J.F. SEMBLAT, A.

PECKER

Waves and Vibrations in Soils:

Earthquakes, Traffic, Shocks,
Construction works

1x M 544256 0-1-2
Houtschild International Booksellers

Waves and Vibrations in Soils. Earthquakes

IJllll 11111111111111I111111111111111111111111111111 IIll

000186015900476070010010544256
mpsmn
551
2009

·~mm]
WAV
IUSS Press
9 78886 1 980303 ) lstituto Universitario di Studi Superiori di Pavia
 JEAN-FRAN<;OIS SEMBLAT ()EJECTIVE OF THE BOOI~

an-Fran.:;:ois Semblat is Head of the Dynamics, Waves and Vibrations
nit in the Division for Soil and Rock Mechanics at Laboratoire
entral des Ponts et Chausst'es (Paris, France) and Associate Professor
Ecole Polytechnique (Palaiseau, France). He has published over 100
::hnical papers in journals and conferences. He is on the editorial
)ard of the International Journal of Geomechanics (ASCE) and
iropean Journal of Environmental and Civil Engineering. He is
ember of the Scientific Con11nittees of the French Association for
rrhquake Eng. and the French Society of Soil Mechanics and
:otechnical Engineering. He is member of the board of the
ternational Association for Computer Ivfethods and Advances in
:omechanics and is associate 1nember of the American Society of
vil Engineers (ASCE).
: has received several awards for his work: French .Association for
rthquake Eng., European Association of Geoscientists and Eng.,
:ernational Association for Computer Methods in Geomechanics.
'fhe 1nain scientific and engineering goal of this book is to deal simul-
taneously with soil dynamics/vibrations and wave propagation in
soils (including seis1nic 'Naves). These various fields are generally con-
sidered separately and the important links benveen then1, bor.h fror_n
scientific and practical points of vie"'" are unfortunately not investi-
gated. They arc usua!ly considered in separate disciplines such as
eanhguake geotechnical engineering, civil engineering, mechanics,
geophysics, seisn1ology, nurnerical r:iode~ling, etc.. . .
The objective of the book is to offer in a single pt~bhcat1on a~ ovcrv1e'.v
of soil dynamics and \.vave propagation in soils with ernpha.51s on engi-
neering applications. It starts frotn a \vide variety o.f prac:ical ?ro~lerns
(e.g. traffic induced vibrations, dynan1ic con1pact1on,. v1b:anon isola-
tion), then deals with 1D and 2I)/3D \Vave propaganon u1 heteroge-
neous and attenuating media (\.vith application to laboratory and in
situ dvnamic characterization of soils), gives an overviC'\v of various
nume;'ical methods (e.g. FEi\11, BEM) to sirnulate wave propagation
(including nun1crical errors, radiation/absorbing con~tio~s, ~tc) and
finally investigates seis1nic \Vave propagation and amplification in com-
plex geological structures (e.g. irregular topographies, alluvial deposits).

ALAIN PECKER

in Pecker is. ~airman and Managing Director ofGe'odynamique

)tructure, Professor at Ecole Nationale des Ponts et ChaussCes and
iting Faculty at the Centre for Post-Graduate Training and
:earch in Earth.quake Engineering and Engineering Seismology
)SE School, !USS Pavia). He is member of the French National
Ldemy of Technologies, Honorary President of the French
ociation for Earthquake Engineering, Past President of the French
iety of Soil Mechanics and Georechnical Engineering, President
he French Committee for Seismic Building Codes. He is also a
nber of several international technical committees dealing with
hquake geotechnical engineering. He has also published over 80
tnical papers in journals and conferences and has been invited as
Lirer or State of the Art speaker in several international events. He
l the editorial boards of three international journals. He has recei-

several awards for his work, among which one from the French
ional Academy of Sciences.
 Contents

Introduction 11

Acknowledgements 13

List of symbols 15

1 Waves and vibrations in soils 19

1.1 Various fields and applications 19
1.2 Vibrations due to construction works 20
1.2.l Pile driving . 20
1.2.2 Dynamic compaction . 24
1.2.3 Vibratory compaction 28
Nessuna parte di questa pubblicazione puO esscre riprodotta o trasmessa in qualsiasi forma o con qualsiasi mezzo
1.3 Vibrations induced by wind turbines 31
elettronico, meccanico o altro senza l'autorizzazione scritta dei proprietari dei diritti e dell'editore.
1.4 Blast induced vibrations 33
No parts of this publication may be copied or transmitte@ in any shape or form, and by any type of electronic,
1.4.1 Vibrations induced in mines 33
tnechanical or different means, without the prior \Vritten·permission of the copyright holder and the publisher.
1.4.2 Vibrations induced in quarries 34
©Copyright 2009 · IUSS Press 1.5 Traffic induced vibrations . 36
1.5.1 Vibrations due to road traffic 36
1.5.2 Vibrations due to railways . 39
1.5.3 Theoretical analyr.;is of rr1oving loads 40
prodotto e distribuito da: 1.6 Vibration isolation ..... 42
produced and distributed by:
1.6.1 Practical problem 42
Fondazione EUCENTRE 1.6.2 Expcrin1ental results 42
Via Fcrrata 1 - 27100 Pavia, Italy 1.6.3 Nurnerical rnodels 44
Tel.: (+39) 0382.516911 -fax: (+39) 0382.529131
Email: info@iusspress.it - web: www.iusspress.it
1.6.4 Values of. the mechanical parameters 48
1. 7 Earthquake engineering and seismology 49
1. 7 .1 Analysir.; at various scales 49
1.7.2 Seisrnic VI.rave propagation in soils 50
ISBN: 978-88-6198-030-3
1.7.3 Values of the rnechanical parameters 51
UNIVERSITY OF W.A. 1.8 Synthesis of the variour.; parameters . . . . 52
LIBRARY
 CONTENTS CONTENTS
4

53 2.10 Application 1: dyna1nic characteti;.-;ation on resonant colu1nn 97

2 ID-wave propagation 97
53 2.10.1 Principles of the test
2.1 Introduction. 98
53 2.10.2 Description of the speci1nen n1otion
2.2 Dynarnic equilibriurn of a bca1n
53 2.10.3 Actual resonant colu1nn test . 99
2.2.1 Kinematics and n1ain assumptions
54 2.10.4 Estimation of damping . 101
2.2.2 Virtual rate of work by internal forces
56 2.10.5 ReRults fron1 resonant column tests 102
2.2.3 Virtual rate of vvork by external force8
57 2.11 Application 2: dynamic characterization under fast loadings 103
2.2.4 Virtual rate of \Vork by quantitieH of acceleration
57 2.11.1 Split Hopkinson Pressure Bar test 103
2.2.5 Equilibriurr1 equation .
59 2.11.2 Experimental device 1114
2.3 Longitudinal vibrations of bearns
59 2.11.3 Stret>s \orave in the specimen 104
2.:).1 Dyr1a1nic equilibriurn .
611 2.11.4 Determination of the rnechanical paran1eterH 1116
2.3.2 Hon1ogcneous equation .
62 2.11.5 3D Split Hopkirnon Pressure Dar test . 108
2.3.3 Solution in terms of stresses
62 2.11.6 3D fast dynan1ic response of sand. llO
2.3.4 Eigernnodes of the bar
64 2.12 Application 3: response of a heterogeneous soil profile 112
2.3.5 Exa1nple 1: pile driving
68 2.12.1 Soil behaviour ll2
2.3.6 r:.xarnplc 2: characteriLJation of a heterogeneous bar
70 2.12.2 \~~ave equation for a heterogeneous profile 113
2.3.7 Absorbing boundaries
71 2.12.3 Boundary conditions . 114
2.4 Torsional vibrations of bca1ns
71 2.12.4 Eigenfrequencies ancl 1node participation factors 118
2.4.1 Dynarnic equilibriurn .
72 2.13 Application 4: t>oil-structure interaction 119
2.4.2 IIornogencous equation.
73 2.13.J Basic principles 119
2.4.3 Stresses in the bea1n
73 2.13.2 EquationH of motion for the soil. 120
2.4.4 Eigenrnodes of the bea1n .
75 2.13.3 Case of a surface excitation 120
2.5 Shear vibrations of bea.rns
75 2.1~).4 Influence of the wave velocity in the soil 122
2.5.1 Bending-shear vibrationR
75 2.14 E·xperiinental estin1ation of <lan1ping . 122
2.5.2 From bending-shear to pure shear
76 2.14.1 Various experilnental rnethods 122
2.6 Behaviour of dissipative n1edia
76 2.14.2 J\!Iethods for the estirnation of attenuation 124
2.6.l DiRsipative phenornena.
76 2.14.3 Definitions of attenuation: synthesis 125
2.6.2 ViRcoelastic behaviour
79 2.14.4 Variations of attenuation 12G
2.6.3 Rheological modeb
84 2.14.5 Characteri2Jation of the varioufi approaches 127
2. 7 V\Tave propagation in viscoelaRtic 1nedia
84 2.14.6 Cornparison of the governing parameters . 127
2. 7.1 Dyna1nic equilihriun1 .
2.7.2 Viscoelastic behaviour 84
3 2D/3D-wave propagation 131
2.7.3 \~~ave equation in viscoelastic media 85
3.1 Introduction . 131
2. 7.4 Cornplex "\vavenurnbcr . 85
86 3.2 Dyna1nic equilibriu1n of a continuous rnc<liu1n 131
2.7.5 Relationship betv.reen o: and Q-J 132
87 3.2.1 Equilibriu1n equation - principle of virtual work .
2.7.6 Dispersion la\vs 133
88 3.2.2 Constitutive eq11ation
2.8 Exarnplet> of propagation in viscoelastic rncdia . 134
88 ~-3.2.3 Equilibriurn equations in terms of <lisplacernents
2.8.1 Exarnple 1 propagation of a triangular signal 134
91 3.2.4 Decornposition of the displacernent field
2.8.2 Example 2 : propagation of a t>eisrnic wave.
91 3.2.5 Uncoupled wave equations . 135
2.9 Other linear and nonlinea.r rnodels 135
91 3.2.6 I3ocly waves
2.9.1 Constant Q (CQ) model 137
92 3.2. 7 \~Tave propagation in anisotropic rnedia
2.9.2 Frequency dependent Q model
92 3.3 V\Tave propagation in unbounded meclia . 138
2.9.3 Nearly Constant Q (NCQ) model . 138
94 3.3.1 Wave equations for plane 1vaves
2.9.4 Equivalent linear viscoelasticity 140
95 3.3.2 Plane monochrornatic waveR .
2.9.5 Frequency dependent rnodels 142
96 3.3.3 Reflection-refraction of plane 1vaves at an interface
2.9.6 Nonlinear viscoelastic models .
 CONTENTS CONTENTS 7
6

152 4.2.6Spectral elc1ncnts . 255

3.3.4 Plane waves in layered rr1cdia: vibration isolation
17'1 4.3 Numerical dispersion 257
3.4 Spherical 1..vaves
174 4.3.1 Physical dispersion and attenuation 257
3.4.1 \Vavc equation
174 4.3.2 Nurncrical errors for wave propagation 258
3.4.2 Solution \vaveficld
175 4.3.3 Theoretical numerical dispersion 260
3.4.3 Geornetrical damping
176 4.3.4 Tirne-stcp estimates for some simple cases 264
3.5 V\laves in a homogeneous or heterogeneous half-space
176 4.3.5 Numerical dispersion for low order elements 264
3.5.1 Surface waves: SH case
177 4.3.6 Influence of the geornetrical arrangernent 265
3.5.2 Su1'facc V\raver:;: P / SV case .
180 4.3.7 Influence of rnass matrix forrnulation . 268
3.5.3Propagation of a plane SH-¥.rave in a surface layer
Amplification of seismic -,,vaves in layered media . 186 4.3.8 Efficiency of higher order ele1nentH 269
3.5.4
188 4.4 Physical and nurnerical damping 275
3.6 Application 1: waves in centrifuged models
188 4.4.1 Rayleigh and Caughey damping . 275
3.6.1 HiHtorical sumrr1ary
188 4.4.2 Rheological interpretation of R,ayleigh darnping 276
3.6.2 Equivalence principle .
189 4.4.3 Attenuation models for geomaterials 277
3.6.3 Calculation of the scaling factors
190 4.4.4 Numerical darnping 281
3.6.4 Dyna1nic experirr1ents in the centrifuge .
192 4.5 Modelling wave propagation in unbounded rnedia 283
3.6.5 Exarnples of dynamic centrifuge experirnents
194 4.5.1 Absorbing boundaries in lD . 283
3.6.6 AnalysiH of the threedimensional \vavefield .
196 4.5.2 Abt->orlJing boundaries in 2D . 284
3.6.7 ::viodelling propagation in dit->Hipati.ve soils
199 4.5.3 Infinite elerr1ents 286
3.6.8 Si1nulations for drop-hall experirr1ents
203 4.5.4 Absorbing layers (PML) 289
3.6.9 Influence of frequency on attenuation factor
204 4.5.5 Coupled approaches 296
3.6.10 N urnerical modelling of centrifuge experirnents
206 4.6 The Boundary Element Method . 298
3.6.11 Re1noving reficctionH by homornorphic filtering
207 4.6.1 Interest of the method in dyna1nics . 298
3.6.12 Analysis of diHpcrsion
208 4.6.2 J'v:Ia.xwell-Betti theorern . 298
3.7 Application 2: Spectral Analysis of Surface \i\Taves and in situ tests
208 4.6.3 fntegral equations in elastoclynamics 298
3.7.1 l)ispersion of Love waves in a single-layered half-space
214 4.6.4 DiscretiLJation and regulari:6ation principle 301
3.7.2 Dispersion of surface waves in a heterogeneous half-space
218 4.6.5 V\Tave propagation in unbounded media 302
3.7.3 Steady State Rayleigh method
218 4.6.6 Nurr1erical Implementation 303
3.7.4 Spectral Analysis of Surface V\Taves: experiments
224 4.6.7 Validation and influence of the regularization 304
3.7.5 Seisrnic refraction .
226 4.6.8 Advanced formulation: the Fast JVIultipole Method 306
3.7.6 In-hole tests .
235 4.6.9 Elastodynarnics in time dorr1ain . 308
3.7.7 1v1icrotremor inethods
236 4.7 Applications to wave propagation in soil 308
3.7.8 Conclusions on field tests
4.7.1 Diffraction of a plane vvave in unbounded media 308
4 Modelling wave propagation 239 4.7.2 Vibrations of a foundation . 312
4.1 Nurnerica.l rnethods for wave propagation 239 4.7.3 Vibration isolation using piles or trenches 313
4.1.l Jlv:Iodelling V.'ave propagation . 239 4.7.4 Traffic induced vibrations in railway tunnels . 321
4.1.2 Nurncrical J\1odelling in Elastodynamics 240
240 5 Seismic wave propagation and amplification 323
4.1.3 Time dornain vs frequency dornain
241 5.1 Introduction . 323
4.1.4 Actual or synthetic signals .
246 5.2 Seis1nic -,,o,rave amplification . 324
4.2 The Finite Element Method
246 5.2.1 l\l:Iain governing phenomena 324
4.2.1 Strong for1nulation
247 5.2.2 Experimental characteriLJation . 327
4.2.2 \!\Teak for1nulation
248 5.3 Seismic wave an1plification in layered media 328
4.2.3 Approxin1ate minirniza.tion: Galerkin method
248 5.3.1 Frorn transfer function to time-domain response 328
4.2.4 Finite clements
253 5.3.2 Amplification in single-layered media . 330
4.2.5 'Ti1ne integration algorithms .
 CONTENTS CONTENTS 9
8

333 5.9.7 Conclusions on site effects in Volvi 408

5.3.3 Arnplificatiou in rnulti-layerc<l n1edia
33:J 5.10 \Vave-structure interaction . 410
5.1± A1nplification due to the topography ..
5.4.1 1\'lain phenomena and Hin1plified analysis . 333 5.10.1 Frorn soil-structure to \Vave-structnre interaction 410
5.4.2 Amplification by crests and hills :J36 5.10.2 Seis1nic analysis for buildings . 410
5.4.:J Amplification by ca.nyono 340 5.10.3 Seismic interaction 10.1ith underground structures 412
5.4.4 Amplification on actual topographies . 345 5.10.4 Seisrnic interactions at the local and global scales . 413
5 .5 Amplification of Reir:nnic vvave8 in 2D alluvial basins . 351
5.5.1 Amplification by wedges 351
5.5.2 Tl1coretical basinH 356 Appendices 419
5.5.:i Cylindrical basins 361
5.5.4 Cylindrical basin vs horizontally layered soil . :J62 A Several operators in mechanics 419
5.5.5 Elliptical basins V\rith variable shape ratio 365 A.l Vectors and product of vectors 419
5.5.6 21) /lD aggravation factor 366 419
A.2 Tensors and product .
5.6 Arnpliflcation of Heisrnic 1..vaves in 31) alluvial basins. 367
A.:-3 Gradient and Laplacian
,ng
5.6.1 Semi-spherical basin 367
A.3.1 Definitions 419
5.6.2 Sine-shaped basin 369
A.3.2 Examples 420
5.6.~i Semi-spherical basin and obliqne incidences 370
5.6.4 Semi-ellipsoidal basin 372
B Synthetic wavelets 421
5.6.5 Moon-valley model 372
374 B.l Ricker wavelet 421
5. 7 Modal approaches to analyze site effects
374 B.2 Gabor vvavelet 424
5.7.1 A1nplification of the seir.;rnic inotion and resonance
5.7.2 Various types of inodal approaches 374 B.3 1\Javroeidis & Papageorgiou vvavelet 425
5.7.3 Simplified modal method 375 D.4 Generalized R.ayleigh \Va:velet 426
5.7.4 FeatureH of the various ino<lal methods . 376 B.5 l{iipper \vavelet 427
5.7.5 F11nda1nental frequency of a. geological structure :J76 B.6 Ormsby wavelet . 428
5.7.6 l\1odal cstirr1ation of the fundarncntal frequency 378 429
B.7 Morlet wavelet
5.7.7 Si1nplified rnodal n1ethod vs experi1nental spectral ratios. 379 430
B.8 Meyer wavelet .
5.8 Arnplification in shallow basins (e.g. ]\Tice) :JSO 432
B. 9 Double-M wavelet
5.8.1 Analysis of site effects in Nice (France) 380
5.8.2 Amplification from lD transfer functions 382 433
:J83 C Spectral analysis and filtering
5.8.3 2D model of the geological profile . 433
384 C.1 Fourier transforrn .
5.8.4 Amplification of a plane SH-wave
386 C .1.1 Definitions 433
5.8.5 Influence of attenuation
387 C.1.2 Main properties. 434
5.8.6 lD and 2D a1nplification vs cxperirncntal results
5.8.7 2D/1D aggravation factor . . . 388 C.1.3 Usual transforrns 434
5.8.8 Comparison \vith the sirnpli:fied rnodal rncthod 389 C.l.!1 Fourier transforn1s of synthetic \Vavelets 435
5.8.9 Ti1n.e-dornain sirnulations 390 C.1.5 VVave propagation in viscoelastic rncdia 435
5.8.10 Cornparisons ¥.1ith a deep alluvial basin 390 C.1.6 Fourier transforrns of actual signals . 438
5.9 Arnplifica\ion in a deep basin (Volvi) 393 C.2 Filtering 439
5.9.1 The Volvi EuroSeisTcst 393 439
C.2.1 Classical filters
5.9.2 Simplified and complete rnode!H of the Volvi basin 396 4:l9
C.2.2 Filtered signals: examples
5.9.3 SI-I wave amplification in the Volvi basin . 397 440
C.:-J I-Iilbert transform and envelope curve .
5.9.4 Cornparisons betvveen simplified and complete n1odels MJO
C.3.1 Definition 440
5.9.5 SV-vvave arnplification in the Volvi basin 406
406 C.3.2 Envelope curves 441
5.9.6 2D/1D aggravation factor .
1() CONTENTS

D Propagating waves: duration, velocity, echoes 443

D.1 Frorn acceleration to displacement 443
D.1.1 Integration of accelerograrns. 443
D.1.2 Baseline correction 443
D. l.3 Spectral methods 443
D.2 PGA, PGV, PGD 444
D.3 Signal duration
D.4 Estirnation of -i,vavc velocity
445
447 Introduction
D .4.1 Peak to peak estirr1ation 447
D.4.2 Cross-correlation function 448
D.5 Detection of reflected waves and echoes 449
D.5.1 Autocorrelation function 449 Various types of rnechanica.l vva:vcs inay propagate through soils: traffic induced vibra-
D. 5. 2 R.eal cepstrum 449 tions, seiHrnic -,,vaves, blast or impact induced loadings, etc. The frequency ranges 1 the
D.5.3 Tirnc phase 449 wavclengtlu~, the amplitudes are different from one application to another. However 1 the
funtlan1ental theory of mechanical waves in heterogeneous elastic or attenuating media
E Echo removal by homomorphic filtering 451 may he considered to investigate such various fields of interest. ·The related experimen-
E.l Basic principles ..... . 451 tal and nu1nerical techniques may also apply to a wide range of dyna1nic prohlerns in soils.
E.2 Co1nputation of the complex cepstrurn 451
E.3 Removing echoes thanks to cepstra.l filtering 453 The main scientific and engineering goal of this book is thus to deal simultaneously
E.3.1 Principle of the method 453 with soil dyna1nics/vibrations and vv~avc propagation in soils (including seismic waves).
E.3.2 Horr1ornorphic filtering: sunrmary . 453 These various fields are generally considered separately and the in1portant links bet\veen
E.3.3 Application to actual rT1easurements 454 thern, from both scientific as vvell as practical points of views, are unfortunately not
investigated. They arc usually considered in separate disciplines such as earthquake
References 459
gcotechnical engineering, civil engineering, mechanics) geophysics, seisrnology, cornputa-
493 tional engineering 1 etc.
Index
The objective of the book is to offer in a single publication an overvie"\v of soil dynamics
a.nd Virave propagation in soils Virith emphasis on engineering application8. The structuTe
of the book is the following:

• It starts (Chapter 1) \.Vith the presentation of a wide variety of practical proble1ns:

traffic induced vibrations, 1,vind turbines, dynamic cornpaction 1 earthquake engi-
neering, etc. Pictures fro1n practical applications are sho"\vn together with signal
recordings fro1n the field and numerical simulations frorr1 various methods.

• The lD-1,vavc theory is detailed in Chapter 2 for various types of \vavcs (axial 1
torsion, shear). It is then applied to the dynamic characterization of soil from lD
laboratory tests (resonant colurnn, Hnpkin8on pressuTe bars), the dynamic response
of a heterogeneous soil profile as vvel.l as dynamic soil-structure interaction.

• Chapter 3 deals with the analysis of 2D /3D "\vave propagation in layered media
(body Viraves as well as surface '°''aves). Sorne simple equations are derived for
applications in the field of vibration isolation as "\vell as surface ¥lave propagation.
Using the 2D /3D theory, the tlyna1nic characterization of soils is considered. ¥lave
propagation in reduced-scale rnodels is first discussed. 'l'he dispersion of surface
12 Introduction

"\vaves in layered soils is then investigated and field SAS\\r teRts 1 arc presented.
()thcr types of in situ tests (e.g. in-hole testR) arc also investigated.

• l\tfodelling \Vave propagation in soils n1ay be perforrncd using various typeH of nu-
n1erical methods. Chapter 4 gives an overview of various nurnerical approaches and
investigates tvvo of thcrn: the Finite E1e1nent l\ifethod and the Boundary J<~lement
fvlcthod. Such important ist>ucs as nurncrical errors and radiation conditions are
discussed. Several numerical examples arc then proposed (e.g. vibration isolation). Acknowledgements
• Finally, Chapter 5 investigateH Heisrnic "-'ave propagation and arnplification in corn-
plex geological structures such as irregular topographies or alluvial d12posits. Some
classical theoretical results are recalled first. Several exarr1ples are also presented: ·The authors vvould like to thank first Pr lVIichael Pendcr 1 frorn the University of Auckland,
J\1cxico, Los Angeles, Carflcas, Nice, Volvi 1 etc. Such pararnetcrs as the topographic Ne'N-Zealancl, for his careful and detailed revic\v of this book. Elis valuable suggestions
aggravation factor (for irregular topographies) or the 2D/1D aggravation factor (for
i1nproved significantly our original rnanuscript.
alluvial basins) are discussed.

In addition to these five chapters, five appendices deal \Vith various types of synthetic \\Te \Vould like to acknowledge the help of several colleagues \vith v,rhorr1 vve shared
wavelets, spectral analysis, signal integration 1 signal duration or echo rcn10\ral. cornrnon researches or teaching activities throughout the yea.rs in various fields of dy-
namics. 1\.1any foreign colleagues frorn various places in the vvorld v,rere also supporting
our project providing fruitful ideas, discussions, figures, papers.

()ur deepest thanks, also, to our fa1niliel:l and relatives for their continuouR patience.
V.,le \Vere often thinking and ta.lking about waves in soils during the last n1onths and they
were alvvays supporting us to\vards our goal.

1 Spcctral Analysis of Surface Waves

List of symbols

Notation conventions
a scalar variable
a time derivative of scalar variable a
a* I~'ourier transform of variable a
an real part of complex variable a
a1 imaginary part of cornplcx variable a
lal modulus of complex variable a
a, normalized scalar variable
g._ vector (analytical expressions)
ij, virtual (vectorial) field
a.i jth component of vector g
aj,k spatial derivative of the jth component of vector g._
with respect to the kth spatial coordinate
{a} vector (discretized expressions)
A 2nd order tensor
[A.] matrix (discretized expressions)

Symbols
Latin symbols
g_ acceleration vector
A.; amplitude of incident j-wave (.i = P, SV, SH)
c wave velocity
C elasticity tensor

C complex cepstrum (homomorphic filtering)

4 strain rate tensor
d. polarization vector
 List of syrnbols 17
16 List of syrr1bols

velocity vector
E Young's inodulus
>.vavc velocity of P-wavcs
f frequency
wave velocity of Rayleigh 1..vaves
t body force vector
>.vave velocity.of S-v,raves
:F F·ourier transfor1n
position vector
)-{ Hilbert transforn1
x, y, z spatial coordinates
square root of -1
1vavc vector Greek symbols
norrr1a]i7'ed Vi'ave vector
er attenuation factor
'mu unrelaxed inudulus
-y shear strain (in Chapters 2 and 5)
·m= relaxed n1odulus
2nd para1neter of the Nev,r1nark scherne (in Chapter 4)
J.\1 co1nplcx modulus
0 logarithn1ic decre1nent
111 rnoment
!:::.h finite clen1ent size
1"'vit torsional rnon1ent
!:!. t time step
lVI& bending rno1nent
f;_ strain tensor
Tl norrna1 vector
Tl viscosity
f\T nor1nal force
Pi, Pe rate of ¥.rork by internal (resp. external) forces
e angle
n, aspect ratio
Q quality fa.ctor
A 1st Lame ehu>ticit.y constant
Q- 1 attenuation
A wavelength
R.i amplitude of reflected J-wavc (j ~ P, 8V, SH)
p shear modulus or 2nd Lame elasticity constant
S cross-section area
v Poisson's ratio
t tiJne
~ da1nping ratio
tb llickcr \.vavclet breadth
p rnass density
tp Ricker wavelet fundan1ental period
Q: (Cauchy) stress tensor
ts Ricker vvavelct delay
2=7j fundarnental solution (stress tensor)
T tretnsfer function as the ratio betvveen the surface rr1otion and the
interface inotion
strain relaxation ti1ne
transfer function as the ratio bet\.veen t.he surface inotion and the Ta stresH rela.xation tilne
outcrop 1notion ¢ creep hmction (in Chapter 2)
~ri.j transfer function bet1,vcen layers i an<l j scalar potential for ?-waves (in Chapter 3)

1~; 11111plitude of trnnsmittecl/refracted j-wavc (j = P, SV, SH) X velocity ratio

T, transn:1itision coefficient for the ,]-wave (j = P, SV, S.H) 1/J relaxation function (in Chapter 2)
Tik funda1nental solution (traction) :£.. vectorial potential for S-waves (in Chapter 3)
v shear force w circular frequency
displace1nent vector O. clornain
funcla1ncntal solution (clisplacen1ent) ori boundary of domain ~2

11 (particle) velocity
Chapter 1

Waves and vibrations in soils

Actual problems

1.1 Various fields and applications

The analysis of soil response under dynamic loadings iti of great interest for various
applications in the field of Civil and Environn1cntal engineering. In rnany practical
situations, dynamic or vibratory excitations rrc
very ilnportant since:

• they allovv some improvement of the in~hanical properties of the soil (e.g. vibratory
or dynamic cornpaction): the aim is t,,6~ptimize the irnprovement process of the soil
mechanical features,

• they rnay originate so1ne annoyances to the public (e.g. traffic induced vibrations):
the goal is to reduce the vibration level through isolation techniques,

• they can lead to some int:itabilities in the considered soil m<:-u~s or to the da1naging
of a structure resting on it (for instance in case of seismic loading): the objective
is then to limit their consequences.

