to 0 .115 and is relative to a value of Ł0 In summary, for the reference case A it appears that there
equal to Łh, that is, to a log-spiral degenerating to a planar are two concurring mechanisms at stake. The prevailing one
surface; this planar surface has an inclination on the hori- mobilises the resistance of about 75% of the reinforcements,
zontal of 908 � Ł0 + �9 � 418 and reaches the backfill in the extends to the upper part of the backfill and can be
upper 25% of the structure’s height: it is shown with a interpreted by a two-block scheme. The secondary one is
dashed line in Fig. 4(a), and it is seen to be in a fair fully internal, mobilising the resistance of all the reinforcing
agreement with the prevailing plastic mechanism for case A. layers; it is fully consistent with a kinematic log-spiral
Inspection of the contour lines of Fig. 6(a) also shows that solution, and is associated with similar strains levels; it
there is a relative minimum of kc ¼ 0 .132 corresponding to should be deemed capable of producing a significant amount
Ł0 ¼ 528 and Łh ¼ 838. The resultant log-spiral is shown of energy dissipation during an earthquake. Conversely, in
with a dash-dotted line in Fig. 4(a) and approximates very case B there is a single plastic mechanism, which mobilises
well the internal, secondary mechanism evidenced by the the resistance of about half the reinforcing elements and
pseudo-static numerical analyses. The continuous line in Fig. extends well behind the reinforced structure; it is interpreted
satisfactorily by the two-block kinematic solution, as the
ω relative displacement between the two blocks is quite evi-
O
θ0
dent. For this case, a smaller amount of energy dissipation is
θh expected, since there are no different concurring plastic
θi r0 mechanisms. Case C cannot mobilise any internal dissipation
ri B mechanism; therefore it can reach a plastic mechanism only
B C D
rh v21 W1 activating the strength of the external soil, including the
zi � backfill and the foundation. This idealised wall provides no
φ v φ
1 internal energy dissipation and it is expected to show the
H W2 worst performance under seismic conditions.
α1
v φ v2 φ
�
A DYNAMIC BEHAVIOUR
α2
(a) (b) Analysis method and seismic input
Starting from the end-of-construction stage, time-domain
Fig. 5. Kinematic solutions based on upper bound limit analysis dynamic analyses were carried out by applying time histories
28
a(t ) of the horizontal acceleration to the bottom boundary of Table 2. Mechanical properties of the soil
the same finite-difference grid used for the pseudo-static
analyses. For these analyses the horizontal fixities at the c9: kPa �9: B: kPa C D
lateral sides of the grid were replaced by Flac free-field degrees
boundary conditions.
The cyclic behaviour of the soil was described through Foundation soil 10 28 12 750 1397 .4 0 .790
Backfill soil 1 35 5100 5329 .5 0 .500
the hysteretic damping model implemented in Flac, coupled
with the same Mohr–Coulomb plasticity criterion as was
used in the pseudo-static analyses. The hysteretic damping
model is essentially an extension to two dimensions of the
non-linear soil models that describe the unloading–reloading Assisi (discussed by Callisto & Soccodato, 2010) was used
stress–strain cycles using the Masing (1926) rules. Some as a seismic input; for case A the additional seismic record
details on the hysteretic damping model are provided by of Darfield (taken from the Darfield High School (DFHS)
Callisto & Soccodato (2010). The model requires the values station in New Zealand) was employed. Some properties of
of the small-strain shear stiffness and a backbone curve. The the records are reported in Table 3, where amax is the peak
small-strain shear stiffness was expressed as a function of ground acceleration, IA is the Arias intensity, Ts is the
the mean effective stress p9 significant duration and Tm is the mean period as defined by
Rathje et al. (1998). Fig. 8, showing the acceleration–time
G0 ¼ B þ C 3 (p9)D (1) histories (Fig. 8(a)) and the 5% damped elastic spectra (Fig.
8(b)), evidences that the amplitudes of the Darfield record
Values for the coefficients B, C and D were selected to are much larger than in the Assisi record, particularly in the
reproduce the typical small-strain stiffness for a medium large-periods interval of 0 .3–1 s. Fig. 8(b) shows also the
plasticity sandy-silt for the foundation soil and a dense sand elastic spectra of the horizontal acceleration computed at the
for the backfill. The backbone curve was calibrated to
reproduce a modulus decay curve typical for coarse-grained
Table 3. Properties of the input seismic records
materials. Fig. 7 shows a comparison of the modulus decay
curve obtained with the hysteretic damping model and those Record amax: g IA: m/s Ts: s Tm: s
published by Seed & Idriss (1970) and by Vucetic & Dobry
(1991) for coarse-grained soils. The figure also shows a Assisi 0 .28 0 .75 4 .28 0 .24
comparison between the equivalent damping ratio evaluated New Zealand (NZ-DFHS) 0 .53 2 .62 20 .70 0 .42
from the model hysteresis loops and that provided by the
same authors. Although for ª . 0 .02% the equivalent damp-
ing ratio predicted by the hysteretic model is larger than the
0·5
Seed & Idriss (1970) one, it seems still compatible with the
Vucetic & Dobry (1991) curve.
Acc.: g
In the soil model, damping results entirely from the Assisi

0
hysteretic unloading–reloading cycle. A small amount of
additional viscous damping was used to attenuate the soil
response at very small strain and to reduce spurious high- 0·5
frequency noise (Joyner & Chen, 1975; Callisto et al.,
0·5
2013). This was obtained by specifying a Rayleigh damping
with both mass and stiffness components corresponding to a
NZ-DFHS
damping ratio of 1% at a central frequency of 1 .02 Hz,
Acc.: g
0
which is the fundamental frequency of the soil deposit,
including the backfill. Table 2 lists the values of the strength
and stiffness parameters adopted in the analyses.
0·5
In most of the analyses, the scaled seismic record of
0 5 10 15 20 25 30 35
t: s
(a)
1·0 50
1·6
Assisi – bedrock
0·8 40
1·2 Assisi – free field
NZ-DFHS – bedrock
0·6 30
Damping: %
Flac model NZ-DFHS – free field

Sa : g
0·8
G/G0
Seed & Idriss (1970)

0·4 Vucetic & Dobry (1991) 20
0·4
0·2 10
0
0 1 2 3 4
0 0 T: s
0·0001 0·001 0·01 0·1 1 (b)
γ: %
Fig. 8. Acceleration–time histories and 5%-damped elastic re-
Fig. 7. Modulus decay curve and equivalent damping ratio sponse spectra for the selected seismic records and for the free-
predicted by the hysteretic damping model field surface motion
29
soil surface in the free-field, evidencing a significant amplifi- 1·0
εu ∞
cation for periods in the interval of about 0 .5 to 2 s.
Essential to the present discussion is that both records
proved sufficiently intense to activate the plastic mechanisms 0·8 εu 20%
found in the previous pseudo-static analyses. It was found,
as expected, that the deformation pattern induced by the two
records is quite similar, because it is controlled by the T
0·6
diffuse plastic flow in the soil rather than by the soil
u: m
stiffness; therefore, the following discussion is restricted to
the effects of the Assisi record. However, the computed final 0·4
T
seismic displacements, discussed in the next section, are
quite different, reflecting the largest intensity of the DFHS
record. G
G
0·2
Results of the dynamic analysis

0
The dynamic analyses of the reference retaining structure
0 2 4 6 8 10 12 14 16
A were carried out under the two hypotheses that the t: s
reinforcing elements either have an infinite ductility, or are
characterised by a finite ultimate strain u ¼ 20%, beyond Fig. 10. Time histories of the horizontal displacements computed
which they lose their strength completely (see Fig. 3(a)). with the Assisi record for case A, with reinforcements of infinite
Fig. 9 depicts the contours of the shear strains computed at ductility (dashed lines), and with reinforcements of limited
the end of the Assisi seismic record, while Fig. 10 shows ductility (continuous lines)
the corresponding time histories of the horizontal displace-
ment u of two selected points, located, respectively, within
the reinforced zone (centre of gravity G) and at the top (T) internal log-spiral evaluated with both the FLAC pseudo-
of the retaining structure. For the case of ductile reinforce- static analysis and the kinematic approach; the second sur-
ments, the deformation pattern (Fig. 9(a)) is similar to the face is somewhat lower than the corresponding pseudo-static
plastic mechanisms obtained from the pseudo-static analyses one, and engages a larger portion of the foundation soil. The
(Fig. 4(a)), with the development of two intensely sheared corresponding displacements, indicated in Fig. 10 with
surfaces. The internal surface is very close to the critical dashed lines, show a regular increase during the strong
motion part of the seismic record, reaching maximum values
of about 0 .43 m and 0 .71 m for points G and T, respectively,
γ Case A corresponding to about 2 .9% and 4 .7% of the wall height. It
2
is interesting to note that point T is located to the left of the
internal sliding surface, while the position of point G is
0·20
intermediate between the two surfaces. Therefore, the
displacement of point G is produced only by the activation
0·16 of the two-block plastic mechanism, while the larger
0·12 displacement of point T results from the activation of both
plastic mechanisms.
0·08
It is interesting to note that the computed seismic
0·04 displacements have the same order of magnitude as those
resulting from empirical relationships developed for the
0
Italian territory: according to the results published by Calli-
sto & Rampello (2013), relative to Italian seismic records,
εu ∞ for a class B subsoil as defined by Eurocode 8, part 5 (EN
(a) 1998-5, CEN, 2003) values of the critical seismic coefficient
Case B of 0 .06–0 .07 correspond, for a maximum bedrock accelera-
tion of 0 .28g, to a permanent seismic displacement of
0 .4 4 0 .6 m.
For the case of reinforcements of infinite ductility, Fig. 11
shows the profiles of the mobilised strength T/TT of the
reinforcement layers and of the corresponding maximum
tensile strain , divided by the yield strain y. The open
symbols represent the initial static condition, while the filled
symbols indicate the post-seismic values. For case A (Fig.
11(a)) after the construction the reinforcements located in
the lower third of the wall height are already mobilising
their strength, the corresponding strains being larger than y.
Limit analysis After the earthquake, the strength of 90% of the reinforcing
εu 20 % Trasl. – kc 0·101 layers has been attained. The post-seismic axial strain in-
Rot. – kc 0·115 creases with depth: in the lower half of the structure it varies
from 15 to 20 times the yield strain y, but in the lowest two
Rot. – kc 0·132
(b) layers it is as large as 30 to 50 y.
Since the ultimate tensile strain u assumed for the
Fig. 9. Contours of final shear strains computed at t 9 s with the reinforcements is equal to 10y, it is unsurprising that the
Assisi record for case A, with: (a) indefinitely ductile reinforce- dynamic behaviour of the reinforced structure A changes
ments; (b) reinforcements of limited ductility dramatically if the analysis accounts for the limited ductility
30
Case A Case B Case B
T/TT T/TT γ T
_
0 0·5 1·0 0 0·5 1·0 2
0
End of 0·20
construction
0·16
0·2 Post-seismic G
condition 0·12
0·08
0·4
0·04
εu /εy10
z/H
0
εu /εy10
0·6
Limit analysis
ε/εy Trasl. – kc 0·086
0·8
T/TT Rot. – kc 0·106
(a)
1·0 Case C
T
0 20 40 60 0 20 40 60
ε/εy ε/εy
(a) (b)
Fig. 11. Profiles of the mobilised strength T/TT and of the

G
corresponding maximum strain ratio /y , computed for
reinforcements of infinite ductility: (a) case A; (b) case B
of the reinforcing elements. As shown in Fig. 9(b), in this

case the shear strains concentrate along the internal sliding
surface where, starting from the bottom layers, the largest
proportion of the reinforcements reach their ultimate strain
progressively. Fig. 10 shows that, in this case (continuous (b)
lines), the horizontal displacements of point T diverge: this
is a progressive failure, triggered by the attainment of the Fig. 12. Contours of final shear strains computed with the Assisi
ultimate strain in the lower reinforcement levels. Since point record for: (a) case B with reinforcements of infinite ductility;
(b) case C
G is located just outside the internal mechanism, its
displacements are scarcely affected by the ductility of the
reinforcements.
Figure 11(b) relates to the idealised structure B analysed accommodate the deformation of the structure. A compari-
assuming reinforcements of infinite ductility. In this case, son of the three different mechanisms depicted in Fig. 4
fewer reinforcements located at the bottom of the wall indicates that, although the critical seismic coefficient for
mobilise their strength after construction, with axial strains the three structures is about the same, in progressing from
larger than y. After the earthquake, the amount of reinfor- case A to case C the internal plastic deformations become
cing layers that reach their strength is lower than for case A, less and less important; therefore, a general deterioration of
but the extent of the structure in which the post-seismic the seismic performance should be expected as smaller
strains are larger than 10y is still significant. Therefore, the amounts of kinetic energy can be dissipated by the internal
dynamic behaviour of structure B reinforced with elements plastic strains. This perception is substantiated by the inspec-
of limited ductility (not shown here for brevity) is similar to tion of the seismic performance of the three different
that for case A – that is, a progressive failure occurs with a structures subjected to the same Assisi seismic input, as
rupture surface, which for case B starts within the reinforced shown in Fig. 13(a). With reference to points G and T, the
zone but also reaches the backfill. displacements exhibited by the structure B are from 1 .4 to
Moreover, if case B is analysed assuming that the 2 .6 times larger than the corresponding displacements com-
reinforcements have an infinite ductility, then its seismic puted for case A; the non-dissipative structure C undergoes
behaviour is controlled by the activation of the same plastic displacements that are from 2 .3 to 4 .5 times larger than
mechanism obtained from the pseudo-static analysis: Fig. those computed for structure A. Differently from case A, for
12(a), depicting the shear strain contours at the end of the the other two structures the displacement at the top (point
earthquake, shows that a two-block-mechanism emerges, T) is always lower than that of the centre of gravity G. This
which is analogous to that shown in Fig. 4(b) for the happens because for cases B and – obviously – C, the
pseudo-static critical condition. internal log-spiral concurrent mechanism is absent. This
The dynamic analysis of the non-dissipative structure C finding is consistent with the fundamental assumption of the
shows the activation of a plastic mechanism (Fig. 12(b)), present paper that most of the seismic displacements are
which is totally external to the wall, and very similar to the generated by consecutive, transient activations of the plastic
corresponding pseudo-static mechanism of Fig. 4(c). mechanisms evidenced by the pseudo-static analyses. Fig.
Therefore, it appears that the search for the relevant 13(b) shows the deformed profiles of the wall facing. The
plastic mechanisms that the different structures activate average rotation for cases A and B is similar (about 18),
when subjected to a uniform acceleration field is sufficiently whereas that for case C is about three times larger, since in
indicative of the actual seismic response, provided that the this case only a global plastic mechanism is activated, which
reinforcing elements have a plastic ductility sufficient to implies a significant rotational movement.
31
2·0 Case A
T
γ
_ T
G Case C 2
1·6
0·40
0·32
1·2 G
0·24
Case B
u: m
0·16
0·8 T 0·08
Case A 0
G
0·4 Limit analysis
Trasl. – kc 0·101
εu ∞ NZ-DFHS Rot. – kc 0·115

0
0 2 4 6 8 10 12 14 16 Rot. – kc 0·132
t: s
(a) Fig. 14. Contours of final shear strains computed with the DFHS
Case A Case B Case C record for case A with indefinitely ductile reinforcements
u: m 15
0 3 CONCLUSIONS
12 The behaviour of an earth retaining structure subjected to
an intense seismic loading can be conveniently regarded as a
succession of transient activations of plastic mechanisms,
9 each producing an incremental permanent displacement (Ri-
chards & Elms, 1979). For a geosynthetic-reinforced struc-
y: m
ture retaining a coarse-grained backfill, typical wavelengths

6 are generally much larger than the wall height, and therefore
the analysis of the plastic mechanism can be carried out
with a sufficient accuracy using a pseudo-static method and
3 assuming a uniform distribution of the inertial forces. Limit
analysis and limit equilibrium methodologies can be advan-
tageously extended to pseudo-static conditions and used
0 iteratively to analyse a specific mechanism.
However, in some cases the critical plastic mechanism for
(b)
a given structure may not be obvious, and there may be
more than one active mechanism. The actual deformation
Fig. 13. (a) Time histories of the horizontal displacements pattern can emerge more accurately from a pseudo-static
computed with the Assisi record for cases A and B with analysis performed with a numerical discretisation of the
reinforcements of infinite ductility, and for case C; (b) deformed entire soil domain, carried out up to the seismic coefficient
profiles of the wall façades
that mobilises critical conditions (e.g. Callisto, 2014). Im-
plementing this approach, it was shown that the reference
structure A undergoes a composite deformation pattern that
Since the seismic displacements derive from the activation includes two concurrent plastic mechanisms; conversely, it
of plastic mechanisms that are characterised by a given was found that the seismic behaviour of both structures B
critical seismic coefficient, it is logical that a different and C is controlled by a single, well-defined mechanism,
seismic input would induce a different seismic performance which is partly internal for structure B and totally external
resulting from the activation of similar plastic mechanisms. for structure C. It was shown that the dynamic behaviour of
In fact, a dynamic analysis of the retaining structure A with these retaining structures, resulting from the time-domain
reinforcements of infinite ductility, carried out using the dynamic analyses, can be profitably interpreted on the basis
large-amplitudes DFHS seismic record of Fig. 8, produces of the corresponding pseudo-static plastic mechanisms; dif-
the larger displacements reported in Table 4, but nearly the ferent earthquakes may induce different displacements,
same deformation pattern, as shown by a comparison of which, however, derive essentially from the same plastic
the contours of shear strains plotted in Fig. 14, relative to mechanisms.
the DFHS record, with those of Fig. 9(a), referring to the The three different structures examined in this paper are
Assisi record. characterised by about the same critical seismic coefficient,
but under the actual dynamic loading they show a very
different behaviour. The reference structure A is conceived
to mobilise large plastic internal strains; although during
Table 4. Dynamic analyses. Final horizontal displacements in- construction it undergoes the largest displacements, it shows
duced by the input seismic records (m) the best seismic performance with maximum horizontal
displacements of 0 .03–0 .05H, close to the range computed
Case B/H Record Point G Point T with the empirical Callisto & Rampello (2013) relationship.
The alternative structure B has shorter and stronger
A 0 .75 Assisi 0 .43 0 .71 reinforcements and, under the same seismic input, undergoes
A 0 .75 NZ-DFHS 3 .70 5 .60
B 0 .53 Assisi 1 .12 1 .00
larger displacements (0 .075H). This happens because the
C 0 .38 Assisi 1 .92 1 .61 larger reinforcement strength inhibits the development of the
internal plastic mechanism, subtracting an important source
32
of energy dissipation from the structure. In this respect, it t time
appears that an important criterion for an effective seismic u horizontal displacement
design is to prefer long reinforcements with a relatively low v velocity vector
strength. z depth
When analysing the seismic behaviour of a reinforced- � inclination of the wall facing
ª shear strain
earth structure it is recommendable that limit equilibrium � axial strain
and limit analysis methods be used iteratively to study the �u ultimate strain
critical conditions and to inspect the corresponding plastic �y yield strain
mechanism, rather than to evaluate the safety coefficient Łh angle of log-spiral radius at the wall toe
corresponding to a specific value of the seismic coefficient Łi angle of log-spiral radius
(Callisto, 2014). Contour lines similar to those of Fig. 6 can Ł0 angle of log-spiral radius at the top of the wall
be used to see whether there is a single, well-defined �9 angle of shearing resistance
mechanism, or to detect eventual local minima indicating
concurring mechanisms. Structure layouts entailing more
than one mechanism should be preferred for design.
Structure C is an extreme case in that, having the same REFERENCES
Ambraseys, N. N. & Menu, J. M. (1988). Earthquake-induced
overall seismic strength of structure A, it can mobilise only
ground displacements. Earthquake Engng Struct. Dynam, 16,
external mechanisms. As there cannot be any internal energy No. 7, 985–1006.
dissipation, its seismic displacements are comparatively large Ausilio, E., Conte, E. & Dente, G. (2000). Seismic stability analysis
with respect to those shown by the structures made of of reinforced slopes. Soil Dynam. Earthquake Engng 19, No. 2,
reinforced earth (0 .13H). This finding is consistent with field 159–172.
observations, indicating a generally superior performance of Basha, B. M. & Basudhar, P. K. (2010). Pseudo-static seismic
reinforced earth structures if compared to the behaviour of stability analysis of reinforced soil structures. Geotech. Geol.
more traditional retaining structures. Engng 28, No. 6, 745–762.
Typically, a reinforcing layer intersecting a sheared surface Bathurst, R. J. & Cai, Z. (1995). Pseudo-static seismic analysis of
yields because its structural resistance is attained, rather than geosynthetic reinforced segmental retaining walls. Geosynth. Int.
2, No. 5, 789–832.
because of slippage at the soil–reinforcement interface. This Biondi, G., Grassi, F. & Maugeri, M. (2008). Earthquake effect on
is why the seismic behaviour of both retaining structures A the earth pressure coefficient at active limit state for geogrid
and B, relying to a different extent on the strength of the reinforced wall. Proceedings of the 1st pan-American geosyn-
reinforced areas, was found to be critically dependent on the thetic conference and exhibition, Cancun, Mexico, (eds R. J.
ductility capacity of the reinforcing layers. While the actual Bathurst and E. M. Palmeira), pp. 878–887. Roseville, FL,
mechanical behaviour of a geosynthetic depends on multiple USA: IFAI.
factors, including ageing and environmental effects, it is Biondi, G., Cascone, E. & Rampello, S. (2011). Valutazione del
clear that the choice of the most appropriate reinforcement comportamento dei pendii in condizioni sismiche. Riv. Ital.
should be made principally on the basis of the maximum Geotec. 45, No. 1, 11–34.
Cai, Z. & Bathurst, R. J. (1996). Deterministic sliding block meth-
elongation that it can maintain without important strength
ods for estimating seismic displacements of earth structures. Soil
reduction. For the cases analysed in the present work, it Dynam. Earthquake Engng 15, No. 4, 255–68.
appears that this maximum strain can be evaluated as about Callisto, L. (2010). A factored strength approach for the limit states
ten times the ratio umax /H of the maximum horizontal design of geotechnical structures. Can. Geotech. J. 47, No. 9,
displacement to the wall height. 1011–1023.
Callisto, L. (2014). Capacity design of embedded retaining struc-
tures. Géotechnique 64, No. 3, 204–214, http://dx.doi.org/
ACKNOWLEDGEMENT 10.1680/geot.13.P.091.
The research work presented in this paper was partly Callisto, L. & Rampello, S. (2013). Capacity design of retaining
structures and bridge abutments with deep foundations. J. Geo-
funded by the Italian Department of Civil Protection under
tech. Geoenviron. Engng, ASCE 139, No. 7, 1086–1095.
the ReLUIS 2009–2012 research project. Callisto, L. & Soccodato, F. M. (2010). Seismic design of flexible
cantilevered retaining walls. J. Geotech. Geoenviron. Engng,
ASCE 136, No. 2, 344–354.
NOTATION Callisto, L., Rampello, S. & Viggiani, G. M. B. (2013). Soil–
a acceleration structure interaction for the seismic design of the Messina Strait
amax peak ground acceleration Bridge. Soil Dynamics Earthquake Engng 52, No. 1, 103–115.
B length of geo-grid Cascone, E., Maugeri, M. & Motta, E. (1995). Seismic design of
B, C, D coefficients representing small-strain stiffness earth-reinforced embankments. Proceedings of the 1st interna-
c9 cohesion of foundation soil tional conference on earthquake geotechnical engineering, To-
G operative shear stiffness kyo, Japan (ed. K. Ishihara), pp. 1129–1134. Rotterdam, the
G0 small-strain shear stiffness Netherlands: Balkema.
H fill height CEN (European Committee for Standardisation) (2003). Eurocode
IA Arias intensity 8: EN 1998-5: Design of structures for earthquake resistance –
Ks stiffness Part 5: Foundations, retaining structures and geotechnical
kc critical seismic coefficient aspects. Brussels, Belgium: CEN.
kh seismic coefficient Conte, E. & Rizzo, G. (1996). Un’analisi probabilistica degli
L reinforcement length spostamenti permanenti indotti nei pendii da carichi sismici. Riv.
p9 mean effective stress Ital. Geotec. 40, No. 1, 62–70.
rh log-spiral radius at the wall toe El-Eman, M. & Bathurst, R. J. (2004). Experiment design, instru-
ri log-spiral radius mentation and interpretation of reinforced soil wall response
r0 log-spiral radius at the top of the wall using a shaking table. Int. J. Phys. Modelling Geotech. 4, No. 4,
Sa spectral acceleration 13–22.
T axial force El-Eman, M. & Bathurst, R. J. (2005). Facing contribution to
Tm mean period seismic response of reduced-scale reinforced soil walls. Geo-
Ts significant duration synth. Int. 12, No. 5, 215–238.
TT tensile strength El-Eman, M. & Bathurst, R. J. (2007). Influence of reinforcement
33
parameters on the seismic response of reduced-scale reinforced soil sismico delle opere in terra rinforzata: esperienze da recenti
retaining walls. Geotextiles Geomembranes 25, No. 1, 33–49. terremoti, modellazione fisica e numerica. In Geosintetici in
Itasca (2005). FLAC fast Lagrangian analysis of Continua v. 5. 0. rilevati ed opere di sostegno: atti del XX convegno nazionale
User’s manual. Minneapolis, MN, USA: Itasca Consulting geosintetici, Bologna, Italy (ed. D. Cazzuffi), pp. 63–78. Bol-
Group. ogna, Italy: Patron Editore (in Italian).
Izawa, J., Kuwano, J. & Ishihara, Y. (2004). Centrifuge tilting and Michalowski, R. L. (1998). Soil reinforcement for seismic design of
shaking table tests on the RSW with different soils. Proceedings geotechnical structures. Comput. Geotech. 23, No. 1–2, 1–17.
of the 3rd Asian regional conference on geosynthetics, Seoul Michalowski, R. L. & You, L. (2000). Displacements of reinforced
(eds J. B. Shim, C. Yoo and H. Y. Jeon), pp. 803–910. Seoul, slopes subjected to seismic loads. J. Geotech. Geoenviron.
Korea: Korean Geosynthetic Society. Engng, ASCE 126, No. 8, 685–694.
Joyner, W. B. & Chen, A. T. F. (1975). Calculation of nonlinear Motta, E. (1996). Earth pressure on reinforced earth walls under
ground response in earthquakes. Bull. Seismol. Soc. Am. 65, No. general loading. Soils Found. 36, No. 4, 113–117.
5, 1315–1336. Newmark, N. M. (1965). Effects of earthquakes on dams and
Koseki, J., Nakajima, S., Tateyama, M., Watanabe, K. & Shinoda, M. embankments. Fifth Rankine lecture. Géotechnique 15, No. 2,
(2009). Seismic performance of geosynthetic reinforced soil re- 139–193, http://dx.doi.org/10.1680/geot.1965.15.2.139.
taining walls and their performance-based design in Japan. In Rampello, S., Callisto, L. & Fargnoli, P. (2010). Evaluation of slope
Proceedings of performance-based design in earthquake geotech- performance under earthquake loading conditions. Riv. Ital.
nical engineering, Tokyo (eds T. Kokusho, Y. Tsukamoto and M. Geotec. 44, No. 4, 29–41.
Yoshimine), pp. 149–161. London, UK: Taylor and Francis Group. Rathje, E. M., Abrahamson, N. A. & Bray, J. D. (1998). Simplified
Kramer, S. L. & Paulsen, S. B. (2001). Seismic performance of frequency content estimates of earthquake ground motions. J.
MSE structures in Washington State. Proceedings of the interna- Geotech. Geoenviron. Engng, ASCE 124, No. 2, 150–159.
tional geosynthetic engineering forum, Taipei, Taiwan, pp. 145– Richards, R. & Elms, D. G. (1979). Seismic behavior of gravity
173. Taiwan: International Geosynthetics Society. retaining walls. J. Geotech. Engng Div., ASCE 105, No. GT4,
Kramer, S. L. & Paulsen, S. B. (2004). Seismic performance 449–464.
evaluation of reinforced slopes. Geosynth. Int. 11, No. 6, 429– Seed, H. B. & Idriss, I. M. (1970). Soil moduli and damping factors
438. for dynamic response analysis, Report No. EERC 70-10. Berke-
Leshchinsky, D. (2001). Design dilemma: use peak ore residual ley, CA, USA: University of California.
strength of soil. Geotextiles Geomembranes 19, No. 2, 111–125. Shinoda, M., Watanabe, K., Kojima, K., Tateyama, M. & Horii, K.
Lin, J. S. & Whitman, R. V. (1986). Earthquake induced displace- (2007). Seismic stability of reinforced soil structure constructed
ments of sliding blocks. J. Geotech. Engng, ASCE 112, No. 1, after the mid Niigata prefecture earthquake. In New Horizons in
44–59. Earth Reinforcement, pp. 783–787. London, UK: Taylor and
Ling, H. I. & Leshchinsky, D. (1998). Effect of vertical acceleration Francis.
on seismic design of geosynthetic-reinforced soil structures. Tatsuoka, F. (2008). Geosynthetics engineering, combining two
Géotechnique 48, No. 3, 347–373, http://dx.doi.org/10.1680/ engineering disciplines. Proceedings of the 4th geosynthetics
geot.1998.48.3.347. Asia, Shanghai, China, Special Lecture, pp. 1–35.
Ling, H. I., Leshchinsky, D. & Perry, E. B. (1997). Seismic design Vucetic, M. & Dobry, R. (1991). Effect of soil plasticity on cyclic
and performance of geosynthetic-reinforced structures. Géotech- response. J. Geotech. Geoenviron. Engng, ASCE 117, No. 1, 89–
nique 47, No. 5, 933–952, http://dx.doi.org/10.1680/geot.1997. 107.
47.5.933. Wartman, J., Rondinel-Orviedo, E. A. & Rodriguez-Marek, A.
Ling, H. I., Mohri, Y., Leshchinsky, D., Burke, C., Matsushima, K. (2006). Performance and analysis of mechanically stabilized
& Liu, H. (2005). Large-scale shaking table tests on modular – earth walls in the Tecoman, Mexico Earthquake. J. Perf. Constr.
block reinforced soil retaining walls. J. Geotech. Geoenviron. Facilities, ASCE 20, No. 3, 287–299.
Engng, ASCE 131, No. 4, 465–476. Watanabe, K., Munaf, Y., Koseki, J., Tateyama, M. & Kojima, K.
Masing, G. (1926). Eigenspannungen und Verfertigung bim Mes- (2003). Behaviour of several types of model retaining walls
sing. Proceedings of the 2nd international congress on applied subjected to irregular excitation. Soils Found. 43, No. 5, 13–27.
mechanics, pp. 322–355. Zurich, Switzerland and Leipzig, Ger- Yegian, M. K., Marciano, E. A. & Ghahraman, V. G. (1991).
many: Füssli (in German). Earthquake induced permanent deformations probabilistic ap-
Maugeri, M. & Biondi, G. (2008). Analisi del comportamento proach. J. Geotech. Engng, ASCE 117, No. 1, 1158–67.
34
Verrucci, L. et al. (2015). Géotechnique 65, No. 5, 359–373 [http://dx.doi.org/10.1680/geot.SIP.15.P.012]
Cyclic and dynamic behaviour of a soft pyroclastic rock

L . V E R RU C C I , G . L A N Z O , P. TO M M A S I † a n d T. ROTO N DA
The mechanical behaviour under cyclic and dynamic conditions of a weakly cemented pozzolana,
taken from the subsoil of a historical town in a seismically active area, was investigated through
laboratory tests and in situ measurements. The geotechnical characterisation at laboratory scale is of
particular importance as the small- and large-scale behaviour of these deposits is comparable, owing
to the large discontinuity spacing. These soft pyroclastic rocks, which are found in many volcanic
regions worldwide, are particularly interesting as they are often alternated with stronger volcanic rocks
in thick sequences, constituting a significant impedance contrast that could increase the ground-
shaking hazard. A cross-hole test and a spectral analysis of surface waves test, carried out on the
ground surface and at the floor of an underground cavity, respectively, provided in situ vertical profiles
of the shear wave velocity. Cyclic and dynamic properties were investigated in the laboratory through
velocity measurements of ultrasonic pulses, cyclic simple shear, torsional shear and resonant column
tests. Most testing procedures and devices, which were originally conceived for soils, posed some
challenges. The collected data were used to analyse the influence of mean confining stress, strain
amplitude and number of cycles on both shear stiffness and material damping.
KEYWORDS: dynamics; in situ testing; laboratory tests; particle crushing/crushability; soft rocks
INTRODUCTION grain surfaces, which in turn affects the micro-mechanisms

Pyroclastic deposits cover wide areas, with considerable dominating at the grain contacts (local breakage). The be-
local heterogeneity that derives from the variable modes of haviour of volcanic soils for increasing strain amplitude
deposition and post-depositional alteration processes. Alter- appears more linear than those of other granular materials
nation with other volcanic products, such as effusive rocks, with similar grain size distribution (e.g. Marks et al., 1998)
further increases the geotechnical variability and can result and the linear elastic threshold ªl (Vucetic, 1994) is higher.
in high impedance contrasts and other conditions that influ- Senetakis et al. (2013) suggest that these features of the
ence the local seismic response. cyclic behaviour are possibly due to crushing of the aspe-
The rock-like pyroclastic materials (e.g. welded tuffs and rities at the particle contact, which would also have an
many ignimbrites) are usually dynamically characterised, as impact on energy dissipation mechanisms.
with other rocks, only by measuring the small-strain shear Soft tuffs and weakly cemented pozzolanas form a parti-
wave velocity VS and damping ratio D0. These materials are cular class of pyroclastic materials that present a transitional
so stiff that their stress–strain behaviour is linear even during behaviour between rocks and non-cohesive pyroclastites.
strong earthquakes. Nevertheless comprehensive dynamic Laboratory dynamic testing on undisturbed specimens of
characterisations of ash-flow tuffs and ignimbrites from these materials is not common, owing to difficulties in
Nevada and New Mexico (US) can be found in Choi (2008) sampling and in adapting the traditional cyclic and dynamic
and Jeon (2008), extending up to the mildly non-linear range testing devices. Guadagno et al. (1988) and Papa et al.
(up to strains of about 0 .02%). (1988) tested some cemented pozzolanas together with other
The dynamic properties of incoherent medium- to coarse- incoherent specimens. Similar to other non-cohesive volcanic
grained pyroclastic soils, mainly composed of pumice grains soils, the stiffness and dissipative behaviour of cemented
of low density and high porosity, have been investigated pozzolanas over the medium strain range is more linear than
through bender element measurements (Liu & Yang, 2014), that of quartz sands. Furthermore, a significant and simul-
cyclic triaxial tests (Marks et al., 1998; Miura et al., 2003; taneous reduction of both stiffness and damping ratio with
Sahaphol & Miura, 2005), resonant column tests (Senetakis the number of cycles was attributed to the progressive break-
et al., 2012; Senetakis et al., 2013) and cyclic torsional age of particle bonds which occurs without a significant
shear tests (Orense et al., 2012). Comparison with reference modification of spatial arrangement and size distribution
quartz sands has shown that volcanic soils exhibit signifi- (Papa et al., 1988).
cantly lower small-strain stiffness and slightly lower small- This study reports the dynamic and cyclic characterisation
strain damping. Differences in the observed response are of a weakly cemented pozzolana through laboratory and in
attributed to the higher void ratio and to the lower dry situ tests. Correlation of stiffness and damping with basic
density of volcanic soils, as well as to the morphology of physical properties and their dependence on the stress level,
applied strain amplitude and cycle number have been inves-
tigated. Furthermore, the different results obtained from
Manuscript received 1 April 2014; revised manuscript accepted 4 various types of test apparatus have been highlighted. Analo-
February 2015. gies with the mechanical behaviour in static or monotonic
Discussion on this paper closes on 1 October 2015, for further details loading conditions and possible relationships with the micro-
see p. ii.
Department of Structural and Geotechnical Engineering, Sapienza structure of the material have also been explored.
University, Rome, Italy. Certain particular behaviours of the investigated material
† Institute for Environmental Geology and Geo-Engineering, National can be extended to many other slightly cemented pyroclas-
Research Council, c/o Faculty of Engineering, Sapienza University, tites in peninsular Italy with similar lythological features and
Rome, Italy. physical properties (Pellegrino, 1969; Cecconi et al., 2010),
35
Downloaded by [ University of Liverpool] on [12/09/16]. Copyright © ICE Publishing, all rights reserved.
VERRUCCI, LANZO, TOMMASI AND ROTONDA
although the wide variation in pyroclastic materials should of the experimental investigation presented. Tuff and pozzo-
not be misestimated. lana facies were probably deposited by pyroclastic flows
with different water vapour contents (Faraone & Stoppa,
1988). More specifically, tuff is believed to be the result of
PHYSICAL AND STATIC PROPERTIES a vapour phase alteration given the widespread growth of
The products of the extinct volcanic apparata of central zeolite minerals, which are absent in the pozzolana facies.
western Italy, located north of Rome, form a wide plateau of Although the spatial distribution of the two main facies
layered pyroclastic materials and effusive rocks, which over- within the slab has not been fully ascertained, a homoge-
lies a thick Plio-Pleistocene sedimentary substratum of neous stratigraphy is recognisable over a wide sector of the
clayey and sandy units. The hydrographic network has southern side of the hill (site SF in Fig. 1). Here the cliff
eroded the plateau, especially at its distal areas, exposing the (35–40 m high) is almost entirely formed by pozzolana,
sedimentary substratum and producing a landscape charac- overlying a layer of tuff that is a few metres thick (Fig. 2).
terised by gorges, canyons and mesas. Many old towns are At the SF site the pozzolana can be subdivided into an
perched on top of these mesas, often dating back to the upper weaker lithotype (SFW) and a lower, richer in scoriae,
Middle Ages. Owing to its inestimable artistic heritage, stronger (SFH) lithotype. Pozzolana samples were also col-
Orvieto is the most famous of these. It extends over the lected 200 m to the east (site SB in Fig. 1). Here, too, the
entire top (about 0 .8 km2) of a pyroclastic slab (Fig. 1), pozzolana, similar to the SFW lithotype, overlies the tuff.
delimited by vertical scarps up to 60 m high, which caps a Although from a macroscopic point of view only the
clay base with gentle slopes (Tommasi et al., 2013). Be- scoria content can differentiate the pozzolana lithotypes,
tween the slab and the clay there is a thick layer of fluvio- laboratory tests and petrographic analyses highlighted some
lacustrine sediment which hosts a perched groundwater, so differences. Focus was on the SFW lithotype, the only one to
that fully saturated conditions do not occur within the be investigated through all the test methods. Nevertheless,
pyroclastic slab. results of tests conducted on the other lithotypes (SFH and
The whole slab consists of pyroclastic products belonging SB), especially in situ tests and laboratory measurements of
to the Tufo di Orvieto e Bagnoregio formation, deposited ultrasonic pulse velocity, offer other insights into the cyclic
about 330 000 years BP (before present), in the final period and dynamic behaviour of the pozzolana overall.
of activity of the Bolsena caldera (Santi, 1991), during a A physical and mechanical characterisation under static
major eruption covering a 200 km2 area (Nappi et al., 1994). conditions of the Orvieto pozzolana is reported in Rotonda
From the geotechnical point of view two materials are et al. (2002) and Tommasi et al. (2015), covering all the
observed: a rock-like, red–yellow-coloured facies, and a lithotypes as well as providing some comparisons with the
slightly cohesive, grey-coloured facies, hereafter termed tuff facies.
‘tuff’ and ‘pozzolana’, respectively. The latter is the object At a macroscopic observation pozzolana appears to be
Pyroclastic SF site
slab SB site
Fig. 2
Clayey
slope
Fig. 1. View of the Orvieto Hill from the south with indication of sampling sites
Duomo
San Francesco
Cross-hole
Cavity 536 (SASW) Cavity 508
SFW pozzolana
SFH pozzolana
Tuff
Fig. 2. View of the southern cliff underneath the S. Francesco church (SF zone) with location of
the investigation and sampling sites
36
CYCLIC AND DYNAMIC BEHAVIOUR OF A SOFT PYROCLASTIC ROCK
constituted of clasts immersed in a grey aphanitic ground Silt Sand Gravel
mass (Fig. 3(a)). The size of pumices and scoriae, the most
100
numerous clasts, varies from a few millimetres to several
centimetres, whereas that of lithic fragments does not exceed
10 mm. The material does not disaggregate when completely SFH
80
Percentage passing: %
soaked, but under an appreciable finger pressure it is slightly SFW
friable at any water content. The ground mass is responsible
for the disaggregation under loads, pumices and scoriae 60 SB
being markedly tougher.
Scanning electron microscopy (SEM) analyses of SF 40
specimens (Tommasi et al., 2015) showed that the ground
mass is formed by extremely small glass lumps that are
welded to each other over very small areas (Fig. 3(b)). These 20
act as struts of a weak but strongly three-dimensional frame.
The size of interparticle pores is similar to that of the glass 0
lumps. At high magnification each lump appears to be an 0·01 0·1 1 10
aggregate of closely welded minuscule glassy fragments, in d: mm
turn containing extremely small voids. The groundmass of Fig. 4. Grain size distribution curves of the pozzolana lithotypes
the pozzolana sampled at the SB site is more porous and is
formed by less continuous and smaller glass particles (Fig.
3(c)). high-strain range (0 .01–0 .1%, Fig. 5). All specimens exhibit,
Owing to the high porosity n of most of its constituents presumably after the coupling effects between testing ma-
(ground mass, pumices, scoriae), the Orvieto pozzolana is chine and specimen are extinguished, a decrease of the
highly porous (Table 1), much more than other similar Young’s modulus up to an axial strain of 0 .02–0 .05%. A
materials, such as the pozzolanas from the Rome area further gradual increase of the Young’s modulus, up to large
(Cecconi et al., 2011). According to the grain size distribu- axial strains (about 0 .2%), is then observed.
tion (Fig. 4), all of the lithotypes are classified as sandy
gravel.
Based on the main static strength and deformability index LABORATORY AND IN SITU INVESTIGATIONS
parameters (Table 1), the material can be considered to be a A couple of vertical boreholes were drilled to perform
rock of extremely low uniaxial strength f and medium to cross-hole (CH) measurements in the SFW and SFH pozzola-
high modulus ratio (Et50 /f). na. A spectral analysis of surface waves (SASW) test was
Under oedometric and isotropic compression, permanent carried out in an ancient sub-horizontal underground pozzo-
strains are observed from the lowest stress values; hence it lana quarry excavated in the upper portion of the cliff
is not possible to define a linear elastic phase. Of greater (cavity 536 in the census of Regione Umbria, Fig. 2). The
note is the variation of the secant Young’s modulus measured laboratory tests were conducted on specimens retrieved at
during uniaxial monotonic compression in the medium- to the SF site from the CH borehole cores and from two block
10 mm 80 μm 80 μm
(a) (b) (c)
Fig. 3. (a) Picture of the pozzolana; SEM microphotographs of ground mass of the (b) SFW
and (c) SB lithotypes
Table 1. Physical and static mechanical properties of the three pozzolana lithotypes SFW, SFH and SB (< x >: mean value; STD: standard
deviation; Nd: number of data)
SFW SFH SB
,x. STD Nd ,x. STD Nd ,x. STD Nd
rs: Mg/m3 2 .59 – 4 2 .63 – 2 2 .54 – 1

rd: Mg/m3 1 .07 0 .04 57 1 .16 0 .05 37 1 .11 0 .09 48
n 0 .59 0 .02 57 0 .56 0 .02 37 0 .57 0 .03 48
e0 1 .47 0 .09 57 1 .27 0 .09 37 1 .30 0 .15 48
p9y : MPa 4 .4 – 3 – – – 3 .7 – 4
Et50: MPa 321 132 4 1106 391 17 394 187 15
f: MPa 1 .18 0 .32 4 2 .28 0 .51 17 1 .49 0 .71 15
t: MPa 0 .56 0 .11 7 0 .90 – 3 0 .14 0 .03 10
Note: rs ¼ density of the solid particles; rd ¼ dry bulk density; n ¼ porosity; e0 ¼ void ratio; p9y ¼ isotropic yielding stress; Et50 ¼ Young’s
modulus at 50% of failure stress; f ¼ uniaxial strength; t ¼ tensile strength.
37
1500
the specimen preparation are reported in Tommasi et al.
(2015). All tests were carried out on dry specimens; the
specimen sizes are shown in Table 2.
50% of failure stress
Measurements of the ultrasonic pulse velocity. Ultrasonic P-

1000 and S-wave velocities were measured in the laboratory on
specimens prepared for static and cyclic/dynamic tests. The
apparatus consisted of a Pundit ultrasonic pulse generator, a
Esec: MPa
pair of piezoelectric contact transducers (diameter 12 .7 mm

and natural frequency 1 MHz), an amplifier with gain up to
SB 60 dB, and a digital oscilloscope. A square wave was used as
500 input pulse. To improve the specimen–transducer coupling, a
pressure of 320 kPa at contact was applied and a thin film of
cane molasses was interposed between the transducers and
the specimen ends, which filled the surface voids of the
material.
0
0 0·05 0·10 0·15 0·20 0·25 Double-specimen direct simple shear apparatus. Tests were
εa: % conducted through a double-specimen direct simple shear
(a) apparatus (DSDSS), an evolution of the single-specimen
1500 device developed at the Norwegian Geotechnical Institute
(NGI-DSS) (Bjerrum & Landva, 1966). The apparatus, which
is schematically illustrated in Fig. 6, is designed for testing
soil specimens under simple shear strain conditions. A
SFH
Table 2. Sizes of specimens for cyclic/dynamic mechanical tests

1000
Test type UPV DSDSS TS–RC

Esec: MPa
Diameter, D: mm 35 .7–66 .6 66 .6 35 .7
Height, H: mm 20 .0–100 .0 20 .0 70 .0
500
Note: UPV ¼ ultrasonic pulse velocity measurement; DSDSS ¼ dou-
ble specimen direct simple shear test; TS ¼ torsional shear test;
SFw RC ¼ resonant column test.
0
0 0·05 0·10 0·15 0·20 0·25
εa: % Top cap
(b)
Fig. 5. Secant Young’s modulus Esec plotted against axial strain a D

from uniaxial monotonic tests conducted on the (a) SB and (b) SFW
and SFH lithotypes Proximity Specimen h
transducer
δ
samples that were manually extracted from the cavity 508 F
wall (Fig. 2). Other core specimens were obtained from a Middle cap
horizontal borehole at the SB site. Load cell
Specimen Reinforced
O-ring seal rubber
Laboratory test equipment and testing programme membrane
The experimental investigation comprised a series of
dynamic (ultrasonic pulse velocity (UPV) and resonant col-
umn (RC)) as well as cyclic (simple shear and torsional Bottom cap
shear) tests. The ultrasonic pulse velocity tests were per- Frame
formed on the three pozzolana lithotypes, but only the SFW
lithotype was used for the other cyclic and dynamic tests.
Owing to the high drillability contrast between the ground
mass and the clasts, the usual coring and grinding proce-
dures could not be used. Therefore, specimens were retrieved Fig. 6. Section of the DSDSS apparatus (Doroudian & Vucetic
from partially saturated frozen samples. On a similar pozzo- (1995), modified). Adapted, with permission, from Geotechnical
lana (Cecconi, 1998) the decrease of the dynamic Young’s Testing Journal, vol. 18, no. 1, March 1995, copyright ASTM
modulus induced by such a preparation method did not International, 100 Barr Harbor Drive, West Conshohocken, PA
exceed 12% of that of the original material. Details about 19428
38
description and details of the DSDSS device are given mean effective confining stress p9 between 0 and 500 kPa. In
elsewhere (Doroudian & Vucetic, 1995; D’Elia et al., 2003). the cyclic TS test a series of five torsional cycles (frequency
To summarise, the vertical load P is applied to the top cap, f ¼ 0 .1 Hz) was applied to the specimen, with ªc increasing
whereas the horizontal cyclic load F is applied by a from about 0 .0001 to 0 .005%. For the dynamic RC tests,
micrometer connected to the horizontal piston. A proximity some thousands of high-frequency cycles of the torque
transducer measures the horizontal displacement � of the (variable frequency f ¼ 30–90 Hz) were applied, yielding a
middle with respect to the bottom cap. Minor false ªc from about 0 .0005 to 0 .05%. In these cases the stiffness
deformations can be induced only by the shear deformability was calculated from the resonance conditions and the damp-
of the caps and possible slips at the specimen–cap interface. ing from the frequency response curve, using the half-power
These are minimised by using stainless steel caps and by bandwidth method.
sticking the specimens’ ends to all the caps with a two- As TS and RC tests were performed consecutively on the
component adhesive. same specimen, the sequence of the applied static and
When used on deformable soils the lateral confinement of dynamic loads was chosen so that the specimen would
the specimens is achieved through wire-reinforced rubber undergo an increasing maximum strain amplitude. Therefore,
membranes that ensure the absence of lateral strains. Owing the TS tests at all times preceded the RC tests at the same
to the stiffness of the pozzolana and to the unavoidable, torsional load.
although extremely thin, annular gap between the lateral
surface of the cored specimens and the membrane, the
lateral constraint cannot be completely achieved by such a In situ measurements
method. Nevertheless, a significant horizontal confinement is The CH test was performed between two 35 m deep
applied at the end faces, which are firmly fixed to the caps, vertical boreholes, spaced at 3 .6 m and 15 m from the cliff
especially for very short specimens. In some check tests no edge. To measure the P-wave transit time a spark generator
noticeable differences arose between stress–strain curves and hydrophone were utilised as pulse source and receiver,
obtained with or without membranes. Therefore, an unknown respectively. Regarding the S-wave, an electro-dynamic
horizontal stress �h acts during the tests, which does not shockwave generator and a geophone (natural frequency
exceed �h,oed ¼ Koed�v, where Koed is the coefficient of earth 14 Hz) were used. The measures were spaced 1 m along the
pressure at rest in oedometric conditions. borehole axes.
The underground quarry where the SASW test was per-
formed (Rotonda et al., 2002) extends perpendicular to the
Torsional shear and resonant column apparatus. The appa- cliff wall at about 12 m below the top surface of the
ratus used for the torsional shear (TS) and the resonant pyroclastic slab and 30 m east of the CH test site. A 40 m
column (RC) tests is an evolution of the first free-fixed type survey line on the floor of the quarry was investigated; the
machine designed at the University of Texas at Austin impact on the ground was obtained with a 170 N weight
(Isenhower, 1979); its use is widespread in dynamic research falling from a height of 1 .5 m. In the instrumentation
on soils (e.g. Allen & Stokoe, 1982; Ni, 1987; Silvestri, scheme, two piezoelectric transducers, with a 50 kHz reso-
1991). The test system consists of a cylindrical specimen nant frequency, positioned along the same alignment with
with the bottom end (passive end) fixed on a base pedestal, the impacting weight, were utilised. The common receiver
while the top active end is attached to an excitation device mid-point scheme (Nazarian, 1984) was used. The transmis-
capable of applying a torquing excitation. Transducers are sion velocity for each frequency, related to different depths,
used to measure the rotational vibration amplitudes of the was calculated using Fourier analyses. The inversion of the
driving system. dispersion curves provided the vertical profile of the Ray-
The specimen ends for the TS–RC tests were also stuck leigh wave velocity. Shear velocity was determined assuming
to the platens, thus ensuring full compliance. The vibrating a dynamic Poisson coefficient �dyn ¼ 0 .3.
system was positioned in an air pressure chamber where an
isotropic pressure p was applied.
EXPERIMENTAL RESULTS
Laboratory tests
Testing programme. Table 3 summarises the testing pro- UPV tests. Elastic wave velocities are reported in Fig. 7 for
gramme applied entirely on dry specimens. VP and VS values specimens longer than 25 mm for the three lithotypes.
measured with the UPV test are listed, together with the Measures for the shortest specimens were rejected because
derived dynamic elastic moduli Gdyn, Edyn and Poisson ratio of the difficulty in establishing the arrival time, owing to the
�dyn. To assess the Young’s modulus, laterally constrained near-field source. Moreover, very short ray paths have a high
conditions for the P-wave propagation were assumed. probability of developing almost entirely through the stiffest
In the DSDSS tests, each series of tests on a pair of components, thus overestimating the material global elastic
specimens consisted of several loading stages, with a � v9 properties.
varying between 100 and 800 kPa. Assuming K0 ¼ 0 .43 (co- The mean elastic properties of the three lithotypes ob-
efficient of pressure at rest for an elastic material with tained from the UPV measurements are reported in Table 4.
Poisson coefficient � ¼ 0 .3 deformed in laterally constrained Despite the differences in wave velocity, the Poisson ratio
conditions) the estimated mean effective stress p9 varied �dyn of the pozzolana does not vary much from 0 .3. The
between about 60 and 500 kPa. For each loading stage, the glass forming the scoriae has a very high dynamic stiffness,
specimens were subjected to several steps of strain-con- although the density is similar to that of the overall material.
trolled cyclic shearing, gradually increasing the amplitude of The main physical character influencing the dynamic
shear strain ªc from 0 .0004% to 0 .1%. In one test only, a elastic properties is the abundance of voids, which is well
maximum value of ªc equal to 0 .3% was reached. The represented by porosity n. Fig. 8 shows that the scattering of
number of cycles (N) applied in each step, at constant ªc a single lithotype is noticeable, but the difference in porosity
values, varied between 5 and 10. The shape of cyclic from one lithotype to another is associated with a significant
loading was approximately sinusoidal, with a frequency change in dynamic stiffness. Taking into account all the
ranging between 0 .1 and 0 .3 Hz. measures, a correlation between wave velocities and porosity
The low-amplitude TS and RC tests were performed at a is observed, which is more marked for the longitudinal wave
39
Table 3. Ultrasonic pulse velocity (UPV), double-specimen direct simple shear (DSDSS) and torsional shear and resonant column (TS–RC) testing programme
Code names rd: n e0 UPV test DSDSS–TS–RC tests

of specimens Mg/m3
VP: m/s VS: m/s �dyn Gdyn: Edyn: ªc: % f: Hz N p9: G0: D0: %
MPa MPa kPa MPa
SF-BL2ag 1 .07 0 .60 1 .52 1218 618 0 .32 408 1080 DSDSS 0 .0004–0 .3 0 .1–0 .3 10 223 226 1 .2
446 235 1 .3
SF-BL2bc 1 .12 0 .58 1 .40 1271 633 0 .34 456 1215 124 180 1 .5
248 198 1 .3
495 284 1 .6
SF-BL2de 1 .05 0 .61 1 .57 1255 574 0 .37 347 948 248 194 1 .8
495 237 1 .4
SF-BL2n 1 .11 0 .59 1 .42 1060 556 0 .31 345 904 TS–RC 0 .0001–0 .01 (TS) 0 .1 (TS) 5 (TS) 0 150 2 .6
0003–0 .1 (RC) 30–80 (RC) . 1500 (RC) 60 241 1 .8
120 223 1 .6
250 288 2 .1
40
500 359 1 .1
SF-BL2o 1 .07 0 .60 1 .51 916 439 0 .35 207 560 0 128 2 .1
60 211 1 .7
120 201 0 .8
250 272 1 .2
500 321 1 .4
SF-BL2q 1 .11 0 .59 1 .43 852 394 0 .36 173 471 0 111 1 .9
60 187 1 .7
120 210 0 .6
250 277 1 .5
500 337 0 .7
SF-BL2r 1 .11 0 .59 1 .42 970 468 0 .35 244 658 0 162 1 .6
60 223 2 .3
120 242 1 .3
250 290 1 .8
500 348 1 .8
Note: VP, VS ¼ elastic wave velocity from UPV tests; �dyn, Gdyn and Edyn ¼ elastic properties from UPV tests; ªc, f, N ¼ strain amplitude range, test frequency and maximum number of cycles of the DSDSS, TS
and RC tests; G0, D0 ¼ shear modulus and damping ratio at minimum strain amplitude from DSDSS and TS tests.
2000
Scoriae
2000
SFW
ν0
VP: m/s
SFH
1600 1500
SB 0·1
1000
VS: m/s
Scoriae
1200
0·3
0·50 0·55 0·60 0·65
n
(a)
0·4
800 1500
Scoriae
1000
VS: m/s
400
500
800 1200 1600 2000
Vp: m/s
0
0·50 0·55 0·60 0·65
Fig. 7. Longitudinal (VP) plotted against shear (VS) elastic wave n
velocities from UPV tests for all the pozzolana lithotypes (b)
3000
Scoria
velocity VP. The data trend is consistent with that predicted
by models of poroelastic materials with isometric pores 2000
Kdyn: MPa
(Rotonda & Ribacchi, 1995).
1000
DSDSS tests. The results of a typical DSDSS test are shown
in Fig. 9 in terms of stress–strain loops for cyclic shear strain
amplitude ªc, ranging from 0 .0004% to 0 .3%. The figure 0
0·50 0·55 0·60 0·65
highlights that, at strains as small as 0 .0004%, both shear n
modulus and damping have been accurately determined. (c)
Owing to ambient vibrations, measurements of damping ratio
Scoria
at very small strains were rejected in some tests. 1500
Plots in Fig. 9 show a typical decrease in shear modulus
SFW
and an increased damping ratio for increasing shear strain
amplitude, up to ªc , 0 .1%. At the highest strain amplitudes 1000
Gdyn: MPa
SFH
(ªc ¼ 0 .1% and 0 .3%) the shape of the loop tips changes.
SB
This is more evident in the enlarged picture referring to
500
ªc ¼ 0 .3%, where the stress–strain curve shows an upwards
curvature, indicating a stiffening of the material close to the
tips of the loops. This strain-hardening behaviour has also 0
0·50 0·55 0·60 0·65
been observed in static tests at comparable strain levels (Fig. n
5). (d)
Average values of the shear modulus G and damping ratio
D for the three pairs of specimens tested in the DSDSS Fig. 8. Wave velocities ((a) longitudinal Vp and (b) shear VS) and
elastic moduli ((c) bulk Kdyn and (d) shear Gdyn) from UPV tests
apparatus are plotted against ªc in Fig. 10 for different
plotted against porosity n of the specimens
values of p9. G and D data points are connected with solid
lines. The maximum shear modulus G0 calculated by extra-
polating the stiffness decay curves to ªc ¼ 0 .0001% (Table
3) varies between 180 and 280 MPa in the range of tion for the other two specimens (SF-BL2bc, SF-BL2de) is
p9,100–500 kPa. At the maximum ªc values (ªc ¼ 0 .1%) smaller, especially at lower values of confining stress
one specimen (SF-BL2ag) experienced an approximately (p9 ¼ 60 kPa and p9 ¼ 124 kPa).
50% reduction in shear modulus with respect to G0. Reduc- Damping ratio is under 2% at small-strain amplitudes and
Table 4. Elastic properties of pozzolana lithotypes from UPV tests (< x >: mean value; STD: standard deviation; Nd: number of data).
VP: m/s VS: m/s rd: Mg/m3 Gdyn: MPa Edyn: MPa �dyn
,x. STD Nd ,x. STD Nd
SFW 1000 148 35 502 80 33 1 .07 275 734 0 .33

SFH 1516 166 30 751 100 29 1 .16 652 1743 0 .34
SB 1283 154 9 766 73 9 1 .05 618 1510 0 .22
Scoriae 2216 – 2 1312 – 2 1 .00 1715 4220 0 .23
41
2 4 10
γc 0·0004% γc 0·001% γc 0·004%
N1 N2 N2
τ: kPa
τ: kPa
τ: kPa
0 0 0
2 4 10
0·001 0 0·001 0·001 0 0·001 0·005 0 0·005
γ: % γ: % γ: %
20 80 200
γc 0·01% γc 0·04% γc 0·1%
N2 N2 N2
τ: kPa
τ: kPa
τ: kPa
0 0 0
20 80 200

0·010 0 0·010 0·040 0 0·040 0·100 0 0·100
γ: % γ: % γ: %
600
γc 0·3%
N2
100
τ: kPa
τ: kPa
600
0·300 0 0·300 0 0·200
γ: % γ: %
Fig. 9. Stress–strain cycles (�–ª) of a typical series of DSDSS tests (specimen SB-BL2bc) at increasing shear strain amplitude ªc
attains maximum values of about 6% at ªc � 0 .1%. The at the sampling site (corresponding to a mean stress value of
decrease in damping ratio with strain amplitude, observed at 440 kPa). In Fig. 12, degradation indexes �G ¼ GN/G1 and
high strains in some tests (e.g. specimen SF-BL2bc), could �D ¼ DN/D1 are plotted against N for ªc ¼ 0 .1%, G1 and GN
be associated with the change in shape of the stress–strain being the shear moduli at the first and at the Nth cycle, and
loop. Further, a general decrease of D as the mean confining D1 and DN the analogous damping ratio values. A progres-
stress increases can be noted. sive reduction in stiffness (,15%) and damping (. 20%)
The evolution of stiffness and damping parameters with can be observed. Although these results refer to a lithotype
the number of cycles N, at a given strain amplitude ªc, is (SB) characterised by a more delicate structure (Tommasi et
shown in Fig. 11 for the highest stress levels (p9 ¼ 240 kPa al., 2015), they indicate the noticeable effect of a large
and p9 ¼ 495 kPa). A maximum of ten cycles was applied number of cycles on the stiffness and damping degradation
during each strain-controlled test. The reduction in damping of the pozzolana, similar to the behaviour observed by Papa
ratio with the number of cycles is significant in all tests, et al. (1988) in cemented pozzolanas.
particularly in the earliest cycles and at higher strain ampli-
tudes. A small reduction of the G modulus with cycles is
noticeable only for ªc ¼ 0 .1%. TS/RC tests. Results from both TS and RC tests are
Of particular interest are the results of a test on the SB illustrated in Fig. 13 in terms of variation of G and D with
pozzolana lithotype conducted with the specific aim of the shear strain amplitude at different values of the mean
investigating the effect of a large number (130) of cycles. confining stress p9. The strain ranges investigated by TS and
The adopted vertical stress corresponds to that acting in situ RC tests span from 0 .0001% to about 0 .01% and from
42
500 8
SF-BL2ag SF-BL2ag
400
6
300
G: MPa
p 446 kPa
D: %
4
200 p 223 kPa
2 p 223 kPa
100
p 446 kPa
0 0
0·0001 0·001 0·01 0·1 0·0001 0·001 0·01 0·1
γc: % γc: %
500 8
SF-BL2bc SF-BL2bc
400
6
300 p 495 kPa

G: MPa
D: %
4
p 248 kPa
200
p 124 kPa
p 62 kPa 2
100 p 124 kPa p 495 kPa
p 248 kPa
0 0
0·0001 0·001 0·01 0·1 0·0001 0·001 0·01 0·1
γc: % γc: %
500 8
SF-BL2de SF-BL2de
400
6
300
G: MPa
p 495 kPa
D: %
4
p 248 kPa
200
p 124 kPa
2
100 p 62 kPa p 248 kPa
p 495 kPa
0 0
0·0001 0·001 0·01 0·1 0·0001 0·001 0·01 0·1
γ c: % γc: %
Fig. 10. Shear modulus (G) and damping ratio (D) plotted against strain amplitude ªc from DSDSS tests
0 .0004% to about 0 .1%, respectively. The stiffness values increases from about 400 to 570 m/s, with a low, steady
obtained from the TS tests are in satisfactory accordance with gradient from the surface down to 29 m depth, where the
those from dynamic RC tests. boundary between the two different pozzolana lithotypes is
Generally in the medium strain range, where experimental apparent, as already observed on the cliff wall. The values
data from the two tests overlap, G values from RC tests are measured in the underlying SFH lithotype vary from 650 to
slightly smaller than those from TS tests for the same mean 730 m/s. P-wave measurements were rejected because a
stress. In fact such a weak material suffers significant strong attenuation of the signal, typical for this kind of
damage during RC tests, which apply a very high number of materials, caused uncertainty regarding arrival time.
cycles. Therefore G from RC tests is underestimated with The VS profile obtained with the SASW test (Fig. 14) reaches
respect to that obtained from the other test types. a depth of about 8 m below the floor of the underground quarry,
At small-strain amplitudes the shear modulus G0 varies in that is, 20 m below the ground surface. A 2 .5 m thick layer
the range of 150–350 MPa for all the tested specimens. below the floor of the underground cavity shows very low
The increase of the mean stress results in a general dynamic characteristics, due to both loosening and excavation-
increase in shear stiffness. In some tests, at high values of induced damage around the cavity. In the 15–20 m depth range,
p9 the slope of the decay curves decreases, similarly to that VS varies between 460and 500 m/s and this is in close agree-
observed in the DSDSS tests. ment with that obtained from the CH test. Thus the SASW
Damping ratios from TS and DSDSS tests show a similar technique is reliable for these materials and can be used when
trend. Conversely, damping ratios from RC tests are smaller borehole tests cannot be performed.
than those obtained from TS tests at a comparable mean In Fig. 14 UPV measurements on specimens recovered
confining stress and often show a significant decrease at the from the CH boreholes are also reported. Within the SFW
highest strain amplitudes. lithotype, VS exhibits the same gradient that was described
in CH measurements. Laboratory UPV values are only
slightly larger than field values, because of the lower bulk
In situ tests density of dry specimens. This evidence suggests that the
In Fig. 14, S-wave velocities from CH and SASW tests widely spaced discontinuities crossing the pozzolana scarcely
are plotted against depth, together with the borehole strati- have an influence on the propagation of elastic waves, at
graphy. In the SFW pozzolana layer, VS from the CH test least at the scale of the borehole distance.
43
250 8
p 240 kPa p 240 kPa
200
γc 6
0·01%
150 0·04%
G: MPa
D: %
0·1% 4 γc
0·1%
100 0·04%
2 0·01%
50
0 0
0 4 8 12 0 4 8 12
Number of cycles Number of cycles
250 8
γc p 495 kPa
0·01%
200
0·04%
6
0·1%
150
G: MPa
D: %
4
γc
100 0·1%
0·04%
2
50 0·01%
p 495 kPa
0 0
0 4 8 12 0 4 8 12
Number of cycles Number of cycles
Fig. 11. Variation of shear modulus (G) and damping ratio (D) plotted against number of cycles in a typical series of
DSDSS tests (specimens SF-BL2de) at different strain amplitudes ªc and for the two highest mean stresses
1·0
DISCUSSION
The values of the small-strain shear modulus G0 from the
cyclic tests (DSDSS and TS) on the SFW pozzolana are
0·9 reported in Fig. 15 plotted against the mean confining stress
p9. The G0 values from TS tests are generally higher than
GN /G1
those from DSDSS tests, which are measured at a slightly
higher strain. Owing to the early development of permanent
0·8 strains and the consequent decay of the shear modulus, the
GN /G1, DN /D1
most precise measure of G0 is provided by TS tests, and

determined even at a strain amplitude as small as 0 .0001%.
Variation of G0 with the mean stress p9 can be interpolated
0·7 by a linear or a power law function with stress exponent
equal to 0 .5. A mean shear modulus G0 ¼ 127 MPa might
DN /D1
be assumed for a null confinement. A general decrease in
0·6
the small-strain modulus as the void ratio increases is ob-
served for both TS and DSDSS data sets (Fig. 15).
This behaviour is confirmed by UPV tests (Fig. 16), which
indicate an apparent inverse correlation between Gdyn and
0·5 porosity n. Data show a moderate dispersion due to the
20 40 60 80 100 120 140 heterogeneity of the material. The values of Gdyn obtained
Number of cycles
from the UPV tests generally agree with the G0 values for a
Fig. 12. Stiffness (�G) and damping (�D) degradation indexes mean pressure p9 ranging from 60 to 250 kPa.
plotted against number of cycles in a DSDSS test at ªc 0 .1% Plots of normalised shear modulus G/G0 and damping
(specimen SB10-5) ratio D against ªc for all the cyclic and dynamic tests
44
500 6
SF-BL2n SF-BL2n
p 500 kPa TS
400
UPV RC
4 250
300 250
G: MPa
D: %
120 p 0
200 60
2 60
p 0
120
100
p 500 kPa
0 0
0·0001 0·001 0·01 0·1 0·0001 0·001 0·01 0·1
γc: % γc: %
500 6
SF-BL2o SF-BL2o
400
500 kPa 4
300 250 p from RC
G: MPa
D: %
UPV 120 p 0
60
200 120
60 2 120
250
100 p 0 60 500 kPa
250
500
0 0
0·0001 0·001 0·01 0·1 0·0001 0·001 0·01 0·1
γc: % γc: %
500 6
SF-BL2q SF-BL2q
400
500 kPa
4 120
300
G: MPa
250
D: %
p 0
120
200
2 60
UPV 60
100
p 0 250
500 kPa
0 0
0·0001 0·001 0·01 0·1 0·0001 0·001 0·01 0·1
γc: % γc: %
500 6
SF-BL2r SF-BL2r
400 500 kPa