This chapter presents and discusses several irnportant domains of application related
to soil dynam,ics 1 an expression designating a wide range of practical situations since the
main mechanical para1neters are very different from one case to another (stress and strain
level, frequency, strain rate ... ). As depicted in Fig.1.1, the different practical problems
are related to various shear strain ranges:

• in the case of rnachine vibrations or traffic ind11,ced vibrations, the shear strain
ranges betvveen 10- 5 and 10- 5 and the soil behaviour remains in the linear l'ange,

• for strong earthquake motions, the shear strain n1ay be larger than 10- 4 thus leading
to nonlinear effects in the soil response,

• for very strong loadings such as explosions or impacts 1 the shear strain may be as
large as 10- 3 or 10- 2 which is beyond the scope of this book.
 l

\¥aves and vibrations in soils 1.2 Vibrations due to construction works 21

20

shear strain y static approach to deterrnine, for instance, the bearing capacity of the pile or pile group
and a dynarnic approach to analyze the transient and resulting stress field in the pile
10~ 1ff' and the soil around it. After construction, pile foundations 1n.ay also undergo dynamic
I I I
excitation in the case of: structures in seisn1ic areas, ta.11 structures subjected to \Vind
machine/traffic strong seismic explosions loads, offshore structuTes, etc.
vibrations motion
------------------------------------------------- Basic principles of the driving process
resonant column
Dynamic pile driving. As depicted in Fig.1.2 (left)) dyuarnic pile driving conRist in
triaxial tests 11
dr·i'uing 11 a pile with a. falling inass of knovvu kinetic energy. The soil resistance dur-
------------------
ing the driving proceRs can be estin1ated through various parameters characteriiing the
simple shear test: number of blo\vs, penetration per blovir, kinetic energy per blow (mass and falling
height); penetration resistance, etc. The penetration resistance of a sand increases for
hollow cylinder test larger relative densities or lateral pressure) but decreases for larger water contents. T\vo
types of resistance are generally considered: the tip resistance and the total resistance.
shaking table
I I I The bearing capacity of piles changes in variou.s \vays for coheRive and cohesionleRR
.soils. For sandR, tvvo n1ain phenomena have to be conRidered: on the one hand the den-
Fig. 1.1: Ranges of various problems in terms of shear strain (top) and related sity modificationH around the pile because of con1paction phenon1ena 1 on the other hand
experimental methods (bottom). larger in depth tip resistance and lateral friction. From a dynarnic point of vie\v, the
in1pact of the han1n1er on the pile head generates transient phenomena in the pile and at
the soil-pile interface. A simple exarnple of such phenornenon is proposed in the following
sections.
The experimental rncthods available in the laboratory for the dyna1nic characteriza-
tion of soils in such strain ranges are also detailed in Fig.1.1. They allovv the identification
The detailed analysiH of the soil-pile interaction (axial or radial loading) \\ras proposed
of the soil dynamic/cyclic behaviour. Nevertheless, they generally involve hornogeneous
by many authors (Bourdin ct al., 1989; Huang ct al., 1992; el Naggar and Novak, 1994;
strain states in a Roil specin1en whereas actual Viraves propagating in real soils lead to J'viasoumi et al., 2007). They analyzed the driving process and deterrnined the static
cornplex ~iD stress-strain paths varying -,,vith tirne. Field cxperirnentH allu\v the detailed (bearing capacity) as ¥.rell aH the dynamic (stiffnesH and da1nping) para1neters. During
analysis of such propagation process but do not lead to the full identification of the the driving proceHs, the dynamic loadings correspond to fa.st shear and pressure forces.
soil behaviour. Laboratory experirnents 'Nill be discussed in Chapter 2 and several field
The pressure dynamic forces are mainly Higuificant around the pile tip. The lateral
experirnents 'Nill also be presented in Chapter 3. friction induces shear stresses in the soil. El Naggar and Novak (1994) proposed to
define an annular area around the pile with v,reaker soil mechanical properties than in
the neighbouring soil (lo\ver shear inodulus). FUrtherrnore 1 they take into account a
1.2 Vibrations due to construction works dependency on the strain rate. Dynarnic soil response for large strain rates vvas also
studied in det'1ils by Scmblat et al. (1999) using Split Hopkinson Pressure Bar systems.
1.2.1 Pile driving This experimental approach \Vill be presented in Chapter 2.
Interest and related problems
lD wave propagation in a pile. To estimate the stresses induced in the pile 1 it is
Piles are involved in many foundation syRterns for large structures. Two different t:ypes
necessary to investigate \Vave propagation phenorucna in the pile. As depicted in Fig.1.:3,
of problems then arise: on the one hand it is necessary to estiinatc the bearing capacity of
the dynarnic loading inay be decoruposed into several transient parts: the falling 1naR8
the pile(R), on the other hand; one has to cleterrnine both the stresses induced in the pile
first generate.s in the pile a step load involving both a loading wave and au 'unloading
and the dynarnic excitation of the neighbouring soil during the driving process. The first
'Wave. AH this exarnplc neglects the lateral friction (beginning of the pile driving), this
problerr1 corresponds to the design of geotechnical structures \o.1hereas the second deals
step load propagates several tin1es along the pile: each time the \vave reaches the pile
with the question of vibratory nuisances.
tip the strain energy is partly transrnitted to the soil and partly reflected into the pile.
For each hlcnv at the pile head, a cornplex \Vavefield is thus generated leading to a pile
These t\vo types of problems generally raif;e the need for t\vo different approaches: a
 \Vaves and vibrations in soils
22
Fig. 1.2: Dynamic pile driving (left) and vibratory pile driving (right).
penetration in several (dynarr1ic) steps (see Fig.1.3). The penetration process is thus not
governed by an instantaneous pheno1nenon. In actual cases) the lateral friction rapidly
plays an important role since a significant part of the energy is dissipated or radiated
along the soil-pile interface (Niasoumi et al. 1 2007). However, the fast dynamic processes
discusRe<l previously are still significant in all cases.
The pile driving process is then controlled by propagation phenomena in the pile itself
as vrcll as n1ultiple floil-pile interactionfl. Indeed each penetration step depends on the
rnechanical characteristics of the soil, of the pile and of the soil-pile interface. J'viany
theoretical resultR vvill be given in the next chapter to analyse lD vvave propagation in
piles.
Vibratory pile driving. In the case of vibratory pi.le driving, the excitation at the
pile head is given through a vibratory load (Fig.1.2, right). The main difference with the
dynarr1ic pile driving technique is that the energy of the vibratory pile driving: equipment
1.2 Vibrations due to construction Vi.rorks 23
time
Fig. 1.3: Propagation of loading and unloading waves in a pile during the
driving phase, from (Meunier, 1979).
is trans1nitted progressively to the pile as vvell as to the neighbouring soil. There arc no
more alternate dynamic and Htatic steps in the process but a vibratory/ cyclic continuous
excitation at the pile head. The problern of pile hearing capacity and vibratory nuisances
are cornbined: the balance bct¥reen the dissipated energy and the radiated energy is the
key issue for such a problcn1.
As sho-\Vn in Fig.1.4) for vibratory pile driving experin1ents perforrne<l by Rocher-
Lacostc et al. (2007), the analysis of the experimental results along the pile and in
the neighbouring soil is very complicated. The a1nplitude n1casured at the free-surface
depends on the penetration depth. l)ynamic phcno1nena take place in the pile, along the
soil-pile interface and the soil layers around the pile (-\vaves propagating in the soil up to
the free surface). Vcu"ious types of theoretical and experirncntal approaches are propoRed
in (Transvib, 2006).
Values of the mechanical parruneters
For the theoretical and numerical analysis of pile driving 1 the following range rr1ay be
considered for the rnechanical parameters: l'.1~
• particle velocity (soil): several mm/s,
~
I
 25 I,
VVaveH and vibrations in soils 1.2 Vibrations due to construction v.rorks
24

10'
n
D5m
.
D 10m1
10' 8 20m ;;;

••
L

• 30m
s__§ 1o" ;' -----
.... ....
(l) .
impact t- - - - - - - . . . ~1 H
""';:J
+-' ve1ocity
I
'------
/
·~
~
10·' ro::rr ;;;
s'"ro v, =v2gH •

~~4:-
10·'
~ \ \~) J J
·,, ** 11\,.
\ '•¥..... ... ___ .........
/••'
///

10·' *
~~~~~~~~~~~~~~~~~~~~~~~---"~~~~~---" ,., .... _._ ,,•
~~

.....
.,..~ I

0 10 20
frequency (Hz)
30 40 50
'------;
soil to densify

Fig. 1.4: Vibratory pile driving: amplitude at the free-surface for various
penetration depths (Rocher-Lacoste et al., 2007). Fig. 1.5: Dynamic compaction of soils: picture of the experimental arrange-
ment (left) and principle of the process (right).

• force at the pile head: from 10 to 10,000 kN,

• nature of the soil, perrr1cability,
• driving energy: 10 to :JOOO J (per blow),
• mass of the impactor m,,
• displacement: from several min to several drn (per blov.r).
• impact velocity Vi 1

1.2.2 Dynamic compaction • number of shock steps (delay between each step).

Principle of the process Experimental results

The main goal of this teclmiquc, mainly developed by L. Menard (Menard, 1974), is to
in1prove 1 dovvn to a 4 to 6 n1 depth, the density of the soil. To <lo so 1 as depicted in
Fig.1.5 1 a rnass of several tens of tons is dropped on the soil to con1pact it. The falling
height is generally very large (10 to 40 rn). To in1prove the soil properties in a given
area, it is necessary to define different lines along 1.vhich the dynamic compaction "\Vill
be pcrforrned at regular spa.cings. The improverncnt of the mechanical properties of the
compacted soil rnay be very important since the soil modulus can 1 for instance, be rnul-
tiplied by a factor 2 or 3 (Menard, 1974).
So1ne experimental rer.;ults in terrns of particle velocities are displayed in Fig.1.6 (JVI6nard-
Soltraiten1ent, 2002). These signals correspond to a falling height H=23 111) a falling
rnass m,=16 t) that is an energy E=3)610 1 080 J. 1Wo sensors \Vere located at the soil
surface at the follov;,ring distances: d1 =97 m for sensor 1 and d2=197 m for sensor 2
(1I6nard-Soltraiterncnt; 2002). 'fhe duration of the signal is larger at sensor 2 than at
sen8or 1 and the maxirnu1n an1plitudes arc not reached for the same co1nponent (J~vI6nard­
Soltraitement, 2002). The \.Vavefield generated by the impact is very cornplex and it is
difficult to assess the efficiency of the process only considering the radiated waveD.

T'l1e characteristic pararneters of dynamic compaction are the follovring: In dry soils, the irr1pact generates transient elastic waves as \.VCll aR plastic loading
and unloading waves. ·The forrncr are rnainly responsible for the dynamic compaction.
• depth of soil to dcnsify,
 VVaves and vibrationH in soils 1.2 \Tibrations due to construction \.VorkH 27
26

1992; Scott and Pearce, 1975; Sieffert et al., 1990). Nowacki and Raniccki (1987) pro-
4.00 sensor 81
'if; posed a theoretical study on dyna1nic compaction by spherical explosive sources. Hat.a
8 2.00 and rrateyan1a (1991) inveHtiga.ted plastic Virave propagation due to the irr1pact. Several
.§ researches on rnetals have. some sirnilarities with dynamic compaction in soilf; (Brothers
0.00 et al., 1977; Edlingcr et al., 1993; Hardy et al., 1971; Sinclair et al., 1985; Studman et
"'u""'"'
-~
al., 1977).
~
D -2.00
'"
> -4.00 (vertical)
Experimental point of view: dynamic stiffness
During dynamic con1paction tests, Foran et al. (1992) measured several mechanical
0.30

-ss
~

~
w
0.13
sensor 82
parameters and characterized the dynamic response of the soil under impact. The ac-
celeration (or deceleration) of the falling 1nass, the pressure at the impact location, the
settlernent of the inass alloVir the calculation of the Dynarnic Settlernent Modulus Md·
-0.05 This sett.lenient rr1o<lulus involves the global dynarr1ic interaction between the soil and
"'u""'"'
-~
the falling mass.
~
D -0.22
'" -0.40
> (vertical)
The displacement of the inass dt as a function of tirne is estirnated from acceleration
18.0 18.5 19.0 19.5 20.0 20.5 rneasurcs (initial conditions). The stress at the irnpact location 1 denoted Pt, is derived
17.5
time (s) from the following expression:
rn.at
Pt = ----::;-[ (1.1)
Fig. 1.6: Velocities measured in the soil during a dynamic compaction exper- where m and A are the rnass and the surface area of the falling Vi'eight and at its accel-
iment (Menard-Soltraiternent, 2002).
eration.

Foran et al. (1992) determined the dynamic settlement modulus of the soil M,, from
In saturated soils) pressure waves propagates in the fluid phase. The pore pressure rnay the experirncntal curves using the folloViring cxpresRion:
increase until the soil skeleton collapses. A grain rearrangement in a denser configuration D.p,
is then possible. (1.2)
Ma= D. ('1J)
Several err1pirica] formulae have been proposed to rnakc the link between the soil where D rcpresentH the diameter of the falling Vireight.
thickness for densi:fication and the in1pact energy. Some authors also tried to esti1nate
the optin1aI distance betvveen tv1.ro in1pact pointr.; (1viayne and Jones, 1983; Nienard, 1974; Poran et al. (1992) proposed a co1nparison between the dynarnic settlerr1ent modulus
Gambin, 1985). and the elastic properties of the soil. Furthermore, they related the rneasurements of the
density of the compacted soil to the impact energy for several tests.
Theoretical analysis of the impact
Values of the mechanical parameters
L. Menard was one of the first to develop dynamic compaction (Menard, 1974). He
proposed a con1parison betVi'een Htatic and dynamic co1npaction. The fundamental as- The main parameters of dyna.rnic compaction tests arc the folloViring (also sec the cornplete
surnptions of these two types of compaction arc different on inany aspects: friction, Table 1.1):
stiffness and drainage. All these phenomena are strongly influenced by the con1paction
• particle velocity: frorn 10 to 30 n1s- 1,
velocity. The dynarnic soil reHponsc inay be different depending on the strain rate. Such
results \Vere demonHtratcd on fast <lyna1nic testR performed in the lab by Sernblat et al. • acceleration of the masH: frorn 50 to 500 n1s- 2 ,
(1999).
• impact energy: several J\!LJ 1
:tviany authors perforrned theoretical analyses of the soil-1nasR interaction during the • stre8ses: fro1n 10 to 500 kPa.
dynamic con1paction process (Chou et al., 1991; Chovv et al., 1990 1 1992; Foran et al. 1
 VVaves and vibrations in soils 1.2 Vibrations due to construction 1.vorks 29
28

100
1.2.3 Vibratory compaction ~

~
w
8 50
Prin.ciple and example
s 0
The vibratory con1paction technique is v,ridely used in civil engineering. As depicted in
Fig.1. 7, it leads to soil densification thanks to a heavy rolling and vibrating cylinder. "'
+'
·~

"
0
~ -50
The 1nain parameters of the soil-cylinder interaction are the frcquency1 the force at the
surface, the speed of the cylinder axis and the nurr1bcr of passes (Pictshch and Poppy,
'"
>
-100
1992; Yoo and Selig, 1979). The vibrati.onH of the cylinder on the soil surface induce a
cyclic loading 1,vhich is supposed to increase the density of the mediurri. The 1nain goal
~
100
of the works dedicated to this topic is the optimization of the cornpaction pararneters to '"s°
~

get the larger densification of the soil.

~
s 0

For such experi1nents, as shoV>.rn in Fig.l.8 1 the nurnber of loading cycleti is very large "'"
+'
·~

0
-100
Hir1ce steady-state cyclic loadings are applied to the soil. Hovvever, since the cylinder ~

111oves along the soil surface at various speeds, the source o:f vibrations changes \vith '"> -200
0 10 20
ti1ne. The particle velocities displayctl in Fig.1.8 arc thus very complex signals also de-
time
pending on the propagation proces8 through the soil.
Fig. 1.8: Horizontal (top) and vertical (bottom) particle velocities measured
So111e experirnental examples are given in Fig.1.9 for three different types of vibratory during vibratory compaction (from CER-Rouen, France).
rollers and two distances from the road. In Fig.1.9, only the 1nax.imum particle velocity
is thus considered. It is very large close to the track (up to 40 rrnn/s) and it decreases
\Vith the distance (around 6 mrn/s at a larger distance). The 111axi1nurr1 particle velocity
generally increa.seH vvith the nurnber of loadings: it is due to the fact that the soil being
progressively densified, the radiated wavcfield is rr1ore easily propagated through the soil. Main phenomena
Ho¥lever, the influence of the type of vibratory roller is very strong since there is a large The way the soil is co1npacted thus depends on the applied vibratory loading. For in-
discrepancy in the rneasured particle velocities. stance, Pietzsch and Poppy (1992) consider that the vibrating mass may be uplifted
above the soil vi/herea8 Yoo and Selig (1979) assurne a perrnanent contact between the
soil and the vibrating cylinder.

Further1nore 1 one has to contiider tvvo different loading cases:

• stationary case: the sanrc soil volurne undergoes the vertical harmonic force,

I I I • non stationary case: the cylindric rnass moves and the vibratory excitation it> ap-
geophones plied at various points.

In the latter case, the behaviour pararncters of the soil cannot be deter1nined \vith clas-
sical testti because the loading is non stationary.

Fig. 1.7: Principle of vibratory compaction and type of measurements per-
formed (from CER-Rouen, France).
The displacc1nent of the \vhole compacting syste1n changes vvith frequency. 1'he vi-
bratory feature.s of the systerrr strongly influence the stress level at the soil tiurfacc. The
displacc1nent-frequency curve has two peakti: the first one is mainly influenced by the
vibratory features of the ¥,rhole systern ¥,rhereas the forrrrer actually depends on the vi-
bratory retiponse of the cylinder. All these remarks allo\v the previous authors to relate
the motion of the vibrating rnass and the compaction level in the soil.
:m \'Tave::; and vib rations in ::;oils 1.3 Vibrations induced by wind tnrbiue::; 31

1.3 Vibrations induced by wind turbines

40
,-._
r:n
ss Wind turbines are very large structures undergoing strong wind loading::;. When the
turbine is turning, it may induce acoustic waves (noise) as well as vibrations iu the soil.
'-' 30 The latter depend on the type of fouwlation and t he properties of the soil.
>.
...., type 2 __ _

-
Tl
0
Q) --- ---- --- --------- As depicted in Fig.] .10, experiments to assess t he seismic wavefield radiated by wind
> 20
-Q)

.....
( .)
....,
lo-<
ell 10
type 3
-------·-----·-------- ... ....................... .
farms wer e for instance performed by Schofield (2004). He measured the grou11d mo-
tion at d ifferent di:-;tances from a wind far m located nearby the Laser Interferometric
Gravitational wave Observatory (LICO) located in Hanford, Washington , USA .. Since
0. .· the observatory is very sensitive to gronnd vibrat ions, a detailed :-;tudy was performed.
.··
0
6

,-._
5 ,. ___ type 2
ss
(/J
/

.._,
/
/
/
- ------- --- --- - --
4 /

>.
....,
..... - - ---- ... "'
-(.)
0

>
Q)
3

- Q)

.....
( .)
....,
id
2
type 1

.... ----
0. ··· · type 3
-..... .... _______ ____ _....

00 2 4 6 8 10 12 14
number of passes

Fig. 1.9: Maximum vertical particle velocity measured at two distances,

d 1 =1.35 m (top) and d2=8 m (bottom), for different types of vibratory rollers
(from CER-Rouen, France) .

Values of the mechanical parameters

The main parameters for problems in the field of vibratory compaction are the following:
• frequency: 10 to 80 IIz,
• force: 50 to 200 k;'J, Fig. 1.10: Picture of a wind turbine and a seismometer used for the assess-
ment of turbine induced vibrations in the soil nearby the Laser Interfero-
• speed of t he cylinder axis: several m::;- 1,
metric Gravitational wave Observatory (LIGO) in Hanford, USA (Schofield ,
2004).
• vertical displacement: several mm.
 V\Taves and vibrations in soils 1.4 Blast induced vibrations 33
32

Andersen and Clausen (2008) proposed a numerical model to analyze the impedance
1.4 Blast induced vibrations
of a wind turbine foundation. They showed t he strong influence of soil layering on the
impedance of a r igid hexagonal foundaLion footing for a wind turbine. As discussed by 1.4.1 Vibrations induced in mines
Andersen and Clausen (2008), for low freqnencics around 0.25 Hz, corresponding t o the
first resonance frequency of a 3 MW wind turbine, changes in the impedance are stronger Mines and quarr ies are important sources of waves and vibrations during the exploi tation
for deep soft layers overlying a st iff hal f-space. process (use of explosive sources) or even after it has been exploited (failure of mines roof
generating a, seismic wave in the soil). As depicted in Fig.l.12 , waves may be generated
The amplitude of the ground rnot,ion measured by Schofield (2004) is displayed versus into the geological layers due to the exploitation process (Dowding, 1985).
frequen cy in F ig.l.11 for distances ranging from 24 m to 18 km from the wind farm.
Schofield (2004) also proposed some simple a Ltenuation relationships to e:;timate the For explosive sources located at the free-snrface (q11arries, construction works, etc),
ground motion for future wind forms projects. As shown in Fig.1.11, the motion ampli-. pressure waves ure a lso generated into the air. Furt hermore, the various vvaveficlcls prop-
tude reaches its maximum at 4.3 Hz wlrich is not close to the fundarnenta.J frequency of agating into the soil as well a.'l into the air may interact leading Lo so-called ground-borne
the seismic isolation system of the obser vatory. The moLion amplitudes range between noise. Gro11nd vibrations and ground-borne noise may t hus be a source of annoyances in
4.10-to to 2.10- 7 at, the various sites. Depending on the wind speed, the ground motion the environment at the surface.
induced by the wind turbines may influence the background noise level at the observa-
tory (Schofield, 2004). He also pointed out Lhe potential influence of t he l_ocal ~eologi_c~l
structure (e.g. bas alt ridge). For sites located close to a wind farm, the vibrations ong1-
nated by the wind turbines into t he soil may imply some annoyances to Lhe b uildings or

~·~ i,:~';;ced
the public. . .··cz::.---
.·· ... -- quarry
''_)\.--
____m_1_n_e_ _ _
: .: ; ; ; waves
~-- waves
:. :. ~ :. : .
induced

Fig. 1.12: Mine and quarry induced waves propagating into the soil: descrip-
tion of the problem.

Several experimental results are displayed in Fig. 1.13. They correspond to the spec-
t ral ratio:; between the horizonta l and vert ical components of motion measured by Driad-
Lebeau et al. (2009) above the Gardanne coal mine located in Provence, France. T his

r
mine is nearly 1 km deep and the motion is measured at several locations aL the free-

~·: ~f'~ : :~: ~1:: f :j T

surfacc. Fi:orn Fig.1.1:3, it can be noticed that some locations lead to larger sp ectral
ratios: amplitudes range between 3 and 13 in the freqnency band 3-8 Hz. As shown by
· + -i Driad-Lcbeau et al. (2009), such locations correspond to shallow alluvial sites overlying
10" 0 L_J__L_j__.l__J__J__i__i____L__!.-1.-1.-1.--,l:---'--:!:--'----::---'---:":::--'--:7--'--:'. marls of lower Rognacicn . The impedance contrast b etween these two formations learls
0 2 3 4 5 6 7 8 9 10 11 12
frequency (Hz) to a significaut motion amplification (also see Chapter 5) . The annoy2.nccs caused in the
environment may be large in the frequency range characterized by t he spectral ratios.
Numer ical simulations were also performed by Lokmane et al. (2003) for the Ga rdanuc
Fig. 1.11: Vibrations induced in the soil by a wind farm: ground motion
mine taking into account the soil layering as well as t he layers geometry.
spectra at different distances (Schofie ld, 2004).
34 \Vaves and vibrations in soils 1.4 Blast induced vibrations 35

As shown in F ig.1.15 (CETE-Lyon , France) , t he signal duration is less than 1 s. The

20 maximum particle velocity 'Vz = 10 mm/ s is ob ta ined for the vertical cmnponent, (left).
0
...... 15 The mechanical waves propagaLing into the soil may have a signi6cant influence on the
...., 15
ro neighbouring structures.
~
10 10
c\Il
---... -- _,___ __ ., ...
5 5

0 0
0.1 1.0 10 0. 1 1.0 10

20 20

0 15 15
....,
• rl

ro
~
10
c\Il 5 5 ... ______ ___ ....._

0.-..........................~..,_.-..-............"'--o,......I
1.0 0.1 1.0

.s
.....,
1 5 1 - - - - - - - - - -- -----l
ro
~ 101---------~-\-------l
c\Il 51-~~~~¥-~~;.,-
Fig. 1.14: Picture of the blasting phase in construction works (from CETE-
o..__.......................................wp....;:;p...1 Lyon, France).
1.0 10 0 .1 1.0 10
frequency (Hz) frequency (Hz)
10 5 8
Fig. 1 .13: Mine induced vibrations measured at the soil surface for the Gar-
8 4
danne coal mine, France, by Driad-Lebe au et al. (2009). ,.-.,. 6

--ss
en 3

.._,
1.4.2 Vibrations induced in quarries -1
-2
The energy of explosions in quarries or construction works is generally very large and -3
may generate seismic waves propagating along long distances . As depicted in Fig.1. 14 -4
in the case of construction works, the blas t generates two wave typ es: mechanical waves - 10
1.5 - 5
0.5 1.0 0.5 1.0 1 .5 0.5 1.0 1.5
t hrough the soil (vibrations) and pressure waves through the air (e.g. noise).
time (s) time (s) time (s)
In the case of a quarr y, several experimental results are displayed in Fig.1.15 showing
Fig. 1.15: Ground motion (velocities) due to an explosive source close to a
the ground motion measured close to a bridge foundation (from CETE-Lyon, France).
bridge foundation (from CETE-Lyon, France).
The mechanical waves are generated by explosive sources located in 60 holes having a
6 rn length and using a 25 ms delay.
36 ·waves and vibrations in soils l .5 Tl·affic induced vibrations 37

1.5 Traffic induced vibrations The parLicle velocities measured at di fferent poinLs for a truck dr iving at 50 km/ h
arc displayed in Fig.1.1 7: the first signal (top) was meas ured at t he soi l surface 10 m
The main problem for traffic induced vibrations is t he fact Lhat t,hey are originated by from t he road; the second signal (middle) was measured al the ceutcr of the first floor
a moving source. Fmthermore, the interaction between the vehicle and the road/track of a neighbouring building; t he third signal (botLom) was measured at the center of the
generally implies contact/sliding procesHes and may be very complex t o describe. Lt is second floor of the same building (from CETE-Lyon, France).
thus more difficult to characterize the dynamic loading or to model t he wave propagation
procesHt ha n in the previous cases. As shown iu Fig.1.17, the ma.ximnm vclocily is approxi mately 0.1 mm/ sat t he soil
surface whereas it reaches 0.3 m m/s at the second floor in t he building . The annoyance
1.5.1 Vibrations due to road traffic due to traffic induced vibrations thus depends on the source, lhe propagation through
t he soil, soil-slructm e interaction as well as the dynamic features of the structure itself.
Experimental assessment Finally, Fig.l.17 also allows the estimation of t he delay between Lhe passage of each
Vibrations induced by road traffic are cl ue to the regular paHHage of vehicleH as well as t ruck axle (from the signals measured at the soil surface as well as those recorded in the
the interaction with some element s of the road structure. As Hhown in Fig.1.16, if there buildi ng). T he parameters govern ing the signal features arc Lhus also t he vehicle mass,
are some variations in the pavement s trncLure, it may indncc significant interactions with dimensions and speed. As it will be shown hereafter, t his iHsue may be critical for train
the vehicle and its susp ension system. Such interactions will generate dynamic loadings induced vibrations.
on t he pavement that may be propagaLed to t he soil and the neighbouring st ructures.
,.-.._0.3
As illustrated in Fig.1. 16, the vibrations induced by road t raffic may be measured (/)
-...
uHing geophones. This picture was taken from an exper imental study performed in an s 0.2 soil (lOm)
urban environment by the CET E-Lyon (France). Several types of loadings related to s
.__,. 0.1
differen t, vehicle features were analyzed. >.
.._;. 0
......
u
0 -0.1
.......
~ -0.2 delay between axles
,.-.._ 0.3
(/)
-... first floor
s 0.2
s
.__,. 0 .1
.c
...... 0
u
0 -0.1
.......
Q.)
> -0.2
,.-.._ 0.3

ss
(/)
0.2 second floor
,...... 0.1
>. 0
......
.._;.
u
0 -0.1
.......
~ -0.2
0 1.0 2.0 3.0 4.0 5.0
time (s)
Fig . 1.16: Measurement of traffic induced vibrations around a variation of
the pavement structure (from CETE-Lyon, France) . Fig. 1.17: Traffic induced vibrations: particle velocities measured along the
road for different types of vehicles (from CETE-Lyon, France).
 Waves a nd vibrations iu soils 1.5 Ttaffic induced vibrations 39
38

N u merical simulat ion
It is also possible to investigate traffic induced vibrations in soils considering uu merical
methods (Ekevid and Wiberg, 2002; Lombaert and Degrande, 2003; Sheng el a.l., 2006) .
Since such problems involve moving loadings, special numerical fo rmulations have been
proposed (Clout eau et al., 2001).
As shown in Fig.1.18 (Frangois et al., 2007), the passage of a two-axle truck on a
traffic plateau generates vibrations into the soil.
T he bending st iffness of the road being negligible with resp ect Lo the stiffness of
the soil, cylindrical R ayleigh waves can clearly be observed (Fig.l. 18) . Fram;ois et al.
(2007) considered t he influence of dynam ic soil-structure interaction since such waves
may excite neighbouring structures. They analy7.ed the response of a two-st orey single
family dwelling for different foundation types. Frarn;ois et al. (2007) evidenced that
the potential damages in such structures, due t o excessive deformations, depend on t he
ratio between the soil stiffness and the structure stiffness. As it will be shown in Chap-
ter :3, the inAuence of the soil layering on wave prop<tgation in soils may also be very large .

1.5.2 V ibrations d ue to railways

Rayleigh In t he case oft rains, as described in Fig.l.19, vibrations are generated at t he wheel-track
wave contact and radiate waves into t he soil (as well as noise in the air). Armoyances in the
environment have t hus t.o be dealt with a nd depend on the t rain/track features, the
soil properties and t he soil-bu ilding interaction (Fiala et al. , 2007; Sheng el al., 2006;
Paolucci et al., 2003).

\
•
'• _____ ...... , waves propagating
•
·1····\
••••
int o the soil

F ig. 1.19: Train induced v ibration s : schematic o f the p r oble m .

ln the case of high speed t rains on soft soils, t he train speed V t may be close to the
surface wave velocity Vn (Rayleigh waves). The energy may not be completely radiated
in the environment th us leading to a large energy storage under t he train. The specific
problem of superseisrnic ranges is thus important since t he infrastruct ure itself may be
damaged (see next paragraph).

F ig. 1.18: Waves propagating into the soil d u e the passag e of a t wo-axle truck
(speed 14 m / s): compute d displa cements a t different t imes (Fra n c;ois et al. ,
2007).
Ground motion measurement s from a site on the Swedish V.Test Coast Linc were stud-
ied by Paolucci et al. (2003) . As shown in Fig. 1. 20, the train speed has a strong influence
on the ground motion. At 70 km/h, the Peak Ground Velocity is decreased by almost
two orders of magnit ude at only 7.5 m from the track. The same reduction occurs at
d istances larger than 50 m for a t rain speed of 186 km/h (P aolucci et al., 2003). As it
will be shown later , P aolucci and Spinelli (2006) also performed numerical simulations
to assess the influence of t.he st rain speed on the induced ground vibrations.
40 \i\Taves and vibrations in :;oils 1.5 Traffic induced vibrations 11

0
curved wavefront ----......
100 ···········•·!···············I···············~···············<······· ····;········ ·-<--············· 20
:t.;::::~::::::::::::::::::::::::::==:::::::::::::: ::::::;r::: :::::::::,::::::::::::: -- - -- -- ;~--~·~···········
40
::\\:::.>t::::::::::::::i..::::::::::::L ::::::::F..:.::::.::r:::::::·::J::::::::: -- - - 120 km/h
·\:\·f·,·:-::.··. r ..........;.. ·········t······ ····;······ · ·····;········· -·- -- 160 km/h
,..--.._
60

······-::,:xr··············· -·····r········ - - 186 km/h E

'---'
80
Q)
10-' · ········"··''· ·········i · · ................. u 100
:::::::::::::~i::,:::::::::::r . ,...., ........ ....,........:::::::;:::::::::: ::t:::::::::::::;t::::::::::::::t:::::::::::::: i::::
- --- ·········- ';- -··\' · ········! ··············-~---·····- -- ,--~-- -- -- ----···-:-··············:··············:---···········-:----·········· ....,ell 120
::::::::: ::::::\::::~~:::::1:::::::::::::::~:::::::::::: :::F:::.. . - ---··:········::::::r::::::::::::::r:::::::::::::+:::::::::::::: 00
~

·• ·• · · · ·• J\:t~~1;~j;.::]~•~••·E···· · ·~·· · · · ,. . . . . . .. 140

160
~ ...... ! .. ·--...... .... ~ : . ... . :
10
, Fi i I;·;··: ,·f0''J~:,:1;;~i:1~;B~~~~~~~.~~ 180
200

:; ,··· . . . .. . ... .. ' · ·· • ! -- -------- · ~····

E••· · ·L••• ,. . . . . + ···········
· · ··········- ~
0
Mach cone
20 ~
10·3 L-~--'-~~'--~--'-~~'--~-'-~~'--~-'-~--'~~--'-~--'
40
0 10 20 30 40 50
distance (m) ,..--.._
60 x10o 2.0
E 0.0
'---' 80
F ig. 1.20: G round motion d ecay for various tra in s p eeds: Peak Ground Ve- Q)
-2.0
u
locity normalized by the peak ve locity at the track V.nax (Paolucci et al. , i:::: 100 -4.0
....,
ell
2003). ..... 120 I
00 -6.0
'U
-8.0 s
140 -10 8
-12
160
1.5.3 Theoretical analysis of moving loads -14
180 -16
Various authors studied the influence of t he train speed / wave velocity ratio on Lhe gen- -18
erated wavefield (.Ju and Lin, 2004; Lefeuve-Mesgouez et al. , 1998, 2002 ; Paolucci a nd 200
-20
Spinelli, 2006). When the train speed Vt is not far from Lhe Rayleigh wave velocity V.n, 0 20 40 60 80 100 120
t he energy radiation process may be strongly modified. distance (rn)

As discussed by Lefcuve-Mesgouez et al. (2002) or Paolucci and Spinelli (2006), two Fig. 1.21: Wavefield originated by standard (top, Vt = 0.78VR) and fast train s
different cases ar e possible: (bottom, 'Vt= VR ) , from Paolucci and Spinelli (2006).

• 'sub- Rayleigh' case: for Vt < Vn , the velocity of Rayleigh ·waves propagating inLo
the soil is lar ger than the train speed and the energy is radiated beyond the train
as usual, As shown in Fig.1.21 (Paolucci and Spinelli, 2006), the firs t case (Lop, Vt = 0.78Vn )
leads to a curved wave front whereas the second case (bottom, Vt = Vil) produces .:Vlach
• 'super-Rayleigh ' case: for Vi ? VR , the train speed is larger than the R ayleigh type cones similar to Lhat obtained for supersonic flights. They are called ' super-seismic"
wave velocity and some energy is stored under the train, possibly leading to some Mach cones.
damages in the track.
42 \Vaves and vibrations in soils
1.6 V ibration isolation
1.6.1 Practical problem
T he vibrations generated by various sources (rotating machines, road traffic ... ) may
induce nuiimnces in neighbouring buildings or for sp ecific types of activities (electronic
industries, particle accelerators ... ). Wave propagation due to sources on/ in t he soil may
actually induce particle velocities (or accelerations) disturbing for the people or the in-
dustries in the vicinity.
To solve this problem, iL is necessary to either lirniL the vibrations Lransmi tted t o Lhe
environmen t (active isolation), or Lo modify the propagation in the soil or the transmis-
sion to the structures (passive isolat'ion) . To limit the wave propagatiou , several isolation
methods have been proposed. They generally t ry to change the soil characteristics to
modify the propagation process. It is t hus necessary to insert a different material in the
soil which can consist in:
• a Hoil volume with a different deusity from t he neighbouring soil (e.g. trench mix
technique, jet grouting),
• a linear or cyli ndrical trench which can be empty or filled with another material
(e.g. damping material),
• an embedded screen constituted of a stiff material (e.g. concrete),
• one or several pile groups (or cylindrical cavities) in the soil.
T he pa rameters mainly influenciug the wave propagation process in the modified soil
arc the following:
• the velocity contrast between the soil a n<l the material,
• t he isolation system geometry (depth, wid th),
• in the case of pile grou ps or cylindrical cavit ies, the ratio between the pile diameter
and their spacing.
1.6.2 Experimental results
Several books (Richart et al., 1970; Skipp, 1966) deal with vibration isolation in Civil
and Environ mental Engin eering. A review of codes and standards was proposed by Skipp
(1997).
Isolation by trenches
Experiments perfor med by W oods (1968) and Richart et al. (1970) . As de-
picted in Fig. 1. 22, Woods (1968) and Richart et al. (J 970) analyzed the efficiency of
various trench geometries (circular, rectangular, etc) as an isolation system. A rotating
machine is installed at the center of a circular trench and the amplitude is measured
J.6 VibraLion isolation 43
along various radii originatiug from the trench (Fig.l.22). The amplitude reductiou is
esLimated by comparing Lhe motion amplitude with the trench and Lhe motion amplit1Lde
without the trench.
Their analysis deals wit h surface (Rayleigh) wave propagation generated by harmonic
vibration sources at the soil surface for frequencies ranging from of 200 to 350 I-fa. \Voods
(1968) considered various trench types characterized by their dimensions : depth H , width
w and distance from the source of vibrations R. T hese geometrical parnrneLers arc given
as fractions of Lhe Rayleigh (surface waves) wavelength AR.:
• circular trench: H = 1.45Au and R = 0.73An,
• semi-circular trench : H = 0.60Au and R = 0.60An,
• rectangular trench: H = l. 19AR; R = 2.97AR and length L = l.79AR-
exciter footing
"'-...
R
w
Fig. 1 .22: Vibratory isolation using circula r trench: description of t h e exper-
imental set-up (Woods, 1968).