4 60
300 250
G: MPa
D: %
120
200 UPV 250
60
2 p 0
p 0 120
100
500 kPa
0 0
0·0001 0·001 0·01 0·1 0·0001 0·001 0·01 0·1
γc: % γ c: %
Fig. 13. Plots of shear modulus (G) and damping ratio (D) against strain amplitude ªc for different specimens subjected to TS
(black thin lines) and RC (grey thick lines) tests. Values of the mean stress p9 are indicated. Dynamic shear moduli obtained
through the UPV tests are shown by arrows on the vertical axes
carried out on the SFW pozzolana, grouped according to the their band. Consequently the linear strain threshold ªl, from
applied mean stress, are reported in Fig. 17. The G/G0–ªc DSDSS tests (0 .001–0 .01%), too, is higher than that from
curves from TS and RC tests, merged into a single curve for TS–RC tests (0 .0003–0 .003%). The reduced values of G/G0
each specimen, extend up to a strain of 0 .06%, whereas the for the same stress level, obtained from the RC–TS data,
DSDSS curves reach a maximum ªc of 0 .1–0 .3%. In the can be explained by the fact that the delicate microstructure
same figure, the G/G0–ªc and D–ªc curves proposed by can be irreversibly damaged, even at the lowest strain levels,
Rollins et al. (1998) for granular soils merged into a single when a high number of cycles is applied. In particular,
band, and those obtained by Papa et al. (1988) for cemented damage under repeated strains could concentrate within the
pozzolanas, are plotted as reference. Curves by Rollins et al. very weak ground mass, interposed between the stiffer clasts,
mostly refer to tests on poorly graded clean gravels and which are not in contact each other.
gravelly sands, with few well-graded gravels. DSDSS results also indicate that the rate of decay of shear
G/G0–ªc curves from DSDSS tests lie to the right of those stiffness diminishes at the highest investigated strain ampli-
by Rollins et al., while curves from TS–RC tests lie within tudes (ªc > 0 .1%). This strain-hardening behaviour, which
45
VS: m/s 500
p: kPa
200 400 600 800 1000
0 500 kPa G0 500
250 kPa 250

400
120
5 60
G0, Gdyn: MPa

300 0
10 200
120 kPa
100
60 kPa
SFW
15
0
Gdyn
0
Depth: m
0·58 0·59 0·60 0·61 0·62

20 n
Fig. 16. Small-strain shear modulus G0 from cyclic (DSDSS and

TS) tests and dynamic shear modulus Gdyn from UPV tests plotted
against porosity n. Regression lines (dashed) of G0 from data
25 obtained at different mean stresses p9 are also reported
has already been observed in the upwards curvature of the

30 tips of the stress–strain loops in Fig. 9, might also affect the
damping ratio.
Material damping has been determined in the strain range
from 0 .0001% to about 0 .001% from TS tests, whereas
SFH
35 SASW reliable damping ratio measurements have been obtained

between 0 .0004% and 0 .3% in the DSDSS tests. Values of
CH test damping ratio from TS and DSDSS tests compare very
favourably at the same mean confining stress p9. Conversely,
UPV test (linear interpolation)
40
damping ratios from the RC test are considerably smaller.
Possible explanations for this are currently being explored.
Fig. 14. Profiles of shear wave velocity VS from cross-hole Values of damping ratio from DSDSS and TS plot within
measurements (CH), SASW investigation and UPV tests on the range proposed by Rollins et al. (1998) up to about
specimens from cores recovered in the CH boreholes. The UPV 0 .002–0 .003%. At higher shear strain amplitudes (up to
data of the SFW lithotype are interpolated with a linear regression 0 .1%), values of damping ratio plot well below the literature
(dashed line)
curves, especially for higher stress levels. Again, a similar
trend has been previously reported in several studies on
pyroclastic materials (e.g. Guadagno et al., 1988; Papa et
al., 1988; Senetakis et al., 2013).
Finally, the dependence on the number of cycles observed
400 in cyclic tests at high strain levels is different for stiffness and
damping properties (Fig. 11). The stiffness decrease is moder-
G 126·8 7·8p0·5 1·42
1·42 ate and gradual, while the damping ratio sharply decreases,
1·43 especially in the first cycles at the lowest mean stresses. More
1·51 specifically the damping ratio quickly diminishes down to
300
1·42 20% of the initial value and then decreases at a progressively
1·4
1·51 lower rate. Tests with over 100 cycles show a non-negligible
1·42
residual damping decrease (Fig. 12). This behaviour is corre-
1·57 lated to the different results yielded by DSDSS and TS–RC
1·42 1·52 1·57
G0: MPa
1·51 tests.
200 1·51 1·57
1·4
1·42
1·42 1·4 CONCLUSIONS
1·51 1·57 1·57 The cyclic/dynamic behaviour of the Orvieto pozzolana
1·43 shows a number of peculiarities that deserve special con-
100
sideration in geotechnical earthquake engineering studies.
DSDSS The small-strain stiffness from geophysical measurements
(cross-hole and surface wave techniques) compares quite
TS
well with that obtained from pulse velocity measurements on
0 laboratory specimens. This result, a consequence of the very
0 100 200 300 400 500 600 wide spacing of discontinuity, is of undoubted practical
p: kPa
importance.
Fig. 15. Small-strain shear modulus G0 plotted against mean Small-strain shear moduli obtained from laboratory simple
stress p9 for cyclic (DSDSS and TS) tests. Void ratio values of the shear and cyclic torsional tests also compare satisfactorily
specimens are reported with the dynamic moduli from UPV tests. Overall, the
46
10
p 0
1·0
Papa et al. (1988) 8
Rollins et al. (1998)
0·8 6
D: %
G/G0
4
0·6 DSDSS
2
TS–RC p 0
0·4 0
0·0001 0·001 0·01 0·1 0·0001 0·001 0·01 0·1
γc: % γc: %
10
p 60 kPa
1·0
8
0·8 6
D: %
G/G0
4
0·6
2
p 60 kPa
0·4 0
0·0001 0·001 0·01 0·1 0·0001 0·001 0·01 0·1
γc: % γc: %
10
p 125 kPa
1·0
8
0·8 6
D: %
G/G0
4
0·6
2
p 125 kPa
0·4 0
0·0001 0·001 0·01 0·1 0·0001 0·001 0·01 0·1
γc: % γc: %
10
p 215–250 kPa
1·0
8
0·8 6
D: %
G/G0
4
0·6
2
p 215–500 kPa
0·4 0
0·0001 0·001 0·01 0·1 0·0001 0·001 0·01 0·1
γc: % γc: %
10
p 425–450 kPa
1·0
8
0·8 6
D: %
G/G0
4
0·6
2
p 425–500 kPa
0·4 0
0·0001 0·001 0·01 0·1 0·0001 0·001 0·01 0·1
γc: % γc: %
Fig. 17. Variation of normalised shear modulus G/G0 and damping ratio D with the shear strain amplitude ªc
from all the cyclic and dynamic tests (DSDSS, TS, RC), grouped for different stress level
influence of mean confining stress and void ratio on G0 is pronounced, especially at high strain level and low confine-
similar to that observed in other granular materials. ment stress.
The high linearity threshold and the decay rate of G/G0 Values of damping ratio from cyclic tests (TS and
with ªc obtained from DSDSS tests are comparable with DSDSS) are in satisfactory agreement with those exhibited
those observed for other pyroclastic materials (both unce- by other pyroclastic materials (both uncemented and weakly
mented and weakly cemented). The decay of G/G0 observed cemented). With respect to other granular materials, a more
in TS–RC tests starts at a lower strain level and is more linear dissipative behaviour is therefore observed, with a
47
slight dependence of the damping ratio on shear strain, even rs density of the solid particles
at ªc of about 0 .1%. According to Papa et al. (1988) and �f uniaxial strength
Senetakis et al. (2013), it is possible that micro-crushing at �t tensile strength
the particle contacts is a major mechanism during cyclic �v, �h vertical and horizontal total stress
loading, thus preventing significant particle rearrangement � shear stress
and increase in damping. The Orvieto pozzolana particles
are to be considered not only as single pumices or scoriae,
but also as aggregates of weakly bonded minuscule glassy REFERENCES
masses. Allen, J. C. & Stokoe, K. H. (1982). Development of a resonant
Finally, a dependence of stiffness and damping on the column apparatus with anisotropic loading, Geotechnical Engi-
number of cycles has been observed. Both the cyclic degra- neering Report GR82-28. Austin, TX, USA: The University of
dation indexes and the normalised stiffness provided by Texas at Austin.
dynamic RC tests suggest that the microstructure of the Bjerrum, L. & Landva, A. (1966). Direct simple-shear test on a
pozzolana has a marked influence on the cyclic behaviour. Norwegian quick clay. Géotechnique 16, No. 1, 1–20, http://
In fact, the delicate three-dimensional lattice structure of the dx.doi.org/10.1680/geot.1966.16.1.1.
glassy mass, separating the stiffer and stronger clasts, seems Cecconi, M. (1998). Sample preparation of a problematic pyroclas-
tic rock. Proceedings of international symposium on problematic
to be easily damaged when subjected to a high number of
soils, Sendai, Japan, pp. 165–168. Rotterdam, the Netherlands:
cycles. Balkema.
Further studies of the cyclic/dynamic behaviour of these Cecconi, M., Scarapazzi, M. & Viggiani, G. MB. (2010). On the
pyroclastic materials are in progress to confirm the findings geology and the geotechnical properties of pyroclastic flow
and to be more conclusive regarding the issues at hand. deposits of the Colli Albani. Bull. Engng Geol. Environ. 69, No.
2, 185–206.
Cecconi, M., Rotonda, T., Tommasi, P. & Viggiani, G. M. B.
ACKNOWLEDGEMENTS (2011). Microstructural features and compressibility of volcanic
deposits from Central Italy. Proceedings of the international
The authors would like to give special thanks to Luigi symposium on deformation characteristics of geomaterials,
Callisto (DISG – Sapienza Università di Roma) for sug- Seoul, Korea, pp. 884–891. Amsterdam, the Netherlands: IOS
gestions and helpful insights into various aspects of the TS– Press.
RC testing. The support of Alessandro Pagliaroli, who con- Choi, W. K. (2008). Dynamic properties of ash-flow tuffs. PhD
ducted part of the DSDSS testing, is also gratefully acknow- thesis, The University of Texas at Austin, Austin, TX, USA.
ledged. The help of Silvano Silvani, Maurizio Di Biase and D’Elia, B., Lanzo, G. & Pagliaroli, A. (2003). Small-strain stiffness
Francesco Coni in carrying out mechanical tests is greatly and damping of soils in a direct simple shear device. Proceed-
appreciated. The cross-hole test was carried out by Solgeo ings of the 2003 Pacific conference on earthquake engineering,
Srl, Seriate. Underground sampling was carried out with the Christchurch, New Zealand, paper no. 111. Wellington, New
Zealand: New Zealand National Society for Earthquake Engi-
support of Speleotecnica Srl, Orvieto.
neering.
Doroudian, M. & Vucetic, M. (1995). A direct simple shear device
for measuring small strain behaviour. ASTM Geotech. Testing J.
NOTATION 18, No. 1, 69–85.
A specimen transversal area Faraone, D. & Stoppa, F. (1988). Il tufo di Orvieto nel quadro
D damping ratio dell’evoluzione vulcano-tettonica della caldera di Bolsena, Monti
D0 small-strain damping ratio from cyclic tests Vulsini. Bollettino della Società Geologica Italiana 107, 383–
D1, DN damping ratio at the first, Nth cycle 397 (in Italian).
d diameter of the particles Guadagno, M., Rapolla, A., Ni, S. & Stokoe, K. H. (1988).
Esec Young’s secant modulus Dynamic properties of pyroclastic soils of the Phlegraean Fields.
Et50 Young’s tangent modulus at 50% of failure Proceedings of the 9th world conference on earthquake engi-
e void ratio neering, Tokyo-Kyoto, vol. 3, pp. 35–40. Tokyo, Japan: Maru-
F horizontal force in DSDSS test zen.
f cycle frequency in cyclic tests Isenhower, W. M. (1979). Torsional simple shear/resonant column
G shear modulus properties of San Francisco Bay mud. MSc thesis, The Uni-
Gdyn, Kdyn, Edyn shear, volumic bulk, Young’s modulus from UPV versity of Texas at Austin, TX, USA.
tests Jeon, S. Y. (2008). Dynamic and cyclic properties in shear of tuff
G0 small-strain shear modulus from cyclic tests specimens from Yucca mountain, Nevada. PhD thesis, The Uni-
G1, GN shear modulus at the first and the Nth cycle versity of Texas at Austin, TX, USA.
h specimen height Liu, X. & Yang, J. (2014). Laboratory measurements of small-strain
Koed earth pressure coefficient in oedometric conditions shear modulus of volcanic soils. In Geo-congress 2014 technical
K0 earth pressure coefficient papers: geo-characterization and modeling for sustainability
N number of cycles (eds M. Abu-Farsakh and L. R. Hoyos), Geotechnical Special
Nd number of data in computations of statistical Publication no. 234, pp. 113–122. Reston, VA, USA: ASCE.
parameters Marks, S., Larkin, T. J. & Pender, M. J. (1998). The dynamic
n porosity properties of a pomiceous sand. Bull. New Zealand Nat. Soc.
P vertical load in DSDSS test Earthquake Engng 31, No. 2, 86–102.
p9 effective mean stress Miura, S., Yagi, K. & Asonuma, T. (2003). Deformation–strength
p9y isotropic yielding stress evaluation of crushable volcanic soils by laboratory and in-situ
VS, V P shear and longitudinal wave velocity testing. Soils Found. 43, No. 4, 47–57.
ª shear strain Nazarian, S. (1984). In situ determination of elastic moduli of soil
ªc cyclic shear strain amplitude deposits and pavement systems by spectral-analysis-of-surface-
ªl strain linear threshold waves method. PhD thesis, The University of Texas at Austin,
� horizontal displacement in DSDSS tests TX, USA.
�G, �D degradation indexes of the shear modulus and Nappi, G., Capaccioni, B., Renzulli, A., Santi, P. & Valentini, L.
damping ratio with number of cycles (1994). Stratigraphy of the Orvieto-Bagnoregio Ignimbrite erup-
�a axial strain tion (Eastern Vulsini district, central Italy). Memorie Descrittive
�dyn Poisson coefficient from UPV tests della Carta Geologica d’Italia, vol. XLIX, pp. 241–254. Rome,
rd dry bulk density Italy: Servizio Geologico d’Italia..
48
Ni, S. H. (1987). Dynamic properties of sand under true triaxial volcanic rocks, Madeira, vol. 2, pp. 137–146. Lisbon, Portugal:
stress state from resonant column/torsional shear test. PhD Sociedade Portuguesa de Geotecnia.
thesis, The University of Texas at Austin, TX, USA. Sahaphol, T. & Miura, S. (2005). Shear moduli of volcanic soils.
Orense, R. P., Hyodo, M. & Kaneko, T. (2012). Dynamic deforma- Soil Dynamics Earthquake Engng 25, No. 2, 157–165.
tion characteristics of pumice sand. Proceedings of the New Santi, P. (1991). New geochronological data of the Vulsini volcanic
Zealand Society for Earthquake Engineering conference (2012 district (central Italy). Plinius 4, 91–92.
NZSEE), Christchurch, New Zealand, paper no. 6. Wellington, Senetakis, K., Anastasiadis, A. & Pitilakis, K. (2012). The small-
New Zealand: New Zealand National Society for Earthquake strain shear modulus and damping ratio of quartz and volcanic
Engineering. sands. Geotech. Testing J. 35, No. 6, 964–980, http://dx.doi.org/
Papa, V., Silvestri, F. & Vinale, F. (1988). Analisi delle proprietà di 10.1520/GTJ20120073.
un tipico terreno piroclastico mediante prove dinamiche di taglio Senetakis, K., Anastasiadis, A., Pitilakis, K. & Coop, M. R. (2013).
semplice. Convegno ‘Deformazioni dei terreni ed interazione The dynamics of a pumice granular soil in dry state under
terreno-struttura in condizioni di esercizio’, Monselice, Padova, isotropic resonant column testing. Soil Dynam. Earthquake
Italy, vol. 1, pp. 265–285. Rome, Italy: CNR – Gruppo Nazio- Engng 45, 70–79.
nale di Coordinamento per gli Studi di Ingegneria Geotecnica Silvestri, F. (1991). Analisi del comportamento dei terreni naturali
(in Italian). in prove cicliche e dinamiche di taglio torsionale. PhD thesis,
Pellegrino, A. (1969). Proprietà fisico-meccaniche dei terreni vulca- Federico II University of Naples, Italy (in Italian).
nici del Napoletano. Proceedings of the VIII national congress Tommasi, P., Sciotti, M., Rotonda, T., Verrucci, L. & Boldini, D.
of geotechnics, Cagliari, Italy, vol. 3, pp. 113–145. Napoli, Italy: (2013). The role of geotechnical conditions in the foundation
Edizioni Scientifiche Italiane (in Italian). expansion and preservation of the ancient town of Orvieto
Rollins, K. M., Evans, M., Diehl, N. & Daily, W. (1998). Shear (Italy). In Geotechnics and heritage (eds E. Bilotta, A. Flora, S.
modulus and damping relationships for gravels. J. Geotech. Lirer and C. Viggiani), pp. 49–73. London, UK: CRC Press,
Geoenviron. Engng 124, No. 5, 396–405. Taylor and Francis Ltd.
Rotonda, R. & Ribacchi, R. (1995). Caratteristiche dinamiche di Tommasi, P., Verrucci, L. & Rotonda, T. (2015). Mechanical proper-
rocce porose e fessurate. Rivista Italiana di Geotecnica 1, No. ties of a weak pyroclastic rock and their relationship with
1995, 17–36 (in Italian). microstructure. Can. Geotech. J. 52, No. 2, 211–223, http://
Rotonda, R., Tommasi, P. & Ribacchi, R. (2002). Physical and dx.doi.org/10.1139/cgj-2014-0149.
mechanical characterization of the soft pyroclastic rocks forming Vucetic, M. (1994). Cyclic threshold shear strains in soils. J.
the Orvieto cliff. Proceedings of Eurock 2002, workshop on Geotech. Engng 120, No. 12, 2208–2228.
49
Lee, J. & Green, R. A. (2015). Géotechnique 65, No. 5, 374–390 [http://dx.doi.org/10.1680/geot.SIP.15.P.011]
Empirical predictive relationship for seismic lateral displacement

of slopes: models for stable continental and active crustal regions
J. L E E a n d R . A . G R E E N †
The objective of this study is to develop an empirical predictive relationship for permanent lateral
displacements for use in assessing the seismic stability of slopes, earthen dams and/or embankments
subject to stable continental earthquake motions. The empirical relationship is developed from 620
horizontal motions for stable continental regions, consisting of 28 recorded motions and 592 scaled
motions. For each motion, the permanent relative displacements are computed using the Newmark
sliding block procedure for a suite of yield accelerations. The proposed predictive relationship is
derived by performing separate regression analyses for each yield acceleration. This allows the
relationship to be simply formulated in terms of ground motion characteristic parameters, independent
of yield acceleration, and results in lower standard deviations than those for relations developed by
regressing all the data in a single analysis. The non-linear mixed-effects technique is used to regress
the data as functions of maximum ground accelerations and velocities. To account for the zero
displacement data, logistic regression was conducted to model the probability of zero-displacement
occurrences. Then, the probability models were applied as weighting functions to the non-linear
mixed-effects regression results. Also, a similar relationship for active shallow crustal motions is
developed and compared with the stable continental region relationship. Lastly, the predicted displace-
ments from the proposed model are shown to be in good agreement with those computed using
motions recorded during the Mineral, Virginia earthquake of 23 August 2011.
KEYWORDS: dams; deformation; dynamics; earthquakes; embankments; retaining walls; slopes
INTRODUCTION lated permanent relative displacement to the ratio of ky and

The objective of the study presented herein is to develop an the maximum peak horizontal ground acceleration (Amax), in
empirical predictive relationship for permanent relative addition to other ground motion parameters. Furthermore,
displacements for use in assessing the seismic stability of this study differs from the previous studies in that the
slopes, earthen dams and/or embankments subjected to stable relationships proposed herein were developed by performing
continental earthquake motions (e.g. central eastern North separate regressions for each ky using the non-linear mixed-
America (CENA)) for different site conditions (rock and effects (NLME) technique. Performing separate regressions
soil). The Newmark sliding block method was used to for each ky allowed the relationships to have relatively low
compute the permanent relative displacements. This method standard deviations and to have simple functional forms that
was proposed by Newmark (1965) for evaluating the seismic are independent of ky. Using NLME regression analyses
stability of slopes, wherein the sliding mass is modelled as a results in unbiased fits of the data, irrespective of the
block on an inclined plane. Displacement of the block varying amount of data from different earthquakes. The
relative to the plane initiates when the yield acceleration (ky) authors present a similarly developed relationship for active
is exceeded and continues until the velocities of the block shallow crustal regions (e.g. western North America: WNA).
and ground coincide. The permanent relative displacement is This WNA relationship was developed so that consistent
defined as the cumulative relative displacement at the end of comparisons in permanent relative displacement for the two
ground shaking, as illustrated in Fig. 1. regions could be made. Although there are existing relation-
Numerous empirical relationships for estimating perma- ships for permanent relative displacement for WNA, the
nent relative displacements have been developed over the authors felt that in order to avoid issues related to differ-
past 30 years (e.g. Franklin & Chang, 1977; Richards & ences in predicted displacement due to disparities in data-
Elms, 1979; Ambraseys & Menu, 1988; Ambraseys & base size, analysis techniques and so on, consistently
Srbulov, 1994; Bray & Travasarou, 2007; Jibson, 2007; developed relationships for the two regions were needed to
Saygili & Rathje, 2008; Rathje & Saygili, 2009; Rathje & make valid comparisons. The correlations were developed by
Antonakos, 2011). However, these relationships are for ac- performing NLME regression analyses on data derived from
tive shallow crustal tectonic regimes, whereas the focus of horizontal motion recorded in active shallow crustal regions
the study presented herein is stable continental regimes. (e.g. WNA), and horizontal motions for stable continental
Also, most of these previous studies used fixed-effects regions (e.g. CENA) that were composed of both recorded
regression techniques (e.g. least-squares method) and corre- motions and scaled motions; this is discussed in more detail
later in the paper. (Note that the acronyms CENA and WNA
are used in this paper for convenience to refer to ‘stable
Manuscript received 1 April 2014; revised manuscript accepted 5 continental’ and ‘active shallow crustal’ regions, respectively.
February 2015.
However, the authors contend that the use of these respective
see p. ii. relationships are not solely limited to North America, but
Arup, San Francisco, CA, USA. rather are applicable for use in stable continental and active
† Department of Civil and Environmental Engineering, Virginia Tech, shallow tectonic regions in other parts of the world too.)
Blacksburg, VA, USA. The paper is organised as follows: first, the strong ground
51
LEE AND GREEN
Block
0·1 Yield acceleration
Acceleration: g
Ground
0·1
20
Velocity: cm/s
20
15
Relative displacement: cm
Permanent lateral displacement
10
0
0 1 2 3 4 5 6 7 8 9 10
Time: s
Fig. 1. Example of Newmark sliding block analyses for ky 0 .03g and a ground acceleration–time history
(BES090: M 6 .9; R 49 .9 km) from the 1989 Loma Prieta earthquake
motion databases used in this study are described. Then, 0 .1 km to 199 .1 km, where site-to-source distance is defined
basic concepts of the NLME regression method are re- as the closest distance to the fault rupture plane. Because
viewed, and the proposed functional form of the predictive there are few recorded strong ground motions in stable
model is introduced. Next, the results of the regression continental regions, only 28 of the motions in the CENA
analyses and a comparison of permanent relative displace- dataset are recorded motions, with the remaining 592 mo-
ments predicted by this study’s empirical relationships for tions being ‘scaled’ WNA motions for CENA conditions. A
stable continental and active shallow crustal regions are brief summary of the scaling procedure is provided below,
presented. A comparison of this study’s relationship to with a more detailed description of the scaling procedure
existing relationships is then presented. Additionally, in an given in the Appendix. The moment magnitudes (Mw) for
effort to validate the proposed relationship, the displace- these motions range from 4 .5 to 7 .6, and the site-to-source
ments computed using motions recorded during the recent distances range from 0 .1 km to 199 .1 km. The recorded
2011 Mineral, Virginia earthquake are compared with those motions include the 1988 Saguenay (Mw5 .9 mainshock and
predicted by the CENA model proposed herein. Finally, the Mw4 .5 aftershock), the 1985 Nahanni (Mw6 .8), and the 1989
procedure used to scale the scaled CENA motions is sum- New Madrid, MO (Mw4 .7) earthquakes. Fig. 2 shows the
marised in the Appendix. magnitude and site-to-source distance distributions for both
regions.
McGuire et al. (2001) scaled the WNA motions for
STRONG GROUND MOTION DATA CENA conditions using response spectral transfer functions
The datasets used in this study consist of 324 two- generated from the single-corner frequency point source
component sets of horizontal strong ground motion time model in conjunction with random vibration theory (RVT)
histories from WNA and 310 sets for CENA. Thus, totals of (e.g. Brune, 1970, 1971; Boore, 1983; Silva & Lee, 1987;
648 and 620 horizontal time histories were used to develop McGuire et al., 2001). The transfer functions account for the
the empirical relationships for WNA and CENA, respec- differences in seismic source, wave propagation path proper-
tively. This study adopted the ground motion dataset as- ties, and site effects between the WNA and CENA regions.
sembled by McGuire et al. (2001) from the strong ground Many seismological publications have shown successful re-
motion database processed by Dr Walter Silva of Pacific sults of the RVT point source model for generating strong
Engineering and Analysis. Primarily, this dataset was in- ground motions for both WNA and CENA (Hanks &
tended to provide a library of strong ground motion time McGuire, 1981; Boore, 1983, 1986; McGuire et al., 1984;
histories suitable for engineering analyses. Hence, the data Schneider et al., 1993; Silva, 1993). In generating the scaled
selection criteria (e.g. earthquake magnitude range) were CENA motions, recorded WNA motions were used as ‘seed’
established based on engineering interest. The criteria were motions in the spectral scaling process, resulting in scaled
categorised by: site condition, earthquake magnitude, site-to- motions that have realistic characteristics. In this context,
source distance and strong ground motion duration (McGuire the stochastic point source model is a reliable and reason-
et al., 2001; Lee, 2009). The strong motion data for WNA able approach for estimating spectral characteristics of
were from 49 mainshock events, with the 1999 Chi-Chi strong ground motions for engineering analyses. The scaling
earthquake being the most recent event included in the method, however, should be validated as additional record-
database. The moment magnitudes of these events range ings of stable continental motions become available.
from 5 .0 to 7 .6, and the site-to-source distances range from The ground motions were classified as either ‘rock’ or
52
PREDICTIVE RELATIONSHIP FOR SEISMIC LATERAL DISPLACEMENT OF SLOPES
8·0
dual data points in the regression analyses. The permanent
relative displacements were computed for a suite of yield
7·5 accelerations. The displacement data were then correlated to
the maximum ground acceleration (Amax) and velocity
7·0 (Vmax). Note that the Amax and Vmax are defined as a peak
acceleration and velocity in the opposite direction as the
Magnitude, M
6·5
displacement of a sliding block. They may differ from the
peak ground acceleration (PGA) and the peak ground velo-
city (PGV), which are the maximum ‘absolute’ acceleration
6·0 and velocity.
5·5
REGRESSION ANALYSES
Rock The empirical predictive relationship developed in this
5·0
Soil study consists of two models: a model for conditional
expected non-zero displacement and a model for the prob-
4·5 1
10 100 101 102 103 ability of non-zero displacement occurrence. This can be
Site-to-source distance, R: km expressed using the total probability theorem
8·0 E[D] ¼ E[DjD . 0] 3 p(D . 0) þ E[DjD ¼ 0] 3 p(D ¼ 0)
(1)
7·5
where E[ .] and p( .) represent expected value and probability,
7·0 respectively. The second term on the right side of Equation
1 is zero since E[D|D ¼ 0] is zero. The expected value for
Magnitude, M
6·5
D, therefore, becomes equal to the quantity of the expected
value for a given non-zero displacement (i.e., E[D|D . 0])
multiplied by the probability of non-zero displacement oc-
6·0 currence p(D . 0). The NLME regression was performed
for modeling the expected value of displacement for a given
5·5 non-zero displacement. A logistic regression method was
employed to model the probability of non-zero displacement
Rock occurrence.
5·0
Soil The proposed models were derived by performing separate
regression analyses for each ky. This approach allows the
4·5 1
10 100 101 102 103 predictive relationship to be formulated only in terms of
Site-to-source distance, R: km ground motion characteristic parameters, independent of ky,
and allows the standard deviations to be estimated for each
Fig. 2. Earthquake magnitude and site-to-source distance dis- ky value. This inherently allows the uncertainty in ky to be
tributions (recorded motions for CENA are shown in bold):
treated as epistemic uncertainty, which can be reduced by
(a) active shallow crustal; (b) stable continental
way of field investigations, as opposed to being treated as
aleatory variability in the predicted displacements. Also, this
‘soil’, based on the site conditions at the respective seismo- approach is in contrast to previous studies where permanent
graph stations. The site classification scheme used by relative displacements were correlated to the ratio ky /Amax,
McGuire et al. (2001) is based on the third letter of the which results in complex functional forms and relatively
Geomatrix three-letter site classification system shown in large total standard deviations. This is attributable primarily
Table 1. Site categories A and B were considered to to the large variations in displacements for a given ky /Amax,
represent rock sites, and site categories C, D and E were as shown in Fig. 3. A suite of discrete ky values were
considered to represent soil sites (note that there were only a considered for the separate regressions, which were distrib-
few motions in the McGuire et al. database that were uted with an approximately equal interval in a log scale
recorded in category E sites). (listed in Tables 2–5). A more detailed discussion of the
As may be surmised from Fig. 1, the permanent relative merits/disadvantages of performing separate regression ana-
displacements may vary with the orientation (or sign) of the lyses for each ky against ky /Amax is presented subsequently.
ground motion. Accordingly, permanent relative displace- As mentioned previously, the regression analyses were
ments were computed for both directions (i.e. +/�) of a performed on the permanent relative displacement data of
ground motion. These displacements were treated as indivi- the individual horizontal components of the dataset using the
Table 1. Third letter: geotechnical subsurface characteristics of Geomatrix three-letter site classification
Third Site description Comments

letter
A Rock Instrument on rock (VS . 600 m/s) or , 5 m of soil over rock

B Shallow (stiff) soil Instrument on/in soil profile up to 20 m thick overlying rock
C Deep narrow soil Instrument on/in soil profile at least 20 m thick overlying rock, in a narrow canyon or valley no more than several
kilometres wide
D Deep broad soil Instrument on/in soil profile at least 20 m thick overlying rock, in a broad valley
E Soft deep soil Instrument on/in deep soil profile with average VS , 150 m/s
53
LEE AND GREEN
CENA – rock
103
102 ky 0·01
ky 0·05
ky 0·10
101
ky 0·20
D: cm
Davg.: ky 0·01
100
Davg.: ky 0·05
Davg.: ky 0·10
101 Davg.: ky 0·20
102 2
10 101 100
ky/Amax
ky/Amax 0·05 ky/Amax 0·10 ky/Amax 0·20 ky/Amax 0·50
103 103 102 101
102
2
10 101 100
D: cm
D: cm
D: cm
D: cm
101
101 100 101

0
10
100 101 101 102

0·045 0·050 0·055 0·09 0·10 0·11 0·16 0·18 0·20 0·22 0·4 0·5 0·6
ky/Amax ky/Amax ky/Amax ky/Amax
Fig. 3. Displacement data and averages for ky /Amax 0 .05, 0 .10, 0 .20 and 0 .50 for each ky value considered in this study; plots for
each ky /Amax in semi-log scale (bottom) and a plot for all the ky /Amax in log–log scale (top)
NLME regression technique. The displacements less than inter-event error is defined as the difference between the
1 cm were considered as zero displacements, owing to median for the ith event and the median of the entire
engineering insignificance, and were not included in the database (i.e. model median) and has a mean of zero and a
NLME regression analyses. variance of 2. The intra-event error is designated by ij,
The NLME modelling is a maximum likelihood method where the subscripts ij indicate the jth record of the ith
based on normal (Gaussian) distribution and is primarily event. The intra-event error is defined as the difference
used for analysing grouped data (i.e. databases comprised of between the data value of the jth record and the median for
subsets). The NLME regression method allows regression the ith event and has a mean of zero and a variance of 2.
models to account for both random effects that vary from The total error for the jth record of the ith event is defined
subset to subset and fixed-effects that do not. In this study, a as the sum of the corresponding inter- and intra-event errors
subset consists of motions recorded during a given earth- (i.e. i + ij). The standard deviation of the total error is
quake. In comparison to applying a fixed-effects regression given by
technique (e.g., the least squares method) to the entire pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
dataset, a mixed-effects regression method allows both inter- total ¼ 2 þ 2 (2)
and intra-earthquake uncertainty to be quantified. This re-
gression method produces unbiased fittings for each subset
having different numbers of ground motion recordings. This where total is the standard deviation of total error, also
is important because of the number of motions from each called the total standard deviation. The NLME method
earthquake can widely vary. assumes the normal distribution for intra-event errors and
The NLME modelling estimates the variation in the mean random-effects (Pinheiro & Bates, 2000), which underlies
values among earthquakes (i.e. inter-event variability) and the theoretical formulation of the NLME regression analyses.
the variation in the data for a single earthquake (i.e. intra- The inherent distributional assumptions are checked by the
event variability) by way of the variances of inter-event normal quantile–quantile (Q–Q) plots, wherein the data
errors and intra-event errors, respectively. The inter-event points will plot approximately as a straight line if the data
error is designated by i, where the subscript i represents the are normally distributed. The statistical analysis program-R
ith event (i.e. set of motions from a given earthquake). The (version 3.0.2 (R Foundation, 2013)) was used to perform
54
Table 2. Regression results and data statistics: CENA – rock
ky: g Regression results Statistics of used data for D > 1 cm
C1 C2 C3 ln ln (ln)total 0 1 2 No. No. (ky /Amax)min (ky /Amax)max (Amax)min: g (Amax)max: g (Vmax)min: (Vmax)max:
data event cm/s cm/s
0 .01 1 .272 0 .564 1 .020 0 .355 0 .476 0 .593 11 .51 132 .26 1 .448 532 31 0 .002 0 .218 0 .046 4 .72 0 .92 127 .31
0 .014 0 .992 0 .618 1 .049 0 .299 0 .493 0 .577 8 .53 78 .88 0 .912 506 29 0 .003 0 .305 0 .046 4 .72 1 .04 127 .31
0 .02 0 .690 0 .702 1 .058 0 .267 0 .509 0 .575 11 .08 84 .48 0 .718 451 28 0 .004 0 .311 0 .064 4 .72 1 .51 127 .31
0 .03 0 .177 0 .645 1 .215 0 .194 0 .508 0 .543 9 .70 62 .92 0 .307 375 26 0 .006 0 .394 0 .076 4 .72 2 .00 127 .31
0 .04 0 .459 0 .685 1 .205 0 .240 0 .459 0 .519 8 .30 34 .42 0 .279 294 26 0 .008 0 .408 0 .098 4 .72 3 .42 127 .31
0 .05 0 .465 0 .818 1 .145 0 .221 0 .486 0 .534 10 .52 35 .12 0 .362 259 25 0 .011 0 .334 0 .150 4 .72 3 .42 127 .31
0 .06 0 .473 0 .967 1 .101 0 .187 0 .514 0 .547 9 .52 23 .88 0 .355 234 25 0 .013 0 .355 0 .169 4 .72 4 .56 127 .31
0 .08 0 .465 1 .119 0 .995 0 .262 0 .530 0 .591 10 .26 26 .40 0 .148 202 22 0 .017 0 .393 0 .204 4 .72 4 .56 127 .31
0 .1 0 .678 1 .177 0 .965 0 .304 0 .525 0 .607 11 .86 27 .46 0 .113 178 21 0 .021 0 .491 0 .204 4 .72 6 .76 127 .31
0 .14 1 .412 1 .197 1 .023 0 .352 0 .555 0 .657 9 .91 17 .30 0 .074 142 19 0 .030 0 .399 0 .351 4 .72 7 .80 127 .31
0 .18 2 .320 1 .183 1 .163 0 .333 0 .597 0 .684 11 .08 15 .42 0 .080 120 17 0 .038 0 .468 0 .384 4 .72 10 .40 127 .31
0 .2 2 .356 1 .231 1 .124 0 .408 0 .559 0 .692 9 .37 11 .65 0 .070 113 17 0 .042 0 .520 0 .384 4 .72 10 .80 127 .31
0 .28 2 .582 1 .297 1 .021 0 .338 0 .531 0 .630 9 .51 6 .64 0 .092 84 14 0 .059 0 .501 0 .559 4 .72 19 .88 127 .31
0 .36 1 .528 1 .411 0 .614 0 .385 0 .514 0 .642 12 .88 7 .70 0 .104 69 13 0 .076 0 .523 0 .689 4 .72 28 .04 127 .31
0 .4 1 .535 1 .618 0 .511 0 .403 0 .524 0 .661 12 .76 7 .86 0 .078 64 12 0 .085 0 .428 0 .935 4 .72 28 .04 127 .31
0 .5 0 .751 1 .801 0 .165 0 .437 0 .484 0 .652 9 .72 5 .66 0 .028 47 12 0 .106 0 .535 0 .935 4 .72 28 .04 127 .31
55
Table 3. Regression results and data statistics: CENA – soil
C1 C2 C3 ln ln (ln)total 0 1 2 No. (ky / (ky /Amax)max (Amax)min: g (Amax)max: g (Vmax)min: (Vmax)max:
data Amax)min cm/s cm/s
0 .01 1 .673 0 .516 1 .055 0 .489 0 .361 0 .608 12 .85 75 .46 3 .496 606 37 0 .006 0 .31 0 .032 1 .55 1 .78 146 .42
0 .014 1 .556 0 .671 1 .062 0 .484 0 .400 0 .628 28 .26 158 .84 6 .939 603 37 0 .009 0 .44 0 .032 1 .55 2 .28 146 .42
0 .02 1 .044 0 .711 1 .134 0 .448 0 .411 0 .608 10 .32 99 .16 0 .897 581 37 0 .013 0 .49 0 .041 1 .55 2 .52 146 .42
0 .03 0 .557 0 .837 1 .189 0 .398 0 .418 0 .577 14 .68 122 .67 0 .775 550 37 0 .019 0 .50 0 .060 1 .55 2 .62 146 .42
0 .04 0 .210 0 .974 1 .236 0 .350 0 .439 0 .562 13 .16 66 .78 0 .790 497 35 0 .026 0 .55 0 .073 1 .55 3 .87 146 .42
0 .05 0 .145 1 .133 1 .209 0 .343 0 .452 0 .568 13 .10 61 .74 0 .578 456 35 0 .032 0 .54 0 .092 1 .55 4 .37 146 .42
0 .06 0 .044 1 .237 1 .228 0 .288 0 .469 0 .551 10 .73 40 .93 0 .408 409 29 0 .039 0 .59 0 .103 1 .55 4 .37 146 .42
0 .08 0 .731 1 .277 1 .322 0 .168 0 .471 0 .500 9 .62 32 .66 0 .201 332 29 0 .052 0 .62 0 .130 1 .55 6 .75 146 .42
0 .1 1 .450 1 .327 0 .073 0 .521 0 .526 11 .95 280 23 0 .064 0 .66 0 .152 1 .55 8 .82 146 .42
0 .980 33 .04 0 .220

0 .14 0 .927 1 .775 1 .186 0 .201 0 .533 0 .570 12 .27 28 .20 0 .146 207 21 0 .090 0 .75 0 .188 1 .55 9 .41 146 .42
0 .18 1 .337 1 .870 1 .178 0 .210 0 .560 0 .599 11 .36 24 .70 0 .076 165 19 0 .116 0 .67 0 .267 1 .55 9 .41 146 .42
0 .2 1 .479 1 .845 1 .159 0 .247 0 .547 0 .600 10 .82 20 .24 0 .069 142 15 0 .129 0 .74 0 .269 1 .55 13 .28 146 .42
0 .28 1 .967 1 .946 1 .088 0 .305 0 .585 0 .660 13 .93 21 .37 0 .058 96 14 0 .180 0 .74 0 .378 1 .55 17 .28 146 .42
0 .36 2 .361 2 .406 1 .072 0 .263 0 .550 0 .609 12 .86 16 .74 0 .044 67 10 0 .232 0 .73 0 .494 1 .55 30 .95 146 .42
0 .4 1 .673 0 .516 1 .055 0 .489 0 .361 0 .608 16 .02 20 .28 0 .043 54 10 0 .258 0 .74 0 .540 1 .55 30 .95 146 .42
Table 4. Regression results and data statistics: WNA – rock
0 .01 1 .084 0 .571 1 .208 0 .240 0 .491 0 .547 7 .14 86 .94 1 .473 531 30 0 .006 0 .54 0 .019 1 .58 1 .32 125 .13
0 .014 0 .901 0 .652 1 .210 0 .230 0 .493 0 .544 8 .35 101 .86 0 .844 454 29 0 .009 0 .51 0 .027 1 .58 2 .02 125 .13
0 .02 0 .897 0 .788 1 .158 0 .239 0 .492 0 .547 8 .97 81 .98 0 .695 378 27 0 .013 0 .61 0 .033 1 .58 3 .03 125 .13
0 .03 0 .362 0 .842 1 .221 0 .213 0 .486 0 .531 10 .79 83 .75 0 .499 305 25 0 .019 0 .52 0 .058 1 .58 3 .33 125 .13
0 .04 0 .444 0 .985 1 .139 0 .243 0 .485 0 .543 11 .64 71 .79 0 .460 267 24 0 .025 0 .57 0 .070 1 .58 4 .50 125 .13
0 .05 0 .536 1 .125 1 .072 0 .265 0 .503 0 .569 14 .98 75 .61 0 .551 239 24 0 .032 0 .55 0 .091 1 .58 4 .66 125 .13
0 .06 0 .499 1 .170 1 .021 0 .313 0 .480 0 .573 10 .46 52 .90 0 .201 214 23 0 .038 0 .62 0 .096 1 .58 6 .16 125 .13
0 .08 0 .0017 1 .226 1 .072 0 .335 0 .470 0 .578 12 .61 58 .18 0 .135 182 19 0 .050 0 .67 0 .120 1 .58 6 .16 125 .13
0 .1 0 .077 1 .377 1 .031 0 .365 0 .487 0 .609 14 .31 52 .87 0 .152 155 18 0 .063 0 .72 0 .139 1 .58 8 .22 125 .13
0 .14 0 .810 1 .622 1 .115 0 .350 0 .572 0 .671 12 .26 36 .84 0 .100 126 18 0 .088 0 .75 0 .187 1 .58 9 .82 125 .13
0 .18 1 .812 1 .465 1 .252 0 .420 0 .509 0 .660 11 .33 25 .95 0 .084 96 14 0 .114 0 .70 0 .259 1 .58 16 .38 125 .13
0 .2 2 .021 1 .410 1 .242 0 .428 0 .497 0 .657 13 .21 27 .93 0 .092 86 14 0 .126 0 .69 0 .290 1 .58 16 .38 125 .13
0 .28 1 .195 1 .893 0 .884 0 .484 0 .458 0 .667 9 .12 13 .11 0 .073 67 12 0 .177 0 .79 0 .353 1 .58 24 .48 125 .13
0 .36 0 .959 2 .473 0 .732 0 .408 0 .491 0 .638 7 .85 9 .02 0 .032 37 10 0 .227 0 .74 0 .485 1 .58 26 .24 125 .13
56
Table 5. Regression results and data statistics: WNA – soil
LEE AND GREEN
0 .01 1 .687 0 .633 1 .116 0 .446 0 .424 0 .615 11 .96 229 .51 1 .820 583 36 0 .011 0 .52 0 .019 0 .94 1 .73 263 .21
0 .014 1 .325 0 .715 1 .177 0 .391 0 .444 0 .591 9 .36 126 .82 1 .092 541 36 0 .015 0 .56 0 .025 0 .94 2 .24 263 .21
0 .02 0 .922 0 .809 1 .234 0 .352 0 .436 0 .560 9 .24 104 .83 0 .733 495 32 0 .021 0 .57 0 .035 0 .94 3 .16 263 .21
0 .03 0 .493 0 .966 1 .299 0 .286 0 .441 0 .526 11 .07 75 .03 0 .760 410 29 0 .032 0 .65 0 .046 0 .94 4 .12 263 .21
0 .04 0 .892 1 .244 1 .172 0 .295 0 .437 0 .527 12 .00 64 .47 0 .628 341 27 0 .043 0 .68 0 .059 0 .94 4 .35 263 .21
0 .05 0 .681 1 .374 1 .204 0 .229 0 .481 0 .532 10 .21 53 .58 0 .335 298 23 0 .053 0 .73 0 .068 0 .94 5 .61 263 .21
0 .06 0 .519 1 .441 1 .204 0 .140 0 .494 0 .513 10 .10 46 .58 0 .231 263 20 0 .064 0 .71 0 .085 0 .94 6 .76 263 .21
0 .08 0 .107 1 .668 1 .245 0 .070 0 .519 0 .524 11 .89 58 .69 0 .139 240 20 0 .085 0 .69 0 .116 0 .94 6 .76 263 .21
0 .1 0 .032 1 .967 1 .244 1 .85 3 105 0 .561 0 .561 13 .08 54 .25 0 .116 205 18 0 .106 0 .72 0 .139 0 .94 8 .91 263 .21
0 .14 0 .245 2 .236 1 .161 0 .188 0 .577 0 .607 19 .45 58 .78 0 .120 142 16 0 .149 0 .70 0 .201 0 .94 12 .22 263 .21
0 .18 0 .609 2 .288 1 .121 0 .219 0 .578 0 .618 14 .75 43 .40 0 .025 110 15 0 .192 0 .73 0 .247 0 .94 15 .59 263 .21
0 .2 0 .725 2 .354 1 .105 0 .199 0 .566 0 .599 13 .63 36 .57 0 .026 98 13 0 .213 0 .80 0 .251 0 .94 15 .59 263 .21
0 .28 0 .722 2 .241 0 .866 0 .267 0 .618 0 .673 14 .33 31 .51 0 .014 63 10 0 .298 0 .80 0 .348 0 .94 22 .23 176 .65
0 .36 1 .607 1 .661 0 .866 0 .174 0 .501 0 .531 17 .02 30 .09 0 .010 35 8 0 .383 0 .74 0 .484 0 .94 36 .33 127 .18
the NLME and logistic regression analyses (Pinheiro & was, however, consistent with many other functional forms
Bates, 2000; Lee, 2009). attempted in the regression analyses. Thus, the functional
form of equation (3) was kept because of its simplicity and
associated lower standard deviation.
PROPOSED MODEL AND REGRESSION RESULTS As described above, the distributional assumptions for
After observing the trends from the data and considering intra-event errors and random effects were assessed by the
numerous functional forms (Lee, 2009), the following model normal Q–Q plots. Fig. 5 shows the Q–Q plots for CENA
was selected for the predictive relationship for all the ky and ky ¼ 0 .05. In these figures, the theoretical quantiles of
values: the standard normal distribution against the standardised
E[DjD . 0] ¼ exp[C 1 þ C 2 ln (Amax ) þ C 3 ln (V max )] (3) intra-event errors (i.e. intra-event errors divided by their
standard deviation) and random effects are plotted. As
where D is the permanent relative displacement (cm); Amax shown in Fig. 5, both intra-event errors and random effects
is the maximum ground acceleration (g); Vmax is the maxi- plot approximately as straight lines, indicating that the data
mum ground velocity (cm/s); and C1, C2, C3 are regression follow normal distributions, consistent with the assumptions
coefficients. The selected functional form (equation (3)) inherent to NLME modelling. The Q–Q plots for the other
produced a smaller standard deviation than the others, and N values, and so on, were similar to Fig. 5 and thus are not
the resulting residuals showed no bias and had a normal shown herein.
distribution (presented later this section). The regression To estimate the probability of non-zero displacement oc-
coefficients and standard deviations determined from NLME currence, logistic regressions were implemented separately
regression analyses are listed in Tables 2–5 for CENA – for each ky, tectonic regime and site condition as a function
rock, CENA – soil, WNA – rock and WNA – Soil, respec- of Amax and Vmax. Using the logistic function, the probability
tively. Also, the Amax and Vmax ranges associated with the of non-zero displacement for a given ky, tectonic regime and
displacement data used in the regression analyses for each ky site condition is expressed as follows
are listed in these tables. It is recommended that equation 1
(2) be used only for Amax and Vmax values that are within the p(D . 0jAmax , V max ) ¼
ranges listed in Tables 2–5. 1 þ exp [1 þ 2 Amax þ 3 V max ]
Figure 4 shows the comparison of the non-zero displace- (4)
ments computed directly from the motions (i.e. Dobserved)
data and the predicted displacements (i.e. E[D|D . 0]) from The regression coefficients (1 through 3) determined
the NLME regression results for CENA and three ky values: from logistic regression analyses are listed in Tables 2–5 for
0 .01, 0 .1 and 0 .2. Also Fig. 4 shows a straight line with a y CENA – rock, CENA – soil, WNA – rock and WNA – soil,
intercept of zero and a slope of 1 representing the predicted respectively. Also, the resulting weighting functions (prob-
values equal to the observed data. Overall the NLME results ability of non-zero displacement) for different tectonic re-
with the functional form of equation (3) represent the ob- gions and site conditions are shown in Fig. 6.
served data well without a significant bias except for small As shown in equation (1), the proposed model for perma-
displacements. The over-estimation for small displacements nent relative displacements is defined as the multiplication
CENA – rock: ky 0·01 CENA – rock: ky 0·10 CENA – rock: ky 0·20

7
5 4
6
E[D|D > 0] in ln unit
5 4 3
4
3
2
3
2
2 1
1 1
0 0
0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 0 1 2 3 4
ln Dobserved ln Dobserved ln Dobserved
CENA – soil: ky 0·01 CENA – soil: ky 0·10 CENA – soil: ky 0·20

5
7
5
6 4
4
5
3
4 3
2
3 2
2 1 1
1
0 0
0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 0 1 2 3 4 5
ln Dobserved ln Dobserved ln Dobserved
Fig. 4. Predicted displacements from NLME regression compared with observed data in a natural log unit for CENA
57
LEE AND GREEN
CENA – rock (ky 0·05 g) CENA – soil (ky 0·05 g)
3
3
2 2
Quantiles of standard normal

1 1
0 0
1 1
2 2
3
3
4 2 0 2 3 2 1 0 1 2
Standardised residuals Standardised residuals
2
2

1 1
0 0
1 1
2
2
0·4 0·2 0 0·2 0·4 0·5 0 0·5

Random effects Random effects
(a) (b)
Fig. 5. Normal Q–Q plots of intra-event errors (top) and random-effects (bottom) for (a) CENA – rock and (b) CENA
– soil; ky 0 .05g
of equations (3) and (4). The permanent relative displace- given that the combination of Amax and Vmax has been used
ments predicted for CENA and WNA motions using equa- to quantify the characteristic period of the ground motion,
tions (3) and (4), in conjunction with the coefficients listed and Vmax has been shown to be better correlated to the
in Tables 2–5, are shown in Fig. 7. As may be observed energy of the motions (e.g. Green & Cameron, 2003). How-
from this figure and as expected, for a given site condition ever, as noted in Green et al. (2011), quantifying the
and tectonic setting the permanent relative displacements frequency content of earthquake ground motions by a single
decrease with increasing ky /Amax (i.e. increasing ky for a characteristic period is very approximate. As a result, using
given Amax) but increase with increasing Vmax. In comparing a functional form for the regression equation that includes
the permanent relative displacements for CENA and WNA both Amax and Vmax is not able to account completely for the
for a given site condition, WNA motions have greater D influence in the variation of frequencies in WNA as opposed
than CENA motions, especially for rock sites. In comparison to CENA motions or rock as opposed to soil motions.
between rock and soil sites, larger D values are estimated For ky values other than those listed in Tables 2–5,
for soil sites than for rock sites. Finally, the curves in Fig. 7 logarithmic interpolation of D computed using the bounding
show the general trend that longer period motions produce ky values listed in Tables 2–5 is recommended.
larger permanent relative displacements (D) than shorter
period motions (e.g. WNA as opposed to CENA motions
and soil as opposed to rock motions). This trend makes COMPARISON WITH EXISTING RELATIONSHIPS
sense in that for the same peak acceleration, long-period One of the earliest ground motion predictive relations for
accelerations result in a larger displacement than short- permanent relative displacements for evaluating seismic
period accelerations. To better capture this trend in their slope stability was proposed by Newmark (1965). Subse-
predictive relation, Saygili & Rathje (2008) correlated the D quent relations have been proposed by Franklin & Chang
with Amax and Tm (mean period of the ground motion). (1977), Richards & Elms (1979), Jibson (2007), Saygili &
However, as shown in Saygili & Rathje (2008) the inclusion Rathje (2008) and NCHRP (2009), to name a few, where all
of Tm resulted in larger standard deviations than similar of these relations were developed using displacements com-
models that correlated D with Amax and Vmax (this is puted using Newmark’s sliding block analysis. However, as
independent of the additional uncertainty associated with noted above, the distinguishing characteristics of the rela-
estimating Tm versus Vmax). This is not altogether surprising tions proposed herein are that they are for stable continental
58
Rock: Vmax 25 cm/s
1·0 (14%). However, using screening criteria, Jibson (2007)
compiled a subset of 875 motions that were evenly distrib-
uted across regressed seismic demand parameters.
0·8 Jibson (2007) computed the permanent relative displace-
ments for this subset of motions using the Newmark sliding
P[D > 0|Amax, Vmax, ky
block analysis for ky equal to 0 .05, 0 .1, 0 .2, 0 .3 and 0 .4g.

0·6 However, no information is given with regard to whether the
polarity of the motions was considered (i.e. whether or not
Amax was scaled by factors of 1 and �1 prior to computing
0·4
D) or what regression technique was used, although it is
assumed that the linear least-squares technique was used.
0·2 Jibson (2007) used several different functional forms for his
regression equation having varying numbers and combina-
tions of parameters (i.e. Mw, ky and Ia, where Ia is Arias
0 intensity), with three of the five proposed models having
102 101 100 terms that are functions of ky /Amax and neither of the
Amax: g
remaining two models having terms with ky and Amax con-
Soil: Vmax 25 cm/s sidered independently.
1·0 Of the relationships proposed by Jibson (2007), only one
did not include Mw or Ia; this model is only a function of
ky /Amax. It is this relation that is used for comparison in Figs
0·8
8 and 9. As may be observed in these figures, the Jibson
P[D > 0|Amax, Vmax, ky
(2007) relation tends to predict larger displacements at lower

0·6 values of ky /Amax and smaller displacements at higher values
of ky /Amax, relative to the relations proposed herein for
WNA – rock (Fig. 8) and WNA – soil (Fig. 9). However, the
0·4 ky /Amax values corresponding to the crossover points for
larger and smaller predicted displacements varied with Amax,
Vmax and site classification. Also, the largest differences in
0·2 the predicted displacements tended to be at lower values of
ky /Amax.
0
102 101 100
Amax: g Saygili & Rathje (2008)
Saygili & Rathje (2008) developed their predictive relation
WNA: ky 0·05 using next generation attenuation (NGA) ground motion
WNA: ky 0·10 database for active shallow crustal tectonic regimes. The
magnitude and distance ranges for the motions in the NGA
WNA: ky 0·20
database are Mw5 .0 to Mw7 .9 and R ¼ 0 .1 to 100 km, respec-
CENA: ky 0·05 tively. Saygili & Rathje (2008) scaled the Amax of the motions
CENA: ky 0·10 by factors of 1, 2 and 3, but excluded scaled motions having
CENA: ky 0·20 Amax . 1g (no explicit scaling was simultaneously applied to
Vmax, as was done by Newmark (1965) and Franklin &
Chang (1977)). They computed the permanent relative
Fig. 6. Probability of permanent relative displacement (D) great-
displacements for scaled motions using the Newmark sliding
er than zero as a weighting function for WNA and CENA for rock
and soil sites; Vmax 25 cm/s block analysis for ky equal to 0 .05, 0 .1, 0 .2 and 0 .3g and
performed a linear least-squares regression analysis on the
displacements. They used several different functional forms
regimes, NLME regression analysis was used to account for for their regression equation having varying numbers and
biases in the ground motion databases, and that the relation- combinations of ground motion parameters (i.e. Amax, Vmax,
ships were derived by performing separate regression ana- Tm and Ia, where Ia is Arias intensity), but all of the forms
lyses for each value of ky. This latter characteristic is based included several terms that are functions of ky /Amax. Of the
on the observation that the influence of N cannot be relationships having two ground motion parameters, the one
completely normalised by Amax as assumed by Newmark expressing displacements as a function of Amax and Vmax had
(1965) and subsequent researchers, as shown in Fig. 3. the lowest standard deviation. Accordingly, it is this relation
However, given the prevalence of existing relationships, Figs by Saygili & Rathje (2008) that is used for comparison in
8–10 compare the relations proposed herein only with Figs 8 and 9.
selected relations from the literature, namely Jibson (2007), As may be observed in Figs 8 and 9, the Saygili & Rathje
Saygili & Rathje (2008) and NCHRP (2009). (2008) relation varies similarly with Amax, Vmax, and ky /Amax
to the relations proposed herein for WNA – rock (Fig. 8) and
WNA – soil (Fig. 9). However, from Fig. 8 it can be seen
Jibson (2007) that there is a general tendency for the Saygili & Rathje
Jibson (2007) developed his predictive relation starting (2008) relation to predict larger displacements at intermedi-
with 2270 single-component horizontal motions recorded ate values of ky /Amax (i.e. ,0 .08 to ,0 .4) and to predict
during 30 earthquakes, having a range in magnitude from smaller displacements at lower and higher values of ky /Amax
Mw5 .3 to Mw7 .6. The latest motions in the database were for rock sites. For soil sites in WNA (Fig. 9), the predicted
from the 2004, Mw6 .6 Niigata-Ken-Chuetsu, Japan earth- displacements by Saygili & Rathje (2008) compare well to
quake. The site classes of the recording stations were hard the relation proposed herein for Amax ¼ 0 .25g and
rock (10%), soft rock (27%), stiff soil (49%) and soft soil Vmax ¼ 25 cm/s over the entire range of ky /Amax considered.
59
LEE AND GREEN
Vmax 25 cm/s Vmax 50 cm/s
103 103
Amax 0·25g: WNA – rock
Amax 0·25g: CENA – rock
Amax 0·25g: WNA – soil
Amax 0·25g: CENA – soil
2
10 102
D: cm
D: cm
101 101
100 100
102 101 100 102 101 100
ky/Amax ky/Amax
(a)
Vmax 25 cm/s Vmax 50 cm/s
10 3
103
Amax 0·5g: WNA – rock
Amax 0·5g: CENA – rock
Amax 0·5g: WNA – soil
Amax 0·5g: CENA – soil
10 2
102
D: cm
D: cm
101 101
100 2 100 2
10 101 100 10 101 100
ky/Amax ky/Amax
(b)
Fig. 7. Permanent relative displacements (D) predicted for WNA and CENA for rock and soil sites, with Vmax 25 cm/s (left) and
50 cm/s (right): (a) Amax 0 .25g; (b) Amax 0 .50g
However, for Amax ¼ 0 .25g and Vmax ¼ 50 cm/s and for squares technique was used). Also, the form of the NCHRP
Amax ¼ 0 .5g and Vmax ¼ 25 cm/s, the Saygili & Rathje (2008) (2009) proposed relations have multiple terms that are func-
relation tends to predict smaller and larger displacements, tions of ky /Amax.
respectively, than the relation proposed herein. Finally, for As may be observed in Figs 8–10, the NCHRP (2009)
Amax ¼ 0 .5g and Vmax ¼ 50 cm/s, the Saygili & Rathje (2008) relations vary similarly with Amax, Vmax and ky /Amax to the
relation matches very closely to the displacements predicted relations proposed herein for WNA – rock (Fig. 8), WNA – soil
by the relation proposed herein for intermediate ranges of (Fig. 9), CENA – rock (Fig. 10), and CENA – soil (Fig. 10).
ky /Amax (i.e. ,0 .13 to ,0 .4), but tends to predict smaller From Fig. 8 it can be seen that the NCHRP (2009) for WNA
displacements for smaller and larger values of ky /Amax. rock sites and the relation proposed herein match very closely
for Amax ¼ 0 .25g and Vmax ¼ 25 cm/s and for Amax ¼ 0 .5g and
Vmax ¼ 50 cm/s over the entire range of ky /Amax considered.
NCHRP (2009) However, for Amax ¼ 0 .5g and Vmax ¼ 25 cm/s and for
The NCHRP (2009) relations for permanent relative Amax ¼ 0 .25g and Vmax ¼ 50 cm/s, the NCHRP (2009) WNA –
displacement used the same ground motion database as was rock relation tends to predict smaller and larger displacements,
used in the present study (i.e. McGuire et al., 2001). As a respectively, than the relation proposed herein. As shown in
result, they developed relations for WNA – rock, WNA – soil, Fig. 9, the trends identified for WNA rock sites between the
CENA – rock and CENA – soil similar to those proposed displacements predicted by NCHRP (2009) and the relations
herein. However, as with Jibson (2007), no information is proposed herein hold for WNA soil sites, except the match
given with regard to whether the polarity of the motions was between the predicted displacements for Amax ¼ 0 .5g and
considered (i.e. whether or not Amax was scaled by factors of Vmax ¼ 50 cm/s for WNA soil sites are not as good as they are
1 and �1 prior to computing D) and what regression tech- for WNA rock sites.
nique was used (although it is assumed that the linear least- A comparison of the predicted permanent relative displace-
60
Vmax 25 cm/s; Amax 0·25g Vmax 50 cm/s; Amax 0·25g
103 103
This study
NCHRP (2009)
Jibson (2007)
Saygili & Rathje (2008)
102 102
D: cm
D: cm
101 101
100 100
102 101 100 102 101 100
ky/Amax ky/Amax
(a)

103 103
This study
NCHRP (2009)
Jibson (2007)
102 102
D: cm
D: cm
101 101
100 100
102 101 100 102 101 100
ky/Amax ky/Amax
(b)
Fig. 8. Comparison of this study’s model and the existing relationships (Jibson, 2007; Saygili & Rathje, 2008; NCHRP, 2009) for
WNA – rock: (a) Amax 0 .25g; (b) Amax 0 .50g
ments by the NCHRP (2009) and the relations proposed COMPARISON WITH THE 2011 MINERAL, VIRGINIA
herein for CENA rock and soil sites are shown in Fig. 10. EARTHQUAKE MOTIONS
(Note that of the existing predictive relations presented here- The Mw5 .8 Mineral, Virginia earthquake of 23 August
in, only the NCHRP (2009) is shown in Fig. 10 because it is 2011 occurred in the Piedmont region of Virginia (central
the only one that applies to CENA.) As shown in Fig. 10, for Virginia seismic zone). The epicentre was located about
Amax ¼ 0 .25g and 0 .5g and for Vmax ¼ 25 cm/s and 50 cm/s, 61 km northwest of Richmond and 8 km south-southwest of
the NCHRP (2009) relations tend to predict similar displace- the town of Mineral. Its fault mechanism was a reverse slip
ments to those predicted by the relations proposed herein at fault on a north to northeast striking plane dipping to the
smaller values of ky /Amax (i.e. ky /Amax , ,0 .05) and to southeast, and the estimated focal depth was 8 .0 km (Chap-
predict smaller displacements than the relationships proposed man, 2013). The fault rupture comprised three slip events
herein at larger values of ky /Amax for CENA rock sites. On that occurred over about 1 .57 s, with the second slip episode
the contrary, for Amax ¼ 0 .25g and for Vmax ¼ 25 cm/s and releasing approximately 60% of the total moment (Chapman,
50 cm/s, the NCHRP (2009) relation for CENA soil sites 2013).
tends to predict larger displacements than the relation pro- The permanent relative displacements (D) computed from
posed herein for ky /Amax values less than about 0 .4, while the the recorded motions from the mainshock of the 2011
predicted displacements are similar at larger ky /Amax values. Mineral earthquake are compared with those predicted by
Finally, for Amax ¼ 0 .5g and for Vmax ¼ 25 cm/s and 50 cm/s, the proposed CENA model. The ground motion records were
the NCHRP (2009) for CENA soil sites matches very closely obtained from the Center for Engineering Strong Motion
the displacements predicted by the relation proposed herein Data (www.strongmotioncenter.org), except for those re-
for the range of ky /Amax considered. corded at North Anna nuclear power plant, which were
61
LEE AND GREEN
103 103
This study
NCHRP (2009)
Jibson (2007)
102 102
D: cm
D: cm
101 101
100 100
102 101 100 102 101 100
ky/Amax ky/Amax
(a)
3
10 103
This study
NCHRP (2009)
Jibson (2007)
102 102
D: cm
D: cm
101 101
100 100
102 101 100 102 101 100
ky/Amax ky/Amax
(b)
Fig. 9. Comparison of this study’s model and the existing relationships (Jibson, 2007; Saygili & Rathje, 2008; NCHRP, 2009) for WNA –
soil: (a) Amax 0 .25g; (b) Amax 0 .50g
obtained directly from Dominion Power. Table 6 lists the data from the Mineral earthquake shows that the displace-
2011 Mineral earthquake motions used herein and their ments generally increase with decreasing distance and in-
recording stations. Also, Fig. 11 shows the locations of the creasing Amax and Vmax, as expected. In comparison with the
epicentre and the recording stations within 200 km from the proposed CENA – rock model shown in Fig. 12, the perma-
epicentre, where 200 km is approximately the maximum nent relative displacements computed from the Mineral,
applicable distance of the proposed relationships. Virginia earthquake motions are in overall good agreement
No information about the site conditions at the recording with those predicted by the proposed model, except for those
stations was yet available at the time of writing of this at CW026 station. For this station, the proposed model
paper; thus, for comparison purposes, it was assumed that predicts greater displacements than those computed from the
the sites were ‘rock’ (i.e. site class A or B in Table 1). Fig. recorded earthquake motions. There are many factors that
12 shows the permanent relative displacement comparisons may have resulted in the CW026 motions being outliers (e.g.
of both horizontal components of motion recorded at each the recorded motions at this site showed pronounced near-
station during the Mineral earthquake with those predicted fault directivity velocity pulses) and further study is required
by the proposed CENA – rock model, along with the range to examine this in more detail.
of +/ one standard deviation of the proposed model. Note
that only the recorded motions with D greater than 1 .0 cm
are considered in this comparison. It is also noted that the CONCLUSIONS
2011 Mineral earthquake motions were not used to compute Empirical predictive relationships for permanent relative
displacements that were included in the regression analyses displacement for use in assessing the seismic stability of
to develop the proposed model herein. slopes, earthen dams and/or embankments subjected to stable
The overall trend of the permanent relative displacement continental motions have been developed. The predictive
62
CENA – rock: Amax 0·25g CENA – soil: Amax 0·25g
103 103
Vmax 25 cm/s; Present study Vmax 25 cm/s; Present study
Vmax 25 cm/s; NCHRP (2009) Vmax 25 cm/s; NCHRP (2009)
102 102
D: cm
D: cm
101 101
100 100
102 101 100 102 101 100
ky/Amax ky/Amax
(a)
CENA – rock: Amax 0·5g CENA soil: Amax 0·5g

103 103
102 102
D: cm
D: cm
101 101
100 100
102 101 100 102 101 100
ky/Amax ky/Amax
(b)
Fig. 10. Comparison of this study’s model and the NCHRP (2009) model for CENA – rock (left) and soil (right) sites: (a) Amax 0 .25g;
(b) Amax 0 .50g
Table 6. 2011 Mineral earthquake records (CESMD, 2012)
No. Station Code/ID Network Repic: km PGA: g PGV: cm/s
1 North Anna NPP CW026 – 18 .7 0 .11; 0 .26 4 .0; 14 .0

2 VA, Corbin (Fredricksberg Obs.) CBN NEIC 58 .2 0 .14; 0 .08 7 .1; 5 .1
3 VA, Charlottesville CVVA NMSN 53 .5 0 .10; 0 .12 1 .1; 1 .7
4 VA, Reston Fire Station 25 2555 USGS 121 .6 0 .09; 0 .04 3 .0; 1 .2
5 PA, Philadelphia – Drexel University 2648 USGS 326 .1 , 0 .01 , 1 .0
6 NY, Albany – VA Med 2653 USGS 629 .9 , 0 .01 , 1 .0
7 VT White River Junction VAMC 2655 USGS 787 .9 , 0 .01 , 1 .0
8 VA Pearisburg – Giles County CH 2549 USGS 254 .5 , 0 .01 , 1 .0
9 SC Columbia – VA Hospital 2554 USGS 519 .3 , 0 .01 , 1 .0
10 SC Charleston – Cha Pla Hotel 2544 USGS 603 .1 , 0 .01 , 1 .0
11 SC Summerville – Fire Station 2552 USGS 584 .5 , 0 .01 , 1 .0
12 NY Buffalo – VA Medical Center 2654 USGS 557 .3 , 0 .01 , 1 .0
13 MA Bedford – VA Hospital 2602 USGS 759 .6 , 0 .01 , 1 .0
14 Manchester – VA Medical Center 2652 USGS 787 .0 , 0 .01 , 1 .0
15 MA Boston – Jamaica Plains 2649 USGS 758 .0 , 0 .01 , 1 .0
63
LEE AND GREEN
79°00W 78°00W 77°00W
39°00N
N 2555 station 39°00N
CBN station
CW026 station
CVVA station
38°00N
Epicentre 38°00N
0 5 10 20 30 40
km
79°00W 78°00W 77°00W
Fig. 11. Locations of the 2011 Mineral earthquake epicentre and the recording stations within 200 km from the epicentre (CESMD,
2012)
relationships proposed herein differs from existing relation- sions or recommendations expressed in this material are
ships in that the NLME regression technique was used and those of the authors and do not necessarily reflect the views
separate regression analyses were performed for each ky. The of the National Science Foundation. The authors sincerely
resulting relationships have simple functional forms and thank Francisco Ciruela-Ochoa of Arup for generating the
correlate permanent relative displacement to Amax and Vmax. GIS map.
The predicted median permanent relative displacements de-
crease with increasing ky /Amax but increase with increasing
Vmax. In comparing predicted displacements for WNA and
CENA, WNA motions produce larger permanent relative NOTATION
Amax maximum ground acceleration (g)
displacements than CENA motions. In comparing rock and
C1 regression coefficient
soil sites, soil motions have larger displacements than rock C2 regression coefficient
motions. Lastly, the permanent relative displacements pre- C3 regression coefficient
dicted by the proposed model are shown to be in good D permanent relative displacement (cm)
agreement (mostly in +/ one standard deviation) with the ky yield acceleration (g)
displacements computed from motions recorded during the Mw moment magnitude
Mineral, Virginia earthquake of 23 August 2011, giving Q0, n regional-dependent parameters for the frequency-dependent
credence to the validity of the proposed model. quality factor Q( f )
R site-to-source distance
Repic epicentral distance
Vmax maximum ground velocity (cm/s)
ACKNOWLEDGEMENTS VS shear wave velocity
This material is based upon work supported in part by the 0 shear wave velocity of the crust at the source
National Science Foundation under grant numbers CMMI 1 regression coefficient
1030564 and CMMI 1435494. All support is gratefully 2 regression coefficient
acknowledged. However, any opinions, findings and conclu- 3 regression coefficient
64
1000 1000
CW026 CH1 CW026 CH1 CW026 CH3 CW026 CH3
μ (cm) μ (cm)
μ 1σ (cm) μ 1σ (cm)
100 100
D: cm
D: cm
10 10
1 1
0 0·05 0·10 0·15 0 0·05 0·10 0·15
ky ky
1000 1000
CBN – H1 CBN – H1 CBN – H2 CBN – H2
μ (cm) μ (cm)
100 100
D: cm
D: cm
10 10
1 1
0 0·05 0·10 0·15 0 0·05 0·10 0·15
ky ky
1000 1000
CVVN – H1 CVVA – H1 CVVN – H2 CVVA – H2
μ (cm) μ (cm)
100 100
D: cm
D: cm
10 10
1 1
0 0·05 0·10 0·15 0 0·05 0·10 0·15
ky ky
Fig. 12. Permanent relative displacements (D) of the recorded motions in two horizontal components (left and right) at
the three stations (CW026, CBN and CVVA) during the 2011 Mineral earthquake compared to the present study’s model
for CENA – rock; only the records with D greater than 1 cm are shown
˜t time interval (s) Table 7. Point source parameters for WNA and CENA motions
˜ stress drop at source (bars) (McGuire et al., 2001)
k parameter representing damping in shallow crust directly
below site (s) WNA CENA
r0 crustal density in the source region (g/cm3)
standard deviation of intra-event error ˜: bars 65 120
total the standard deviation of total error k: s 0 .040 0 .006
standard deviation of inter-event error Q0 220 351
n 0 .60 0 .84
0: km/s 3 .50 3 .52
r0: g/cm3 2 .70 2 .60
APPENDIX
Scaling procedure for CENA motions
The scaling procedure used by McGuire et al. (2001) consists of
the following computation processes the stress drop at the source; Q0 and n are regional-dependent
(a) determination of response spectral transfer function parameters for the frequency-dependent quality factor, Q( f ); r0 is
(b) computation of response spectrum for a given ground motion crustal density in the source region; and 0 is shear wave velocity of
(c) determination of target response spectrum the crust at the source. Next, random vibration theory (RVT) was
(d ) spectral matching of the time history. used to generate response spectra from the FAS (e.g. Boore, 1983;
Boore & Joyner, 1984; Silva & Lee, 1987). The ratio of these two
A response spectral transfer function was obtained by first using the response spectra is the spectral transfer function. The response
single-corner frequency point source model (Brune, 1970, 1971) to spectral transfer functions were generated for each site condition;
compute smoothed Fourier amplitude spectra (FAS) for both the horizontal/vertical components; earthquake magnitudes of 5 .5, 6 .5
CENA and WNA. The values of the point source model parameters and 7 .5 (i.e. centre value of magnitude bins); and distances of 1, 5,
used are listed in Table 7, where k is a parameter that represents 30, 75 and 130 km. In total, 60 different transfer functions were
damping in the shallow crust directly below the site; ˜ represents therefore developed. Example transfer functions for M6 .5 cases are
65
LEE AND GREEN
10 Horizontal, Mw6·5, rock 10 Horizontal, Mw6·5, soil
Amplification
Amplification
1 1
R 130 km R 130 km
R 75 km R 75 km
R 30 km R 30 km
R 5 km R 5 km
R 1 km R 1 km
0·1 0·1
0·1 1 10 100 0·1 1 10 100
Frequency: Hz Frequency: Hz
10 Vertical, Mw6·5, rock 10 Vertical, Mw6·5, soil
Amplification
Amplification
1 1
R 130 km R 130 km
R 75 km R 75 km
R 30 km R 30 km
R 5 km R 5 km
R 1 km R 1 km
0·1 0·1
0·1 1 10 100 0·1 1 10 100
Frequency: Hz Frequency: Hz
Fig. 13. Response spectral transfer functions for M6 .5, rock and soil sites, horizontal and vertical components, and each of
the distance cases (from McGuire et al., 2001)
shown in Fig. 13. The response spectrum (5% damping) of a WNA Chapman, M. C. (2013). On the rupture process of the 23 August
‘seed’ acceleration–time history is then computed. Next, the CENA 2011 Virginia Earthquake. Bull. Seismol. Soc. Am. 103, No. 2A,
target response spectrum is obtained by multiplying the ‘seed’ 613–628.
motion’s response spectrum by the appropriate response spectral Franklin, A. G. & Chang, F. K. (1977). Earthquake resistance of
transfer function. Lastly, the ‘seed’ acceleration time history is earth and rock-fill dams. In Report 5: Permanent displacements
scaled to match the target CENA response spectrum (Silva & Lee, of earth embankments by Newmark sliding block analysis, Mis-
1987). In the spectral matching process, a sample time interval ˜t of cellaneous Paper S-71-17. Vicksburg, MS: Soils and Pavements
0 .005 s (the corresponding Nyquist frequency is 100 Hz) was used to Laboratory, U. S. Army Engineer Waterways Experiment Sta-
avoid aliasing effects in the frequency range of interest. tion.
Green, R. A. & Cameron, W. I. (2003). The influence of ground
motion characteristics on site response coefficients. Proceedings
of the 7th Pacific Conference on earthquake engineering, Uni-
REFERENCES versity of Canterbury, Christchurch, New Zealand, paper number
Ambraseys, N. N. & Menu, J. M. (1988). Earthquake-induced 90. Wellington, New Zealand: New Zealand Society of Earth-
ground displacements. Earthquake Engng Structl Dynam. 16, quake Engineering.
No. 7, 985–1006. Green, R. A., Lee, J., Cameron, W. & Arenas, A. (2011). Evaluation
Ambraseys, N. N. & Srbulov, M. (1994). Attenuation of earthquake- of various definitions of characteristic period of earthquake
induced ground displacements. Earthquake Engng Structl Dy- ground motions for site response analyses. Proceedings of the
nam. 23, No. 5, 467–487. 5th international conference on earthquake geotechnical engi-
Boore, D. M. (1983). Stochastic simulation of high-frequency neering, Santiago, Chile. London, UK: International Society of
ground motions based on seismological models of the radiated Soil Mechanics and Geotechnical Engineering.
spectra. Bull. Seismol. Soc. Am. 73, No. 6A, 1865–1894. Hanks, T. C. & McGuire, R. K. (1981). The character of high-
Boore, D. M. (1986). Short-period P-and S-wave radiation from frequency strong ground motion. Bull. Seismol. Soc. Am. 71,
large earthquakes: implications for spectral scaling relations. No. 6, 2071–2095.
Bull. Seismol. Soc. Am. 76, No. 1, 43–64. Jibson, R. W. (2007). Regression models for estimating coseismic
Boore, D. M. & Joyner, W. B. (1984). A note on the use of random landslide displacement. Engng Geol. 91, No. 2–4, 209–218.
vibration theory to predict peak amplitudes of transient signals. Lee, J. (2009). Engineering characterization of earthquake ground
Bull. Seismol. Soc. Am. 74, No. 5, 2035–2039. motions. PhD thesis, University of Michigan, Ann Arbor, MI,
Bray, J. D. & Travasarou, T. (2007). Simplified procedure for USA.
estimating earthquake-induced deviatoric slope displacements. J. McGuire, R. K., Becker, A. M. & Donovan, N. C. (1984). Spectral
Geotech. Geoenviron. Engng 133, No. 4, 381–392. estimates of seismic shear waves. Bull. Seismol. Soc. Am 74,
Brune, J. N. (1970). Tectonic stress and spectra of seismic shear No. 4, 1427–1440.
waves from earthquakes. J. Geophys. Res. 75, No. 26, 611–614. McGuire, R. K., Silva, W. J. & Costantino, C. J. (2001). Technical
Brune, J. N. (1971). Correction. J. Geophys. Res. 76, No. 20, 1441– basis for revision of regulatory guidance on design ground
1450. motions: Hazard-and risk-consistent ground motion spectra
CESMD (2012) Center for Engineering Strong Motion Data. See guidelines. Washington, DC, USA: US Nuclear Regulatory
http://strongmotioncenter.org (accessed 04/02/2012). Commission.
66
NCHRP (2009). National Cooperative Highway Research Program Richards, R. & Elms, D. G. (1979). Seismic behavior of gravity
(NCHRP) Report 611: Seismic analysis and design of retaining retaining walls. J. Geotech. Engng Div., ASCE 105, No. GT4,
walls, buried structures, slopes, and embankments. Washington, 449–464.
DC, USA: The National Academies Press. Saygili, G. & Rathje, E. M. (2008). Empirical predictive models for
Newmark, N. M. (1965). Effects of earthquakes on dams and earthquake-induced sliding displacements of slopes. J. Geotech.
embankments. Géotechnique 15, No. 2, 139–160, http:// Geoenviron. Engng 134, No. 6, 790–803.
dx.doi.org/10.1680/geot.1965.15.2.139. Schneider, J. F., Silva, W. J. & Stark, C. (1993). Ground motion model
Pinheiro, J. C. & Bates, D. M. (2000). Mixed-effects models in S for the 1989 M 6 .9 Loma Prieta earthquake including effects of
and S-PLUS. New York, NY, USA: Springer. source, path, and site. Earthquake Spectra 9, No. 2, 251–287.
R Foundation (2013). Program-R (version 3.0.2) A language and Silva, W. J. (1993). Factors controlling strong ground motion and
environment for statistical computing and graphics. Vienna, their associated uncertainties. In Dynamic analysis and design
Austria: R Foundation. See http://www.r-project.org/ (accessed considerations for high-level nuclear waste repositories (ed. Q.
07/10/2013). A. Hossain), pp. 132–161. New York, NY, USA: American
Rathje, E. M. & Antonakos, G. (2011). A unified model for Society of Civil Engineers.
predicting earthquake-induced sliding displacement of rigid and Silva, W. J. & Lee, K. (1987). WES RASCAL Code for Synthesizing
flexible slopes. Engng Geol. 122, No. 1–2, 51–60. Earthquake Ground Motions: State-of-the-art for assessing
Rathje, E. M. & Saygili, G. (2009). Probabilistic assessment of earthquake hazards in the United States, Report 24. Vicksburg,
earthquake-induced sliding displacements of natural slopes. Bull. MS, USA: US Army Engineering Waterways Experiment Sta-
New Zealand Soc. Earthquake Engng 42, No. 1, 18–27. tion.
67
Tsinidis, G. et al. (2015). Géotechnique 65, No. 5, 401–417 [http://dx.doi.org/10.1680/geot.SIP.15.P.004]
Dynamic response of flexible square tunnels: centrifuge testing and

validation of existing design methodologies
G . T S I N I D I S , K . P I T I L A K I S , G . M A DA B H U S H I † a n d C . H E RO N †
A series of dynamic centrifuge tests were performed on a flexible aluminium square tunnel model
embedded in Hostun dry sand. The tests were carried out at the centrifuge facility of the University of
Cambridge in order to further improve knowledge regarding the seismic response of rectangular
embedded structures and to calibrate currently available design methods. The soil–tunnel system
response was recorded with an extensive instrumentation array, comprising miniature accelerometers,
pressure cells and position sensors in addition to strain gauges, which recorded the tunnel lining
internal forces. Tests were numerically analysed by means of full dynamic time history analysis of the
coupled soil–tunnel system. Numerical predictions were compared to the experimental data to validate
the effectiveness of the numerical modelling. The interpretation of both experimental and numerical
results revealed, among other findings: (a) a rocking response of the model tunnel in addition to
racking; (b) residual earth pressures on the tunnel side walls; and (c) residual internal forces after
shaking, which are amplified with the tunnel’s flexibility. Finally, the calibrated numerical models were
used to validate the accuracy of simplified design methods used in engineering practice.
KEYWORDS: centrifuge modelling; earthquakes; numerical modelling; soil/structure interaction; tunnels &
tunnelling
INTRODUCTION issues that significantly affect the seismic response (Pitilakis &
Recent earthquake events have demonstrated that underground Tsinidis, 2014). Seismic earth pressures and shear stresses
structures in soft soils may undergo extensive damage or even distributions along the perimeter of the embedded structure
collapse (Dowding & Rozen, 1978; Sharma & Judd, 1991; Iida and complex deformation modes during shaking for rectangu-
et al., 1996; Kawashima, 2000; Wang et al., 2001; Kontoe et lar cross–sections (e.g. rocking and inward deformations) are,
al., 2008). These failures increased interest in further investi- among other issues, still not entirely understood.
gation of the seismic response of these types of structures. The knowledge shortfall motivated a range of experimen-
Generally, the seismic response of embedded structures is quite tal (e.g. Chou et al., 2010; Shibayama et al., 2010; Chian &
distinct from that of above-ground structures, as the kinematic Madabhushi, 2012; Cilingir & Madabhushi, 2011a, 2011b,
loading induced by the surrounding soil prevails over inertial 2011c; Lanzano et al., 2012; Chen et al., 2013), numerical
loads stemming from the oscillation of the structure itself (e.g. Anastasopoulos et al., 2007, 2008; Amorosi & Boldini,
(Kawashima, 2000). In addition, large embedded structures are 2009; Anastasopoulos & Gazetas, 2010; Kontoe et al., 2011;
commonly stiff structures to withstand static loads. Hence, Lanzano et al., 2014) and analytical (e.g. Huo et al., 2006;
during earthquake shaking, strong interaction effects are mobi- Bobet et al., 2008; Bobet, 2010) research studies over recent
lised between the structure and the surrounding soil, especially years, investigating the effects of seismic shaking and earth-
for structures of rectangular cross-section. These interaction quake-induced ground failures (e.g. liquefaction) on the
effects are mainly affected by two crucial parameters, namely: response of embedded structures. In some cases, the effi-
(a) the soil to structure relative flexibility and (b) the soil– ciency of different design methods has been investigated by
structure interface characteristics. In general, both are chang- comparing the outcomes of the methods (e.g. tunnel distor-
ing with the amplitude of seismic excitation, as they depend on tions or dynamic internal forces) with each other (e.g.
the soil shear modulus and strength, which are related to the Hashash et al., 2005, 2010; Kontoe et al., 2014).
ground strains and the non-linear behaviour of the soil. This study presents a series of dynamic centrifuge tests that
Several methods are available in the literature for the evalua- were performed on a flexible aluminium square tunnel model
tion of the response of underground structures and tunnels embedded in dry sand. The soil–tunnel system response was
under seismic shaking (e.g. St John & Zahrah, 1987; Wang, recorded with an extensive instrumentation array comprising
1993; Penzien, 2000; AFPS/AFTES, 2001; Hashash et al., miniature accelerometers, pressure cells and position sensors,
2001; ISO, 2005; Anderson et al., 2008; FHWA, 2009). The in addition to strain gauges, which recorded the tunnel lining
results of these methods may deviate, even under the same internal forces. The test case is also numerically analysed by
design assumptions, especially in case of rectangular structures means of a full dynamic time history numerical analysis of
(e.g. cut and cover tunnels), owing to both inherent epistemic the coupled soil–tunnel system. Numerical predictions are
uncertainties and a knowledge shortfall regarding some crucial compared to the experimental data to validate the effec-
tiveness of the numerical modelling. The calibrated numerical
models are finally used to validate the accuracy of available
Manuscript received 31 March 2014; revised manuscript accepted 23 simplified design methods used in engineering practice.
December 2014. Published online ahead of print 20 March 2015.
see p. ii.
Department of Civil Engineering, Aristotle University, Thessalon- DYNAMIC CENTRIFUGE TESTING
iki, Greece. The test was carried out on the 10 m diameter Turner
† Schofield Centre, University of Cambridge, Cambridge, UK. beam centrifuge of the University of Cambridge (Schofield,
71
TSINIDIS, PITILAKIS, MADABHUSHI AND HERON
1980) under a centrifuge acceleration of 50g (scale factor Table 3. Model tunnel mechanical properties
n ¼ 50). A large equivalent-shear-beam (ESB) container was
used to contain the model (Zeng & Schofield, 1996). Scaling Unit weight, Young’s modulus, Poisson Tensile strength, fbk:
laws that are applied to convert the measured quantities from ªt: kN/m3 E: GPa ratio, v MPa
model to prototype scale are summarised in Table 1 (Scho-
field, 1981). 2 .7 69 .5 0 .33 220
The soil deposit was made of uniform Hostun HN31 sand
with 90% relative density. The physical and mechanical
properties of the sand are summarised in Table 2. Sand
pouring was performed in layers using an automatic hopper model tunnel, creating a rough surface. Two polytetrafluor-
system (Madabhushi et al., 2006), while the model tunnel oethylene (PTFE) rectangular plates were placed at each end
and the instruments were properly positioned during con- of the tunnel to avoid the entry of sand into the tunnel. The
struction. plates, which were marginally larger than the model tunnel,
The model tunnel, manufactured using 6063A aluminium were connected to each other by a rod which passed through
alloy, was 100 mm wide and 220 mm long, having a lining the tunnel (Fig. 1(b)).
thickness of 2 mm (Fig. 1(a)). The aluminium alloy mech- A dense instrumentation array was implemented to moni-
anical properties are summarised in Table 3. According to tor the soil–tunnel response (Fig. 2). Miniature piezoelectric
the scale factor, the model corresponds to a 5 3 5 (m) accelerometers were used to measure the acceleration in the
square tunnel having an equivalent concrete lining thickness soil, on the tunnel and on the container. The soil surface
equal to 0 .13 m (assuming Ec ¼ 30 GPa for the concrete). settlements were recorded in two locations using linear
This thickness is obviously unrealistic in practice, as the variable differential transformers (LVDTs), while two posi-
design analysis for the static loads will result in a much tion sensors (POTs) were attached to the upper edges of the
thicker lining. However, this selection was made in order to tunnel walls to capture the vertical displacement and the
study the effect of high flexibility on the tunnel response, as possible rocking of the model tunnel. Both the LVDTs and
well as to obtain clear measurements of the lining bending the POTs were attached to gantries running above the ESB
and axial strains. To simulate more realistically the soil– container. Two miniature total earth pressure cells (PCs)
tunnel interface, sand was stuck on the external face of the were attached to the left side wall of the tunnel, allowing
the measurement of the soil earth pressures on the wall.
Strain gauges were attached to the inner and outer faces of
Table 1. Centrifuge scaling laws (Schofield, 1981) the tunnel to measure the lining bending moment and axial
force at several locations (Fig. 2). Unfortunately, the bending
Parameter Model/Prototype Dimensions moment strain gauge at the middle of the roof slab (SG-B3)
malfunctioned during testing. All the instruments were prop-
Length 1/n l erly calibrated before and checked after testing. The strain
Mass 1/n3 m gauges were carefully calibrated for static loading patterns
Stress 1 ml�1t�2
using the procedure outlined in Tsinidis et al. (2014a). The
Strain 1 1
Force 1/n2 mlt�2 data were recorded at a sampling frequency of 4 Hz during
Time (dynamic) 1/n t the swing up of the centrifuge and at 4 kHz during shaking.
Frequency n t�1 A series of air hammer tests was performed to estimate
Acceleration n lt�2 the soil shear wave velocity profile (Ghosh & Madabhushi,
Velocity 1 lt�1 2002). A small air hammer was introduced close to the base
of the soil layer, while a set of accelerometers (AH, Fig. 2)
were placed above it, forming a vertical array, allowing a
record of the arrival times of the waves emanating from the
Table 2. Hostun HN31 physical and mechanical properties air hammer. To ensure that the arrival times were adequately
recorded, the accelerometers along this array were attached
rs: emax emin d10: mm d50: mm d60: mm �cv: degrees
g/cm3
to a different acquisition system that allowed for a sampling
frequency of 50 kHz.
2 .65 1 .01 0 .555 0 .209 0 .335 0 .365 33 The dynamic input was provided at the container base by
a stored angular momentum actuator, designed to apply
PTFE plate
PTFE plate
(a) (b) (c)
Fig. 1. (a) Model tunnel; (b) model tunnel placement in the equivalent shear beam container; (c) completed model in the equivalent shear
beam container
72
DYNAMIC RESPONSE OF FLEXIBLE SQUARE TUNNELS
145 45 96·5 100 141·5 145
LVDT1 LVDT2
POT1 POT2
AH5 A10 A8 A3 Strain gauges set-up
85
SG-A4
A15 A14 A16 SG-B3
AH4 A7
SG-B4
100
A6
427
AH3 PC2 A13

370
A2
AH2 PC1 A5 SG-A3
A12 SG-A1
SG-B2
110
SG-B1 SG-A2
AH1 (Dimensions in mm)
A11
60
Air hammer
A9 A4 A1
673
Accelerometer Pressure cell LVDT POT Strain gauge
Fig. 2. Model layout and instrumentation scheme (A: accelerometer; AH: accelerometer above air
hammer; LVDT: linear variable differential transformer; POT: position sensor; SG-A: axial strain
gauge; SG-B: bending moment strain gauge; PC: pressure cell)
sinusoidal or sine-sweep wavelets (Madabhushi et al., 1998). ducted in the frequency domain. In particular, acceleration–
The model was subjected to a total of eight earthquakes time histories were filtered at 10 to 400 Hz using a band
during two flights: EQ1 to EQ5 were fired during a first pass eighth-order Butterworth filter. All the other data were
flight, whereas EQ6 to EQ8 were fired during a subsequent filtered using a low-pass eighth-order Butterworth filter at
flight. Fig. 3 presents the input motion–time histories, while 400 Hz.
Table 4 tabulates their characteristics. During each flight, the
centrifuge was spun up in steps until 50g and then the
earthquakes were fired in a row, leaving some time between NUMERICAL ANALYSIS
them to acquire the data. Numerical model
To interpret the experimental results, the data were wind- The test was numerically simulated by means of full
owed, neglecting the parts of the signals before and after the dynamic time history analyses, using the finite-element code
main shake duration, while a filtering procedure was con- Abaqus (Abaqus, 2012). The analyses were performed in
EQ1 EQ2 EQ3 EQ4

0·30 0·30 0·30 0·30
0·15 0·15 0·15 0·15

A/50g
A/50g
A/50g
A/50g
0 0 0 0
0·15 0·15 0·15 0·15
0·30 0·30 0·30 0·30

0 0·2 0·4 0·6 0·8 0 0·15 0·30 0·45 0·60 0 0·15 0·30 0·45 0·60 0 0·15 0·30 0·45 0·60
t: s t: s t: s t: s
EQ5 EQ6 EQ7 EQ8

0·30 0·30 0·30 0·30
0·15 0·15 0·15 0·15

A/50g
A/50g
A/50g
A/50g
0 0 0 0
0·15 0·15 0·15 0·15
0·30 0·30 0·30 0·30

0·5 0·6 0·7 0·8 0 0·15 0·30 0·45 0·60 0 0·35 0·70 0 0·175 0·350 0·525
t: s t: s t: s t: s
Fig. 3. Input motion–time histories
Table 4. Input motions characteristics (bracketed values in prototype scale)
EQ ID EQ1 EQ2 EQ3 EQ4 EQ5 EQ6† EQ7† EQ8†
Frequency, f: Hz 30 (0 .6) 45 (0 .9) 50 (1) 50 (1) 60 (1 .2) 50 (1) 50 (1) 50 (1)

Amplitude, a: g 1 .0 (0 .02) 4 .0 (0 .08) 6 .5 (0 .13) 12 .0 (0 .24) 12 .0 (0 .24) 5 .8 (0 .116) 6 .0 (0 .12) 11 .0 (0 .22)
Sine sweep.
† Fired during a second flight.
73
prototype scale assuming plane strain conditions. Fig. 4 (Bilotta et al., 2014). Several research groups have simulated
presents the numerical model layout. the tests using different numerical codes and constitutive
The soil was meshed with quadratic plane strain elements, models of different complexity (Amorosi et al., 2014; Conti
while the tunnel was simulated with beam elements. The et al., 2014; Gomes, 2014; Hleibieh et al., 2014; Tsinidis et
element size was selected in a way that ensured efficient al., 2014b). Among the most interesting results of this
reproduction of the waveforms of the whole frequency range comparative effort is that even sophisticated constitutive
under study. models produced results that deviated considerably from the
The base boundary of the model was simulated as rigid recorded data. Part of the difference was attributed to
bedrock (shaking table), while for the side boundaries kine- calibration issues and determination of constitutive param-
matic tie constraints were introduced, forcing the opposite eters.
vertical sides to move simultaneously, simulating, in that The input motion was introduced at the model base in
simplified way, the container. terms of acceleration–time histories, referring to the motion
The soil–tunnel interface was modelled using a finite recorded by the reference accelerometer (A1, Fig. 2). The
sliding hard contact algorithm embedded in Abaqus (Aba- analyses were performed in two steps: first the gravity loads
qus, 2012). The model constrains the two media when were introduced, while in a second step the earthquake
attached, using the penalty constraint enforcement method motions were applied in a row, replicating each test flight.
and Lagrange multipliers, while it also allows for separation. To this end, the loading history for the sand was accounted
The interface friction effect on the soil–tunnel system re- for.
sponse was investigated by applying different Coulomb fric-
tion coefficients �, namely � ¼ 0 for the full slip and 0 .4
and 0 .8 for non-slip conditions. In a final series of analyses, Sand stiffness and strength
the soil and the tunnel were fully bonded, assuming no slip The sand small-strain shear modulus (Gmax) was described
conditions, precluding separation. according to Hardin & Drnevich (1972), which fits reason-
The model tunnel was modelled using an elastic–perfectly ably well with the air hammer test results and also results of
plastic material model, with yield strength equal to laboratory tests (resonant column) that were performed on
220 MPa, while the soil response under seismic shaking was the specific sand fraction (Pistolas et al., 2014). Fig. 5
simulated in two ways. In a first series of analyses, a visco- compares the estimated small-strain shear wave velocity
elastic model was implemented, introducing a degraded gradient from different methods and the distribution pro-
shear modulus distribution and viscous damping (e.g. follow- posed according to Hardin & Drnevich (1972). It is worth
ing the equivalent linear approximation method). In the noting that these results refer to the ‘free-field’ conditions
second series of analyses, a non-associated Mohr–Coulomb away from the model tunnel. The exact properties of sand in
model was used to account for the permanent deformations the area close to the tunnel are not well known. The reason
of the soil. The latter model, embedded in Abaqus, allows is that considering the model’s formation (i.e. sand pouring
for simulation of certain hardening or softening responses from a height to achieve the desired relative density of the
after yielding. Elastic properties were assumed the same soil specimen), the existence of the model tunnel may affect
with the visco-elastic analyses, following a similar procedure the density of the sand in the adjacent zone, thus affecting
as in Amorosi & Boldini (2009). This elasto-plastic model the mechanical properties of the sand at this location. How-
has been implemented by several researchers (e.g. Pakbaz & ever, it is believed that after the first shakes the soil in this
Yareevand, 2005; Hwang & Lu, 2007), while it has been particular region will have reached a reasonable degree of
recently used by Cilingir & Madabhushi (2011a, 2011b, densification comparable to the rest of the soil sample.
2011c) for the simulation of similar dynamic centrifuge tests To estimate the real sand stiffness and viscous damping
on model tunnels in dry sand, revealing reasonable compari- during shaking a trial-and-error procedure was applied. More
sons between the recorded data and the numerical results. specifically, one-dimensional (1D) equivalent linear (EQL)
The implemented models were selected as they are proposed soil response analyses of the soil deposit were performed,
in guidelines for dynamic analysis of embedded structures using different sets of G–ª–D curves for cohesionless soils
(e.g. equivalent linear approximation in FHWA (2009)) and (e.g. Seed et al., 1986; Ishibashi & Zhang, 1993; Pistolas et
are commonly used in tunnelling design practice owing to al., 2014). The analyses were performed in the frequency
their easy calibration and control. Recently, a series of domain using EERA (equivalent-linear earthquake site
dynamic centrifuge tests on a flexible circular model tunnel response analyses) (Bardet et al., 2000). The computed
embedded in dry sand (Lanzano et al., 2012) has been used
as a benchmark of a numerical round robin on tunnel tests VSO: m/s
0 90 180 270 360
0
Soil–tunnel interface Displacement constraints
5
Depth: m
10
AH flight 1
15 AH flight 2
RC results
Empirical formulation
20
a(t)
Fig. 5. Small-strain shear wave velocity profiles estimated from
air hammer tests (AH) and resonant column tests (RC) compared
Fig. 4. Numerical model in Abaqus to the Hardin & Drnevich (1972) empirical formulation
74
horizontal acceleration–time histories and amplification were where e is the void ratio, � 9 is the mean effective stress (in
compared to the recorded data of the free field array MPa), G is the degraded shear modulus (in MPa) and Æ is
(sensors A4 to A8 in Fig. 2). The adopted G–ª–D curves the reduction value for each shake, ranging between 0 .3 and
were those that resulted in the best fitting of the numerical 0 .4. For the computation of the mean effective stress the
predictions with the experimental results (Ishibashi & Zhang earth coefficient at rest (K0) was evaluated as (Jaky, 1948)
(1993) for small confining pressure). Comparisons of the
K 0 ¼ 1 � sin � (2)
adopted G–ª–D curves with empirical ones (Seed et al.,
1986) and laboratory results from resonant column and where � is the sand friction angle.
cyclic triaxial tests for the specific sand fraction (Pistolas et The reduced values for the sand shear modulus come in
al., 2014) are provided in Fig. 6. The adopted numerical agreement with the shear moduli computed from the stress–
curves compare reasonably well with the laboratory test strain loops, estimated using the recorded acceleration–time
results over a wide range of strain amplitudes. histories across the free-field array (A4–A8 in Fig. 2),
One-dimensional equivalent linear soil response analyses following Zeghal & Elgamal (1994). It is noteworthy that
for the finally selected Gmax and G–ª–D curves revealed that this high decrease of the soil stiffness and increase of
a reduced Hardin and Drnevich distribution adequately re- damping in this type of test is also reported by other
produced the degraded sand shear modulus during shaking. researchers (Kirtas et al., 2009; Pitilakis & Clouteau, 2010;
To this end, the following expression was used for the Lanzano et al., 2010, 2014; Li et al., 2013).
description of the degraded strain shear modulus In the final two-dimensional (2D) full dynamic analysis,
the degraded elastic stiffness of the sand material for each
(3 � e)2 . shake was introduced through a Fortran user subroutine,
G ¼ Æ 3 100 (� 9)0 5 (1)
1þe which correlates the stiffness with the confining pressure at
each soil element integration point. To this end, the effect of
1·00
the tunnel on the surrounding sand stiffness was explicitly
accounted for.
In both visco-elastic and visco-elasto-plastic analyses,
viscous damping was introduced in the form of the fre-
quency dependent Rayleigh type. ‘Target’ damping (15%)
0·75
was estimated through the 1D equivalent linear response
analyses, as discussed before. For the calibration of the
Rayleigh parameters, the double frequency approach was
implemented. The Rayleigh parameters were properly tuned
G/G0
0·50 for different ‘important frequencies’ (e.g. soil deposit domi-

nant frequencies or signal dominant frequencies). The finally
selected parameters were those that resulted in good com-
parisons between the computed and recorded acceleration
0·25 data. The importance of proper calibration for the Rayleigh
coefficients is discussed in Kontoe et al. (2011). In the
elasto-plastic analyses, additional energy dissipation was
introduced by the hysteretic soil response.
0 Regarding the strength parameters of the sand, a friction
104 103 102 101 100 angle � equal to 338 (critical friction angle for the specific
γ: % sand fraction) was used, while the dilatancy angle ł was
(a)
assumed equal to 38 (Schanz & Vermeer, 1996). These
40
strength parameters correspond to the specific sand fraction
and are found to give reasonable comparisons with the
Seed et al. (1986) recorded response. A slight cohesion (c ¼ 1 kPa) was intro-
Analysis duced to avoid numerical problems.
σ 25 kPa, RC
30
σ 50 kPa, RC
NUMERICAL PREDICTIONS COMPARED WITH
σ 200 kPa, RC
EXPERIMENTAL RESULTS
σ 50 kPa, TX Representative comparisons between the recorded and the
D: %
20 σ 200 kPa, TX computed response are presented in this section. Through

the presentation of relevant data several crucial aspects of
the soil–tunnel response are discussed. Results are generally
shown at model scale, if not stated otherwise.
10
Horizontal acceleration
Figure 7 presents time windows of typical comparisons
0
between the recorded and the computed acceleration–time
104 103 102 101 100 histories at two representative locations (middle section of
γ: % left side wall, A13; top receiver of tunnel accelerometer
(b) array, A10). In Fig. 8 representative comparisons between
Fig. 6. Adopted G–ª–D curves compared to resonant column test the computed and recorded horizontal acceleration amplifica-
results (RC) (Pistolas et al., 2014), cyclic triaxial test results (TX) tion along the free-field and the tunnel vertical accelerometer
(Pistolas et al., 2014) and empirical proposals (Seed et al., 1986): arrays are depicted. Generally, both visco-elastic and elasto-
(a) G/G0 plotted against shear strain; (b) damping plotted against plastic analyses reveal similar responses and amplification,
shear strain while numerical predictions are in good agreement with the
75
A10 – EQ2 A10 – EQ4 A10 – EQ5 A10 – EQ7
0·4 0·4 0·4 0·4
0·2 0·2 0·2 0·2
A/50g
A/50g
A/50g
A/50g
0 0 0 0
0·2 Test 0·2 0·2 0·2

Analysis
0·4 0·4 0·4 0·4
0·20 0·25 0·30 0·20 0·25 0·30 0·50 0·55 0·60 0·20 0·25 0·30
t: s t: s t: s t: s
(a)
A13 – EQ2 A13 – EQ4 A13 – EQ5 A13 – EQ7
0·4 0·4 0·4 0·4
0·2 0·2 0·2 0·2

A/50g
A/50g
A/50g
A/50g
0 0 0 0
0·2 0·2 0·2 0·2
0·4 0·4 0·4 0·4

0·20 0·25 0·30 0·20 0·25 0·30 0·50 0·55 0·60 0·20 0·25 0·30
t: s t: s t: s t: s
(b)
Fig. 7. Time windows of representative acceleration–time histories recorded and computed for different earthquake input
motions; experimental data compared with visco-elasto-plastic results: (a) accelerometer A10 at the soil surface above the
model tunnel; (b) accelerometer A13 on the tunnel side-wall (notation according to Fig. 2)
A50g – EQ2 A50g – EQ3 A50g – EQ5 A50g – EQ8

0 0·175 0·350 0 0·175 0·350 0 0·175 0·350 0 0·175 0·350
0 0 0 0
0·1 0·1 0·1 0·1

Depth: m
Depth: m
Depth: m
Depth: m
0·2 0·2 0·2 0·2
0·3 0·3 0·3 0·3
0·4 0·4 0·4 0·4
Test Analysis
(a)
A50g – EQ2 A50g – EQ3 A50g – EQ5 A50g – EQ8

0 0·175 0·350 0 0·175 0·350 0 0·175 0·350 0 0·175 0·350
0 0 0 0
0·1 0·1 0·1 0·1

Tunnel level
Depth: m
Depth: m
Depth: m
Depth: m
0·2 0·2 0·2 0·2
0·3 0·3 0·3 0·3
0·4 0·4 0·4 0·4
(b)
Fig. 8. Horizontal acceleration amplification along (a) the soil free-field accelerometers vertical array, and (b) the tunnel
accelerometers vertical array, for different earthquake input motions; experimental data compared with visco-elasto-plastic
results
records both in terms of amplitude and frequency content deformations, discussed in the following section, are likely
(Fig. 9). The differences, generally minor, are attributed to to have caused a malfunction of the accelerometer at this
the inevitable differences between the assumed soil mechani- location. It worth mentioning the higher frequencies of the
cal properties (stiffness and damping) and their actual values signals observed in the Fourier spectra shown in Fig. 9.
during the test, especially near the tunnel. The larger devia- Significant energy content is associated with higher frequen-
tion observed at the tunnel roof slab is attributed to an cies than with the predominant one. These higher frequen-
erroneous record at this location. Actually, the slab inward cies, which are attributed to the experimental equipment’s
76
A8 – EQ1 A8 – EQ2 A8 – EQ5 A8 – EQ8
0·2 0·6 0·8 1·0
Test
Amplification
Amplification
Amplification
Amplification
Analysis
0·1 0·3 0·4 0·5
0 0 0 0
0 3 6 0 3 6 0 3 6 0 3 6
f/50: Hz f/50: Hz f/50: Hz f/50: Hz
(a)
A10 – EQ1 A10 – EQ2 A10 – EQ5 A10 – EQ8
0·2 0·6 0·8 1·0
Amplification
Amplification
Amplification
Amplification
0·1 0·3 0·4 0·5
0 0 0 0
0 3 6 0 3 6 0 3 6 0 3 6
f/50: Hz f/50: Hz f/50: Hz f/50: Hz
(b)
Fig. 9. Fourier spectra of acceleration–time histories recorded and computed at locations of (a) accelerometer A8 and
(b) accelerometer A10 for different earthquake input motions; experimental data compared with visco-elasto-plastic results
mechanical response (Brennan et al., 2005), are described Tunnel initial shape
quite efficiently by the numerical model.
Tunnel deformed shapes

Figure 10 presents time windows of typical comparisons
between the recorded and computed vertical accelerations at
the sides of the tunnel roof slab. Experimental results are
slightly larger than the numerical predictions. The difference
is attributed to the parasitic yawing movement of the whole
model on the shaking table during shaking, which may
amplify vertical acceleration and cannot be reproduced by
the numerical analysis. The no-slip condition analysis results
are closer to the recorded response. Generally, signals are
out of phase, indicating a rocking mode of vibration for the Tunnel deformed shape
(a) (b)
tunnel, in addition to the racking mode. Fig. 11 presents
typical computed deformed shapes of the tunnel during Fig. 11. Shape of deformed tunnel for time steps of the computed
shaking, verifying this complex racking–rocking response. maximum racking distortion; EQ4 earthquake, elasto-plastic
Owing to the high flexibility of the tunnel, inward deforma- analysis for no slip conditions: (a) motion towards left; (b) motion
tions are also observed for the slabs and the walls. towards right (deformations scale 360)
Test Analysis – Full slip Analysis – No slip

0·2 0·2 0·2
0·1 0·1 0·1

A15 A16
A/50g
A/50g
A/50g
0 0 0
0·1 0·1 0·1
A15 A16
0·2 0·2 0·2
0·30 0·33 0·36 0·30 0·33 0·36 0·30 0·33 0·36
t: s t: s t: s
Fig. 10. Time windows of recorded and computed vertical acceleration–time histories at the sides of the tunnel roof slab
for EQ4 earthquake; experimental data compared with visco-elasto-plastic results
77
Dynamic earth pressures efficiency of the constitutive models, and also to recording
Typical comparisons between the computed and recorded issues that are related to the response of the miniature earth
dynamic earth pressures–time histories at the left side wall pressures cells in the case of granular dry sands. Accurate
are presented in Fig. 12. The effect of the soil–tunnel inter- measurement of earth pressures in sands with miniature
face characteristics on the computed earth pressures is also pressure cells is always difficult, as the relative stiffness of
highlighted. Residual values are presented in records after the sensing plate may affect the readings, while there are
shaking as a result of the soil yielding and densification also problems related to the grain size effect (Cilingir,
around the tunnel. This post-earthquake residual response 2009). Moreover, inward deformations of the tunnel wall
has also been reported during similar centrifuge tests (Cilin- may slightly change the recording direction (small inclina-
gir & Madabhushi, 2011a, 2011b) and is amplified with the tion of the pressure cell) and therefore the recorded earth
flexibility of the tunnel. In addition, dynamic pressure incre- pressure may be different from the ‘normal’ value computed
ments are found to be larger near the stiff corners of the by the analysis. Considering the aforementioned points, the
tunnel. Generally, numerical predictions for no-slip condi- comparisons indicate a reasonable agreement.
tions are closer to the recorded response. The comparison is Figure 13 presents typical dynamic earth pressure distribu-
more satisfactory, especially for the last shakes. Observed tions around the tunnel’s perimeter, referring to the time step
differences in amplitude can be attributed to the discrepan- of the tunnel maximum racking distortion. Soil yielding
cies between the assumed and the actual in test mechanical around the tunnel results in stress redistributions leading to a
properties of the sand and the soil–tunnel interface, the slightly different response between elasto-plastic and visco-
PC1 – EQ6 PC1 – EQ6 PC2 – EQ6 PC2 – EQ6

100 100 100 40
50 50 50 20 PC2
PC1
σ: kPa/m
σ: kPa/m
σ: kPa/m
σ: kPa/m
0 0 0 0
50 50 50 20
100 100 100 40

0 0·3 0·6 0·08 0·13 0·18 0 0·3 0·6 0·08 0·13 0·18
t: s t: s t: s t: s
(a)
PC1 – EQ8 PC1 – EQ8 PC2 – EQ8 PC2 – EQ8

160 160 160 40
80 80 80 20
σ: kPa/m
σ: kPa/m
σ: kPa/m
σ: kPa/m
0 0 0 0
80 80 80 20
160 160 160 40

0 0·3 0·6 0·06 0·11 0·16 0 0·3 0·6 0·06 0·11 0·16
t: s t: s t: s t: s
(b)
Test Analysis – No slip Analysis – Full slip
Fig. 12. Dynamic earth pressure–time histories recorded and computed by visco-elasto-plastic analyses on the left side wall for
different earthquake inputmotions; effect of the soil–tunnel interface characteristics: (a) earthquake EQ6; (b) earthquake EQ8
120 120
D C
60 A B 60
σ: kPa/m
σ: kPa/m
0 0
60 Full slip, elasto-plastic 60

No slip, elasto-plastic
No slip, elastic
120 120
A B C D A A B C D A
(a) (b)
Fig. 13. Effect of soil–tunnel interface characteristics and soil yielding response on the dynamic
earth pressures distributions computed along the perimeter of the tunnel at the time step of
maximum racking distortion: (a) earthquake EQ2; (b) earthquake EQ4
78
elastic analyses (effect on distributions). Moreover, soil– case. Fig. 17 illustrates this effect on the residual bending
tunnel interface properties seem to affect the soil yielding moments computed for different shaking scenarios. It is
response in the area adjacent to the tunnel (Fig. 14) and noteworthy that the recorded residual bending moments for
therefore the pressure distributions. This relation between EQ7 are much lower than EQ6, although both input motions
the soil yielding response and the soil–tunnel interface share the same amplitude and frequency characteristics. This
properties is also reported by Huo et al. (2005). is attributed to the fact that the largest part of soil plastic
strain that is induced by the specific input motion amplitude
(the same for both shakes) is accumulated during the first
Soil dynamic shear stresses loading circles of the first shake (EQ6). This phenomenon is
Figure 15 portrays representative soil dynamic shear stress simulated reasonably well by the implemented elasto-plastic
distributions around the tunnel computed for the time step model.
of maximum racking distortion. As for the earth pressures,
soil yielding affects the soil shear stress around the tunnel.
Generally, shear stresses tend to increase near the tunnel Lining dynamic axial force
corners due to the higher earth pressures (confining pressures Similar to the dynamic bending moments, residual values
for the tunnel) at these locations. As expected, interface were recorded for the lining axial forces (Fig. 18). Residuals
friction plays an important role on the shear stress distribu- were generally smaller than the ones of the bending mo-
tion and magnitude. An increase of the soil–tunnel interface ment, but were larger along the slabs. In addition, dynamic
friction results in an increase of the soil shear stresses along axial forces recorded on the side walls were out of phase,
the middle sections of the tunnel slabs and walls. verifying the racking–rocking response of the tunnel during
shaking (Tsinidis et al., 2014a). Numerical results revealed
similar tendencies. The effect of the mobilised friction
Lining dynamic bending moment (along the interface) on the lining axial forces is quite
Representative comparisons between recorded and com- important (Fig. 18). Similar to the dynamic earth pressures,
puted by elasto-plastic analyses dynamic bending moment– recorded axial forces were found to be in better agreement
time histories are presented in Fig. 16. Both experimental with the numerical predictions assuming no-slip conditions.
data and numerical predictions indicate a post-earthquake This observation may be attributed to the inward deforma-
residual response, similar to that of the earth pressures. This tions of the model tunnel that are amplified by the tunnel’s
residual response is highly affected by the tunnel’s flexibility. high flexibility. The surrounding sand is actually squeezing
Different assumptions for the soil–tunnel interface character- the tunnel, leading to a more rigid soil–tunnel interface (no
istics may affect the computed bending moments both in separation–no-slip conditions).
terms of residuals and dynamic increments, mainly due to Generally, both the visco-elastic and the elasto-plastic
the different soil yielding response around the tunnel in each analyses reproduce the recorded dynamic internal forces
increments (reversible component of force increments) rea-
No slip Full slip
sonably well (Fig. 19). These increments, which are com-
puted as the half of the amplitude of the maximum values
of the loading cycles in the internal forces–time histories,
are in both cases amplified near the tunnel corners (Fig. 19).
SIMPLIFIED ANALYSIS METHODS

Simplified methods are commonly used in design practice,
especially during preliminary stages of design, mainly due to
their simplicity and reduced computational cost compared to
the non-linear full dynamic analysis. The majority of these
methods rely on the assumption that the seismic load is
0·018 0·009 0
introduced on the tunnel in a quasi-static manner, and there-
Fig. 14. Soil plastic deformations computed by the visco-elasto- fore they do not account for the dynamic soil–structure
plastic numerical analyses around tunnel at end of first flight interaction effects (Pitilakis & Tsinidis, 2014). In this section
(deformations scale 310) two of the most commonly used methods are discussed,
60 60
D C Full slip
μ 0·4
45 45 μ 0·8
A B
No slip
τ: kPa/m
τ: kPa/m
30 30
15 15
0 0
A B C D A A B C D A
(a) (b)
Fig. 15. Effect of the soil–tunnel interface properties on the soil dynamic shear stress
distributions computed along the perimeter of the tunnel at the time step of maximum
racking distortion: (a) earthquake EQ4; (b) earthquake EQ7
79
SG-B1 – EQ2 SG-B1 – EQ2 SG-B4 – EQ2 SG-B4 – EQ2
M: N mm/mm 5 2·5 5 2·5
M: N mm/mm
M: N mm/mm
M: N mm/mm
0 0 0 0
5 2·5 5 2·5
10 5·0 10 5·0

0 0·3 0·6 0·10 0·15 0·20 0 0·3 0·6 0·10 0·15 0·20
t: s t: s t: s t: s
(a)
5 2·5 5 2·5
SG-B4
M: N mm/mm
M: N mm/mm
M: N mm/mm
M: N mm/mm
0 0 0 0
5 2·5 5 2·5
SG-B1
10 5·0 10 5·0

0 0·3 0·6 0·12 0·17 0·22 0 0·3 0·6 0·12 0·17 0·22
t: s t: s (b) t: s t: s

5 5 5 5
M: N mm/mm
M: N mm/mm
M: N mm/mm
M: N mm/mm
0 0 0 0
5 5 5 5
10 10 10 10

0 0·3 0·6 0·08 0·13 0·18 0 0·3 0·6 0·08 0·13 0·18
t: s t: s t: s t: s
(c)
Fig. 16. Dynamic bending moment–time histories recorded and computed by visco-elasto-plastic analysis for different earthquake input
motions, effect of the soil–tunnel interface properties: (a) earthquake EQ2; (b) earthquake EQ3; (c) earthquake EQ8
5·0 EQ2 12 EQ4
2·5 6
M: N mm/mm
M: N mm/mm
0 0
2·5 6
5·0 12
A B C D A A B C D A
(a) (b) D C
EQ6 EQ7
A B
10 2
5 1
M: N mm/mm
M: N mm/mm
0 0
5 1
10 2
A B C D A A B C D A
(c) (d)
Full slip μ 0·4 μ 0·8 No slip Test
Fig. 17. Residual dynamic bending moment distributions along the perimeter of the tunnel, recorded and
computed by visco-elasto-plastic analyses at the end of shaking: (a) earthquake EQ2; (b) earthquake EQ4;
(c) earthquake EQ6; (d) earthquake EQ7
80
SG-A3 – EQ3 SG-A3 – EQ3 SG-A4 – EQ3 SG-A4 – EQ3
2·50 2·50 1·0 0·50
1·25 1·25 0·5 0·25

N: N/mm
N: N/mm
N: N/mm
N: N/mm
0 0 0 0
1·25 1·25 0·5 0·25
2·50 2·50 1·0 0·50

0 0·3 0·6 0·11 0·16 0·21 0 0·3 0·6 0·11 0·16 0·21
t: s t: s t: s t: s
(a)

2·50 2·50 1·0 0·50
1·25 1·25 0·5 0·25 SG-A4

N: N/mm
N: N/mm
N: N/mm
N: N/mm
0 0 0 0 SG-A3
1·25 1·25 0·5 0·25
2·50 2·50 1·0 0·50

0 0·35 0·70 0·07 0·12 0·17 0 0·35 0·70 0·07 0·12 0·17
t: s t: s t: s t: s
(b)
2·50 2·50 1·0 0·6
1·25 1·25 0·5 0·3

N: N/mm
N: N/mm
N: N/mm
N: N/mm
0 0 0 0
1·25 1·25 0·5 0·3
2·50 2·50 1·0 0·60

0 0·3 0·6 0·07 0·12 0·17 0 0·3 0·6 0·07 0·12 0·17
t: s t: s t: s t: s
(c)
Fig. 18. Dynamic axial force–time histories recorded and computed by visco-elasto-plastic analysis for different earthquake input
motions, effect of the soil–tunnel interface properties: (a) earthquake EQ3; (b) earthquake EQ7; (c) earthquake EQ8
6·0 6·0
D C Elasto-plastic analysis
Visco-elastic analysis
2|ΔM|
4·5 A B 4·5 Test
|ΔM|: N mm/mm
|ΔN|: N mm/mm
2|ΔN|
3·0 3·0
1·5 1·5
0 0
A B C D A A B C D A
(a) (b)
Fig. 19. Internal forces dynamic increments along the tunnel perimeter: (a) bending moment for EQ3;
(b) axial force for EQ4
namely, the design procedure proposed by Wang (1993) and modelled as an equivalent static load or pressure that is
the pseudo-static seismic coefficient deformation method imposed on the frame (Fig. 20(a)). This ‘structural’ racking
(FHWA, 2009) or detailed equivalent static analysis method distortion is evaluated by the free-field ground racking
(ISO, 2005). distortion, which is properly adjusted, through the so-called
According to the first methodology, the tunnel seismic racking ratio (structural to ground racking distortions), in
response is evaluated through a simple static frame analysis. order to account for the soil–tunnel interaction effects. The
The structural racking distortion due to ground shaking is racking ratio is correlated with relative flexibility of the soil
81
Δstr
Finertia
αff δ
P
Δstr Δstr Δff
H H
Δstr
B B
(a) (b) (c)
Fig. 20. Schematic representation of the simplified analysis methods: (a) Wang (1993) simplified method, (b) detailed equivalent
static analysis method, distributed inertial loads, (c) detailed equivalent static analysis method, imposed deformations at model
boundaries
to the tunnel that is expressed through the flexibility ratio F separately for each earthquake scenario, using the numerical
(Wang, 1993) model presented in Fig. 4. Although simplified methods
G3B propose an equivalent linear approximation (e.g. degraded
F¼ (3) shear modulus computed from site response analysis) to
S3H account for the soil non-linear response under ground shak-
where G is the soil shear modulus, B and H are the width ing (e.g. FHWA, 2009), both elastic and elasto-plastic ana-
and the height of the structure, respectively, and S is the lyses are performed, using the constitutive models presented
required force to cause a unit racking deflection of the before, in order to check the effect of the soil yielding
structure. response on the results. Moreover, to study the effect of the
According to NCHPR611 regulations (Anderson et al., soil–tunnel interface properties, the analyses are carried out
2008) the racking ratio can be computed as under full slip and no-slip conditions. Sand mechanical
˜str 2F properties (e.g. stiffness and strength) are selected in order
R¼ ¼ (4) to correspond with those of the dynamic analysis, while the
˜ff (1 þ F) equivalent seismic loads (e.g. inertia forces or ground
displacements) are computed from the dynamic analysis,
In the detailed equivalent static analysis method, a 2D referring to the free field and for the time step of maximum
soil–tunnel numerical model is proposed for the analysis, tunnel racking distortion. To investigate the effect of the
similar to the dynamic analysis (ISO, 2005; FHWA, 2009). input motion amplitude, the analyses are performed for EQ3
The seismic load is introduced in a pseudo-static manner, as (0 .13g) and for EQ4 (0 .24g) according to Table 4, whereas
equivalent inertial load throughout the entire model that to study the input motion frequency content on the response,
corresponds to the ground free-field acceleration amplifica- a final set of analyses is performed using the Japanese
tion profile (Fig. 20(b)). In an alternative to this method, Meteorological Agency (JMA) record from the 1995 Kobe
equivalent seismic load is introduced as a ground deforma- earthquake scaled down to 0 .24g. The following presented
tion pattern on the numerical model boundaries (Fig. 20(c)), results refer to extreme scenarios regarding the tunnel flex-
corresponding to the free-field ground response (Kontoe et ibility and therefore they should be interpreted as limit
al., 2008; Hashash et al., 2010). cases. Soil strength parameters may affect the soil yielding
The test case presented herein is used as a case study to response and therefore may alter the results of non-linear
verify the effectiveness of the aforementioned simplified analyses. Considering the relatively low strength estimated
methods. More specifically, the results of the implemented in the examined cases and the associated increased yielding
simplified methods are compared to the calibrated dynamic response, the results may be considered conservative.
analysis that is used as the benchmark case. The compari- Table 5 presents representative comparisons of racking
sons are made in terms of computed racking ratio and ratios estimated from different approaches for EQ4, assum-
dynamic bending moment in the lining, which are con- ing elastic soil response. Generally, the numerical results for
sidered to be representative parameters for the validation. no-slip conditions resulted in larger racking ratios (12–35%
The flexibility ratio for the given case is estimated equal to larger) compared to the full slip conditions. Moreover, rack-
F ¼ 62 .5, indicating a quite flexible structure compared to ing ratios computed from the equivalent static analyses seem
the surrounding soil. To further extend the comparisons, a to be slightly lower (15–20%) compared to the dynamic
second series of analyses are performed, increasing the analysis results. The NCHPR611 analytical relation (Ander-
tunnel lining thickness, in order to model a rigid tunnel son et al., 2008) overestimates the racking ratio for the
(F ¼ 0 .29). Both static and dynamic analyses are performed flexible tunnel, while for the rigid tunnel, assuming no-slip
Table 5. Racking ratios estimated by different methods under the assumption of elastic soil response for EQ4
Case Dynamic Equivalent static analysis – Equivalent static analysis – NCHPR611 Anderson et al. (2008) –
analysis force displacement (R ¼ 2F/(1 + F))
Flexible tunnel – full slip 1 .3 1 .27 1 .22 1 .96

Flexible tunnel – no slip 1 .46 1 .42 1 .40 1 .96
Rigid tunnel – full slip 0 .5 0 .47 0 .40 0 .45
Rigid tunnel – no slip 0 .74 0 .72 0 .65 0 .45
82
conditions, numerical analyses result in a ratio larger than Table 6. Comparisons between recorded and computed – from
the analytical estimation. An underestimation of the racking different design methods – bending moments at receivers’
ratio will result in underestimation of the lining forces (e.g. positions (EQ3 elasto-plastic analyses for full slip conditions)
implementing Wang’s method). On the contrary, an over-
estimation of the racking ratio may lead to an overdesign Position M: N mm/mm
that may be considered as a conservative ‘safe’ design
Full Equivalent static Equivalent static Test
concept. However, overdesign is not only needlessly expen- dynamic analysis – force analysis –
sive but may lead to the stiffening of the structure, which analysis deformation
may in turn change the whole response pattern in a detri-
mental way. SG-B1 3 .90 2 .55 1 .74 4 .16
Figure 21 presents representative comparisons of the SG-B2 1 .59 0 .25 0 .20 3 .59
dynamic bending moment distributions along the tunnel’s SG-B4 4 .00 1 .10 0 .25 4 .21
perimeter, computed with different design methods, assum-
ing no-slip conditions. The elasto-plastic analyses numerical
results for the flexible tunnel case are also compared with
the experimental data (Fig. 21(b)). Table 6 tabulates similar the soil permanent response. Similar to the elastic analyses,
comparisons between the recorded and the computed dy- the differences are higher for the cases where the equivalent
namic bending moment at the locations of the strain gauges. seismic loads are introduced in terms of imposed ground
Numerical results correspond to full-slip conditions in this displacement at the boundaries. Local yielding at these
case. Generally, for the assumption of elastic soil response, boundary locations may affect the tunnel loading.
the equivalent static analyses reproduce well the computed Figure 22 plots static to dynamic bending moment ratios
bending moment distribution from the dynamic analysis. that are computed at a crucial lining section (joint C, Fig.
However, the maximum bending moment is underestimated 21) under different assumptions regarding the soil–tunnel
for both the flexible and the rigid tunnel, especially when interface properties, the soil response (elastic and elasto-
the equivalent seismic load is introduced in terms of defor- plastic) and the input motion characteristics. Generally,
mation at the model boundaries. equivalent static analyses underestimate the bending moment
In the case of the elasto-plastic analyses, bending moment compared to the full dynamic analysis. For the elastic
distributions are more complex, especially for the flexible analyses, the differences may reach 20 to 40%. The discre-
tunnel, due to the associated larger soil yielding. Experimen- pancies are even higher for the elasto-plastic analyses (dif-
tal data are generally closer to the dynamic analysis results ferences up to 60%), especially for the flexible tunnel case.
(Fig. 21(b) and Table 6). Actually, equivalent static analyses The differences are generally higher for the cases where the
results barely follow the experimental data and the bending equivalent seismic load is introduced in terms of ground
moment distribution computed by the dynamic analysis, displacements at the model boundaries. This may be attrib-
exhibiting values which are considerably lower. For the rigid uted to the relatively large distance between the tunnel and
tunnel case, simplified analyses results are closer to the the numerical model boundaries (14 .3 m for the side bound-
dynamic analysis, but again the differences are quite notice- aries), where the ground deformation is imposed. By in-
able. It is obvious that simplified methods cannot reproduce creasing this distance it is possible that a greater amount of
the soil loading history during shaking as efficiently as the induced ground strain is artificially absorbed by the soil
dynamic analysis. This loading history affects significantly elements, thus ‘relieving’ the structure and altering the
Elastic analyses, F 62·5 Elasto-plastic analyses, F 62·5

4 12
2 6
M: N mm/mm
M: N mm/mm
0 0
2 6
4 12
A B C D A A B C D A
(a) (b)

60 60
30 30
M: N mm/mm
M: N mm/mm
0 0
30 30
60 60
A B C D A A B C D A
(c) (d)
D C
Dynamic analysis EQL static analysis – force
EQL static analysis – deformation Test

A B
Fig. 21. Dynamic bending moment distributions along the tunnel perimeter computed from different methods for EQ3:
(a) flexible tunnel–elastic analysis; (b) flexible tunnel–elasto-plastic analysis; (c) rigid tunnel–elastic analysis; (d) rigid tunnel–
elasto-plastic analysis
83
1·0 1·0
0·8 0·8
Mstatic /Mdynamic
Mstatic /Mdynamic 0·6 0·6
0·4 0·4
0·2 0·2
0 0
EQ3 EQ4 JMA EQ3 EQ4 JMA
(a) (b)