E fficie ncy of trenches as isolation system. The efficiency of Lhe isolation system
is assessed by estimating the amplitude reduction factor. In Fig.1.23, the amplitude re-
uuction factor is plotted againHt the diHt ance frorn Lhe source for various trench depth to
wavelength ratios: H / Au= 0.363 , 0.726 and 1.452 (resp. ). T he efficiency of the t rench is
found to be larger for deeper trenches since t he maximum reduction is found to be larger
than 90%.
The amplitude reduction all around the trench is displayed in Fig.1. 24 for a semi-
circular trench and a rectangular Lrench. T hese results show that low amplitude areas
(< 0.125) appear beyond the trench, whereas some areas may experience a significant
motion amplification (> 1.25) before the trench. The efficiency of t he isolation system
studied by Woods (1968) and Ri chart ei al. (1970) is thus very good.
The op timal t rench dimensions depend on the main wavelength of t he problem. The
minimum values of the ratios between the system dimensions and the wavelength (that
44 ·w aves and vibralions in soib 1.6 Vibration isolation 45

is H / A R and w/An) only allow app licatiollS at frequencies down to several hundreds of
H7. (keeping wavelengths not t oo lar ge). Another remark on the isovalue plots displayed
in Fig. 1.24 concerns the non-symmetric features of the results . In actual soil dep osits,
the soil layering is not perfecily fla t and some heterogeneities may occur: the isolation
efficiency is thus not fully symmetric.

0.7
\
H!An=0.363 '' / \

. ',
o----<>
0.6 '' /
/
\
~-- - -e H/Att=0 .726 '' /
/
\
""'
0
.;...>
• --------• H/An=l .452 '' /
/ \
u \
<S 0.5 '' /
/
/

>::
.....0 - - __ _- -·~ - __ _.... - - _,,_
.. - .. .. .. .. .. ..
''
•a
'
.;...>
u 0.4
;:l
'"Cl
''
,,
Q)
\
""'Q) 0.3 \

~
'"Cl
::l
.....
.;...>
.-<
0.2
'
P..
sro
0.1

• >1.25 D 1.25-o.50 D o.50-0.25 D 0.25-0.125 D <0.125

00 2 4 6 8 10 12
distance from source (ft) Fig. 1.24: V ib ratory isolation using different tre n ch typ es: semi-circular (le ft )
and r e ctangular (r ight) , from Woods (1968) .
F ig. 1.23: Amplitude r eduction factor for different normalized tre n ch d epths,
from Woods (1968).
Other experimental works were performed by Woods et al. (1974) to analyze vibra-
t ory isolaLion by rows of (empty or filled) cylindrical wells. To estimate the influence of
t he isolation system on t he wave propagat ion process, Woods considered an holographic
1.6.3 Numerical models tedmique. Woods gives the global efficiency of the system as a function of the well:;
diameter mid their relative distance. Numerical siruula.tions of such a problem are pro-
Various approache s in differe nt fields posed in Chapter 4.
Many different authors h ave studied wave isolation methods and systems (Aviles and In recent years, the extensive use of numer ical tech niques in 2D and 3D has motivat ed
Sanchez-Sesma, 1983; Banerj ee ct al., 1988; Beskos et al., 1986; Ha upt, 1977; Liao and new researches in this field:
Sa ngrey, 1978; May and Bolt , 1982; Scgol el al. , 1978; Woods et al., 1974).
• Karlslrom and Bostrom (2007) , Sheng et al. (2006); Andersen and Nielsen (2005)
Haupt (1977) mainly investigat ed isolat ion using stiff embedded screens. Ba nerjee or .Tu and Lin (2004) for train induced vibrntious,
el al. (1988) considered a Boundar y Element Method to simulate the fi eld exper iments
perfor med by Richart et al. (1970) (sec F ig.l.24 and Chapter 1). Compar ing different • Fl:arn;ois el al. (2007), Lombacrt a nd Degrande (2003) or Cloutcau et al. (2001)
resulLs indicat e th at an isolation system involving an empty trench is generally more for road traffic induced vibrat ions,
efficient.
• Chehab and cl Naggar (2003) for ham mers and presses indnced vibrat ions.
 Waves and vibrations in soils 1.6 Vibration isolation 47
46

Modelling the efficiency of trenches for trains

Karlstrom and Bostrom (2007) studied the efficiency of trenches on the isolation of wave
induced vibrations propagat ing into the soil. Three cases are compared in Fig.1.25: the
ca.sc without any isolation t echnique (right), the case involving two trenches (middle)
and the case using one trench only (left). As shown in Fig.1. 25 by the numerical resul ts
of Karlstrom and Bostrom (2007), the efficiency of the trench is very good. However,
the vibration level being reduced in Lhe environment , the mechanical waves seem to be
trapped in the track st ructure (Fig.J .25, middle). It may lead to some damages Lo the
track structure.

y (m) y (m) Y (m)

.h r0 N -"" .h N N -"" .h N
0 0
N
0
-""
0
()10 0 0 0 0 oiQ 0 0 0 0 ()10
0 0 0

r. ~ ~

0 0
0

>< >< ><

,....._ ,....._ ,..-._

e..- s
......,. s
......,.

&, U1
I
&,
0 0
0

0' ...__ _ _ _ _ _ __. Fig. 1.26: Numerical a nalysis of various isolation techniques for train induced
.'.... .'.... ~

0 ------~-- 0
0
--------~-' 0 vibrations by Andersen and Nielsen (2005): (a) origina l track and s ubsoil; (b)
0
soil improvement; (c) open trench a nd (d) trench backfilled with concrete.
-w (mm) L
5 0 5

F ig. 1.25: Numerical a nalysis of various isolation techniques for train in-
duced vibrations by Karlstrom and Bostrom (2007): no isolation (right), two
trench es (middle) and one trench (left).
Modelling various isolation techniques
Andersen and Nielsen (2005) studied numerically various isolation techniques. As shown
in Fig.1.26, the wavefield radiated in t he vicinity of the track is significantly different
from one case to the other. T he efficiency of ihe open t rench (bottom left) and the con-
crete screen (bottom right ) appears to be larger than that of the improved soil (top right).
Considering different lines parallel Lo the track (denoted 2, 3 and 4 in F ig.1.26) ,
Andersen and I\ielsen (2005) assessed the effic.iency of each isolation system along these
lines. For a pair of vertical forces, the results are displayed in F ig. 1.27 in terms of pseudo-
resultanL displacement (PRD = Jluxl 2 + !1iyl 2 + luzl2 ) at two frequencies: 10 Hz (left)
and 20 Hz (right). Along lines 3 and 4 (i.e. beyong the isolation system) , the efficiency of
the deepest open trench (c2) is the largest and t hat of the concrete screen (d) is close to
the deep trench far fro m the track. The efficiency is larger at 20 Hz (Fig. 1.27, right) since
the related wavelength is shorter (the isolation system geometry thus "appears" larger).
The soil improvement technique (b) does not appear as a very efficient t echnique in
terms of isolation away from the track. However, before the isolation system (Fig.1.27,
top) , the soil improvement technique (b) leads to smaller displacements close t o the track
 Waves and vibrations in soils 1.6 Vibration isolation 47
46

Modelling the efficiency of trenches for trains

Karlstr()rr1 and BostrOrn (2007) studied the efficiency of trenches on the isolation of 1,vavc
induced vibrations propagating into the soil. Three cases are compared in Fig.1.25: the
case \vithout any isolation technique (right), the case involving two trenchefl (rniddle)
and the case using one trench only (left). As shown in J.i.,ig.1.25 by the numerical results
of KarlstrOm and Bostr6m (2007)~ the efficiency of the trench is very good. Ho-,,vevcr,
the vibration level being reduced in the environment, the mechanical \vaves see1n to be
trapped in the track structure (Fig.J.25, middle). It may lead to some damages to the
track structure.

y (m) y (m) y (m)

0 0
0

"' "' s"'

~ ~

s
~

~ 2-. ~

&, &, &,

0 0 0

Fig. 1.26: Numerical analysis of various isolation techniques for train induced
.'... .'... ~

0 0
0
0
0
vibrations by Andersen and Nielsen (2005): (a) original track and subsoil; (b)
0
soil improvement; (c) open trench and (d) trench backfilled with concrete.
-w (mm) C-
5 0 5

Fig. 1.25: Numerical analysis of various isolation techniques for train in-
duced vibrations by Karlstrom and Bostrom (2007): no isolation (right), two
trenches (middle) and one trench (left).
Modelling various isolation techniques
Considering different lines parallel to the track (denoted 2, 3 and 4 in Fig.1.26),
A.ndersen and Nielsen (2005) assessed the efficiency of each isolation system along these
lines. For a pair of vertical forces, the results a.re displayed in Fig.1.27 in terms ofpscudo-
rcsultant displacement (PRD ~ vluxl 2 + luyl 2 + lv·zl') at two frequencies: 10 Hz (left)
and 20 Hz (right). Along lines 3 and 4 (i.e. bcyong the isolation system), the efficiency of
the deepest open trench (c2) is the largest and that of the concrete screen (d) is close to
the deep trench far from the track. The efficiency is larger at 20 Hz (Fig.1.27, right) since
11 11

Andersen and Nielsen (2005) studied numerically various isolation techniques. As sho\vn
in Fig.1.26, the wavcfield radiated in the vicinity of the track is significantly different
from one case to the other. The efficiency of the open trench (bottom left) and the con-
crete screen (bottom right) appears to be larger than that of the improved soil (top right).
the related wavelength is shorter (the if;olation systcn1 geometry thus appears larger).
'"rhe soil improvernent technique (b) does not appear as a very efficient teclrr1ique in
terms of isolation away from the track. However 1 before the isolation system (Fig.1.27,
top), the soil improvement technique (b) leads to sm'1ller displacements close to the track
 \;\.Taves and vibrations in soils l. 7 Earthquake engineering and sei.srnology 49
48

and may reduce the potential darnages vvhen cornpared to other isolation techniques. 1.7 Earthquake engineering and seismology

(a) (b) ··--------- (c) (c,l (d) 1.7.l Analysis at various scales
V\Tavc propagation in surface soil layers is a key isr-n1e in the field of engineering scisn1ology
and earthquake geotcchnical engineering. Investigating seismic \Vave propagation in soilH
is necessary to esti1nate the scisn1ic ground rnotion at the surface, to study the response
of large constructionH (bridge) clam 1 poVi.rer plant ... ) or the behaviour of geotechnical
structures such as cmbankn1ents or retaining walls. The assess1nent of the propagation
-10 0 10 process in the Rnrficial soil layers it-i thus rnandatory to characterize the seisrnic 1vavefield
in such cases. As Vi.rill be discussed in details in Chapter 5, scisn1ic vvave arnplification
] 2 x10·10
(or site effects) and slope stability are important topics (Bard and Bouchon, 1985; Seed 1
······ .......•.,·.:.-.·.- .. 1974; Semblat et al., 2005) .
. ..-'
1 ,_-;.-:·---~-·-- ..·.'i~
_..'l-::::_·-,..,..--......... ,,--<~',
. ' ,..~.":>---:.:-
' --
-- ........ __ .......... :::: ......
_,..
~
------ ,.. ·"' ,..... As depicted in Fig.J .28, seismic hazard and risk rnay be considered at various scales.
~20 -10 0 10 -10 0 10 Frorn a phenomenological point of view, seismic 1,vave propagation can generally be ana-
. 15 lyzed considering one of the foll.o\ving points of vieV1.r:

0 ~~x10::~:~:~::~--~--~--~--~--~-.;:~·.·~·:>•@I
-... 1.5 x1 o-10
~ ____, __________ . - .-;;;...-.·=--.-=--------- ., • regional propagation frorn the seismic source 1
-it ,.-
;.a 0.5 -
./....--~ - ___ ,.. .... - -- --- - ·,.
-
- - _.::-... --- ..--........ _--- - - - - - - . . _- - - --=--- .___ • local propagation in surficial geological structures 1
~ OL..~-...,..-~-,__~--c::-______o OL.-~-~

-20 -10 0 10 -20 -10 0 10 • propagation and interactions at the scale of one or several constructions.
distance from the source (m) distance from the source (m)

Fig. 1.27: Efficiency of various isolation techniques for train induced vibra-
tions: pseudo-resultant displacement in the vicinity of the track (lines 2, 3
and 4); (a) original track and subsoil; (b) soil improvement; (c1) 2 m deep
open trench; (c 2 ) 6 m deep open trench and (d) trench backfilled with con-
crete (Andersen and Nielsen, 2005).

1.6.4 Values of the mechanical parameters

The main mechanical pararneters for vibration isolation problerns are the follovving :

• frequency: 10 to fiOOO Hz,

• amplitude reduction: divided by a factor 4 or 8 (sec figure (1.24)),

• trench or inclusion depth: up to one wavelength,

• corresponding Vl.'idth: from 0.1 to 0.5 Vi.ravelength.

Fig. 1.28: Various scales to analyze seismic wave propagation in geological
structures.
 \\Taves and vib~:ations in soils
50
1.7.2 Seismic wave propagation in soils
Tl1c Hecond topic of the previous list rr1ai.nly conHiders the local geological (and geornet-
rical) conditions near the free-surfa,cc since they may influence the propagation process.
J)ue to complex refraction and scattering phenornena detailed in Chapter 5, seis1nic waves
1nay be a.rnplificd in aJ.luvial deposits or on topographic irregularities (Ba.rd and Bouchon,
1985·, Dobrin et al., 1954:. Moeen-Vaziri a.nd Trifunac, 1988; Sanchc"-Sesrna. et al., 2000).
As shown in Fig.1.29 for the Kobe basin, the geological structure (top) is strongly
heterogeneous and leads to complex propagation phenorncna (Ka\\rase et al., 2000). In
such cases, as computed by J'vfatsushirna and Kawase (2006), the seismic vvavefield is thus
amplified in the surficial layers as shown in the Peak Ground Velocity plot (bottom).
JMA
.I
42km
Fig. 1.29: Seismic wave propagation in the Kobe area (Japan): geological
structure (top) and computed Peak Ground Velocity for the Kobe 1995 earth-
quake (Matsushima and Kawase, 2006).
1.7 Earthquake engineering and seisrnology 51
The seismic ground inotion to consider for the design of buildings or bridges is thus .in-
fluenced by such cornplcx propagation phenomena. Scis1nic code provisions are generally
based on simplified lD analyses (hori2'ontally layered soils). However, actual an1plifi-
ca.tion levels due to the local geology rru1y raise the need for improverncnts to current
code provisions (e.g. aggravation factors) or for n1ore corr1plex experi1nental/nu1nerical
analyses (Gelis et al., 2008; r,,1Iakra ct al., 2005). Such issue.s Virill be extensively discussed
in Chapter 5.
The analysis of soil-structure interaction generally considers seisn1ic -Virave propagation
under a si1nplified forrn. The inain issues to be ernphasized are the boundary conditions
at the base of the structure (fixed base assumption not valid for soft soilfi) and the en-
ergy radiation into the soil (influence on the overall darnping in the structural ret>ponse).
The detailed analysis of the \Va:ve propagation process is generally not mandatory except
for large structure8 such as dan18) bridges or poV1rer plantH. In such cases, the influence
of the variability of the scisn1ic \Vavefield rnay be significant and may lead to different
assumptions for their design. For instance, Sextos et al. (2003) sho-\ved the influence of
the seisrnic wave passage on the design of long span bridges.
For Htrong seismic rnotions 1 the analysis of the dynamic soil response strongly de-
pends on the soil constitutive behaviouT as >..vell as on the loading histor:y. Such con1plex
phenon1ena as the liquefaction process may lead to very large strains as \vell as very poor
bearing capacities. 1\!Iany various con1plex constitutive rnodcls have been proposed in
the literature (Bonilla, 2000; Delepine et al., 2007; Ha.shash and Park, 2001; Kausel and
Assimaki) 2002; Iai et al., 1995; Prevost, 1985) but this issue is beyond the scope of thiH
book mainly dedicated to the propagation process in soils.
1. 7.3 Values of the mechanical parameters
Earthquake engineering proble1ns are characterized by the following mechanical param-
eters:
• frequency: fro1n 0.1 to 20 H:0,
• acceleration: fron1 0.1 to 20 n1s- 2 1
• vvavelength: from several tens to several hundreds of 1n,
• strain: from 10-4 to 2.10- 3 m/m.
52 vv~avet-i and vibrations in soilt-i

1.8 Synthesis of the various parameters

For the varion8 dornains of application detailed previo1uily, the 1nain pararneters corre-
spond to very different ranges of the mechanical parameters (frequcncy1 strain rate, etc).
Usual va]uet-i of these various pararncters are synthctized in Table 1.1.

Table 1.1: Characteristic values of the main mechanical parameters.

Chapter 2
Type of Ivlain n1echanical pararnetcn.;
problern
energy
charactcriJ-;ing the problc1n
force at JJ'ile-head
lD-wave propagation
pile
driving 10 to 3000 J 10 to 10000 kN
particle velocitu displacem,cnt Application to the dynamic
several mrn/ s Reveral rnm, to several drn
dyna1nic energy rnass acceleration characterization of soils
_______.s_.,c_.,_.'e:;:r:::a::_l_.JV,:;:f_:J~_________5_0_t_o_5_,0~0:-r_n_.8_-__ __
2
corr1pactio11
particle velocity stress
10 to 30 m,s- 1 HJ to 500 kPa 2.1 Introduction
vibratory frequency force
In this chapter, lD 1,va:vc propagation is Rtudic<l considering the cquilibriu1n of curvilinear
compaction -----~10:'...'"tc_.i..':8..':0_'H:':z""--------_c5:-.:0c_t;.co:,.-20.:_(;.cl.:_k.:_N_--,--
cylindcr ax'i8 speed vertical displacement n1edia.. Such 1nedia can be described as bearns and ba.rs 1 for 1,vhich the systen1 of coor-
around 1 rn.8-l several rnrn dinates is reduced to the location of the cross-section in terms of curvilinear abscit-isc.t s.
frequency ampl'itude reduction The dynamic cquilibriu1n equation of such systerns leads to 2nd order partial differential
vibratory
10 to ,5000 Ih divided by 4 or 8 equations with the following independent variables: the _point location ;r_, in a specific
isolation
incl'Usion depth inclv,sion VJidth coordinate syste1n, and the tin1e t. The problen1 unknowns arc represented by the dis-
up to one vva:velength from 0.1 to 0.5 1,vavelength placernent field 1L = g( s, t) for the crosR-Rection of abscissa s.
earthquake freq'Uency acceleration
0.1 to 20 Hz 1 to 20 m,8-
2 The lD problem is firstly solved for various types of motion (longitudinal, trant-iverse,
engineering
wavelength strain etc). The corresponding results then allow the dyna1nic characterization of t-ioils and
several tens to several hnndreds of rn 10- 4 to 2.10- 1n/m
3 rocks (reRonant column, Split ITupkinson PressuTe I3ar, etc). VVave propagation in dissi-
pative media as well as layered rnedia are also discussed in details.

The rnain interest of the lD analysis is to lead to scalar wave equations. In the next
chapter, 2D /3D theory may be derived under a si1nplifiecl lD forrn to be co1npared to
the present lD analysis.

2.2 Dynamic equilibrium of a beam

2.2.1 Kinematics and main assumptions
One firstly considers the dynarnic cquilibriun1 of a. Learn. To write the equations of the
dynarnic equilibriurn of the syste1n, the follovving assumptions are considered:

• the syRtem corresponds to a linear bean1,

 lD--,vave propagation 2.2 Dyna1nic equilibriurn of a bea1n 55
54

and the contact forces are \Vritten using the streHH tensor a and the outvvard normal n
e the transforrnations re1nain SJnall, to the section: = -
e the cross-r:;cctions of the bean1 rernain plane during the deformation (N avicr-Bernoulli
P,(U) - { (rz rr). dfl_ dO
assu1nption). Jn(s) - ds
'fhe mechanical fonnulation is \vritten thanks to the Principle of Virtual VV-ork (Salen- - { (rz.rr). [d0. A(:!i-1:(8))] d(l (2.5)
gon~ 2001) involving, for any virtual velocity fic]d fJ, the rate of Vi.rork by internal forces Jn(s) - ds
Pi(U), the rnte of work by external forces P,(U) and the rate of work by quantities of
acceleration A(U). It leads to:
+j U(s)
(grr) [0.A D~(s)l dO
OS

(2.1) Taking into a.ccount the follcn~ring identity:

·The bearn kinc1natic is fully described by a translation Qy(s) of a point of the neutral
(2.6)
axis and a rotation ~(s) of the crosR-scction (Fig. 2.1). In such con<litiuns 1 the virtual
and the relation :
velocity fiel<lR at point },;[ can be written: Dr.(s)
--=n (2. 7)
08
-
(2.2)
equation (2.5) can be rno<lified in the follo>.ving form:

where£ is the location of point A1, r_(s) the curvilinear abscissa of the cross-section and
/\ the vectorial product of t\.VO vectors.
, J
Pi(U) = -
Cl(<)
(g 11) dv
d-T d'2
8
-
1 Cl(>) -
dw
[(I-r.(B)) A (rzrr)] .-=&l
ds
(2.8)
Q,(s) +1 (g 11) [0. A rr] dO
n(.~)

M At this stage 1 it is possible to introduce the generalized resulting force acting on the
------- ---+--·- crosti-section. It is represented by hoth the force V and the 1no1nent 11v1:
r(s) fD(s)
JI:= { (grr)dS
.fs(,l
(2.9)
Fig. 2.1: Virtual velocity fields in a beam. M= { ((:!i-r.(s))A(rzrr)]dS
Js(,) -

\.vhcre S(s) is the cross-sectional area at abscisRa s.

2.2.2 Virtual rate of work by internal forces Introducing these qua.ntitics into Eq. (2.8) 1 the virtual rate of \Vork by internal forces
The virtual rate of work by internal forces Pi(U) for a beam volume O(s) of thickness betvveen tvvo sections A and B is >.vritten:
ds is obtained by integrating a volume density p.,(U) over n(s) (Saleni;on, 2001):
P,([z_) = - LB [v ( d:~ -0.Arr) + M. ~~] d8 (2 JO)
Pi(U) = { Pi(Q)dO (2.3)
Jo(,) The force V and the 1no1nent Jo..![ can be split into t>.vo components: one oriented along
the section nor1nal J} and one belonging to the croDs-section:
1'his volume density of the virtual >.vork by internal forccs 1 Pi(f!_) 1 corresponds to
(2.11)
the virtual work by the contact forces in the gradient of the virtual velocity field. The
gradient of the virtual velocity fields is derived from Eq. (2.2): Under this form 1 the classical parameters of the beam theory can be identified: the nor-
mal force !V- 1 the shear force V.l . 1 the torsional rno1nent 1Vft and the bending rnornent A1 _L.
dfl_ = dfir(s) + dQ.(s) ;\ (± _ r.(s)) _ 0.(s) ;\ D~(s) (2.4)
ds ds ds iJs
 lD-vvave propagation 2.2 Dyna1nic equilibriun1 of a bea1n 57
56

'rhe sa1ne deco1nposition can be performed on the cornponents of the virtual velocity 2.2.4 Virtual rate of work by quantities of acceleration
fields 1'r(s) and &(s) by introducing Eq. (2.11) into Eq. (2.10) and noticing that the To sin1plify the presentation, vvc neglect the inertial terrns in rotation. J)enoting p the
scalar product ri.I is zero. J:<"""'or every vector j;_ located in the cross-section, the virtual rate n1aRH density of the Lca1n and S its cross-sectional area, the virtual rate of \Vork by
of \vork by internal forces becomes: quantities of acceleration is vvritten:

AT
P.(G)=
i,_
·
13
A
a·
j [-N---V.
d.s
(d.ds
!ln
-J_
1'.l_
--wf -\n
-
A
) d&n
-M1· - --M
ds
d&J_l
_J_ . -
ds- ds
·
(2.12) A(ll) = l 13
pS [t(s).1'(s)] ds (2.19)

Under this form 1 the various tcrn1s in the right hand terrn of l<~q. (2.12) correspond
to the classical paran1eter::cJ of the bearn theory:
2.2.5 Equilibrium equation
Introducing Eqs (2.12), (2.18) and (2.19) into Eq. (2.1) leads to the uynamic equilibrium
• first terrn: elongation of the neutral fiber;
equations of the bea.1n. The various tern1s involved in these equations can be grouped into
• tiecond terrn: shearing in the section; three categories: the ter1ns involving co1nponents norn1al to the cross-section, co1nponents
around an axis nor1nal to the cross-section and cornponents belonging to the cross-section.
• third tcrn1: torsion around the axis nor1nal to the section, The virtual velocity fields being arbitrary, it is possible to choose fields succesHively vvith
cornponents norrnal to the ,':'lection, around an axis norrnal to the Hection and in the cross-
• fourth term: bending around the axis located in the crosi'l-section.
section (Salcngon, 2001). Vle thus write the principle of virtual v,rork for each of these
fields. The equilibrium equations are detailed in the follol'lring.
2.2.3 Virtual rate of work by external forces
A.ssu1nir1g the bea1n being subjected to a volurnic density of external force8 f, the virtual Component normal to the cross-sectio11: elongation
rate of -,,vork by external forces may be written: - F'or the co1nponent nor1nal to the cross-section, the follo-,,ving equation is obtained:

(2.13)
(2.20)

Taking into account the expression of the virtual velocity field given by Eq. (2.2) 1 it considering free ends and integrating by parts the first term:
yields:
l_.ll = l_.i'r(s) + l_. [&(s) /\ (:i: - L(s))]
that is, considering Eq. (2.6):
(2.14)
LB (~~ + fn - pSu,,) Dnds = 0 (2 21)

(2.15) Virhich should be valid for every virtual velocity Dn. As a result) the equilibriu1n
equation becomes:
8N
Introd11cing the linear density of external forces: -;:;-
vs
+ .fn - pSii,, = o (2.22)

L= j'S(>)
[_dS (2.16)
Component around an axis normal to the section: torsion
For this co1nponent, the equilibriun1 equation is written as follo-,,vs:
and the linear density of 1noment fr·om external forces:
13
rn = ;· [:i: - z:(s)] f\ [_ dS (2.17)
1A
(- M, dw,
ds
+ m,w,) ds = O (2.23)
. s(,l
integrating by parts the first tern1 (free ends assumption) and \Vriting that this relation
The virtual rate of work by external forces bct\veen tvvo cross-sections .11. and B is
shoulcl be valid for every virtual velocity Wt, the cquilibriu1n equation is derived as:
written:
113
P 0 (ll) = j·B
A f_,.i'ds + A rn&ds (2.18) 8Mt
as+m,=0 (2.24)
58 lD-vvave propagation 2.3 Longitudinal vibrations of beams 59

or, taking into account the rotational inertia neglected Vi.rhen vvriting the virtual rate rnon1cnts are obtained, hence lea.ding to the global equilibrium:
of work by qua.11tities of acceleration, the follcnving form is obtained:

oM, ··
OS + m, - pJe, = 0 (2 25)

(2.31)
\vhcre J is the geo1netrical torsional inertia of the beam and Ot is the torsional angular
acceleration.

Component in the cross-section: bending-shear

rrhe follovving equation is obtained: 2.3 Longitudinal vibrations of beams

2.3.1 Dynamic equilibrium
In thiti section; we consider the longitudinal vibration of beams, that is phenomena con-
(2.26)
trolled by Eq. (2.22). Equation (2.22) was obtained as a specific case of the general
equations governing the vibration of bearr1s. It should be noticed that it rnay also be
derived by identifying the axial forces acting on a beam slice and \Vriting that their sun1
1'he previous relation should be valid :for every virtual velocity fields ~--1- and ~J_.
is zero (Fig. 2.2).
Ir1tegrating by parts (free ends assurnption) and using Eq. (2.6) lead to the t\vo following
equilibriurn equations: ds
(2 27)

•8N
N+-ds
(2.28) as
1'he previous vectorial equations can be expressed under a scalar form by choosing,
in the cross-sectio11 1 a syste1n of norn1al axes for which the co1nponents of the various Fig. 2.2: Axial equilibriurr1 of a beam.
paran1ctcrs arc \vritten:
This equilibriun1 equation is \vrittcn:

MJ_= { M1
~2 } , v J_ =
Vi
{ }
Vi
0
ffiJ_ =
{ }
rn1
rn2
0
n=
{~ } (2.29) N
oN +(Jn - fr )d8 -
+ -,-d8
Os
N = 0 (2.32)

in vvhich 1 denoting S' the section area and p the densit:y1 the inertial forces fr a.re
Equation (2.28) time yields the following equations: expressed as:
(2.33)
To derive the proble1n Holution, in addition to Eq. (2.22) describing the equilibritun 1

l
Vi = - oJV,.f,
08 - m2.
it is necessary to \VTite the cont->titutive lav.r of the beam for purely axial loads. This
(2.30)
0M1 conr.;titutive law relates the axial force to the elongation of the neutral fiber. Considering
Vi =+--+m 1 linear elasticity and Hrnall strains, it is written as follows:
08
OU (2.:J4)
then 1 introducing into Eq. (2.27) 1 the tl'.ro equations governing the equilibrium of N=ES-
08
60 ID-vvave propagation 2.3 Longitudinal vibrations of bca1ns 61

Virhcrc E is the Yonng't1 rr1o<lulus of the incdiuni. In E·q. (2.31) and all equations in the Equation (2.:l7) is then equivalent to the difforenti;iJ system:
preHent section) index n is 01nitted to make the presentation clearer; u. correRponds to
the dit~placerncnt oriented along the section nurn1al n_.
(2.40)
Introducing Eq. (2.34) into (2.22) 1 the equation describing the axial vibrationR of the
bearr1 is:
2
8 11
Dt'
_ _I!_~
pS(s) as
(s( au)
as 8) = _!_
pS(s)
(2.35)

For a 11011-conHta.nt cro8s-scctiunal area 8(8), it is thuR necessary to dctcrrninc the DU + [A] of!_ = 0 (2.41)
spatial derivative of the S(s) ~: product. For a cylindrical beam (S(8) constant), the at m
equation describing the axial vibrationM is as follov1.rs:
1'he operator [A] has the follovving expression:
2
8 2 ·u
'b l
E8 11. f
------
[)t2 P Ds2 pS
(2.36)
[A] = [ ~ (2.42)

2.3.2 Homogeneous equation and its eigenvalues arc +c anrl. -c. The general solution of the differential sys-
tem (2.40) is:
Assu1ning the linear density of forces f to be zero, the hornogeneons equation of vibrations 11.(s, t) = ·u1(s - ct)+ v.2(8 +ct) (2.43)
is the follo1.ving:
Solution (2.43) represents the superposition of two 1,vaves, vvith a1nplitudes u1 and
ii(8, t) - c2u 0 (8, t) = 0 (2.37)
8 > 0 and direction 8 < 0 respectively.
u 2 , propagating at velocity c, to\\rards direction
in \.Vhich 1.ve introduced the lD longitudinal wave velocity c, defined as 1 : During their propagation, both "''aves keep the san1e shape (Fig.2.:1); these \Vaves a.re
called non di8pcrs'ive 1.va:ves.

c=~ (2.38) Another for1n of the solution of differential syste1n (2.40) 1nay be:

·ii and u" denote the second order partial derivatives with respect to time and space u(s,t) = u 0 (t-D +u4 (t+~) (2.44)
(respectively).
which is equivalent to solution (2.43).
Classical wave velocity values for usual n1aterials arc:
Quantities u 1 and 'u 2 arc deter1nined by the initial conditions. The differential equa-
• steel: c=5200 m/s, tion governing the axial vibration of bearns being of second order, tvvo initial conditions
are needed:
• concrete: c=3600 n1/s, at t = 0 u(s, 0) and ·u(s, 0) (2.45)

• dry sa.nd: c=300-500 m/ s.

u
Equation (2.37) corresponds to a first order hyperbolic differential system. Its solution
is classically obtained by introducing the two unk1,vovvns [!1 et U2 defined as: u,(s-ct) u,(s+ct)
U1 ='U U2 = - "
[)
us.
/''
0
u(s,T)dT (2.39)
s
1As \Ve vi'ill see in Chapter 3, the lD longitudinal wave velocity c differ:,; front the 2D j:~D pressure
wave velocity Vp because the lateral boundary conditions (free-surface) in a beani arc noi. the same as
in 2D/3D
Fig. 2.3: Longitudinal waves propagating in a bar.
 1

62 11)-vvave propagation 2.3 Longitudinal vibrations of bearns 63