1·0 1·0
0·8 0·8
Mstatic /Mdynamic
Mstatic /Mdynamic
0·6 0·6
0·4 0·4
0·2 0·2
0 0
EQ3 EQ4 JMA EQ3 EQ4 JMA
(c) (d)
Full slip, force Full slip, displacement No slip, force No slip, displacement
Fig. 22. Static to dynamic bending moment ratios computed at the right side-wall–roof slab corner: (a) flexible tunnel–
elastic analysis; (b) flexible tunnel–elasto-plastic analysis; (c) rigid tunnel–elastic analysis; (d) rigid tunnel–elasto-plastic
analysis
analysis results (Pitilakis & Tsinidis, 2014). It is worth earth pressures on the side walls and the lining forces, which
noting that Hashash et al. (2010) propose this distance to be were amplified by the increased flexibility of the tunnel.
significantly smaller. Soil–tunnel interface properties and This complex response associated with residual deformations
input motion characteristics seem to have a minor effect on and internal forces in the lining cannot be reproduced by the
the computed ratios in case of the elastic analyses, whereas equivalent linear approximation method that is often pro-
these parameters become more important in the case of the posed in regulations and used in engineering practice. There-
elasto-plastic analyses (especially in the case of the flexible fore, this approach should be used with caution, especially
tunnel), owing to their effect on the soil yielding response. when the tunnel is quite flexible and high soil non-linearity
is expected, as in the case of strong earthquakes.
The calibrated dynamic numerical models were finally
CONCLUSIONS used as a benchmark to validate the accuracy of currently
The paper presented representative experimental results used simplified methods. Racking ratios computed from the
from a series of dynamic centrifuge tests on a flexible model equivalent static analyses were found to be slightly lower
tunnel embedded in dry sand, along with results from compared to the dynamic analysis results, while the
numerical simulations of the tests. Numerical models were NCHPR611 analytical relation (Anderson et al., 2008) was
found capable of reproducing the recorded response with found to overestimate the racking ratio for the flexible tunnel
reasonable engineering accuracy. Some inevitable differences case. In general, simplified methods underestimated the
between the recorded and the computed response are attrib- tunnel lining forces compared to the full dynamic analysis.
uted to the difficulties in ascertaining precisely the soil, Assuming an elastic soil response, the differences were up
tunnel and soil–tunnel interface mechanical properties of the to 30%, and the discrepancies were much higher for the
centrifuge model. To a certain degree this also depends on cases when the soil permanent deformation was accounted
the constitutive models used; however, these models are for. Equivalent static analyses, where the load is introduced
adequately calibrated. All constitutive models actually con- in terms of distributed inertial loads throughout the model,
stitute an approximation of the actual sand behaviour under were found to be more efficient. The main conclusion drawn
seismic loading. Their accuracy depends on numerous is that simplified methods should be used with caution,
parameters, which are mainly affected by the typology and mainly during preliminary stages of design, and for cases
the complexity of the problem modelled. Sometimes model- where high soil non-linearity is not expected (e.g. rather low
ling very complex problems, such as the one in this paper, to medium seismic intensities).
using complicated constitutive models for the soil, which
could not be well calibrated, may increase considerably the
uncertainties and reduce the accuracy of the results. The use
of the models implemented herein and the comparisons to ACKNOWLEDGEMENTS
the experimental data are an additional verification of their The research leading to the presented experimental results
efficiency to model complicated problems such as the one has received funding from the European Community’s Se-
presented in this paper. venth Framework Programme (FP7/2007–2013) for access to
With regard to the tunnel’s response: both the experimen- the Turner beam centrifuge, Cambridge, UK, under grant
tal and the numerical data revealed a rocking mode of agreement no. 227887 (SERIES – Seismic Engineering Re-
vibration for the tunnel in addition to the racking distortion. search Infrastructures for European Synergies, http://
Inward deformations were also observed due to the high www.series.upatras.gr/). The excellent technical support re-
flexibility of the tunnel. Post-earthquake residual values were ceived by the technicians at the Schofield Centre is grate-
recorded experimentally and predicted numerically for the fully acknowledged.
84
NOTATION Anastasopoulos, I. & Gazetas, G. (2010). Analysis of cut-and-cover
A acceleration amplitude tunnels against large tectonic deformation. Bull. Earthquake
a input motion amplitude Engng 8, No. 2, 283–307, http://dx.doi.org/10.1007/s10518-009-
aff soil free-field horizontal acceleration 9135-4.
B tunnel width Anastasopoulos, I., Gerolymos, N., Drosos, V., Kourkoulis, R. &
c cohesion Georgarakos, T. (2007). Nonlinear response of deep immersed
D damping of sand tunnel to strong seismic shaking. J. Geotech. Geoenviron. Engng
d10 sand grain diameter at 10% passing 133, No. 9, 1067–1090, http://dx.doi.org/10.1061/(ASCE)1090-
d50 sand grain diameter at 50% passing 0241(2007)133:9(1067).
d60 sand grain diameter at 60% passing Anastasopoulos, I., Gerolymos, N., Drosos, V., Georgarakos, T.,
E aluminium alloy Young’s modulus Kourkoulis, R. & Gazetas, G. (2008). Behaviour of deep im-
Ec concrete Young’s modulus mersed tunnel under combined normal fault rupture deformation
e sand void ratio and subsequent seismic shaking. Bull. Earthquake Engng 6, No.
emax maximum sand void ratio 2, 213–239, http://dx.doi.org/10.1007/s10518-007-9055-0.
emin minimum sand void ratio Anderson, D. G., Martin, G. R., Lam, I. & Wang, J. N. (2008).
F soil to tunnel flexibility ratio NCHPR611: Seismic analysis and design of retaining walls,
Finertia equivalent to acceleration inertial load buried structures, slopes and embankments. Washington, DC,
f input motion dominant frequency USA: National Cooperative Highway Research Program, Trans-
fbk aluminium alloy tensile strength portation Research Board.
G sand reduced shear modulus Bardet, J. B., Ichii, K. & Lin, CH. (2000). EERA: a computer
Gmax sand small-strain shear modulus program for equivalent-linear earthquake site response analyses
H tunnel height of layered soil deposits. Los Angeles, CA, USA: University of
K0 earth coefficient at rest Southern California, Department of Civil Engineering.
l length Bilotta, E., Lanzano, G., Madabhushi, S. P. G. & Silvestri, F.
M lining bending moment per unit length (2014). A numerical round robin on tunnels under seismic
Mdynamic lining bending moment evaluated through dynamic actions. Acta Geotechnica 9, No. 4, 563–579, http://dx.doi.org/
analysis 10.1007/s11440-014-0330-3.
Mstatic lining bending moment evaluated through equivalent static Bobet, A. (2010). Drained and undrained response of deep tunnels
analysis subjected to far-field shear loading. Tunnell. Underground Space
m mass Technol. 25, No. 1, 21–31, http://dx.doi.org/10.1016/j.tust.2009.
N lining axial load per unit length 08.001.
n scale factor Bobet, A., Fernandez, G., Huo, H. & Ramirez, J. (2008). A
P equivalent to tunnel racking distortion force practical iterative procedure to estimate seismic-induced defor-
R racking ratio mations of shallow rectangular structures. Can. Geotech. J. 45,
RC resonant column tests No. 7, 923–938, http://dx.doi.org/10.1139/T08-026.
S required force to cause a unit racking deflection of the Brennan, A. J., Thusyanthan, N. I. & Madabhushi, S. P. G. (2005).
tunnel Evaluation of shear modulus and damping in dynamic centrifuge
t time tests. J. Geotech. Geoenviron. Engng 131, No. 12, 1488–1497,
TX cyclical triaxial tests http://dx.doi.org/10.1061/(ASCE)1090-0241(2005)131:12(1488).
Vso small-strain shear velocity gradient of sand Chen, G., Wang, Z., Zuo, X., Du, X. & Gao, H. (2013). Shaking
Æ reduction coefficient for sand shear modulus during table test on seismic failure characteristics of a subway station
shaking structure in liquefiable ground. Earthquake Engng Structl Dy-
ª shear strain nam. 42, No. 10, 1489–1507, http://dx.doi.org/10.1002/eqe.2283.
ªt aluminium alloy unit weight Chian, S. C. & Madabhushi, S. P. G. (2012). Effect of buried depth
˜ff free-field ground racking distortion and diameter on uplift of underground structures in liquefied
˜str tunnel racking distortion soils. Soil Dynam. Earthquake Engng 41, No. 1, 181–190,
|˜M| lining bending moment dynamic increment http://dx.doi.org/10.1016/j.soildyn.2012.05.020.
|Ñ| lining axial force dynamic increment Chou, J. C., Kutter, B. L., Travasarou, T. & Chacko, J. M. (2010).
� horizontal deformation at soil surface Centrifuge modelling of seismically induced uplift for the BART
� soil–tunnel interface friction coefficient transbay tube. J. Geotech. Geoenviron. Engng 137, No. 8, 754–
� aluminium alloy Poisson ratio 765, http://dx.doi.org/10.1061/(ASCE)GT.1943–5606.0000489.
rs sand density Cilingir, U. (2009). Seismic response of tunnels. PhD thesis, Uni-
� dynamic earth pressure per unit length versity of Cambridge, UK.
�9 mean effective stress Cilingir, U. & Madabhushi, S. P. G. (2011a). A model study on the
� dynamic shear stress per unit length effects of input motion on the seismic behaviour of tunnels. Soil
� sand friction angle Dynam. Earthquake Engng 31, No. 3, 452–462, http://
�cv sand critical friction angle dx.doi.org/10.1016/j.soildyn.2010.10.004.
ł sand dilatancy angle Cilingir, U. & Madabhushi, S. P. G. (2011b). Effect of depth on the
seismic response of square tunnels. Soils Found. 51, No. 3, 449–
457, http://dx.doi.org/10.3208/sandf.51.449.
Cilingir, U. & Madabhushi, S. P. G. (2011c). Effect of depth on the
REFERENCES seismic response of circular tunnels. Can. Geotech. J. 48, No. 1,
Abaqus (2012). Abaqus: theory and analysis user’s manual version 117–127, http://dx.doi.org/10.1139/T10-047.
6.12. Providence, RI, USA: Dassault Systèmes Simulia. Conti, R., Viggiani, G. M. B. & Perugini, F. (2014). Numerical
AFPS/AFTES (2001). Guidelines on earthquake design and protection modelling of centrifuge dynamic tests of circular tunnels in dry
of underground structures, Version 1. Paris, France: Working sand. Acta Geotechnica 9, No. 4, 597–612, http://dx.doi.org/
Group of the French Association for Seismic Engineering (AFPS) 10.1007/s11440-13-0286-8.
and French Tunnelling Association (AFTES). Dowding, C. H. & Rozen, A. (1978). Damage to rock tunnels from
Amorosi, A. & Boldini, D. (2009). Numerical modelling of the earthquake shaking. J. Geotech. Engng Div., ASCE 104, No. 2,
transverse dynamic behaviour of circular tunnels in clayey soils. 175–191.
Soil Dynam. Earthquake Engng 29, No. 6, 1059–1072, http:// FHWA (Federal Highway Administration) (2009). Technical manual
dx.doi.org/10.1007/s11440-013-0295-7. for design and construction of road tunnels – Civil elements,
Amorosi, A., Boldini, D. & Falcone, G. (2014). Numerical predic- Publication No. FHWA-NHI-10-034. Washington, DC, USA:
tion of tunnel performance during centrifuge dynamic tests. Acta U. S. Department of Transportation, Federal Highway Adminis-
Geotechnica 9, No. 4, 581–596, http://dx.doi.org/10.1007/ tration.
s11440-013-0295-7. Ghosh, B. & Madabhushi, S.P.G. (2002). An efficient tool for
85
measuring shear wave velocity in the centrifuge. In Proceedings Lanzano, G., Bilotta, E., Russo, G., Silvestri, F. & Madabhushi, S.
of international conference on physical modelling in geotechnics P. G. (2010). Dynamic centrifuge tests on shallow tunnel models
(eds R. Phillips, P. J. Guo and R. Popescu), pp. 119–124. St in dry sand. Proceedings of the VII international conference on
Johns, NF, Canada: Balkema. physical modelling in geotechnics (ICPMG 2010), pp. 561–567.
Gomes, R. C. (2014). Numerical simulation of the seismic response Zurich, Switzerland: Taylor and Francis.
of tunnels in sand with an elastoplastic model. Acta Geotechnica Lanzano, G., Bilotta, E., Russo, G., Silvestri, F. & Madabhushi, S.
9, No. 4, 613–629, http://dx.doi.org/10.1007/s11440-013-0287-7. P. G. (2012). Centrifuge modelling of seismic loading on tunnels
Hardin, B. O. & Drnevich, V. P. (1972). Shear modulus and in sand. Geotech. Testing J. 35, No. 6, 10–26, http://dx.doi.org/
damping in soils. J. Soil Mech. Found. Div., ASCE 98, No. 7, 10.1520/GTJ104348.
667–692. Lanzano, G., Bilotta, E., Russo, G. & Silvestri, F. (2014). Experi-
Hashash, Y. M. A., Hook, J. J., Schmidt, B. & Yao, J. I.-C. (2001). mental and numerical study on circular tunnels under seismic
Seismic design and analysis of underground structures. Tunnell. loading. Eur. J. Environ. Civ. Engng, http://dx.doi.org/10.1080/
Underground Space Technol. 16, No. 2, 247–293, http:// 19648189.2014.893211.
dx.doi.org/10.1016/S0886-7798(01)00051-7. Li, Z., Escoffier, S. & Kotronis, P. (2013). Using centrifuge tests
Hashash, Y. M. A., Park, D. & Yao, J. I. C. (2005). Ovaling data to identify the dynamic soil properties: Application to
deformations of circular tunnels under seismic loading, an Fontainebleau sand. Soil Dynam. Earthquake Engng 52, No. 1,
update on seismic design and analysis of underground structures. 77–87, http://dx.doi.org/10.1016/j.soildyn.2013.05.004.
Tunnell. Underground Space Technol. 20, No. 5, 435–441, Madabhushi, S. P. G., Schofield, A. N. & Lesley, S. (1998). A new
http://dx.doi.org/10.1016/j.tust.2005.02.004. stored angular momentum (SAM) based actuator. In Proceedings
Hashash, Y. M. A., Karina, K., Koutsoftas, D. & O’Riordan, N. of the international conference, centrifuge 98 (eds T. Kimura, O.
(2010). Seismic design considerations for underground box Kusakabe and J. Takemura), pp. 111–116. Tokyo, Japan: Balk-
structures. In Proceedings of the earth retention conference 3 ema.
(eds R. Finno, Y. M. A. Hashash and P. Arduino), pp. 620–637, Madabhushi, S. P. G., Houghton, N. E. & Haigh, S. K. (2006). A
http://dx.doi.org/10.1061/41128(384)64. Bellevue, WA, USA: new automatic sand pourer for model preparation at University
ASCE. of Cambridge. In Physical modelling in geotechnics – 6th
Hleibieh, J., Wagener, D. & Herle, I. (2014). Numerical simulation ICPMG 08 (eds C. W. W. Ng, L. M. Zhang and Y. H. Wang),
of a tunnel surrounded by sand under earthquake using a pp. 217–222. London, UK: Taylor and Francis.
hypoplastic model. Acta Geotechnica 9, No. 4, 631–640, http:// Pakbaz, M. & Yareevand, A. (2005). 2-D analysis of circular tunnel
dx.doi.org/10.1007/s11440-013-0294-8. against earthquake loading. Tunnell. Underground Space Tech-
Huo, H., Bodet, A., Fernández, G. & Ramirez, J. (2005). Load nol. 20, No. 5, 411–417, http://dx.doi.org/10.1016/j.tust.2005.01.
transfer mechanisms between underground structure and sur- 006.
rounding ground: evaluation of the failure of the Daikai station. Penzien, J. (2000). Seismically induced racking of tunnel linings.
J. Geotech. Geoenviron. Engng 131, No. 12, 1522–1533, http:// Earthquake Engng Structl Dynamics 29, No. 5, 683–691.
dx.doi.org/10.1061/(ASCE)1090-0241(2005)131:12(1522). Pistolas, G. A., Tsinaris, A., Anastasiadis, A. & Pitilakis, K.
Huo, H., Bodet, A., Fernandez, G. & Ramirez, J. (2006). Analytical (2014). Undrained dynamic properties of Hostun sand. Proceed-
solution for deep rectangular structures subjected to far-field ings of 7th Greek geotechnics conference, Athens, Greece.
shear stresses. Tunnell. Underground Space Technol. 21, No. 6, Athens, Greece: Hellenic Society of Soil Mechanics and Geo-
613–625, http://dx.doi.org/10.1016/j.tust.2005.12.135. technical Engineering (in Greek).
Hwang, J.-H. & Lu, C.-C. (2007). Seismic capacity assessment of Pitilakis, D. & Clouteau, D. (2010). Equivalent linear substructure
old Sanyi railway tunnels. Tunnell. Underground Space Technol. approximation of soil–foundation–structure interaction: model
22, No. 4, 433–449, http://dx.doi.org/10.1016/j.tust.2006.09. presentation and validation. Bull. Earthquake Engng 8, No. 2,
002. 257–282, http://dx.doi.org/10.1007/s10518-009-9128-3.
Iida, H., Hiroto, T., Yoshida, N. & Iwafuji, M. (1996). Damage to Pitilakis, K. & Tsinidis, G. (2014). Performance and seismic design of
Daikai subway station Japanese Geotechnical Society, Japan. underground structures. In Earthquake geotechnical engineering
Soils Found., Special Issue on Geotechnical Aspects of the design (eds M. Maugeri and C. Soccodato), Geotechnical Geologi-
January 17 1995, Hyogoken-Nambu Earthquake, 283–300. cal and Earthquake Engineering, vol. 28, pp. 279–340, http://
Ishibashi, I. & Zhang, X. (1993). Unified dynamic shear moduli dx.doi.org/10.1007/978-3-319-03182-8_11. Geneva, Switzerland:
and damping ratios of sand and clay. Soils Found. 33, No. 1, Springer.
182–191. Schanz, T. & Vermeer, P. A. (1996). Angles of friction and
ISO (International Organization for Standardization) (2005). ISO dilatancy of sand. Géotechnique 46, No. 1, 145–151, http://
23469: Bases for design of structures – Seismic actions for dx.doi.org/10.1680/geot.1996.46.1.145.
designing geotechnical works, International Standard ISO TC Schofield, A. N. (1980). Cambridge geotechnical centrifuge opera-
98/SC3/WG10. Geneva, Switzerland: International Organization tions. Géotechnique 30, No. 3, 227–268, http://dx.doi.org/
for Standardization. 10.1680/geot.1980.30.3.227.
Jaky, J. (1948). The coefficient of earth pressure at rest. J. Union Schofield, A. N. (1981). Dynamic and earthquake centrifuge model-
Hungarian Engrs Architects 78, No. 22, 355–358. ling. Proceedings of international conference on advances in
Kawashima, K. (2000). Seismic design of underground structures in geotechnical earthquake engineering and soil dynamics, pp.
soft ground: a review. In Geotechnical aspects of underground 1081–1100. Rolla, MO, USA: University of Missouri.
construction in soft ground (eds O. Kusakabe, K. Fujita and Seed, H. B., Wong, R. T., Idriss, I. M. & Tokimatsu, K. (1986).
Y. Miyazaki). Rotterdam, the Netherlands: Balkema. Moduli and damping factors for dynamic analyses of cohesion-
Kirtas, E., Rovithis, E. & Pitilakis, K. (2009). Subsoil interventions less soils. J. Geotech. Engng 112, No. 11, 1016–1032, http://
effect on structural seismic response. Part i: validation of dx.doi.org/10.1061/(ASCE)0733-9410(1986)112:11(1016).
numerical simulations. J. Earthquake Engng 13, No. 2, 155– Sharma, S. & Judd, W. R. (1991). Underground opening damage
169, http://dx.doi.org/10.1080/13632460802347463. from earthquakes. Engng Geol. 30, No. 3–4, 263–276, http://
Kontoe, S., Zdravkovic, L., Potts, D. & Mentiki, C. (2008). Case dx.doi.org/10.1016/0013-7952(91)90063-Q.
study on seismic tunnel response. Can. Geotech. J. 45, No. 12, Shibayama, S., Izawa, J., Takahashi, A., Takemura, J. & Kusakabe,
1743–1764, http://dx.doi.org/10.1139/T08-087. O. (2010). Observed behavior of a tunnel in sand subjected to
Kontoe, S., Zdravkovic, L., Potts, D. & Mentiki, C. (2011). On the shear deformation in a centrifuge. Soils Found. 50, No. 2, 281–
relative merits of simple and advanced constitutive models in 294, http://dx.doi.org/10.3208/sandf.50.281.
dynamic analysis of tunnels. Géotechnique 61, No. 10, 815–829, St John, C. M. & Zahrah, TF. (1987). Aseismic design of under-
http://dx.doi.org/10.1680/geot.9.P.141. ground structures. Tunnell. Underground Space Technol. 2, No.
Kontoe, S., Avgerinos, V. & Potts, D. M. (2014). Numerical valida- 2, 165–197, http://dx.doi.org/10.1016/0886-7798(87)90011-3.
tion of analytical solutions and their use for equivalent-linear Tsinidis, G., Heron, C., Pitilakis, K. & Madabhushi, S. P. G. (2014a).
seismic analysis of circular tunnels. Soil Dynam. Earthquake Physical modelling for the evaluation of the seismic behavior of
Engng 66, No. 1, 206–219, http://dx.doi.org/10.1016/j.soil- square tunnels. In Seismic evaluation and rehabilitation of struc-
dyn.2014.07.004. tures (eds A. Ilki and M. Fardis), Geotechnical Geological and
86
Earthquake Engineering, vol. 26, pp. 389–406, http://dx.doi.org/ due to the Taiwan Chi-Chi earthquake. Tunnell. Underground
10.1007/978-3-319-00458-7_22. Geneva, Switzerland: Springer. Space Technol. 16, No. 3, 133–150, http://dx.doi.org/10.1016/
Tsinidis, G., Pitilakis, K. & Trikalioti, A. D. (2014b). Numerical S0886-7798(01)00047-5.
simulation of round robin numerical test on tunnels using a Zeghal, M. & Elgamal, A. W. (1994). Analysis of site liquefaction
simplified kinematic hardening model. Acta Geotechnica 9, No. using earthquake records. J. Geotech. Engng ASCE 120, No. 6,
4, 641–659, http://dx.doi.org/10.1007/s11440-013-0293-9. 996–1017, http://dx.doi.org/10.1061/(ASCE)0733-9410(1994)120:
Wang, J. N. (1993). Seismic design of tunnels: A simple state 6(996).
of the art design approach. New York, NY, USA: Parsons Zeng, X. & Schofield, A. N. (1996). Design and performance of an
Brinckerhoff. equivalent shear beam (ESB) container for earthquake centrifuge
Wang, W. L., Wang, T. T., Su, J. J., Lin, C. H., Seng, C. R. & modelling. Géotechnique 46, No. 1, 83–102, http://dx.doi.org/
Huang, T. H. (2001). Assessment of damage in mountain tunnels 10.1680/geot.1996.46.1.83.
87
Bertalot, D. & Brennan, A. J. (2015). Géotechnique 65, No. 5, 418–428 [http://dx.doi.org/10.1680/geot.SIP.15.P.002]
Influence of initial stress distribution on liquefaction-induced settlement of

shallow foundations
D. B E RTA L OT a n d A . J. B R E N NA N †
During earthquakes, saturated sandy soils may generate significant excess pore pressures and approach
a state of liquefaction. Structures founded on shallow foundations above such soils may consequently
undergo large settlements. Recent case history analysis has shown that the stress imposed by the
foundation is a key factor in the estimation of such settlements. However, the case history data
showed that although increasing bearing pressure caused an increase in settlements as expected, this
was only true up to a point, and that very heavy structures appeared to settle less than some lighter
structures. This work aims to investigate these counter-intuitive results by means of controlled
experimental testing using a geotechnical centrifuge. Results of the centrifuge tests show that the trend
derived from case histories is correct and that liquefaction-induced settlements peak for a given
bearing stress (90 kPa for the models tested) and reduce for greater applied stresses. Further, by
analysis of excess pore pressure distributions beneath the foundations it is shown that the main factor
inhibiting pore pressure generation beneath the footings is not so much the confining pressure as the
in-situ static shear stress around the edge of the foundation. This is supported by element test data
from the literature. When this initial static shear stress is so high that the applied cyclic shear stress
cannot exceed it (i.e. the direction of shear stress does not reverse) then pore pressure generation is
greatly reduced, thus causing the observed reduction in expected settlements.
KEYWORDS: centrifuge modelling; footings/foundations; liquefaction
INTRODUCTION motions and interpretation, meaning that further under-

During earthquakes, an increase in pore pressure may standing would be better derived from controlled model
occur in saturated sandy/silty soils as a result of the soil testing.
structure collapse and consequent transfer of load to the Most of the current understanding of the mechanics
pore fluid. This phenomenon results in a reduction of the governing soil liquefaction has been derived by laboratory
effective stresses acting on the soil and a consequent element testing (mainly cyclic triaxial and cyclic simple
degradation of the soil’s shear stiffness and strength, shear tests), leading to the recognition of three main initial
facilitating settlement of structures having shallow founda- state variables controlling the cyclic resistance of liquefiable
tions. If the generated excess pore pressure is high enough soils: relative density, confining pressure and static shear
to induce full liquefaction underneath the footing (i.e. stress. The effect of static shear stress in particular has been
equals the effective overburden stress), very large deforma- the object of debate over the last 40 years, and as the stress
tions may take place, with one resultant phenomenon fields beneath shallow foundations contain such static shear
being the excessive settlement of structures with shallow stresses, an understanding of this is required to capture the
foundations. A number of examples of this exist through behaviour of such systems. Initially, cyclic triaxial tests on
the literature. isotropic consolidated samples or cyclic simple shear tests
Data from this earthquake helped to identify an omis- on one-dimensionally consolidated samples were commonly
sion from the existing methods of assessing such settle- used to assess the behaviour of saturated sands. However,
ments. Current practice is based primarily on the variables despite satisfactorily reproducing the behaviour of level
of footing width B and the depth of liquefiable soil DL saturated soil deposits (where static shear stress is zero),
(Liu & Dobry, 1997). Analysis of both new and previously these techniques could not capture the soil behaviour in
published case histories by Bertalot et al. (2013) led to sloping ground or underneath a footing, where significant
the identification of footing bearing pressure q as a signifi- static shear stresses are induced (Vaid & Finn, 1979). Since
cant additional variable. Further, it was also identified then, many authors have recognised that static shear stresses
from this analysis that, although observed settlements in- may, under given conditions, increase the soil’s resistance to
creased with q, this was only true for lower imposed liquefaction by limiting the generation of excess pore pres-
stresses and that structures imposing stresses of the order sure during cyclic loading (Lee & Seed, 1967; Vaid & Finn,
of 100 kPa and above on the soil appeared to undergo 1979; Boulanger & Seed, 1995; Vaid et al., 2001; Kam-
comparatively reduced settlements. However, case history merer, 2002).
data contain significant variability in local site factors, The aim of the current study is therefore to present an
explanation of the apparently counter-intuitive relationship
between settlement and footing bearing stress by means of a
Manuscript received 27 March 2014; revised manuscript accepted 9 series of dynamic centrifuge tests. This will be achieved by
January 2015. Published online ahead of print 7 April 2015.
first validating measured settlements against the case study
see p. ii. data, then looking more closely at developed excess pore
D’Appolonia S.p.A., Genova, Italy; also Civil Engineering Depart- pressures, and consequently determining whether existing
ment, University of Dundee, UK. knowledge about static shear stresses can explain the ob-
† Civil Engineering Department, University of Dundee, UK. served pore pressures and settlement trends.
89
BERTALOT AND BRENNAN
CENTRIFUGE MODELLING action and resultant ratcheting of the rigid footing into the
To achieve this aim, three centrifuge tests have been soil, footings have identical low heights and bearing pres-
performed on the University of Dundee geotechnical centri- sures achieved by controlling material density (aluminium,
fuge. Centrifuge modelling enables small-scale models to be steel and lead). All of the footing models tested are placed
tested at stress levels equivalent to a larger prototype. In this on the ground surface, with an embedment of 0 .5 m.
case, a length scale factor of 50 has been adopted, and a All three models consist of 15 m deep deposits of clean
corresponding centrifugal acceleration of 50g applied. All sand with a top loose liquefiable layer (DR ¼ 40%) and a
data are here reported in prototype scales; a full discussion bottom dense layer (DR ¼ 80 %). The thickness of the
of scaling may be found in, for example, Muir Wood (2004) liquefiable layer varies in each model as the analysed tests
or Schofield (1980). This is a 3 m diameter Actidyn C-67-2 are part of a broader testing campaign whose goal is, among
centrifuge, equipped with the Actidyn QS-67-2 one- others, the investigation of the role of DL on footing settle-
dimensional servo-hydraulic shaker capable of applying user- ment. The presence of the bottom dense sand layers is
defined motions up to 0 .4g within the range of 0 .6–8 Hz necessary to achieve the same overall depth of the soil
prototype scale at the 50g acceleration considered (Brennan deposit in the three centrifuge models. The effect of these
et al., 2014). layers was seen to have limited effects both on the overall
Each centrifuge model tested consists of four rigid square vertical strain of the soil deposit and on the vertical propa-
footings (B ¼ 2 .75 m at prototype scale) resting on level gation of the base motion. In test BD8 the top liquefiable
loose liquefiable sand (DR ¼ 40%) deposits of different sand layer has a depth of 8 .25 m in prototype scale, whereas
thickness. Despite having the same dimensions, each model in test BD5 the liquefiable layer depth is reduced to 4 m
footing exerts a different bearing pressure, namely 30, 60, measured from the foundation plan. In test BD7 the contain-
90 and 130 kPa (Figure 1). To minimise the component of er was divided into different sectors by means of thin
the settlement due to moment-induced soil–structure inter- aluminium walls to achieve different stratification profiles
PPT6
130 kPa
PPT7 PPT5
PPT4
DR 40% PPT3 PPT2
PPT1
Test BD5
10·25 15
DR 80%
PPT6
90 kPa
DR 40%
PPT7 PPT5
PPT4
PPT3 PPT2
DR 40%
Test BD7
13·3
15
DR 40%
8·25
4·25
DR 80% DR 80% DR 80%
60 kPa
PPT5 PPT7
PPT6
Test BD8 15
DR 40% PPT1
DR 80%
Fig. 1. Layout of centrifuge tests and instrument positions
90
INITIAL STRESS DISTRIBUTION AND SETTLEMENT OF SHALLOW FOUNDATIONS
within the same model. Thin aluminium walls (i.e. 0 .3 mm Table 1. Properies of HST95 Congleton sand
thick) were used for this purpose in order to create a low
lateral stiffness impermeable boundary between the different Property Measured Lauder (2010)
compartments, able to stop pore fluid migration without
imposing constraints to the soil model. Because of the Gs: g/cm3 – 2 .63
ª d,min: kN/m 3
14 .34 14 .59
limited bending stiffness of the walls, lateral deformations at
ªd,max: kN/m3 17 .6 17 .58
these boundaries are assumed to be controlled by the soil. In �crit: degrees 33 32
particular, liquefiable layers of 6 and 10 m have been �peak: degrees� 46 44
reproduced in this test.
In order to compare the results from different tests, � Indicates a property evaluated for soil at D ¼ 90%.
R
normalised settlements (S/DL) are considered. This normal-
isation has been previously used to compare field cases of
liquefaction-induced foundation settlement (Yoshimi & Toki- Soil models are contained in an equivalent shear beam
matsu, 1977; Liu & Dobry, 1997). Dashti et al. (2010) and box, consisting of stacked aluminium rings separated by thin
Bertalot et al. (2013) pointed out that the assumption of a rubber layers providing the desired flexibility (Bertalot
linear relationship between the induced settlement and lique- (2012), and shown schematically in Fig. 1). The container is
fied soil thickness may be misleading for small B/DL ratios. designed so that the shear stiffness of the end walls matches
This limits the value of quantitative comparison of normal- that of the soil model before soil softening due to shaking-
ised footing settlement in different tests. induced excess pore pressure build-up, thereby minimising
Besides footing settlement, pore pressure and acceleration dynamic boundary effects. Upon liquefaction the soil stiff-
are measured within the soil. In each model the pore ness drops significantly and this condition is no longer met;
pressure measurements were concentrated underneath a dif- however, such stiffness reduction also inhibits the propaga-
ferent footing, providing detailed information about the tion of parasitic waves generating at the model boundaries
excess pore pressure generation pattern in an area with width due to the marked stiffness contrast.
B from the footing axis, and depth B from the foundation The input motion chosen was that experienced at the
plan. A schematic layout of these tests is shown in Fig. 1. primary case study site during the 2010 Maule earthquake.
The soil used for these models was HST95 Congleton The initial reason is to better compare data obtained against
silica sand; this is a fine, uniformly graded sand with that from the case study site (Brennan et al., 2014). This
rounded particle shape (Table 1). In order to achieve a motion was also seen as useful due to its long duration and
correct scaling of time in both inertial and seepage con- large magnitude (Mw ¼ 8 .8), meaning that settlements
trolled phenomena, a pore fluid with a viscosity 50 times achieved are likely to be an upper bound on those experi-
higher than that of water was required (Stewart et al., 1998). enced in lower magnitude events, while also being less
This was achieved by using a solution of methylcellulose in damaging than the sinusoidal shaking applied in older
water as pore fluid, at a concentration of 2 .3% of methylcel- centrifuge based investigations. Fig. 2 shows the target
lulose by mass. motion record used in both time and frequency domain,
Test BD3
4
0·015
0 0·010
0·005
4
0
4 Test BD4
0·015
0 0·010
Amplitude: g/Hz
0·005
4
0
a: g
4 Test BD6
0·015
0 0·010
0·005
4
0
4 Test BD8
0·015
0 0·010
0·005
4
0
0 10 20 30 40 50 60 70 80 90 100 0 1 2 3 4 5 6 7 8
Time: s Frequency: Hz
Fig. 2. Acceleration record measured at the ‘Colegio Concepción’ station (San Pedro de la Paz) during 2010 Maule earthquake
and earthquake motions reproduced in the centrifuge tests
91
together with the motion recorded on the centrifuge models, EXCESS PORE PRESSURE GENERATION
indicating a very good match between target motion and The effective confining stress acting on the soil has been
achieved motion, as well as good repeatability. observed to influence its potential for excess pore pressure
(Epp) generation. Steedman et al. (2000) performed a series
of centrifuge tests investigating the cyclic behaviour of
SETTLEMENTS saturated sands under high confining stress (� v9 ) by testing
The initial objective of this work was to validate the trend deep level soil models with a uniform, freely draining
observed in field data concerning the liquefaction-induced surcharge. A reduction of the measured excess pore pressure
settlement of buildings having shallow foundations. Ob- ratio (ru) with depth (i.e. with increasing � v9 ) was observed.
served settlements of such buildings from the Chile earth- Similar conditions are present beneath a footing resting on a
quake, as well as past events in Niigata (Yoshimi & liquefiable sand deposit. However, in this case the confining
Tokimatsu, 1977) and Luzon in the Philippines (Adachi et effect exerted by the footing is localised to the soil under-
al., 1992; Acacio et al., 2001), are plotted as a function of neath it, generating a more complex stress distribution.
bearing pressure in Fig. 3. The analysis of the documented Beside the horizontal dis-homogeneity of confining stress,
case histories of this specific phenomenon suggests that high shallow foundations also generate shear stresses in the
foundation bearing pressure may result in reduced settlement foundation soil, significantly influencing its cyclic response.
in the case of liquefaction of the foundation soil (Bertalot et In order to investigate the excess pore pressure generation in
al., 2013). However, it is seen that there are limited field such conditions, in each of the centrifuge tests performed
data for such heavy structures, necessitating the confirmation the soil underneath one of the footing models (test BD8–
of this with the centrifuge models. 60 kPa footing; test BD7–90 kPa footing; test BD5–130 kPa
Figure 3 therefore also includes the total normalised footing) has been instrumented with an array of pore pres-
settlements recorded during tests BD5, BD7 and BD8 sure transducers (PPTs).
plotted against the footing model bearing pressure. As instruments located at shallow depth underneath a
The hypothesised bearing pressure dependence of the footing are displaced proportionally to footing settlement
liquefaction-induced settlement is confirmed by the experi- during testing, and excess pore pressure requires reference to
mental results. Footing settlement was indeed seen to be a measurement location, the pore pressure readings have
directly proportional to the footing’s bearing pressure for been corrected for instrument displacement. The positions of
bearing pressure up to about 100 kPa, as had been ex- the instruments prior to and after liquefaction were carefully
pected based on the field data. However, the heaviest measured. Correction of the excess pore pressure measure-
(130 kPa) footing also experiences less settlement than the ment (Epp) was then carried out by subtracting the hydro-
lighter ones. Data from similar experimental works from static surplus due to the instrument being pushed deeper into
Hausler (2002) and Dashti et al. (2010) have also been the foundation soil
included in Fig. 3 for comparison. In particular, Dashti et Epp(t) ¼ Epp0 (t) � (z � z0 )[S av (t)=S av,final ]ªw
al. tested two types of footing models characterised by
different bearing pressure (namely 76 and 120 kPa). In where Epp0 represents the uncorrected measurement, z and
line with what was observed in this study, increasing the z0 the final and initial instrument positions, respectively
footing bearing pressure from 76 to 120 kPa did not result (below foundation plan), Sav(t ) and Sav,final are the settlement
in further footing settlement in Dashti et al.’s experiment. at time t and the final settlement, respectively, and ªw is the
Evidence from centrifuge test results implies that the trend unit weight of water. It is noted that instruments did not
identified by Bertalot et al. (2013) is due to a real displace laterally during testing.
physical phenomenon rather than arising by inconsistencies Figure 4 shows the instrument initial positions and the
and omissions in the field data set. The remainder of this excess pore pressure measurements during shaking from the
work therefore attempts to explain this trend by investigat- centrifuge tests. It is acknowledged that the thickness of
ing the pore pressures beneath the footings. the liquefiable layer was different for different models.
0·250
q 90 kPa
0·225
S/DL 0·362
Niigata earthquake (Yoshimi & Tokimatsu, 1977)
0·200 Luzon earthquake (Adachi et al., 1992)
Luzon earthquake (Adachi et al., 2001)
Normalised settlement, S/DL
0·175 Maule earthquake (2010)

Test BD8 (DL 8·25 m)
0·150
Test BD5 (DL 4 m)
0·125 Test BD7 (DL 1 m)
0·100 Test BD7 (DL 6 m)

Test BD7 (DL 10 m)
0·075
Dashti et al. (2010) (DL 3 m)
0·050 Dashti et al. (2010) (DL 6 m)
Hausler (2002) (DL 20 m)
0·025
0
20 30 40 50 60 70 80 90 100 110 120 130 140
Foundation bearing pressure, q: kPa
Fig. 3. Footing settlement measured in centrifuge tests BD5, BD7 and BD8 against footing bearing pressure (q)
92
60
45
PPT1 – free field
Epp: kPa
30
PPT2 – footing axis
15
PPT3 – footing edge
0
(a)
40
30 PPT6 –footing edge

Epp: kPa
20 PPT7 – near footing
10
0 0·500
5
0
(b)
20
Depth from foundation plan: m

PPT5 PPT6 PPT7
15
0·750
Epp: kPa
10
5 PPT4
1·875
0
5
(c) PPT2 PPT3
Acceleration: g
3·000
0·4
0
0·4 4·000
0 10 20 30 40 50 60 70 0 1·375 2·750
Time: s Distance from footing axis: m
Fig. 4. During shaking Epp measurements in the soil underneath: (a) 130 kPa footing; (b) 90 kPa footing; (c) 60 kPa footing
Although this is more likely to affect the long-term pore Comparing the settlement–, acceleration– and pore pressure–
pressure dissipation rather than the short-term pore pressure time histories during shaking relative to the 130 kPa footing
generation, there may still be variation between tests, pre- and the underlying soil, four main phases can be identified
venting direct comparison of quantitative values. However, (Figure 5).
inspecting the distribution of pore pressures during shaking
in each case is still instructive. The excess pore pressure (a) Significant excess pore pressures are generated under-
traces in Fig. 4 show that, for the heavier footings, the neath the central axis of the footing, exceeding the
lowest excess pore pressures are recorded not directly under- vertical effective stress in the free-field (,12 kPa at
neath the footing, but below the footing’s edge. Significant z ¼ 0 .75 m below foundation plan). On the contrary, zero
positive excess pore pressures were initially generated be- or negative excess pore pressures are generated in the
neath the central axis of the 130 kPa footing, while small proximity of the footing edges. Footing settlement starts
negative excess pore pressures were recorded below the as soon as excess pore pressures are generated in the
footing edge and at a distance of B/2 outside the footing foundation soil and reaches its maximum rate during this
(Figure 4(a)). Because of the pressure gradient generated as phase.
a consequence of the dis-homogeneous excess pore pressure, (b) Significant pressure gradient exists between the axis and
cross-drainage occurred during cyclic loading between dif- the edge of the footing. Cross-drainage towards the area
ferent areas of the foundation soil. All of the excess pore underneath the footing edge, and possibly dilative
pressures recorded show their maximum absolute value after behaviour in the foundation soil caused by the settling
15 to 20 s from the start of earthquake shaking; during this footing, act to reduce the excess pore pressure under the
time interval a maximum pressure difference of approxi- centre of the footing. During this phase the seismic
mately 55 kPa exists between PPT5 (footing axis) and PPT7 loading on the soil reduces because of excess-pore-
(B/2 metres outside footing). As a consequence of this pressure-induced soil softening. During this phase, the
gradient, drainage occurs from the area below the centre of rate of footing settlement also reduces.
the footing towards the soil below the footing edge, leading (c) Following cross-drainage, a more homogeneous excess
to an equalisation of excess pore pressure in the foundation pore pressure distribution in the foundation soil has been
soil (after 30 s for the 130 kPa footing in Fig. 4(a)). Similar reached. However, a steady and relatively slow increase
behaviour is observed underneath the 90 kPa footing; how- in the pore pressure is observed underneath the footing. A
ever, in this case the pressure gradients generated between possible explanation for this is the vertical dissipation of
the axis area and the edge area are smaller. Negative excess excess pore pressure from the underlying soil. It is
pore pressures were recorded only below the footing edge hypothesised that, in the short term, fluid would migrate
(PPT6), determining a less marked post-peak drainage effect toward areas characterised by higher effective stresses,
(Figure 4(b)). For the lighter, 60 kPa footing, this effect which can accommodate higher pore pressure with
seems to have been suppressed and only small differences in respect to the free field. Footing settlement rate further
measured excess pore pressures are observed. In particular reduces during this phase, reaching its final equilibrium.
no negative excess pore pressures were observed below the (d ) This phase corresponds to the post-shaking behaviour.
footing edge, resulting in a fairly homogeneous excess pore The drainage toward the foundation soil observed in
pressure response across the foundation soil (Figure 4(c)). phase (c) does not stop at the end of shaking, but rather a
93
PPT2 – footing axis (z 3 m)
PPT3 – footing edge (z 3 m)
50 PPT4 – footing axis (z 1·875 m)
PPT5 – footing axis (z 0·75 m)
40
PPT5 – footing edge (z 0·75 m)
Excess pore pressure: kPa
PPT5 – near footing (z 3 m)

30
20
(b)
10
0
(a) (c) (d)
10
0
Footing D
0·5
S: m
1·0
1·5
0·2 ACC11 (Footing axis; z 1·875 m)
a: g
0·2
0 10 20 30 40 50 60 70 80 90 100 110 120
Time: s
Fig. 5. During shaking Epp, settlement and horizontal acceleration measurements in the soil underneath the 130 kPa footing in test BD5
slight increase in its rate is observed, possibly due to the & Finn (1979) and Vaid & Chern (1983, 1985), who showed
ceasing of the cyclic dilative response associated with the the effect of static shear to be strongly dependent on the
co-seismic soil behaviour. Pore pressures in the free field initial confining stress and soil density. The results of these
seem unaffected from phase (c). Footing settlement studies show that the initial state variables considered (static
relative to the soil ceases with the end of shaking; shear, confining stress, relative density) control the cyclic
however, further vertical footing displacement occurs in behaviour of saturated sands, and that their effects are
the longer term due to the dissipation of excess pore mutually dependent. The soil relative density in particular
pressure associated with the post-liquefaction reconsoli- will determine the mechanism of strain development, strain
dation of sands (i.e. volumetric settlement). softening taking place in loose and strongly contractive soils
and ‘cyclic mobility’ (or limited strain liquefaction) in
These data therefore show that the main cause for limiting denser soils (Castro, 1975). Again, the threshold density
excess pore pressure generation is associated with the soil between these two mechanisms, other than on the soil type,
beneath the edge of the footing. As this is where initial depends on the confining stress and static shear acting on
static shear stresses are at a maximum, and as static shear the soil.
stresses have previously been identified with increased cyclic The addition of an initial shear stress in loose contractive
resistance under certain conditions, their potential correlation materials would therefore lower its resistance to liquefaction
with these results is examined further in the next section. as it moves it closer to the failure envelope. However, if
This mechanism has not been previously identified in similar such initial shear stress (�i) is higher than the cyclic shear
works (Whitman & Lambe, 1982; Liu & Dobry, 1997; stresses (�cyc) applied, than no stress reversal takes place in
Hausler, 2002; Dashti et al., 2010) as the instrument dis- the soil, significantly increasing its cyclic resistance (Vaid &
tribution was not targeted to pick up variations in pore Finn, 1979; Boulanger & Seed, 1995; Kammerer, 2002).
pressure within the space beneath the foundations. In order to adapt the empirical liquefaction triggering
curves used in current practice, which are based on level
ground conditions case histories, to sloping ground condi-
INFLUENCE OF INITIAL SHEAR STRESS tions (i.e. presence of static shear), Seed (1983) first pro-
The effect of static shear stress on the cyclic resistance to posed a correction factor KÆ dependent on the static shear.
liquefaction has been investigated by several authors since Several sets of curves relating KÆ values to static shear
the late 1960s. Most of the experimental work on this topic stress ratio (Æ ¼ �i/� v9 ) have subsequently been published
is based on soil element testing, in particular cyclic triaxial (Seed & Harder, 1990; Harder & Boulanger, 1997).
testing. Lee & Seed (1967) first hypothesised that the Similarly, an independent factor (K�) is used in current
presence of static shear increases the soil’s cyclic resistance practice to account for the influence of confining stress on
to liquefaction by performing a series of cyclic triaxial tests the soil liquefaction resistance. Based on the collective
on anisotropically consolidated samples. Subsequently, Cas- dependency of cyclic resistance on all initial state variables,
tro (1969, 1975) and Castro & Poulos (1977) verified that Vaid et al. (2001) question the use of independent factors in
the presence of static shear may result in a reduction of the order to account for confining stress and static shear. Results
cyclic resistance of sand. These apparently contradictory of a series of cyclic triaxial tests on Fraser River sand,
results were unified in a more complex framework by Vaid presented by the authors, show that the empirical method
94
currently used underestimates the cyclic resistance, the de- The mechanical parameters required by the model are shown
gree of conservatism being more dramatic for loose soils in Table 2. The computed stress distribution describes the
(Fig. 6). static case before earthquake shaking, and is shown in Fig. 7
(to the right of the axis of each footing in the figure).
During shaking the soil–structure interaction may result in a
Effect of initial shear stress in the centrifuge tests different, time-dependent, distribution of such stresses, re-
Despite significant differences between the loading and flecting the foundation tendency to rock and translate hor-
boundary conditions of an element test and those of an izontally, and thus cyclically extending both laterally and
element of soil in a centrifuge model (or in the field), these downward. Therefore, the computed stress distribution has to
findings can still be used to inform interpretation of the soil be considered only indicative. The typical Æ distribution
behaviour observed in the centrifuge tests. One of the main induced by a shallow foundation consists of ‘bulbs’ of high
differences in soil behaviour is that element tests are usually Æ beneath the footing edges, the extension of such bulbs
performed in fully undrained conditions, while results show being proportional to the foundation bearing pressure.
that partially drained conditions apply to centrifuge tests Also plotted on Fig. 7 (to the left of the axis of each
(Madabhushi & Haigh, 2012; Lakeland et al., 2014). Stress footing in the figure) are contours of the peak excess pore
distribution in the two cases may also vary significantly. In pressure before drainage, measured in the foundation soil
particular, in cyclic triaxial tests the soil is subjected to below the 60, 90 and 130 kPa footings, respectively. Con-
different admissible soil deformations and the total stress is tours are obtained by interpolating between data points using
kept constant, while in centrifuge tests it may vary during a linear interpolation scheme. Contours are plotted such that
cyclic loading. darker colours represent lower pore pressures.
The effect of static shear stresses on the cyclic response Comparison between excess pore pressures and static
of the soil depends on the ranges of state variables repre- shear stresses in Fig. 7 shows that the observed reduction in
sented in the centrifuge tests (� v9 varying between approxi- the generated excess pore pressure below the footing edge
mately 30 and 160 kPa between foundation plan and a depth corresponds, as suggested above, with areas characterised by
equal to B metres below foundation plan; and DR ¼ 40%). high values of Æ. In particular, the minimum excess pore
According to the KÆ values proposed by Seed & Harder pressures were recorded in portions of the foundation soil
(1990), the presence of the footing-induced static shear characterised by Æ . 0 .4 (Figs 7(a) and 7(b)). The heavier
stress within these ranges of state variables should result in footing (q ¼ 130 kPa) generates significant shear stresses in
a reduction of the foundation soil’s cyclic resistance (Fig. 6). a broad portion of the foundation soil. The ‘bulb’ of soil
On the contrary, in the centrifuge tests performed, reduced characterised by a value of Æ higher than 0 .4 extends to a
excess pore pressure generation was observed in soil asso- depth of , B (footing width) and has a maximum width of
ciated with high static shear stresses (i.e. increased cyclic , B/2 (Fig. 7(a)). Pore pressure measurements in correspon-
resistance). This is in accordance with the KÆ curves pro- dence with this area show no generated excess pore pres-
posed by Vaid et al. (2001) for relative density of 40% and sures. On the contrary, small negative peak excess pore
a similar initial confining stress. Results from Vaid et al. pressures have been recorded by PPT6 (footing edge) and
(2001) may be considered more reliable as, unlike the Seed PPT7 (near footing), possibly due to dilation induced by the
& Harder (1990) formulation, they account for the combined footing settlement. Similar observations have been made for
effect of confining stress and relative density. the 90 kPa footing; in this case, only records from PPT6
In order to identify the initial distribution of static shear (footing edge) show near-zero excess pore pressure, while
stress present in the centrifuge model, the contours of Æ significant excess pore pressure was measured by PPT7, in
along a longitudinal section of the model have been com- line with those recorded at the same depth in correspon-
puted based on a finite-element simulation performed using dence with the footing axis (Fig. 4). A possible explanation
the software Plaxis2D. The built-in elastic perfectly-plastic for these observations may be that the bulb of soil with
Mohr–Coulomb material model was used for this purpose. Æ . 0 .4 induced by the 90 kPa footing is smaller; as a
consequence the soil near PPT7 is subjected to lower shear
2·25 stresses, resulting in lower cyclic resistance (Fig. 7(b)).
Excess pore pressures generated at a depth of 0 .75 m
DR 55–70% underneath the lighter footing analysed (q ¼ 60 kPa) show
2·00
little or no horizontal variability from the footing axis to a
1·75
distance of B/2 from the footing edge (Fig. 7(c)). In
particular, the excess pore pressure generation pattern ob-
1·50
served is similar to what is observed in the free-field case
(i.e. �i ¼ 0), suggesting a reduced influence of the footing-
induced stresses for lower bearing pressure.
1·25
Kα
In all cases the highest excess pore pressure values were
DR 45–50%
recorded beneath the footing axis, where the shear stresses
1·00
are a minimum. Despite full liquefaction (i.e. ru ¼ 100 %)
being observed in the free field in all of the models tested,
0·75
Vaid et al. (2001)

0·50 Table 2. Plaxis 2D linear-elastic perfectly-plastic Mohr–Coulomb
DR 25% DR 35%
DR 35% model parameters (HST95 Congleton sand at DR 40%)
0·25 DR 40% Note:
Shaded areas from E: MPa 25
DR 60%
Seed & Harder (1990)
0 ı 0 .3
0 0·05 0·10 0·15 0·20 0·25 0·30 0·35 0·40 0·45 0·50 ªd: kN/m3 15 .49
α τi /σv,0 �crit: degrees 33
ł: degrees� 1
Fig. 6. KÆ plotted against Æ for Fraser River sand at different
density states and �9v of 100 kPa (Vaid et al., 2001) � Measured at D ¼ 40%.
R
95
Excess pore pressure: kPa
0 10 20 30 40
Footing D
t 16 s
q 130 kPa α τi /σv
0

1
0·5
2
0·4
0·3
3
0·1
0·2
4
0·1
5
5 4 3 2 1 0 1 2 3 4 5
(a)
t 18 s Footing C
q 90 kPa
0
1
0·5
0·4
2
0·3
0·2
3
0·1
4
α τi /σv
5
5 4 3 2 1 0 1 2 3 4 5
(b)
t 18 s Footing B
q 60 kPa
0
0·5
0·4
1
0·3
0·2
2
0·1
3 α τi /σv
5
5 4 3 2 1 0 1 2 3 4 5
Distance from foundation axis: m
(c)
Fig. 7. Contours of peak Epp before drainage against contours of footing-

induced static shear stress ratios: (a) 130 kPa footing; (b) 90 kPa footing;
(c) 60 kPa footing
96
all the PPTs positioned in the soil below or in the proximity pressure generation (Vaid & Finn, 1979; Boulanger & Seed,
of the footings recorded peak excess pore pressure ratios 1995; Kammerer, 2002). Shear stress reversal occurs when
(ru,max ¼ Eppmax/� v9 ) significantly lower than 100%. This is the oscillating earthquake-induced shear stress (�cyc) exceeds
plotted in Fig. 8, which shows the ru,max recorded beneath the the static shear stress (�i), crossing a state of zero shear and
footings against the estimated initial shear stress ratio at thus reversing direction. However, the magnitude of �i and
the measurement location. Also in Fig. 8, the results from �cyc is not sufficient information to evaluate the degree of
the centrifuge tests are compared to those from cyclic shear stress reversal, as the direction of such stresses has to be
tests performed by Kammerer (2002) and Boulanger et al. taken into consideration as well. Kammerer (2002) points
(1991) for different values of Æ. For higher Æ, all three sets out that, in the case of sloping ground, �i will be oriented
of data show a decrease of the recorded excess pore pressure parallel to the dip direction of the slope, while the direction
ratio for increasing initial static shear stress ratio. However, a of �cyc is variable depending on the ground motion. In this
substantial difference exists between centrifuge and cyclic scenario, the potential for stress reversal is higher when the
shear data for low values of initial static shear stress ratio, principal direction of seismic loading is parallel to the dip
which is ascribed to the different boundary conditions. direction of a slope. However, the author observes that
Theoretically, applying horizontal fixity and no-flow bound- significant excess pore pressures are generated in the soil
ary conditions to the soil column below the footing, if even when the direction of loading is parallel to the strike
liquefaction is triggered the entire weight of the footing direction of the slope. This is ascribed to the correspondent
would be transferred to the pore fluid, reaching a state of full complementary shear stresses acting perpendicularly to the
liquefaction. These idealised boundary conditions are unre- loading direction. Unlike those generated in slopes, the
presentative of reality where significant cross-drainage occurs direction of the maximum shear stresses induced by a
during shaking, resulting in peak ru values lower than unity. footing in the foundation soil changes with depth. This
A threshold value of Æ exists, above which excess pore increases the complexity in the evaluation of the effect of
pressure generation is impeded. Such a threshold value is not shear stress reversal on the excess pore pressure generation
unique but depends, other than soil type, on relative density and below shallow foundations.
confining stress. For the range of confining stress and relative Figure 9 compares the estimated stress paths in two soil
density considered in the models tested, Æ values higher than elements located at the same depth below the foundation
,0 .4 showed zero or negative excess pore pressure generation. plan (z ¼ 0 .75 m), the first corresponding to the central axis
Results from a series of cyclic bi-directional simple shear tests of the 130 kPa footing (PPT5) and the second beneath the
performed by Kammerer (2002) on Monterey #0/30 sand show footing edge (PPT6). Since no stress measurement is avail-
a similar threshold value (Fig. 8). able at the selected locations, the proposed stress paths
These observations suggest that, in all of the analysed represent an estimate based on the following assumptions.
cases, the foundation soil underwent excess pore pressure-
(a) The initial static shear stress (from static finite-element
induced softening without reaching a condition of full
simulation) is assumed to be maintained during the entire
liquefaction (i.e. ru ¼ 100%). Moreover, in the presence of
earthquake.
significant static shear stress, areas of non-softened soil may
(b) Total vertical stress variations in the soil during earth-
initially exist below the edges of the footing during earth-
quake shaking (˜�v) are accounted for based on the
quake shaking, even in the case where the motion is strong
measured vertical footing accelerations. A vertical
enough to trigger liquefaction in the free field.
‘piston-like’ footing movement is assumed and total
vertical stress variations in the soil calculated according
to Newton’s law and Boussinesq theory.
Effect of shear stress reversal
(c) The static (�i) and cyclic (�cyc) shear stresses are assumed
As mentioned above, many authors have stressed the
to act on the same plane (condition yielding maximum
influence of shear stress reversal on cyclic excess pore
expected amount of shear stress reversal).
Kammerer (2002) The proposed linear failure envelope (Fig. 9) corresponds to
Boulanger et al. (1991) a friction angle of ,328 (Lauder, 2010), yielding a critical Æ
130 kPa footing (centrifuge test BD5) value of 0 .61. Under these premises, the stresses in a soil
90 kPa footing (centrifuge test BD7) element at the selected locations can be estimated according to
60 kPa footing (centrifuge test BD8)
100 [M f av (t)]
˜� v (t) ¼ Bq ˜q(t) ¼ Bq
Maximum excess pore pressure ratio, ru,max: %
Af
80
� v9 (t) ¼ � v9,0 (t) þ ˜� v (t) � Epp
60
40 �cyc (t) ¼ ah (t) rd [� v9 (t) þ Epp(t)]
20 �(t) ¼ �i þ �cyc (t)

0 where Bq represents the ratio ˜�v /˜q evaluated at the
20 selected locations according to Boussinesq theory, ˜q is the
variation in footing bearing pressure due to the footing
40 vertical acceleration, av. ah is the horizontal acceleration
measured in the ground at 0 .75 m below the foundation
60
plane and rd is a stress reduction factor evaluated according
80 to the formulation of Idriss & Boulanger (2004).
0 0·05 0·10 0·15 0·20 0·25 0·30 0·35 0·40 0·45 0·50
Normalised initial shear stress, α τi /σv
According to the computed stress paths, no stress reversal
occurs below the footing edge, whereas beneath the footing axis
Fig. 8. Æ values at Epp measurement point against ru,max together shear stress reverses direction in most of the loading cycles. It is
with data from bi-directional direct shear tests from Kammerer also worthwhile to notice that the magnitude of the cyclic shear
(2002) and Boulanger et al. (1991) stress imposed to the foundation soil depends on the acting
97
60
50
40 PPT5 footing axis

(z 0·75 m)
30
20
10
τ: kPa
PPT6 footing edge

10 (z 0·75 m)
20
30
40 Pre-earthquake stresses

Stresses at peak Epp (t 16 s)
50 Stresses after cross-drainage (t 30 s)
Failure envelope
60
50 60 70 80 90 100 110 120 130 140 150 160
σv : kPa
Fig. 9. Approximate stress paths in the –9v plane, occurring during shaking beneath the footing’s
central axis (PPT5) and below the footing edge (PP6) (see Fig. 5(a))
vertical effective stress; therefore it is higher below the centre and below the footing edge suggests that the absence of
of the footing where v9 is maximum. shear stress reversal when initial shear stress is greater than
High i brings the soil’s stress state closer to failure, but cyclic shear stress (verified if i . cyc) may provide an
at the same time reduces its potential for softening as a explanation to the limited excess pore pressure generation
consequence of excess pore pressure generation. The stress observed below the edges of the footing.
path corresponding to PPT6, despite starting closer to the These results clarify the hypothesis derived previously
failure envelope than PPT5, shows no initial softening as from case study data that, although increasing the bearing
excess pore pressure generation is impeded. Instead a slight pressure of shallow foundations on liquefiable soil causes an
hardening behaviour is initially observed, possibly due to increase in likely settlement, this is only true up to a certain
dilation occurring in the soil and consequent development of stress level (here ,100 kPa), above which settlements may
negative excess pore pressure, increasing the v9 : However, no longer increase and may even reduce. It appears from the
significant softening occurs following cross-drainage, bring- testing performed that this is attributable to the initial shear
ing the soil state closer to failure (Fig. 9). On the contrary, stresses in the soil inhibiting pore pressure generation,
PPT5 shows significant softening taking place from the first apparently through a lack of shear stress reversal. Moreover,
cycles, reducing the acting v9 of about 35%. Cross-drainage the soil tendency to dilate at high stress ratio may also
in this case takes place from the footing axis toward the contribute to inhibiting pore pressure generation.
areas below the edge of the footing, resulting in a reduction The generality of the observed behaviour would be further
in excess pore pressure and hence in a regain of v9 (Fig. 8). tested by consideration of very wide footings, where the
areas of high initial shear stress are proportionally less than
the more uniformly loaded soil beneath the footing, and also
CONCLUSIONS by consideration of a wider range of input motions to
Excess pore pressure measurements from a series of three investigate the relationship between i and cyc, and whether
centrifuge tests showed that, in the presence of a footing this affects the bearing pressure at which peak settlement
resting directly on liquefiable soil, limited or no softening occurs.
occurred in the soil below the footing edge during earth-
quake shaking. In particular, these observations were seen to
be valid for footings exerting a relatively high bearing
pressure (i.e. . ,100 kPa) on the soil. Areas of the founda- NOTATION
tion soil which experienced least pore pressure increase were ah horizontal acceleration measured in ground at 0 .75 m
seen to correspond with those where the footing-induced below foundation plane
amax peak ground acceleration (g)
initial shear stresses i are a maximum. Partially drained
av footing vertical acceleration
loading conditions were observed in all of the centrifuge B building width (m)
tests performed, with significant drainage taking place during Bq ratio ˜v /˜q evaluated at selected locations according to
shaking to equalise the initial inhomogeneous excess pore Boussinesq theory
pressure generation. The comparison of the approximate DL liquefied soil thickness (m)
stress paths between soil beneath the footing’s central axis DR relative density (%)
98
E Young’s modulus (MPa) ings with shallow foundation on liquefiable soil. J. Geotech.
Epp excess pore pressure (kPa) Geoenviron. Engng, ASCE 136, No. 1, 151–164.
Eppmax peak excess pore pressure before cross-drainage (kPa) Harder, L. F. Jr & Boulanger, R. W. (1997). Application of K� and
Epp0 uncorrected excess pore pressure (kPa) KÆ correction factors. Proceedings of the NCEER workshop on
KÆ static shear correction factor evaluation of liquefaction resistance of soils, report NCEER-97-
K� overburden correction factor 0022, pp. 167-190. Buffalo, NY, USA: National Center for Earth-
q foundation bearing pressure (kPa) quake Engineering Research, State University of New York.
rd stress reduction factor evaluated according to formulation Hausler, E. A. (2002). Influence of ground improvement on settle-
of Idriss & Boulanger (2004) ment and liquefaction: a study based on field case history
ru excess pore pressure ratio, Epp/� v9 (%) evidence and dynamic geotechnical centrifuge tests. PhD thesis,
ru,max peak excess pore pressure ratio before cross-drainage, University of California, Berkeley, CA, USA.
Eppmax/� v9,0 (%) Idriss, I. M. & Boulanger, R. W. (2004). Semi empirical procedures
Sav average liquefaction induced settlement (m) for evaluating liquefaction potential during earthquakes. In
Sav(t ) settlement at time t Proceedings of the 11th international conference on soil dy-
Sav,final final settlement namics and earthquake engineering, and 3rd international con-
� Poisson ratio ference on earthquake geotechnical engineering (eds D. Doolin,
z depth from foundation plane (m) A. Kammerer, T. Nogami, R. B. Seed and I. Towhata), pp. 33–
z0 initial instrument depth (m) 56. Singapore: Stallion Press.
Æ static shear stress ratio (or normalised initial shear stress), Kammerer, A. M. (2002). Undrained response of Monterrey 0/30
�i/� v9,0 sand under multidirectional cyclic simple shear loading condi-
ªd dry unit weight (kN/m3) tions. PhD thesis, University of California, Berkeley, CA, USA.
ªw unit weight of water Lakeland, D. L., Rechenmacher, A. & Ghanem, R. (2014). Towards
˜q variation in footing bearing pressure due to footing vertical a complete model of soil liquefaction: the importance of fluid
acceleration, av flow and grain motion. Proc. A, R. Soc. 470, No. 2165,
� v9 vertical effective stress (kPa) 20130453, http://dx.doi.org/10 .1098/rspa.2013 .0453.
� v9,0 initial vertical effective stress (kPa) Lauder, K. D. (2010). The performance of pipeline ploughs. PhD
�cyc cyclic shear stress (kPa) thesis, University of Dundee, UK.
�i initial (or static) shear stress (kPa) Lee, K. L. & Seed, H. B. (1967). Dynamic strength of anisotropi-
�xz shear stress acting on the x–z plane (�i + �cyc) (kPa) cally consolidated sand. J. Soil Mech. Found. Div., ASCE 93,
�crit friction angle at critical state (degrees) No. SM5, 169–190.
�P peak friction angle (degrees) Liu, L. & Dobry, R. (1997). Seismic response of shallow foundation
�R residual friction angle (degrees) on liquefiable sand. J. Geotech. Geoenviron. Engng, ASCE 123,
ł dilation angle (degrees) No. 6, 557–567.
Madabhushi, S. P. G. & Haigh, S. K. (2012). How well do we
understand liquefaction. Indian Geotech. J. 42, No. 3, 150–160.
Muir Wood, D. (2004). Geotechnical modelling. London, UK: Spon
REFERENCES Press, Taylor and Francis.
Acacio, A., Kobayashi, Y., Towhata, I., Bautista, R. T. & Ishihara, Schofield, A. N. (1980). Cambridge geotechnical centrifuge opera-
K. (2001). Subsidence of building foundation resting upon tions. Géotechnique 30, No. 3, 227–268, http://dx.doi.org/
liquefied subsoil: case studies and assessment. Soils Found. 41, 10 .1680/geot.1980 .30 .3 .227.
No. 6, 111–128. Seed, H. B. (1983). Earthquake resistant design of dams. In Seismic
Adachi, T., Iwai, S., Yasui, M. & Sato, Y. (1992). Settlement and design of embankments and caverns (ed. T. R. Howard), pp. 41–
inclination of reinforced-concrete buildings in Dagupan-City due 64. New York, NY, USA: American Society of Civil Engineers.
to liquefaction during the 1990 Philippine earthquake. Proceed- Seed, R. B. & Harder, L. F. (1990). SPT-based analysis of cyclic
ings of the 10th world conference on earthquake engineering, pore pressure generation and undrained residual strength. In
vol. 1, pp. 147–152. Rotterdam, the Netherlands: Balkema. Proceedings of the Seed memorial symposium (ed. J. M. Dun-
Bertalot, D. (2012). Behaviour of shallow foundations on layered can), pp. 351–376. Vancouver, BC, Canada: BiTech Publishers.
soil deposits containing loose saturated sands during earth- Steedman, R. S., Ledbetter, R. H. & Hynes, M. E. (2000). The
quakes. PhD thesis, University of Dundee, UK. influence of high confining stress on the cyclic behavior of
Bertalot, D., Brennan, A. J. & Villalobos, F. (2013). Influence of saturated sand, ASCE Geotechnical Special Publication No.
bearing pressure on liquefaction-induced settlement of shallow 107, pp. 35–57. Reston, VA, USA: American Society of Civil
foundations. Géotechnique 63, No. 5, 391–399, http://dx.doi.org/ Engineers.
10 .1680/geot.11.P.040. Stewart, D. P., Chen, Y. R. & Kutter, B. L. (1998). Experience with
Boulanger, R. W. & Seed, R. B. (1995). Liquefaction of sand under the use of methylcellulose as a viscous pore fluid in centrifuge
bi-directional monotonic and cyclic loading. J. Geotech. Engng, models. Geotech. Testing J. 21, No. 4, 365–369.
ASCE 121, No. 12, 870–878. Vaid, Y. P. & Chern, J. C. (1983). Effect of static shear on
Boulanger, R. W., Seed, R. B., Chan, C. K., Seed, H. B. & Sousa, resistance to liquefaction. Soils Found. 23, No. 1, 43–60.
J. (1991). Liquefaction behaviour of saturated sands under Vaid, Y. P. & Chern, J. C. (1985). Cyclic and monotonic undrained
unidirectional and bi-directional monotonic and cyclic simple response of sands. Advances in the art of testing soils under
shear loading, Geotechnical Engineering Report No. UCB/GT/ cyclic loading conditions. In Advances in the art of testing soils
91-08. Berkeley, CA, USA: University of California. under cyclic conditions (ed. V. Khosla), pp. 171–176. New York,
Brennan, A. J., Knappett, J. A., Bertalot, D., Loli, M., Anastasopou- NY, USA: American Society of Civil Engineers.
los, I. & Brown, M. J. (2014). Dynamic centrifuge modelling Vaid, Y. P. & Finn, W. D. L. (1979). Static shear and liquefaction
facilities at the University of Dundee and their application to potential. J. Geotech. Engng Div., ASCE 105, No. GT10, 1233–
studying seismic case histories. In ICPMG2014 – physical 1246.
modelling in geotechnics (eds C. Gaudin and D. White), pp. Vaid, Y. P., Stedman, J. D. & Sivathayalan, S. (2001). Confining
227–234. Boca Raton, FL, USA: CRC Press. stress and static shear effects in cyclic liquefaction. Can.
Castro, G. (1969). Liquefaction of sands. PhD thesis, Harvard Geotech. J. 38, No. 3, 580–591.
University, Cambridge, MA, USA. Whitman, R. V. & Lambe, P. C. (1982). Liquefaction: consequences
Castro, G. (1975). Liquefaction and cyclic mobility of saturated for a structure. In Soil dynamics and earthquake engineering:
sands. J. Geotech. Engng, ASCE 101, No. 6, 551–569. proceedings of the conference on soil dynamics and earthquake
Castro, G. & Poulos, S. J. (1977). Factors affecting liquefaction and engineering (eds A. S. Cakmak, A. M. Abdel-Ghaffar and C. A.
cyclic mobility. J. Geotech. Engng Div., ASCE 103, No. GT6, Brebbia), vol. 2, pp. 941–949. Rotterdam, the Netherlands:
501–516. Balkema.
Dashti, S., Bray, J. D., Pestana, J. M., Riemer, M. & Wilson, D. Yoshimi, Y. & Tokimatsu, K. (1977). Settlement of buildings on
(2010). Mechanism of seismically induced settlement of build- saturated sands during earthquakes. Soils Found. 17, No. 1, 23–28.
99
Knappett, J. A. et al. (2015). Géotechnique 65, No. 5, 429–441 [http://dx.doi.org/10.1680/geot.SIP.14.P.059]
Seismic structure–soil–structure interaction between pairs of adjacent

building structures
J. A . K NA P P E T T , P. M A D D E N a n d K . C AU C I S †
Structure–soil–structure interaction between adjacent structures, which may occur in densely populated
urban areas, has received little attention compared to the soil–structure interaction of single isolated
structures. Additionally, recent earthquakes in/near such areas (e.g. the Christchurch series, 2010–
2011) have shown that large motions can be followed by strong aftershocks. In this paper, the seismic
behaviour of isolated structures and pairs of adjacent structures under a sequence of strong ground
motions has been investigated using a combination of centrifuge and finite-element modelling. The
latter utilised an advanced constitutive model that can be parameterised from routine test data, making
it suitable for use in routine design. The finite-element models were shown to accurately simulate the
centrifuge-measured response (in terms of surface ground motion and structural sway, settlement and
rotation) even after multiple strong aftershocks, so long as the buildings’ initial conditions were
reproduced accurately. For the case of a building structure with a close neighbour, structural drift and
co-seismic settlement could be reduced or increased as a result of structure–soil–structure interaction,
depending chiefly on the properties of the adjacent structure. This suggests that careful arrangement
of adjacent structures and specification of their properties could be used to control the effects of
structure–soil–structure interaction. In all cases where adjacent structures were present, permanent
rotation (structural tilt) was observed to increase significantly, demonstrating the importance of
considering structure–soil–structure interaction in assessing the seismic performance of structures.
KEYWORDS: centrifuge modelling; earthquakes; numerical modelling; sands
INTRODUCTION they are limited in that the soil–structure interaction is

Current seismic design of building structures considers the always linear elastic. This may be an acceptable assumption
response of a building in isolation from its neighbours. in a very small earthquake, but when the ground motions
Many of the most damaging earthquakes over the last 20 become large (such as in the strong recent earthquakes
years, however, have struck heavily populated and highly mentioned previously) the soil would be expected to be
urbanised areas, including those in Kobe (1995), Kocaeli highly non-linear with significant plastic strains. Under these
(1999), Athens (1999), Wenchuan (2008) and the Christ- circumstances, not only would the effects of the SSSI on the
church series (2010–2011). Although damage may be ex- dynamic structural response (e.g. inter-storey drift or spectral
pected to be high in these areas as there are many more acceleration) be expected to change, but there may also be
‘targets’ for the earthquake, the close spacing of the building significant permanent settlement and rotation of the struc-
stock will result in interaction between adjacent structures tures, which could be as damaging as the structural motions.
through the ground, a phenomenon which is here termed The ability to capture the permanent behaviour is particu-
structure–soil–structure interaction (SSSI). There have been larly important considering that many of the earthquakes
few previous attempts to study SSSI and these have often mentioned previously were associated with strong after-
been highly simplified; however, it may be expected that, shocks, the most notable recent example being the Darfield
depending on the layout of adjacent buildings and their (2010) and Christchurch (2011) earthquakes, which occurred
relative dynamic properties, the effects of SSSI may have less than 6 months apart, that is, before a substantial amount
either a beneficial or detrimental effect on the overall of repair could be conducted on damaged structures follow-
response of the structures, through changes in the local soil– ing the first earthquake. Permanent soil deformations will
structure interaction. change the behaviour of the underlying soil. Settlement may
Previous studies of SSSI have generally focused on under- lead to soil stiffening as a result of densification, making it
standing changes in dynamic characteristics of adjacent rigid potentially better able to transmit ground motions into the
blocks (e.g. Tsogka & Wirgin, 2003) or simple oscillators structure (and thereby potentially increasing the structural
(Alexander et al., 2013), in each case on a linear elastic response). A bias in permanent rotation may be amplified in
medium as a representation of the soil. Both shallow (Betti, subsequent earthquakes owing to P–˜ effects. Current linear
1997) and deep foundations (Padrón et al., 2009) have elastic approaches cannot incorporate such effects.
previously been considered. Although these studies have This paper, therefore, aims to study SSSI utilising meth-
provided useful information in terms of changes in funda- ods that can incorporate non-linear elasto-plastic soil behav-
mental periods of vibration and elastic spectral response, iour. First, dynamic centrifuge modelling is used to create a
database of physical test data of pairs of adjacent low-rise
simple structures having shallow foundations on sand, con-
Manuscript received 4 April 2014; revised manuscript accepted 10
February 2015. Published online ahead of print 13 April 2015.
sidering situations (a) where the structures have similar
Discussion on this paper closes on 1 October 2015, for further details properties and (b) where the properties are dissimilar. Tests
see p. ii. of the structures in isolation are also performed for compari-
University of Dundee, Dundee, UK. son. In all cases, series of strong motions are applied to the
† Arup, Edinburgh, UK; formerly University of Dundee, Dundee, UK. models as an idealised representation of a sequence of
101
KNAPPETT, MADDEN AND CAUCIS
100
strong aftershocks. Non-linear finite-element modelling is
then used to simulate these tests and is validated against the
centrifuge test data. This will utilise the constitutive model-
ling approach outlined in Al-Defae et al. (2013), which was
shown to closely simulate the dynamic constitutive behav- 24
iour of soil identical to that used herein at similar relative
density in sequences of strong earthquakes for slope models 6
Aluminium
atop a level bearing layer. The influence of modelling alloy Steel 60
assumptions will also be discussed. Finally, the finite-ele- 60
3
Steel
ment approach is used to gain insight into the influence of 6
building arrangement and aftershocks on the seismic re-
sponse of pairs of adjacent structures. 12 12
40 40 40
CENTRIFUGE MODELLING (a)
All of the tests were conducted using a model scale of 100
1:50 and tested at 50g using the 3 .5 m radius beam centri-
fuge at University of Dundee, UK. All subsequent param-
eters are given at prototype scale, unless otherwise stated.
Scaling factors for centrifuge modelling can be found in
Muir Wood (2004). 48
Model structures
Two different structural models were produced that were 6
designed to represent some of the key characteristics of low-
rise buildings. Such buildings will generally form a much
larger proportion of the building stock within an urban area
compared to high-value, high-rise structures and are also less
likely to have undergone detailed seismic design. They may 140
therefore be more vulnerable and contribute more signifi- 138 6
cantly to the overall cost of seismic damage following a
major earthquake.
Each structure was modelled as a single-bay, single-
degree-of-freedom (SDOF) sway frame on separated strip
foundations, 100 mm long and 40 mm wide at model scale
(5 m 3 B ¼ 2 m at prototype scale). Steel mass plates were
used to represent the dynamic mass of the structure, and 6
vertical aluminium alloy plates, the same length as the 12 12
foundations, were used to represent the sway stiffness of the
structures (that would be provided by the columns). The 40 40 40
centre-to-centre spacing between the foundations (s) was (b)
kept the same between the models, but the height was varied
to change the position of the centre of mass above the soil Fig. 1. Single-degree-of-freedom model structures: (a) ‘short’
surface and therefore alter the proportion of rocking to sway period structure; (b) ‘long’ period structure. All dimensions at
model scale in mm
deformation between the two structures. The model struc-
tures are shown in elevation with model scale dimensions
(in mm) in Fig. 1. The fundamental natural period (Tn0) of a
building structure is typically related to its height, with taller founding plane). From equation (1), Tn0 ¼ 0 .33 s and 0 .65 s
buildings being laterally more flexible per unit mass and for the ‘short’ and ‘long’ period structures, respectively. The
hence having higher Tn0 (e.g. Goel & Chopra, 1997; BSI, characteristics of the models are compared to the measured
2005). To design the model structures, the relationship given properties of a variety of real structures (after Goel &
in Eurocode 8 (BSI, 2005) was used for building structures Chopra (1997) and Stewart et al. (1999)) in Fig. 2.
less than 40 m high The supported mass and stiffness of the plates represent-
. ing the columns were then selected to match these values of
T n0 ¼ C t H 0 75 (1) Tn0. The mass at the top of each structure (Meq) was selected
where H is the overall height of the building and Ct ¼ 0 .085 first, which fixed the value of the bearing pressure (q) for a
(as an approximation for a steel moment resisting frame given static factor of safety (FSv). The thickness of the
(MRF)). In each case, the structures are assumed to be vertical plates was then selected to provide a bending
SDOF idealisations of multi-storey structures that have a stiffness, EI (and therefore lateral sway stiffness, Keq) such
uniform distribution of mass and stiffness with height. The that the required Tn0 was achieved for the mass selected,
centre of mass of such structures would be approximately at using
the mid-height, and so in order to ensure the correct amount sffiffiffiffiffiffiffiffi
of overturning moment for a given value of H, the structure M eq
T n0 ¼ 2 (2)
with the shorter period (hereafter termed ‘short structure’) K eq
represents a prototype structure H ¼ 6 m high (centre of
mass 3 m above founding plane). The structure with the A summary of the properties of the two structures at
longer period (hereafter termed ‘long structure’) represents a prototype scale is provided in Table 1. This table includes
structure H ¼ 15 m high (centre of mass 7 .5 m above the an estimate of FSv when surface bearing on a uniform
102
SEISMIC STRUCTURE–SOIL–STRUCTURE INTERACTION BETWEEN BUILDING STRUCTURES
3
in the following section). Tests PM005 and PM006 tested
Field data (Stewart et al., 1999)
the same adjacent pair of two long structures, but with
Short structure different edge-to-edge (inter-building) spacing between the
Fundamental natural period, Tn0: s
Long structure structures; PM008 was comparable to PM006 in terms of

Eurocode 8 (Ct 0·085)
spacing, but tested the two short structures as an adjacent
2 pair. Finally, tests PM009 and PM011 tested pairs of one
short and one long structure; by reversing the order of the
structures in the box, the effect of earthquake direction
relative to the arrangement of structures could be considered,
as the ground motions were always applied in the same
1 direction. Spacing between structures in the adjacent cases
was kept to a minimum to consider the condition where
SSSI effects will likely be most significant (assuming that
these reduce as spacing increases). A summary of the test
configurations is given in Table 2.
0
0 10 20 30 40
Building height, H: m Model preparation and soil properties
As the tests reported here focus on the effects of structur-
Fig. 2. Fundamental period of model structures and comparison
to field data al properties and arrangement on SSSI, a single set of soil
properties was used in all of the tests. Dry HST95 Congle-
ton silica sand was air-pluviated into an equivalent shear
Table 1. Properties of model structures (prototype scale) beam (ESB) container to a target relative density of
Dr ¼ 55–60% (the range accounts for the accuracy with
Parameter: units ‘Short’ structure ‘Long’ structure which this property can be replicated and measured within a
model soil bed) to produce a uniform layer with a prototype
Total height, H: m 6 15 thickness of 10 m. The measured relative density achieved in
Height to C-of-M: m 3 7 .5 each test is shown in Table 2. The design and performance
Natural period, Tn0: s 0 .33 0 .65
of the ESB container is described by Bertalot (2012) and the
Bearing pressure, q: kPa 161 276
Static factor of safety, FSv 9 .5 5 .5 basic soil properties for HST95 are given in Table 3 after
Meq: t 235 469 Lauder (2011). As the rigid base of the container lay below
Keq: MN/m 87 .2 44 .1 the layer of sand, the ground profile represents ground type
Vertical plate EI: MN m2/m 19 .6 155 .1 E according to Eurocode 8 (BSI, 2005).
Footing spacing, s: m 4 4 During pluviation, the soil was instrumented with type
ADXL78 MEMS accelerometers (� 70g range) manufac-
tured by Analog Devices, as shown in Fig. 3 (PM003 and
PM006 are shown as examples). These were used to measure
deposit of dry sand as used in the centrifuge tests (relative the input motion (point E), free-field ground motion 1 m
density Dr ¼ 58%, unit weight ª ¼ 16 .2 kN/m3 and peak below the ground surface (point F) and accelerations beneath
friction angle �9p ¼ 408), which was determined using (D) and between (G) the structures at the same depth as the
0.5ªBsª N ª
FSv ¼ (3)
q
Table 3. State-independent physical properties of HST95 silica
sand (after Lauder, 2011)
In determining FSv, Nª and sª were calculated for
�9p ¼ 408 following Salgado (2008) and Lyamin et al. Property Value
(2007), respectively, although it should be noted that almost
identical values are found using the relationships provided in Specific gravity, Gs 2 .63
Eurocode 7 (BSI, 2004). The same foundation type and size D10: mm 0 .09
was used for both structures, which resulted in a higher D30: mm 0 .12
value of FSv for the short structure due to its lower applied D60: mm 0 .17
Cu 1 .9
bearing pressure. Cz 1 .06
These model structures were used in a total of seven Maximum void ratio, emax 0 .769
centrifuge tests. Two of these, PM003 and PM004, tested the Minimum void ratio, emin 0 .467
single structures in isolation (ground properties are described
Table 2. Centrifuge testing programme
Test ID Configuration Structure Structure Inter-building Relative

type (left) type (right) spacing: m density: %
PM003 Isolated Short N/A 58

PM004 Isolated Long N/A 59
PM005 Adjacent, similar Long Long 2 57
PM006 Adjacent, similar Long Long 1 58
PM008 Adjacent, similar Short Short 1 58
PM009 Adjacent, dissimilar Long Short 1 59
PM011 Adjacent, dissimilar Short Long 1 56
103
Direction of shaking Initial tilt: degrees
Accelerometer 40
[A] LVDT PM003
20
[B] [C] 1·5 1·0 0·5 0 0·5 1·0 1·5 PM004
0
PM006
Initial settlement: mm
[D] 1 (20) [F]
Medium dense sand
20 PM008
Dr 55–60%
PM009 (short)
40
10 (200)
PM009 (long)
60 PM011 (short)
80 PM011 (long)
[E] 1 (20)
100 1/250
33·5 (670) 1/50
120
(a)
140
[A] [K] Fig. 4. Initial conditions of model structures (settlement and
Direction of shaking rotation) measured following centrifuge spin-up to 50g and prior
Accelerometer to earthquake shaking
[B] [C][H] [I] LVDT
1 (20)
was band-pass filtered between 0 .8 Hz and 8 Hz (40–400 Hz
Medium dense sand [G] [F]
Dr 55–60%
at model scale) using a zero-phase-shift digital filter to
remove components of the signal that were outside the range
10 (200)
1 (20)
that can be accurately controlled by the EQS. The time
history of this demand motion, normalised by peak accelera-
[E] 1 (20)
tion, is shown in Fig. 5(a). The Kobe earthquake was known
to be particularly damaging to infrastructure, and the motion
33·5 (670)
selected has a number of repetitive acceleration peaks close
(b) to the peak ground acceleration (ag), for example, between 9
and 14 s in Fig. 5(a). In each test a 0 .1g motion was initially
Fig. 3. Layout of centrifuge tests: (a) isolated structure case applied to the model to characterise the dynamic behaviour
(PM003 shown); (b) adjacent structure case (PM006 shown). All of the system when the soil strains were small (negligible
dimensions at prototype scale in m (model scale in mm) permanent settlement and rotation occurred during these
motions). Subsequently, three nominally identical 0 .5g mo-
tions were applied, representing a strong earthquake and two
free-field instrument, to examine near-field effects of SSSI strong aftershocks. The initial 0 .1g motion could also be
on the ground motion. Identical accelerometers were also
attached to the structures, as shown in Fig. 3. These were
1
used to measure the dynamic motion of the foundations and
equivalent mass. Through high-pass filtering and integration,
Normalised acceleration
the dynamic displacements were determined at these loca-

tions with rotational bias removed, and the difference be-
tween them represented the dynamic inter-storey drift. 0
The container and model soil beds were loaded onto the 0 10 20 30 40
centrifuge and the structures were placed on the surface after Time: s
loading using a post-level to place them as accurately as
possible. In all cases the structures occupied a maximum of
the central 33% of the ESB (Fig. 3(b)) to minimise any 1
potential boundary effects within the container. An overhead (a)
gantry was then placed above the structures allowing linear 4
variable differential transformers (LVDTs) to be placed as 0·5g
shown in Fig. 3 to measure average settlement and global
Normalised spectral acceleration, Se/ag
0·1g
rotation (tilt) of the structures. During spin-up of the centri- Eurocode 8 (ground type A)
fuge the response of the LVDTs was recorded such that the 3
initial settlement and tilt of the structures, prior to earth-
quake shaking, were known. These initial conditions are
presented in Fig. 4.
2
Dynamic excitation
Following spin-up, a sequence of strong ground motions 1
was applied to each of the models using the Actidyn QS67-2
servo-hydraulic earthquake simulator (EQS), the performance
of which is detailed in Bertalot et al. (2012) and Brennan et
al. (2014). In each case, a horizontal motion recorded at the 0
Nishi-Akashi recording station in the Mw ¼ 6 .9 Kobe earth- 0 1 2 3 4
quake (1995) was used. This record was downloaded from Period: s
(b)
the PEER (Pacific Earthquake Engineering Research) NGA
database and, unscaled, had a peak acceleration of Fig. 5. Input ground notion from Nishi-Akashi recording station
ag ¼ 0 .43g. For the purpose of these tests, the motion was (Kobe in 1995): (a) normalised time history; (b) response spectra
rescaled to 0 .1g and 0 .5g nominal peak accelerations and for nominal 5% damping
104
10
considered as a small pre-shock prior to the 0 .5g main
shock. A final 0 .1g motion was then applied to provide a re- Finite-element model, equation (4)
characterisation of the behaviour at smaller strains following Short structure
8
Structural viscous damping ratio: %

the substantial changes imparted to the soil fabric by the
preceding motions. The use of the same underlying record to Long structure
represent all ground motions applied was an idealisation of Typical steel MRF
the earthquakes being generated by the same source (and 6
therefore having similar characteristics). The repeatability of
the motions is demonstrated in Fig. 5(b) in terms of the
nominal 5% damped response spectrum of the actual re- 4
corded motion at point E in Fig. 3, normalised by ag (12
0 .5g motions and four 0 .1g motions are shown). As this
represents the ‘bedrock’ input motion, the design spectrum
from Eurocode 8 for such material (ground type A) is also 2
shown in Fig. 5(b) for context (BSI, 2005).
Range of input motion
0
FINITE-ELEMENT MODELLING 0 2 4 6 8 10
Frequency: Hz
As the centrifuge model structures and foundations were
long in the direction perpendicular to shaking, finite-element Fig. 6. Modelling of structural damping in finite-element simula-
simulations of all of the tests were conducted in plane strain tions
using Plaxis 2D 2012. Compared to the centrifuge model
shown in Fig. 3, the dimensions of the model domain were
extended laterally to 100 m and combined with non-reflecting incorporates non-linear elastic behaviour which is dependent
boundary elements controlling the dynamic stresses along the on both confining stress level and induced strain, with
vertical boundaries (after Lysmer & Kuhlmeyer, 1969) to Mohr–Coulomb plasticity having isotropic hardening. To
represent semi-infinite soil conditions; that is, boundary examine the influence of soil parameter correlations, two
deformations at the location of the centrifuge container wall different sets of parameters were used in simulations of the
which are controlled by the dynamic deformation of the centrifuge tests. The first, by Brinkgreve et al. (2010), is
adjacent soil. This boundary condition can also be modelled non-soil specific, having been developed based on fitting to
by horizontal node-to-node ties between the two vertical a database of historical element test data for different sands.
boundaries of a model the width of the soil tested in the This only requires relative density as an input parameter
centrifuge. Compared to this alternative, the method used has from which all of the constitutive parameters are obtained,
a higher element requirement for the same mesh density, but potentially allowing for complete parameterisation from rou-
allowed future extension of the model beyond the two tine in-situ tests that can be used to estimate relative density,
structures considered here (although the extended results are such as the standard penetration test (SPT) or cone penetra-
not reported in this paper). This same approach has been tion test (CPT) if further laboratory testing data were not
used previously in modelling the behaviour of slopes during available. The second set of parameters used was specifically
a sequence of strong aftershocks in the same soil and model calibrated for the HST95 sand used in the centrifuge tests
container, as described by Al-Defae et al. (2013). through additional (routine) soil element testing, including
The equivalent mass and vertical plates of the structures direct shear tests and oedometric compression tests of the
were modelled numerically using elastic plate elements hav- sand across a wide range of relative densities. A complete
ing the same bending stiffness per metre length and mass as description of this set of parameters and the methods used
the centrifuge models (Table 1). The footings were modelled to find them is given in Al-Defae et al. (2013). This second
as an elastic continuum with Young’s modulus of 210 GPa, set of parameters also uses relative density as the input
Poisson ratio of 0 .3 and unit weight of 76 .5 kN/m3 to match parameter. A summary of both sets of correlations is pro-
the properties of the steel footings in the centrifuge tests. vided in Table 4. For the simulation of a particular centri-
Damping in each structure was modelled using Rayleigh’s fuge test, the actual relative density from Table 2 was used
approach to determine the equivalent viscous damping () to obtain the constitutive parameters. Previous finite-element
modelling of the behaviour of the test soil at a similar
1
¼ cm þ ck ( f n ) (4) density in the same ESB container at 50g by Al-Defae et al.
4 f n (2013) has shown that some additional Rayleigh damping
was required in addition to the implicit hysteretic damping
where fn are natural frequencies (of the structures). The included in the constitutive model, to control higher fre-
mass- and stiffness-proportional coefficients were found by quency components of the deformation and replicate dy-
fitting equation (4) to measured damping of the model namic accelerations accurately within the soil. Therefore, the
structures determined using the logarithmic decrement meth- same mass- and stiffness-proportional Rayleigh parameters
od applied to impulse test data, and simultaneously ensuring from this previous study were used for the soil in all
that the relationship is relatively flat across the full range of simulations presented here, namely, cm ¼ 0 .0005 and
input motion frequencies. This is shown in Fig. 6, where ck ¼ 0 .005.
mean values are shown, along with error bars representing Simulations using the Brinkgreve et al. (2010) correlations
the maximum and minimum range of the experimental represent those that could be achieved without doing an
impulse test data. From this figure, cm ¼ 0 .4 and ck ¼ 0 .001 extensive amount of site investigation; those using the Al-
were selected for use in both structures. It should also be Defae et al. (2013) correlations would potentially allow
noted that both models represent the damping of typical improved predictions at the cost of performing an additional
steel structures (,2%) well, in addition to the natural period soil-specific calibration. Their comparison later in the paper
matching shown in Fig. 2. will demonstrate how important such a calibration is to the
The soil was modelled using the ‘Hardening soil model accurate prediction of the dynamic soil and structural re-
with small-strain stiffness’ (Benz, 2006). This soil model sponse. In each case, the simulations were conducted in two
105
Table 4. Constitutive model parameters for finite-element modelling
Parameter Brinkgreve et al. (2010) HST 95 (Al-Defae et al., 2013) Units
�9p 12.5Dr þ 28 20Dr þ 29 degrees

c9 0 0 kPa
ł9 12.5Dr � 2 25Dr � 4 degrees
ref
Eoed 60Dr 25Dr þ 20.22 MPa
E50ref
Eref
oed 1.25Eref
oed MPa
Eurref 3Eref
oed 3Eref
oed MPa
�ur 0.2 0.2 –
G0ref 68Dr þ 60.00 50Dr þ 88.80 MPa
�s,0 .7 2 � Dr (310�4 ) 1.7Dr þ 0.67(310�4 ) –
Rf 1 � 0.13Dr 0.9 –
m 0.7 � 0.31Dr 0.6 � 0.1Dr –
ª 4Dr þ 15.0 3Dr þ 14.5 kN/m3
Note: All reference stiffness parameters (indicated by superscript ‘ref’) are defined at
p9 ¼ 100 kPa; the hardening soil model subsequently adjusts the stiffness as the confining
stress changes within the continuum.
stages: Following initial stress generation to K0 conditions An example of the sensitivity of the rotation of isolated
(where K0 ¼ 1 � sin �9p ), the structure is ‘built’ during a structures to initial conditions is shown in Fig. 8, where the
static stage; that is, the weight of the structure(s) is used to three large earthquakes for test PM004 are shown, along
load the soil, obtain the initial stress and deformation fields with the input and free-field surface time histories of ground
within the model and achieve static equilibrium. Once this is acceleration. It can be seen that if the initial rotation is
complete the second stage applies a dynamic input motion matched, the finite-element model produces a very good
to the bottom of the model, in each case matching that prediction of the development of rotation throughout the first
recorded in the corresponding centrifuge test. The motions large shock and the subsequent large aftershocks. The pre-
were input as ground displacement histories, determined by dicted rotations are completely different using the idealised
high-pass filtering and integration of the accelerometer re- initial rotation condition. This sensitivity is exacerbated
cords; filtering before integration to obtain velocity, and when adjacent structures are considered. Fig. 9 shows a
again, before integrating velocity to obtain displacement comparison similar to Fig. 8 but for test PM006. In this
ensured that there was no permanent ‘wander’ due to any particular case, the rotation of the right-hand structure is
offset in the accelerometer recordings or integration of matched well in the first motion, but the left-hand structure
random noise within the signal. Dynamic displacement data is not (likely due to a local heterogeneity in this particular
were then obtained at points in the finite-element model centrifuge test). As soon as the rotations begin to deviate
which matched those in the centrifuge tests, as shown in from the centrifuge result, the match in subsequent strong
Fig. 7 (compare to Fig. 3). Where acceleration data were aftershocks cannot be good, as the initial conditions in the
required, these were obtained by numerically double differ- subsequent earthquakes will be different. This does not
entiating the appropriate displacement–time history. imply that the finite-element model is wrong per se, just that
For each of the simulations using the different parameter the assumption of a uniform deposit of soil cannot model
sets, the initial conditions were first determined direct from the highly subtle variations in the real soil of the centrifuge
the initial static stage – this represents ‘ideal’ soil conditions. test, and that the rotation behaviour is highly sensitive to
These were not the same as those in the centrifuge tests this. This is discussed further in the next section.
(Fig. 4), as it was impossible to achieve perfectly level
placement of the structures and avoid small variations in soil
properties in preparing the real soil. As the rotational behav- VALIDATION OF FINITE-ELEMENT MODEL AGAINST
iour is likely to be highly influenced by any initial bias in CENTRIFUGE TEST DATABASE
the system, a third set of simulations was also conducted, As there is a substantial amount of test data (seven tests,
using the HST95 parameter set, but with additional vertical each having five earthquakes’ worth of data), the perform-
point loads applied to the foundation (points B, C, H and I) ance of the finite-element model and effects of both material
during the initial static step to generate a couple which properties and initial conditions are here summarised in terms
forces the structure to have the initial rotations shown in of the following key performance indicators: (a) peak ground
Fig. 4. Such a couple, superimposed on the initially equally acceleration near the soil surface in the free-field (point F);
divided vertical loads, simulates the difference in vertical (b) peak ground acceleration at point G (in the case of
loads between footings induced by the resultant static mo- adjacent structure models); (c) peak cyclic drift across the
ment on the structure that is consistent with the measured superstructure (between the foundation and the mass plates)
structural rotation. The magnitude and direction of the in each earthquake; (d ) post-earthquake settlement and
couple in each case were determined by trial and error (e) post-earthquake structural tilt (global rotation). It should
within the initial static step of the corresponding finite- be noted that (c) is a measure of super-structural demand,
element model, so the resulting static moment and rotation whereas (d ) and (e) are measures of foundation performance
are consistent with the non-linear behaviour of the founda- which may affect the post-earthquake serviceability of the
tion soil. By adding a couple, the average bearing pressure structure.
across the whole structure is unchanged (it has consistent Figure 10(a) shows the soil amplification factor in the free
mass), despite the loads on the different footings being field (SFF) from the centrifuge tests and Fig. 10(b) presents
distributed differently. This third set of simulations will the performance of the numerical simulations in replicating
demonstrate the importance of knowing the initial conditions this parameter using the different sets of material param-
of a structure prior to an earthquake (such as could be eters. The factor SFF is the ratio of the peak ground
measured by surveying). acceleration at point F divided by that at point E, and the
106
1m
7m
Positive rotation
Centre of mass, A
Footings: B (left) and C (right)

G Free field, F
D
1m
Soil base, E
1m
(a)
Centre of mass: A (left) and K (right)

Positive rotation
G – 1 m deep in soil
– equidistant between structures
Footings: B (left) and C (right) Footings: H (left) and I (right)

20 m Free field, F
1m
D (left) and J (right)
1m
Soil base, E
1m
(b)
Fig. 7. Example layouts and finite-element mesh for numerical simulations: (a) PM003; (b) PM006.
All dimensions at prototype scale
value of 1 .4 suggested by Eurocode 8 for ground type E is attenuation of ground motion close to (between) the adjacent
also shown for context (BSI, 2005). Examples of the ampli- structures. In the two finite-element cases using the ideal
fication in the time histories of motion can be seen in Fig. initial conditions (Fig. 11(b)), this behaviour is not well
8(b) and Fig. 9(c). Only the cases with ideal initial condi- represented, showing instead a predominance towards ampli-
tions are shown for the finite-element model data, as the fication in the near field. When the initial conditions are
initial conditions were not found to affect this parameter. correctly replicated, this tendency is reduced and a better
Generally, both sets of material parameter correlations pro- match to the centrifuge data is obtained. This may suggest
duce similar predictions of the centrifuge data, including the that the non-symmetrical changes to the stress distribution
observation from the centrifuge that the amplification is beneath the structures induced by the non-uniform load
generally larger in the smaller earthquakes. There is also a distribution between the footings increases the asymmetry in
noticeable increase in SFF in the smaller earthquake follow- the interactions between the incident and reflected waves
ing the strong aftershock sequence in Fig. 10(a), presumably beneath the structures, encouraging destructive interference
as a result of soil densification and a resulting stiffening of as these waves are superimposed.
the soil response. The amplification factors are generally Figure 12 compares the magnitude of peak drift recorded
lower in the higher strength earthquakes as there is increas- for each structure in each earthquake. These are clustered
ing soil inelasticity, which limits the transfer of cyclic shear into a group of smaller values, representing the response of
stress. (Ultimately shear decoupling may occur if the ratio the short structures, and a larger set for the long structures.
(S 3 ag /g) becomes equal to tan 9p ; that is, the cyclic shear The variation in the magnitude of the drift is partially
stresses become equal to the shear strength of the soil. This associated with the actual achieved strength of the input
does not happen here as the peak friction angle of the soil motion, variations in soil amplification in each test (e.g. Fig.
would require S ¼ 1 .68 for ag ¼ 0 .5g in the free field and 10) and SSSI. Fig. 12 suggests that the dynamic response is
none of the measured values for this strength of input best simulated when the initial conditions are correctly
motion is this high in Fig. 10(a).) simulated, with the idealised cases leading to an under-
Figure 11 shows the changes to the soil amplification in prediction of super-structural response.
the near field of the structures (only data for the adjacent Figure 13 shows the post-earthquake structural settle-
structure models are shown). SNF is the ratio of the peak ments. These are over-predicted by the Brinkgreve et al.
ground acceleration at point G divided by that at point E. (2010) set of material parameters, and under-predicted using
The centrifuge data in Fig. 11(a) show, in general, a slight the soil-specific HST95 parameters. The over-prediction
107
1·5 2·5 Centrifuge test
Centrifuge test
HST95 ideal initial conditions HST95 ideal initial conditions

2·0
1·0
HST95 measured initial conditions HST95 measured initial conditions
1·5
Structural tilt: degrees

0·5
1·0
Structural tilt: deg
0·5
0
0 50 100 150
Time: s 0
0 50 100 150
0·5 Time: s
0·5
1·0 1·0
1·5
(a)
1·5
(a)
1·5 Centrifuge test
1·0
Free field, F HST95 ideal initial conditions
Input, E 1·0
HST95 measured initial conditions
Structural tilt: degrees

0·5
0·5
Acceleration: g
0
0 50 100 150
Time: s
0 0·5
0 50 100 150
Time: s
1·0
0·5 1·5
(b) (b)
Fig. 8. Effect of initial conditions on simulated rotational 1·0

response of an isolated structure during a sequence of strong Free field, F
earthquakes (data for PM004 shown): (a) rotations; (b) ground Input, E
motions at bedrock (E) and free field (F) (as recorded in the
centrifuge)
0·5
Acceleration: g
within the former is likely to be attributable to the lower soil

strength (peak friction angle – see Table 4) meaning that soil
yield occurs earlier and greater settlements are accrued. The
Brinkgreve et al. (2010) set of parameters was similarly 0
found by Al-Defae et al. (2013) to over-predict permanent 0 50 100 150
Time: s
seismic slope deformations in the same sand (at a similar
relative density). These observations demonstrate the benefits
of investing additional effort and resources in performing a
0·5
soil-specific model calibration on accurate prediction of (c)
permanent deformations. The importance of correct simula-
tion of the initial conditions is also highlighted, as for peak Fig. 9. Effect of initial conditions on simulated rotational
drift. Using the idealised initial conditions resulted in a response of adjacent identical structures during a sequence of
starting position with comparative amounts of settlement of strong earthquakes (data for PM006 shown): (a) rotation of left
each foundation (low initial rotation). Applying the couple structure; (b) rotation of right structure; (c) ground motions at
to generate the measured initial conditions generally resulted bedrock (E) and free field (F) (as recorded in the centrifuge)
in an increase in the load on one of the foundations, while
reducing it on the other. As the soil response is non-linear,
the foundation under greater compressive loading will be data, and shows a number of significant outliers, particularly
pushed into a more inelastic part of the load–settlement along the x-axis (i.e. the finite-element model is predicting
curve, resulting in greater settlements. This would be ex- large rotations). These points are associated with the later
pected in the centrifuge too, assuming that the stress dis- earthquake motions of the adjacent structure tests when
tribution is similarly altered as a result of the measured substantial permanent rotations had accrued in the finite-
initial conditions. As settlement of the structure is the element model owing to the successive strong shaking, and
average of the settlements of the two foundations, this may are to be expected given the example results from Fig. 9. In
explain the lower settlements for the idealised initial condi- Fig. 14(b), only the data from the first 0 .1g and first 0 .5g
tions compared to both the case with measured initial motions are plotted, which results in a strong positive
conditions and the centrifuge data, which match well. correlation, but only when the initial conditions are correctly
Figure 14 compares the post-earthquake permanent rota- modelled. This also serves to highlight the data for the two
tions (tilts) of the structures. Fig. 14(a) includes all of the sets of simulations which used idealised conditions, where it
108
2·5 2·5
Centrifuge Centrifuge
Eurocode 8, ground
type E
2·0 2·0
Free-field amplification, SFF
1·5 1·5
SNF/SFF
1·0 1·0
0·5 0·5
0·1g EQ 0·5g EQs 0·1g EQ 0·1g 0·5g 0·1g
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Earthquake event no. Earthquake event no.
(a) (a)
2·5 1·5
2·0
SNF/SFF, centrifuge test

1·0
SFF, centrifuge test
1·5
1·0
Brinkgreve et al. (2010) 0·5

Brinkgreve et al. (2010)
HST95 ideal initial conditions
0·5 HST95 ideal initial conditions
Parity (1:1)
Eurocode 8, ground type E Parity (1:1)
0 0
0 0·5 1·0 1·5 2·0 2·5 0 0·5 1·0 1·5
SFF, finite-element model SNF/SFF, finite-element model
(b) (b)
Fig. 10. Effect of soil material parameter correlations on pre- Fig. 11. Effect of soil material parameter correlations and
diction of site effect in the free field through a sequence of structural initial conditions on amplification/attenuation of
strong earthquakes: (a) centrifuge data, adjacent structure tests; ground motion in the near field (point (G), adjacent structure
(b) comparison of finite-element method predicted values to cases): (a) centrifuge data; (b) comparison of finite-element
centrifuge data method predicted values to centrifuge data
can be seen that neither set of material properties provides 100

any correlation with the centrifuge results.
|Peak drift|, centrifuge test: mm
80
Validation summary
Linear relationships were fitted to the individual data sets 60
in Figs 10(b), 11(b), 12, 13 and 14(b) (and also for data of
SNF plotted as centrifuge against finite-element model), using
a least-squares fitting procedure. The gradient of these 40
relationships demonstrates, on average across the full data- Brinkgreve et al. (2010)
set, the degree of over- or under-prediction. The inverse of
20 HST95 ideal initial conditions
these gradients, plotted as percentages, are summarised in
Fig. 15 for the performance indicators (a)–(e). This shows HST95 measured initial conditions
that in order to achieve the best simulation of soil structure Parity (1:1)
0
interaction and SSSI (at least for pairs of structures), it is
0 20 40 60 80 100
necessary to both obtain a soil-specific set of model param- |Peak drift|, finite-element model: mm
eters (as also concluded by Al-Defae et al. (2013)) and to
model the actual initial (rotation) conditions of the struc- Fig. 12. Effect of finite-element method modelling assumptions on
ture(s). When applied in practice to field structures, this estimation of peak drift
could be measured based on structural surveying of the
building stock, and would need to be updated if this varied
with time since construction. When both material properties there are some outlying points, which are perhaps to be
and initial conditions are correctly modelled, all five of the expected given the extensive amount of earthquake shaking
performance indicators can generally be predicted within applied to each model, and therefore the potential for small
10% averaged error across the 35 different earthquake and differences to become amplified by the end of the earth-
structure combinations considered in this paper, although quake sequence.
109
300 200
Post-EQ settlement, centrifuge: mm
Finite-element model/centrifuge: %
150
Parity (1:1)
200 Finite-element method over-prediction
10%
100
10%
Finite-element method under-prediction
100
50
0 0
0 100 200 300 SFF SNF |Peak Post-EQ Post-EQ
Post-EQ settlement, finite-element model: mm drift| settlement tilt*
Fig. 13. Effect of finite-element method modelling assumptions on *Tilt value for initial
estimation of post-earthquake settlement earthquakes only
All other values include
HST95 measured initial conditions aftershock data
3·0
Fig. 15. Summary of finite-element method validation against

Post EQ tilt, centrifuge: degrees
2·0
centrifuge test data
1·0
for perfectly ideal and identical ground. In this section,
0 finite-element analyses using the HST95 material model and
3·0 2·0 1·0 0 1·0 2·0 3·0
ideal initial conditions are therefore compared to demon-
1·0 Brinkgreve et al. (2010) strate the effects of the presence of an adjacent structure on
HST95 ideal initial conditions the resulting SSSI and structural response, all other condi-
2·0 HST95 measured initial tions being equal, using the same Kobe earthquake motion
conditions
Parity (1:1) and order of consecutive motions as described previously.
3·0 Tilt 1/50 The results are summarised in Figs 16–18 and will be
Post-EQ tilt, finite-element model: degrees
discussed together at the end of the section. Fig. 16 shows
(a) the peak drifts normalised by those of the isolated structures.
1·0
The data are separated by earthquake strength and by
structure type, and each data point represents an average
Post EQ tilt, centrifuge: degrees
across the different motions; for example, for the case of

0·5 similar tall structures, the 0 .5g point in Fig. 16 is the
average of the EQ2, EQ3 and EQ4 responses for both
structures. Fig. 17(a) shows a comparison of the permanent
0 movements (settlement and tilt) of a long structure when it
1·0 0·5 0 0·5 1·0 is either on its own (‘isolated’), adjacent to an identical
structure (‘Similar’, with ‘L’ and ‘R’ denoting left and right
0·5 Brinkgreve et al. (2010)
Series 1 130
1·0 SSSI detrimental
Parity (1:1)
Post-EQ tilt, finite-element model: degrees
(b) 120
Change in drift due to SSSI, compared
Fig. 14. Effect of finite-element method modelling assumptions on

to an isolated structure: %
estimation of post-earthquake permanent rotation (structural 110

tilt): (a) all earthquakes; (b) earthquakes EQ1 (first 0 .1g motion)
and EQ2 (first 0 .5g motion) only
100
INSIGHTS INTO SSSI OF PAIRS OF ADJACENT

STRUCTURES 90
Long structures (0·5g)
Although the centrifuge test data are valuable as a means
of validating the finite-element model and understanding the Long structures (0·1g)
importance of the modelling assumptions (material proper- 80
Short structures (0·5g)
ties and initial conditions), direct comparison across tests,
particularly in terms of settlement and tilt, is not ideal, Short structures (0·1g)
SSSI beneficial
owing to the different initial conditions in the tests (Fig. 4). 70
The rotation behaviour has also been shown to be highly Similar Similar Dissimilar Overall
(s 2 m) (s 1 m)
sensitive to the exact ground conditions which match well,
but not perfectly across the different centrifuge models (rel- Fig. 16. Summary of effect of adjacent structure properties and
ative density in Table 2). The finite-element method, how- earthquake strength on drift amplification/attenuation due to
ever, presents an opportunity to compare model behaviour SSSI (ideal initial conditions)
110
Static EQ1 EQ2 EQ3 EQ4 EQ5 Static EQ1 EQ2 EQ3 EQ4 EQ5
0 0
Isolated
Final settlement: mm
Final settlement: mm
Similar, L
50 50
Similar, R
Dissimilar, L
100 100
Dissimilar, R
150 150
1·00 1·00
Final rotation: degrees

0·50 0·50 Isolated
Similar, L
0 0
Similar, R
0·50 0·50 Dissimilar, L
Dissimilar, R
1·00 1·00
Static EQ1 EQ2 EQ3 EQ4 EQ5 Static EQ1 EQ2 EQ3 EQ4 EQ5
(b)
(a)
Fig. 17. Effect of adjacent structure properties, earthquake strength and strong aftershocks on development of settlement and tilt due to
SSSI (ideal initial conditions): (a) long structures; (b) short structures
1·0
of the pair relative to the direction of the earthquake and the
Isolated geometry shown in Fig. 7), or next to a smaller structure
Similar (‘Dissimilar’). Fig. 17(b) shows similar data for a short
structure when it is isolated, next to an identical structure, or
Dissimilar
next to a larger structure. As the rotation may be different
depending on whether the structure is on the left or right of
0·5 the arrangement (this essentially represents the earthquake
motions being in opposite directions), Fig. 18 shows the
average of the absolute rotations of the ‘L’ and ‘R’ cases
from Fig. 17. Based on Figs 16–18, the following insights
can be drawn.
(a) Insights for the case where a structure is situated next to

0 an identical neighbour.
Static EQ1 EQ2 EQ3 EQ4 EQ5 (i) The drift may or may not be increased, depending on
(a)
the natural period of the structure in question and the
strength of the earthquake. In this study, in small
earthquakes inducing a smaller strain soil response,
1·0
SSSI increased drift for long period structures and
Isolated reduced it for short period structures; in larger
Similar earthquakes with a strong elasto-plastic near-field
response, SSSI increased drift for both types of
Dissimilar
structure by between 2 and 10% (Fig. 16).
(ii) The settlement is either unaffected, or slightly
0·5 reduced due to the adjacent structure providing
additional confinement to the soil beneath the
foundations (Fig. 17).
(iii) The magnitude of the rotation of the structure
increases (Fig. 18), and it rotates away from its
neighbour (compare hollow circle and hollow square
markers in Fig. 17(a) or Fig. 17(b); the sign
0 convention is shown in Fig. 7). This outward
Static EQ1 EQ2 EQ3 EQ4 EQ5 ratcheting is thought to occur as plastic soil
(b) deformation is confined by the adjacent structure
while the structure is rotating towards its neighbour,
Fig. 18. Magnitude of structural tilt (average of both earthquake whereas soil deformation while rotating outwards
directions): (a) long structures; (b) short structures (away from its neighbour) is not. This means that
111
there would be a net rotation outwards in a given either similar or dissimilar type. This demonstrated that the
cycle of deformation, which would then progres- structural drift and co-seismic settlement could be reduced
sively accrue in subsequent cycles due to P–˜ or increased as a result of SSSI, depending chiefly on the
effects. properties of the adjacent structure (building height was
(b) Insights for the case where a structure is situated next to a considered here, in terms of changes to the fundamental
shorter neighbour (having lower natural period and period and foundation bearing pressure). This suggests that,
bearing pressure). through further study, it may be possible in the future to
(i) SSSI increases drift compared to the isolated case, prescribe dynamic properties in seismic design to exploit
but by less than when the neighbouring structure is beneficial effects of SSSI with the surrounding urban envir-
identical, and in both small and large earthquakes onment. However, in all cases, permanent rotation (tilt) of
(Fig. 16). the structure was observed to increase compared to the
(ii) The settlement is increased (Fig. 17(a)). isolated case as a result of SSSI, and so consideration must
(iii) The magnitude of rotation of the structure increases, also be given to effective ways of remediating this.
by more than when the neighbouring structure is
identical (Fig. 18(a)). It rotates towards its neighbour
(positive rotation if on the left of the pair and
negative rotation if on the right, Fig. 17(a)). ACKNOWLEDGEMENTS
(c) Insights for the case where a structure is situated next to a This work was supported by the Engineering and Physical
taller neighbour (having higher natural period and Sciences Research Council (EPSRC) under grant no. EP/
bearing pressure). H039716/1, which the authors acknowledge with thanks. The
(i) SSSI appears to increase drift in larger earthquakes authors would additionally like to express their sincere
but reduce it in smaller earthquakes compared to the gratitude to Mark Truswell and Colin Stark at the University
isolated case. Irrespective of earthquake strength, of Dundee for their assistance in performing the centrifuge
however, drift is larger than the case when the tests.
neighbouring structure is identical (Fig. 16).
(ii) Settlement is reduced (Fig. 17(b)).
(iii) The magnitude of rotation of the structure increases,
by less than when the neighbouring structure is NOTATION
identical (Fig. 18(b)). It rotates away from its ag peak ground acceleration at bedrock/input
neighbour (negative rotation if on the left of the B footing width (in plane of shaking)
pair and positive rotation if on the right, Fig. 17(b)). Ct empirical period determination factor
Cu coefficient of uniformity
These conclusions should not be considered to be general, Cz coefficient of curvature
as there is a need to perform further simulations with differ- ck stiffness-proportional Rayleigh damping coefficient
ent types of structure (particularly to investigate the effect of cm mass-proportional Rayleigh damping coefficient
c9 cohesion intercept
different building widths and foundation types) and on dif-
Dr relative density
ferent types of ground to demonstrate generality. However, D10 particle diameter at which 10% is smaller
they do demonstrate that the presence nearby of even a D30 particle diameter at which 30% is smaller
single adjacent structure can have a dramatic effect on a D60 particle diameter at which 60% is smaller
structure’s seismic response compared to a consideration of Eoed oedometric tangent stiffness (in compression)
the same structure and underlying ground in isolation. These Eur unloading–reloading stiffness
effects appear to be either beneficial or detrimental, depend- E50 triaxial secant stiffness (at 50% of deviatoric failure
ing on the relative dynamic properties of the adjacent stress in drained triaxial compression)
structures and strength of the earthquake (i.e. soil response). EI elastic bending stiffness
This suggests that through further study it may be possible emax maximum void ratio
emin minimum void ratio
to exploit the beneficial effects of SSSI and avoid the FSv static vertical factor of safety
detrimental ones to improve the seismic performance of the fn natural frequency
built environment. Gs specific gravity of soil grains
G(0) (small strain) shear modulus
g acceleration due to gravity (¼ 9 .81 m/s2)
CONCLUSIONS H height
This paper has examined how the performance of a simple Keq equivalent lateral sway stiffness
structure is altered when it is situated close to an adjacent K0 coefficient of lateral earth pressure at rest
structure, as a first step towards a better understanding of Meq equivalent mass
Mw moment magnitude
the seismic response of densely packed urban areas. Dy- m power-law index for stress-dependency of stiffness
namic centrifuge modelling was conducted such that the full Nª footing bearing capacity factor
non-linear behaviour of the soil could be incorporated into p9 mean effective stress
the SSSI. This generated a database of performance data q bearing pressure
against which non-linear finite-element models were vali- Rf deviatoric failure ratio
dated. The importance of both generalised or soil-specific Se spectral acceleration
material properties and the initial geometric configuration of S(FF,NF) Eurocode 8 equivalent soil factor (free-field, near-field)
the structure (initial conditions) was investigated, and it was s footing spacing (centre-to-centre)
demonstrated that accurate simulations could be achieved so sª footing shape factor
Tn0 fundamental natural period
long as soil-specific material properties can be determined
ª soil unit weight (dry)
and the initial conditions are known. (This would require �s,0 .7 shear strain at G/G0 ¼ 0 .7
building surveys for field application and laboratory testing �ur Poisson ratio (unload–reload)
of soils to generate site-specific soil property calibrations.) � equivalent viscous damping
The finite-element approach was subsequently used to inves- �9p (secant) peak angle of friction
tigate the effects of the presence of an adjacent structure of ł9 dilation angle
112
REFERENCES BSI (2004). BS EN 1997-1:2004: Eurocode 7: Geotechnical design
Al-Defae, A. H., Caucis, K. & Knappett, J. A. (2013). Aftershocks – Part 1: General rules. British Standards Institution, London,
and the whole-life seismic performance of granular slopes. UK.
Géotechnique 63, No. 14, 1230–1244, http://dx.doi.org/10.1680/ BSI (2005). BS EN 1998-1:2005: Eurocode 8: Design of structures
geot.12.P.149. for earthquake resistance – Part 1: General rules, seismic actions
Alexander, N. A., Ibraim, E. & Aldaikh, H. (2013). A simple and rules for buildings. British Standards Institution, London,
discrete model for interaction of adjacent buildings during earth- UK.
quakes. Comput. Structs 124, 1–10. Goel, R. K. & Chopra, A. K. (1997). Period formulas for moment-
Benz, T. (2006). Small-strain stiffness of soils and its numerical resisting frame buildings. J. Structl Engng 123, No. 11, 1454–
consequences. PhD thesis, University of Stuttgart, Germany. 1461.
Bertalot, D. (2012). Behaviour of shallow foundations on layered Lauder, K. (2011). The performance of pipeline ploughs. PhD
soil deposits containing loose saturated sands during earth- thesis, University of Dundee, UK.
quakes. PhD thesis, University of Dundee, UK. Lyamin, A. V., Salgado, R., Sloan, S. W. & Prezzi, M. (2007). Two-
Bertalot, D., Brennan, A. J., Knappett, J. A., Muir Wood, D. & and three-dimensional bearing capacity of footings in sand.
Villalobos, F. A. (2012). Use of centrifuge modelling to improve Géotechnique 57, No. 8, 647–662, http://dx.doi.org/10.1680/
lessons learned from earthquake case histories. Proceedings of geot.2007.57.8.647.
the 2nd European conference on physical modelling in geotech- Lysmer, J. & Kuhlmeyer, R. L. (1969). Finite dynamic model for
nics, Eurofuge 2012, Delft, the Netherlands. infinite media. ASCE J. Engng Mech. Div. 95, No. 4, 859–887.
Betti, R. (1997). Effects of the dynamic cross-interaction in the Muir Wood, D. (2004). Geotechnical modelling. Abingdon, Oxford-
seismic analysis of multiple embedded foundations. Earthquake shire, UK: Spon.
Engng Structl Dynam. 26, No. 10, 1005–1019. Padrón, L. A., Aznárez, J. J. & Maeso, O. (2009). Dynamic
Brennan, A. J., Knappett, J. A., Bertalot, D., Loli, M., Anastasopou- structure–soil–structure interaction between nearby piled build-
los, I. & Brown, M. J. (2014). Dynamic centrifuge modelling ings under seismic excitation by BEM–FEM model. Soil Dy-
facilities at the University of Dundee and their application to namics Earthquake Engng 29, No. 6, 1084–1096.
studying seismic case histories. In Proceedings of the 8th Salgado, R. (2008). The engineering of foundations. New York, NY,
international conference on physical modelling in geotechnics USA: McGraw-Hill.
(eds C. Gaudin and D. J. White), pp. 227–233. London, UK: Stewart, J. P., Seed, R. B. & Fenves, G. L. (1999). Seismic soil-
Taylor & Francis Group. structure interaction in buildings, Part II: Empirical findings. J.
Brinkgreve, R. B. J., Engin, E. & Engin, H. K. (2010). Validation of Geotech. Geoenviron. Engng 125, No. 1, 38–48.
empirical formulas to derive model parameters for sands. In Numer- Tsogka, C. & Wirgin, A. (2003). Simulation of seismic response in
ical methods in geotechnical engineering (eds T. Benz and S. an idealized city. Soil Dynam. Earthquake Engng 23, No. 5,
Nordal). Rotterdam, the Netherlands: CRC Press/Balkema. 391–402.
113
Heron, C. M. et al. (2015). Géotechnique 65, No. 5, 442–451 [http://dx.doi.org/10.1680/geot.SIP.15.P.020]
A new macro-element model encapsulating the dynamic moment–rotation

behaviour of raft foundations
C . M . H E RO N , S . K . H A I G H † a n d S . P. G . M A DA B H U S H I †
The interaction of shallow foundations with the underlying soil during dynamic loading can have both
positive and negative effects on the behaviour of the superstructure. Although the negative impacts are
generally considered within design codes, seldom is design performed in such a way as to maximise
the potential beneficial characteristics. This is, in part, due to the complexity of modelling the soil–
structure interaction. Using the data from dynamic centrifuge testing of raft foundations on dry sand,
a simple moment–rotation macro-element model has been developed, which has been calibrated and
validated against the experimental data. For the prototype tested, the model is capable of accurately
predicting the underlying moment–rotation backbone shape and energy dissipation during cyclic
loading. Utilising this model within a finite-element model of the structure could potentially allow a
coupled analysis of the full soil–foundation–structure system’s seismic response in a simplified manner
compared to other methods proposed in literature. This permits the beneficial soil–structure interaction
characteristics, such as the dissipation of seismic energy, to be reliably included in the design process,
resulting in more efficient, cost-effective and safe designs. In this paper the derivation of the model is
presented, including details of the calibration process. In addition, an appraisal of the likely resultant
error of the model prediction is presented and visual examples of how well the model mimics the
experimental data are provided.
KEYWORDS: centrifuge modelling; dynamics; earthquakes; footings/foundations; soil/structure interaction
INTRODUCTION protection, a comprehensive model of the soil–foundation

Performance-based seismic design, whereby a system is interaction behaviour is required.
designed based on deformation limits rather than load limits, Previous researchers have investigated different methods
offers the potential for safe economical design in seismic of incorporating these beneficial characteristics into the de-
engineering. Permitting a certain amount of ductility in a sign process. Ultimately all methods have to be incorporated
system minimises the cost of the structure when designing into a numerical model of the overall soil–foundation–
for extreme but infrequent load cases such as earthquakes. structure system. This can be done either by detailed numer-
This in turn allows for a more efficient design and construc- ical simulations of the soil behaviour (Abate et al., 2010;
tion procedure. With the ability to accurately model the Gelagoti et al., 2012), by equivalent simplified springs
behaviour of manufactured structural elements, this design (Raychowdhury & Hutchinson, 2009; Wotherspoon & Pen-
philosophy has to date been utilised extensively in structural der, 2010; Anastasopoulos & Kontoroupi, 2014) or by using
designs. Examples of such a method are the introduction of macro-element models (Paolucci et al., 2007; Grange et al.,
ductility in the design of beams and columns of a steel- 2009; Chatzigogos et al., 2011). The latter will form the
frame structure or the use of unbonded tendons to give focus of this paper. The use of a macro-element simplifies
ductile behaviour of beam–column joints (Holden et al., the modelling process by reducing the large number of
2003; Pampanin, 2005; Ou et al., 2010; Smith et al., 2011). elements and non-linear relationships (such as the elasto-
With designs being led by structural engineers combined plastic behaviour of the soil and non-elastic soil–structure
with a perceived or real lack of ability to characterise the interface movements) down to a single element which
seismic response of the soil, ductility has not been widely combines these non-linearities into one constitutive law. In
included into the design of the soil–foundation system. How- order for such a model to be accepted for use in design, it
ever, recent research (Gajan & Kutter, 2008; Anastasopoulos needs to be fully validated against physical modelling and/or
et al., 2010; Pender, 2010; Gelagoti et al., 2012) has in- field data. Despite soil–foundation interaction being capable
creasingly shown the potential merits of doing so. Utilising of providing seismic protection to a variety of foundation
ductility within the underlying ground can potentially reduce types, this research focuses on shallow raft foundations
the cost of the overall system and can provide significant located on dry sand beds.
levels of seismic protection. In order to optimise designs Paolucci et al. (2007) made use of data from 1g pseudo-
and provide quantitative assessments of the level of seismic dynamic tests and tests conducted on a large 1g shaking
table to compare against predictions from a macro-element
model. The variations between the physical model and
numerical analysis results are similar in both cases in that
Manuscript received 4 April 2014; revised manuscript accepted 17 the numerical analysis over-predicts permanent rotations and
February 2015. Published online ahead of print 15 April 2015. under-predicts settlements compared to the physical model
see p. ii.
results. The authors note that the error in the numerical
Department of Civil Engineering, University of Nottingham, analysis increases with the magnitude of excitation. The
Nottingham, UK. results presented in their paper do, however, have one of the
† Department of Engineering, University of Cambridge, Cambridge, best correlations between experimental and numerical data
UK. presented in the literature.
115
HERON, HAIGH AND MADABHUSHI
Similarly, Grange et al. (2009) present a macro-element of allowing the response to simple motions to be analysed
model which is validated against data from a large-scale 1g and understood prior to examining the behaviour when more
shaking table test. In this case, the model predicted the complex earthquake traces are used. This does, however,
rotations of the structure relatively accurately; however, it limit the extent to which the model developed in this paper
over-predicted settlement by approximately 50%. This is can be validated for use with real earthquake motions. The
contrary to the results presented by Paolucci et al. (2007) sand and foundations were contained within a rigid container
where the model under-predicted the settlements. with a Perspex window, allowing the movements of the
The model presented by Gajan & Kutter (2009) tracks the foundation and soil to be imaged at a high rate during the
geometry of the soil–foundation interface and determines the test. To reduce the effect of the rigid boundary conditions, a
loading on the foundation, thus the moment, rotation and plastic material, Duxseal (Steedman & Madabhushi, 1991),
settlement are evaluated simultaneously. Six user-defined was used to limit the amount of energy which was reflected
parameters are required, as well as nine non-user-defined from the container walls. In addition to high-speed photo-
parameters, these having been back-calibrated against centri- graphy, an extensive instrumentation array, consisting of
fuge data. Gajan & Kutter (2009) present several compari- miniature piezoelectric accelerometers and micro-electro-
sons between the predictions made by the contact interface mechanical system (MEMS) accelerometers, was used to
model and experimental data obtained from pseudo-dynamic monitor the response of the ground, foundation and super-
centrifuge tests – with moments, shear forces and settle- structure.
ments all being predicted accurately. However, despite per-
forming true-dynamic centrifuge testing, the authors did not
present any comparisons between the model predictions and Test details
these data; such a validation is vital in order for the model The structure tested was a single-degree-of-freedom (when
to be adopted. In addition, the large number of user and the base is fixed) stiff structure with a high centre of gravity
non-user-defined parameters adds to the complexity of im- and was located on a shallow raft foundation. The founda-
plementing this model for alternative prototype scenarios tion was located on dry Hostun HN31 sand (Flavigny et al.,
and reduces confidence that the prediction can be extrapo- 1990) prepared to loose and dense states using an automated
lated beyond the precise situation studied – 15 independent sand pourer (Zhao et al., 2006). A typical model layout is
parameters allowing almost any behaviour to be replicated shown in Fig. 1. Table 1 summarises details of the structure
with appropriate values chosen. used and Table 2 summarises the tests conducted. Included
Although there have been numerous other macro-elements in Table 1 is the vertical static factor of safety for each test,
presented in literature, with varying degrees of validation which was calculated assuming a Coulomb soil, a rough
and capabilities, the three outlined above are prominent in foundation–soil interface (Davis & Booker, 1971) and a
the field and exemplify the challenges of developing such shape correction factor as detailed in Eurocode 7 (BSI,
models. The majority of published models have to be ques- 2004). The pseudo-dynamic factor of safety for each test is
tioned because they have been calibrated using data from
tests in which the soil stress-state and loading were not
accurately replicated. Ideally, a simple model, rigorously
calibrated and validated using data from tests which repre- Model
structure
sent the prototype stress-state, would be available to practis-
ing engineers to use in the seismic design of foundations.
In this paper a simplified macro-element model developed
based on fundamental geotechnical principles will be pre- Model
sented. The calibration and validation of the model against a container
collection of centrifuge data, collected from true-dynamic
testing of a raft foundation located on dry sand, will also be
shown, concluding with a realistic appraisal of the new
model, its capabilities and limitations. The model consists of High-speed
two separate components; a backbone curve which forms the camera
overall shape of the moment–rotation cycles and an energy
dissipation component which converts the model backbone (a)
into a fully developed moment–rotation cycle. Prior to
presentation of the model, the experimental programme used
to derive and validate the model will be described.
EXPERIMENTAL PROGRAMME
Data from two dynamic centrifuge tests, incorporating
numerous excitations, were used to develop the model
described in this paper. The tests were conducted using the
240 mm
10-m diameter Turner beam centrifuge (Schofield, 1980),

which was operated such that the g-level in the region of the
foundation was 44g. A stored angular momentum (SAM) 500 mm
actuator (Madabhushi et al., 1998) was used to subject the
models either to constant frequency, constant amplitude
sinusoidal ground motions or to subject them to a constant Accelerometer
displacement and decreasing frequency ground motion (re-
(b)
ferred to as a sine-sweep). The amplitude, frequency and
duration of each shake can be adjusted in-flight, allowing a Fig. 1. Model layout: (a) photograph of model set-up on
range of motions to be tested. Subjecting the models to centrifuge; (b) schematic diagram of instrument and structure
constant frequency and amplitude motions has the advantage layout
116
DYNAMIC MOMENT–ROTATION BEHAVIOUR OF RAFT FOUNDATIONS
Table 1. Details of model structure (model scale values in by the difference in displacement between the two vertical
brackets) accelerometers on either side of the foundation divided by
the distance between the instruments. The moment was
Bearing pressure: kPa 82 (82) taken as the overturning force provided by the inertial
Fixed base natural frequency: Hz Stiff: . 9 (. 400) acceleration of the structure mass multiplied by the distance
Height of centre of gravity: m 2 .4 (0 .054)
between the base of the foundation and the centre of gravity
Base width: m 2 .2 (0 .050)
Overall height: m 4 .0 (0 .092) of the structure. The moment contribution from the rota-
Length: m 9 .4 (0 .214) tional inertial component was found to be negligible com-
Construction material – (Steel) pared to the overturning moment and was therefore not
Static vertical factor of safety (loose soil) 7 .2 included in the production of the plots presented in this
Static vertical factor of safety (dense soil) 7 .8 paper.
DEVELOPMENT OF THE BACKBONE MODEL

Fundamentally, the soil beneath the foundation is experi-
detailed in Table 2 and was calculated using the approach encing a load–unload cycle as the foundation rocks during
presented by Butterfield & Gottardi (1994), as previously seismic loading, with the load being provided by the mo-
implemented for a similar purpose by Loli et al. (2014). ment loading transmitted to the ground by the foundation. It
is intuitive therefore initially to consider an established
model for the stress–strain behaviour of soil, such as that
Data processing developed by Oztoprak & Bolton (2013) (equation (1)).
The results presented in this paper are derived using data Utilising a mobilisable strength design framework, as pro-
collected from the MEMS accelerometers located on the posed by Osman & Bolton (2005), the applied moment from
structure, with quoted base input accelerations being deter- the foundation can be considered to induce a representative
mined from the piezoelectric accelerometer attached to the shear stress in the deformation mechanism within the soil,
outside base of the model container. The data were initially �rep. The rotation can also be shown to be compatible with a
processed by filtering out high-frequency noise using an certain magnitude of shear strain in the mechanism, ªrep.
eighth-order Butterworth filter with a cut of frequency of Equation (1) can then be used to link these representative
400 Hz. The MEMS accelerometers used have inbuilt filters stresses and strains and hence to link moments to rotations.
at 400 Hz, hence no data were eliminated. Although the If the stresses and strains are assumed to be proportional to
fixed base natural frequency of the stiff structure exceeded the moments and rotations, respectively, as shown in equa-
400 Hz, negligible internal deformation would have occurred tions (2) and (3), combining equations (1), (2) and (3)
during these tests due to the excitation frequency (50 Hz) allows the formulation of an overall relationship between
being significantly below the natural frequency and therefore moment and rotation, as shown in equation (4). The choice
no information is lost by not examining frequencies above of values for the R and S proportionality constants will be
400 Hz. To obtain the experimental moment–rotation behav- the main focus of this section.
iour, the acceleration data were double integrated and then
high-pass filtered above 10 Hz in order to remove the G 1
¼ (1)
accumulation of displacement error created by the integra- G0 ª � ªe a
tion process. These calculated displacements were validated 1þ
ªr
against displacements obtained from imaging (particle image
velocimetry (PIV)) techniques. The rotation was evaluated �rep ¼ R 3 M (2)
Table 2. Details of tests conducted (model scale values in brackets)
Test name Relative Earthquake details

density
EQ number Input acceleration: g Frequency: Hz Duration: s Pseudo-dynamic factor
of safety
CH10 50% EQ01 0 .14 (6 .0) 1 .1 (50) 22 (0 .5) 2 .2