Deriving the displacernent and the velocity at tin1e t = 0 from solution (2.43)) for Introducing this relation into Eq. (2.:37), fonction X(s) should satisfy:
every section of abscjssa s) it gives:
(2.52)
u(s, 0) = ur(s) + u2(s) 2
{ u(s,o) = -cu\(.s) + cuS(s) ( .46) The solution of Eq. (2.52) is of the form:

It is then possible to write H1 and u2 as functions of the initial conditions: X(s) =A cos(~ s) + Ilsin (~ s) (2.53)

1'
l
Hr(s) = -l u(s, 0) - -1 c/1((3, 0) d{! +a
2 2c 0
(2.47)
u 2 (s) = -1 u(s, 0)
2
+ -1
2c
1·= u(!J, 0) d(J - a
0

where a is an integration constant. L

Fig. 2.4: Cantilever beam to estimate the axial eigenmodes.

2.3.3 Solution in terms of stresses
The solution being kncnvn in terms of displacement, the stress in the beam can be derived
The integration constants A and fl are dctcr1nined by the boundary conditions. In
from the constitutive law (2.34):
the case of the cantilever beam depicted in Fig. 2.4, these boundary conditions are given
by:
er = E ~~ = - E [u~ (s - ct) + u; ( s + ct)] = 91 ( s - ct) + .92 (s + ct) (2.48)
X(O) = 0
(2.54)
The axial t:itrcss in the hean1 is due to the superposition of tvvo non dispersive vvaves) { N(L) = ESdX(L) = 0
propagating towards directions s > 0 and s < 0 respectively. This stress can also be ds

()
1vritten as a function of velocities 'ili and ii,2 called particle velocities. Considering the Conditions (2.54) imply:
follovving relations:
8H1(s.t) '( )

l
'ii1s,t = at' =-CU1::3-Ct
(2.49) (2.55)
. 8u2(s, t) ,
H2(8, t) = iJt = +cu 2 (s +ct)

and noting that E = pc2 , the axial stresH can be ¥.rritten as: The eigenfrequencics arc solutions of Eq. (2.55b), that is:

(2.50) JfC
a= -pc(ur - ·u2) Wn = ( 2n -1 ) L· (2.56)
2

2.3.4 Eigenmodes of the bar The related eigenmodes are then given by:

An exa1nplc dealing vvith pile driving (§2.3.5) Virill noVir illustrate the difficulty to dcter-
(2.57)
n1inc the solution if multiple reflections take place during the propagation process. The
calculation of all these reflections, to express the general solution, becorr1es rapidly in1-
posHiblc. It is often inore suitable to try to solve the problem using the decornposition The eigenmodes defined by Eq. (2.57) fulfil the following condition and thus corre-
of the solution on the eigenrnodcs base. The cigenmodcs represent the periodic solution Hponcl to an orthogonal base:
of the hornogeneous equation (2.37). These solutions take the following form:
·L

v.(s, t) = X(s)eiwt (2.51) j 0

Xm(s)Xn(s)ds = 0 if m f n (2.58)
64 lD-vvavc propagation 2.3 Longitndinal vibrations of bca1ns 65

T11c solution in terms of eigcnrnodcs can also be written as the superposition of a wave a 0 (t)=-Asin(wt)
propagating in the positive direction and a v.'ave propagating in the negative direction.
Considering the trigono1netric identity, for the n/h mode 1 1.ve get:
3
b
(2.59) U}
U}

tJ'"
U}

The general solution of the equation of vibrations (2.36) can then be obtained b;y
superpoHition of all the cigenmodes: time

u(s, t) = L X.,,(8)qn(t) (2.60)

n=l Fig. 2.5: Example of dynarnic pile driving and variations of the axial stress.

=
and f(s,t) = LX,,(s)fn(t) (2.61)
n=l
T'he steel pile has a longitudinal \vavc velocity c = 5200n1/s and the shock duration
is T = O.Ols (that is w = 21T /T = 628.3rad/s).
Introducing expressions (2.60) and (2.61) into Eq. (2.36) and taking into account the
orthogonality property (2.58), qn(t) fulfils the following differential equation:
The impact generates an incident wave, propagating to\va.rch~ the pile end (s > 0).
ijn(t) + w;,qn(t) = f.,,(t) with n = 1, +cc (2.62) No wave can of course propagate tu1,vard direction s < 0 since there is no pile in this
direction. Considering Eq. (2.18)) the stress due to the incident \vave is as follo-1,vs:
2
with: fn = - - (' X,,(s)f(s, t)ds o-(s, t) = .91(s - ct) (2.64)
pSL j 0
At section s = 0) the expresr.;ion of the axial stress is:
The sirnilarity with the aru:Lly8is of systems \Vith a finite nurnber of degrees of freedorn
is obviouR: the dcco1nposition to eigenrr1odeti leads to the deter1nination of the solution of o-(O,t)=o-o(t)=g1(-ct) =? g1(-ct)=-Asin(~ct) (2 65)
ordinary 2nd order differential equatiorrn (Eq. (2.62)). The only difference is the infinite
number of equations corresponding to the infinite nurnbcr of degrees of freedorn of such fro1n \Vhich the stress for1nulation in the pile iH:
a continuous system.
o-(s, t) =-A sin [~(ct - sJ] (2.66)

2.3.5 Example 1: pile driving
The various results established in the previous sections Vi.rill be detailed in the simplified
case of pile driving. As already t>hovvn in Chapter 1, various -,,:vaves propagate in the pile
at each blovv. These waves 1,vill be analyzed in details hereafter.
Wave propagation in the pile
Consider a pile of length L undergoing, at ti1ne t = 0, an irnpact force at the top (falling
mass) 1 generally called an 1:m,pactor· (Fig. 2.5). This in1pact is supposed to induce the
folloV\ring axial stress:
o-o(t) = -Asin(wt) (2.63)
and the pile is not subjected to any other force. 'I'his asRumption irnplies that frictional
forces generated at the soil-pile interface are negligible (a rough description of reality). In
the previous relation) the sign / - 1 stan<ls for the fact that the axial stress is cornpressive.
At tin1c t = 11 /w, the irnpact has ended (Fig. 2.G) and the stress vvave propagates for
larger tirnes -,,vithont any changes along the pile. It reaches the pile end at tirne t = L/c.
Wave reflection at the pile end
Reaching the pile end (L = L/c), the incitlcnt -,,vave generate8 a refiected Vi.rave (u 2 (s) t),
o-2(s) t)). Its nature dependt> on the houndary condition at sections=]~. Let us consider
t1,oro extren1e cases:
• Fixed end: it correRponds to the case where the pile reaches a very stiff layer, such
as beclrock. The boundary condition is expressed through u(L, t) = 0 for all t.
• Pile end located in a very soft Tnediurn: the force acting in section s = L can be
considered to be nearly zero ; the boundary condition is expressed by JV(I_j, t) = 0
for au
l:.
66 lD-Vi'<-lVC propagation 2.:) Longitudinal vibrations of beams 67

t=O.Olls
...
nc
·· ........ ...
a(s,n/m)
2m ...............
i--~~~-"-~~~~-·
A
L=40m
...... ;

.-- ----- --- T

'
R)
.''

a(s,L/c) stresses (I,R) stresses (R,T)

Fig. 2.7: Stresses during the reflection phase - fixed end case (I incident
Fig. 2.6: Axial stress distribution in the pile (t S L/c). R : reflected ; T : total).

Fixed end: In the case of a fixed end, the Holution is straightforvvard: Free end: The free cn<l condition can be Virritten:

u(L,t) =0 * 1' 1 (L-ct) = -v.,(L+ct) (2.67) u(L,t) = 0 * 1i.1(L-et) =1i2(L+ct) ¢? ui(L-ct) = -uS(L+ct) (2. 71)

that is, after integration:

a(L, t) = -pc(u1 - u 2 ) = -2pc1i.1(L, t) (2.68)
u1 (L - ct) = 11.2(L +ct) (2. 72)
The pressure v,Tave o-1 (s, t) is reflected as a pressure 1vave a2(s, t) \Vith the sarne arnpli-
tude. During the reflection phase, that is L / c S t S L / c + 7r / w, the contributions of the 'I'he pressure incident wave o-1 (s 1 t) is reflected as a tentiile vvave v1lith the sa1ne an1-
t"\VO waves are cornbined. For larger tirnes 1 only the reflected vvavc exists; it propagates
plitude.
towards tlre pile cap where it Virill. undergo another reflection.
Sirnilarly to the case of a fixed end, the follcnving equations give the axial stress value
The following equations give the value of the axial stress in the pile until the reflected in the pile until the second reflection takes place at the pile cap:
wave reaches the pile cap:
[~(ct -

I
u1(s, t) =-A sin s)]
u1(s, t) =-A sin [~(ct - sJ] (2.73)
(2.69) . w wL

1 0"2(s, t) =-A sin [~(ct+

c

T'l1at is 1 superposing the contributions of both waves:

s) - 2w_Ll
c
<T2(s,t) = +Asm [-;:-(ct+s)-2-z-]

The '+' sign in the expression of the reflected wave stands for a tensile stress. Super-
posing the contributions of both ¥.raves:

O<t<L/c: u(s, t) = u1 (s, t) O<t<L/c: u(s, t) = u1 (s, t)

L/c<tSL/c+K/w: a(s,t) = 0"1(s,t) +u2(s,t) (2. 70) L/c < t S L/c+K/w: a(B,t) = u1(s,t)+u2(s,t) (2.74)
{ {
L/c+Ir/w < t S 2L/c: O"(s, t) = 0"2(s, t) L/c+Ir/w < t S 2L/c: a(B,t) =a2(s,t)

F'igure 2. 7 displays the stress diagram corresponding to the reflection phase. Figure 2.8 displays the stress diagram obtained during the reflection phase.
 1D-\vave propagation 2.3 Longitudinal vibrations of beams 69
68

1-- ---!l>~T
t=O.Olls
R--~

Fig. 2.9: Interface conditions between two media.

L=40m
Application to Hopkinson bar tests

A practical exa1nplc is proposed in Fig. 2.10 where the experimental device called Split
l Hopkinson Pressure Bar (SHPI3) is described (Klepaczko et a.I., 1991; Zhao et al., 1997).
R/
.. A cylin<lrical specirnen of knovvn di1nenHions) but v,rith unknown mechanical properties,
is int'ltalled betv,reen t\VO cylindrical steel bars 2 . The irnpact is irnposed at the end of a
e ~
"8 ~ bar; the ineasurement of the displaccn1cnt due to the incitlcnt and reflected 1,vaves allows
stress stress the estilnation of the wave velocity in the specirnen. The Split Hopkinson Pressure Bar
is pre8ented in details at the encl of this chapter and fast dynamic experin1cnts on soils
Fig. 2.8: Stresses during the reflection phase - free e11d case (I incident
using SHPB are alw discussed (§2.11).
R : reflected ; T : total).

specimen steel bar

2.3.6 Example 2: characterization of a heterogeneous bar 1-+"'Jc' i
R ~--i?£;'~----<• T
Propagation in a heterogeneous bar

The previous section presented simple exarnples of various boundary conditions. A 1nore
Fig. 2.10: Schematic of the Split Hopkinson Pressure Bar test.
con1plex exa1nple corresponds to the case where an incident Vi'ave propagates through an
interface bet"\vecn t"\vo rnedia "\Vith different properties (:Fig. 2.9). One Rhould no"\v enforce
the continuity of both displacements and forcer.; at the interface. l)enoting the variables
Denoting x = p1c1/ p2c2 the ratio of the properties between medium 1 (steel bar)
of mediun1 1 with index 1 and those of rnedium 2 Virith index 2, these conditions can be
and rnedium 2 (specimen), \Ve successively apply conditions (2.75). V\Tc thus obtain the
\Vritten under the following for1n: arnplitude ratio bet"\veen the ~·ave transmitted in the steel bar beyond the spccirnen and
the incident wave in the steel bar before the speci1nen:
• continuity of displacc1nent : u.1 = 'U2

• continuity of forces : Jvr1 = (2. 76)

]\TL
\Vhen the incident ¥.rave reaches the interface, a part of it8 energy is reflected and
carried by the reflected ¥.'ave propagating back\vard in inediurn 1. The cornplen1entary
energy crosses the interface and is carried by another "\Vave, called refracted or transn1itted
wave, propagating in n1cdiu1n 2. The variables related to rncdiun1 1 may be clecornposecl
into incident variables, denoted with index I, and refiected variables) denoted with in-
dex R. The variables related to the transmitted vvavc are written, Virith index 1'. The
continuity conditions thu8 write:
(2.75)
(1 + x) 2
This relation allows the estimation of paran1eter x as \¥ell as the ¥.'ave velocity c2 in
the specimen. It n1ay depend on the ratio bet¥.rcen the cross-sectional area8 of the bar
an<l the speci1nen if they are different. It is due to the expression of the norrnal forceH
(Eq. (2.34)), involving the cross-sectional area S, and their continuity at the inter.faces
(Eq. (2.75)b).
'l'he previonH exa1nple is a si1nplified version of the Split Hopkinson I3ar experirr1ent
coni:;idering linear elasticity. This test is indeed very general and actually perrnitH the
identification of dynan1ic constitutive lavv·s well beyond linear elasticity (Hee §2.11).
2
Stecl bars are inainly <lcdicatcd to uxperirnents on stiff materials such as metals or rocks whereas
softer bars (e.g. PJVIMA/Plexiglas) can be used for soft materials such as soils (see §2.11)
70 lD-\vavc propagation 2.4 Torsional vibrations of beams 71

2.3.7 Absorbing boundaries of infinite length.

Nurncri.cal con1putations for v,rave propagation problems, for instance in the framework Proof : Denoting u(L, t) the displaccn1cnt at section x = L of the rnodel depicted in
of the Finite Elerncnt 1-Iethod (see Chapter 4), raise the need for an accurate description Fig. 2.ll(b), Vire consider a beam of finite length connected to a damper.
of v,ca-vc radiation at infinity. Free or fixed boundaries originate wave reflections in their
vicinity. These reflections do not generally correspond to realistic phenornena and in- From Eq. (2.50)) the stress in this section can be written:
fiuence the solution accuracy. In the onc-dilnenr.;ional case 1 it is possible to control such <T(L,t) = -pc[u1(L,t)-u2(L,t)] (2.80)
reflections by introducing) at the model boundaries, specific ele1ncnts knovvn as absorb-
ing bo'undaries. 'Their role is to approximate the exact stress conditions at the 1nc<liun1
and, due to the effect of the da1nper) it is equal to:
boundaries (Fig. 2.lla).
rf(L, t) = -pcv.(L, t) (2.81)
Furthermore, the displacen1ent u(L, t) may also be derived as the mnn of the contributions
of both the incident Vi'avc and the reflected v-rave:
(a) u(L, t) = u 1 (L, t) + v. 2 (L, t) (2.82)
dimension of the model: L
The con1bination of Eqs (2.80) and (2.82) yield8:
V.1(L,t)-1i2(L,t) =·u.1(L,t) +·u2(L,t) (2.83)

which lead8 to U2 (L, t) = 0 and then (T 2 (L, t) = 0 and u 2 (L, t) = 0.

(b)

Fig. 2.11: Absorbing boundaries in a lD medium. 2.4 Torsional vibrations of beams

In the exarnple of Fig. 2.11.(a), the airn is to represent at sections= L the conditions
tha.t \Vould exist if the bar \Vas of infinite dimension. In this cross-section, due to the
finite dimension of the model, the displacernent (or stress) results frorn the superposition
of two waves, an incident one denoted \\rith index 1 and a refiected one denoted with
index 2. The goal is to remove the contribution of the reflected wave, that is to satisfy
the follo\ving relations:
2.4.1 Dynamic equilibrium
In this section, we focus on the torsional vibration of bearns 1 that is phenomena governed
by Eq. (2.25). Equation (2.25) was obtained as a part.icnlar case of the general equations
governing the vibration of beams. It can also be derived through the identification of the
torsional forces acting on a beam slice and considering its equilibriun1 (Fig.2.12).

u(L, t) = ·u1 (L, t) + u2(L, t) = u1(L, t)

{ a(L, t) = cr1 (L, t) + cr2 (L. t) = cr1 (L, t)
(2. 77) IJJ illr
ill__ _
\}1j
M +aM,d
-- s
The stresses at the mediurn free boundary can be Vilritten aH a function of the particle
velocity:
m,\J] t as
• incident wave: cr1(L, t) = -pc-it1(L, t) (2. 78)
• reflected wave: " 2 (L, t) = +pcui(L, t) (2.79) Fig. 2.12: Torsional equilibrium of a beam.

'"fo rernove the contribution of the reflected wave 1 it is then necessary to impose in
this croHs-section a stress equal to -pcU 1 (L, t). This streHs is precisely that developed ·This equilibriu1n equation iH written:
in a viscous darnpcr having a viscosity TJ = pc. Adding such a darnper at the beam end 8M1
removes the contribution of the reflected \vavc; this systen1 is then equivalent to a bearr1 M,+-'ds+(m,-m 1 )ds-M,=0 (2.84)
8s
 lD-vvave propagation 2.4 Ton.;ional vibrations of beamH
72

J<=:quation (2.88) rnay be \VTittcn as a first order hyperbolic differential system for
in whicl1 the inertial forces rn1 are as follovvs:
"\vhich the general solution is:
(2.85)
B(s, t) = B1(s - Vst) + e,(s + Vst) (2.90)
In addition to Eq. (2.84) 1 it iR necessary to consider the constitutive equation of the
beam for torsional excitations. 'fhis constitutive equation relates the torsional morr1ent
to tbe cross-section rotation. In the fra1nework of sn1all strains in linear elasticity, it is Solution (2.90) corresponds to the superposition of t\vo wavcfields, of arnplitudes (] 1
and fJ'2 1 propagating \vith velocity Vs to\vaxds directions 8 > 0 and 8 < 0 respectively.
1vrittcn:
M, = /tlc,-
ae (2.86)
DuTing propagation 1 the shapes of both waveficlds re1nain unchanged ; these \vavcs are
us said to be non-dispersive.
1
in vvhich µ.. is the shear modulus of the rnaterial 1 also known as L-arn6 s inocluluR. In
Eq. (2.86) and other subsequent equations of this section 1 index t is on1ittecl to make the 2.4.3 Stresses in the beam
presentation clearer; (] is the rotation around the normal to the cross-section 1 denoted 11-
Introducing relation (2.86) into Eq. (2.25), the equation governing the torsional vibrations The solution being known in tcr1ns of tlisplacerucnt 1 the shear stresses in the bearn are
obtained from the constitutive relation (2.86):
of beams is Lhe following:
0 2 e µ. 2 ae
rnt
(2.87)
at 2 p os 2 pJ T = µr ~~ = p.r 1e; (s - Vst) + e;(8 + Vst)] = g1(s - Vst) + 92(8 + Vst) (2.91)

T11e presentation rnade in this section is similar to that related to longitudinal vi-
In the previous equation 1 r denotes the radial distance to the torsion axis of the beam.
bratiorn since the etructure of Eqs (2.36) and (2.87) are identical, only the governing
para1neters are different.
The shear stresR in the beam is derived from the t>uperposition of t\vo non dispersive
\vavcs propagating tcnvards directions 8 > 0 and 8 < 0 respectively. This stress can also

.()
2.4.2 Homogeneous equation be expressed a.s a function of velocities iJ L and 82 known as the ang11Jar particle velocities.
Taking into account the follcnving relations:
Assurning the linear denRity of moment Tnt to be zero, the homogencour:; equation of
vibratior1 can be written: 8B 1 (s,t) ,

l
ii(s: t) - vffe" (s, t) = 0 (2.88) B1 s, t = iJt = -Vs0 1(8 - Vst)
(2.92)
in \;v~hich the shear wa~,1e velocity Vs iR introduced 3
. It is defined as follo\vs: · ae2(s, t) ,
B2(s,t)= at =+Vs0 2 (s+Vst)
(0.
Vs= y-p (2.89)
and recalling that G = pV§, the shear stress can be \vrittcn:

-pV.sr(B1 - li2) (2.93)

(j and e
11 denote the second order partial derivatives vvith respect to tirr1c and space T =

(resp.).

Typical values of the shear wave velocity are proposed hereafter for various civil
engineering materials: 2.4.4 Eigenmodes of the beam
• steel: Vs= 3100 m/s, The cigcnmodes correspond to the periodic solutions of Eq. (2.88). These solutions are
expressed under the forrn:
• concrete: Vs= 2200 rn/s, O(s, t) = El(s)eiwt (2.94)

• sand: Vs= 200-400 rn/s. Introducing this relation into Eq. (2.88), function C-l(s) must satisfy:
3 The shear 'Wave velocity V, \vill be tbe same in 2D/3D (Chapter :~) since the lateral boundary
5
conditions in a beam do not influence the solution for pure torsion or pure shear (see §2.3)
V,2 El"(s) + w2 Ecl(s) = 0 (2.95)
74 lD-\vavc propagation 2.5 Shear vibrations of bcan1s 75

yielding the follovving solutioru.;: 2.5 Shear vibrations of beams

8(s) =A cos ( ;, s) + B sin ( ;, s) (2.96) 2.5.1 Bending-shear vibrations

In this section, the bending-shear vibrations of beams arc firstly considered. These
'fhe integration of constants A and B is performed using the boundary conditions. In phenornena are governed by Eqs (2.27) and (2.28). Such vibrations are due to rnotions
the case of the cantilever bearn depicted in Fig.2.4, these boundary conditions are given of the neutral fiber of the plane cross-section as well as rotations of the section around
by: axes located in its plane. Equations (2.27) and (2.28) are vectorial equations which
8(0) = () can, by an adequate choice of the syste1n of orthogonal axes located in the plane of the
(2 97) croHs-scction, be replaced by four scalar equations. In the folluvving, the analyHis Vl.'ill be
{ 2v1, (L) = 11.J-d8_-(_L_) = o rnacle on tvvo equations only (corresponding to one of the two aJCes (e.g. axis 1)). Indices
ds
1..vill be disregarded to si1nplify the notation. The two equations governing the dynamic
Conditions (2.97) lead to:
equilibriun1 are thus 'hTitten as follovvs:
8(0) = O=?A= O
iW
(2.98) - + f-pSii= 0 (2.105)
{ 8'(L) = 0 =?cos (wL) = O
v~
OS

(2.106)
The eigenfrequencics arc solutions of Eq. (2.98b), that is:
KY; \vhere the shear force V, the linear density of tranr.;verse external forceR f 1 and the
Wn = ( 2n - 1) £ (2.99) transverse displacc1nent u. correspond to the cornponents along direction 1 ; the bending
2
n101nent l'vfb and the linear density of external moments m correspond to the components
The associated eigenrnodes are given by: along direction 2.

8n(s) =sin [(2n - 1) ;n (2.100)

To fully detern1ine the solution of the proble111 1 it is necessary, in addition to equi-
librium equations (2.27) am] (2.28), to consider the constitutive laws of the beam for
The eigenmodes defined by Eq. (2.100) fulfil the following relation and thus form an
bending and shear loadi.ngH. These conr.;titutive laws relate the shear force and the bend-
orthogonal base: ing moment to the r.;hear strain and rotation w. ConRidering linear elasticity and small
(2.101) strains, they are \Vritten as folknvs:

The general solution of the equation of vibrations (2.87) can thns be obtained through V = /LS' o'U - w) (2.107)
the superposition of the eigerunodes:
( -,-
iJs

aw
8(s, t) = L 8n(s)qn(t) (2.102) NI,, = EI -
0s
(2.108)
n=l
vvhere µ, is the shear rnodulus of the mediurn, S' and I the shear area of the cross
00
section and the bending inertia of the beam. To derive the differential equation governing
and m,(s, t) = L 8n(s)m1n(t) (2.103) the transverse displacement of the neutral fiber, denoted ll, it is necessary to elirninate
n=l
quantities V, Mb and w from Eqs (2.27), (2.28), (2.107) and (2.108).
Introducing expressions (2.102) and (2.103) into Eq. (2.87) and taking into account
the orthogonality property (2.101), qn(t) satisfies the differential equation:
2.5.2 From bending-shear to pure shear
iin(t) +w?,qn(t) = m1n(t) with n = 1, +oo (2 104) 'fo analyze shear '°'raves in soils and rocks 1 it is generally possible to consider the local

with: Tntn
2-
= -.
pJL o
1·t 0ln(s)m,(s, t)dt rotation (as well as the bending moment) negligible. Equation (2.107) may thus be
wTitten under the following for1n:
o'U (2.109)
V=µS-
These results will be used in section 2.10 for the analysis of the resonant column test. os
76 lD- \vavc propagation 2.6 I3chaviour of dissipative 1nedia n
4
The transverse vibrations of a cylindrical bea1n are finally expressed as :

[J'u 11 !Pu f (2.113)

(2.110)
i)t2 - p [)32 pS
Vl.'here a(t) and c(t) are specific con1ponents of the stress and strain tensors 5, Cr(t) and
It leads to a homogeneous equation sirnilar to Eq (2.36) considering the shear vvave
i(t) their tirr1e derivatives.
velocity Vo, clcfincd as:
Vs= 1¥ (2.111) Functions 'ljJ(t) and qJ(t) can be estirr1ated cxperin1entally:

A.xial and transverse \vavcs are different Rince axial Viravcs generates motions along the
bearn axis vvhereas traru;;versc waves genernteR rr1otions perpendicular to the axis. \Ve
vvill discuss the significance of the differences betVireen these two \vave types in the next
chapter (2D/3D propagation).
• '¢(t) is called the relaxation function and is defined as the stress rer.;ponse for a
Heaviside type strain variation,
• </>(t) is called the creep .function and is defined as the strain response for a Heaviside
type stress variation.

Equations (2.112) and (2.ll~i) indicate that, for viscoelastic n1aterials, the response
2.6 Behaviour of dissipative media depends on the loading history \:Vhereas in linear elasticity, there is not such a dependency.
This phenon1enon is often called the m,ernory effect.
2.6.1 Dissipative phenomena
Cyclic, vibratory and dyna1nic problerns have a com1non feature since they generally Causality principle
involve energy dissipation phenomena. In soils and rocks, the energy dissipation for a
Equation (2.112) (as well rm Eq. (2.113)) aoournes that the stress at time t, denotecl
dynamic or vibratory loading rnay someti1nes be very large (Hardin, 1965; I{okusho,
O"(t), linearly depends on the strain history c(T) until time t (Bonrhi6 et al., 1987).
1980; Seed et a.l., 1986; Stoll, 1979). For wave propagation problems, the attenuation
It consequently involves si1nultaneously the superposition principle and the causality
phenornena generally appear under both foll.owing forrns:
principle. Causality implies that, in Eq. (2.112), the stre88 O"(t) is not infiuencecl by the
• geometrical damping due to the wavefront expansion (effect of radiation) 1 future values of strain c(T) for T > t. Equation (2.112) can then be rc\:vritten under the
follo\ving forrn:
• material or intrinsic attenuation due to energy dissipation inechanisrns in the +=
rnaterial itself. !7(l) = j-= </J(t - T)E(T)dT (2.114)
Only rnaterial attenuation is cliscusscd in this section (geornetrical da1nping \Vill be pre-
Rented in the next chapter). A convenient v..ray to model attenuation is to consider linear where 1 due to the causality principle, v..re aRRnrue that:
viscoelastic constitutive ]a\vs.
r/;(x) = 0 for x < 0 (2115)

2.6.2 Viscoelastic behaviour Using integration by parts, Eq. (2.114) can be rewritten under the follo\:ving for1n:
Principles of viscoelasticity +=
For a linear isotropic viscoelastic material 1 the stress-strain relation is derived fro1n the
Boltzn1ann principle (Bonrbi6 et al., 1987; Fung and Tong, 2001). In the one-dimensional
!7(t) =
1-,x rn(t - T)E(T)dT (2.116)

case, it is Vflitten under one of the follovving forrns: v..rhere rn(t) standH for the stress rer.;ponse to a strain variation :follo\ving a Dirac function
(Fig. 2.13), leading to the expression:
(2.112)
rn(t) = d'ljJ(t) (2.117)
4For a non-constant cross-sectional area. S(s), the equation of n1otion has a different fonn: dt

D
2

ot2
u__pS(s) _iJ_ (s(s) au.)
µ
OS as
~ _f_
pS(s)
5 0uc inay consider traction/cornpression components for longitudinal vibrat.ions or shear coniponcnLs
for transverse vibrations.
78 lD- -,,vavc propagation 2.6 Behaviour of dissipative 1ncdia 79

Characterization of m( t) and M (w)

The function m(t) (Eq.(2.118)) and the complex modulus M(w) (Eq.(2.121)) characteri"e
the viscoelastic behaviour of the rnaterial (Fig.2.13). Various paran1eters n1ay be used
to identify these functions:·
om
• the unrelaxed (elastic) rnodulus6 rrio which gives the instantaneous elastic response
of the n1atcrial, that is, in the tirnc <lon1ain:
m~ mo= Jim m(t) (2.123)
t--+0

or in the Hpectral <lo1nain:

mo= Jim M(w) (2.124)
w--+=
t=O time
• the relaxed m,odnlHs m= vvhich gives the responHe for the long tern1 equilibrium of
Fig. 2.13: Typical m(t) function for a viscoelastic material. the rru-1terial 1 that is, in the time dornain:

m= = Jim m(t) (2125)

t~=
Constitutive law in the spectral domain
or in the spectral dornain:
From a rnathcn1atical point of vieVir, Eq. (2.116) represents a convolution in time bet1,vcen m= =Jim M(w) (2.126)
w~o
rn and c. In the following) this relation vvill be vvritten under the following condensed
forn1:
Quality factor and attenuation
u(t) = m(t) * E(t) (2.118)
ConHidering tin1e harmonic variations for the mechanical variables: The quality factor Q(w) is derived from the complex modulus J'vf(w):

u(t) = u*(w) exp(iwt) and c(t) = c'(w) exp(iwt) (2.119) Q(w) = Mn(w) (2.127)
Mr(w)
equation (2.118) inay be written in the frequency dornain: where l11[R and 1VI1 correspond to the real and irnaginary parts of the eornplex rnodulus.
O"*(w) = M(w)E*(w) (2.120)
The inverse of the quality factor Q- 1 (w) is called attenuation and characterizes the
where u*(w) and E'(w) arc the Fourier transforms of u(t) and c(t), M(w) is the complex da.rnping properties of viscoelastic 1natcrials.
viscoelastic modulus, depending on frequency.
2.6.3 Rheological models
Starting from Eq.(2.117), 'ljJ(t) and M(w) may be related ao:
Constitutive laws
!(·)-
t - :F_ 1
"ljJ
{NI(w)}
.
1.w
(2.121) Rheological 1nodels are often used to describe the viscoelastic constitutive laws of dissipa-
tive iuaterials. These allow the description of various types of attenuation-frequency de-
v;rhere :F- 1 (f) is the inverse Fourier transforru of function f. pendencies (Bourbie et aL, 1987; Emmerich and Korn, 1987; Semblat, 1997; Semblat ancl
Luong, 1998). As depicted in Fig.2.14, the rheological rnodels considered for viscoelastic
In the case of an isotropic three-dimenRional rncdium, the generalization of Eq.(2.114) materials arc composed of sin1ple rheological cells: the spring and the dashpoL Figure 2.15
is straightforward (Bourbie ct al., 1987): displayR several classical viscoelastic models considering uniaxial traction/ cornpression (CJ
being the uniaxial tcnsilc/co1npressive stress and£ the corresponding unim::ial strain).

(2.122) 6 In the field of geophysics, the uurcl1:txcd rnodulus is often denoted 1Vfu whereas the relaxed rnodulus
is denoted 1Vln. The latter may be confusing when considering the real and imaginary parts of tlw
leading to the definition of cornplcx Laine parameters. coniplex rnodulus. It. is thus avoided herein.
 l

2.6 Behaviour of dissipative media 81

80 lD-vvave propagation

Q'

~
• Maxwell 0.5
Spring Dashpot
Maxwell
a = Es .§ 0.4
1;l
g 0.3 Kelvin
Fig. 2.14: Elementary rheological models in viscoelasticity. E (1) Voigt
• Kelvin·Voigt ~ ~
0.2

The Maxwell model (Fig.2.15) is composed of a spring (modulus E) and a dashpot

(viticosity 77) in series. For this inodel, considering cu1nulativc uniaxial stresses in each
[ M(co)=E+ico11 J 1J
0.1

0 -
-- --
cell, the constitutive la\·\' writes as Tollovvs: E E" Q'
I) . .
•Zener ~ 0.5 Zener
u + Ea=r7c (2.128)
M(co) = E(Ev +icori) 1J
The Kelvin-Voigt model (Fig.2.15) is composed of a spring and dashpot associated
in parallel (i.e. cumulative axial strains). For this model, the constitutive laVi.r is of the
E +Ev +icoY)
E
\\ -
, ___ .... ................ .
follovving forrn: generalized
• Generalized
(2.129) 0.1 Maxwell
Maxwell
OL.._J__L__J~J__J__L__JL_J__L._

The Zener 1nodel (Fig.2.15) , also called the three pararneters rnodel or the standard 0 1.0 2.0 3.0 4.0
frequency (Hz)
solid m,odel (Fu.ng and Tong, 2001), is co1nposed of a Kelvin-Voigt cell associated in serieti