EQ02 0 .25 (11 .0) 1 .1 (50) 22 (0 .5) 0 .8
EQ03 0 .18 (8 .0) 1 .1 (50) 22 (0 .5) 1 .3
EQ04 0 .16 (7 .0) 1 .1 (50) 22 (0 .5) 1 .7
EQ05 0 .18 (8 .0) 1 .1 (50) 22 (0 .5) 1 .6
EQ06 0 .14 (6 .0) 1 .1 (50) 22 (0 .5) 1 .9
EQ07 0 .09 (4 .0) 0 .9 (40) 22 (0 .5) 2 .8
EQ08 0 .09 (4 .0) 0 .9 (40) 22 (0 .5) 2 .6
CH11 80% EQ01 0 .11 (5 .0) 1 .1 (50) 22 (0 .5) 2 .2
EQ02 0 .23 (10 .0) 1 .1 (50) 22 (0 .5) 0 .9
EQ03 0 .16 (7 .0) 1 .1 (50) 22 (0 .5) 0 .9
EQ04 0 .14 (6 .0) 1 .1 (50) 22 (0 .5) 0 .9
EQ05 0 .14 (6 .0) 1 .1 (50) 22 (0 .5) 0 .9
EQ06 0 .11 (5 .0) 1 .1 (50) 22 (0 .5) 1 .4
EQ07 0 .11 (5 .0) 1 .1 (50) 22 (0 .5) 1 .8
EQ08 0 .11 (5 .0) 1 .1 (50) 22 (0 .5) 2 .4
EQ09 0 .11 (5 .0) 0 .9 (40) 22 (0 .5) 1 .9
EQ10 0 .36 (16 .0) � 1 .1 (60) � 440 (�10) � 0 .9
� Indicates sine-sweep motion.
117
ªrep ¼ S 3 Ł (3) The vertical stress can be split into two components; a
constant component across the entire base and a linear
G0 (S=R)Ł
M¼ � � (4) component, which varies from zero at one end to a maxi-
S jŁj � ªe a mum at the other end of the foundation. It is this linear
1þ
ªr component that provides the restoring moment to the foun-
dation. If the maximum difference in stress between the two
Concern may arise from the lack of cut-off in the adopted edges of the foundation is 2rŁb, where r is a scalar to
hyperbolic model; however, the peak shear strains within the convert displacement into vertical stress (i.e. the subgrade
soil were approximately 5% (as determined from the image reaction modulus), the moment about the centre of the
analysis) and therefore it is being assumed that no cut-off is foundation being provided by the stress distribution can be
required prior to this level of induced strain. It is worthwhile evaluated using equation (5). A further conversion factor, f,
to comment on the intrinsic parameters of the hyperbolic is used to convert the peak gradient of the vertical stress
model proposed by Oztoprak & Bolton (2013). Although into a representative shear stress, as shown by equation (6).
average values for the ªe and ªr parameters are proposed, Combining equations (2), (5) and (6) allows the parameter R
relationships linking their value to the confining stress are to be evaluated as shown by equation (7). This indicates
also presented. Similarly, parameter a is related to the coef- that, at low rotation, before uplift has occurred, R is
ficient of uniformity of the sand. Values for parameters G0, independent of the rotation and peak normal stress, only
a, ªe and ªr were hence calculated as presented in Table 3. being a function of the factor converting normal stress to
The confining stress was calculated at a depth of a quarter shear stress, f. Further comments on the choice of parameter
of the foundation width which, from image analysis of the f will be made later.
deformation mechanism, appeared to be a sensible represen- 2rŁb3
tative depth. The small change in confining stress with M¼ (5)
changing density did not vary the values calculated for ªe 3
and ªr, as shown in Table 3, and hence a single value can be �rep ¼ f 3 (2rŁb) (6)
used in the subsequent analysis. � 3f
It could be postulated that the strain in the mechanism R(Ł) ¼ ¼ 2 (7)
M b
would be directly proportional to foundation rotation and
hence S might be constant. Conversely, the shear stress With regard to the uplift case, a linear variation in the
conversion parameter, R, would be anticipated to be a func- bearing pressure is again assumed in order to maintain
tion of the rotation magnitude Ł, as the mechanism changes simplicity in the overall model, as shown in Fig. 3. In this
size and shape. R will also depend on whether the founda- case an extra parameter, �, defining the distance to the uplift
tion is in full contact with the ground (non-uplift – ‘NUL’) point from the centre of the foundation, has been added.
or has rotated sufficiently to form a gap between its bottom The value of � can be determined by resolving forces
surface and the underlying sand (uplift – ‘UL’). vertically, as shown in equation (8), which results in � being
In order to understand the relationship between shear a function of the foundation width (b), the nominal bearing
stress and moment, the bearing pressure distribution beneath pressure (�n), the rotation (Ł) and the subgrade reaction
the foundation (and how it changes as rotation increases) modulus (r). By calculating the restoring moment about the
needs to be considered. Turning attention first to the non- centre of the foundation, the restoring moment can be
uplift case, a simple linear distribution of bearing pressures evaluated as shown in equation (9). Similarly to the NUL
is adopted, as shown in Fig. 2. It may be postulated that a case, the representative shear stress is taken as a factor, f,
more complex pressure distribution, similar to that found times the variation in normal stress across the foundation
under a static, rigid, shallow foundation, should be con- (equation (10)). Combining equations (2), (9) and (10)
sidered. Despite the static case being well documented, there results in the formulation for parameter R shown in equation
is little information on the stress distribution under a founda- (11). Unlike the NUL case, R is a function of r, Ł and f.
tion sited on sand being subjected to true dynamic loading, � �0.5
4b� n
with the inertia of the foundation and soil both likely to ?:�¼ �b (8)
affect the true distribution. Therefore, in the lack of a proven rŁ
alternative, a linear distribution is the most sensible choice. 2b� n
M¼ (2b � �) (9)
3
2b
�rep ¼ f 3 [rŁ(b þ �)] (10)
Constant � �0.5
component 4b� n
θ frjŁj
� rjŁj
Gradient R(Ł) ¼ ¼ " � �. # (11)
component: M 2b� n 4b� n 0 5
Vertical
stress rθ2b 3b �
3 rjŁj
Fig. 2. Assumed bearing pressure distribution under foundation � b

for non-uplift case
Table 3. Values taken for intrinsic model parameters

Gradient
component:
Relative density 50% 80% rθ(b �)
Vertical
p9: kPa 60 .0 60 .6 stress
ªe: % 6 .5 3 10�4 6 .5 3 10�4
ªr: % 3 .7 3 10�2 3 .7 3 10�2
G0: MPa 75 99 Fig. 3. Assumed bearing pressure distribution under foundation
for uplift case
118
The modified hyperbolic model parameter R can therefore rotation stiffness percentage errors. These two tests have
be evaluated for different rotation magnitudes if values of r different relative densities but the same magnitude of earth-
and f are known. With reference to the f parameter, it was quake, with the good comparison being for dense sand. It is
found that when different values were implemented in the interesting to note that if the small-strain shear modulus
optimisation code, the other model parameters, r and S, (G0) calculated for the dense test was used for the loose test
countered the change, resulting in approximately the same model then the model would predict the data more accu-
overall quality of fit between the model and the data. Thus rately. It could therefore be the case that the density was not
parameter f has been set to be equal to one in the following achieved precisely for the loose test, or the previously
analysis, as this was found to result in the best fit after induced earthquake had caused sufficient densification as to
optimisation of the r and S parameters. This leaves the increase the shear modulus towards the dense sand magni-
parameters r and S to be determined by calibration of the tude. Fig. 5 shows the error distribution curves resulting
model against the experimental data. from implementing the generalised parameters to model the
tests used to obtain the parameters. Based on these distribu-
tions, with a 95% confidence level, the model predicts the
Choosing optimal model parameters peak moments to within 30% and the small-rotation stiffness
Having established the appropriate equations it is now to within 50%. The large stiffness error is likely to be a
possible to calibrate the model against the centrifuge data in result of inaccurate values for the small-strain shear mod-
order to determine the two unknown parameters, r and S. ulus. The model could be used to back-calculate appropriate
Owing to the non-linear response to changes in the r and S G0 values by using the generalised model parameters and
parameters, it is not possible simply to determine the ideal determining the best choice of G0 to best fit the model to
parameter choices for each dataset and then take an average the data. With a formulated, calibrated and validated model
of the obtained values. Instead a least-squares analysis based for the moment–rotation backbone curve it is now possible
on the error between the model prediction and the experi- to examine how best to include energy dissipation.
mental moment backbone curve was performed and the best
choice for the r and S parameters was determined. The
model assumes some uplift and hence only tests in which DEVELOPMENT OF ENERGY DISSIPATION MODEL
uplift was observed were taken for the back-calculation of Damping in soils is known to be largely independent of
the r and S parameters. It worth noting that the tests with loading frequency (Pyke, 1979) and therefore a hysteretic
evident uplift were only the ones with pseudo-dynamic factor damping model is most suitable for including damping in
of safety values of less than one (Table 2). Generalised the moment–rotation cycles. Hysteretic damping could be
parameter values for r and S of 4 .3 3 107 and 0 .21, respec- included within the model through a mathematical construct
tively, were obtained from the back-calculation analysis. which includes a damping coefficient and an imaginary unit.
Figure 4 compares the data and model moment–rotation The imaginary unit is required to synchronise the damping
curves for two datasets: one with the greatest and one with with the rotational velocity as opposed to the rotation.
the least error when using the generalised model parameters. Alternatively, the hysteretic damping can be included
Also quoted in the figure are the peak moment and small- through a purely analytical method as proposed by Masing
(1926) and Pyke (1979). Masing proposed a set of rules
which use the initial backbone load–unload curve to develop
0·2 Stiffness error: 47% a representation of the fully damped cyclic response. The
Moment error: 30% rules proposed by Masing are outlined below and are shown
diagrammatically in Fig. 6 for a simple stress–strain cycle.
0·1
Moment: MN m/m
Test: CH10 EQ2
0 3·5
3·0
0·1
2·5
0·2
0·04 0·02 0 0·02 0·04
(a) 2·0
Probability
0·2 Stiffness error: 13%

Moment error: 3·8% 1·5
0·1
Moment: MN m/m
1·0
Test: CH10 EQ2
0 0·5
0·1 0
Data 60 40 20 0 20 40 60
Error: %
Model
0·2 Moment error (normal distribution)
0·04 0·02 0 0·02 0·04
Rotation: rad Stiffness error (normal distribution)
(b)
Fig. 4. Data–model moment–rotation backbone comparisons: Fig. 5. Error distributions resulting from use of generalised
(a) worst case; (b) best case model parameters
119
1·0 0·4
Step 1: Loading backbone

0·2
0·5
Moment: MN m/m
Step 3: Loading
backbone scaled by 2
0
τ/τy
0
Step 2: Unloading
backbone scaled by 2
0·2
0·5
0·4
0·04 0·02 0 0·02 0·04
1·0 (a)
5 0 5 0·4
γ/γy
Fig. 6. Application of original Masing rules
0·2
1. The shear modulus on each loading reversal assumes a
Moment: MN m/m
value equal to the initial modulus for the initial loading
curve.
2. The shape of the unloading or reloading curve is the same as 0
that of the initial loading curve, except that the scale is
enlarged by a factor of two in both the x- and y-directions.
The second of the original Masing rules detailed above 0·2

fails to deal with asymmetrical loading, for example when
there is a smaller load–unload cycle within a larger overall
load–unload cycle. Such a loading pattern results in accumu-
lation of shear stress beyond the yield shear stress. Pyke 0·4
0·04 0·02 0 0·02 0·04
(1979) proposed modifications to the original Masing rules Rotation: rad
in order to deal with such scenarios. These modifications (b)
would need to be considered when real earthquake motions
are modelled; however, for the constant amplitude ground Fig. 7. Application of (a) original and (b) modified Masing rules
motions used during the experimental testing presented in to moment–rotation backbone
this paper, the original Masing rules are sufficient. Direct
implementation of the Masing rules to the moment–rotation however, the cycle still does not follow precisely the char-
backbone results in a cycle as shown in Fig. 7(a). Although acteristic moment–rotation shape, as shown in Fig. 7(b).
significant damping has been included, as shown by the area The scaling of the sections that follow a load reversal is
enclosed within the loops, the cycles no longer follow the key in defining the cycle shape. Although scaling these legs
typical shape of the moment–rotation curves. The Masing by a factor of two for simple stress–strain cycles, as
rules were developed for a simple load–unload cycle applied originally proposed by Masing, is applicable for those situa-
to a soil column, hence at all times during the loading and tions, the complex moment–rotation behaviour being mod-
unloading cycle there will be a geometrically identical elled and the requirement to divide the backbone into two
mechanism at work. As discussed previously, the moment– sections necessitates further modifications to the original
rotation loading consists of two mechanisms; pre-uplift and rules. In order to obtain a moment–rotation model more
post-uplift. Hence the simple application of the Masing rules representative of the true moment–rotation behaviour, the
to the entire moment–rotation backbone curve does not magnitude of the scaling applied to the section after a load
result in a correct representation of the energy dissipation. reversal point requires further examination.
As Masing rules apply to a scenario in which a consistent It should be noted that there is some asymmetry in the
mechanism acts during loading and, given the two mechan- experimental data, but owing to the inherent symmetry of
isms (non-uplift and uplift) at work during moment–rotation the model, a symmetric rotation profile is input. This results
loading, modifications to the rules are required to allow in some difference in the backbone curves but this approach
them to accurately model the moment–rotation cycles. The is required so that when energy dissipation is added to the
initially proposed modifications involve splitting the back- model, drift of the cycles along the moment or rotation axis
bone curves at the uplift point and thus having four separate does not occur. A peak rotation, which is the mean of the
sections instead of two; non-uplift loading and unloading, positive and negative experimental peak rotations, was used
and uplift loading and unloading. The moment–rotation path for the analysis.
would now follow the listed sequence of sections: loading After thorough investigation it was found that bespoke
uplift – unloading uplift – double unloading non-uplift – scaling was required post load-reversal in order to model the
unload uplift – loading uplift – double loading non-uplift true moment–rotation cycles with different scalars being
(Fig. 7(b)). This set of modified rules still results in the applied in the rotation (RS) and moment (MS) directions, as
same change in moment and rotation as the original rules; shown in Fig. 8. The magnitude of moment reduction post
120
RS values, the error between the predicted cycle and an
average experimental cycle was calculated. A least-squares
δM
method was applied to determine the optimal parameters.
δθ Values of 0 .84 and 1 .07 for RS and MS, respectively, were
obtained for the dense sand tests, while for loose sand,
values of 0 .84 and 1 .50 were obtained. This indicates that
indeed there is an effect of relative density on the overall
cycles beyond its influence on the backbone curve. A larger
value for MS implies a larger area within the moment–
MS δM rotation cycle, indicating more energy dissipation – an
expected characteristic for loose sand. Fig. 9 shows the
RS δθ damping errors (between the model and the data) obtained
for ‘optimal choice’ and generalised parameters with the
dense sand tests. Further improvements to the accuracy of
the model could be made with further experimental testing,
providing more data against which the model parameters can
be calibrated.
Moment
DISCUSSION AND VALIDATION OF MODEL

Although Figs 5 and 9 give some indication of the success
of the model, it is worthwhile to examine more closely the
accuracy of the predictions made by the model. Using the
generalised parameters, as summarised in Table 4, different
measures of error between the model and the experimental
data can be obtained as shown in Fig. 10. It should be noted
that the presented generalised parameters are currently only
Rotation valid for prototype scenarios similar to what was modelled
during this testing programme – a rigid, shallow, raft founda-
Section Scaling
Load UL (RS, MS)
tion sited on dry sand and subjected to horizontal sinusoidal
excitations, up to 0 .25g in magnitude, which propagate
Unload UL (2–RS, 2–MS)
2 1·0
NUL
Fig. 8. Proposed modifications to Masing rules involving situation-

0·8
dependent scaling
Normalised probability
load reversal is the moment scalar, MS, times the moment

0·6
change across the uplift backbone section, with the same
rule being applied to the rotation magnitude. Following this
initial scaled uplift reversal section a double non-uplift
(NUL) section follows, as this was found to follow the 0·4
pattern of the data successfully. To ensure moment or
rotation drift does not occur, the end point of the reversal
path must coincide with the opposite end of the backbone 0·2
curve. Therefore, the uplift section prior to the next reversal
point is scaled by 2-RS and 2-MS. In this way, the same
change in moment and rotation occurs as if the original 0
Masing rules were followed. As can be observed in Fig. 8, 40 20 0 20 40
Error: %
the shape formed following these rules now appears much
more like the expected form of a moment–rotation cycle. Best RS, MS, r and S per test
Therefore this set of modified Masing rules has been
All generalised parameters
adopted and the optimal RS and MS values can be deter-
mined by calibration against the experimental dataset.
Fig. 9. Damping error distributions resulting from use of
In a similar manner to the determination of the r and S optimum and generalised backbone model parameters with
parameters for the backbone curve, the optimum values for obtained RS and MS values for dense tests
the moment and rotation scalars can be determined by find-
ing the combination of values that best fits the experimental
data. The ability to include damping accurately through this Table 4. Summary of generalised model parameters
modified Masing method relies heavily upon an accurate
backbone curve being available. In order to calibrate the Parameter Value
moment and rotation scalars reliably, the optimum values for
r and S were used for each test instead of the generalised Subgrade reaction modulus, r 4 .3 3 107
parameters, thus ensuring the highest possible accuracy of Rotation-shear strain scalar, S 0 .21
the backbone shape. Soil density will affect the amount of Dense rotation scalar, RS 0 .84
Dense moment scalar, MS 1 .07
damping obtained and despite the backbone model taking Loose rotation scalar, RS 0 .84
account of density, it is diligent to examine different relative Loose moment scalar, MS 1 .50
densities separately in the analysis. For a range of MS and
121
4
be expected to reduce as anomaly tests would not bias the
overall error statistics to the same degree. In fact, further
3 testing was performed which included a further four centri-
fuge tests and 45 dynamic excitations. Unfortunately during
Probability
these tests an issue with the experimental set-up resulted in

2 some restriction being applied to the free-movement of the
foundation, hence these data have been excluded from the
previous discussion. If an identical analysis procedure, as
1 was described previously, is used with the data from these
additional tests, the model is still shown to be effective at
modelling the response, albeit with different r and S param-
0
100 50 0 50 100 eter values (due to the experimental issue) and an increase
Error: % in the standard deviation of the errors obtained (due to the
(a)
inconsistency of the experimental issue). This does add
1·5 further reassurance as to the validity of the proposed model.
However, further testing is still advisable, not only for the
reasons outlined above but also because testing a more
diverse range of prototype scenarios would allow the model
1·0
to be refined and validated for use in a wider range of
Probability
design cases.
It is also worthwhile to examine some comparisons be-
0·5 tween the model and experimental moment–rotation cycles.
Fig. 11 shows the results from the test with the minimum
damping error, with an error of 3 .5%. Fig. 12, on the other
hand, shows the case resulting in the largest damping error
0 (from within the subset of tests used to calibrate the model)
100 50 0 50 100
Error: % between the model and experimental data – with an error of
(b) 23%. This is again the test shown in Fig. 4, which showed
Moment error (at peak rotation) the largest error with the generalised backbone parameters.
Small-rotation stiffness error It was, however, found that modifying the small-strain shear
Damping error
modulus to the value used for the dense tests corrected the
error in the backbone. As shown in Fig. 12, when the small-
Fig. 10. Error distributions according to different measures for: strain shear modulus is again increased to the same level as
(a) a subset of tests; (b) all tests
0·15
upwards through the sand layer. The subset of six tests
which exhibited uplift behaviour was used in the calibration 0·10
of the final model and the error distributions using the
generalised parameters from Table 4 are shown in Fig.
10(a), with the mean and standard deviation values obtained 0·05
Moment: MN m/m
being summarised in Table 5. Therefore, the maximum
CH11F3E3
expected error with a 95% confidence level would be 33%, 0
51% and 34% for the moment, stiffness and damping,
respectively. Fig. 10(b) on the other hand shows the distribu-
tion of error resulting from applying the generalised param- 0·05
eters to a larger subset of tests, in which sufficiently large
rotations occurred such that the model was able to be 0·10
applied (ten tests), with the mean and standard deviation
values being given in Table 5. As observed from the figure
and the values quoted in Table 5, the errors increase by 0·04 0·02 0 0·02 0·04
approximately three times when the entire set of tests is Rotation: rad
examined. However, it must be remembered that several of
the earthquakes fired during the tests were very small and Experiment
hence uplift was minimal, resulting in difficulty determining Model
accurate values for the damping within the cycles. With
further testing and an expansion of the useable dataset Fig. 11. Comparison between experimental data and model –
against which the model can be calibrated, the errors would smallest error
Table 5. Mean and standard deviation of errors
Error type Subset of six tests Subset of ten tests
Mean: % Standard deviation: % Mean: % Standard deviation: %
Moment 7 .4 12 .7 20 .8 26 .7
Small-rotation stiffness 7 .9 21 .4 25 .3 35 .5
Damping 9 .2 12 .5 33 .7 41 .2
122
0·15
remembered that they were obtained from applying a model
developed to capture uplift behaviour to experimental situa-
0·10 tions in which significant amounts of uplift did not always
occur and hence increased comparative errors are inevitable.
0·05
It has been shown, however, that further validation and
Moment: MN m/m
calibration of the model, especially regarding the selection
CH10F2E2
of small-strain shear modulus, could lead to a rigorous and
0 easily implementable model. The ability of the model pre-
sented in this paper to model the changing mechanism under
0·05 the foundation in such a simple manner, and yet still to
accurately replicate experimental data acquired from true
dynamic centrifuge testing, thus presents a novel contribu-
0·10 tion to this field.
It should be noted that many proposed models, such as
those proposed by Gajan & Kutter (2009) and Chatzigogos
0·04 0·02 0 0·02 0·04 et al. (2011), are capable of predicting the full moment–
Rotation: rad
rotation–settlement behaviour, whereas the model presented
Experiment here predicts the moment–rotation behaviour only. However,
Model this new model is comparatively simpler, with significantly
fewer user-defined input parameters required.
Model with higher G0
Naturally there are limitations that require further explora-
tion in the proposed model, given that one specific prototype
Fig. 12. Comparison between experimental data and model –
scenario was tested; these include, for example, how the
largest error
rigidity of the foundation affects the accuracy of the model.
In addition, experimental data on the stress distribution
under a rocking foundation would be useful to further refine
that used for the dense sand tests, the model very accurately the relationship between applied moment and representative
predicts the peak moment and energy dissipation. This high- shear stress.
lights the importance of accurate evaluation of the small-
strain shear modulus when implementing the model de-
scribed in this paper. CONCLUSIONS
As discussed in the introduction to this paper, there have In this paper a model for the moment–rotation behaviour
been numerous other macro-element models proposed by of shallow raft foundations located on dry sand beds and
researchers striving to encapsulate the moment–rotation re- subjected to medium-sized seismic excitations has been
sponse of shallow foundations. The majority of these mod- developed and validated. This is intended to be included as
els, including the three presented earlier (Paolucci et al., a macro-element within an overall numerical model of the
2007; Gajan & Kutter, 2009; Grange et al., 2009), are entire soil–foundation–structure system. Appropriate simpli-
validated from data acquired from tests performed at an fying assumptions have been made such that the final model
incorrect stress level and/or without true dynamic loading did not become overly complex, a key element in the
being applied. For example, Paolucci et al. (2007) present novelty of this model. Even with these simplifying assump-
moment–rotation cycles obtained from small-scale tests per- tions the model was found to be able to reliably predict the
formed at 1g with pseudo-dynamic loading being applied to observed experimental behaviour obtained from centrifuge
a square shallow pad foundation. The peak magnitude of testing, provided the small-strain shear modulus could be
rotation applied to the foundation was 3 mrad, compared to accurately determined. The peak moments and energy dis-
a rotation magnitude of around 20 mrad recorded during the sipated were replicated reliably with a maximum damping
experimental programme described in this paper. As Paoluc- error of around 20%. The ability of such a simplified model
ci et al. (2007) did not subject the foundations to substantial to perform reliably potentially paves the way, following
rotation, uplift did not occur and therefore only the almost validation against a wider range of prototypes, for it to be
linear non-uplift behaviour is observed. Although the numer- included within an appropriate model of an overall soil–
ical model prediction successfully mimics the experimental structure–foundation system.
data, it is unclear what would happen if the model were to
be used for situations with larger magnitudes of rotation
when uplift does occur. Similarly, Chatzigogos et al. (2011) ACKNOWLEDGEMENTS
develop a theoretically rigorous model and provide compari- The authors would like to acknowledge the collaborative
sons against numerical simulations, other proposed macro- and financial support received through the European Com-
element models and two different sets of data, with generally munity’s Seventh Framework programme (FP7/2007–2013)
favourable comparisons being presented. However, the data- under grant agreement number 227887 (SERIES – Seismic
sets used for validation are again from testing in which the Engineering Research Infrastructures for European Syner-
complicated nature of true dynamic loading was not con- gies). The support from the staff at the Schofield Centre,
sidered and hence the validation is thus far limited. A University of Cambridge is also gratefully acknowledged.
particular strength of this new model arises from the fact
that it was calibrated using true dynamic centrifuge data and
thus questions regarding differences between the model and
the real design scenario are avoided. The data used for the NOTATION
a dimensionless parameter from Oztoprak & Bolton (2013)
comparisons presented in this paper have not been specially b half footing width
selected to show particularly favourable correlations; instead f model parameter – vertical to shear stress ratio
an open appraisal of the model has been presented, at least G shear modulus
for the prototype examined. Errors such as those presented G0 small strain shear modulus
in Fig. 10 might initially cause concern; however, it must be g acceleration due to gravity (taken to be 9 .81 ms2)
123
M moment Holden, T., Restrepo, J. & Mander, J. B. (2003). Seismic perform-
MS model parameter – moment scalar ance of precast reinforced and prestressed concrete walls. J.
p9 effective mean normal stress Structl Engng 129, No. 3, 286–296.
R model parameter – shear stress scalar Loli, M., Knappett, J. A., Brown, M. J., Anastasopoulos, I. &
RS model parameter – rotation scalar Gazetas, G. (2014). Centrifuge modeling of rocking-isolated
r subgrade reaction modulus inelastic RC bridge piers. Earthquake Engng Structl Dynam. 43,
S model parameter – shear strain scalar No. 15, 2341–2359.
� distance from centre of footing to uplift point Madabhushi, S. P. G., Schofield, A. N. & Lesley, S. (1998). A new
ª shear strain stored angular momentum (SAM) based earthquake actuator. In
ªe strain parameter from Oztoprak & Bolton (2013) Centrifuge ’98: proceedings of the international conference, IS-
ªr strain parameter from Oztoprak & Bolton (2013) Tokyo ’98 (eds T. Kimura, O. Kusakabe and J. Takemura), pp.
ªrep representative shear strain level within the soil deformation 111–116. Rotterdam, the Netherlands: Balkema.
mechanism Masing, G. (1926). Eigenspannungen und verfestigung beim mes-
ªy yield shear strain sing. In Proceedings of the 2nd international congress of
�Ł change in rotation applied mechanics (ed. E. Meissner), pp. 332–335. Zürich,
�M change in moment Switzerland: Oren Füssli.
Ł rotation Osman, A. & Bolton, M. (2005). Simple plasticity-based prediction
�n nominal bearing pressure of the undrained settlement of shallow circular foundations on
� shear stress clay. Géotechnique 55, No. 6, 435–447, http://dx.doi.org/
�rep representative shear stress level within the soil deformation 10.1680/geot.2005.55.6.435.
mechanism Ou, Y.-C., Wang, P.-H., Tsai, M.-S., Chang, K.-C., & Lee, G. C.
�y yield shear stress (2010). Large-scale experimental study of precast segmental
unbonded posttensioned concrete bridge columns for seismic
regions. J. Structl Engng 136, No. 3, 255–264.
REFERENCES Oztoprak, S. & Bolton, M. D. (2013). Stiffness of sands through a
Abate, G., Massimino, M. R., Maugeri, M. & Muir Wood, D. laboratory test database. Géotechnique 63, No. 1, 54–70, http://
(2010). Numerical modelling of a shaking table test for soil– dx.doi.org/10.1680/geot.10.P.078.
foundation–superstructure interaction by means of a soil consti- Pampanin, S. (2005). Emerging solutions for high seismic perform-
tutive model implemented in a FEM code. Geotech. Geol. ance of precast/prestressed concrete buildings. J. Advd Concrete
Engng 28, No. 1, 37–59. Technol. 3, No. 2, 207–223.
Anastasopoulos, I. & Kontoroupi, T. (2014). Simplified approximate Paolucci, R., Prisco, C. & Vecchiotti, M. (2007). Seismic behaviour
method for analysis of rocking systems accounting for soil of shallow foundations: large scale experiments vs. numerical
inelasticity and foundation uplifting. Soil Dynam. Earthquake modelling and implications for performance based design. Pro-
Engng 56, 28–43. ceedings of the 1st US–Italy seismic bridge workshop, Pavia,
Anastasopoulos, I., Gazetas, G., Loli, M., Apostolou, M. & Geroly- Italy.
mos, N. (2010). Soil failure can be used for seismic protection Pender, M. J. (2010). Integrated earthquake resistant design of
of structures. Bull. Earthquake Engng 8, No. 2, 309–326. structure–foundation systems. Proceedings of the 5th interna-
BSI (British Standards Institution) (2004). BS EN 1997-1:2004: tional conference on recent advances in geotechnical and earth-
Eurocode 7. Geotechnical Design. General Rules. London, UK: quake engineering and soil dynamics, San Diego, CA, USA,
British Standards Institution. paper no. SOAP 7.
Butterfield, R. & Gottardi, G. (1994). A complete three-dimensional Pyke, R. M. (1979). Nonlinear soil models for irregular cyclic
failure envelope for shallow footings on sand. Géotechnique 44, loadings. J. Geotech. Engng Div., ASCE 105, No. 6, 715–726.
No. 1, 181–184, http://dx.doi.org/10.1680/geot.1994.44.1.181. Raychowdhury, P. & Hutchinson, T. (2009). Performance evaluation
Chatzigogos, C., Figini, R., Pecker, A. & Salencon, J. (2011). A of a nonlinear Winkler-based shallow foundation model using
macroelement formulation for shallow foundations on cohesive centrifuge test results. Earthquake Engng Structl Dynam. 38,
and frictional soils. Int. J. Numer. Analyt. Methods Geomech. 35, No. March, 679–698.
May, 902–931. Schofield, A. N. (1980). Cambridge geotechnical centrifuge opera-
Davis, E. H. & Booker, J. R. (1971). The bearing capacity of strip tions. Géotechnique 30, No. 3, 227–268, http://dx.doi.org/
footings from the standpoint of plasticity theory. Proceedings of 10.1680/geot.1980.30.3.227.
the 1st Australian–New Zealand conference in geomechanics, Smith, B. J., Kurama, Y. C. & McGinnis, M. J. (2011). Design and
Melbourne, pp. 276–282. Sudney, Australia: Institution of En- measured behavior of a hybrid precast concrete wall specimen
gineers Australia. for seismic regions. J. Structl Engng 137, No. 10, 1052–1062.
Flavigny, E., Desrues, J. & Palayer, B. (1990). Le sable d’Hostun. Steedman, R. S. & Madabhushi, S. P. G. (1991). Wave propagation
Rev. Française Géotechnique 53, 67–70 (in French). in sand medium. Proceedings of the 4th international conference
Gajan, S. & Kutter, B. L. (2008). Capacity, settlement, and energy on seismic zonation, Stanford, CA. Oakland, CA, USA: Earth-
dissipation of shallow footings subjected to rocking. J. Geotech. quake Engineering Research Institute.
Geoenviron. Engng 134, No. 8, 1129–1141. Wotherspoon, L. & Pender, M. (2010). Effect of uplift modelling
Gajan, S. & Kutter, B. L. (2009). Contact interface model for on the seismic response of shallow foundations. Proceedings of
shallow foundations subjected to combined cyclic loading. J. the 5th international conference on recent advances in geotech-
Geotech. Geoenviron. Engng 135, No. 3, 407–419. nical and earthquake engineering and soil dynamics, San Diego,
Gelagoti, F., Kourkoulis, R., Anastasopoulos, I. & Gazetas, G. CA, USA, paper no. 5 .12a.
(2012). Rocking isolation of low-rise frame structures founded Zhao, Y., Gafar, K., Elshafie, M. Z. E. B., Deeks, A. D., Knappett,
on isolated footings. Earthquake Engng Structl Dynam. 41, No. J. A. & Madabhushi, S. P. G. (2006). Calibration and use of a
7, 1177–1197. new automatic sand pourer. In Physical modelling in geotech-
Grange, S., Kotronis, P. & Mazars, J. (2009). A macro-element to nics: proceedings of the 6th international conference on physical
simulate dynamic soil–structure interaction. Engng Structs 31, modelling in geotechnics (eds C. W. W. Ng, Y. H. Wang and L.
No. 12, 3034–3046. M. Zhang), pp. 265–270. Boca Raton, FL, USA: CRC Press.
124
Bilotta, E. et al. (2015). Géotechnique 65, No. 5, 391–400 [http://dx.doi.org/10.1680/geot.SIP.15.P.016]
Importance of seismic site response and soil– structure interaction in

the dynamic behaviour of a tall building founded on piles
E . B I L OT TA � , L . D E S A N C T I S † , R . D I L AO R A † , A . D ’ O N O F R I O � a n d F. S I LV E S T R I �
A tall public building in Naples (Italy) has recently undergone a seismic vulnerability assessment,
following the new Italian code requirements. The building is about 100 m high and is founded on a
piled raft floating in a thick layer of soft pyroclastic and alluvial soils. On the basis of a conventional
subsoil classification, the inertial seismic actions on the building would lead to expensive measures
for seismic retrofitting. By contrast, if site effects and soil–structure interaction are adequately
addressed the picture is completely different. First, free-field seismic response analyses highlighted the
beneficial effects of a peat layer, acting as a natural damper on the propagation of shear waves. Finite-
element analyses of pile–soil kinematic interaction were then carried out to define the foundation
input motion, which was found not to be significantly affected. The effects of inertial interaction were
evaluated accounting for soil–foundation compliance; they resulted in an increase of the structural
period of vibration, while the overall damping did not change compared to that of the fixed-base
structure. The increased structural period led to further reduction of spectral acceleration. The results
could lead to significant impacts on the seismic assessment of slender buildings founded on piles
embedded in deformable soils.
KEYWORDS: dynamics; footings/foundations; piles; soil/structure interaction
(a) assessment of the seismic input on the foundation

INTRODUCTION accounting for kinematic interaction (FIM ¼ foundation
Soil–structure interaction (SSI) may be an important issue in input motion)
the assessment of the seismic vulnerability of a building. (b) calculation of the dynamic impedance functions asso-
Depending on the relative stiffness between the structure and ciated with vertical and horizontal translation, as well as
the soil–foundation system, it is generally expected that the with torsional rotation and rocking
dynamic SSI induces a significant increase of the fundamen- (c) analysis of the inertial interaction of the building
tal period of the structure and an increase of damping, thus subjected to the FIM and supported by visco-elastic
reducing the seismic demand on the structure (e.g. Veletsos springs, characterised by the impedance functions
& Meek, 1974). determined above.
Recent studies (e.g. Han, 2002) have shown that the
seismic response of a tall building supported on a pile In this paper stages (a) and (b) are described in detail, with
foundation may be difficult to predict correctly, if the com- reference to a case study of a tall building on a pile founda-
plex dynamic interaction problem is not handled with care. tion floating in a deformable subsoil (section entitled ‘Case
Neglecting such interaction, for instance by modelling the study’). The seismic actions on the building are calculated in
tall building as having a fixed base, cannot represent the terms of response spectra by free-field seismic response
actual seismic response, since the overall stiffness of the (section entitled ‘Seismic site response’) and kinematic inter-
system is overestimated and the damping is underestimated. action (section entitled ‘Foundation input motion’) analyses.
Equally, simplifying the problem by modelling a real pile The numerical calculation procedure of the (six-components)
foundation as a fictitious equivalent footing leads to no impedence matrix and the relevant modifications of the
better prediction. Particularly in the case of large-diameter spectral ordinates due to the changes of the natural frequen-
piles, the important contribution of the foundation system to cies and the overall damping are finally assessed (section
the rocking stiffness of the building would be neglected. As entitled ‘Influence of pile–foundation compliance’).
a consequence, too low natural frequencies and too large
displacements would be calculated. In such a case, the
assessment of the seismic vulnerability of the building would CASE STUDY
be inaccurate, hence expensive and likely useless retrofitting The analysed building, located in the eastern area of
could be undertaken to meet the seismic safety requirements. Naples (Italy), is a 29-storey reinforced concrete tower, with
An adequate procedure to consider the soil–foundation– a height of 107 .4 m, built in the early 1980s. The tower,
building interaction is based on the substructures method with a stiffening core, is rigidly connected to a pile founda-
(Gazetas, 1984; Makris et al., 1996; Mylonakis et al., 1997) tion by a reinforced concrete box structure, made up by a
and it is implemented by subdividing the analysis into three lower raft of thickness up to 1 m and an upper 40 cm slab,
different stages joined by vertical reinforced concrete walls 6 m high. The
82 piles are unevenly distributed on a large area of 3300 m2
(Fig. 1); they were drilled in alluvial and volcanic soils with
Manuscript received 1 April 2014; revised manuscript accepted 11
March 2015. a length of 42 m and a diameter varying between 1800 mm
Discussion on this paper closes on 1 October 2015, for further details and 2200 mm (Viggiani & Vinale, 1983; Mancuso et al.,
see p. ii. 1999).
� University of Napoli Federico II, Naples, Italy. The reconstruction of the subsoil layering was based on
† University of Napoli Parthenope, Naples, Italy. the results of boreholes and cone penetration tests (CPTs)
125
BILOTTA, DE SANCTIS, DI LAORA, D’ONOFRIO AND SILVESTRI
actions on the building was hence evident; therefore, the
N d 1·8 m seismic site response (SSR) in ‘free-field’ conditions was
d 2·0 m analysed. Also, the complex SSI involving the building, the
d 2·2 m
ground and the pile foundation was recognised as deserving
of greater insight. The former (SSR) aspect will be sum-
marised briefly in the next section, whereas the latter (SSI)
will be investigated in greater detail in the later sections
entitled ‘Foundation input motion’ and ‘Influence of pile–
foundation compliance’.
Y SEISMIC SITE RESPONSE

On the basis of the available data from geotechnical
investigation, a regular layering, characterised as shown in
Fig. 2, was adopted to carry out one-dimensional SSR
analyses with the linear equivalent approach in the frequency
X domain, by using the code EERA (Bardet et al., 2000).
The decay of normalised shear modulus, G/G0, and the
70 m variation of the damping ratio, D, with the shear strain, ª,
Fig. 1. Piled raft – plan view were defined (Fig. 3) by resonant column tests carried out
on undisturbed specimens of pozzolana (Vinale, 1988) or
based on data from literature for the other soils. Peat behav-
carried out in that area, during the design and construction iour was characterised by using experimental data reported
of the building (Vinale, 1988). by Wehling et al. (2003).
The schematic east–west stratigraphic section (Fig. 2(a)) A visco-elastic bedrock was assumed at 60 m depth, with
shows that the foundation subsoil profile consists of made shear wave velocity Vsb ¼ 800 m/s and damping ratio
ground (R), laying above volcanic ash (C), and pyroclastic D ¼ 0 .5% assigned to the stiff ‘A’ formation, based on
silty sand (cohesionless pozzolana, Ps), alternating with preliminary analyses (Bilotta et al., 2013b) showing that the
alluvial materials (peat, T, and sand, S). Underneath, the amplification function of the subsoil was largely independent
Neapolitan yellow tuff (NYT) is replaced in some zones by of any reasonable assumption about the variability of the VS
weakly cemented pozzolana (Pc), which can be viewed as a below 60 m.
weathered and weaker kind of the same soft rock. This Seven natural accelerograms were extracted from the
formation rests on stiff alternating layers of ash, sand and European strong motion database (ESD), through the soft-
pozzolana (A) with uncertain depth. ware Rexel 3 .5 (Iervolino et al., 2009), compatible with the
The shear wave velocity profile, VS, shown in Fig. 2(b), is spectrum specified by the code for the life safety limit state
based on interpretation of cross-hole and down-hole tests, criteria. Fig. 4 shows with thin lines the response spectra
carried out in the same area down to 60 m (Vinale, 1988). associated with each accelerogram, scaled to amax ¼ 0 .19g.
Below such a depth, no direct measurements of VS were It is worth noting that at structural periods higher than 1 s
available and the profile was extrapolated to about 100 m on the average spectrum (thick grey line) is fully compatible
the basis of the results of deep CPT by means of regional with the code-specified reference spectrum for a ground type
correlations between qc and VS (Rippa & Vinale, 1983). A (black line).
According to the national technical code (NTC, 2008), A comparison between the profiles of the initial as
adopting similar soil classification criteria as Eurocode 8 opposed to the mean mobilised stiffness resulting from the
(CEN, 2003), the ground type might be classifiable between one-dimensional SSR analyses is shown in Fig. 5.
C and D, because the equivalent velocity VS,30 is about The response spectra calculated at the surface for each
180 m/s. In a preliminary study carried out by Bilotta et al. input signal (thin grey lines) and their average (thick black
(2013a), linear pseudo-static finite-element method analysis line) are plotted in Fig. 6. The average input spectrum
of the building was performed by assuming a fixed base and (dotted line) and the code-defined spectra for any possible
by simplifying the complex structure into main seismo- ground type are also shown in the figure. It may be observed
resistant elements modelled with one- and two-dimensional that the type C and D spectral ordinates at periods higher
elements (‘frame’ and ‘shell’, respectively). than 2 s overestimate the mean values predicted by the SSR
The standard code seismic actions relevant to the life- analyses.
safety limit state (SLV) were first used; in this case they
corresponded to a return period as high as 712 years and a
peak reference acceleration of 0 .19g at this site. A para- FOUNDATION INPUT MOTION
metric study of the seismic response of the building was The effect of kinematic interaction between the foundation
carried out by changing the input spectra according to the and the surrounding soil is generally of reducing the motion
different ground types. These first analyses showed that the transmitted to the superstructure. The amount of such a
seismic performance of the tall building was unsatisfactory reduction depends mainly on: excitation frequency, pile
when the code-specified demand spectra corresponded to the diameter and soil stiffness. Increasing the values of the first
most unfavourable ground conditions, namely subsoil class two parameters or decreasing the soil stiffness, will lead to
C or D. higher mismatch between free field and foundation input
Nevertheless, it is worth remembering that the standard motion.
classification criteria may be reliable only when VS continu- The reduction of spectral acceleration due to pile–soil
ously increases with depth: this is not the case, since a clear kinematic interaction was expressed by Di Laora & de
inversion in the velocity profile can be observed in Fig. 2(b), Sanctis (2013) with simplified formulations applicable to a
due to the presence of a relatively shallow layer of peat single pile embedded in homogeneous and two-layer subsoil
between 10 and 12 m. models. For the layered subsoil at hand, such simple for-
The need for the definition of more realistic seismic mulations cannot provide sufficiently accurate results. There-
126
IMPORTANCE OF SEISMIC SITE RESPONSE AND SOIL–STRUCTURE INTERACTION
Building H 107 m
E W
Made ground (R)
Volcanic ash (C)

15 Peat (T) Sand (S)
30
Cohensionless
pozzolana (Ps)
Neapolitan
45
Yellow tuff
Cemented (NYT)
60 pozzolana (Pc)
Alternating layers of
75 ash, sand and pozzolana (A)
90
(a)
VS: m/s
0 200 400 600 800
Made ground (R)
Volcanic ash (C)
Peat (T)
20
Sand (S)
Cohensionless
pozzolana (Ps)
40
Depth: m
Cemented
pozzolana (Pc)
60
Alternating
layers of
ash, sand 80
and
pozzolana (A)
100
(b)
Fig. 2. (a) Ground conditions and (b) shear wave velocity profile
fore the kinematic interaction problem has been analysed by between displacements and forces at the heads of a couple
way of numerical simulations in the frequency domain, by of piles is expressed using flexibility coefficients, by assum-
assuming a linear visco-elastic model for the soil and ing that the presence of a second pile does not affect the
including the fixed head single pile or the pile group. deformation of the loaded pile. For horizontal modes of
To this aim, the code Dynapile 2.0 (Ensoft, 1999) has vibration, it is assumed that the deflections of both piles are
been used, based on the ‘consistent boundary matrix’ method identical. Finally, the raft is assumed to be rigid and clear to
(Kausel, 1974; Blaney et al., 1976). This is a hybrid the soil.
procedure that models the soil–pile interaction through the The mobilised stiffness profile from SSR analyses (bold
finite-element method in the vertical direction and applies line in Fig. 5) has been adopted in the analyses.
closed-form solutions along the horizontal direction. Group Figure 7 shows the ratio between the peak accelerations
effects are taken into account by means of frequency-depen- of the pile head and the free-field ground motion as a
dent interaction factors, according to the superposition ap- function of the loading frequency. It can be noted that, for
proach. For vertical and rocking behaviour, the relationship low frequency, the foundation and the free-field motions are
127
1·0 1·0
Rock input signals
0·8 Average
0·8
Code spectrum, subsoil type A (NTC)
0·6 0·6
Sa : g
G/G0
0·4 0·4
0·2 0·2
0 0
105 104 103 102 101 100 101 0 1 2 3 4
γ: % Structural period, T: s
(a)
25 Fig. 4. Spectra associated with the input signals
Made ground and volcanic ashes
Peat
20 0 .5 Hz, the kinematic interaction is certainly negligible. The
Sand
results shown in Fig. 7 demonstrate that piles are unable to
Pozzolana (PS, PC)
modify the free-field seismic motion, which is instead
15
Alternating layers of ash, strongly affected by the presence of the peat layer. In the
D: %
sand and pozzolana following the foundation input motion has been therefore
10
assumed coincident with the free-field ground motion.
5 INFLUENCE OF PILE–FOUNDATION COMPLIANCE

Simplified SSI model
The dynamic response of a building founded on piles
0 embedded in a deformable soil may be different from that of
105 104 103 102 101 100 101
a similarly excited, identical structure resting on a rigid
γ: %
(b) ground. The factors responsible for such a different behav-
iour are: (a) the flexibility of the pile–foundation system; (b)
Fig. 3. (a) Normalised shear modulus and (b) damping ratio the vibrational energy dissipated by the wave radiation and
plotted against shear strain by the internal soil damping.
The above factors were both addressed for the case under
coincident, whereas only at high frequencies the discrepancy examination. Fig. 8 shows a simple oscillator on a flexible
is noticeable; in addition, minor group effects are observed. foundation, whose dynamic compliance is modelled by two
As a matter of fact, considering that for this specific case springs (K and KR) associated to translational and rotational
study the structural fundamental frequency is of the order of oscillations and a pair of dashpots (c and cR) attached in
amax: g G: MPa
0 0·1 0·2 0·3 0 100 200
Made ground (R)
Volcanic ash (C) Mobilised

stiffness
10 10 from EERA
Peat (T)
Initial
stiffness
Sand (S) 20 20
Back-analysis
of pile load test
Depth: m
Depth: m
30 30
Cohensionless
pozzolana (Ps)
40 40
Cemented 50 50
pozzolana (Pc)
60 60
Fig. 5. Average profiles of initial and mobilised stiffness in SSR analyses and profile from pile load test back-analysis
128
1·4
From the dynamic equilibrium of the replacement oscilla-
Surface signals tor, the fundamental period along the i-axis, T~ i , of a building
1·2
Average surface signals modelled as a single-degree-of-freedom (SDOF) system on a
Average rock signals compliant base, and the associated apparent damping, ~ �i , can
1·0
be expressed in the form (Veletsos & Meek, 1974; BSSC,
2004)
0·8 type D vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u !
Sa: g
C u ki K i h2
0·6 ~ t
Ti ¼ Ti 1 þ 1þ (1a)
B Ki K Łj
A
0·4 !�3
~i
T
~
�i ¼ �0,i þ �ci (1b)
0·2 Ti
0 where Ti is the fundamental period of the fixed-base struc-
0 1 2 3 4
Structural period, T: s ture; ki is the stiffness of the fixed-base structure; h is the
height of the centre of mass of the building computed from
Fig. 6. Acceleration response spectra computed at surface plotted the top of the foundation plate; Ki is the horizontal stiffness
against mean input spectrum and code-defined seismic actions for of the foundation; KŁj is the rotational stiffness around the j-
different ground types axis perpendicular to i; �ci is the structural damping asso-
ciated with oscillations along the i-axis; �0,i is the foundation
1.25
damping factor, that is, the contribution due to both radia-
tion and hysteretic damping of the foundation system.
1.00 Considering that Ti is lower than T~ i , the interaction
reduces the effectiveness of the structural damping. When Ti
Acceleration ratio, ap /aff
is small compared to T~ i , the contribution of the structural

0.75 damping may be significantly reduced. However, this reduc-
tion is usually compensated by the increase in the apparent
damping due to the foundation. Depending on the ratio of T~ i
0.50 over Ti, the apparent damping might be larger or smaller
than the structural damping. The expression for �0,i supplied
Single pile by Veletsos & Meek (1974) is only applicable to the case of
0.25 a shallow foundation resting on an elastic halfspace and,
Pile group hence, it is not suitable for the case of a piled foundation. A
novel exact formulation for structures resting on generic
0
0 1 2 3 4 5 springs and dashpots has been proposed by Maravas et al.
Frequency, f: Hz (2007, 2014).
According to this method, the damping of the overall
Fig. 7. Ratio between foundation and free-field accelerations as a system for vibration mode along the x-axis, for example,
function of loading frequency can be expressed by the following equations
" #
parallel to the springs. The overall system is commonly ~ �X �Ł �CX
�X ¼ S X 2 þ þ
referred to as a ‘replacement oscillator’. In the same figure, øX (1 þ 4�2X ) ø2ŁY (1 þ 4�2ŁY ) ø2CX (1 þ 4�2CX )
the ratio between the mass acceleration, ast, and that of the
free-field motion, aff, is plotted against frequency. The foun- (2a)
dation compliance acts as a low-pass filtering device; as a 2
~ 2X
ø ¼ S X =(1 þ 4~
�X ) (2b)
result, the fundamental frequency of the replacement oscilla-
tor is shifted far apart from the natural frequency of the with
fixed-base structure. In addition, the energy dissipated by the " #�1
piled foundation might lead to an increase of the damping 1 1 1
SX ¼ þ 2 þ 2
ratio of the replacement oscillator, which is referred to as øX (1 þ 4�X ) øŁY (1 þ 4�ŁY ) øCX (1 þ 4�2CX )
2 2 2
‘apparent damping’; thus, a reduction of the peak acceleration
corresponding to the natural frequency is usually expected. (3)
U X UR U
ast/aff
mg
Fixed base
Compliant
h
base
K
c f
KR cR
Fig. 8. Replacement oscillator (left) and transfer functions of the fixed as opposed to the
compliant base structure (right)
129
where by a total of 76 piles (Fig. 9), all having a diameter of 2 m.
rffiffiffiffiffiffiffi Such idealisation is a reasonable approximation for engineer-
KX ing purposes.
øX ¼ (4)
m The Dynapile analyses have been performed by assuming
rffiffiffiffiffiffiffiffi
K ŁY the same profile of the mobilised soil stiffness as for the
øŁY ¼ (5) kinematic interaction analysis (Fig. 5). The code allowed
mh2 first the determination of the impedance functions of a single
are fictitious uncoupled natural frequencies of the system pile and then those of the whole pile group.
under rocking and swaying oscillation of the base, and The elastic properties of the soil representative of the
rffiffiffiffiffiffi response of single piles under vertical loads, and hence the
kX rotational impedance, are also affected by the pile installa-
øCX ¼ (6)
m tion technique. In order to evalute the modification of soil
properties induced by pile installation, the procedure sug-
is the natural oscillation frequency of the undamped fixed- gested by Mandolini & Viggiani (1997) was adopted. Hence
base structure. the small-strain stiffness of each layer was first evaluated
In equations (2) and (3), �X and �ŁY are the damping terms from cross-hole and down-hole investigations (Vinale, 1988).
of the foundation in the vibrational modes along the x-axis The stiffness profiles were then normalised by the stiffness
and around the y-axis, respectively, while �CX is the damping of the first layer to obtain a non-dimensional profile. Finally,
ratio related to the horizontal motion of the structure along the mobilised stiffness was calculated by performing itera-
the x-axis. tively an elastic analysis of the single pile, until matching
By approximating to unity the terms expressed as the initial slope of the experimental load–settlement curve.
(1 þ 4�2 ), with � being any foundation or structural damping The initial vertical stiffness of the single pile as deduced by
term, it is possible to rearrange equations (2) and (3) into averaging four load tests was KZ0 ¼ 2381 MN/m (Mandolini
� � !�1 & Viggiani, 1997). Interestingly, the back-figured stiffness
2 1 1 1 �1 ~ 1 h2 1 profile, shown in Fig. 5, is very similar to the mobilised
ø
~X ¼ þ þ ) kX ¼ þ þ stiffness profile deduced from free-field SSR analyses. This
ø2X ø2ŁY ø2CX K X K ŁY k X
similarity led to the choice of the mobilised stiffness as the
(7a) equivalent stiffness for the subsequent linear elastic analyses.
! ! !
Figure 10 illustrates the real and imaginary part of the
~�X ¼ k~X ~
kX h2 ~
kX
� þ �ŁY þ � impedance functions, (Kh( f ), Ch( f )) and (Kv( f ), Cv( f )), for a
KX X K ŁY k X CX (7b) single pile for horizontal (Fig. 10(a)) and vertical (Fig. 10(b))
modes of vibration. The two components of the impedance
¼ ÆX �X þ ÆŁY �ŁY þ ÆCX �CX function are plotted against frequency up to a value of 5 Hz.
The vertical static stiffness, obtained by extrapolating the
where ~k X is the translational stiffness along the x-axis of the dynamic stiffness function to zero frequency, meets exactly
overall system. the experimental initial stiffness KZ0 ¼ 2381 MN/m back-
The same procedure may be applied to the motion along figured from loading test. The static horizontal stiffness of
the y-direction. the fixed-head single pile is KH0 ¼ 683 MN/m.
Rewriting the expressions by Maravas et al. (2014) as in Figure 11 shows the real and imaginary parts of the
equations (7a) and (7b) has the advantage of offering an impedance functions of the pile group associated with the
insight into the physics of the interaction phenomenon. The horizontal modes of vibration along the x-axis (KX) and
apparent damping is a linear combination of the damping y-axis (KY) and with rocking modes of vibration around the
ratios pertaining to the fixed-base structure, the swaying y-axis (KŁY) and x-axis (KŁX). It is worth noting that the
oscillation and the rocking oscillation of the foundation,
weighted for the three coefficients ÆX, ÆŁY and ÆCX. It can be
verified that the sum of above coefficients is 1. As a result,
if damping ratios �X and �ŁY are equal to �CX, the apparent
damping must be equal to �CX. For very stiff foundation
systems, ÆX and ÆŁY are negligible, and the apparent damp-
ing of the replacement oscillator coincides with that of the d 2·0 m
fixed-base structure.
Since stiffness and damping terms pertaining to the foun-
dation are frequency-dependent, an iterative procedure is
necessary to obtain the apparent damping of overall system.
To this aim, frequencies and damping ratios corresponding
to the natural circular frequency of the structure øci can be
used as starting values, thereby calculating by means of
equation (7a) a new value for the frequency ø ~ 2i : Thereafter,
new estimation of impedances may be obtained for ø ~ 2i , until
convergence. Generally, two or three iterations are sufficient Y
to get accurate results. Finally, the value of the apparent
damping may be calculated from equation (7b).
Evaluation of translational and rotational impedance X

The rotational and horizontal stiffness components of the 56·8 m
dynamic compliance have been evaluated by means of the
code Dynapile 2.0 (Ensoft, 1999). The analyses have been Fig. 9. Simplified pile layout considered in the analyses carried
performed by referring to a symmetric layout characterised out with Dynapile
130
1000
The exact solution supplied by Maravas et al. (2014) has
Real part
been then applied to evaluate both the modified period and
Single pile impedance, Kh, Ch: MN/m
Imaginary part the overall damping of the complete system, in order to assess
750 the role of the radiation and hysteretic damping associated
with the foundation motion. The frequency-dependent terms
�i and �Łj, associated with swaying and rotational modes of
500
vibration of the pile foundation, can be easily obtained from
the impedance functions by the following expression
C i (ø)
�i ¼
250 2K i (ø)
(8)
C Łj (ø)
�Łj ¼
2K Łj (ø)
0
0 1 2 3 4 5 where Ci(ø) and Ki(ø) are the imaginary and real parts of the
Frequency, f: Hz dynamic stiffness associated with swaying along the i-axis,
(a)
while CŁj and KŁj are those associated with rocking around
3000
the j-axis.
For the case at hand, the method by Maravas et al. (2014)
Single pile impedance, Kv, Cv: MN/m
provides the results shown in Table 1.

It would be straightforward to show that the increase in
2000 oscillation period along both directions is coincident with
that evaluated by the approach proposed by Veletsos &
Meek (1974); that is, by taking the static stiffness of the
foundation, in agreement with the findings by Maravas et al.
(2014). This implies that the adopted procedure can be
1000 applied without any iteration; in other words, simply by
computing with equation (1a) the natural frequency of the
replacement oscillator. Also, the contribution of the founda-
tion motion to the overall damping is very small in compari-
0
son with that associated with the structure. For example, ÆY
0 1 2 3 4 5 and ÆŁX are only about 5% and 10%, respectively, while ÆCY
Frequency, f: Hz is 85%. Taking into account that �X and �ŁY are very close to
(b)
�CX, the overall damping is practically coincident with that
Fig. 10. Impedance functions associated with (a) horizontal and of the structure (i.e. ~
�X ¼ ~
�Y ¼ 5%). This is mainly due to
(b) vertical modes of vibration for the single pile the fact that the stiffness of the structure, kX, is small in
comparison to the stiffness of the foundation.
horizontal impedance functions, KX and KY, are practically
coincident. The same is true for the rotational components.
The vertical and the torsional components of the stiffness Effect of soil–foundation compliance on seismic actions
matrix, KZ and KT, are not reported since they do not affect The reduction of the seismic actions achieved by taking
the increase of the oscillation period and the apparent into account the site response and the dynamic SSI in the
damping. problem at hand is presented in Fig. 12. The average
For both directions, the value of the real part at zero- spectrum, computed from the SSR accelerograms by assum-
frequency is the static horizontal stiffness, corresponding to the ing ~�X ¼ ~�Y ¼ 5%, is compared to the spectrum for ground
raft restrained against rotation, that must be introduced into type D.
equation (1) in order to evaluate the vibration period of the By considering the effects of SSR on the fixed-base
replacement oscillator. From the Dynapile analyses, the follow- structure, at the first vibration mode (TY ¼ 2 .28 s) the spec-
ing are obtained: KŁY ¼ 1 .316 3 107 MNm (rotational static tral acceleration is reduced by 47 .2% compared to that
stiffness around y-axis), KŁX ¼ 1 .351 3 107 MNm (rotational predicted by the code. An additional reduction (9 .5%) of the
static stiffness around x-axis), KX ffi KY ¼ 6215 MN/m (hori- inertial action can be achieved by accounting for the in-
zontal static stiffnesses around x- and y-axes). crease in the structural period (T~ Y ¼ 2 .47 s). For the second
The natural periods of the fixed-base building along the vibration mode (TX ¼ 1 .62 s), the reduction due to SSR is
two directions x and y were already calculated by Bilotta et equal to 28 .3%; a further significant reduction of 15 .1% is
al. (2013a). The stiffnesses ki of the fixed-base SDOF due to the deformability of the pile group.
equivalent to the building were computed from the first Summarising, the overall reduction of inertial action due
vibration periods assuming a seismic mass equal to 0 .7W/g, to both seismic response and SSI is 56 .8% along the y-axis
with W being the weight of the building (BSSC, 2004). The and 43 .5% along the x-axis.
height was assumed according to BSSC (2004) as h ¼ 0 .7H,
with H being the total height of the building. From such
values, the ratios between the fundamental periods of the CONCLUSIONS
building on compliant and fixed bases, T~ =T , were computed A tall public building in Naples (Italy) recently underwent
according to equation (1) along both directions. The result- a seismic vulnerability assessment, following the new Italian
ing ratios equalled 1 .16 along the x-axis and 1 .08 along the code requirements. After a preliminary unsatisfactory evalua-
y-axis, representing an appreciable increase of the structural tion based on code-specified spectra, the seismic actions
period due to foundation compliance. A reduction of the were re-evaluated by a more sophisticated approach, giving
seimic action on the central core of the building and on the credit to seismic site effects and dynamic SSI.
foundation can be expected on the basis of such an incre- Seismic site response was evaluated by one-dimensional
ment. equivalent linear analyses, which highlighted the beneficial
131
4
3 10 3 104
Real part
Imaginary part
2 104 2 104
KX, CX: MN/m
KY, CY: MN/m

1 104 1 104
0 0
1 104 1 104
2 104 2 104
0 1 2 3 4 5 0 1 2 3 4 5
Frequency, f: Hz Frequency, f: Hz
(a) (b)
3 107 3 107
2 107 2 107
KθY, CθY: MN/m
1 107 KθX, CθX: MN/m 1 107
0 0
1 107 1 107
0 1 2 3 4 5 0 1 2 3 4 5
Frequency, f: Hz Frequency, f: Hz
(c) (d)
Fig. 11. Real and imaginary parts of the impedance functions associated with horizontal swaying along the (a) x-axis and (b) y-
axis, and rocking around the (c) y-axis and (d) x-axis
Table 1. Results of calculations according to Maravas et al. (2014)
TX: s T~ X: s ~f X: Hz KX (~f X): MN/m CX ( ~f X): MN/m �X (~f X) KŁY (~f X): MN m CŁY (~f X): MN m �ŁY (~f X)
1 .62 1 .89 0 .528 5 .311 3 103 5 .461 3 102 5 .14% 1 .312 3 107 1 .185 3 106 4 .52%
TY: s T~ Y: s ~f Y: Hz KY (~f Y): MN/m CY (~f Y): MN/m �Y (~f Y) KŁX (~f Y): MN m CŁX (~f Y): MN m �ŁX (~f Y)
2 .28 2 .47 0 .404 5 .697 3 103 5 .308 3 102 4 .66% 1 .372 3 107 1 .221 3 106 4 .45%
effects of a peat layer, acting as a natural damper on the In order to evaluate the contribution of the combined
propagation of seismic waves. As a consequence, the spec- radiation and hysteretic damping, an exact solution recently
tral ordinates at the surface were reduced by as much as proposed in the literature was adopted, requiring the evalua-
47% and 28%, for the first two vibration modes of the tion of the frequency-dependent impedance components as-
building. sociated with swaying and rocking oscillation of the
The attention was then focused on the role of SSI. Pile– foundation. The overall damping of the compliant base
soil kinematic interaction analyses were first carried out to system was found to be practically coincident with that of
assess the so-called foundation input motion (FIM): the the fixed-base structure. Such a result can be attributed to
results showed that, in this specific case, the filtering action the larger foundation stiffness compared to that of the
exerted by the piles did not affect the FIM. structure.
The pile foundation compliance relevant to inertial inter- Owing to SSR effects and dynamic SSI, the inertial
action was then computed by the numerical code Dynapile, actions were overall reduced by 57% and 43% for the first
referring to mobilised stiffness profile. and second vibration modes. It is therefore inferred that the
The rotational and translational dynamic stiffnesses were assessment of the above factors is mandatory for reliable
found to be poorly affected by the frequency in the range of and sustainable predictions of the seismic performance of
periods corresponding to the first vibration modes of the buildings like the one considered in this study.
building. The swaying and rocking components of the foun-
dation impedance made it possible to evaluate the increase
of the structural period of the compliant base system, which ACKNOWLEDGEMENTS
was found to be 1 .08 and 1 .16 for the first and second The activity was carried out as part of WorkPackage 5
modes, respectively. ‘Soil–foundation–structure interaction’ of the sub-project on
132
1·0
Kv( f ), Cv( f ) real and imaginary parts of the single pile vertical
impedance
KZ0, KH0 static axial and horizontal stiffness of single pile
0·8 Subsoil type D (NTC)
KŁj rotational stiffness of the foundation around j-axis
Seismic response analysis ki stiffness of the fixed-base structure along i-axis
m mass
0·6 qc CPT cone resistance
Sa : g
Sa spectral acceleration
SX model parameter
0·4 Ti fundamental period of the fixed-base structure
T~ i fundamental period of the compliant-base structure
UX, UR, U displacement components of the SDOF system
0·2 VS shear wave velocity in the soil
VS,30 equivalent shear wave velocity defined by Eurocodes
~ Vsb bedrock shear wave velocity
TY TY
0 ÆCX,ÆŁY,ÆX weighting coefficients
0 1 2 3 4 ª shear strain
Structural period, T: s ~
�i apparent damping ratio associated to oscillations
(a) along i-axis
1·0 �ci structural damping ratio associated to oscillations
along i-axis
�i foundation damping ratio associated to oscillations
0·8 along i-axis
�Łj foundation damping ratio associated to oscillations
around j-axis
0·6 �0,i foundation damping factor associated to oscillations
along i-axis
Sa : g
øCX natural circular frequency of the fixed-base structure

along x-axis
0·4
øX,øŁY fictitious frequencies
ø ~ i natural circular frequency of the replacement
oscillator
0·2
~
TX TX
0
0 1 2 3 4 REFERENCES
Structural period, T: s Bardet, J. P., Ichii, K. & Lin, C. H. (2000). EERA a computer
(b) program for equivalent-linear earthquake site response analyses
of layered soil deposits. Los Angeles, CA, USA: University of
Fig. 12. Reduction of the inertial actions for (a) the first and Southern California, Department of Civil Engineering.
(b) the second building vibration modes Bilotta, A., Sannino, D., Fretta, A., Nigro, E. & Manfredi, G.
(2013a). Influenza della categoria di sottosuolo sulla vulnerabil-
‘Earthquake geotechnical engineering’, in the framework of ità sismica di edifici alti. Proceedings ANIDIS conference atti
the research programme funded by Italian Department for del convegno ANIDIS 2013, Padova. Padova, Italy: Padova
University Press (in Italian).
Civil Protection through the ReLUIS (University Network of Bilotta, E., Bilotta, A., Del Prete, I., d’Onofrio, A., Nigro, E. &
Seismic Engineering Laboratories) consortium. Silvestri, F. (2013b). Influenza delle condizioni locali di sottosuo-
lo sulla risposta sismica di un edificio pubblico di notevole
altezza. Proceedings ANIDIS conference 2013, Padova. Padova,
Italy: Padova University Press (in Italian).
NOTATION Blaney, G. W., Kausel, E. & Roesset, J. M. (1976). Dynamic
free field acceleration
aff stiffness of piles. Proceedings of the 2nd international conference
maximum (absolute) acceleration of the time history
amax on numerical methods in geomechanics, Blaksburg, Virginia.
foundation acceleration
ap BSSC (2004). NEHRP recommended provisions for seismic regula-
mass acceleration of the replacement oscillator
ast tions for new buildings and other structures, FEMA 450.
damping coefficient associated to the oscillation of
Ci Washington D.C., USA: Building Seismic Safety Council, Na-
foundation along i-axis, with i ¼ X, Y tional Institute of Building Sciences.
CŁj damping coefficient associated to the oscillation of CEN (2003). (pr)EN 1998-1:2003: Eurocode 8: Design of structures
foundation around j-axis, with j ¼ Y, X for earthquake resistance – Part 1: General rules, seismic actions
c viscous dashpot coefficient associated to the and rules for buildings. Brussels, Belgium: CEN European
translational oscillation of the foundation Committee for Standardization.
cR viscous dashpot coefficient associated to the Di Laora, R. & de Sanctis, L. (2013). Piles-induced filtering effect
rotational oscillation of the foundation on the foundation input motion. Soil Dynam. Earthquake Engng
D soil damping ratio 46, 52–63.
fi natural frequency of the fixed-base structure along i- Ensoft (1999). Dynapile 2.0: A program for the analysis of piles
axis and drilled shafts under dynamic loads. Austin, TX, USA:
~f i natural frequency of the compliant-base structure Ensoft Inc.
along i-axis Gazetas, G. (1984). Seismic response of end-bearing single piles.
G shear modulus Int. J. Soil Dynam. Earthquake Engng 3, No. 2, 82–93.
G0 initial shear modulus Han, Y. (2002). Seismic response of tall building considering soil–
h height of the centre of mass of the building pile–structure interaction. Earthquake Engng Engng Vibration 1,
K horizontal stiffness of the foundation No. 1, 57–64.
Kh( f ), Ch( f ) real and imaginary parts of the single pile horizontal Iervolino, I., Galasso, C. & Cosenza, E. (2009). REXEL: computer
impedance aided record selection for code-based seismic structural analysis.
Ki horizontal stiffness of the foundation along i-axis Bull. Earthquake Engng 8, No. 2, 339–362, http://dx.doi.org/
KR rotational stiffness of the foundation 10.1007/s10518-009-9146-1.
133
Kausel, E. (1974). Forced vibration of circular foundations. ScD Mylonakis, G. E., Nikolaou, A. & Gazetas, G. (1997). Soil-pile-bridge
thesis, Massachusetts Institute of Technology, Cambridge, MA, seismic interaction: kinematic and inertial effects. Part I: soft soil.
USA. Earthquake Engng Structural Dynam. 27, No. 3, 337–359.
Makris, N., Gazetas, G. & Delis, E. (1996). Dynamic pile–soil– NTC (2008). D. M. 14 Gennaio 2008, ‘Norme tecniche per le
foundation–structure interaction: records and predictions. Géo- costruzioni’. Gazzetta Ufficiale della Repubblica Italiana no. 29,
technique 46, No. 1, 33–50, http://dx.doi.org/10.1680/geot.1996. 4 February 2008 (in Italian).
46.1.33. Rippa, F. & Vinale, F. (1983). Experiences with CPT in Eastern
Mancuso, C., Viggiani, C., Mandolini, A. & Silvestri, F. (1999). Naples area. Proceedings of the 2nd European symposium on
Prediction and performance of axially loaded piles under work- penetration testing (ESOPT), Amsterdam. London, UK: CRC
ing loads. In Pre-failure deformation characteristics of geoma- Press.
terials (eds M. Jamiolkowski, R. Lancellotta and D. Lo Presti), Veletsos, A. S. & Meek, J. W. (1974). Dynamic behaviour of
pp. 801–808. Rotterdam, the Netherlands: Balkema. building foundation systems. Earthquake Engng Structural Dy-
Mandolini, A. & Viggiani, C. (1997). Settlement of piled founda- nam. 3, No. 2, 121–138.
tions. Géotechnique 47, No. 4, 791–816, http://dx.doi.org/ Viggiani, C. & Vinale, F. (1983). Comportamento di pali trivellati
10.1680/geot.1997.47.4.791. di grande diametro in terreni piroclastici. Rivista Italiana di
Maravas, A., Mylonakis, G. & Karabalis, D. L. (2007). Dynamic Geotecnica 17, No. 2, 59–84 (in Italian).
characteristics of structures on piles and footings. Proceedings Vinale, F. (1988). Caratterizzazione del sottosuolo di un’area cam-
of the 4th international conference on earthquake geotechnical pione di Napoli ai fini di una microzonazione sismica. Rivista
engineering, Thessaloniki, Paper 1672. Dordrecht, the Nether- Italiana di Geotecnica 22, No. 2, 77–100 (in Italian).
lands: Springer. Wehling, T. M., Boulanger, R. W., Arulnathan, R., Harder, L. F. &
Maravas, A., Mylonakis, G. & Karabalis, D. L. (2014). Simplified Jr Driller, M. W. (2003). Nonlinear dynamic properties of a
discrete systems for dynamic analysis of structures on footings fibrous organic soil. J. Geotech. Geoenviron. Engng 129, No. 10,
and piles. Soil Dynam. Earthquake Engng 61, No. 62, 29–39. 929–939.
134
Conti, R. et al. (2014). Géotechnique 64, No. 1, 40–50 [http://dx.doi.org/10.1680/geot.13.P.031]
Some remarks on the seismic behaviour of embedded cantilevered

retaining walls
R . C O N T I � , G . M . B. V I G G I A N I † a n d F. B U R A L I D ’ A R E Z Z O ‡
This paper is a numerical investigation of the physical phenomena that control the dynamic behaviour
of embedded cantilevered retaining walls. Recent experimental observations obtained from centrifuge
tests have shown that embedded cantilevered retaining walls experience permanent displacements even
before the acceleration reaches its critical value, corresponding to full mobilisation of the soil strength.
The motivation for this work stems from the need to incorporate these observations in simplified
design procedures. A parametric study was carried out on a pair of embedded cantilevered walls in
dry sand, subjected to real earthquakes scaled at different values of the maximum acceleration. The
results of these analyses indicate that, for the geotechnical design of the wall, the equivalent
acceleration to be used in pseudo-static calculations can be related to the maximum displacement that
the structure can sustain, and can be larger than the maximum acceleration expected at the site. For
the structural design of the wall, it is suggested that the maximum bending moments of the wall can
be computed using a realistic distribution of contact stress and a conservative value of the pseudo-
static acceleration, taking into account two-dimensional amplification effects near the walls.
KEYWORDS: design; earth pressure; earthquakes; numerical modelling; retaining walls; shear strength
INTRODUCTION either provided by the technical codes or obtained by one-

In the recent literature several cases are reported of damage dimensional seismic response analyses.
or failure of gravity and cantilevered retaining walls during If the permanent displacement at the end of the earth-
earthquakes (Fang et al., 2003; Madabhushi & Zeng, 2007; quake is taken as a performance indicator, the choice of the
Koseki et al., 2012). Following the pioneering works by equivalent acceleration should be related to the maximum
Okabe (1926) and Mononobe & Matsuo (1929), several displacements that the structure can sustain, with respect to
studies have tackled the problem of computing dynamic different levels of design earthquake motion. The reliability
earth pressures on retaining structures with a theoretical of the choice of an equivalent acceleration depends crucially
(Steedman & Zeng, 1990; Lancellotta, 2007; Mylonakis et on the ability to predict the displacements experienced by
al., 2007; Kim et al., 2010), experimental (Atik & Sitar, the wall during the earthquake.
2010) or numerical approach (Gazetas et al., 2004; Evange- Permanent displacements of retaining walls are usually
lista et al., 2010). In the last decade, following the seminal computed through Newmark (1965) rigid-block analysis
works by Newmark (1965) and Richards & Elms (1979), (Richards & Elms, 1979). According to this method
more and more works have been devoted to the computation
(a) the critical (yield) acceleration of the wall, ac – that is,
of wall displacements, in the light of a performance-based
the acceleration corresponding to which the strength of
design (Ling, 2001; Huang et al., 2009; Basha & Babu,
the soil is fully mobilised – is computed with respect to
2010).
an assumed collapse mechanism, assuming rigid-
In recent years, new performance-based strategies have
perfectly plastic behaviour for both the soil and the wall
been proposed in the literature and included in current codes
(b) for accelerations a(t ) < ac , no relative displacements
of practice (PIANC, 2001; CEN, 2003; NTC, 2008) for the
occur between the soil and the wall, and both the inertia
seismic design of retaining structures. Although charac-
forces into the soil wedge–wall system and the internal
terised by different levels of complexity, all these methods
forces in the structure increase with the applied
rely on the idea that the structure may experience permanent
accelerations
displacements during the earthquake, provided the behaviour
(c) for accelerations a(t ) . ac , the wall experiences perma-
of the system is ductile.
nent displacements, but the internal forces remain
The simplest way to embody the performance-based phil-
constant and equal to the maximum value they attained
osophy in the seismic design of retaining structures is by an
for a(t ) ¼ ac
appropriate choice of the equivalent acceleration to be used
(d ) the permanent displacements are computed by integrating
in pseudo-static calculations, which has to be proportional to
the relative acceleration, a(t ) � ac , twice over the time
the maximum acceleration expected at the ground surface,
intervals in which the relative velocities are non-zero.
The critical acceleration is a key ingredient not only for the
Manuscript received 19 February 2013; revised manuscript accepted computation of the permanent displacements experienced by
11 July 2013. Published online ahead of print 13 September 2013. the wall, but also for its structural design, as it defines the
Discussion on this paper closes on 1 June 2014, for further details see
maximum internal forces that the structure may ever experi-
p. ii.
� Dipartimento di Ingegneria Civile, Università di Roma Tor Vergata, ence during an earthquake.
and International School for Advanced Studies, Trieste, Italy. Centrifuge dynamic tests have shown that Newmark rigid-
† Dipartimento di Ingegneria Civile, Università di Roma Tor Vergata, block analysis provides good results when applied to gravity
Italy. retaining structures (Zeng & Steedman, 2000; Huang et al.,
‡ University of Cambridge, UK. 2009). Moreover, experimental dynamic tests carried out on
137
CONTI, VIGGIANI AND BURALI D’AREZZO
reduced-scale models (Neelakantan et al., 1992; Richards & mobilisation of plastic strain). Based on the results of the
Elms, 1992) and dynamic numerical analyses (Callisto & parametric study, suggestions are provided for the seismic
Soccodato, 2010) have shown that a Newmark-type calcula- geotechnical and structural design of embedded cantilevered
tion may also be adopted, at least qualitatively, to interpret walls using simplified methods.
the dynamic behaviour of embedded cantilevered walls or
retaining walls with one level of support, where the wall can
rotate when a state of limit equilibrium is attained in the ANALYSIS PROCEDURE
adjacent soil. Constitutive soil model
Results from both numerical (Callisto & Soccodato, 2007, The soil is modelled as an elastic-perfectly plastic mater-
2010) and experimental (Zeng, 1990; Zeng & Steedman, ial with Mohr–Coulomb failure criterion, in which, during
1993; Conti et al., 2012) work on the dynamic behaviour of the dynamic stage, non-linear and hysteretic behaviour is
embedded retaining walls, however, have shown that the introduced for stress paths within the yield surface through a
Newmark approach does not describe the observed behaviour hysteretic model available in the library of FLAC 5.0 (Itasca,
satisfactorily. In fact, if the critical acceleration of the 2005). This strategy makes it possible to take into account
system is computed with standard limit equilibrium methods, both the cyclic soil behaviour and the possibility of full
such as those adopted in European countries and the USA mobilisation of soil strength close to the excavation, at the
(Blum, 1931; Padfield & Mair, 1984; King, 1995; Powrie, same time avoiding the cumbersome calibration process and
1996; Day, 1999; Osman & Bolton, 2004), seeking the the high computational costs of more advanced constitutive
pseudo-static coefficient corresponding to which limit condi- models (Kontoe et al., 2012).
tions are attained in the system, two findings are of major The hysteretic model, which is used to update the tangent
concern. shear modulus of the constitutive law for the soil at each
calculation step, consists in an extension to general strain
(a) Embedded walls may accumulate significant rigid
conditions of the one-dimensional non-linear models that
permanent displacements concurrently with an increase
make use of the Masing (1926) rules to describe the unload-
of the internal forces in the structural members: that is,
ing–reloading behaviour of soil during cyclic loading. If the
permanent displacements occur even before the critical
simplified assumption is made that the stress state does not
acceleration is attained.
depend on the number of cycles, the relationship between
(b) Internal forces in cantilevered walls may be substantially
shear stress, �, and shear strain, ª, can be written as
larger than those computed with conventional limit
equilibrium methods in critical conditions. � ¼ GS (ª) � ª
(1)
It follows that, at least for cantilevered walls, standard ¼ G0 M S (ª) � ª
pseudo-static approaches do not provide reliable or conserva-
tive values of the yield acceleration, neither for a displace- where GS (ª) is the secant shear modulus, G0 is the small-
ment-based analysis nor for a pseudo-static calculation. strain shear modulus, and MS (ª) is the normalised secant
On the basis of centrifuge dynamic tests carried out on shear modulus, defined as
pairs of embedded propped and cantilevered walls in dry a
sand, Conti et al. (2012) have shown that a Newmark analy- MS ¼ (2)
1 þ exp �(log10 ª � x0 )=b
sis carried out using the limit equilibrium value of the
critical acceleration would yield displacements that are much where a, b and x0 are model parameters that can be
smaller than observed, as the analysis would overlook the determined from the best fit of a specific modulus degrada-
displacements experienced by the wall before the acceleration curve. The tangent shear modulus, Mt , can be evaluated
tion reaches the limit equilibrium critical value. According by differentiating equation (1) with respect to ª. Strain
to the authors, the observed behaviour may be justified by a reversals during cyclic loading are detected by a change of
stress redistribution and a progressive mobilisation of the the sign of the scalar product between the current strain
soil strength on the passive side of the wall produced by the increment and the direction of the strain path at the previous
earthquake. time instant. At each strain reversal, the Masing rule is
Recent numerical studies of the dynamic behaviour of invoked, and stress and strain axes are scaled by a factor of
embedded retaining walls, both cantilevered (Madabhushi & 0.5, resulting in hysteresis loops in the stress–strain curves
Zeng, 2006, 2007) and with one level of support (Iai & with associated energy dissipation.
Kameoka, 1993; Callisto et al., 2008; Cilingir et al., 2011) As already outlined by Callisto & Soccodato (2010), an
have shown interesting aspects related to the soil–structure advantage of using a truly non-linear soil model for dynamic
interaction and the constitutive modelling of the mechanical numerical simulations is that energy dissipation emerges
behaviour of the soil under cyclic loading. Useful guidelines from the hysteretic behaviour of the soil, and is not intro-
for the seismic design of cantilevered retaining walls may be duced artificially by including a frequency-dependent viscos-
found in Callisto & Soccodato (2010), but these are still not ity in the equilibrium equations.
exhaustive: the simplified procedure proposed by the authors
does not seem conservative, as it is based on a standard
pseudo-static calculation of the critical acceleration. Seismic input
This work is a numerical investigation of the physical Three different acceleration time histories were used in
phenomena that control the dynamic behaviour of embedded the analyses, all registered on rock outcrop during real
cantilevered retaining walls, aimed at developing suitable earthquakes: Tolmezzo (T) from the Friuli earthquake of
simplified procedures to be incorporated in recommendations 1976, Assisi (A) from the Umbria-Marche earthquake of
and codes of practice. A parametric study was carried out 1997, and Arcelik (N) from the Kocaeli earthquake of 1999.
on a pair of embedded cantilevered walls in dry sand, The choice of these three earthquakes is motivated by the
subjected to real earthquakes scaled at different values of fact that they are characterised by substantially different
the maximum acceleration. The earthquakes were chosen to frequency contents. Table 1 shows the maximum values of
represent a significant range of dominant frequency (govern- acceleration, amax,r , duration, T5–95 , mean period, Tm (Rathje
ing local amplification and resonance phenomena) and peak et al., 1998), and Arias intensity, Ia : Fig. 1 shows the
acceleration (governing non-linearity of soil behaviour and acceleration time histories and the Fourier spectra of the
138
THE SEISMIC BEHAVIOUR OF EMBEDDED CANTILEVERED RETAINING WALLS
0:5
Table 1. Ground motion parameters of input earthquakes G0 p9
¼ KG (3)
pref pref
Record amax,r : g T5–95 : s Tm : s Ia : m/s
Assisi 0.28 4.28 0.25 0.78 where p9 is the mean effective stress, pref ¼ 100 kPa is a
Tolmezzo 0.35 4.19 0.40 0.80 reference pressure, and KG is a stiffness multiplier, set equal
Arcelik 0.14 7.23 1.09 0.31
to 1000. Soil parameters a ¼ 1.0, b ¼ �0.6 and x0 ¼ �1.5
were used for the normalised secant shear modulus in equa-
tion (2), derived from the best fit of the modulus degradation
curve proposed by Vucetic & Dobry (1991) for cohesionless
three signals. The recorded signals were baseline-corrected (PI ¼ 0) soils. Fig. 3 shows a comparison between the
and low-pass-filtered at 15 Hz for compatibility with the modulus decay curve and the equivalent damping ratio of
dimension of the grid zones in the numerical domain; more- the adopted model, and that suggested by Vucetic & Dobry
over, they were scaled at maximum accelerations ranging (1991).
from 0.05g to 0.5g. The retaining walls were modelled as elastic beams con-
nected to the grid nodes with elastic-perfectly plastic inter-
faces with a friction angle � ¼ 208. The bending stiffness of
Numerical model the walls was EI ¼ 2.7 3 105 kN m2 /m, corresponding to
Two-dimensional, plane-strain, finite-difference analyses of that of a wall consisting of 0.6 m diameter and 0.7 m
a rectangular excavation of width B ¼ 16 m and depth spacing bored piles. The walls can be considered, for all
h ¼ 4 m, in a layer of dry sand with thickness Z ¼ 16 m,
were carried out. The excavation was retained by a pair of
cantilevered retaining walls. 1·0 30
Figure 2 shows the grid adopted in the numerical ana- 0·8 25
lyses, with an extension of 80 m, consisting of a total of 20
0·6
4838 elements, with a minimum size of 0.33 m near the
D: %
G/G0
Vucetic & Dobry (1991) 15

walls. Both the refinement of the mesh and the extension of 0·4
Numerical 10
the grid were chosen after a preliminary parametric study, in
0·2 5
order that they did not influence the numerical results during
either the static or dynamic stages. 0 0
0·0001 0·001 0·01 0·1 1
The soil was modelled with a constant friction angle γ: %
� ¼ 358, cohesion c9 ¼ 0, and density r ¼ 2.04 Mg/m3 : A
standard non-associated flow rule was used, with angle of Fig. 3. Modulus decay and damping ratio curves (from Conti &
dilatancy ł ¼ 0. The small-strain shear modulus is given by Viggiani, 2013)
0·4 0·02
0·2 Tolmezzo
A: g
a: g
0 0·01
0·2
0·4 0
0·4 0·02
0·2 Assisi
A: g
a: g
0 0·01
0·2
0·4 0
0·4 0·02
0·2 Kocaeli
A: g
a: g
0 0·01
0·2
0·4 0
0 4 8 12 16 20 0 4 8 12 16
t: s f : Hz
Fig. 1. Acceleration time histories and Fourier amplitude spectra of input earthquakes
32 m B 32 m
d
16 m
Bedrock
Fig. 2. Finite difference grid (from Conti & Viggiani, 2013)
139
practical purposes, as infinitely rigid (Callisto & Soccodato, A time increment of ˜t ¼ 6.25 3 10�7 s was adopted
2010). during the dynamic stage, in order to guarantee the stability
The in situ stress state was prescribed in terms of the of the explicit time integration scheme. Moreover, a small
earth pressure coefficient at rest, h9 = v9 ¼ K 0 (¼ 1 � sin ). Rayleigh viscous damping (1%) was adopted to remove the
During the static stage, standard boundary conditions were high-frequency noise deriving from the numerical integra-
applied to the model: that is, zero horizontal displacements tion, but not otherwise affecting the results of the analyses.
along the lateral boundaries, and fixed nodes at the base of A total of 39 numerical analyses were carried out in this
the grid. The excavation was carried out in four successive study, as reported in Table 2. Three different values were
steps, chosen after a preliminary study in order to have the adopted for the embedded depth of the walls: d ¼ 3 m (Nos
numerical results unaffected by the calculation sequence, 1 to 9), d ¼ 4 m (Nos 10 to 21) and d ¼ 5 m (Nos 22 to
during which the soil elements corresponding to 1 m of the 36). For the analysis with d ¼ 4 m and the Tolmezzo (T)
excavated volume were removed. In this stage the shear record scaled to 0.35g, the soil stiffness multiplier was
modulus of the soil was set equal to 0.3G0 , corresponding to halved and doubled (Nos 37 and 38), and a thickness of
a shear strain level of about 0.1% (see Fig. 3), characteristic Z ¼ 30 m was considered for the soil layer (No. 39).
of the expected level of deformation during the excavation
stage (Atkinson et al., 1993).
After the excavation, static constraints were removed from DISCUSSION OF RESULTS
the boundaries. The selected acceleration–time histories Static stage
were applied to the bottom nodes of the grid, together with At the end of the static stage, the soil around the excava-
a zero velocity condition in the vertical direction, thus tion is approximately in limit equilibrium conditions, both in
simulating the presence of an infinitely rigid bedrock. Stan- front and behind the walls. Fig. 4 shows, for the case
dard dynamic constraints (Zienkiewicz et al., 1988) were d ¼ 4 m, the contour lines of: (a) the mobilised shear
applied to the nodes on the lateral boundaries of the grid: strength into the soil, defined as the ratio, /lim between the
that is, they were free to move in both the vertical and maximum shear stress and the corresponding available
horizontal directions, while being tied to one another in strength; and (b) the maximum shear strain (in %) close to
order to enforce the same displacements of the two bound- the left wall. Fig. 4(a) also shows the horizontal stress
aries. distribution in the soil elements at the contact with the wall,
Table 2. Summary of numerical analyses
No. Record ainp : g d: m Z: m EI: kN m2 /m] KG
1 Assisi 0.05 3 16 2.7.105 1000

2 Assisi 0.10 3 16 2.7.105 1000
3 Assisi 0.20 3 16 2.7.105 1000
4 Tolmezzo 0.05 3 16 2.7.105 1000
5 Tolmezzo 0.10 3 16 2.7.105 1000
6 Tolmezzo 0.20 3 16 2.7.105 1000
7 Arcelik 0.05 3 16 2.7.105 1000
8 Arcelik 0.10 3 16 2.7.105 1000
9 Arcelik 0.20 3 16 2.7.105 1000
10 Assisi 0.05 4 16 2.7.105 1000
11 Assisi 0.10 4 16 2.7.105 1000
12 Assisi 0.20 4 16 2.7.105 1000
13 Assisi 0.35 4 16 2.7.105 1000
14 Tolmezzo 0.05 4 16 2.7.105 1000
15 Tolmezzo 0.10 4 16 2.7.105 1000
16 Tolmezzo 0.20 4 16 2.7.105 1000
17 Tolmezzo 0.35 4 16 2.7.105 1000
18 Arcelik 0.05 4 16 2.7.105 1000
19 Arcelik 0.10 4 16 2.7.105 1000
20 Arcelik 0.20 4 16 2.7.105 1000
21 Arcelik 0.35 4 16 2.7.105 1000
22 Assisi 0.05 5 16 2.7.105 1000
23 Assisi 0.10 5 16 2.7.105 1000
24 Assisi 0.20 5 16 2.7.105 1000
25 Assisi 0.35 5 16 2.7.105 1000
26 Assisi 0.50 5 16 2.7.105 1000
27 Tolmezzo 0.05 5 16 2.7.105 1000
28 Tolmezzo 0.10 5 16 2.7.105 1000
29 Tolmezzo 0.20 5 16 2.7.105 1000
30 Tolmezzo 0.35 5 16 2.7.105 1000
31 Tolmezzo 0.50 5 16 2.7.105 1000
32 Arcelik 0.05 5 16 2.7.105 1000
33 Arcelik 0.10 5 16 2.7.105 1000
34 Arcelik 0.20 5 16 2.7.105 1000
35 Arcelik 0.35 5 16 2.7.105 1000
36 Arcelik 0.50 5 16 2.7.105 1000
37 Tolmezzo 0.35 4 16 2.7.105 500
38 Tolmezzo 0.35 4 16 2.7.105 2000
39 Tolmezzo 0.35 4 30 2.7.105 1000
140
1984; Bica & Clayton, 1989; King, 1995; Day, 1999), and a
limit equilibrium method outlined herein. Fig. 5(d) shows
the horizontal displacements of the wall computed at the end
of each earthquake, while in Figs 5(e)–5(i) the computed
contact stresses are plotted together with the dynamic active
1·00
and passive limit values. The earth pressure coefficients, kAE
0·95 (Okabe, 1926; Mononobe & Matsuo, 1929) and kPE (Lancel-
0·90 lotta, 2007), are computed using, at each depth, a pseudo-
0·85
static acceleration equal to the acceleration resulting from
σh: kPa
the numerical analyses (kh ¼ a/g).
0·80
During the earthquakes, amplification phenomena around
0·75 Active and passive the excavation cause the surface accelerations behind the
limit states
0·70 100 50 0 50 100 150
wall to be substantially larger than the maximum input
accelerations, with amplification factors between 1.5 (No.
(a) 17) and 3.8 (No. 14). The distribution of accelerations into
the soil is not uniform, owing to both amplification and
phase shift between the top and the bottom of the wall; in
the time instants when the acceleration reaches its maximum
value, it varies almost linearly with depth behind the wall, at
least in the retained part of the soil, while it is approxi-
mately constant below dredge level, and substantially lower
0·28
than the maximum value at surface (Fig. 5(a)).
0·24 The inertia forces in the retained soil induce an increment
0·20 of the contact stresses behind the wall, the soil being in
0·16
active limit conditions. As a consequence, the wall rotates,
mobilising the passive resistance of the soil below dredge
0·12
level progressively, until the system reaches a new equili-
0·08 brium configuration. The stronger the earthquake, the greater
0·04 the depth down to which the passive resistance of the soil is
fully mobilised (Figs 5(b), 5(f), 5(g), 5(h) and 5(i)). Both
(b) the increment of the contact stresses behind the wall and the
lower position of the resultant of the pressure distribution in
Fig. 4. Static stage (d 4 m). Contours of (a) mobilised shear front of the wall cause a significant increase of the bending
strength, /lim , and (b) maximum shear strain (%) close to left
moments in the wall (Fig. 5(c)). At the end of the earth-
wall
quakes, the horizontal displacements of the wall correspond
to an approximately rigid rotation around a pivot point
together with the theoretical values of the static active and between 0.8d and 0.9d (Fig. 5(d)).
passive pressure, computed with Lancellotta’s closed-form As far as the internal forces in the walls are concerned,
solutions (Lancellotta, 2002). The soil behind the wall is in the maximum bending moments may occur at instants of
active limit state down to 5 m from the surface, whereas in time when the acceleration at surface behind the wall is not
front of the wall the passive resistance is fully mobilised maximum. As an example, Fig. 6 shows, for analysis No. 17
only immediately below dredge level, the horizontal stresses and for the time instants at which a ¼ amax (t ¼ 5.64 s) and
being approximately constant in the remaining part of the M ¼ Mmax (t ¼ 5.70 s): (a) the acceleration profile, (b, c) the
embedded depth. The distribution of maximum shear strain earth pressure distribution, (d) the bending moment distribu-
into the soil is similar to that of the mobilised strength. tion, and (e, f) the contour lines of the ratio /lim : As
Moreover, while a maximum shear strain of 0.3% is mobi- already discussed, when the acceleration behind the wall
lised just below dredge level, an average strain level of reaches its maximum value (t ¼ 5.64 s), the retained soil is
about 0.1% is mobilised into the whole soil volume interact- in active limit state conditions, at least down to z ¼ 6 m,
ing with the wall during the static stage. while all the available resistance in front of the wall is
mobilised down to 2 m from the excavation bottom. At this
time instant, the occurrence of a soil wedge may be ob-
Dynamic behaviour of retaining walls served both behind (active) and in front (passive) of the wall
In this section, the dynamic behaviour of cantilevered (Fig. 6(e)). By contrast, for t ¼ 5.70 s the soil on both sides
retaining walls is discussed with reference to analyses Nos of the wall is far from limit conditions (Fig. 6(f)): that is,
14 to 17, where the walls with d ¼ 4 m are subjected to the the horizontal stresses are higher than the corresponding
Tolmezzo earthquake scaled to maximum input accelerations active values on the retained side, while a constant fraction
ainp ¼ 0.05g, 0.10g, 0.20g and 0.35g. of the passive resistance is mobilised below dredge level
Figure 5 shows, for the time instant when the acceleration (Fig. 6(c)). This stress distribution results in higher bending
behind the right wall reaches its maximum value moments in the wall, even if the accelerations in the retained
soil are about 20–30% lower than the values computed for
(a) the distribution of accelerations immediately behind and
t ¼ 5.64 s. The behaviour exhibited by the soil–wall system
in front of the wall
at this time instant would be hardly reproduced by a
(b) the contact horizontal stresses
simplified limit equilibrium approach.
(c) the bending moment distribution in the wall.
Figs 5(a) and 5(c) also show the critical acceleration of the
wall and the corresponding bending moment distribution, Critical acceleration and limit equilibrium analysis
computed using the Blum (1931) method, customarily The critical acceleration is computed with respect to an
adopted in the UK and other European countries, and assumed failure mechanism, generally a rigid rotation about
described extensively in many works (e.g. Padfield & Mair, a point close to the toe for embedded cantilevered walls,
141
a: g σh: kPa M: kN m/m u: m
0·2 0 0·2 0·4 0·6 300 200 100 0 100 0 200 400 0·2 0·1 0
0
Static
ainp 0·05g (No. 14)
ainp 0·10g (No. 15)
ainp 0·20g (No. 16)
ainp 0·35g (No. 17) 2
Conti & Viggiani (2013)
Blum (1931)
a: g
0·2 0 0·2
z: m
4
ac 0·283g
ac 0·481g
6
8
(a) (b) (c) (d)
0
Static ainp 0·05g ainp 0·10g ainp 0·20g ainp 0·35g
Limit cond. Limit cond. Limit cond. Limit cond. Limit cond.
2
z: m
4
6
8
200 0 200 0 200 0 200 0 200 0
σh: kPa σh: kPa σh: kPa σh: kPa σh: kPa
(e) (f) (g) (h) (i)
Fig. 5. Analyses Nos 14–17 (right wall): (a) distribution of accelerations behind and in front of wall; (b) and (e) to (i) contact horizontal
stresses; (c) bending moment distribution for time instant when acceleration behind wall is maximum; (d) horizontal displacements of
wall at end of each earthquake
a: g σh: kPa σh: kPa M: kN m/m

0·2 0 0·2 0·4 0·6 300200100 0 100 300200100 0 100 0 200 400
0
t 5.64 s
t 5.64 s
Limit cond.
2
t 5·64 s (a amax)
a: g
(e)
0·2 0 0·2
z: m
4
1·00
6 0·90
0·80
0·70
8 t 5·70 s (M Mmax)
(a) (b) (c) (d)
(f)
Fig. 6. Analysis No. 17 (right wall): (a) acceleration profile; (b), (c) earth pressure distribution; (d) bending moment distribution;
(e), (f) contour lines of ratio /lim , for time instants at which a amax and M Mmax
and depends solely on the geometry of the system and the (a) In the Blum method, the same pseudo-static acceleration is
strength of the soil. assumed for the soil in front and behind the wall, whereas,
The results obtained by Callisto & Soccodato (2010) show in general, the acceleration below dredge level is only a
that the Blum method does not provide a reliable or small fraction of the maximum value on the retained side,
conservative estimate of the critical acceleration, and hence and always lower than about 0.1g (see Fig. 5(a)).
of the maximum bending moment that the wall may ever (b) The pivot point is taken to be at a depth of 0.8d from
experience during an earthquake. There are two main rea- dredge level, whereas it may be as deep as 0.9d during
sons for this. strong earthquakes (see Fig. 5(d)).
142
Moreover, for pseudo-static accelerations smaller than the F RHS ¼ 12ªK AE (h þ d 0 )2 þ 12ªK P (2h þ d þ d 0 )(d � d 0 ) (4)
critical value, the Blum method assumes a constant fraction
1 2 1 þ K AE d 0 )(d 0 � d)
of the soil passive resistance in front of the wall, whereas F LHS ¼ 2ªK P d þ 2ª(K P d
during real earthquakes this is progressively mobilised from (5)
the bottom of the excavation, depending on the amplitude of þ 12ªK AE (d þ d 0 )(d � d 0 )
the acceleration applied (see Fig. 5(b)).
To take into account all these observations, a different Similarly, the moment equilibrium can be established by
pseudo-static distribution of contact stresses is used in this taking the moment about the toe of the wall, generated by
work (see Fig. 7). According to this approach (e.g. Conti & the forces acting on the right-hand side (MRHS ) and on the
Viggiani, 2013) left-hand side (MLHS ), to give
� �
(a) the active earth pressure coefficient, KAE , is computed as M RHS ¼ 12ªK AE (h þ d 0 )2 13(h þ d 0 ) þ (d � d 0 )
a function of the assumed pseudo-static coefficient kh, (6)
while a static earth pressure coefficient, KP , is adopted for þ 16ªK P (3h þ d þ 2d 0 )(d � d 0 )2
the passive resistance 2
(b) the strength of the soil in front of the wall is progressively M LHS ¼ 12ªK P d (d � 23d)
mobilised down to a depth d, as a function of kh
þ 12ª(K P d þ K AE d 0 )(d 0 � d)
(c) the position of the pivot point, d0 , depends on kh , being
about 0.9d for kh ¼ kc : "� � ! #
d 0 � d K AE d 0 þ 2K P d
Using Fig. 7, the force equilibrium of the wall can be 3 þ (d � d 0 )
3 K AE d 0 þ K P d
established by considering the forces acting on the right-
hand side (FRHS ) and on the left-hand side (FLHS ) of the
þ 16ªK AE (d � d 0 )2 (d þ 2d 0 )
wall, as follows.
(7)
By equating FRHS ¼ FLHS and MRHS ¼ MLHS one obtains a

kh k c system of two equations in the two unknowns d and d0 :
Figure 8 shows the maximum bending moment on the
kh kc (d d0)
walls computed from all the numerical analyses, as a func-
h
tion of the surface acceleration behind the walls at the same
time instant; analyses 37 to 39 are those carried out using
different values of the small-strain stiffness of the soil, G0 ,
and of the total height of the wall, H. Fig. 8 also shows the
limit equilibrium maximum bending moment, as a function
of kh , computed according to both the Blum method and the
d proposed method. For completeness, the results obtained
with the earth pressure distribution adopted customarily in
K Pγ d the USA, and described in detail by Bowles (1988), King
d0
d (1995) and Day (1999), are also reported in the figure.
Internal forces on the walls increase as a function of the
KAEγd0 KAEγ(h d0) KPγ(h d0)
applied acceleration, until the critical acceleration is reached.
As an example, for d ¼ 3 m and d ¼ 4 m the numerical
maximum bending moments reach a plateau for accelera-
KAEγd KPγ(h d)
tions greater than 0.3g and 0.48g respectively, their maxi-
mum values being about 185 kN m/m and 330 kN m/m
Fig. 7. Distribution of seismic earth pressures in proposed limit respectively. In this case, critical accelerations predicted by
equilibrium method (from Conti & Viggiani, 2013) the proposed approach, and the corresponding maximum
800
d: m 3 4 5 No. 37 Proposed method
700 Blum method
Tolmezzo No. 38
Assisi No. 39 US method
600
Kocaeli
500 d5m
Mmax: kN m/m
400
300
d4m
200
100
d3m
0
0 0·2 0·4 0·6 0·8 1·0
a: g
Fig. 8. Maximum bending moment as a function of surface acceleration behind walls computed
at the same time instant, and limit equilibrium results
143
0·6
bending moments, are ac ¼ 0.29g and Mc ¼ 175 kN m/m for
ac 0·48g
d ¼ 3 m, and ac ¼ 0.48g and Mc ¼ 342 kN m/m for d ¼ 4 m, 0·3 ac_N 0·16g
showing a very good agreement with the numerical data. A
a: g
0
more dispersed trend is observed for d ¼ 5 m, where the
maximum bending moments do not seem to reach a limit 0·3
value for the earthquakes applied. The fact that the pseudo- 0·6
static approach does not always predict the numerical ob- (a)
servations satisfactorily, at least for a , ac , is due mainly to 0·15
the fact that the soil around the excavation is not always in
limit conditions when bending moments on the walls attain 0·10
u: m
their maximum value, as already discussed. However, the Numerical
proposed method provides an accurate estimate of the maxi- 0·05 Newmark analysis
mum (critical) internal forces that the walls may ever attain
during an earthquake, whereas the Blum and US methods 0
0 4 8 12 16 20
always underpredict the maximum bending moments sub- t: s
stantially. It is worth noting that the critical acceleration, and (b)
hence the maximum bending moment, increase with d – that
is, the longer (and safer) the wall, the larger the bending Fig. 10. Analysis No. 17: (a) acceleration time histories computed
moment it has to sustain under a strong earthquake. at soil surface; (b) horizontal displacements of top of right wall.
Comparison between results from numerical and Newmark
analysis
Newmark analysis
The horizontal displacement of the top of the wall is an fraction of the critical acceleration ac that corresponds to the
important parameter in performance-based seismic design, as complete mobilisation of the soil passive strength; in other
the settlements of the ground surface behind retaining struc- words, had the Newmark analysis been carried out using ac ,
tures, and hence the potential damage to adjacent buildings, the displacements of the wall would have been zero.
are related to the horizontal displacements of the wall (Mana Figure 11 shows the computed values of acN as a function
& Clough, 1981; Hsieh & Ou, 1998; Kung et al., 2007; of the maximum acceleration amax , both normalised by the
Wang et al., 2010). Fig. 9 shows the final horizontal limit equilibrium value of the critical acceleration, ac : Data
displacements of the top of the wall, u, normalised by the from centrifuge tests (Conti et al., 2012) have been also
total height of the wall, H, as a function of the ratio ac /amax included in the same figure for comparison. Both numerical
between the critical acceleration computed with the pseudo- and experimental data indicate that acN /ac increases with the
static approach outlined above, and the maximum accelera- maximum acceleration applied, up to about amax /ac ¼ 1.0,
tion behind the walls. As expected, the displacements of the and it is then approximately constant, and equal to about
walls increase as the intensity of the applied earthquake 40%. This result is not significantly affected by the charac-
increases. However, as shown by Conti et al. (2012) on the teristics of the applied earthquakes, such as frequency
basis of centrifuge tests, the walls can accumulate significant content, duration or Arias intensity.
displacements (u/H . 1–2%), even for accelerations lower
than the critical one, and hence before the available soil
passive resistance is fully mobilised in front of the wall. GUIDELINES FOR DESIGN
These displacements cannot be computed by a conventional The seismic design of embedded cantilevered retaining
Newmark (1965) analysis, that is, assuming a yield accelera- walls for a given value of the maximum acceleration expected
tion equal to the critical value provided by the limit equili- at the site, amax,1D , must address the issues both of: (a) the
brium analysis. Following the procedure adopted by Conti et geotechnical design of the wall, that is, the calculation of a
al. (2012) for the interpretation of centrifuge dynamic tests, depth of embedment such that the permanent horizontal
a Newmark calculation was carried out for each analysis, in displacement of the top of the wall at the end of the earth-
which the yield acceleration, acN , was found by trial and quake, taken as an indicator of the performance of the
error to match computed and numerical displacements at the retaining structure, is less than or equal to an admissible
end of each earthquake. As an example, Fig. 10 shows, for value; and (b) the structural design of the wall, that is, the
analysis No. 17, (a) the acceleration time histories computed definition of the structural section needed to sustain the maxi-
at the soil surface, and (b) the horizontal displacements of mum internal forces experienced by the wall during the earth-
the top of the right wall. The yield acceleration, acN , quake. In the following, the authors try to provide guidelines
required to match the computed final displacement is only a for the seismic design of retaining structures using simplified
methods, in the light of the results presented so far.
6
d: m 3 4 5 1·0
5 Tolmezzo No. 37 Centrifuge data No. 37 d: m 3 4 5
Assisi No. 38 No. 38 Tolmezzo
4 Kocaeli No. 39 No. 39 Assisi
Kocaeli
u/H: %
ac_N/ac
3 0·5
1
0
0 0·5 1·0 1·5 2·0 2·5
0
0 1 2 3 4 5 6 7 amax /ac
ac /amax
Fig. 11. Computed values of acN as a function of maximum
Fig. 9. Final normalised horizontal displacements of top of walls, acceleration amax , both normalised by limit equilibrium value of
as a function of the ratio ac /amax critical acceleration, ac
144
4 d: m 3 4 5
Geotechnical design
Tolmezzo No. 37
Figure 12 is similar to Fig. 9, but this time the final Assisi No. 38
3
normalised displacements of the top of the wall, u/H, are
amax /amax,1D
Kocaeli No. 39
plotted as a function of the ratio ac /amax,1D , where amax,1D is 2
the maximum acceleration at the surface (free field) com-
puted from one-dimensional site response analyses using 1
FLAC. Similar to Richards & Elms (1979), the data were
interpolated with a power function, which is plotted in Fig. 0
12 together with the 98% confidence intervals. 0 0·1 0·2 0·3 0·4 0·5 0·6
amax,1D: g
For each design earthquake motion, corresponding to
which an allowable displacement is prescribed, the upper-
Fig. 13. Ratio amax /amax,1D between maximum acceleration com-
bound curve in Fig. 12 can be used to obtain the ratio puted behind walls and maximum free-field acceleration obtained
¼ ac /amax,1D and then the required critical acceleration. by one-dimensional seismic response analyses, as a function of
Clearly, the effective value of the critical acceleration for the amax,1D
wall must be the maximum among all the design earth-
quakes. Once ac is known, the depth of embedment of the
wall can be computed iteratively using the proposed limit seismic response analyses, amax,1D , as a function of amax,1D :
equilibrium approach, with kh ¼ ah /g ¼ ac /g, until the (criti- Two-dimensional phenomena clearly induce a stronger
cal) condition d ¼ d 0 is satisfied. Any introduction of a (further) amplification than that merely associated with one-
safety factor at this stage would ensure that the displace- dimensional shear wave propagation, but, at least for the
ments experienced by the wall are less than the admissible range of geometrical and mechanical factors considered in
value. this parametric study, the ratio amax /amax,1D is always less
Allowable displacements less than about 3% of H would than about 2, and not significantly affected by the ground
result in a ratio ac /amax,1D . 1: that is, the wall should be motion parameters of the earthquakes applied.
designed to have a critical acceleration larger than the maxi- Figure 14 shows the numerical values of the maximum
mum acceleration expected at the site (i.e. using an equiva- bending moment, Mmax , normalised by the maximum static
lent acceleration that is larger than the maximum free-field bending moment, Mstat , as a function of amax /ac , together
acceleration). This is completely different from the perform- with the pseudo-static bending moments computed according
ance-based design of gravity retaining walls, as they will to the proposed method. The figure shows a good agreement
experience permanent displacements only if their critical between numerical and pseudo-static results, for all the
acceleration is lower than the maximum acceleration of the applied earthquakes. Moreover, limit equilibrium clearly
design earthquake ( , 1). As already discussed, this differ- provides an accurate estimate of the maximum (critical)
ence arises from the fact that embedded cantilevered walls internal forces that a wall with a given value of the critical
begin to rotate for accelerations lower than the critical value. acceleration may ever attain during an earthquake, which is
therefore always a conservative value for the structural de-
sign of the wall. For this purpose, pseudo-static bending
Structural design moments can then be computed using the earth pressure
For the structural design of retaining walls, the maximum distribution outlined in this work, with a pseudo-static coef-
bending moment, Mmax , must be computed under a realistic ficient kh ¼ amax /g. A conservative value of amax would be
distribution of contact stresses between the soil and the amax ¼ 2amax,1D :
structure. Its value is closely related to the maximum accel-
eration at surface behind the walls, amax : As already ob-
served by Callisto & Soccodato (2010) values of amax can CONCLUSIONS
only be computed taking into account soil–structure inter- This paper has addressed the issue of the seismic design
action effects, as the maximum accelerations behind retain- of embedded cantilevered retaining walls, in the light of the
ing structures depend not only on the dynamic properties and results obtained from an extensive set of numerical analyses
thickness of the soil layer, but also on a number of factors, of a pair of cantilevered walls in dry sand, subjected to real
such as the geometry of the excavation, the bending stiffness earthquakes scaled at different values of the maximum
of the wall and the embedded depth, which all affect the acceleration. The results of the analyses confirm that em-
natural frequency of the soil–wall system. As an example, bedded cantilevered retaining walls experience permanent
Fig. 13 shows the ratio amax /amax,1D between the maximum displacements even before the acceleration reaches its criti-
acceleration computed behind the walls, amax , and the maxi- cal value, corresponding to full mobilisation of the shear
mum free-field acceleration obtained by one-dimensional strength of the soil.
7 8
d: m 3 4 5 d: m 3 4 5 No. 37
6 Tolmezzo No. 37 7 Tolmezzo No. 38
Assisi No. 38 Assisi No. 39
5 6
Kocaeli No. 39 Kocaeli
5
Mmax/Mstat
4 u/H 2·9 (ac/amax,1D)1·9 Limit equilbrium

u/H: %
98% interval confidence 4

3
3
2
2
1 1
0 0
0 1 2 3 4 5 6 7 0 0·2 0·4 0·6 0·8 1·0 1·2 1·4 1·6
ac/amax,1D amax /ac
Fig. 12. Final normalised horizontal displacements of top of Fig. 14. Maximum normalised bending moments, as a function of
walls, as a function of the ratio ac /amax,1D amax /ac : numerical and limit equilibrium results
145
For the geotechnical design of the wall, if the permanent Callisto, L., Soccodato, F. M. & Conti, R. (2008). Analysis of the
displacement at the end of the earthquake is taken as a seismic behaviour of propped retaining structures. In Geotechni-
performance indicator, the choice of the equivalent accelera- cal earthquake engineering and soil dynamic IV (eds D. Zeng,
tion to be used in the pseudo-static calculations should be M. T. Manzari and D. R. Hiltunen), GSP 181, pp. 1–10. Reston,
related to the maximum displacements that the structure can VA, USA: ASCE.
CEN (2003). EN 1998-5:2003: Eurocode 8: Design of structures for
sustain, with respect to different levels of design earthquake earthquake resistance – Part 5: Foundations, retaining structures
motion. In this case, the relationship between the final and geotechnical aspects. Brussels, Belgium: European Commit-
displacements of the top of the wall and the ratio ac /amax,1D tee for Standardization.
(Fig. 12) can be used to obtain the equivalent acceleration, Cilingir, U., Haigh, S. K., Madabhushi, S. P. G. & Zeng, X. (2011).
for any given allowable displacement of the wall. This can Seismic behaviour of anchored quay walls with dry backfill.
be expressed as ah ¼ �amax,1D , where � (¼ ac /amax,1D) can be Geomech. Geoengng 6, No. 3, 227–235.
larger than 1. Conti, R. & Viggiani, G. M. B. (2013). A new limit equilibrium
For the structural design of the wall, the maximum bend- method for the pseudostatic design of embedded cantilevered
ing moments can be computed using a realistic distribution retaining walls. Soil Dynam. Earthquake Engng 50, 143–150.
Conti, R., Madabhushi, S. P. G. & Viggiani, G. M. B. (2012). On
of contact stresses, such as that proposed in this work, and a
the behaviour of flexible retaining walls under seismic actions.
conservative value of the pseudo-static acceleration: Géotechnique 62, No. 12, 1081–1094, http://dx.doi.org/10.1680/
kh ¼ amax /g ¼ 2amax,1D : geot.11.P.029.
The data discussed in the paper refer only to cantilevered Day, R. A. (1999). Net pressure analysis of cantilever sheet pile
walls in dry sand; further research is required to clarify the walls. Géotechnique 49, No. 2, 231–245, http://dx.doi.org/
role of the presence of the pore water, for either saturated or 10.1680/geot.1999.49.2.231.
unsaturated soils. Evangelista, A., Scotto di Santolo, A. & Simonelli, A. L. (2010).
Evaluation of pseudostatic active earth pressure coefficient of
cantilever retaining walls. Soil Dynam. Earthquake Engng 30,
ACKNOWLEDGEMENTS No. 11, 1119–1128.
Fang, Y. S., Yang, Y. C. & Chen, T. J. (2003). Retaining walls
The work presented in this paper was developed with the damaged in the Chi-Chi earthquake. Can. Geotech. J. 40, No. 6,
financial support of the Italian Department of Civil Protec- 1142–1153.
tion within the ReLUIS research project. Gazetas, G., Psarropoulos, P. N., Anastasopoulos, I. & Gerolymos,
N. (2004). Seismic behaviour of flexible retaining systems
subjected to short-duration moderately strong excitation. Soil
NOTATION Dynam. Earthquake Engng 24, No. 7, 537–550.
a, b, x0 constants of the hysteretic model Hsieh, P. G. & Ou, C. Y. (1998). Shape of ground surface
d depth of embedment settlement profiles caused by excavation. Can. Geotech. J. 35,
d0 depth of pivot point No. 6, 1004–1017.
d depth of full mobilisation of passive soil pressure Huang, C. C., Wu, S. H. & Wu, H. J. (2009). Seismic displace-
GS secant shear modulus ment criterion for soil retaining walls based on soil strength
G0 small-strain shear modulus mobilization. J. Geotech. Geoenviron. Engng 135, No. 1,
h excavation depth 74–83.
KAE dynamic active earth pressure coefficient Iai, S. & Kameoka, T. (1993). Finite element analysis of earthquake
KG stiffness multiplier induced damage to anchored sheet pile quay walls. Soils Found.
KP static passive earth pressure coefficient 33, No. 1, 71–91.
MS normalised secant shear modulus Itasca (2005). FLAC Fast Lagrangian Analysis of Continua. v. 5.0.
p9 mean effective stress User’s Manual. Minneapolis, MN, USA: Itasca Consulting
pref reference pressure Group.
ª shear strain; soil unit weight Kim, W. C., Park, D. & Kim, B. (2010). Development of a general-
� shear stress ised formula for dynamic active earth pressure. Géotechnique
60, No. 9, 723–727, http://dx.doi.org/10.1680/geot.09.T.001.
King, G. J. W. (1995). Analysis of cantilever sheet-pile walls in
REFERENCES cohesionless soil. J. Geotech. Engng Div., ASCE 121, No. 9,
Atik, L. A. & Sitar, N. (2010). Seismic earth pressures on cantilever 629–635.
retaining structures. J. Geotech. Geoenviron. Engng 136, No. 10, Kontoe, S., Zdravkovic, L., Menkiti, C. O. & Potts, D. M. (2012).
1324–1333. Seismic response and interaction of complex soil retaining
Atkinson, J. H., Coop, M. R., Stallebrass, S. E. & Viggiani, G. systems. Comput. Geotech. 39, January, 17–26.
(1993). Measurement of stiffness of soils and weak rocks in Koseki, J., Koda, M., Matsuo, S., Takasaki, H. & Fujiwara, T.
laboratory tests. Proceedings of the 25th annual conference of (2012). Damage to railway earth structures and foundations
the Engineering Group of the Geological Society, Leeds, UK, caused by the 2011 off the Pacific Coast of Tohoku Earthquake.
pp. 21–27. Soils Found. 52, No. 5, 872–889.
Basha, B. M. & Babu, L. S. B. (2010). Seismic rotational displace- Kung, G. T. C., Juang, H., Hsiao, E. C. L. & Hashash, Y. M. A.
ments of gravity walls by pseudodynamic method with curved (2007). Simplified model for wall deflection and ground-surface
rupture surface. Int. J. Geomech. 10, No. 3, 93–105. settlement caused by braced excavation in clays. J. Geotech.
Bica, A. V. D. & Clayton, C. R. I. (1989). Limit equilibrium design Geoenviron. Engng 133, No. 6, 731–747.
methods for free embedded cantilever wall in granular materials. Lancellotta, R. (2002). Analytical solution of passive earth pressure.
Proc. Inst. Civ. Engrs, Part 1 86, No. 5, 879–898. Géotechnique 52, No. 8, 617–619, http://dx.doi.org/10.1680/
Blum, H. (1931). Einspannungsverhältnisse bei Bohlkwerken. geot.2002.52.8.617.
Berlin, Germany: Wil. Ernst und Sohn (in German). Lancellotta, R. (2007). Lower-bound approach for seismic passive
Bowles, J. E. (1988). Foundation analysis and design, 4th edn. New earth resistance. Géotechnique 57, No. 3, 319–321, http://
York, NY, USA: McGraw-Hill. dx.doi.org/10.1680/geot.2007.57.3.319.
Callisto, L. & Soccodato, F. M. (2007). Seismic analysis of an Ling, H. I. (2001). Recent applications of sliding block theory to
embedded retaining structure in coarse-grained soils. Proceed- geotechnical design. Soil Dynam. Earthquake Engng 21, No. 3,
ings of the 4th international conference on earthquake geotech- 189–197.
nical engineering, Thessaloniki, Greece, paper no. 1197. Madabhushi, S. P. G. & Zeng, X. (2006). Seismic response of
Callisto, L. & Soccodato, F. M. (2010). Seismic design of flexible flexible cantilever retaining walls with dry backfill. Geomech.
cantilevered retaining walls. J. Geotech. Geoenviron. Engng 136, Geoengng 1, No. 4, 275–289.
No. 2, 344–354. Madabhushi, S. P. G. & Zeng, X. (2007). Simulating seismic
146
response of cantilever retaining walls. J. Geotech. Geoenviron. embedded in stiff clay, CIRIA Report R104. London, UK:
Engng 133, No. 5, 539–549. Construction Industry Research and Information Association.
Mana, A. I. & Clough, G. W. (1981). Prediction of movements for Powrie, W. (1996). Limit equilibrium analysis of embedded retain-
braced cuts in clay. J. Geotech. Engng Div., ASCE 107, No. 6, ing walls. Géotechnique 46, No. 4, 709–723, http://dx.doi.org/
759–777. 10.1680/geot.1996.46.4.709.
Masing, G. (1926). Eigenspannungen und Verfertigung bim Mes- Rathje, E. M., Abrahamson, N. A. & Bray, J. D. (1998). Simplified
sing. Proceedings of the 2nd international congress on applied frequency content estimates of earthquake ground motions.
mechanics, Zurich, Switzerland, pp. 332–335. J. Geotech. Geoenviron. Engng 124, No. 2, 150–159.
Mononobe, N. & Matsuo, H. (1929). On the determination of earth Richards, R. & Elms, D. G. (1979). Seismic behavior of gravity
pressure during earthquake. Proceedings of the 2nd world en- retaining walls. J. Geotech. Engng Div., ASCE 105, No. 4, 449–
gineering conference, Tokyo, Japan, vol. 9, pp. 177–185. 464.
Mylonakis, G., Kloukinas, P. & Papantonopoulos, C. (2007). Richards, R. & Elms, D. G. (1992). Seismic passive resistance of
An alternative to the Mononobe–Okabe equations for seismic tied-back walls. J. Geotech. Engng 118, No. 7, 996–1011.
earth pressures. Soil Dynam. Earthquake Engng 27, No. 10, Steedman, R. S. & Zeng, X. (1990). The influence of phase on the
957–969. calculation of pseudo-static earth pressure on retaining wall.
Neelakantan, G., Budhu, M. & Richards, R. (1992). Balanced Géotechnique 40, No. 1, 103–112, http://dx.doi.org/10.1680/
seismic design of anchored retaining walls. J. Geotech. Engng geot.1990.40.1.103.
118, No. 6, 873–888. Vucetic, M. & Dobry, R. (1991). Effect of soil plasticity on cyclic
Newmark, N. M. (1965). Effects of earthquakes on dams and response. J. Geotech. Geoenviron. Engng 117, No. 1, 89–107.
embankments. Géotechnique 15, No. 2, 139–160, http:// Wang, J. H., Xu, Z. H. & Wang, W. D. (2010). Wall and ground
dx.doi.org/10.1680/geot.1965.15.2.139. movements due to deep excavations in Shanghai soft soils.
NTC (2008). Norme Tecniche per le Costruzioni. Gazzetta Ufficiale J. Geotech. Geoenviron. Engng 136, No. 7, 984–994.
della Repubblica Italiana 04/12/2008, Supplemento Ordinario Zeng, X. (1990). Modelling the behaviour of quay walls in earth-
29, No. 30 (in Italian). quakes. PhD thesis, University of Cambridge, Cambridge, UK.
Okabe, S. (1926). General theory of earth pressure and seismic Zeng, X. & Steedman, R. S. (1993). On the behaviour of quay
stability of retaining wall and dam. J. Jpn. Soc. Civ. Engng 10, walls in earthquakes. Géotechnique 43, No. 3, 417–431, http://
No. 6, 1277–1288. dx.doi.org/10.1680/geot.1993.43.3.417.
Osman, A. S. & Bolton, M. D. (2004). A new design method for Zeng, X. & Steedman, R. S. (2000). Rotating block method for
retaining walls in clay. Can. Geotech. J. 41, No. 3, 453–469. seismic displacement of gravity walls. J. Geotech. Geoenviron.
PIANC (2001). Seismic design guidelines for port structures (Work- Engng 126, No. 8, 709–717.
ing Group N34 of the Maritime Navigation Commission, Inter- Zienkiewicz, O. C., Bianic, N. & Shen, F. Q. (1988). Earthquake
national Navigation Association). Rotterdam, the Netherlands: input definition and the transmitting boundary condition. In
Balkema. Conference on advances in computational non-linear mechanics
Padfield, C. J. & Mair, R. J. (1984). Design of retaining walls (ed. I. St Doltnis), pp. 109–138. Berlin, Germany: Springer.
147
Ling, H. et al. (2014). Géotechnique 64, No. 5, 400–404 [http://dx.doi.org/10.1680/geot.12.T.019]
TECHNICAL NOTE
Revisiting Nigawa landslide of the 1995 Kobe earthquake

H . L I N G � , H . I . L I N G † a n d T. K AWA BATA ‡
A simplified pseudo-static limit equilibrium approach is used to analyse a full-scale slope failure that
occurred during the 1995 Kobe earthquake. The slope is 28 m high, having a gentle angle of 26 .58,
and has a factor of safety as high as 2 .43 in the absence of an earthquake. In addition to a factor of
safety, the critical failure surface obtained from pseudo-static analysis is compared with that of the
actual failure surface. Contradictory to some design codes that use only a horizontal seismic
coefficient, this study finds that a rational analysis must take into consideration both horizontal and
vertical seismic accelerations. The analysis including vertical acceleration may give a shallower or
deeper failure surface than cases where vertical acceleration is ignored. An analysis combining
horizontal and upward accelerations results in a deeper failure surface than when only downward or
no vertical acceleration is considered. The effects of soil properties on the slope stability are also
investigated. The study suggests that for near-fault earthquakes, slopes may be subjected to vertical
(upward) seismic coefficient as large as 90% that of horizontal, larger than that specified in existing
design codes.
KEYWORDS: earthquake; landslide; slope
INTRODUCTION drained ring shear tests to investigate possible rapid loss of

The 1995 Kobe earthquake (Hyogoken Nambu earthquake, strength due to liquefaction. Although ground water was
M ¼ 7 .2) resulted in many landslides along the vicinity of observed during post-failure investigation, the writers felt
the fault, among which Nigawa landslide was the largest that it was unlikely that the water table stayed constantly
slope failure induced by an earthquake, in terms of scale high during shaking, as was the degree of saturation in the
and casualties. The slope was 28 m high and gentle, around soil mass, to allow for liquefaction at the site. Statistically,
26 .58 (more details provided subsequently). The landslide 1994 was reported to be one of the driest years. Thus, the
covered the width and length, both of 100 m, and a maxi- water table was not included in the main analysis (effects of
mum thickness of 15–20 m. A total volume of 120 000 m3 which were found to be small), and a different mechanism
of soil was involved. The sliding soil mass travelled rapidly of sliding of the gentle slope could be involved, such as
over a maximum distance of 175 m in a single event to the could be attributable to large horizontal and vertical accel-
toe of the slope and blocked the Nigawa River. The land- erations.
slide destroyed 11 houses and killed 34 residents. In this study, a series of pseudo-static analyses were
The geological profile for the main section of the slope conducted on the Nigawa slope. Both horizontal and vertical
(A–A9) is shown in Fig. 1. (Sassa, 1996). The base rock accelerations were included in the analyses. The results were
was of stable granite, underlain by the Osaka group layer, compared with the observed failure surface and discussed.
which was of limnic and marine deposit of granite sands
and clays from the Pliocene to Middle Pleistocene age. The
terrace deposits were seen at the toe of the slope. The METHODOLOGY OF ANALYSIS
surface layer where sliding occurred was a fill consisting of Limit equilibrium methods are typically used to analyse
Osaka group coarse, sandy soil. Kawabe (1996) investigated slope stability (e.g. Fredlund & Krahn, 1977; Bromhead,
the topographical maps of the site that were published by 1992; Duncan & Wright, 2005). By including an inertia force
the Geospatial Information Authority of Japan in 1886, 1947 due to an earthquake, several mechanisms, such as planar
and 1990. It became apparent that the slope was a man- (e.g. Seed & Goodman, 1964; Ling et al., 1999), circular and
made high embankment, with two filling records. log-spiral (e.g. Leshchinsky & San, 1994; Ling et al., 1997)
There has not been a rigorous failure analysis for Nigawa have been proposed to assess seismic slope stability. The
landslide, except by Sassa and colleagues (e.g. Sassa, 1996; earthquake inertia force is considered through a seismic
Sassa et al., 1996), which focused on the mechanism of coefficient, which is a percentage of earth gravity. Because
long-distance sliding. Sassa et al. (1996) conducted un- of the randomness of earthquake motions, it is difficult to
determine the value of the coefficient, although the peak or a
reduced value may be used according to the importance of
Manuscript received 3 December 2012; revised manuscript accepted structures as stated in the design codes. The design can also
10 October 2013. Published online ahead of print 2 May 2014. be linked to slope performance expressed in terms of earth-
quake-induced displacements, so that different values of
see p. ii.
� Columbia College, Columbia University, New York, NY, USA. seismic coefficient can in principle be used for different
† Department of Civil Engineering and Engineering Mechanics, levels of slope performance (e.g. Ling & Leshchinsky, 1995,
Columbia University, New York, NY, USA. 1997; Bray & Rathje, 1998; Stewart et al., 2003). In design,
‡ Geotechnical and Environmental Engineering, Faculty of Agricul- while the horizontal acceleration is considered, the vertical
ture, Kobe University, Nada-ku, Kobe, Japan. acceleration is usually ignored, since a factor of safety may
149
LING, LING AND KAWABATA
Granite
Osaka group
Terrace deposits
Secondary deposits (fill)

110 B-6 Sliding surface
B-7 Initial landslide mass (before motion)
100 Deposited landslide mass
B-8
90 B-9 B-23
B-5
Height: m
80 B-10
70
60
0 50 100 150 200

Width: m
Fig. 1. Nigawa landslide: slope geometry of section A–A9 (Sassa, 1996)
be adequate to smear the effects of vertical excitation. How- portion of slope was 6 .3 m high and had an angle of 13 .138.
ever, relatively large vertical components of acceleration, The toe was inclined at 4 .68.
larger than 30% that of the horizontal, have been recorded The basic soil properties were obtained from a post-failure
during recent major earthquakes (e.g. Ling & Leshchinsky, investigation (Kawasaki Chishitsu, personal communication,
1998), such that their effects may no longer be ignored. Ling 2010). Fig. 2(a) shows the grain size distribution curves of
et al. (1997) indicated that effects of vertical acceleration two soil samples, where the mean diameter D50 was between
may become prominent when combined with large horizontal 0 .5 and 0 .7 mm and the percentage of fines was between
component. 25 .5 to 28 .4%. The soil was classified as silty sand and its
In Eurocode 8 (CEN, 2005), both horizontal and vertical specific gravity was 2 .645. The void ratio was 0 .558 and the
components of acceleration have to be considered in seismic natural water content was 11 .65%, which corresponded to a
design, with kv ¼ �0 .33kh (for avg /ag < 0 .6) or kv ¼ �0 .5kh degree of saturation of 55 .3%.
(for avg /ag . 0 .6), where ag is the design value of accelera- A series of direct shear tests was conducted using undis-
tion and avg is the design value of vertical acceleration. kh turbed soil specimens of diameter 15 cm and height 6 .2 cm
and kv are the horizontal and vertical seismic coefficients, at the natural water content (unsaturated condition). The
respectively. A serviceability evaluation, such as using the specimen was sheared at a displacement rate of 1 mm/min.
rigid sliding block method, is also required. Although there Four normal stresses of 19 .62 kPa, 39 .24 kPa, 58 .86 kPa and
are several design codes in Japan (e.g. RTRI, 2007), a 78 .48 kPa were applied. Fig. 2(b) shows the relationships
pseudo-static analysis of slope is also followed by a perma- between the shear stress and horizontal displacement of
nent displacement evaluation. The vertical component of
acceleration is typically not used for slope stability analysis
in Japan. Because of a large seismic coefficient used in 100
designing against level II earthquakes, a specified factor of
safety may no longer be satisfied, thus the design has to be 80 Sample 1
Sample 2
Percent finer:%
based on serviceability. In the USA, as documented in the

60
NCHRP (2008), seismic design of a slope does not require a
vertical component of acceleration.
40
In this study, a circular mechanism of the modified Bishop
method (Bishop, 1955) is used. The analysis was conducted
20
using the program ReSSA (Leshchinsky, 2002). Because of
the length restriction, the procedure of incorporating inertia
0
forces into the modified Bishop method is not elaborated. 0·001 0·01 0·1 1 10
Note that, in this study, the vertical seismic coefficient is Grain size: mm
(a)
considered positive when vertical inertia force is acting
100
upward. The silty sand disintegrated after failure, thus this
study did not aim to evaluate the permanent displacements,
80 σn: kPa
Shear stress, τ: kPa
which are usually obtained by assuming a rigid body mo-

tion. 78·5
60
58·9
40 39·2
SLOPE GEOMETRY AND SOIL PROPERTIES 19·6
20
The slope geometry was obtained from the failure section Sample 1 Sample 2
(Fig. 1) as reported by Sassa et al. (1996), but interpreted 0
slightly differently here (see Fig. 4(a), in the results section, 0 2 4 6 8 10 12 14 16 18 20
Horizontal displacement: mm
for the geometry used in analyses). The survey showed that (b)
the total height of the slope was 28 .1 m, and was composed
of two parts. The bottom part of the slope (main slope) was Fig. 2. Soil properties: (a) grain size distribution, (b) direct shear
21 .8 m high and at an angle of 26 .58, whereas the top test results
150
REVISITING NIGAWA LANDSLIDE OF THE 1995 KOBE EARTHQUAKE
samples 1 and 2. Sample 1 exhibited slight softening behav- Railway) Takarazuka Station was located closest to the
iour. The cohesion (should be known as apparent cohesion Nigawa landslide site (note that this station has been relo-
in lieu of unsaturated conditions) and the angle of internal cated after the earthquake). The seismic coefficients may be
friction were obtained as c ¼ 16 .7 kPa and ¼ 38 .78, re- selected based on the design codes as explained earlier,
spectively. Sample 2 was of natural water content of 20 .65% using reduced values. Since this study analysed a failure
or degree of saturation of 89 .1%, rendering strength para- case, coefficients corresponding to actual peak earthquake
meters of c ¼ 7 .65 kPa and ¼ 36 .18. The results were of records are considered (as will also be seen later from the
similar trend as reported by Ling & Ling (2012), where results of analysis that reduced seismic coefficients based on
apparent cohesion reduced with increasing water content for design codes would not render reasonable results).
unsaturated soil. The authors felt that direct shear tests, Figure 3(b) shows the records of horizontal (east–west
which allowed principal stress rotation, are more relevant and north–south directions) and vertical accelerations. The
than triaxial tests in representing the soil strength used in seismic coefficients based on peak acceleration in various
slope stability analysis. In the analyses shown subsequently, directions are: north ¼ 0 .586, south ¼ 0 .613, east ¼ 0 .365,
using ¼ 38 .78 (sample 1), the best fit cohesion was west ¼ 0 .697. From the orientation of the major cross-sec-
c ¼ 20 kPa, which is very close to that of sample 1. tion of failure (A–A9) along the north–east direction, and
assuming a linear variation of coefficients between the north
and east directions, the peak horizontal seismic coefficient
SEISMIC COEFFICIENTS was obtained as kho ¼ 0 .476. The seismic coefficients for the
The site was 37 .5 km from the epicenter, but merely vertical inertia force (upward and downward) were
500 m away from the fault (Figure 3(a)). Among the accel- kvo ¼ 0 .426 and �0 .306 (i.e. kvo /kho ¼ 0 .89 and �0 .782),
eration records of the 1995 Kobe earthquake, JR (Japan respectively.
In the analyses, several different combinations of seismic
40 coefficients were used: horizontal coefficient only, horizontal
km and upward vertical coefficients, horizontal and downward
N
vertical coefficients.
30
km
RESULTS AND DISCUSSION
Figure 4 shows the slope geometry and results of stability
analysis. Under static conditions, the slope had a factor of
20
km safety as high as 2 .43, which indicated a very stable slope.
Considering the Kobe earthquake with peak horizontal and
vertical accelerations, which acted upward and downward,
respectively, the factors of safety are 0 .782 and 1 .17. For the
10
km seismic analysis without considering vertical acceleration,
the yield horizontal seismic coefficient (giving a factor of
safety of unity) is 0 .5. The horizontal yield seismic coeffi-
cients including vertical accelerations (upward and down-
ward) are 0 .38 and 0 .65, respectively. Thus, the yield
6·3 m
13·13°
A: Epicentre
28·1 m
B: Landslide site
21·8 m
C: Takarazuka station
Potential failure
(a) surface
26·5°
0·8 c 20 kPa
North–south
kho 0·586 4·6° φ 38·7°
0·4 Kobe earthquake (1995)
γt 17·45 kN/m3
0 (a)
0·4 JR Takarazuka station
3·0
0·8
0·8 East–west 2·5 Fs 2·43 (static)
Acceleration: g
kho 0·365
0·4
kv /kh
Factor of safety: Fs
0 2·0
0·895 (up)
0·4 0
1·5
0·8 Stable 0·782 (down)
Upwards–downwards 1·0 Failure