to a spring 7 . For this mo<lel 1 the constitutive lavv is as follo-\vs:
Fig. 2.15: Various rheological models: expressions of the complex moduli
O" + ., 7) .
O" = E Ev ( C + -17- E·) (2130) Jvl(w) and associated attenuation curves q- 1 .
E+Ev E+Ev Ev
This relationship can be vvritten under the general forrn 8 allowing the determination of the general expression of the co1nplex rnodulus ]\If ( w)

(2 131)
1 + iu)Tc
M(w) = . rn= thus: IM(w)I= (2.133)
\\-There rn= is the relaxed n1odulus, r er is the stress relaxation tin1e under conRtant strain 1 + r,WTcr
and Tc is the strain relaxation time under contitant stret>s.
As shovvn in Fig.2.15, the con1plex modulus JJI(w) for these rheological rnodels is then:
Considering uniaxial shear, all previous constitutive lavvs can be \\.Titten replacing
Maxwell: JVI(w) = iEw77 (2.134)
the unia.xial tensile/con1pressive stress rr by the uniaxial shear stress T, and the uniaxia.l E+1.wr7
longitudinal strain c by the uniaxial shear strain 1 · 1'he tihear viscoelastic properties are
thus introduced (e.g. shear tnodulus µinstead of Young't> rnodulus E). Kelvin-Voigt : M(w) ~ E + iwry (2.135)

Zener : M(w) = E(Ev + iw-r7) (2.136)

Corr1plex modulus E +Ev +iw77
Considering a e+fo.1t tilne har1nonic dependence, Eq.(2.131) 1nay be written in the fre- For the generaliL'ied l\1axv.rell rnodcl, the expression of 1Vl(w) is a bit inore cornplex:
quency do1nain:
(2.132) Ew 2 r7[ + i [E 2w(ry1+7)2) + w 377f-r12]
M (w ) = --"---~~~~~-'--___c'-'---" (2.137)
E2 + w217I
7 H rnay also be built \vith a l\1a)nvell cell associated in parallel to a spring. Tt leads to a si1nilar
constitutive laV\r. The latter rnoclel vvill be also t>tu<liecl in Chapter 4 since it leads to a very convenient
8 The notabons co1rnidcre<l herein arc si1nilar to that used in the field of geophysics but arc differe11t form of the quality factor (Scmblat, 1997).
fron1 that considered in the held of continuum rncchanics (Fung and Tong, 2001), tbat is generally:
a+ Te&= m=(t: +Teri:)
 11)-\vavc propagation 2.6 Behaviour of dir::>sipativc media 83
82

J-\t large times (resp. low frequencies) 1 the asy1nptotic value of the con1plex rnodulus Quality factors and attenuation
for the Zener 1nodcl i.s interesting: Considering the definition of the quality factor (Eq.(2.127)), the expressions of attenua·
tion for the linear viscoelastic models depicted in Fig.2.15 are the following:
] +iwT0 ) (2.138)
Jim M(w) = lirn ( . rn=
w-o W·-.-+O 1 + 1WT(}
-1 E -1 (w) =Wt/ (2.146)
Q,wax(w) = - and QKcl E
WT)
1 1 1
with: - = - + -
rn= E Ev
In the Zener rnodel, the con1plcx 1nodulus J\l[(w) tends to the relaxed rnodulus rrioo (2.147)
at large times and the dashpot has then no effect on the response.

At short tin1cs (resp. high frequencies), the asy1nptotic value of the cornplex rnodulus -1 ( )- E(T/1+T72)~+7/2 (2.148)
QG1\1 w - 2 E w
for the Zener rnodel is as follows: 7/1 w
for the J'vlax1.vell, the Kelvin-Voigt 1 the Zener and the generalized 1\.tlaxvvcll n1odels, re-
Jim M(w) = lim
l + iwT.
. · m=
) Ts
= -m= = E =mo (2.139) spectively.
w--++= l.t.1----7+= ( 1+ iWTCT T er

For the Zener n1odel 1 the complex rnodulus l11l(w) tends to the unrelaxcd (instanta- From Fig.2.15, these rheological rnodels obviously have various attenuation-frequency
neous) modulus rn 0 at short times and the v.rhole Kelvin cell has then no effect on the dependencies. The J\1a.xvvell and generalized lvlax\vell rnodcls do not involve instanta-
neour::> elasticity (infinite attenuation at zero frequency). The Zener and the generalized
response (see also Fig.2.13).
1\tfaxwell rr1o<lcls sho1.v a band-cut and hand-pass effects respectively. Sorr1e analytical
wave propagation sirnulations \vill be considered in the follo1.ving for these various rr1odcls.
Relaxation and creep functions
For a strain loading E(t) defined as a step function H(t) (i.e. Heaviside in time), the As in the previous subsection, the expression of the quality factor can also be deter-
solution of Eqs (2.128) to (2.130) allows tlic determination of the relaxation functions as 1nined under a general for1n derived frorn Eq.(2.133):
follows (Bourbie et al., 1987):
(2.149)
Maxwell: ?j!(t) = Ee-fi'H(t) (2.140)

Kelvin· Voigt: ?jJ(t) = r1b(t) + EH(t) (2.141) The rnaxhnum value of Q- 1 corresponds to the largest attenuation and, for the Zener's
inodel (Ts#- 0 ct Tu#- 0) 1 it iH reached at frequency Wrn:
Zener: 1/;(t) =m= [1-(1- ~:) e-"~] H(t) (2.142)
1
Wm=--- (2.150)
where b(t) is a Dirac function being always zero except fort= 0. ~

Similarly, for stress loading J(t) defined as a. step function H(t), the solution of and leads to: (2.151)
Eqs (2.128) to (2.130) allows the determination of the creep functions under the following
forrn: Defining the normalized frequency [1. = w /wm, Fig.2.16 depicts the variations of both
Maxwell: ¢(t) = + ~t) H(t) (.1.E T/
(2.143) Q- 1 (0) and IM(rl) I considering a semi·logarithmic scale 9 . Attenuation Q- 1 (0) reaches
its maximum at Wm (band·cv.t effect), whereas I M(rl) I tends tom= at low frequencies
(vvhich iH in good a.grce1ncnt with the definition of the complex modulus) and tends to
Kelvin· Voigt: ¢(t) = ~ ( 1- e-fit) H(t) (2.144)
;"' m= = m,o for higher frequencie8.

Zener: ¢(t) =ml= [1-(1- ~) e-*] H(t) (2.145) 9The Q- 1(U) curve is the same as in Fig.2.15 but, considering a se1ni-logarit,hmic scale (inst<Jad of a
linear one), it yields a syrnrnetric attenuation peak.
 ] D-wave propagation 2. 7 \i\lave propagation in viscoelastic rnedia 85
84

2. 7.3 Wave equation in viscoelastic media

The following differential equation in terr11.s of displacement is obtained:
mo E! 2 1i'(s, w) pw 2 ,
(2.156)

~)Meo>[]-- lvl-.. (-)11.

"
vs 2 = -
w (s,w)
.

The solutions of this equation are of the follovving for1n (Aki and Richards, 1980):
I
I v.'(s,w) = Uo(w) exp [-ik'(w)s] (2.157)
I
ffiITT /
------ 1,.vhere k* is a co1nplex number such as:
2
(k'(w)) 2 = _P(<)__ (2.158)
M(w)
0.001 0.01 0.1 10 100 1000
normalized frequency Q From Eq.(2.157), the displacement v.(s, t) solution of Eq.(2.152) is finally written as:

v.(s, t) = Uo(w) cxp[i(wt - k"(w)s)] (2.159)

Fig. 2.16: Attenuation Q- 1 (!1) and complex modulus 1 lvI(!1) I vs normalized
frequency ll = w /Wm for the Zener model. For a non-harn1onic wave, the follo-\ving expression it-> easily obtained Jo:

2. 7 Wave propagation in viscoelastic media v.(s, t) = -1

27f -=
1+= Uo(w) exp[i(wt - k'(w)s)] dw (2.160)

2.7.1 Dynamic equilibrium

The equilibriurn of a 11niaxial 1nediurn under longitudinal loadings writes:
2. 7.4 Complex wavenumber

\7 ) = Da(s, t) = a'v.(s, t) k*(w) fully characterizes the propagation process and its dependency on the frequency
(_.g Tl (2.152) governs attenuation. It is called the complex 1JJa1Jen'urnber. J)enoting the real and in1agi-
as P at 2
nary parts of k'(w) under the form k'(w) = k(w) - frt(w), it is possible to derive them
where \7 is the grndient vector (\7 1 = E!/E!:cj), [[the Cauchy stress tensor and 1]. the
as functions of the co1nplex inodulus Jv[:
normal to the cross-section.

This equation can be revvritten in the frequenc:y· dornain under the folluvving form:
p(IM(w) I +MR(w))
k(w) = w (2.161)
2 llvf(w)l2
aCT'.(s,w) = -pw2v.'(s,w) (2.153)
Ds p(IM(w) I -MR(w))
a(w) =w (2.162)
where o-' (s, w) arnl u' (s, w) are the Fourier trnnsforms of u(s, t) and u( s, t). 2llvI(w)12

2.7.2 Viscoelastic behaviour where IM I is the modulus of the complex variable lvI, that is IlvI I= Jlvl'f, + Mj (with
lvfn an<l 1111 the real and irnaginary parts of the complex rr1odulus).
For a lir1ear viscoelastic rnediurn, the constitutive la\:\r (2.120) is combined \vith Eq.(2.153):
a<r'(s w) D 1 °For a e-iwt hannonic dependence, the expression would have been found as:
, ' = c- [}vl(w)c'(s,w)] = -pw 2 u'(s,w) (2.154)
as as +=
yielding:
2
u(8, t) ~ _1_
2,,. j-= Uo(w) exp[i(k'(w).; - wt)] dw
E!a'(s,w) = "'··f(w)D u'(s,w) 2 '(
-~~~=-pw1t·s,w ) (2.155)
Ds E!s 2 Both vvaves propagate toward the positive s since they are of the forn1 'U(8 - ct) defined in Fig.2.3.
 lD-Vi.rave propagation 2. 7 '\Alave propagation in viscoelastic rncdia 87
86

·The real part kR. = k corresponds to the real v;avenn'mber (characterizing a progressive Denoting A = c/ f the wavelength, it yields:
wave) and the imaginary part k; = o: to the attenuation factor. k* (w) can then be derived
under the following forrn: u'(s,w) = Uo(w)exp[-ik(w)s]exp (
s)
-QA
1r (2.170)
w (2.163)
k'(w) = c(w) - ia(w)
It giveR a sin1ple expression shu\ving the influence of attenuation Q- 1 on the spatial
where w is the frequency1 c(u,.·) the phase velocity and o:(w) the attenuation factor. These decay of the wavcficld.
tvvo terrns (phase velocity, attenuation ratio) may depend on frequency. The dependence
of the phase velocity c(w) on frequency corresponds to the physical dispersion in the 2.7.6 Dispersion laws
rnediu1n.
Wave propagation in dispersive media
The solution of the >Nave equation in viscoelastic rneclia may finally be \vritten as In dispersive inedia) the \Vave velocity is frequency dependent (Aki ancl R,ichards, 1980;
followR: Bourbie et al. 1 1987; Lighthill, 1965). To fully characterize \vave propagation in dispersive
u~(s, w) = exp(-ik~(w)s) = exp(-ik(w)s) exp(-cx(w)s) (2.164) 1ncdia, two types of velocities are defined:
the a(w) variable thus characterizes the spatial decay of the wave. • the group velocity is the velocity at 'which energy is carried. The group velocity "V9
is expressed under the follu\ving :form:
VVhen the viscous effects arc small (1VI1 << JvIR.), it is pos::;ible to approxirnate the
phase velocity, f.::;q.(2.161), as a function of the real part of the co1nplex modulus as (2.171)
follows:
c(w) = JMR~w)) (2.165)
that is under the forn1 of the ratio between a frequency variation and a vvavenumber
variation.

This expression is siinilar to the ¥.'ave velocity :for an elastic n1ediu1n. • the phase velocity is the ·uclocity at 'Which the points of constant phase propagate,
that is, generally, the velocity at which the peaks of the signal propagate. The
1 phase velocity Vq:. has the follo-vving form:
2.7.5 Relationship between a and Q-
w
The atten11ation Q- 1 depending on the complex n1odulus under the follovving forn1: Vq,= - (2.172)
k

(2.166) that is the ratio bet\veen a frequency antl a vvavenurnbcr.

As depicted in Fig.2.17 1 the phase velocity at a given frequency n1ay be estilnated
the relationship hetvveen a; and Q can be written: fro1n a signal filtered around this frequency (individual peaks). For the group velocity,
the envelope of the signal characteriLJes the velocity at V1d1ich the energy is propagated

Q=
w JJ1 + Q1, -1
(2.167)
(Fig.2.17)

Wave dispersion in viscoelastic rnedia

v'2c J1+J,
For viscoelastic models, the deterrnination of the phase velocity c(w) as a function of
In the case of "\veak attenuation (Q- 1
<< 1) 1 the relation is simplified as: frequency is sin1ple since the con1plex \Vave nurnber k*(w) = k(w)-in(r..v•) can be directly
expressed as a function of the cornplex rnodulus characterizing the rheological 1nodel
2nc nc
Q"_ 1 ~-=- (2.168) considered.
- W Kf
Starting from the expressions given in Fig.2.15 and Eq.(2.163), it is possible to deter-
'I'hc solution ,u*(s,r..v•) n1ay then be expressed as a function of Q: mine the attenuation fa.ctor and the phase velocity as functions of frequency, that is o:(w)
and c(w). Figure 2.18 gives the variations of the phase velocity for various viscoelastic
u'(s, w) = Uo(w) exp [-ik(w)s] exp ( - Irf
Qc s ) (2.169) models (Maxwell, Kelvin and Zener):
 11)-\vavc propagation 2.8 Exa1nples of propagation in viscoelastic rncdia 89
88

I
I
I
I
I
I
•, I
I
. _, I
I

,.
·- - Zener
I
I
I
I

I
'' I
/
Kelvin/ /
·-' /
/

4.0 5.0 6.0 7.0 8.0 9.0 ,/

time (ms) ""
-- --
- - - - - - - - - :: :;~~:::::: ::_ -- -... '" -- -- ... "" -- __ ]\l[_".'~:Y.<:~l__ -- --- -- --
Fig. 2.17: Determination of the phase and group velocities from a filtered
signal (solid) and its envelope (dotted).
0.01 0.1 1.0 10 100
• for the lvfa.xwell rnodel, the phase velocity decreases and reaches an asyrnptotic frequency
value c= = j1f;, Fig. 2.18: Phase velocity as a functior1 of frequency for various viscoelastic
• for the Kelvin rnodel. the phase velocity increases rapidly starting from a rninimum models (Maxwell, Kelvin-Voigt and Zener)-
value co = j1f;, ,
'fhe si1nulations displayed in Fig.2.19 clearly illustrate in tilne domain the previous
• for the Zener model, the phase velocity progressively increases fro1n a minirnu1n rernarkR on -,,vave propagation. In a lVIa.xwcll linea.r viscoelastic mediurn (left):
value co =
vr:;;;;;;_ to a rnaxirnunr value c= = vr;;;;;,
p p
• the shape of the triangular signal is unchanged during propagation: the discontinu-
·I'he variations of the phase velocity arc strongly dependent on the viscoelastic para1neters ities (or high frequency components) are slightly attenuated in a ~\.1axvvell mediurn
of the lnodel. rI'his dependency is URUally called material disper::rion. (vireak attenuation) Fig.2.15) 1

Examples of propagation in viscoelastic media
2.8
2.8.l Example 1 : propagation of a triangular signal
Before co11sidering actual signals) a synthetic signal n1ay be chosen to illuRtrate the
influence of the properties of the rheological models. It is thus poHsible to investigate the
way the mc<liurn transrnits slovv or fast amplitude variations (Dero frequency or infinite
frequency). Considering a triarigular Hignal and using Eq. (2.160)) the curves displayed
in Fig.2.19 depict the evolution of the initial shape of the wave during the propagation
process.
• the peak arnplitude decreases rapidly during propagation: the slov; variation.s (or
low frequency components) are strongly attenuated in a :rvia.xvvcll n1ediu1n (strong
attenuation, Fig.2.15).
For a Kelvin-Voigt mediu1n 1 the propagation features arc significantly different (Fig.2.19 1
right):
• the peak shape 1:s altered during propagation: the fast variations (or high frequency
components) are quickly atten'uated in a Kelvin-Voigt mediun1 1
• the peak amplitude slowly decreases during propagation: the slov1 1;ariations (or lo'W
frequency components) are 1veakly attenuated in a Kelvin-Voigt mediuni.

Such differences between both rheological inodels are due to the fact that the attenu-
The signals are referenced frorn an arbitrfily origin, the first signal representing the
ation is proportional to frequency for the I{elvin model v..rhereas 1 for the 1.!Iax\vell rnodel,
excitation at a reference distance do.
it is inversely proportional (see Fig.2.15). The aHyrnptotic behaviour of attenuation is
90 1D-1vavc propagation 2.9 Other linear and nonlinear rnodels 91

thus very different in both caseH. VVe n1ay conclude that the viscoela8tic properties of the
n1edinrn as vvell as the frequency content of the propagating vvave significantly influence
the propagation features in tcrn1s of attenuation and spatial decay.
Maxwell Kelvin-Voigt
2.8.2 Example 2 : propagation of a seismic wave.
V\Te no1v cont>idcr seismic Hignals recorded in centrifuge tests (Sernblat and Luong, 1998).
The wavcfield is nearly plane and 111.ainly corrcspondH to longitudinal v.ravcs. The onc-
di1nensional approximation is acceptable and the propagation process iR the soil mar.;H
can be analyzed thankH to the rheological models presented earlier.

·To simulate the propagation in the soil mass: one consider8 the acceleration signal
n1easured at point Ai ·virith absci8sa Xi, denoted ameas(x,; 1 t) 1 and one cornputcs the Rignal
propagated to point Aj at abscissa :rj, denoted a 8 i 1n(Xj, t). We first write the relationship
bctvveen the spectral cornponents, a~cas(xi,w) and a:im(Xj,u..,'), of the8e acceleration
signals using expression (2.157):
(2.173)
Fron1 expression (2.173), the simulated acceleration asirn(:rj, t) is \.Vritten as follo\.vR:

(2.174)

I,'igure 2.20 compares the signaltJ recorded at different points (left) to the simulations
perfor1ned with Kelvin-Voigt (center) and lviax1,vell n1odels (right). The viscoelastic si1n-
ulations are in good agreernent with the experimental resultR. The Kelvin model never-
time time thele88 leads to overda1nped higher frequency co1nponents. Simulations made vvith the
Zener rnodel are very close to those computed vvith the 1\!Iax1,vell rnodcl (Semblat and
Fig. 2.19: Propagation of a triangular signal in a linear viscoelastic rnedil1rn:
Luong, 1998).
Maxwell model (left) and Kelvin-Voigt model (right).

2.9 Other linear and nonlinear models

Starting frorn the results of the previous Rections, it is possible to develop more cornplex
viscoelastic models with more realistic variations for the attenuation Q- -i . "!"'he ).nj_;ercst
of models built in such a vvay if; first that they arc simple and secondly that tlley fulfil
the fundarncntal aRsurnptions of continuum rnechanics (ca.usality among others). Sorne
of them are presented in the follovving.

Jlvfany constitutive ]av.rs have also been proposed in the framevvork of elastoplasticity
(Bonilla ct al., 2005; Critescu, 1974; Tai et al., 1995; Kramer, 1996; Lade, 1977; Lorct
et al.: 1997; Jlvfatasovic and Vucetic, 1995; l\!lellal and 1\tfodaressi, 1998; Prevost, 1985;
Kelvin Pyke, 1979; To-\vhata, 2008). Since Vire rnainly focus on Vi'ave propagation in the linear
range, such models are not discussed herein.
0.0 1.0 2.0 3.0 0.0 1.0 2.0 3.0 0.0 1.0 2.0
time time time 2.9.1 Constant Q (CQ) model
Fig. 2.20: Analytical simulations from centrifuge experiments (left) for two Kjartansson (1979) proposed a frequency constant Q n1odel having a complex n1odulus
different rheological models (Maxwell: center, Kelvin: right). of the following form:

M(w) ~Mo (:r~ (2.175)

 lD-v,rave propagation 2.9 ()thcr linear and nonlinear models 93
92

where Jvf0 is the rnoduh1s of A1(w) at the reference frequency wo.

Ql (%)
1.0
The quality factor is thus given as (Bourbie et al., 1987):

(2.176) 0.8

2.9.2 Frequency dependent Q model 0.6

J'viiiller (1983) proposed a frequency dependent Q rnodel expressed in the forrn of a power
0.4
lavv:
(2.177)
0.2
"'\vl1cre ''/is an exponent chosen between -1 and +land the related cornplex 111odulus is
given by (Muller, 1983): w' 1o·' 1 o' 1o'
1o" 1
frequency (Hz)
(2.178)
Fig. 2.21: Variation of attenuation Q- 1 with frequency for the NCQ model
("A5") of Liu et al. (1976).
2.9.3 Nearly Constant Q (NCQ) model
Since the attenuation rr1ay often be considered as cont-itant in a. certain frequency range,
Nearly Constant Q (NCQ) rnodelo were proposed by Lomnib (1957), Liu et al. (1976) or For the Lin model 1 the dispersion relationship is \Vritten:
Ermncrich and Korn (1987). T'l1ey are generally built as a cornbination of several sirnplc
c(wi) 1 w1
rheological cc1ls. --':"l+-ln- (2.181)
c(w2) 7rQ w2
For a three pa.rarncter (Zener) rheological rnodel, the constitutive law can be easily Lonuritz (1957) found a si1nilar result by considering an experin1ental creep lav.r of the
written for a sinusoidal disturbance (Liu el al., 1976): for1n:
0 t<0
</JNCQ(t) = { l+qln(J+td) (2.182)
(2.179) ltfo
-,,vhere q 1 d and J11o are constants deterrnined experi1nentally.
Cornbining a finite number of rheological rnodels (or relaxation mechanisrns) 1 the
E1n1nerich 1-_tnd Korn (1987) propoRed a nearly con.stant Q model involving gencralil~ed
conHtitutive lavv can be generalized as follo\.vs:
J'via_xwell bodies. For n lVIax~vell cells, they found the attenuation Q- 1 under the forrn:

(2.180) ~ w/wJ
~1 ' ,f:'_, a11 + (w/w1)2
( )
Q-1 (w) = ~ = um -~-----0,------- (2.183)
In the case of a continuous Rpectrum (infinite nun1ber of 111echanis1ns) 1 we obtain an
MR(w) m= bm ~ (w/w1) 2
l+-~a
m, 00 j=l
1
1+ (I
w Wj
l'
equivalent expression under an integral forrn.

l<"'igurc 2.21 displays the Q- 1 (w) curve corresponding to the cornbinaison of 12 rnech- To sirnplify this expression, they assu1ue Orn << m= yielding:
anisms as proposed by Liu el al. (1976). The attenuation is nearly co11Btant in a \.vide
frequency band. (2.184)
 lD-vvave propagation 2.9 Other linear and nonlinear models 95
94

They finally proposed a least-squares algorithm to optirnizc the Q- 1 variations in the • the average stresses and strains in each layer are then csti1nated (1'f'f.f (j)),
frequency band of interest. As discussed in §2.9.6, their model "\Vas recently generalized
in the nonlinea;r range by Delepinc et al. (2007) • the constitutive parameters arc updated accordingly {µi+l (j), (i+l (j)} starting an
iterative process,
2.9.4 Equivalent linear viscoelasticity
• if the new /I and ~ values are different fro1n the previou.'l ones ({ t::.µ, ,6.(} > c), a
To analyze seismic Virave propagation, it iR often necessary to consider a8HU1nptions be-
nevv stress/strain state is considered and the iterative process is rcsurned until a
yond the field of linear viscoelasticity. As shcnvn by experirr1enta1 results frorn cyclic tests
given tolerance is reached.
(Fig.2.22) or resonant column tests (§2.10), the behaviour of soils can not always be de-
scribed with a linear la\v. From the largeRt and the smallest loops depicted in Fig.2.22,
the shear modulu8 decreases for larger excitations vvhereas clamping strongly increases
(area. of the loops). Such propertieH arc very ilnportant for loadings corresponding to
linear visco
large arnplitude vibrations or strong earthquakes. One should thuH consider inore corn- elastic medium
µ,(j)
plcx constitutive lavvs. The constitutive parameters (8hear modulus, da1nping) rnay be
~,(j) layer j
deterrnined from resonant column tests at various excitation levels (see §2.10).
µ,0), ~,(j)
fo.µ,!i~}>s
·~

\
i=i+l
\
' ' ...
... ...
l-'
---
Ul
I
Ul 0.0 ........... .
----------------------------------~
Cl
JJUl
eff( ")
y i+l J average strain in layer j

-0.5
Fig. 2.23: Iterative procedure for the equivalent linear approach.

-1.0 ' - - - - - - - ' - - - - - - ' - - - - - - ' - - - - - -

-0.010 -0.005 0 0.005 0.010
strain y 2.9.5 Frequency dependent models
Fig. 2.22: Cyclic tests: hysteresis loops in the T - -y diagram. The 1nodel proposed by Kausel and Assirnaki (2002) is a viscoelastic inodel depending
on frequency. It is a kind of generalization of the linear equivalent rnodcl. As sho\vn by
the curves in Fig.2.24 1 the modulus and the damping sirnultaneously depend on strain
To approximate the nonlinear dynamic behaviour of soils, the equivalent b:near ap- and frequency. This dependence corresponds to the higher frequency sn1all loops and
proach consists in considering a linear viscoelastic model having its mechanical propertieR lower frequency large loops shown in Fig.2.22 (cyclic tests).
depending on the induced strain through an iterative procedure (Schnabel et al., 1972).
The equivalent linear approach is depicted in Fig.2.23: As sho1.vn in Fig.2.24) the shear rnodulus p and da1nping ratio ~ arc functions of
both the strain f' and frequency w. Sirnulations perforrr1ed vvith such a strain/frequency
• the viscoelastic constitutive parameters in each layer, {fli(j), (i(j)} (shear modu-
lus and damping ratio resp.), are firstly chosen from cxperirncntal results at low dependent n1odel are much Letter than those obtained with the equivalent linear approach
(Kausel and Assimaki, 2002).
excitation levels,
 lD-\vave propagation 2.10 Application 1: dynamic characterization on resonant colurnn 97

l;(y) rnodel correRpond to a nearly constant Q rnodcl in the frequency do1nain wherear-; at-
tenuation increases for increaRing Rhear .strain.s. Since this rnodel involves a very small
rnunber of paranreters (Delepine et aL) 2007), it allo\VS the analysis of strong scis1nic
motion in large geotechnical inodels.

µ(y)/µo
Q'()
,Y

y
----<-"----
: /
·'
------,,.-L'.
! ;/
:/
---- -
/~ strain
10
---------:---- --~·-··.
- ---- ~~>-/-·+··--
spectrum
frequency /
dependent / shear strain
shear modulus frequency (Hz)
and damping
Fig. 2.25: Nonlinear viscoelastic model proposed by Delepine et al. (2007):
CD variations of attenuation vs shear strain and frequency.

Fig. 2.24: Principles of the frequency dependent model (Assimaki et al., 2000;
Kausel and Assimaki, 2002).

2.10 Application 1: dynamic characterization on res-

2.9.6 Nonlinear viscoelastic models
Starting from the classical Nearly Constant Q model (Emmerich and Korn, 1987), Delepi-
nc et al. (2007) recently proposed a nonlinear model involving both nonlinefil elasticity
and nonlinear viscor-;ity. The constitutive lavv of this rnoclel is the following:
Vilhere SiJ and e,iJ arc the cornponents of the deviatoric stress and strain tensors 1 J2 is
the second invariant of the strain tensor and (1 are relaxation parameters defined as:
c(J,)y1,o
(,(t) + w1(1(t) ~ W/
1 + c(J,)
As <lcpicted in Fig.2.25, the variations of attenuation for this nonlinear viscoelastic
onant column
2.10.1 Principles of the test
The resonant column test is one of the rnost popular inethocl to analyze the vibratory
response of soilR (Cascante et aJ., 2005; Chung ct al., 1984; Drnevich and Richart) 1970;
Luong, 1980; Saxena and Reddy, 1989). The quality of the tests mainly depends on the
(2.185)
hornogencity of the stresses and strains in the specimen. As depicted in Fig.2.26, the
method consists in exciting a soil column vvith vibrations. One rr1ay apply longitudi-
nal1 tranRverse or torsional vibrationti. The transverse vibrations are generally applied
thanks to shaking tables. Longitudinal and torsional vibrations are exerted using reso-
nant colu1nn devices. ~rhis test has been studied in details by Drnevich (1977). The sanre
experirnental device, depicted in Fig.2.26, can be URed for forced a.s \vell as free vibrations.
n Ci.J(t) (2, 186)
Doth rnetho<ls allo\v the estirnation of the shear modulus and the darnping ratio of the
2..:: Yl,O
soil Rpecimen. For instance, Taylor and L·arkin (1978) performed free vibration torsional
l=l
experiinents. For torRional tests 1 the hollow cylinder cxperirncnt is very interesting since
it leads to a nearly constant tangential stress vvithin the cylinder thickness.
 lD--,,vave propagation 2.10 Application 1: dynarnic characteri7'ation on resonant colurnn 99
98

ln the configuration depicted in Fig.2.26 (zero <lisplace1ncnt at the base and zero shear
accelerometer r.;tress at the top) 1 the boundary conditions are:
excitation
system
O(x = 0) = 0
= l) = p.Ip ~: (x = l) = 0
(2 189)
{ T(x

vvhere µ is the shear modulur.; 1 Ip the geo1netrical 1noment of inertia of a cross-section

around the vertical axis of the r.;pecirnen and T the applied torque. ·The cigcnfrequencies
trixial cell arc then given by (Eqs (2.188) and (2.189)):
specimen
wl
cosv =0 (2.190)
8

that is: (2191)

1'he fin.;t eigenrnode corresponds to a quarter vvavelength resonance at frequency f1 .

The shear 1nodulus /I ha,s thus the follo-,,ving expre.ssion:

Ji= p~q = p(4fil) 2 (2.192)

Fig. 2.26: Schematic of the resonant column device. The shear strain 1(r, x) in the specilnen, associated to the first eigenmode 1 is obtained
through:

2.10.2 Description of the specimen motion 'l' = r 80 ='!_Bo r cos

8.x 2 l
(Jrx)
2l
F (7fVst)
21
(2.193)

VVhen this test is perforrncd u.sing forced vibrations, frequencies have to be tuned until
the resonance of the soil specirncn is achieved. The frequency of the electromagnetic
loading system ir.; adjusted to reach the first eigenn1ode of the specimen. The estimation
of the eigenfrcquency of the specimen and the associated eigenmode (generally the lr.;t
one) allows the calculation of the 80il shear modulus. It is also possible to pcrfor1n
free oscillations cxperirnents by stopping the vibratory excitation instantaneously. For
a torsional vibration of the speci1nen) the differential equation governing the motion is
written as follows:
(2.187)
1.vhcre f) is the rotation angle of the specirnen cross-section located at distance x frorn
the base and Vs the shear wave velocity. The general solution of Eq.(2.187) is written:
8= (Ct cos-·wx + C 2 sin-
wx) (D1 cos wt+ Dz sin wt)
Vs V:s
where fJ 0 is the rotation arnplitudc at the top of the specimen. The shear strain is
a function of the distance to the specimen axir.; and of the location of the considered
crosr:;-section. This inhomogeneout> shear strain in the specirnen makes the interpretation
of the test difficult.
2.10.3 Actual resonant column test
In the actual resonant column device 1 the top of the Hpecirnen is not free in the senHe
of condition (2.189). The electromagnetic excitation system 1 as well as the sensors 1 are
attached to the specimen head. Equation (2.189) is then replaced by the condition of
continuity for forces and displacements at the interface between the specirr1en and the
additional 1nass.
Denoting .10 the moment of inertia of the n1ass m and accounting for the conditions
of continuity, the dynan1ic equilibrium of the mass can be vvritten as follows:

that is: 0= (C 1
wx + C2 sm. Iwx)
cos - / F(wt) (2.188)
(2.194)
V:s 's Then, substiting Eq.(2.188) and considering the condition (2.189):
where Ci and Di (i=l,2) are integration constants depending on the boundary con-
ditior1s as well as on the initial conditions. Jp wl
Jo = Vs tan
(wl)
V:s
(2.195)
 lD-vvave propagation 2.10 Application 1: dynan1ic characterization on resonant column 101
100

Forced vibrations Free vibrations

where J p denotes the 1na,ss rr101ncnt of inertia of the specirncn around the vertical axis.

E·quation (2.195) gives the eigenfrcquencies of the speci1nen. I3y choosing a high value ''
for J 0 /Jp it is possible to have a r.;train state nearly independent of the cross-section
' ',, xn
',

+--!-I~
location. Denoting p = wl/Vs the first root of Eq.(2.195), and Bo the rotation at the top
of the speci1nen 1 the strain can he expressed as:

"'/ = De
r- = 80 p
r--.- cos -(P";) F.(pVs
--t) (2.196)
I
.... ··1 ·· I· ··!····I· ··I· ··I
Dx l smp l l I
I
If Jo/ Jp is high, pis srnall and: I
I
[ I
/
~·
1 band 1 ,,,
1Af 1 width ,,
The 1' value generally retained is the average value in a cross-section. It leads to the