0·4 kvo 0·426
Unstable
0 0·5
0·4 kh 0·38 0·5 0·65

0
0 5 10 15 20 0 0·2 0·4 0·6 0·8
Time: s Horizontal seismic coefficient, kh
(b) (b)
Fig. 3. (a) Location of Nigawa landslide with reference to Fig. 4. Stability analysis: (a) slope geometry, (b) factor of safety
epicentre (map data # 2010 Google, Zenrin); (b) earthquake under different combinations of horizontal and vertical accelera-
records obtained at JR Takarazuka station tions
151
LING, LING AND KAWABATA
horizontal accelerations can be larger or smaller than the ward vertical accelerations, when combined with the same
peak value of 0 .476 depending on the direction and magni- magnitude of horizontal acceleration. The horizontal accel-
tude of vertical acceleration. The inclusion of vertical acceleration combined with upward vertical acceleration
eration in the downward direction gives a larger factor of (kh ¼ 0 .38, kv /kh ¼ 0 .895) gave a failure surface closest to
safety compared to that when acting upward. that observed in the field; a smaller value of kv /kh did not
It is clear that in order to obtain the most appropriate lead to acceptable matching of failure surface. The results
value of yield accelerations, the geometry of the failure implied that Eurocode 8, which has kv /kh ¼ 0 .5, is not
surface has to be considered. Fig. 5(a) shows the different conservative, especially in design where the failure surface
failure surfaces obtained under different combinations of is a major concern, such as deciding the reinforcement
horizontal and vertical accelerations. The upward vertical length for a reinforced slope (see Leshchinsky et al., 2014;
acceleration gives a deeper failure surface than the down- Ling & Leshchinsky, 1998).
The shape of failure surface and the yield horizontal
seismic coefficient are greatly affected by the soil cohesion
φ 38·7° as shown in Fig. 5(b). Note that the angle of internal friction
c 20 kPa and the ratio of vertical to horizontal acceleration were kept
60 the same for all cases. When cohesion is increased, a deeper
γ 17·45 kN/m3
failure surface and higher yield horizontal acceleration re-
kv /kh 0, khy 0·5 kv /kh 0·25, khy 0·46 sulted. A cohesion of 20 kPa matched the failure surface of
40 kv /kh 0·25, khy 0·54 kv /kh 0·5, khy 0·425 the lower part of the slope well, whereas it deviated at the
kv /kh 0·5, khy 0·59 kv /kh 0·895, khy 0·38 upper part of the slope. Possible explanations included, but
kv /kh 0·782, khy 0·65 were not limited to, amplification of accelerations along the
20
Observed slope height, as well as the simplified two-dimensional as
opposed to three-dimensional analyses. Note that in Lesh-
chinsky et al. (1985), which used a log spiral mechanism,
three-dimensional analysis gave a deeper failure surface at
0 the top part of slope compared to that of two-dimensional
analysis.
Unit: m
The effect of the internal friction angle of soil was also
0 20 40 60 80 100 investigated by varying it between 208 and 458. The cohesion
(a) (c ¼ 20 kPa) and ratio of vertical to horizontal acceleration
φ 38·7° (kv /kh ¼ 0 .895) were kept the same for all cases. The results
γ 17·45 kN/m3
(Fig. 5(c)) show that a small angle of internal friction
60 rendered a lower yield acceleration, and vice versa. How-
kv /kh 0·895 c 50 kPa, khy 0·493
c 40 kPa, khy 0·46
ever, the difference in the failure surfaces was insignificant.
It has to be noted that the water table was varied in a
c 30 kPa, khy 0·425
40 separate analysis, but showed negligible effect on the failure
c 20 kPa, khy 0·38 surface.
c 10 kPa, khy 0·315
20
Observed CONCLUSION
Nigawa landslide of the 1995 Kobe earthquake was ana-
lysed using the best available information on soil properties
0 and earthquake characteristics. For a reasonable comparison
between the results of analysis and field observation, the
Unit: m
failure surface should be considered in addition to a factor
0 20 40 60 80 100 of safety.
(b) The analysis led to the conclusion that upward vertical
acceleration gave the most critical failure surface compared
φ 38·7° to cases of downward vertical acceleration or when vertical
γ 17·45 kN/m3 acceleration was ignored. The best-fit failure surface was
60
kv /kh 0·895 obtained as kh ¼ 0 .38, which is less than the peak value of
0 .46 (0 .82 instead of 0 .5 for soil factor of 1 in Eurocode 8);
and kv /kh ¼ 0 .9, which is significantly larger than 0 .33 (for
40 φ 45°, khy 0·438 kh , 0 .6) as assumed in Eurocode 8.
φ 38·7°, khy 0·38 This study indicated that vertical (upward) combined with
φ 30°, khy 0·286
horizontal accelerations might have contributed to the actual
φ 20°, khy 0·146
20 failure of Nigawa slope, instead of liquefaction as explained
Observed by Sassa et al. (1996). Additional failure cases should be
analysed for near-fault earthquakes, so as to re-evaluate the
0
recommended design value of kv /kh :
Unit: m
ACKNOWLEDGEMENTS
0 20 40 60 80 100 The writers appreciated the technical information provided
(c)
by Kawasaki Chishitsu Chosa, Co. The location of the
Fig. 5. (a) Effects of vertical acceleration on critical failure failure site is shown by Google Maps in Fig. 3(a). Dr
surface and yield acceleration; (b) effects of cohesion on critical Meguro of the University of Tokyo kindly provided the
failure surface and yield acceleration; (c) effects of angle of earthquake records and Dr Luigi Callisto of the University
internal friction on critical failure surface and yield acceleration of Rome La Sapienza provided information on Eurocode 8.
152
REVISITING NIGAWA LANDSLIDE OF THE 1995 KOBE EARTHQUAKE
ReSSA is licensed by Dov Leshchinsky through Adama of slopes: design charts. J. Geotech. Engng, ASCE 120, No. 9,
Engineering, Inc., Newark, DE, USA. 1514–1532.
Leshchinsky, D., Baker, R. & Silver, M. L. (1985). Three dimen-
sional analysis of slope stability. Int. J. Numer. Analyt. Methods
NOTATION Geomech. 9, No. 3, 199–223.
Leshchinsky, D., Kang, B., Han, J. & Ling, H. I. (2014). Framework
ag design value of acceleration
for limit state design of geosynthetic-reinforced walls and
avg design value of vertical acceleration
slopes. Transportation Infrastructure Geotechnology, 10.1007/
c cohesion
S40515-014-0006-3.
D50 mean diameter
Ling, H. I. & Leshchinsky, D. (1995). Seismic performance of
Fs factor of safety
simple slopes. Soils Found. 35, No. 2, 85–94.
kh horizontal seismic coefficient
Ling, H. I. & Leshchinsky, D. (1997). Seismic stability and
kho horizontal seismic coefficient (peak value)
kv vertical seismic coefficient permanent displacement of landfill cover systems. J. Geotech.
kvo vertical seismic coefficient (peak value) Geoenviron. Engng, ASCE 123, No. 2, 113–122.
ªt bulk unit weight Ling, H. I. & Leshchinsky, D. (1998). Effects of vertical accelera-
�n normal stress tion on seismic design of geosynthetic-reinforced soil structures.
� shear stress Géotechnique 48, No. 3, 347–373.
� angle of internal friction Ling, H. & Ling, H. I. (2012). Centrifuge model simulations of
rainfall-induced slope instability. J. Geotech. Geoenviron. Engng
138, No. 9, 1151–1157.
Ling, H. I., Leshchinsky, D. & Mohri, Y. (1997). Soil slopes under
REFERENCES combined horizontal and vertical seismic accelerations. Earth-
Bishop, A. W. (1955). The use of slip circles in the stability quake Engng Struct. Dynamics 26, No. 12, 1231–1241.
analysis of earth slopes. Géotechnique 5, No. 1, 7–17. Ling, H. I., Mohri, Y. & Kawabata, T. (1999). Seismic stability of
Bray, J. D. & Rathje, E. M. (1998). Earthquake-induced displace- sliding wedge: extended Francais–Culmann’s analysis. Soil Dy-
ments of solid waste landfills. J. Geotech. Geoenviron. Engng, namics Earthquake Engng 18, No. 5, 387–393.
ASCE 124, No. 3, 242–253. NCHRP (National Cooperative Highway Research Program) (2008).
Bromhead, E. N. (1992). The stability of slopes. Glasgow, UK: Seismic analysis and design of retaining walls, buried structures,
Blackie Academic and Professional. slopes, and embankments, Report 611. Washington, D. C., USA:
CEN (European Committee for Standardization) (2005). Eurocode Transportation Research Board.
8: Design of structures for earthquake resistance – Part 5: RTRI (2007). Design standards for railway structures – earth
foundations, retaining structures and geotechnical aspects. Brus- structures. Japan: Maruzen Co., Ltd (in Japanese).
sels, Belgium: European Committee for Standardization. Sassa, K. (1996). Prediction of earthquake induced landslides. In
Duncan, J. M. & Wright, S. G. (2005). Soil strength and slope Landslides (ed. K. Senneset), pp. 115–132. Rotterdam, the
stability. New York, USA: Wiley. Netherlands: Balkema.
Fredlund, D. G. & Krahn, J. (1977). Comparison of slope stability Sassa, K., Fukuoka, H., Scarascia-Mugnozza, G. & Evans, S.
methods of analysis. Can. Geotech. J. 14, No. 3, 429–439. (1996). Earthquake-induced-landslides: distribution, motion and
Kawabe, T. (1996). Liquefaction disasters of the 1995 South Hyogo mechanisms. Soils and Found., Special Issue, 53–64.
earthquake – Special reference to the origin of slope collapse at Seed, H. B. & Goodman, R. E. (1964). Earthquake stability of
Nigawa-Yurinodai, Nishinimiya City. Tohoku J. Natural Disaster slopes of cohesionless soils. J. Soil Mech. Found. Div., ASCE
Sci. 32, 213–218 (in Japanese). 90, No. 6, 43–73.
Leshchinsky, D. (2002). Design software for geosynthetic-reinforced Stewart, J. P., Liu, A. H. & Choi, Y. (2003). Amplification factors
soil structures. Geotech. Fabrics Rep. 19, March/April, pp. 44–49. for spectral acceleration in tectonically active regions. Bull.
Leshchinsky, D. & San, K.-C. (1994). Pseudo-static seismic stability Seismol. Soc. Am. 93, No. 1, 332–352.
153
Kokkali, P. et al. (2014). Géotechnique 64, No. 11, 865–880 [http://dx.doi.org/10.1680/geot.14.P.064]
Static and cyclic rocking on sand: centrifuge versus reduced-scale 1g

experiments
P. KO K K A L I , I . A NA S TA S O P O U L O S † , T. A B D O U N a n d G . G A Z E TA S ‡
Shallow foundations supporting bridge piers, building frames, shear walls and monuments are often
subjected to extreme lateral loading such as wind in offshore environments, or strong seismic shaking.
Under such loading conditions, foundations may experience a host of non-linear phenomena: sliding
on and uplifting from the supporting soil or even soil failure in the form of development of ultimate
bearing capacity mechanisms. This type of response is accompanied by residual settlement and
rotation of the supported structural system. Nevertheless, inelastic foundation performance can provide
potential benefits to the overall seismic integrity of the structure. Thanks to such non-linearities,
energy dissipation at or below the foundation level may eventually limit the seismic demand on
structural elements. Several theoretical and experimental studies have provided encouraging evidence
to this effect. This paper has a dual objective: first, to study the behaviour of shallow foundations
under vertical and lateral monotonic loading and under lateral slow cyclic loading of progressively
increasing amplitude; second, to explore the differences in foundation response between reduced-scale
1g and centrifuge 50g model testing. Emphasis is placed on interpreting their discrepancies by
unveiling the role of scale effects. The role of soil densification due to multiple loading cycles with
uplifting is also highlighted.
KEYWORDS: bearing capacity; centrifuge modelling; footings/foundations; settlement
INTRODUCTION 2012, 2013; Deng et al., 2012; Drosos et al., 2012). Experi-
The importance of non-linear soil–foundation–structure inter- mental studies have significantly contributed to the under-
action under lateral loading has been acknowledged by the standing of the rocking response of shallow foundations.
engineering community, especially for offshore structures Nevertheless, many of them have been conducted at a low
which are typically subjected to multicycle wind and wave confining stress environment (reduced-scale 1g testing).
loading. In the case of seismic shaking, with the recorded Compared to centrifuge model testing, 1g experiments are
acceleration levels by far exceeding the conventional design easier and more economical to perform but cannot reproduce
guidelines in recent seismic events, it has become evident the actual stress field in the soil. Owing to the low prevail-
that inelastic foundation response is often unavoidable. Shal- ing confining stresses in 1g test conditions, the angle of
low foundations supporting bridge piers or building columns shearing resistance and the small strain stiffness of the soil
and shear walls may experience sliding and/or uplifting from are typically much larger compared to realistic stress levels.
the supporting soil, or bearing capacity failure in softer soils. Such issues may have a substantial effect on the measured
Usually, such non-linear response is accompanied by perma- response and therefore 1g tests should be carefully designed
nent settlement and/or rotation. However, such mobilisation and the results should be interpreted accordingly.
of strongly inelastic response also reduces the seismic In an attempt to clarify these issues, commonly referred
demand and may therefore be beneficial for the seismic to as ‘scale effects’, a qualitative and quantitative compari-
performance of the structural system. The potential benefits son of the rocking response of shallow foundations obtained
from these types of non-linearity have been indicated by from centrifuge and reduced-scale 1g experiments is pre-
several researchers (Priestley et al., 1996; Pecker & Pender, sented in this paper. Simple slender systems founded on dry
2000; Gazetas & Apostolou, 2004; Mergos & Kawashima, sand are designed to be equivalent in terms of vertical factor
2005; Gajan & Kutter, 2008; Anastasopoulos et al., 2010a; of safety. They are then subjected to lateral monotonic
Gelagoti et al., 2012; Kourkoulis et al., 2012). loading till overturning and lateral slow cyclic loading. The
Several studies have explored the behaviour of foundations response of the equivalent systems is compared in terms of
under lateral and combined loading, both theoretically (Nova moment capacity, settlement accumulation during cyclic
& Montrasio, 1991; Butterfield & Gottardi, 1994; Paolucci, loading, stiffness degradation and energy dissipation. The
1997; Bransby & Randolph, 1998; Martin & Houlsby, 2001; results elucidate some salient features of the behaviour of
Gourvenec & Randolph, 2003; Chatzigogos et al., 2009) and shallow foundations subjected to large deformations, offering
experimentally (Negro et al., 2000; Gajan et al., 2005; Gajan a quantification of the role of scale effects in cyclic founda-
& Kutter, 2008; Paolucci et al., 2008; Anastasopoulos et al., tion response.
Manuscript received 10 April 2014; revised manuscript accepted 23 PROBLEM DEFINITION AND EXPERIMENTAL
October 2014.
METHODOLOGY
Discussion on this paper closes on 1 April 2015, for further details see
p. ii. A series of centrifuge model tests were conducted in the
Rensselaer Polytechnic Institute, Troy, NY, USA. 3 m radius, 150 g-tonne capacity centrifuge of the Centre for
† University of Dundee, UK; formerly National Technical University Earthquake Engineering Simulation at Rensselaer Polytech-
of Athens, Greece. nic Institute (RPI). The corresponding reduced-scale 1g tests
‡ National Technical University of Athens, Greece. were performed at the Laboratory of Soil Mechanics of the
155
KOKKALI, ANASTASOPOULOS, ABDOUN AND GAZETAS
National Technical University of Athens (NTUA). The ex- sidering the shallow nature of the rocking failure mechan-
perimental investigation focuses on the response of a rigid ism.
single mass slender system supported on a surface footing This prototype system was appropriately scaled down
and subjected to monotonic and slow cyclic lateral loading. according to the relevant centrifuge and 1g scaling laws
The prototype soil–structure system studied is shown in Fig. (Wood, 2003) and the resulting geometries for the centrifuge
1(a). Founded on a square surface footing of width B ¼ 3 m, and the 1g models are shown in Fig. 1. The lateral loading
the system has its centre of mass located at 6 .9 m above tests were conducted at a 50g centrifuge acceleration (1:50
ground. The oscillator is rigid in order to focus on the non- scale) and a scale of 1:20 was selected for the reduced-scale
linear foundation response. The supporting soil consists of 1g tests. Vertical push tests were conducted prior to the
dry sand of adequate thickness D � 3 .3B (¼ 10 m), con- lateral push tests in order to determine the bearing capacity
of the soil–foundation systems and the corresponding verti-
cal factors of safety (FS). The experimental configurations
for the centrifuge and the 1g tests are presented in the
following paragraphs and details about the modelling and
the critical design parameters are also provided.
6·9 m
Experimental set-up for centrifuge tests
The experimental configurations for the vertical and lateral
loading centrifuge tests are depicted in Fig. 2. A four-
degrees-of-freedom in-flight robot designed to perform multi-
3m ple tasks while the centrifuge is spinning was used in the
experimental series. The robot is capable of articulating in
Dry sand 10 m three linear dimensions and rotating around one axis with
variable speed. It can operate in single instruction mode or
follow programmatic scripts. Custom tools for the end of the
robotic arm (robot end-effector) were fabricated for the
vertical and lateral push tests. While the bearing capacity
(a) tests and the lateral monotonic push-over tests were con-
ducted in manual mode, the robot was programmed for the
application of the cyclic loading path assuring control and
consistency of the applied displacement and velocity through
all the loading cycles.
The set-up for the vertical push tests is shown in detail in
13·8 cm Scale 1:50
Fig. 2(a) (plan view) and Fig. 2(b) (side view). A model
square steel foundation of width B connected to the robot
custom tool was placed at the centre of a square container.
Adequate distance from the lateral boundaries (5B) was
assured so that boundary effects were avoided. The founda-
6 cm tion was pushed down until bearing capacity failure was
indicated. The vertical displacement was applied and re-
Dry Nevada sand 20 cm corded by the robot and the reaction load was measured by
the robot load cell.
A rectangular container provided two test locations for the
monotonic and the slow cyclic lateral push tests. Adequate
distance from the box lateral boundaries (4B) and between
(b) the two test locations (5B) was assured to minimise bound-
ary effects and interference between the different tests (Figs
2(c) and 2(d)). The structure was a three-piece unit compris-
ing a steel foundation, a steel column and a steel mass
located at a specified height. Sandpaper was placed beneath
the foundation in order to attain an adequately rough foun-
34·5 cm Scale 1:20 dation–soil interface and minimise sliding. A spherical alu-
minium attachment on top of the structure was pushed
laterally by the robot during cyclic loading. This structure–
sphere assembly was properly designed so that the structure
could freely move without any lateral or vertical restraints.
Details of this connection are shown in Fig. 2(d) and later in
15 cm
Fig. 4(c). After cyclic testing, the structure was moved to
Dry Longstone sand the second test location for the monotonic push-over test,
50 cm
and lateral displacement was applied against the side face of
the structure until it overturned.
A biaxial load cell, connected to the robot custom tool,
measured the horizontal force in both the loading (x) and
the transverse (y) direction, while the vertical force was
(c)
monitored by the robot load cell. On-board cameras captured
Fig. 1. Schematic illustration of the soil–structure systems stud- the horizontal and vertical displacements and rotation of the
ied: (a) prototype system; (b) centrifuge model; (c) reduced-scale structure. Specialised software was used to analyse the
1g model recorded videos and extract displacement–time histories.
156
STATIC AND CYCLIC ROCKING ON SAND
5B Square
foundation
BB Vertical load and
displacement measured
by the robot load cell
5B 5B
5B 5B B 5B
3·3B
Dry Nevada sand
(a) (b)
Location 1: Monotonic loading

Location 2: Cyclic loading
C1, C2, C3, C4: Camera views

for motion tracking
3B
Location 1 Location 2 C4
C1
B
4B B 5B B 4B
3B
C2 C3
(c)
Biaxial load cell on robot

custom tool
Location of the
centre of mass Mass
Detail of
attachment used
in cyclic loading
Column
h 2·3B
Footing
4B B 5B B 4B
3·3B
Dry Nevada sand
(d)
Fig. 2. Centrifuge model containers and experimental set-up for vertical and lateral loading
tests: (a) vertical push test: plan view; (b) vertical push test: side view; (c) lateral push test:
plan view; and (d) lateral push test: side view along the loading axis
Tracking targets were mounted on the structure along each locations, the cameras focused on a square grid in order to
axis, and high-intensity light-emitting diodes (LEDs), placed correct for lens distortion and field of view perpendicularity.
at appropriate angles to minimise glare and reflection, The corrected video was then loaded into the software
enhanced video quality thus facilitating motion tracking. package Tema for tracking. Fig. 3 shows a snapshot during
After the cameras and the structure were placed at their final tracking of a lateral monotonic push-over test.
157
Aluminium plate attached

to the foundation only to
facilitate motion tracking
Column 2
Column 1 Robot custom tool

Foot_left connected to square footing
Foot_ centre for soil surface preparation
Foot_right
Foot_left
Foot_centre
Foot_right
[1 103 m] T 86 500·0 ms
Foot_left (a)
Robot–structure
0
connection for
Foot_centre cyclic lateral
Biaxial load cell on loading SDOF
robot custom tool system
Foot_right
5
Spirit
level
[1 103 ms]
0 10 20 30 40 50 60 70 80 90 100 SDOF system
Fig. 3. Motion tracking during monotonic push-over test (snap-

shot from Tema software)
Camera
An important part of the set-up procedure of the lateral

push tests was the surface preparation and the alignment of (b) (c)
the structure with respect to the robot loading axis. Placing Fig. 4. Centrifuge lateral push tests: (a) soil surface preparation;
the foundation with an initial inclination in any direction (b) alignment and placement of structure on soil surface;
could result in undesired pre-stressing of the soil or biaxial (c) structure located at test location
loading. Therefore, the soil surface was levelled and the
structure was precisely aligned to the robot loading axis, as
shown in Fig. 4. In both cases, the robot custom tools were Experimental set-up for reduced-scale 1g tests
utilised to smooth and level the soil surface without disturb- Similar experimental configurations were developed in the
ing the soil density and to place the footing at the specified reduced-scale 1g test series. Fig. 5(a) shows the set-up of
test location. Fig. 4(c) shows the structure placed at the the bearing capacity tests and Fig. 5(b) the one used in the
cyclic loading test location and the camera view used to lateral push tests. The locations of the models with respect
track the motion of the footing along the loading axis. to the lateral boundaries of the rigid container used in the
Vertical Hinged
slider connection
Screw-jack Load
actuator cell
h 2·3B
Load
cell
5B B 5B 5B B 5B
Dry Longstone sand 3·3B Dry Longstone sand 3·3B
(a) (b)
Laser displacement transducer

Wire displacement transducer
Fig. 5. Model containers and experimental set-up for reduced-scale 1g tests: (a) vertical push test set-up; (b) lateral push test set-up
158
experimental series are also shown. A push-over apparatus,
fixed to a reaction wall and consisting of a servomotor Load cell
attached to a screw-jack actuator, was used to apply the
vertical or horizontal displacement. Four laser displacement
transducers measured the settlement of the four corners of
the footing during the vertical test and a load cell connected
to the edge of the actuator measured the vertical reaction
force.
The foundation–structure model consisted of a square
aluminium footing rigidly connected to a pair of rigid steel
columns that supported an aluminium slab. Steel plates were
symmetrically placed above and below the slab to model the
system’s mass. Sandpaper was used beneath the foundation
to simulate a realistically rough soil–footing interface. In Square foundation
this set-up the free end of the actuator was connected to the
structure model using a vertical slider and a hinged connec-
tion in series. This connection allowed the system to freely
settle, slide or rotate as horizontal displacement was applied.
The horizontal load was measured by a load cell inserted
(a)
between the vertical slider and the hinged connection. Hor-
izontal and vertical displacements were recorded through a
Load
system of wire and laser displacement transducers. Accurate cell
positioning of the structure at the test location without
disturbing the soil surface was achieved by a system of four
mechanical jacks. Photographs of the experimental config-
uration are shown in Fig. 6.
SDOF
system
Soil properties and samples preparation
Nevada 120 sand was dry pluviated to the desired density
with a consistent manual technique in the centrifuge test
series. Dry Longstone sand was layered in the 1g test
container through an electronically controlled raining system,
capable of producing sand specimens of controllable relative
density. The raining system has been calibrated through a
series of pluviation tests documented in Anastasopoulos et Wire
al. (2010b). Nevada 120 sand is a laboratory grade with displacement
D50 ¼ 0 .15 mm and uniformity coefficient Cu ¼ 2 .35. Long- transducers
stone sand is an industrially produced fine and uniform
quartz sand, also having D50 ¼ 0 .15 mm and uniformity (b)
coefficient Cu ¼ 1 .42. The grain size distribution of both
sand specimens is shown in Fig. 7 and their properties are Fig. 6. Photographs of the reduced-scale 1g tests: (a) vertical push
test; (b) cyclic lateral push test
summarised in Table 1.
The stress level prevailing in the 1g tests is unavoidably 100
low; therefore, the strength characteristics of Longstone sand Longstone sand
need to be known at a wide range of stresses. Fig. 8 shows 80
the dependence of the angle of shearing resistance on the
Percentage finer: %
Nevada sand
normal stress level, as described in Anastasopoulos et al.
60
(2010b) based on laboratory tests performed for two relative
densities (Dr ¼ 45% and 80%). The friction angle of Nevada
40
sand at three relative densities (Dr ¼ 40%, 60% and 70%),
at a reference mean effective normal stress ¼ 100 kPa, is
also shown as three points in Fig. 8 (Arulmoli et al., 1992). 20
The two stress ranges prevailing at a depth equal to one
foundation width in each type of test are also depicted. The 0
lower bound corresponds to the average geostatic stress for 0·01 0·1 1 10
Particle size: mm
these densities at this reference depth. The upper bound
takes into account the dead load of the superstructure. This Fig. 7. Grain size distribution of Nevada and Longstone sand
diagram will be revisited in the discussion of the test results
later on. Table 1. Summary of soil properties for Nevada and Longstone
sand
Superstructure modelling Properties Nevada 120 sand Longstone sand

Instead of scaling the dead load of the superstructure, a
different methodology was followed, namely the vertical emax 0 .887 0 .995
load over capacity ratio of the compared systems was kept emin 0 .511 0 .614
D50 0 .15 mm 0 .15 mm
constant in each set of experiments. This ratio is ex- Cu 2 .35 1 .42
pressed through the vertical factor of safety FS and is Gs 2 .67 2 .64
directly correlated to the rocking response of shallow
159
1g tests
stress range vertical stiffness of the 1g soil–foundation systems. Despite
60 these differences, some interesting common trends can be
55 observed: while the capacity of the foundation on dense
50
sand eventually reaches a constant value (plateau of the
Centrifuge tests curve), a continuing increase of the vertical load is observed
45 stress range for the loose and the two-layer soil profiles. This ‘hardening’
φ: deg
40 behaviour can be attributed to the stiffening of the loose

35 supporting soil as the footing is pushed into the ground, in
30
combination with the contribution of the footing embedment
to the bearing capacity (for all four configurations). The
25 vertical load capacity Nu of each system is also shown in
20 Fig. 9. These values were determined in a consistent way for
0 50 100 150 200 250 300 all eight tests. For the tests on loose sand and on the two-
σ: kPa
layer profiles the definition of the ultimate load is not
Longstone sand Nevada sand
straightforward because of the hardening behaviour. In this
Dr 80% peak Dr 70% study, Nu was defined as the load for which the rate of
Dr 45% peak Dr 60% change of the vertical stiffness became constant. In addition,
Dr 80% post peak Dr 40% the choice of the Nu point was consistent when comparing
the same profiles at 1g and high g level.
Fig. 8. Direct shear test results for Longstone sand: dependence A summary of the vertical load capacities as well as the
of the angle of shearing resistance on stress level (Anastasopoulos ratio of the 1g to the centrifuge ultimate loads is provided in
et al., 2010b). Friction angle for Nevada sand: evaluation from Table 2. The 1g vertical load capacities are 30–70% higher
isotropically consolidated undrained compression tests at refer- than the loads sustained in the centrifuge tests. The differences
ence mean effective normal stress 100 kPa (Arulmoli et al., 1992). observed between the two sets of tests can be further explored
The stress ranges at a depth equal to one foundation width are in view of the scale effects affecting the 1g foundation
also depicted. The lower bound corresponds to the geostatic stress response. Nevada and Longstone sands exhibit comparable
at this depth. The upper bound includes the geostatic stress and
shear resistance at large stress levels and similar densities, as
the stress induced by the structure dead weight
evidenced in Fig. 8. Nevertheless, below 50 kPa confining
pressure the shearing resistance of Longstone sand exhibits a
foundations, distinguishing the sinking from the uplifting remarkable increase. As shown in the figure, the stress level
rocking response. Additionally, the location of the centre for the 1g tests falls in this range, hence the overestimation of
of mass is a crucial design parameter related to the the friction angle and thereby of the ultimate vertical loads. To
geometric non-linearity of the system through the slender- further verify this response, the classical expression of Meyer-
ness ratio h/B, which controls the uplifting and overturning hof (1951) for the bearing capacity of a surface rectangular
potential as well as the ultimate moment capacity of the footing was employed to back-calculate the effective friction
foundation. angle of the dense and loose soil profiles. Whereas for the
According to the above, the mass and geometry of the centrifuge tests the friction angle of the loose and the dense
structure models were designed so that the equivalent centrifuge sand was calculated to be 33 .58 and 38 .58, respectively, the
and 1g single degree of freedom (SDOF) systems had the same friction angles of the same profiles at 1g were 478 and 508.
FS and h/B ¼ 2 .3. In terms of the vertical factor of safety FS, the Interestingly, the deviation between the centrifuge and the
bearing capacity tests provided the ultimate vertical load of 1g load–settlement curves becomes more considerable for
each soil–footing system and the mass of the superstructure was the dense soil (Table 2), indicating that the overestimation of
then adjusted to satisfy the FS criterion. As already mentioned, the friction angle is more prominent for a dense sand profile
no attempt was made to model the flexibility of the super- with distinct dilative behaviour. Regarding the two-layer
structure and both models were sufficiently rigid so that the profiles, the response is affected by both soil layers. Revisit-
system’s response was governed by non-linear soil–foundation ing Fig. 9, it can be noticed that the behaviour resembles
response. the response of the underlying loose sand, which is of
dominant importance in this deep failure mechanism.
BEARING CAPACITY TESTS

Two uniform soil profiles were considered: a loose sand
of relative density Dr ¼ 45% and a very dense sand of
Dr ¼ 90%. Two-layer soil profiles were also tested, consisting ROCKING RESPONSE IN VIEW OF SCALE EFFECTS
of a loose (Dr � 45%) bottom layer and a dense (Dr � 90%) As described above, an alternative methodology was fol-
upper layer depth z. In the following, the depth of the upper lowed for the superstructure design in order to maintain
dense layer (z) is normalised by the foundation width: z/B. similitude by keeping the vertical factor of safety and the
Two layered profiles were investigated: z/B ¼ 0 .25 and 0 .5. slenderness ratio of the models constant and directly com-
The model square foundations were subjected to vertical paring the centrifuge and the 1g rocking response. Thus, the
push tests in order to estimate the bearing capacity of four loose soil profile was chosen as a reference case and the
soil–foundation systems and design the centrifuge and the superstructure mass was adjusted and distributed so that a
1g superstructures. In the following paragraphs, a compari- slenderness ratio h/B ¼ 2 .3 and a factor of safety FS ¼ 5
son between the two types of tests is presented. All results were achieved for each system on loose sand. Using the
correspond to prototype units. same superstructures, the two-layer soil–foundation systems
Figure 9 compares the vertical load–settlement curves of yielded factors of safety equal to 5 .5 (for z/B ¼ 0 .25) and 7
each of the four soil profiles. A first comparison reveals (for z/B ¼ 0 .5). The factors of safety on dense sand were
discrepancies between the centrifuge and the 1g response. slightly different and in order to avoid misinterpretations of
The 1g models sustain higher vertical loads than the corre- the test results this case is not be presented here. A
sponding centrifuge systems. The significantly steeper initial summary of the vertical factors of safety is included in
slopes of the 1g load–settlement curves indicate a larger Table 2.
160
Uniform soil profiles Two-layer soil profiles
N Foundation N
width B
s z s Dense sand
Loose or dense
sand Loose sand
20
50
Dr 45%
38·6 MN
15 13·6 MN 40
10·3 MN 30
N: MN
N: MN
10 22·8 MN
20
5
10
0
0
0 0·1 0·2 0·3 0·4 0·5 0·6
0 0·1 0·2 0·3 0·4 0·5 0·6
s: m s: m
(a) (b)
25
25
19·5 MN
20
20
15·6 MN
14·8 MN
15
15
N: MN
11·2 MN
N: MN
10
10
5
5
0
0
0 0·1 0·2 0·3 0·4 0·5 0·6
0 0·1 0·2 0·3 0·4 0·5 0·6
s: m s: m
(c) (d)
Centrifuge tests 1g tests
Fig. 9. Load–settlement curves obtained from monotonic vertical centrifuge and 1g push tests on four soil profiles: (a) loose
sand; (b) dense sand; (c) z/B 0 .25; (d) z/B 0 .5
Table 2. Summary of vertical push-down test results
Soil profile Centrifuge test, 1g test, Nu: FS Nu, 1g /Nu centrifuge

Nu: MN MN
Centrifuge 1g model
model
Loose sand 10 .3 13 .6 5 5 1 .32

Dense sand 22 .8 38 .6 11 14 1 .69
z/B ¼ 0 .25 11 .2 15 .6 5 .5 5 .5 1 .39
z/B ¼ 0 .5 14 .8 19 .5 7 7 1 .32
Monotonic lateral loading systems reaching higher moment capacities and exhibiting a
The soil–structure systems described above were subjected more pronounced uplifting behaviour than those in the
to lateral loading until they overturned. Fig. 10 depicts the centrifuge. These trends are hardly surprising
moment, the settlement and the rotational stiffness of the six
(a) higher angle of shearing resistance results in higher
systems as functions of the footing rotation. The moment is
moment capacity
calculated as the product Fh, using the horizontal force F
(b) the geometric non-linearity (uplifting) in such systems is
measured by the load cell multiplied by the lever arm h.
governed by the stiffness of the soil; the stiffer soil in the
The settlement that is induced only by the lateral loading
1g test leads to greater uplifting.
refers to the centre of the footing. Overall, the two sets of
tests show several differences between them, with the 1g When it comes to the inelastic response of the systems,
161
δ
ϑ
Loose sand: FS 5 z/B 0·25: FS 5·5 z/B 0·5: FS 7

3000 3000 3000
2000 2000 2000
M: kN m
M: kN m
M: kN m
1000 1000 1000
0 0 0
0 0·04 0·08 0·12 0·16 0 0·04 0·08 0·12 0·16 0 0·04 0·08 0·12 0·16
ϑ: rad ϑ: rad ϑ: rad
0·16 0·16 0·16
0·08 0·08 0·08

s: kN m
s: kN m
s: kN m
0 0 0
0·08 0·08 0·08

0·16 0·08 0 0·08 0·16 0·16 0·08 0 0·08 0·16 0·16 0·08 0 0·08 0·16
60 60 60
KR: kN m/rad
KR: kN m/rad
KR: kN m/rad
40 40 40
20 20 20
0 0 0
0·0001 0·001 0·01 0·1 1 0·0001 0·001 0·01 0·1 1 0·0001 0·001 0·01 0·1 1
(a) (b) (c)
Fig. 10. Comparison of monotonic centrifuge and 1g push tests in terms of moment–rotation, settlement–rotation and rotational stiffness
against rotation: (a) loose sand; (b) two-layer soil profile z/B 0 .25; (c) two-layer soil profile z/B 0 .5
that is, the moment degradation and the subsequent overturn The settlement–rotation curves of the two test systems
due to P–� effects, a very good agreement is observed in almost coincide at small rotational amplitudes while dis-
terms of the overturning angle since the systems share the tinct differences are noticed at larger rotations. The 1g
same slenderness ratio. systems uplift from the supporting soil relatively quickly
Unsurprisingly, the moment capacity reached in the 1g (uplift is denoted by the upward change of the slope of the
two-layer system of z/B ¼ 0 .5 is overestimated to a greater settlement–rotation curve), whereas the centrifuge systems
extent than in the other two soil profiles. Since the rocking keep sinking (for loose sand and z/B ¼ 0 .25) or uplift
failure only extends to a very limited depth beneath the slightly and much later (for z/B ¼ 0 .5). Evidently the
foundation, not more than half the width, it seems that scale vertical stiffness of the 1g systems is larger, following the
effects become more adverse with the presence of the larger soil stiffness and friction angle. The comparison
dilative dense upper layer that contains the rotational failure under monotonic loading is concluded with the plots of the
surface. The shape of the 1g moment–rotation curve reflects secant rotational stiffness as a function of the angle of
this behaviour. At low rotational amplitudes and therefore rotation. The accurate measurement of KR is not feasible at
relatively low confining stresses, the 1g moment–rotation small rotations due to limitations in sensor accuracy. The
curve significantly deviates from the centrifuge curve. After real data are plotted for rotational amplitudes larger than
the peak of the curve and when the stresses induced to the 0 .008 rad and curve fitting is adopted to extrapolate the
soil due to lateral loading have increased (around small-strain rotational stiffness (dotted lines). Invariably, all
Ł ¼ 0 .1 rad) the 1g moment–rotation curve starts ap- the 1g systems exhibit significantly higher rotational stiff-
proaching the centrifuge curve and they eventually converge. ness. The deviation becomes smaller at larger rotational
162
0·50
amplitudes where the systems have uplifted from the 0·40
0·35
supporting soil and reduction of the effective contact area 0·30
0·25
has occurred. At that point, the geometric non-linearity 0·25 0·18 0·20
dominates the response and the rotational stiffness quickly 0·14 0·16
0·08 0·10 0·12
degrades in all cases. 0·02 0·04
0·06
δ/δR
0
Cyclic lateral loading

0·25
The performance of the SDOF systems when subjected to
slow cyclic lateral loading is evaluated in this section. Gajan
et al. (2005) have shown through a series of centrifuge tests 0·50
that the response of a rocking foundation to seismic shaking 0 2 4 6 8 10 12 14
can be fairly well predicted from slow cyclic tests. Fourteen Cycle number
cycles of increasing displacement amplitude were applied at
Fig. 11. Normalised cyclic lateral displacement applied at the
the centre of mass of the structures (Fig. 11). The lateral centre of mass of SDOF systems
displacement is normalised to the overturning displacement
of the equivalent rigid block on rigid base (R ¼ B/2 ¼
1 .5 m). Even though the chosen load sequence is not repre- The cyclic response of the compared systems is outlined
sentative of a specific earthquake, it allows the systems’ in terms of moment–rotation and settlement–rotation in Figs
performance to be compared under a wide range of displace- 12–14. Overall, qualitative and quantitative differences are
ment amplitudes, stressing them from their elastic all the observed. The 1g systems consistently exhibit higher cyclic
way into their metaplastic regime. moment capacity and accumulate more settlement than the
s
Centrifuge test 1g test
Loose sand: FS 5 Loose sand: FS 5
4000 4000
rM 80%
2000 2000
rM 60%
M: kN m
M: kN m
0 0
2 000 2 000
4 000 4 000
0·12 0·06 0 0·06 0·12 0·12 0·06 0 0·06 0·12
ϑ: rad ϑ: rad
0 0
0·1 0·1
0·2 0·2
s: m
s: m
0·22 m
0·3 0·3
0·4 0·4
0·43 m
0·5 0·5
0·12 0·06 0 0·06 0·12 0·12 0·06 0 0·06 0·12
ϑ: rad ϑ: rad
(a) (b)
Fig. 12. Moment–rotation and settlement–rotation curves obtained from slow cyclic tests on loose sand (Dr 45%): (a) centrifuge
test; (b) 1g test
163
δ
ϑ
s
z/B 0·25: FS 5·5 z/B 0·25: FS 5·5
4000 4000
rM 75%
2000 rM 50% 2000
M: kN m
M: kN m
0 0
2 000 2 000
4 000 4 000
0·12 0·06 0 0·06 0·12 0·12 0·06 0 0·06 0·12
ϑ: rad ϑ: rad
0 0
0·1 0·1
0·14 m
s: m
s: m
0·2 0·2
0·3 0·3
0·31 m
0·4 0·4
0·12 0·06 0 0·06 0·12 0·12 0·06 0 0·06 0·12
ϑ: rad ϑ: rad
(a) (b)
Fig. 13. Moment–rotation and settlement–rotation curves obtained from slow cyclic tests on two-layer soil profiles (z/B 0 .25):
(a) centrifuge test, (b) 1g test
corresponding centrifuge systems. Nevertheless, the qualita- the loose sand tests but indicated by the decreasing rate of
tive comparison reveals several interesting similarities be- settlement accumulation in all settlement–rotation plots; (b)
tween the response observed in the centrifuge and the 1g due to this compaction-style densification, the footing pene-
cyclic tests. The trend established in the monotonic push- trates into the ground and hence it soon becomes essentially
over tests, where the moment capacity of the 1g tests embedded; (c) on this denser soil layer the embedded
reached higher levels, is also observed during cyclic loading. footing enjoys a greater ultimate moment resistance, hence
No substantial reduction of the moment capacity with num- the great overstrength, especially for the 1g tests; (d) during
ber of cycles takes place. Additionally, a considerable large amplitude rotation angles (Ł . 0 .04 rad), the footing
amount of energy is dissipated in the foundation as sug- (over an already denser soil) tends to uplift, although still
gested by the wide hysteresis loops. Rotational stiffness eventually accumulates settlement. Similar trends have been
degradation is also observed and is discussed later in more noted by Drosos et al. (2012) and Anastasopoulos et al.
detail. (2013) in 1g experiments. But even for footings on saturated
The cyclic moment capacity well exceeds the monotonic clay (under undrained conditions), Panagiotidou et al. (2012)
backbone moment–rotation curve for all systems under com- observed theoretically a cyclic overstrength, which was at-
parison. This apparent moment overstrength can be quanti- tributed to the beneficial role of P–� effects acting in the
fied by an overstrength ratio rM, defined as the increase in opposite to the loading direction. Hence, a portion of the
the cyclic moment capacity divided by the monotonic mo- observed overstrength is not necessarily related to densifica-
ment capacity. For the loosest soil and the 1g tests it reaches tion.
80%; for the centrifuge tests it is smaller than 60%. In both The most evident difference between the centrifuge and the
types of test rM is a function of the soil profile and the 1g tests lies in the accumulation of settlement during cyclic
vertical factor of safety. Several interesting phenomena take loading. The settlement response of the compared systems
place in parallel during cyclic loading: (a) densification of varies upon the vertical factor of safety. The systems on loose
the soil under the footing, a phenomenon most prominent in sand (FS ¼ 5) settle at the very first loading cycles of small
164
δ
s
z/B 0·5: FS 7 z/B 0·5: FS 7
4000 4000
rM 20%
2000 rM 20% 2000

M: kN m
M: kN m
0 0
2 000 2 000
4 000 4 000
0·12 0·06 0 0·06 0·12 0·12 0·06 0 0·06 0·12
ϑ: rad ϑ: rad
0 0
0·1 0·1
0·1 m
s: m
s: m
0·2 0·2
0·2 m
0·3 0·3
0·12 0·06 0 0·06 0·12 0·12 0·06 0 0·06 0·12
ϑ: rad ϑ: rad
(a) (b)
Fig. 14. Moment–rotation and settlement–rotation curves obtained from slow cyclic tests on two-layer soil profiles (z/B 0 .5):
(a) centrifuge test; (b) 1g test
rotational amplitudes and tend to uplift while still accumulating pronounced uplifting behaviour. This could be attributed to
settlement when larger displacement is applied. The systems on the nature of cyclic loading and the potential for soil stiffen-
the two-layer profiles exhibit a more prominent uplifting being after multiple loading cycles. In an attempt to under-
haviour, accumulating less settlement. This is particularly stand the cyclic settlement response, the last loading cycles
evident in the case of z/B ¼ 0 .5 (FS ¼ 7). For all soil profiles of the cyclic loading tests on the z/B ¼ 0 .5 profiles are
considered, the rate of settlement accumulation is larger in isolated and the settlement–rotation curves are compared in
the 1g tests. At the end of the cyclic push test, the 1g systems Fig. 16. The settlement accumulated in the previous cycles
have settled twice as much as the centrifuge systems. The has been subtracted to allow more direct comparison. The
settlement–rotation response is summarised in Fig. 15, plotting two systems follow almost the same path in the first quarter
the settlement per cycle against the cycle rotation half- cycle (A to B) up to the point that the centrifuge system
amplitude. The settlement is normalised to the foundation width enters the unloading quarter cycle, while the 1g system is
B and the rotation is normalised to the overturning angle of the still loaded to a slightly higher rotational amplitude. The
equivalent rigid block on rigid base (ŁR ¼ B/2h, where B is the difference in the response becomes clear during the unload-
block width and 2h is the block height). For all sets of systems ing branch and continues thereafter. The accumulated settle-
(FS ¼ 5, 5 .5 and 7) the settlement per cycle induced during the ment during this half cycle is larger for the 1g system. A
1g slow cyclic push tests is higher at any rotational amplitude. reduced vertical stiffness of the 1g system that affects the
The divergence becomes larger as the rotation increases and the unloading quarter cycle and the accumulation of settlement
settlement obtained from the 1g tests reaches values up to could be implied at this point. This hypothesis for the 1g
double the settlement obtained from the centrifuge tests. system is further illustrated in the sketch of Fig. 16(b) (not
The cyclic settlement response contradicts the monotonic drawn to scale). During the first quarter cycle (loading from
settlement response where the 1g systems exhibited more A to B), large stress concentration takes place beneath the
165
ϑc
zero, and hence the geostatic stresses become prevalent and
sc scale effects play a substantial role. And since the geostatic
stresses are unrealistically low in the 1g tests, soil stiffness
is much lower at this stage and until the foundation regains
full contact with the soil, the potential for settlement is
0·16 substantially larger. This phenomenon takes place in every
Loose sand unloading quarter cycle, leading to an overestimation of the
FS 5
settlement in the 1g tests.
0·12
Stiffness degradation and energy dissipation

sc /B
0·08 As the systems are subjected to large amplitude rotation

cycles, softening occurs and rotational stiffness degradation
is observed. At these amplitudes a gap may form at one side
0·04 of the footing while soil yielding occurs at the other side. In
the following half cycle this phenomenon reverses and the
gap has to close before the opposite corner of the footing
0 starts uplifting. This gap formation and closing that occurs
0 0·1 0·2 0·3 0·4 at large rotational amplitudes is responsible for the rotational
ϑc /ϑR stiffness degradation shown in the moment–rotation plots.
(a) The significant loss of contact between the foundation and
0·12
the supporting soil results in reduction of the rotational
z/B 0·25
FS 5·5 stiffness of the unloading branches and produces the char-
acteristic S-shape hysteresis loops. Even though the rota-
0·09
tional stiffness degradation was evident in both centrifuge
and 1g tests, the comparison with respect to the shape of the
hysteresis loops reveals a fundamental difference: while the
sc /B
0·06
centrifuge tests produce distinct S-shaped loops, a more oval
shape is demonstrated in the 1g plots. S-shaped moment–
rotation response has been previously observed in centrifuge
0·03 and large-scale experiments for systems with relatively high
factor of safety or systems on dense soil (Gajan et al., 2005;
Negro et al., 2000). In the same experimental investigations
0 the moment–rotation curves were more oval-shaped for
0 0·1 0·2 0·3 0·4
systems with lower factor of safety or systems on low-
ϑc /ϑR
density sand.
(b)
0·08
This section explores the rotational stiffness degradation
z/B 0·5 and energy dissipation that takes place during cyclic loading.
FS 7 To this end, the cyclic rotational stiffness and the damping
0·06
ratio were calculated for the different loading cycles that the
six systems were subjected to. Two different approaches
were followed for these calculations and are illustrated in
Fig. 17. According to method I, the rotational stiffness is
sc /B
0·04
defined as the slope of the line connecting the two tips of
the moment–rotation loop. Since the maximum moment and
maximum rotation do not occur simultaneously, an alterna-
0·02
tive rotational stiffness can be calculated as the ratio of the
maximum moment to the maximum cycle rotation (method
II). The respective elastic areas used for the calculation of
0
0 0·1 0·2 0·3 0·4
the damping ratio are shown on this plot.
ϑc /ϑR
Figure 18 focuses on the rotational stiffness degradation
(c)
and Fig. 19 on the dissipated energy during cyclic loading.
In order to avoid misinterpretations due to inaccuracies in
Centrifuge tests 1g tests test measurements, the very first and small amplitude load-
ing cycles are not included in the plots, and only cycles 3 to
Fig. 15. Evolution of settlement during cyclic push tests: normal- 14 are considered. The top graphs of Fig. 18 show the
ised settlement per cycle with respect to normalised cycle rotation rotational stiffness as calculated according to the methods
half amplitude: (a) loose sand; (b) two-layer soil profile z/B 0 .25; previously described. Following the trend noticed in the
(c) two-layer soil profile z/B 0 .5. B is the foundation width and monotonic push tests, the rotational stiffness measured in the
ŁR is the overturning angle of the equivalent rigid block on rigid 1g tests is larger at any rotational amplitude. This agrees
base with the more oval shape of the 1g moment–rotation loops.
In the bottom graphs the cyclic rotational stiffness is nor-
malised to the small strain rotational stiffness as defined in
the monotonic push over tests. Interestingly, the normalised
foundation corner as the latter uplifts from the supporting results from the centrifuge and 1g tests follow identical
soil. As a result, at this stage scale effects are not important, degradation trend. In both absolute and normalised rotation
as the geostatic stresses are not prevalent. As the second plots, no distinct dependence of KR on the soil profile (or
quarter cycle follows (unloading from B to C), the stress alternatively the factor of safety) could be possibly extracted
field under the uplifted footing becomes practically equal to for the tests of the same type (centrifuge or 1g) since there
166
0·08
z/B 0·5: FS 7
D 14th loading cycle B
0·04
Δs: m
A
0 1g curve
Centrifuge
C
curve
E
ΔsAC 2·6 mm ΔsAC 7·7 mm
ΔsCE 6·4 mm ΔsCE 12·7 mm
0·04
0·10 0·05 0 0·05 0·10
ϑ: rad
(a)
Loading quarter Unloading quarter

cycle: A → B cycle: B → C
Initial soil surface
σ0
Soil surface at A
σ 0 during unloading quarter
High stress concentration
around the foundation corner cycle leads to higher settlement
and soil yielding accumulation in 1g test
(b)
Fig. 16. (a) Settlement–rotation curve for the last loading cycle on the z/B 0 .5 profiles and comparison of settlement induced
in each half cycle; (b) schematic illustration of the loading–unloading sequence in the 1g test (not drawn to scale)
Method I Method II rotation plots. Additionally, some dependence on the soil

M M conditions or the factor of safety is present. An increase of
Mmax
M(ϑmax) the damping ratio is noticed for decreasing factor of safety.
ΔEel ϑmax ΔEel ϑmax Overall, the average damping values are slightly larger for
ϑmax ϑ ϑmax ϑ
the 1g tests.
ΔEcycle ΔEcycle
M(ϑmax)
Mmax
SUMMARY AND CONCLUSIONS
2M(ϑmax) 2Mmax
KR KR An experimental study on dry sand was performed for the
2ϑmax 2ϑmax
rocking response of slender systems, and for comparison
ΔEcycle between centrifuge and reduced-scale 1g tests. Equivalent
Damping ratio
4πΔEel (as much as possible) systems on three different soil profiles
were subjected to monotonic and slow cyclic lateral loading.
Fig. 17. Definition of cyclic rotational stiffness and damping ratio
derived from the cyclic moment–rotation diagrams following two
Before the lateral loading tests, the bearing capacity of the
approaches soil-foundation systems was measured through vertical push
tests that provided additional information regarding the role
of scale effects on bearing capacity. Based on the measured
are no substantial differences in the rotational stiffness vertical ultimate force of each foundation, the systems were
values of the loose and the two-layer profiles. designed so as to maintain an analogy between the key
In the centrifuge tests, the damping ratio varies from 20% dimensionless parameters of the rocking response: the verti-
to 40%, being more or less independent of the soil condi- cal factor of safety FS and the slenderness ratio h/B. The
tions or, alternatively, of the FS of the system. The damping key conclusions can be summarised as follows.
ratio is slightly larger for small rotation amplitudes, but for
Ł . 0 .03 rad it remains constant. These observations apply • The low confining pressure prevailing in the 1g tests led
to the results obtained by either method I or II. On the other to overestimation of the bearing capacity, since the
hand, different trends are noted for the 1g tests. Increasing effective friction angle of the soil is highly dependent on
damping ratio with rotation is seen in the 1g tests that the stress level. This overestimation became more
follow method I, while damping remains more or less prominent when a dense sand profile was considered.
constant when calculated with method II. In the first case • The comparison between equivalent centrifuge and 1g
the increase might be attributed to asymmetries in the cyclic lateral loading tests provided insight in several aspects
loading that are also evident in the respective moment– of the problem. The 1g tests exhibited qualitative
167
Loose sand Loose sand
z/B 0·25 z/B 0·25
z/B 0·5 z/B 0·5
30 30
25 25
KR: kN 104 m/rad
KR: kN 104 m/rad

20 20
15 15
10 10
5 5
0 0
0 0·02 0·04 0·06 0·08 0·10 0 0·02 0·04 0·06 0·08 0·10
ϑ: rad ϑ: rad
(a) (b)
1·0 1·0
0·8 0·8
0·6 0·6
KR /KRo
KR /KRo
0·4 0·4
0·2 0·2
0 0
0 0·1 0·2 0·3 0·4 0·5 0 0·1 0·2 0·3 0·4 0·5
ϑc /ϑR ϑc /ϑR
(c) (d)
Fig. 18. Rotational stiffness degradation: top graphs show rotational stiffness plotted against cycle rotation half amplitude
and bottom graphs show ratio of rotational stiffness KR to small-strain rotational stiffness KRo as defined in the monotonic
push tests plotted against the normalised cycle rotation half amplitude. The rotational stiffness was calculated according to
method I in (a) and (c), and according to method II in (b) and (d)
similarities with the centrifuge tests capturing the highly even though the actual rotational stiffness was larger in
ductile cyclic response, the stiffness degradation, the the 1g tests.
high level of energy dissipation of the rocking systems,
as well as the cyclic moment overstrength. A difference Summarising, the comparison presented in this paper
was observed in the moment–rotation hysteresis loops, showed that reduced-scale 1g tests can provide valuable
which were more oval-shaped in the 1g tests, offering insights to the rocking response of SDOF systems only if
larger energy dissipation. properly interpreted, with due consideration to the actual soil
• Densification of the sandy layers during cyclic loading properties at very small confining pressures. This topic could
clearly played a significant role in all the tests, either be further explored with direct comparisons of true seismic
centrifuge or 1g. shaking that could reveal potential differences related to the
• As expected, the quantitative comparison revealed stress dependent dynamic soil behaviour and the character-
discrepancies in terms of vertical and moment capacity istics of the applied ground motion. The experimental find-
as well as settlement accumulation. The increased ings presented in this paper could serve as a baseline to
moment capacity observed in the 1g tests is attributed interpret these differences.
to the overestimation of the angle of shearing resistance.
The increased settlement is most probably due to the
reduced vertical stiffness during the unloading phase of ACKNOWLEDGEMENTS
the loading sequence, when the effective stress exerted to The authors would like to acknowledge financial support
the soil by the foundation is zero and the response is from the EU 7th Framework research project funded through
governed by the geostatic stresses. the European Research Council’s Programme ‘Ideas’, Sup-
• In terms of rotational stiffness degradation, the centrifuge port for Frontier Research – Advanced Grant, under contract
and 1g systems showed the same normalised response number ERC-2008-AdG 228254-DARE.
168
Loose sand z/B 0·25 z/B 0·25 Loose sand z/B 0·25 z/B 0·25
1·0 1·0
0·8 0·8
Damping ratio
Damping ratio
0·6 0·6
0·4 0·4
0·2 0·2
0 0
0 0·02 0·04 0·06 0·08 0·10 0 0·02 0·04 0·06 0·08 0·10
ϑ: rad ϑ: rad
(a)
1·0
1·0
0·8 Damping ratio 0·8

Damping ratio
0·6 0·6
0·4 0·4
0·2 0·2
0 0
0 0·02 0·04 0·06 0·08 0·10 0 0·02 0·04 0·06 0·08 0·10
ϑ: rad ϑ: rad
(b)
Fig. 19. Ratio of energy dissipation during cyclic push tests with respect to cycle rotation half amplitude. Damping was
calculated according to method I in (a) and method II in (b)
REFERENCES modeling of bridge systems designed for rocking foundations.

Anastasopoulos, I., Gazetas, G., Loli, M., Apostolou, M. N. & J. Geotech. Geoenviron. Engng ASCE 138, No. 3, 335–344.
Gerolymos, N. (2010a). Soil failure can be used for seismic Drosos, V., Georgarakos, T., Loli, M., Anastasopoulos, I., Zarzouras,
protection of structures. Bull. Earthquake Engng 8, No. 2, 309– O. & Gazetas, G. (2012). Soil–foundation–structure interaction
326. with mobilization of bearing capacity: An experimental study on
Anastasopoulos, I., Georgarakos, P., Georgiannou, V., Drosos, V. & sand. J. Geotech. Geoenviron. Engng ASCE 138, No. 11, 1369–
Kourkoulis, R. (2010b). Seismic performance of Bar-Mat rein- 1386.
forced-soil retaining wall: Shaking table testing versus numerical Gajan, S. & Kutter, B. L. (2008). Capacity, settlement and energy
analysis with modified kinematic hardening constitutive model. dissipation of shallow footings subjected to rocking. J. Geotech.
Soil Dynam. Earthquake Engng 30, No. 10, 1089–1105. Geoenviron. Engng ASCE 134, No. 8, 1129–1141.
Anastasopoulos, I., Kourkoulis, R., Gelagoti, F. & Papadopoulos, E. Gajan, S., Phalen, J. D., Kutter, B. L., Hutchinson, T. C. & Martin,
(2012). Rocking response of SDOF systems on shallow im- G. (2005). Centrifuge modeling of load deformation behavior of
proved sand: An experimental study. Soil Dynam. Earthquake rocking shallow foundations. Soil Dynam. Earthquake Engng 25,
Engng 40, 15–33. No. 7–10, 773–783.
Anastasopoulos, I., Loli, M., Georgarakos, T. & Drosos, V. (2013). Gazetas, G. & Apostolou, M. (2004). Nonlinear soil–structure inter-
Shaking table testing of rocking-isolated bridge pier on sand. action: Foundation uplifting and soil yielding. In Proceedings of
J. Earthquake Engng 17, No. 1, 1–32. the 3rd joint US–Japan workshop on soil–structure interaction,
Arulmoli, K., Muraleetharan, K. K. & Hossain, M. M. (1992). Menlo Park, CA, pp. 60–66.
VELACS –Verification of liquefaction analyses by centrifuge Gelagoti, F., Kourkoulis, R., Anastasopoulos, I. & Gazetas, G.
studies – Laboratory testing program, Soil data report. Irvine, (2012). Rocking isolation of low rise frame structures founded
CA, USA: The Earth Technology Corporation. on separate footings. Earthquake Engng Struct. Dynam. 41, No.
Bransby, M. F. & Randolph, M. F. (1998). Combined loading of 7, 1177–97.
skirted foundations. Géotechnique 48, No. 5, 637–655, http:// Gourvenec, S. & Randolph, M. F. (2003). Effect of strength non-
dx.doi.org/10.1680/geot.1998.48.5.637. homogeneity on the shape and failure envelopes for combined
Butterfield, R. & Gottardi, G. (1994). A complete three-dimensional loading of strip and circular foundations on clay. Géotechnique
failure envelope for shallow footings on sand. Géotechnique 44, 53, No. 6, 527–533, http://dx.doi.org/10.1680/geot.2003.53.6.
No. 1, 181–184, http://dx.doi.org/10.1680/geot.1994.44.1.181. 575.
Chatzigogos, C. T., Pecker, A. & Salençon, J. (2009). Macroelement Kourkoulis, R., Anastasopoulos, I., Gelagoti, F. & Kokkali, P.
modeling of shallow foundations. Soil Dynam. Earthquake (2012). Dimensional analysis of SDOF system rocking on
Engng 29, No. 5, 765–781. inelastic soil. J. Earthquake Engng 16, No. 7, 995–1022.
Deng, L., Kutter, B. L. & Kunnath, S. K. (2012). Centrifuge Martin, C. M. & Houlsby, G. T. (2001). Combined loading of
169
spudcan foundations on clay: numerical modelling. Géotechni- Panagiotidou, A. I., Gazetas, G. & Gerolymos, N. (2012). Pushover
que 51, No. 8, 687–699, http://dx.doi.org/10.1680/geot.2001.51. and seismic response of foundations on stiff clay: analysis with
8.687. P–Delta effects. Earthquake Spectra 28, No. 4, 1589–1618.
Mergos, P. E. & Kawashima, K. (2005). Rocking isolation of a Paolucci, R. (1997). Simplified evaluation of earthquake induced
typical bridge pier on spread foundation. J. Earthquake Engng permanent displacements of shallow foundations. J. Earthquake
9, No. 2, 395–414. Engng 1, No. 3, 563–579.
Meyerhof, G. G. (1951). The ultimate bearing capacity of founda- Paolucci, R., Shirato, M. & Yilmaz, M. T. (2008). Seismic behavior
tions. Géotechnique 2, No. 4, 301–332, http://dx.doi.org/10. of shallow foundations: Shaking table experiments vs. numerical
1680/geot.1951.2.4.301. modeling. Earthquake Engng Struct. Dynam. 37, No. 4, 577–
Negro, P., Paolucci, R., Pedretti, S. & Faccioli, E. (2000). Large- 595.
scale soil–structure interaction experiments on sand under cyclic Pecker, A. & Pender, M. (2000). Earthquake resistant design
loading. In Proceedings of the 12th world conference on earth- of foundations: New construction. In Proceedings of Geo-
quake engineering, Auckland, New Zealand, pp. 517–523. Upper Engineering 2000 conference, Melbourne, Australia, pp. 313–
Hutt, New Zealand: New Zealand National Society for Earth- 332.
quake Engineering. Priestley, M. J. N., Seible, F. & Calvi, G. M. (1996). Seismic design
Nova, R. & Montrasio, L. (1991). Settlement of shallow foundations and retrofit of bridges. New York, NY, USA: Wiley.
on sand. Géotechnique 41, No. 2, 243–256, http://dx.doi.org/ Wood, D. M. (2003). Geotechnical modelling. Boca Raton, FL,
10.1680/geot.1991.41.2.243. USA: CRC Press.
170

Geotechnical Earthquake Engineering-Géotechnique Symposium in Print 2015

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Geotechnical Earthquake Engineering-Géotechnique Symposium in Print 2015

Transféré par

Droits d'auteur :

Formats disponibles

Geotechnical

Dr Stuart Haigh, University of Cambridge, UK

Related titles from ICE Publishing:

Bio- and Chemo-mechanical Processes in Geotechnical Engineering

Earthquake Design Practice for Buildings, third edition.

Designers’ Guide to Eurocode 8: Design of buildings for earthquake resistance.

Earthworks: A guide, second edition.

ICE Manual of Geotechnical Engineering (2 volumes).

Core Principles of Soil Mechanics.

© Thomas Telford Limited 2016

All rights, including translation, reserved. Except as permitted by the Copyright,

Commissioning Editor: Laura Marriott

Typeset by Manila Typesetting Company

Session 1: Field behaviour of soils during earthquakes

Session 2: Model testing of geotechnical systems during earthquakes

Assessment of CPT-based methods for liquefaction evaluation in a

KEYWORDS: earthquakes; liquefaction; sands; seismicity

INTRODUCTION assumed that the severity of liquefaction manifestation is

OVERVIEW OF ROC ANALYSES

True positive rate, TPR

0·8 (a) Relative trends in model performance, as suggested by

Table 2. Summary of receiver operator characteristic (ROC) analyses

LPI Model All Marginal Moderate Severe Lateral Severe lateral

R&W98 4 .0 0 .73 3 .0 0 .68 5 .5 0 .62 10 .5 0 .69 4 .5 0 .83 10 .0 0 .66

R&W98 (R2  0·98) R&W98

MEA06 prediction error: LPI units

I&B082 prediction error: LPI units

generally consistent between models. In addition, maps 25

Correlation between liquefaction resistance and shear wave velocity of

KEYWORDS: centrifuge modelling; discrete-element modelling; particle-scale behaviour; liquefaction

INTRODUCTION geotechnical engineers a promising alternative to evaluate

Soroush, 2008). Ng & Dobry (1994) were among the first

σm0  100 kPa 20

Deviator anisotropic coefficients, adr , adn, adt

25 Normal contact force anisotropy, and

Sliding fraction, Sc: %

4·0 0 0 adt  0·073

During the liquefaction process, the number of inter-

Cyclic resistance ratio, CRR

4·2 0·1 Non-liquefaction

Dr  50·4%; σm0  100 kPa

Fig. 8. Curves of CSR against number of cycles to liquefaction for

Mobilised angle of internal friction, φ: degrees

υg  0·20 Yimsiri & Soga (2010) (loose, φmax)

μg  0·50 found to be negligible (e.g. V s only increases from 92 .17 m/s

Prediction from equation (5) (Gg  1 GPa)

α  2·205  108 α  4·600  109 α  1·406  109

Fig. 12. Influence of particle shear modulus on CRR–V s1 correlation

�  5·339 together with three pairs of bender elements, two pairs of

(3) Ahmadi & Paydar (2014) (Babolsar sand)

(5) Present study

Gg  6·649 GPa; μg  0·836

equation (8) using both the calibrated parameters described CONCLUSIONS

An interpretation of the seismic behaviour of reinforced-earth retaining

KEYWORDS: design; earthquakes; reinforced soils

INTRODUCTION soil and the reinforcement: in fact, the design is typically

tions using the computer code Flac v.5 (Itasca, 2005), in TT

Case A Case B Case C

kc  0·060 kc  0·066 kc  0·060

Limit analysis Limit analysis

In the soil model, damping results entirely from the Assisi

Flac model NZ-DFHS – free field

Seed & Idriss (1970)

Results of the dynamic analysis

Table 2. Summary of receiver operator characteristic (ROC) analyses

R&W98 (R2 0·98) R&W98

σm0 100 kPa 20

4·0 0 0 adt 0·073

Dr 50·4%; σm0 100 kPa

υg 0·20 Yimsiri & Soga (2010) (loose, φmax)

μg 0·50 found to be negligible (e.g. V s only increases from 92 .17 m/s

Prediction from equation (5) (Gg 1 GPa)

α 2·205 108 α 4·600 109 α 1·406 109

� 5·339 together with three pairs of bender elements, two pairs of

Gg 6·649 GPa; μg 0·836

kc 0·060 kc 0·066 kc 0·060

εu ∞ NZ-DFHS Rot. – kc 0·115

Fig. 5. Secant Young’s modulus Esec plotted against axial strain a D

20 80 200

300 p 495 kPa

101 100 101

100 101 101 102

0 .980 33 .04 0 .220

CENA – rock: ky 0·01 CENA – rock: ky 0·10 CENA – rock: ky 0·20

CENA – soil: ky 0·01 CENA – soil: ky 0·10 CENA – soil: ky 0·20

0·4 0·2 0 0·2 0·4 0·5 0 0·5

Vmax 25 cm/s; Amax 0·50g Vmax 50 cm/s; Amax 0·50g

CENA – rock: Amax 0·5g CENA soil: Amax 0·5g

79°00W 78°00W 77°00W

0·15 0·15 0·15 0·15

0·30 0·30 0·30 0·30

0·15 0·15 0·15 0·15

0·30 0·30 0·30 0·30

EQ ID EQ1 EQ2 EQ3 EQ4 EQ5 EQ6† EQ7† EQ8†