follo1ving choice: r = d/3 where dis the specin1en dian1eter. frequency time

A.s an illustration 1 Fig.2.27 displayR the angular rotation associated to the first cigen- Fig. 2.28: Two types of resonant column tests: forced vibrations (left) and
mode for J 0 / Jp=O (no additional top mass) corresponding to a quarter wavelength vi- free vibrations (right).
bration (left), and for Jo/Jp=lOO corresponding to a typical commercial device (right).

Resonant column devices corresponding to other types of boundary conditions were 2.10.4 Estimation of damping
also developped: finite stiffness at the base of the specimen (Woods, 1978). For practical To estimate material darnping, resonant cohunn tetits rnay be performed considering
situations, the device described previously is the rnost \videly used (Drnevich, 1977). either forced vibrations (Fig.2.28, left) or free vibrations (Fig.2.28, right). In each of
these cases, the estirnation of da1nping is thus pefor1ned in a different Vi.ray:

• forced vibrations: one considers the bancl-\vidth !:J..j of the resonance curve (Fig.2.28,
x eJtl left). The clarnping ratio is related to the band-width by the follovving expression:

(2.198)
0
where Jo is the eigenfrequency of the specimen.
J,
• free vibrations: one considers the amplitude decay in tirne from Vi.rhich the loga-
rithrnic decre1nent 6 can be estimated (Fig.2.28, right). The logaritlnnic decrement
quantifies the amplitude decay between a given vibration cycle) n, and the follovling
one,n+l:
e e 6 = ln Xn+.I (2.199)
Xn
J/J,=O
Fig. 2.27: Resonant column test: 1st eigenmode of the specimen without These tVi.ro quantities are related by:
(left) and with (right) the additional top mass,
(2.200)
102 1 D-wave propagation 2.11 Application 2: <lyna1uic characterization under fast loadings 103

2.10.5 Results from resonant column tests
The resonant column experirncnt is as easy to perforrr1 as the triaxial test allo1.ving control
of drainage, mea8urc1nent of the pore pressure, and the possibility to apply a wide range
of static stresses. It allows the measurerncnt of the dynamic properties of soils for strain
a1nplitude8 ranging frorn 10- 6 to 5.10- 4 for torsional tests, and even lower a1nplitndes
in con1pression. Some experirr1ental devices, such aH the hollow cylinder system, allovv to
reach larger strains around io-- 2 (Anderson, 1974). Such devices have the advantage of
generating a uniforrn strain state in the specirr1en.
70 ------- -:- -- -~ - ~ --....
~ - ~ -:- -:- ~ -:-
For excitations with amplitude 10V1rcr than 10- 4 , soils remain in the elastic range
and the tetit is thus non destructive. It iH then possible to obtain the maxi1nurr1 rnodulus
vvhich can be, in sorne cases; directly compared to that estiinated from in situ geophysical
rneasure1nents. To deterrnine the 1naxirnum modulus 1 only the eigenfrcquency and the
geometrical configuration of the device arc needed. No strain measuren1ent is theoreti-
cally required, even if it is actually perforrned. The accuracy of the resonant colun1n is
thus larger than for tests with direct measurement of the applied force and induced strain.
Some results obtained from resonant colurrrr1 tests arc displayed in Fig.2.29: the
shear n1odulus (top) and the dan1ping ratio (bottorn) arc given as functions of shear
Rtrain (Pecker and Dupas, 1981). For this test, made on a n1ud spcci1ncn, the dynamic
propertieH have been measured for Hhear strains ranging fro1n 2.10- 5 to 3.10- 4 . For
- -:- - ~ -:- -:- ; ~ - strain below 10- 4 , the behaviour of the soil rernains elastic.
~

'
''
' ' ' ' ' '
'
'
'
'
' '
' '
' "
' ' .
'
' .
' ''
' '
'''
' ' '

- -:- .-:- ----- -- ----:- --

' ' ' ' ' ' ' ' ' ' ' ' ' ' '
'ro conclude, one inay notice that the resonant colum.n test is reliable and reprotlucible.
60 ---:- --- ~ -- ~- - ~ ~ -~ ~ ~ ~ --:- - ~ -:- ~- : ~ ---------
'
'
'
'
'
'
'
'
''
' '
'''
' ' '
'
'
'
'
'
'
''
' '
. ''
' ' '
Skoglund et al. (1976) have shovvn a good agreen1ent between results obtained in six
different laboratories, using various devices 1 for experirnents n1ade on the same material
50 under the same experimental conditions.
' '' '''
' ' ' ' ' '
' ' ' ' ' '
40 ' - - - '' - -
- - - - - - - -,- '
~- - '' - ''.', -,-
' ''-,-
. . ' - - '' - -,-' - r -,
' - - - - - - - ,' . - - -,-
. 2.11 Application 2: dynamic characterization under
30 . ----:- --- ~ ---:- -r • .; -:- -:- ~ -:- ---- -" - ~ ----:- --.;- -:- - i- - -:- H- - - - - - - - . fast loadings
' ' ' ' ' ' ' '
' ' ' ' ' '
.'
'
'
'
'
' .
'
'
'
'
In the field of fast dynarnics (shocks; blasts, etc), wave propagation phenomena are often
20'-~~-'---'-'~'---'--'-'~'-'-"'-'--~~~~~~~·~~~~~__J
dorninating. It is thus difficult to directly determine the dynarnic behaviour of materials
under fast loadings. A. specific experirnental device allows such a deterrnination: the
Split Hopkinson Pressure Bar (SHPB) (Davies and Hunter, 1963; Klepaczko et al., 1991;
10 - ' - .; -:- -:- ~ -:- - - - - - - - ~ - - - -:- - - .; " -:- - i- -:- -:- f ~ - - - - - -
Scmblat et al., 1999; Zhao et al., 1997).
'
'
'
'
' '
''
'
'''
.
'

8 ... ' '

·--J----·--~--·-·-~-~·~------
' ' .. ' '
---·---·--·--·-·--·-··----
.
' ' ' ' ' ' ' ' ' ' ' ' ' ' ' '
2.11.1 Split Hopkinson Pressure Bar test
' ' ' ' ' '
6 .......... . -.' -- ·-' - --.' '
' ' ' '
-~ -·-.'' -·-. -------.----·- --.' --·- -.-·- ·-.
'
'
'
'
'
'
'
' ' .
'
'
' '
''
' '
--------- The Split Hopkinson PreHSUTe Bar device is designed to allovv 1 UI1dcr fast loadings, the lon-
'
'
'
'
'
'
'
'
'
'' ''
' '
'''
' '
' ' '
. ''
'
'
.
'
gitudinal stress to quickly become uniforrn in the specimen. It is then possible to perform
'
'
'
'
'
'
''
''
'''
'''
'
'
' behavio'ur experiments at very high Htrain rates. In Hopkinson bar tests, the propagation
4 ----- ---,-- -.,. -~- -~- ;-:- -:-~ -:- ------ -~- ---:- -- ~- _.
... ... ... phenomena are well controlled. Meunier (1974) and Scmblat et al. in France (1999),
Felice et al. (1987, 1991) and Vcycra and Ross (1995) in the USA, Shibusawa and Oida
2 .-;-,--------;----:---.;--:--r-:--:-1-:--------- (1992a,b) in Japan and Bragov and Lomunov (1994) in Russia performed Hopkinson bar
. .' . ' '
tests to determine the response of soils under high strain rateH. This rnethod is very
O'--~~~~~~~~~~~~~~~~~ interesting since it allows reliable rneasuren1ents of both displacements and forces at the
10-' 104 10-' spccirncn ends. However, perforn1ing such tests on soils raises specific difficulties. Since
shear strain the mechanical properties of soils arc generally weak (e.g. low n1oduli), it is necessary to
adapt the classical device to this peculiar material. For instance, 1\tfeunier (1974) used
Fig. 2.29: Typical results from resonant column tests: degradation of the nylon bars, Felice (1985) perforrned oedomctric tests confining the spcciinen in a rigid
shear modulus (top), increasing damping ratio (bottom). cylinder and Zhao et al. (1997) used PMMA viscoelastic bars.
 lD-V11ave propagation 2.11 Application 2: dynamic characterization under faRt loadings 105
104

strain

....... incident bar /

~gauges~
\. transmit. bar

impactor

11
Fig. 2.30: Hopkinson bar dynamic experiment: "classical device.
specimen

2.11.2 Experimental device

'rhc classical expcrirnental device is cornpoRed of two axial bars (an inc1:dent bar and
a transm,itted bar) and an irnpactor launched on the incident bar URing a pressured air
syste1n. Figure 2.30 gives a schcrnatic of the classical H_opkinson device. It -,,oras originally
cornpo::icd of one single bar a.nd has been improved by Kolt1ky (1949) to perform indirect
n1easurernc11ts on both sides of the specirnen.
As shown in Fig.2.30, the shock between the impactor and the incident bar generates
a step 1,vavc co1nposed of a loading \Vave and an unloading wave. 'l'hese >.vaves are nearly
pure pressure \Vaves since the bending of the bars is constrained thanks to several guides
displayed all along their length. The duration of the step wave depends on the pressure
,o;ave velocity and the length of the irnpactor. The end of the transrnitted bar is equipped
with an energy absorbing device.
The measurements perforn1ed in the bars using strain gauges give the values of the
incident (cine), reflected (cref) and transmitted (Etra) strains (I<'ig.2.30). From r.;uch inea-
surements, it is por;sible to detcr1nine the forces and displacen1ents in every cross-section
of the bars (and especially at the bar-specimen interfaces) and at each time.
2.11.3 Stress wave in the specimen
The propagation of the stress wave in the bars and at the bar-specimen interfaces is a
fundarnental aspect of dyna1nic expcri1nents on l{opkinson bars. At both har-speci1nen
interfaces, multiple reflected and trans1nitted v:,raves appear. As discussed in §2.3.6 1 they
are controlled by the rnechanical para1neters o:f the bars and the specimen.
Fig. 2.31: Propagatio11 of various reflected and transmitted stress waves
through both bar-specimen interfaces.
Assurning the specirnen to be purely elastic, it is possible to have a. simple description
of each refiection-trans1nission step. As sho-1,vn in Fig.2.31) the different steps arc as
follcnvs:
• the incident stress \Vave O'i ( co1npression) is partially reflected at the first interface.
Since the rnodulus of the specirncn is lo\:ver than that of the bars, a reflected stress
wave O'r (tension) and a trans1nitted stress \:\rave O't (compression) simultaneously
appear 1
• at the second interface, the sin1ilar refiection-transrnission process occurs and gen-
erates t-,,vo compressive \Vaves: a trans1nitte<l one O'tt and a reflected one O'tr·
The schen1atic given in Fig.2.31 displays several reflected and transmitted -,,vaves vs
tin1e and location. At each reflection-transmission step, the axial stress increases at a
speed depending on the bar/speci1nen velocity ratio. After several r.;teps, the axial stress
in the specin1en is very close to that of the incident streHH wave. 'l'his phase of the ex-
perirnent is called the transient phase: the propagation phenornena don1inate. After this
phase 1 the axial stress becorncs progressively uniforrn in the speci1nen.
As shown in Fig;.2.32, t1,vo different phases can generally be diRtinguished in Hopkinson
bar dynan1ic tests:
• a transient phase: the first refiectetl and traJ1sn1itted "''aves lead to an inho1noge-
neous stress state. The propagation phenomena strongly do1ninate in this transient
 lD-\\rave propagation 2.11 Application 2: <lynamic characterization under fast loadings 107
106

phase of the test. The incident force (Fig.2.32, continuous line) is 1nuch larger tha.n forces and displaccn1ents at the specirr1en boundaries. Considering the results on pressure
the transmitted force (Fig.2.32, dotted line) 1.vaves proposed in §2.3, the expressions of the axial stress and strain are the following:
CY(b) = (!CV (2.201)
• a fast qua8i-static pha8e: after several reflections and transmissions of the loading a:i::

wave at both interfaces, the specimen reaches a stress equilibrium Rtatc. 'I'his step is
cal.led the fast quasi-static phase of the experiment: the axial stress is hon1ogeneous (b) -
E.ax -
1! (2.202)
in the whole specimen and the incident and transmitted forces are equal (Fig.2.32). c
\vherc Cla:i:: is the axial stress, Ca:r the axial strain, p the mass density, v the particle
As sho\Vll in Fig.2.32) a transient phase occurs for both loading and unloading \Vaves velocity and c the pressure wave velocity in the bars.
(phases number (1) and (3)). The time window considered in Hopkinson bar tests is
generally the fast quasi-static phase (denoted (2) and (4) in Fig.2.32). Two equilibrium These expressions arc valid for any medium of propagation. For purely elastic bars)
phar:>es are possible in the r:>pccimen: for loading and for unloading. Original Inethods expression (2.201) yields:
2
allo\v the <lctaile<l analysis of the transient phase of the Hopkinson bar tests (Zhao et al., uax = pc, c
,..,.(Ii) ~ax
(2.203)
1997). and the \Vave velocity c iH expressed as follovii'H:

c=~ (2.204)
Fast
150 F.Q.S
Quasi-Static Stress and strain in the specimen
z
~

.!<1 100
.__,
@ ® The strains being n1easured on the bars 1 the axial strain E.~~J and the axial stress u~·1
in the specirncn may Le deterrr1incd from the incident Ci: reflected CR and transmitted
({] ,
Q.) ,' strains ET (Semblat, 1995; Zhao et al., 1997).
,..,
c.J
,
'8 50 1'he axial strain in the specin1en is determined under the foll.o\ving forn1:

(2.205)
0.0
0.0 50 100 \vherc in and uut denote the first and second bar/specimen interfaces (respectively).

Considering Eq. (2.202), it leads to:

Fig. 2.32: Forces on both sides of the specimen showing the two loading (resp.
(2.206)
unloading) phases ( Semblat et al., 1999).

Since 2;~ (t) = Ei(t) - cR(t) and si:(,(t) = Er(t), the axial strain in the specimen is
finally expressed as:
2.1L4 Determination of the mechanical parameters
Mechanical parameters in the bars
c~·~(t) = *.f (cr(T) + ER(T) - E.;(T))dT (2.207)

In the bars, the parameters related to the behaviour and to Vl.'ave propagation are ex- To determine the axial stress, the normal force should be estimated at both bar-
pressed under a sin1ple forrn. After correction of some dispersive phenomena due to specimen interfaces. The follov,ring expressions are eaHily found:
tl1c three dirnensional geometry of the bars (Zhao et al., 1997), the assun1ption of one- Nin(t) = E~,, [ei(t) + ER(t)J
dimensional propagation is fully justified. The rncasurerncnts perforrned along the bars (2.208)
{ Nm,,(t) = ES,,ET(t)
should be propagated backward to the bar-specin1cn interfaces in order to determine the
108 ID-wave propagation 2.11 Application 2: dynarnic characteriLiation under fast loadingH 109

Finally, the axial stress is cletcrn1ined frorn: • one radial bar to evaluate the <lyna1nic radial stress during the test.

(2.209) For all the tests performed by Scmblat et al. (1999), the specimens are made of dry
Fontainebleau sand. The dynamic ue<lo1nctric testH performed on the device of Fig.2.33
yielding the follo\ving cxpreRsion in terms of measured axial strains: are called rigid confinerncnt tests: the rigid confining cylinder prevents the radial strain.
The confining cylinder must be sufficiently rigid or thick to obtain a Hrnall radial strain.
('l _ SbE This has been checked fron1 the 1neasured radial strer.;ses by Se1nhla.t et al. (1999).
C!ax - - [ci(t) + ER(t) + Er(t)] (2.210)
28'
where Sb is the area of the bar croRs-scction and Ss the area of the specirncn croRs-
radial
r::;cction. bar
Previous expressions may be simplified if the force.s at the ba.r/specirr1eu interfaces ,/
are identical: Nin= }..Tovt (equilibriurn). It corrcspondH to the fc-1st quasi-static phase of

....
the test. The simplified cxpreRsions are as fo1lo¥rs:
strain confining
gauges~ cylinder
, ,/
-
• simplified axial strain: ,/ I
(2.211)

incident bar transmit. bar

• si1nplified axial strain rate:
impactor t
specimen
(2 212)

• sirnplificd axial stresR: Fig. 2.33: 3D-Hopkinson pressure bar device (Sernblat et al., 1999).
(.,) _ S,,E ( ) (2.213)
CJ ax - S ET t
'
2.11.5 3D Split Hopkinson Pressure Bar test
1
- - test 001: £muy=1245 8-
Principles of the test 120
';rj ----test 008 : £moy=771 S.,'
To apply a <lynan1ic loading on soils, it is necessary to rneasure (or control) the axial and 11. 100 ······test 011 : ElllOY=393 s
radial stresses. The dynan1ic responRe of the material may thus be a.naly?:ed considering 6w
w 80
three-dirnensional stress paths. Semblat et al. (1999) perfor1ned clyna1nic oedo1netric
"'w
~ ,,
tests on soils using a rigid confining cylinder. They a.l~so propot'lecl an original experirncntal ~
60 ''
,_,
/
device: the 3D-I-fopkinson pressu1T bar. The use of a rigid confining cylinder avoids radial ''
strain, but it is then possible to estirnate the radial stress. A radial bar, in contact with "'
·,; 40 ,,.--: ...
,.;:·':.-""[
'!'
'
the specirnen through the confining cylinder, allows the measuren1ent of the radial stress "' 20 ,•'I
as a function of tirne (Sc1nblat et al., 1999). FignTe 2.33 gives a schernatic of the 3D-
Hopkinson pressure bar. 00 0.05 0.10 0.15 0.20
axial strain
Experimental device
Fig. 2.34: Examples of dynamic axial responses on the 3D-Hopkinson bar
The special experi1nental device diHpla:yed in Fig.2.33 involves three E:opkinson-type bar,':'l: device (Semblat et al., 1999).
• t\\'O axial bars to n1casure the forces and diRplaccrnents on both Rides of the t1pecirr1cn
( aR for the cla.saical device),
 lD-"\vave propagation 2.11 Application 2: dynarnic characteri?;ation under fast loadingR 111
110

2.11.6 3D fast dynamic response of sand 125

Axial stress ,--.._, 100

F'igure 2.34 displayH the axial stress vs the axial t>train for three oedometric tests (rigid
6
"'
I'.;
75
confinerncnt). These three tests correspond to identical specimen lengths (l=lOmm) but w
w
various strain rates (sec Fig.2.34 and (Semblat et al., 1999)).
] 50
w ()rad
l?or the rigid confinement tests, the axial responses are nearly linear for loading as 25
,.., ......... ---- ......................... ___ _
\vell as for unloading but with very different slopes in both cases. The behaviour of / ,- ..... '
the material under dynarnic oedometric loading is then strongly anelastic. Huvi,rever 1 the ,/
0 50 100 150 200 250 300 350
slopes for loading on one hand and for unloading on the other one arc rather close for
time (µs)
the three examples presented in Fig.2.34.

The elastic part of the response is then not really visible on the curves given in Fig. 2.35: Axial and radial stresses as functions of time (Semblat et al., 1999).
Fig.2.34. The response in the oedo1netric case i8 co1npared aft.crwarrl_8 for various loading
paths.
As discussed by Semblat et al. (1999), Fig.2.36 clearly shows that, for a linear strain
Measurement of the radial stress path (c,/Ev=2/3 for the oedometric tests), the stress path is also linear. In Fig.2.36,
The 3D-Hopkinson pressure bar device presented in Fig.2.33 allo-1,vs the dctern1ina.tion the p - q curves correspond to various specirncn lengths: 10 nun (1), 15 mm (2) and
of both the axial stress and the radial stress. The two curves in Fig.2.35 display the 20 mm (3). For the shortest specimens (1,2), the loading and unloading slopes in the
variations of the axial and radial stresses with tirne. ~rhe radial stress O"rad obviously p - q diagrams arc very different. For the 20 nun long specimen (3) 1 the loading and
changes with tirr1e since) for the results of this test, O"ra.rl reaches a maxin1urn of 30 lVIPa unloading slopes arc very close (Fig.2.36). It shows that the structure of the specimen
at time t=150µs. Beyond this tinre value, the radial stress decreases \.vhercas the axial is different after the loading phase (due to the crushing of grains for instance). The un-
loading behaviour of the rr1atcrial is thuR also different from the loading behaviour. Fro1n
stress is still increasing (Fig.2.35).
the curves displayed in Fig.2.36, this phenon1enon is stronger for shorter specimens, that
Taking into account the variations of the confining pressure (radia.l stress) in the dy- is vvhen grain crushing is important (Scmblat et al., 1999).
namic oedornetric tests is thus crucia.l since the variations of the radial stress during the
axial loading are very important. This aspect of the problem is of great irnportance in
the case of soils since their behaviour is strongly influenced by the three-di1ncnsional
--;;, 100 q s_ 2
loading path (Semblat et al., 1999). I'.; (1)
sv 3
6w 80
w
Dynamic loading: the three-dimensional point of view. "w
H
~ 60
u
(2)
The 3D-Hopkinson bar device (Fig.2.33) gives the axial and radial stresses (denoted (}'ax -~
0 40
~
and Urad respectively). Assurning a horr1og;cneous radial stress around the specimen, it
is possible to determine the three-dimensional loading paths in terrns of mean stre8H p
"'
-~ 20
"
arid deviatoric stress q. I"he mean and deviatoric stresses are defined by: "' p
00 10 20 30 40 50
p=
O'ox + 20'rnd mean stress (Iv[Pa)
3 (2.214)
{
q= O"ax - O"rad
Fig. 2.36: p-q diagrams: deviatoric stress vs mean stress for dynamic oedo-
1.vhcre O"ax and arad are the axial and radial stresses respectively. metric tests on sand (Semblat et al., 1999).
 lD-wave propagation 2.12 Application 3: respon8e of a heterogeneous soil profile 113
112

Values of the dynamic "moduli 11 lalJoratory or to derive frorn field tests through ernpirical correlations, and is routinely
deterrnincd in any soil inve8tigation. In the examples presented herein, a conservative
The rigid confinerncnt tests performed in the 3D-Hopkinson pressure bar by Scrnblat et value of 20% has been assun1ed.
al. (1999) cover a \vidc range of r-;pecirnen siLJes, strain ratcs 1 etc. For all these tests; the
moduli of the stret-is-strain curves \vere deterrnincd in order to characterize the dynan1ic
rcsponRe of sand. The dynarnic response of the rnatcrial being strongly anelastic) the ···~·
n1oduln8 considered by Semblat et al. (1999) corresponds to a hardening n1odulus. The 60
larger values (800 MP a) correspond to the lowest strain rates (200 s- 1 ), whereas the
,
lowest values (:J50-450 MPa) to the highest strain rates (800-1200 s- 1 ). Sem!Jlat e/; a.l. ~ ,,
(1999) concluded that the m,odulus decreases 'when the strain rate increases. It is never- ~ /
/
theless difficult to dcter1nine a quantitative relationBhiµ bct\veen the dynamic moduhu>
6 40 /
/
w /
w
and the strain rate due to the scatter of the results. Furthermore, it \VOuld be necessary (l) /
H I
+.>
to perfor1n tests for various i1npactor lengths (that is loading phase of various duTations) w I
H I
to clearly separate the effects of the stress level and the effectR of the strain rate.
..c:"'w
I
(l) I
20 I
Sc1nblat et aL (1999) also per.formed f::-_t8t dyna1nic tests under 'soft confine1nent' (i.e. I
I
pressure cell) to cornpare the fast dynamic response of soils for various stress-strain paths. I
I

0.0. 2.0 4.0 6.0 8.0 10. 12.

2.12 Application 3: response of a heterogeneous soil shear strain (%)
profile
Fig. 2.37: Soil shear stress-shear strain curve: the dashed line represents lab-
·The approach proposed in this section is Lased on an analytical solution to the \Vave oratory rneasurernents; the solid line represents the elasto-plastic idealization
equation in a heterogeneous soil profile. The assu1ned soil constitutive behaviour is of (Pecker, 2005).
the elastic perfectly plastic typc 1 and is completely defined by the rnaximu1n shear stress
and the yield strain at any depth. This inethodology has been applied to pararnetric
numerical analyses investigating scis1nic ground rnotion (Pecker 1 2005).

2.12.2 Wave equation for a heterogeneous profile

2.12.1 Soil behaviour
Let us consider a soil layer of finite thickness overlying a stiffer baJfspace (referred to
It has long been recognized that t:1oil behaviour fr.; highly non-linear beyond very small as the bedrock in the following), vvhich Vilill Le considered as a rigid Loundary. The
strains, and tl1at a co1nplete description of the behaviour vvould require considerations soil is assurned to be isotropic elastic 1 \vith a shear ¥.rave velocity increasing Vi'ith depth
of srnall as well as large strains. The result of a simple shear test carried out on an according to sorne povver lavv:
undisturbed sample of rncdiun1 plasticity clay, sho\vn in Fig.2.37 1 illustrates the previ-
ous staternent. V\.Tit.hin the context of the sirnplified approach developed herein, only
z+d)l! (2.215)
the shear stress-shear strain c1rrve is needed. The true stress-Rtrain curve is then ap- V(z)=Vs ( d+h
proxiinated by a bilinear curve corresponding to an elastic-perfectly plastic constitutive
relationship (Fig.2.a7); it is defined by two parameters, which may depend on the depth ¥.'here z iR rr1easuTe<l frorn the ground R1rrface 1 h is the layer thickness, p a real positive
z: the shear strength Trnax(z) and the engineering shear strain at \vhich yield occ1u·s 'f J (z ). para1ueter sn1aller than 2 and Vs the shear \\'ave velocity at depth h. d is a Btrictly
positive paran1eter that can be chosen to fit Eq.(2.215) to the experimental data. As a
The shear modulus is then given by G(z) = Tnwc(z)ht(z). The energy dissipated particular case, th.is paran1eter can be set equal to zero giving a zero shear vvave velocity
under cyclic loading docs not explicitly appear in the forrnulation; it is taken into account at the ground surface. It is convenient to rnake a transformation and to define the ne\v
in the analysis throngh an equivalent damping ratio (Pecker, 2005). Unlike the equivalent variaLles:
linear model, this n1odel iR robust and rather reliable becauRe the shear strength is z+d
less affected by san1ple re1nolding than the shear rnodulus, it iR easy to 111easure in the ( = d +h and H = d+h
 1.D-wave propagation 2.12 Application 3: response of a heterogeneous soil profile 115
114

'I'he Vlrave equation for a plane vertically incident shear wave is vvritten as: The first equation gives the mode shapes:

3T(z) _ - a
-- -
oz oz
(c( )au)_
oz
Z - - p--
2
a2 u _ (D'u + .
- p
3z 3t 2
Vg
(t)) (2.216) (2.223)

vvith the boundary conditions:

where pis the rnass density1 T(z) iR the horizontal Rhear stress, G(z) the shear modulus
related to the shear wave velocity by G(z) = pV 2 (z), Uthe absolute soil displacement, u Jim (PX'(() =0
(~(o.
is the horizontal displacc1nent 1 relative to the bedrock; ancl 'Ug the bedrock displacement (2.224)
{ X(l) = 0
motion. With the change of variable z-. ( defined above, Eq. (2.216) becomes:
2 A solution of the form X(() = ("Y((q) is sought for Eq. (2.223) (Pecker, 2005). The
V 2 -3
__lL ( 3 )
p ___}!:_ - _-3!:_
8 .. (2.217)
H2 3( ( 8( - 3t 2 + vg(t) follo-\ving equation iH easily obtained \vith the notation .8 = (q:

Y" (!!) + 2a + q + P - 1 Y'(,6) +

~
['x(a -.1) +pa + (Hw) e-2q-pl Y(;3)
~~ q~
2 = 0 (2.225)

2.12.3 Boundary conditions

An appropriate choice of the paran1eters a and q reduces the previous equation to a
The boundary conditions express that the relative displace1ncnt at the bedrock interface Bessel equation provided the folluviring choice is made:
and the shear stress at the ground surface arc:
2 - 2q-p = 0 q = 2.=.E.
2
(2.226)
( =
d
H = (o T((o) = 0 =? G((o) Dul
D = 0
{ 2a + q+ p -1 = q a-~
- l-]J

( (~(c (2.218)
{ ( = 1 : Eq. (2.225) becomes:
u(l,t)=O
p-1
V\Then the shear 1,vave velocity is different frorn zero at the ground surface (d > 0), //= -- (2.227)
2-p
the first of the two boundary conditioni'l rcduceH to:
The general solution is obtained as:
( =
d
H = (o : Bui -0 (2.219)
3( (~(c -
(2.228)

\\Then the shear wave velocity is equal to Liero at the ground surface, the lirnit of the
V1.rhere ,J1) and Yv are Bessel's function of the first kind and second kind and A = ( 2 ~1..u~)v8 ,
first boundary condition rnust be considered:

OU rraking into account the recurrence formula (Abramovvitz and Stegun, 1970):
(=0 T(( = 0) = 0 =?Jim G(()- = 0 (2.220)
(~O 0(
(2.229)
Considering the homogeneous equation, without the forcing tcr1n V9 (t), and using the
technique of separation of variables u( (, t) = X( ()y(t); Eq. (2.217) becomes: in "\vhich Cv represents either lv or Yv, the derivative of Eq.(2.228) can be expressed
as:

(2.221) (2.230)

The boundary conditions (2.224) together with the two relationships (2.228) and
vvhich can be "\Vritten: (2.230) yield a systern of tvvo equations with two unknowns A and B. ·ThiH systen1 has
a non-trivial Rolution if, and only if, its deterrninant is equal to zero. This condition
2 ((P
Vs .sj_
d(
dX. )
d(
=
.. (t)
L =est= -w 2 (2.222)
provideR the frequency equation, which possesses an infinite number of distinct real pos-
itive roots (Abramowitz and Stegun, 1970). 10 each of these roots is associated a mode
H2 X(() y(t)
116 1D-Virave propagation 2.12 Application~~; response of a heterogeneous soil profile 117

shape Xm,, vv~hich is nor1nalized to 1.0 at the surface for convenience; the solution u((, t) The shear strain at any depth vvithin the profile is then expressed at>:
is expanded in tern1s of the rnode shapes:

/'n1a:r;(z) =
f_(au,(z)) 2
1
(2.238)
u((, t) = L Xm(()um(t) (2.231) i=l Dz H
m=l

Given the ortl1ogonality property of the rnodc shapes, the solution of the wave equa- Zero shear wave velocity at the ground surface.
tion (2.217) is given by: This situation arises when d = () in Eq.(2.215). For the mode Rhape to re1nain finite

Yn(t)
JI

(o
x,;(()d( + w:,.Yn(t) ;·l
· (o
X~(!,)d( = -v9 (t) ;·1
(u
Xn(()d( (2.2'.l2)
at the ground surface, B = 0 n1ust he enforced in Eq.(2.228). 'I'he second boundary
condition (2.224) yields the frequency equation:

(2.239)
v.lhere Lv'n is the nth soil column circular frequency, vrhich is solution of the frequency
equation. Introducing the 1noclc participation fa.ctor O:n: virhich possesses an infinite nurnbcr of distinct roots ,\.i. The eigenfrequencie8 of the
Roil colurnn are:
(1 Xn(()d( -,\Vs(2-p)
f 4IrH (2.240)
.J(o (2.23:J)
i - i

O'.n = 1

.J
r x,; (()d(
(o
and the rnodc shapes, norn1alizcd to 1.0 at the surface (z = 0), are:

(2.241)
Equation (2.232) iH finally written:

Yn(t) + w?,yn(t) = -CYnUg(t) (2.234) V1.rhere r(.) is the Gam1na function. It is easily checked that, as long asp< 2, the first
E·quation (2.234) can be solved by any ccnnputational technique like DuhamePs inte- boundary condition (2.224) is satisfied because XI(() ~ (! J_,_
,_, (Ai(
2
;:p) 1.vhen ( tends
gral. In the context of this section 1 only the ITH-lXirnurn surface acceleration is sought and to zero.
a spectral analysis is used. The rnaxirnun1 displacernent response in each rnode is given
by: Non-zero sl1ear wave velocity at the surface.
(2.235)
The frequency equation it>:
\Vhere 8d(wi: ~i) is the spectral relative displacement of the input n1otion i) 9 for fre-
quency Wi, and darnping ratio ~i· The inaximurn ground surface acceleration due to the (2.242)
ccn1tribution of the first 1\l modes is:

N
\Vhich possesses an infinite nun1bcr of distinct roots ,\i· The frequencies of the soil
ii.mae>Jz = 0) = "D [a.;Sa(w;, ~i)] 2 (2.236) column are still given by Eq.(2.240), and the mode shapes, normalized to 1.0 at the
'i=l
ground surface arc:

where Sa(w;,(i) = w'fSd(w,,~,) is the pseudo acceleration. Equation (2.236) takes into
account the fact that the mode shapes have been normalized to l at the ground surface (2.243)
(X;( (o) = 1). The percentage of modal maHs relative to the total mass of the soil column,
·vvhich is an indicator of the nu1nber of rr1ode8 J\T to retain; is:
The derivation of the previous equation takes advantage of the recurrence relationship
(Abramowitz and Ste gun, 1970):
1
(2.237) 2
m.n = 1- (u Jv+1 (x)Yv(x) - Jv(x)Yv+r (x) = - (2.244)
7fX
 lD-Vl.'ave propagation 2.13 Application 4: soil-structure interaction 119
118

1000 corresponding to the hon1ogcneous layer) given the relationships J _-b; (x) = /fr :Jt and
0

r(~) = y'if, the classical solution is retrieved from Eq. (26) to (28):
I ·ip=Or '
,;.:.
100
0 "
R
:>-; = (2i - 1) ~~ (2.245)

.- .•

10
I-.-· -- '-·· -·· ! "- , - /
I For p cf 0, the eigenfrcqucncies (Fig.2.38, left) arc found to be different from that of
the ho1nogeneous case and the inode participation factors are not coru;tant 1,vith respect
(1 to dirnensionlcss paran1eter (o.
:

1 2.13 Application 4: soil-structure interaction

1000

,
_ II k~-n c ;
1.51===:.t====:.....r=:..:J=:::... 2.13.1 Basic principles
.',./ ' Cone models were initially proposed by Meek and Wolf (1992) and allow the description
100 of dynamic soil-structure interaction under a si1nplified form. This rncthod considers a
_, .. ··/ certain soil volurne vvith a conical geometry to account for wave radiation in the soil. The

0
1-- -_:::.: .......
. c•
:·:: .....
,,,,
.-- - /
:

: schernatic proposed in Fig.2.39 shows the basic principle of the method: a conical soil
volume intersecting the base of the foundation is isolated and seismic Vilave propagation
·•·····
is then studied in this soil volume only (Pecker, 2008). It allows a simplified analysis of
' dyna1nic soil-Htructurc interaction.

1000 .......

I f p=l:O·, .'.••

100 .· r..:·>·/
' '
[::.:---::·~~
.,
-+... - -.
,__ .• /

10 /

1
0 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 I
dimensionless depth ~o dimensionless depth ~c
··················j············ ·········
Fig. 2.38: Roots of the frequency equation (left) and mode participation --· I
factors (right) for the five first modes (numbered top-left) and three p-values: I
p=0.00 (top), p=0.50 (middle) and p=l.00 (bottom), from (Pecker, 2005). I
I
I
2.12.4 Eigenfrequencies and mode participation factors "'Oz

The solutions of Eqs (2.239) and (2.242) and the mode participation factors can be Fig. 2.39: Principle of cone models.
cornputed 1 once and for all, aR a function of (o and p, the paran1eters defining the velocity
profile (Pecker, 2005). The results are presented in Fig.2.38. For the special case p = 0,
 lD-\i1tave propagation 2.1:-3 J-\pplication 4: soil-structure interaction 121
120

2.13.2 Equations of motion for the soil

it i1nplies:
Diix(zo,w)
n
__ 1l:r;ZQ,W
-
1 .
(· . ) ( -+i- w) (2.253)
One thus considers a conical soil colu1nn assun1ed aR a one-di1nensional bar having a uz zu V:s
·variable cros8-section "Vlrith depth (area S(z)). The 1vaveficld in the soil cone rnay then Hence:
be derived frorn the lD Vi.rave equations detailed in §2.5. For a purely horizontal rnotion,
F ( w ) = p,-1rzotan2 (} (
1 +I.wzo)
Vs ux(zu,w) (2.254)
the following expression is obtained:
DV D2 ·u,, Considering that R = zo tan6l and denoting W = wcR and K = p,,-rrRtanO, t=q.(2.254)
(2.246)
Dz = pS Dt 2 becomes:
8
F(w)=K(l+i )ux(zo,w) (2.255)
,vhere: V = µS ,Dux
az is the shear force (bending being neglected ) . tanB
Using the static stiffness I(, 1,al = ~~~, fJ rnay be identific<l:
T~\1rthcrmorc:
S' = S(z) = 7r(ztan8)
2 (2.247) 8
K(w = 0) = 111rRta11G = K,tat, that is: tanB = ~-~ (2.256)
(2 - v)K
hence:
(2.248) 1'he irnpedance of the foundation thus hecornes:

Since S depends on z, it leads to the follo-1,0,ring equation: _ [ .(2-v)K_l (2.257)

K(w) = K,tat l +z W
2
8
2
DS(z) D11x ( ) D nx _ p S( ) D iix (2.249)
- - - - + S z - 2- - - z - -
Dz Dz Dz µ Dt 2
35·~-~-~--~-~------~-------
8
the spatial derivative of the area of the cross-section is: ~~) = 27rz tan
2
(} , yielding:

2 Bu~i; 8 2 u.x 2
p 3 u:r:
(2.250) 3 V,=150m/s
- - + - -2 = - - -
z Dz Dz µ Dt 2 V,=200m/s
The solution is then of the form (Pecker, 2008): V,=300m/s

l .
'o
·.g 2.5
u.x(z, w) nx(zo,w) exp [-1-(z
.w - zo) exp(zwt) (2.251)
= /
z zo Vs '"'
The previous analysis perfor1ned for pure shear n1ay be considered under a si1nilar 2
forrn for pure cornpression. 1~he sa1ne expression holds for the vertical displacc1nent Uz.

2.13.3 Case of a surface excitation 1.5

----
...
Considering a harmonic force F(w) trans1nitted at the free surface by the foundation, --·· -··
.. -·
the boundary condition fulfilled by the solution (2.251) reads:

F(w) = V(zo) = -µS(zo) a:

D-u (2.252)
5 10 15 20 25
radius R
30 35 40 45 50

Fig. 2.40: Kdyn/ I<stal ratio estimated through cone models .as a function of

since: u..,(zo.w)
·· '/,
z zo
exp [ - ,iw (
- z - zo
V'.:S'
l] exp (' )
iwt
the foundation radius R for various shear wave velocities: Vs = 150m/s, V'.:s =
200rn/s and v;, = 300rn/s (u = 0.25, 11.=200 MPa and f = 2w~=2 Hz).

-1
, Vw 11x(zo,
s
/ w) exp [ -ic-V
z zo
. w (z - zo
s
l] exp (, )iwt
122 11)-wave propagation
2.13.4 Influence of the wave velocity in the soil
To assess the influence of the foundation geometry as vvell as of the soil properties on
tl1e dynamic soil-structure interaction 1 the Kdyn/ Kr:ita.t ratio, defined through Eq.(2.257),
is plotted a8 a function of the foundation radius Il for various values of the shear wave
velocity in the soil Vs (Fig.2.40). The Kdyn/ ]{8 tat ratio is larger for 101ver shear wave
velocities and larger foundation. 'The dynamic soil-structure interaction is thur.; stronger
for large structures and soft soils.
2.14 Experimental estimation of damping
2.14.1 Various experimental methods
Ir1 addition to resonant. column tests) several other rnethods arc available for the ex-
perimental characterization of clamping in soils. The n1ain ones are discuRsed in this
section.
Cyclic tests
For various types of experiments (triaxial teHtsi shear tests) hollo\V cylinder), the energy
lo!:!S per cycle can be obtained fro1n the hysteretic stress-strain curves for different loading
and unloading cycles (Das, 1983; Kramer, 1996; Kokusho, 1980; Pecker, 1984). The
modvJns and the darnp'ing are estimated as function8 of the strain level, the void ratio
and, possibly, the nurnber of cycles. As shoVirn in Fig.2.22, the Rhear 1nodulus decreases
for an increasing strain level vvhcreas the dan1ping (related to the area of the loops)
ir1crcases. At high frequencies and lo\v strains, the direct rneasurement of the phase shift
ill is possible on rock specimenR.
Reso11ant bar test
It is one of the 1nost popular teHt for rock specirnens. A rock bar is subrnitted to an
har1nonic excitation at one encl, the displacement of the other end being rneasurcd to
deterrninc its eigenfrequcncy (Lucet, 1989).
The use of various bar lengths alloVirs analyses at different frequencies corresponding to
different modes. For strongly attenuating media (C-2 <3)i thiH method is no longer appli-
cable sir1cc it is nearly irnpossible to reach resonance. Convcrsely1 for weakly attenuating
n1edia, it is possible to consider a transient pulse and to analyze its decay.
Wave propagation based methods
Sonic or ultrasonic waves in rocks and soils. Ultrasonic tests are often used to
study wave propagation in rocks. Several authors perforrned ultrasonic tests on soil spec-
imens (Agarwal and Ishibashi, 1992; Blangy et al., 1993; Koerner et al.,1976; Stephenwn,
1978). VariouR types of testR may be defined:
2.14 Experi1nental esti1nation of da1nping
• acoustic ernission: measurement of acoustic signals emitted by the spccilnen during
loading,
• transm,ission te.sts: excitation of the speci1nen by an ultrasonic source and mea-
surc1nent of the signal transmitted through the 1naterial
Dlangy et al. (1993) proposed son1e comparisons hetvveen variou8 experimental ap-
proaches. The authors also analyze models proposed in various theories of propagation
in porous rnedia. Their 1nain conclusions are the folloViring:
• for consolidated sands: the shear 1nodulus is proportional to poroHity,
• for loose sands: there is an exponential dependence.
The ratio bet'\veen the velocity of P-waves and that of S-waves is constant for clean sands
('Vr /~9=1..5) and depends on the porosity Virhen the clay content increases. Further1nore,
Blangy et al. (1993) define a critical porosity above which the specimen loses its shear
resistance (grainR in suspension).
Kolsky test. 'rhe Kolsky test (1949) iH a dynamic experiinent consisting in launching
a, rnass directly on the specimen. A_ shOVi'n in Fig.2.41, the dyna1nic response is n1easured
on the specirnen itself, generally at tVi.'0 different points. The response of the n1aterial
is analyzed frorn the observed propagation phenon1ena. The two 1neasurements indeed
characteri:.:-;c the way the 'perturbation propagates in the specin1en. The Kolsky test rnay
be conRidered for coherent and stiff n1aterials (rocks, stiff clayH, etc) and the excitation
level should not he to large.
measurement
points
specimen
Fig. 2.41: Principle of the Kolsky test (Kolsky, 1949).
Propagation of plastic waves. Ultrasonic or Kolsky experi1nentR involve low or 1nod-
erate energy levels. ()ther types of tests at higher energy levels n1ay he considered: it
is for instance possible to generate plastic waves and to study their propagation in the
material. J'vieunicr (1974) performed such faHt dyna1nic tests on clay specirnenr:;. Through
fast in1age processing 1 such tests allow the analysis of the plastic '\Vave propagation in
the specimen.
124 11)-v.,rave propagation 2.14 Experirnental estimation of darnping 125

All these tests may give a qualitative idea about the propagation phenornena occuring
in the n1aterial: velocity) dispersion la\ols, etc. 'fhey generally <lo not allo\v a direct access
to tl1c behaviour of the rnatcrial, vvhcreas Hopkinson bar cxpcri1ncnts do.
In situ methods
lvfany different surface or in-hole inethods (reflection, refraction 1 logging) allovv the anal-
ysis of in situ wave propagation. The excitation is generated hy vibrating systen1s or
by explosive sources. The -,,vavelengths are such that the san1plcd soil depths are often
heterogeneous.
These rnctbods raise the need for r.;ophisticated signal processing techniques to ac-
count for tl1e velocity dispersion; the response spectrurn of the sensors, etc. Since 8uch
techniques rnc.tinly involve 2D/3D wave propagation, they will be discussed in Chapter 3.
2.14.2 Methods for the estimation of attenuation
The cxpcri1nental measures give the g(f) = ln ~~ curve vvhich is nearly a line if the
assurnption is fulfilled. The slope of this line giveR tlle attenuation value (Bonrbi6 ct al.,
1987).
• Advantages: T'hiR rnethod is less senr.;itive to noise than the rise tirnc n1ethod.
Furthcr1nore, it only requires two n1easurernents.
• Drn11Jbacks: to apply this rr1ethod, a mini1num signal duration is necesr.;a.ry. The
influence of multiple reflections may be in1portant. Q is generally underestin1ated
(due to scattering effecto).
nPropagation 11 methods.
Senne authors (Badri and Jlviooncy, 1987) ur.;e the causal attenuation operator clue to Fut-
terrnann (1962). ThiH rnethod consists in Hirnulating the propagation of a signal measured
in purely elastic n1cdiu1n and to rnodify it by a filter satisfying the physical causality con-
dition.

Rise tirne method Q is then determined in order to optimize the agreement het~veen the synthetic seis-
This method is based on the crnpirical relation proposed by Gladvirin and Stacey (1974): 1nogram and the n1casured signal. Other s:ynthesis n1ethods n1ay be considered.
T • Ad11antages: the estimation of CJ is independent of the source, the sensor and its
T ~ To + C r Q-l dt
.fo
(2.258) environn1ent .

1vhere Tis the rise thnc of the first peak (To at the source), T the travel tirne aud Ca • Dru'wbacks: the computations arc generally carried out for plane Vi.raves 1 the actual
constant depending on the source (for Q >10). geornetrical attenuation rnust then be Vi.reak. The coupling -,,vith other 1nethods is
often necessary to deterrnine the range of interest for Q.
·This la\v is generally linear:
(2.259) 2.14.3 Definitions of attenuation: synthesis
\larious physical para.rr1etcrs alloVi.r the quantitative analysis of rnechanical \!'\rave attenu-
On a T = f(T) graph, the cxperi1nental rise tirnes arc obtained at different points
ation in solids: the q11,ality factor Q, the atten'Uation factor 0: 1 the logarithrnic decrement
(corresponding to different travel tirnes T) and should theoretically allo-,,v the delermJna-
tion of a line with a slope equal to C /Q. The attenuation values can then be estimated
o, the phase difference Ll<P, etc (Aki and Richanls, 1980; Bourhie et al., 1987). These
various pararnetcrs arc co1npared hereafter:
from C (Bourbie et al., 1987; Jongmans, 1991).
• For a given loading/unloading cycle, the quality factor CJ is defined by the follcnving
• Advantages: only the onset of the signal is considered; the effect of reflections is
relation:
then rc<luce<l. This 1nethod is rather ear.;y to use. 1 L'.lW
(2.261)
• Drn'wbacks: C and To depend on the Ronrce, but C also depends on the Q value. Q(w) 27rW
F'urtherrnore, the rnethod is not adapted to realistic scis1nic pulses (overeRtimation -,,vherc VV is the m,a:rirn'Urn elastic strain ener:qy stored in the considered volume and
of Q) and mrnsable for low Q (Q <l). -6.. i1r is the energy lost in one cycle.
• the attenuation factor ct: it corresponds to the .spatial arnplitude decrease of the
Spectral ratio method \Vave (Vlreakly attenuating inediurn). If the amplitude is known at two points, ct
For two rneasuren1ent points _l\11 ancl 1112 1 assuming s1nall variations of Q over the fre- 111 ay be \\rri tten:
quency range, the spectral a1nplitude ratio reads as follows: o: =
X2 -
1 l
x1 n
(A1)
Az
(2.262)

(2.260) where 6..~c = x2 - x1 is the distance between both points.

126 lD-Virave propagation 2 .14 }~xperirnental cstitnation of darn ping 127

• the logaritlnnic tlccrcn1cnt 6: it corresponds to the tim,e arnplitude decrease for an Influence of the confining pressure
harmonic wave (free vibrations) and it can be >NTittcn: The attenuation decreases for an increasing confining pressure. Saxena and Reddy (1989)
proposed the follo-,,ving empirical relation for soils:
o, = ln ( A(t.n) ) (2.263)
D = KP-o.13co.:33
A(tn+1) (2.269)

• the phase difference <I> gjven the cornpl.ex modulus ]\![ = lvfn + il\/11 1 the phase vvhcrc /) is the attenuation and P is the confining pressure.
Hhift <I> betwcer1 stresses and strains (forced harrnonic vibrations) allovvs the char-
acterization of attenuation:
From the theoretical relation proposed by Stewart et al. (198:J) for rocks, the atten-
1Vfr uation D = Q--i can be written as follo-1,0/s:
tan cl>= - - (2.264)
1Vh

• the \Vidth of the resonance peak 6.f: on a resonance curve (sec §2.10) 1 the cstin1a-
Q- 1~ ;:;1 (2.270)

tion of the peak band-width leads directly to the value of the quality factor: 1vhcre ( fr.; the density of cracks in the rock and k a coefficient inversely proportional to
the contact radius a and the friction coefficient J.
1
(2.265)
Q From these rclations 1 it is obvious that Q- 1 decreases when P increases and that the
or to that of the damping ratio: variations arc very different for soils and rockR. These laws \Vill be corr1parcd in Chapter 4
to investigate seisrnic vvave attenuation in centrifuged soils.
(2.266)

Influence of frequency
Finally, all these parameterR are related as follovvs:
IVIost laboratory experirncnts on various types of soils and rocks shov..r that the attenuation
1
-
cxA
= 2~ = -
o
~ - =
Mr
-
.. 6.f
~tan <I:> = - (2.267) is strongly dependent on fr-cq-uency. From the results obtained by Stoll on sand (1979)
Q 7f 7f i\lh fr and J\1urphy on sandstone (I3ourbie et al.) 1987), the attenuation increases with frequency.

where A is the wavelength. Nevertheless, for soils in the frequency range of seismic waves, the attenuation slightly
depends on frequency. Fron1 the results of Stoll (1979) 1 the attenuation curves at lo-\v
2.14.4 Variations of attenuation frequencies arc very close.
This section gives several ernpirical results on the variations of attenuation for 8oils and
Conversely1 fron1 in situ experirncnts on salt (1 to 200Hi), l'vfacCartor and VVortn1an
rocks. (1990) showed that the attenuation decreases for increasing .frequencies.

Influence of the strain amplitude

From the experirnents perforrncd by Hardin and Drnevich (1972a,b) in the laboratory
on sand, at frequencies f >0.lHz, or Stoll (1979), at frequencies ranging from f=lO
to f =500 H7., the attenuation strongly ·increases with the .strain amplitude. Saxena and
Reddy (1989) even proposed the following empirical law:
(2.268)
2.14.5 Characterization of the various approaches
A synthesis of the various pararnctcrs characterizing each experimental 1ncthod is pro-
posed in Table 2.1. Seismic methods arc also considered Lut are mainly dedicated to
21) /3D in situ characterization. 'I'he deta.ilc<l analysis of such n1cthods is proposed in
Chapter 3.

where E is the strain arnplitude. This increase is lovJer for a large confining pressure.
2.14.6 Comparison of the governing parameters
Mavko (1979) and Stewart et al. (1983) showed experimentally and theoretically that,
above a strain level of 10- 6 or io··- 5 , the attenuation depends on the strain arnplitnde. To compare the practical problems and the investigation n1ethods presented previously;
Thus 1 the attenuation increase8 rapidly vvi.th the strain amplitude. three pararneter8 are considered:
128 lD-V\rave propagation 2.1'1 Experi1nental estirnation of damping 129

The schematic proposed in Fig.2.112 allows a classification of the various approaches.

Table 2.1: Main parameters characterizir1g the various approaches. Some other governing pararnctcrs n1a.J' alflo be chosen, such as the strain rate. AR ex-
1\Iain rncchanical para1n.eters plained previously, two types of approaches obviously appear in this classification. It iH
Type of
characteri;,r,ing the approach possible to directly analyze the beha'V'iour of the m,aterial or to investigate the propagation
approach
phenornena generated in the material. These two types of approaches arc discussed in
cyclic frequency strain dimensions
this book:
tests 10- 0 to 10- 1 Hz from 10- 3 to 10- 2 c1n) drn
resonant frequency strain dimensions • as discussed in this Chapter: the analysis of the soil behaviour under fast dynamic
colurru1 frorn 10 to 500 .Hz 10-4 to 10- 2 crn 1 dm loadings is po8sible on a dynarr1ic IIopkinson bar device,
Hopkinson strain rate strain dimensions
frorn 50 to 1000 s- 1 from 0.05 to 0.2 m/m cm • as it \vill be shown later (Chapter 3), propagation phenornena in soils may also
bars
strain dimensions be investigated through centrifuge experiments or analysis of surface v.ravcs in the
seisn1ic frequency
1 to JOO Hz 10- 0 to 1W 3 m/m dam, lnn, km field.
rnethods
ultrasonic frequency strain dimensions
tests 100 kHz to 1 MHz 10-7 crn, dm
plastic strain_ rate strain dimensions
1.vaves 100 to 5000 s·- 1 0.1 to 2.0 m/m dm

• the fi·equcncy f 1

• the ratio between the \\Tavelcngth and the characteristic cli1nension of the problem
considered 1\/lchar,

• the strain a.mplitu<le co.

The ~/\_/lchar ratio allo\vs to distinguish situations where the propagation phenomena
arc dominating from cases \Vhere the analysis of behaviour inay be directly perforn1ed:

• if tl1e A./lchor ratio is large: the wavelength is large \Vhen con1pared to the charac-
teristic di1ncnsionf'l of the problem. The propagation phenom,ena are negligible,

• if the A/lcho.r· ratio is sm.all: the <lirnensions of the problern are large when cornpared
to the wavelength. The propagalion phenornena dorn'inate.
Fig. 2.42: Comparison of the characteristic parruneters for various experi-
Figure 2.42 displays a 3D diagra1n sho\ving the ranges of the follo\ving three param- mental approaches (Sernblat et al., 1999).
eterr:; for dynarnic cxperilnents: frequency, A/lchnr ratio and strain. The co1nparison
betl'.recn cyclic approaches and ultrasonic tests (Fig.2.42) clearly shlnvs that:

• for cyclic experiments (top right), the frequencies are very llnv 1 the strain a1npli-
t11des are rather large and the A/lchar ratio is very high: the propagation phenomena
are negligible,

• for 'Ultrason:ic tests (botto111 left) 1 the frequencies are very high, the Rt.rain arnpli-
tudes arc extrernely small and the A./lehar ratio is very 1:nnall: the propagation
phenorncna dom,inate.
Chapter 3

2D/3D-wave propagation
Application to the dynamic
characterization of soils

3.1 Introduction
The analysis of vibrations in hean1s, proposed in Chapter 2, allovvs to consider various
vvave typeH in lD structures. ¥lave propagation in three-dirnensional elastic solids cor-
responds to caReS in which one dimension is not significantly larger than the others. It
is often the case for wave propagation in soils: seitnnic wave propagation, irnpact due to
explosions or dynamic compaction, vibrations due to traffic or pile driving.

All these phenornena have a common feature: their sources, located on the free-surface
or at-depth, generate \vavcficlds propagating in the soil layers. The rnain objective of this
chapter is to analyze \O.rave propagation considering 2D /3D heterogeneitiei:'l or obstacles.
Even if the soil has a dissipative (so1netimes) nonlinear behaviour 1 , we shall rnainly
consider linear elasticity and small strains. As evidenced in Chapter 2, the solutions for
2D /3D viscoelastic rnedia 1nay be easily derived from the elastic ones.

3.2 Dynamic equilibrium of a continuous medium

The fir.':'it stage consists in vvriting the equilibriun1 equations starting frorn the Principle
of Virtual Work. The equilibriurn equations are then cornbined with the constitutive law
of the 1nedium in order to obtain expressions involving the displacement vector 1l_(JV1) at
every point lvl in the medium.

1 Sevcral nonlinear models are discussed, or n1ay be found through the citations, in §2.9
 2D (JD-wave propagation 3.2 Dynarnic equilibrium of a continuous rnediun1 133
132

3.2.1 Equilibrium equation - principle of virtual work Equilibrium equation

Virtual rate of work by external forces T'l1c cquilibriun1 equation iR obtained by expressing that the sun1 of the rate of work by
external force8 and the rate of \Vork by internal forces is equal to the rate of \Vork by
External forces are represented by a vol umic denHity of force i in the dornain fl and quantities of acceler'1tion (SalenGOil, 2001), that is:
a s11Tface density of force T_r1 (Fig.3.1) on its boundary 8f1. The rate of v,rork by these
forces considering a virtual velocity field 0-(&) 1 where z_ dcnoteH the por.;ition of a point
1\!l; \.Vritcs as follovi'S:
-l o
g d({;_)dil + ( f_.U dO + ( Id {;_da =
Jo Jan
f pg. udn
lri
(3.4)

Taking into account the Hymrnetry of the stress tensor, the firRt ter1n hecurnes:
(31)
r g: d({;_)dn Jnr g : vu d()
Jn
= (3.5)

and integrating by parts:

an - ( g : dUl)dD
.fo
= j (v.g) .{;_do - Jan( (g n) {;_da
n
(3.6)

"Introducing Eq.(:-3.G) in Eq.(3.4) and writing the equation for every virtual velocity
field (I, we obtain the equilibriurn equations:

v.g + t_ =pg. in D (3 7)
~-Il = T_d on an (3.8)

Fig. 3.1: External forces existing in domain fl. where g is the Cauchy streHH tensor, 1 the body forces and T_rl the surface forces.
3.2.2 Constitutive equation
Virtual rate of work by internal forces The isotropic linear elastic constitutive law in s1nall strains is:

Considerir1g that the rate of vvork by internal forces only depends on the virtual velocity (3.9)
field{;_ and its first gradient (Salcn<;on, 2001), it is po8Sililc to show that:
\vhere ,\ and µ denote the Lame conr.;tantH, ll_ the unitary 2nd order tenRor (i.e. its
(3 2) con1ponents are (\,J), and~ the strain tensor related to the displacc1ncnt field y by:

1
-~= -('vu+vu)
(3.10)
2 -- -
in which g is the Cauchy streHS tensor, g(Tl) ~('vu+ vu) is the strain rate that iR, in terms of curnponents:
tensor, VU is the gradient of the virtual velocity field {;_ (sec appendix A), that is in
~ a·u, + OUj)
components:
- -
d;.1 (U) = -
1
2
(au, + --
-
auj) .
8x.i D.Ti
Etj =
2
(
OXj ax, (3.11)

In Eq.(3.9), the tcr1n tr~ corresponds to the volurnic strain:

Virtual rate of work by quantities of acceleration :J 3 "

~

The rate of vvork by quantities of acceleration is given by (Salengon, 2001): tr~= ~Eii
i=L
= L
i=l
O'Ui
Bx·
i
(3.12)

(3.3) The Lanie constants, ,\ and µ, fully define the isotropic linear elastic behaviour of the
A(U) =
Jn( pg.(32, t) .{;_(32)dn material. In Table ~1.1, Vile recall the relations bet\.veen ,\,µ,the Young's modulus E 1 the
vvhere pis the density and Q. the acceleration vector. Poisson'H ratio v and the bulk rnodulus K.
 -~

2D /3D-wave propagation 3.2 Dynamic equilibriurn of a continuous mediurn 135

134

3.2.5 Uncoupled wave equations

Table 3.1: Relatio11ships between the parameters defining the isotropic linear
Including Eq.(3.15) into the equilibrium equation (Eq.(3.13)) and taking into account
elastic behaviour of the material.
Eq.(3.17), we obtain the following equations for the potentials:
E v ,\ µ=G K
vE E E 1 3 2 dJ
E, v E v to.(¢)= V 2 at; - F (3.18)
(l + v)(l - 2v) 2(1 + v) 3(1 - 2v) p

µ(3,\ + 2µ) ,\ 3,\ + 2µ

,\, µ ,\ µ
3
1 32 w (:0.19)
,\+ µ 2(,\ + 11.) !o.('ljJ_) = VJ 3t2 - G
in vvhich vve consider the notations:

3.2.3 Equilibrium equations in terms of displacements

Including the constitutive law (Eq.(3.9)) into the equilibrium equation (3.7), and taking
Vr = J.\: 2
µ and Vs = ·l't (3.20)

into account the definition of the strain tensor (Eq.(3.10)). we obtain the equilibrium and we apply PoisHon's theorem to j_:
equation expressed as a function of the displacerncnt field 11:
32u [_ = v~v(F) + v§v /\ Q (3.21)
(.\ + µ) [v(Y.it)] +µlo.it= pat-; - [_ (3.13)
In terms of components, Eqs (3.18) and (3.19) may be written:
where the operators \7 and ,6.. repreHent the gradient and the Laplacian respectively (see
appendix A). (3.22)
In terms of components, Eq. (3.13) yields:

(>.+µ)ax 1
3a) +µ:L_332ax
3(Lax Uj Ui
2
32 ?Ii
=pat' -J.; ; i = 1, 3 (3.14) ; i = 11 3 (3.23)
' j=l J j=l J

3.2.6 Body waves

3.2.4 Decomposition of the displacement field
Omitting the boundary conditions, Eqs (3.18) and (3.19) govern the propagation of two
Considering Poisson 1s theorem, it is alv.rays possible to decompose a vectorial field g as
different waves propagating at velocities Vp and Vs. These Vilaves are called body waves
the sun1 of a gradient of a scalar potential rJ; and the rotational of a vectorial potential
and are of t\vo types:
'ljJ_:
it=Y(¢)+Vll'ljJ_ (3.15) • equation (3.18) governs the propagation of the pre8sure v;ave which is also called a
that is, in terrr1s of components: P-wave,

31> + -
3'1/Jk 3'</Jj • equation (3.19) governs the propagation of the shear 'wave 1,vhich is also called an
'U,i = - -- - - , circular permutation for 'i, j k 1
(3.16)
S-wa'ue.
· 3x; 3xj 3xk
Exarnining Eq.(3.15) and starting frorn the three components of vector Jk, we introduce The polarization of these 1,vaves is illustrated in the follo\ving section for plane 1,vaves.
four components by the decon1p0Hition into potentials: the scalar potential and the three
components of the vectorial potential. It is then necessary to specify an additional The comparison of Eqs (3.20) shows that the P-wave velocity Vp is larger than the
condition. Such a condition is generally obtained by writing that the divergence of the S-vvave velocity Vs. The Vp/\!';_9 ratio can be easily expre8sed as a function of Poisson's
vectorial potential 3£ is zero: ratio:
v."¢ = 0 (3.17)
(3 24)
It ineans that the vectorial potential '!£. does not lead to any volume change.
 -,

2D /3D-wave propagation 3.2 Dynarnic equilibrium of a continuous 1nediun1 137

136

5.0~-~--~---,-~--~-~--~-~--~--.- 3.2.7 Wave propagation in anisotropic media

In the case of linear elasticity in sn1all strainR, the constitutive lavv has the following
for1n:
4.0 (3.25)

where g is the stress tensor) Q,_ is the fourth order elasticity tensor and f_ is the strain
tensor.
3.0

In the isotropic case) the elasticity tensor leads to the sarue expresRion as in Eq.(3.9)
since it can be expressed as:
2.0
(3.26)

For an anisotropic merlium 1 Q rnay have up to 21 indepentlcnt co1nponents. (;eologica.l

1. 0 --- --- ---- ---- --. ----- --- --. ----- --- --. -- -- --. ---- --- -- --- --- ---- --- -- --- -- --- ---- -- --- -. --- -- ---
rncdia are often considered as orthotropic rncdia (also called trans'uersely ani.c;otropic
rncdia). Assuming the orthotropy axis as f.zi all the clirectionR in the planes perpendicular
to flz arc then equivalent. Arnong the 21 components of the elasticity tensor, only five are
OL_-~-----'--~-~--~-~--~-~--~-~ independent: Cxxxx = C 11 Y'.IJ'Y' Czzzz 1 Cxxyy, Cyyzz = CJ.r,xzz and Cxzxz = Cyzyz (DourbiC
0 0.10 0.20 0.30 0.40 0.50
el al., 1987). All other components arc hero except ()1:y:ry which is vvritten:
Poisson1s ratio
1
Fig. 3.2: Ratio between the P-wave velocity and the S-wave velocity. C:r:y:ry = 2(Cxxx;i: - C:r:r:yy) (3.27)

We finally obtain four independent velocity values (Dourbie et a.l., 1987):

• for P--vvaves:
Fignre 3.2 displays the variations of the ratio between the P-wave velocity an<l the
s~-vvave velocity. The P-wave velocity is always larger than the ~'J-wave velocity and the
ratio fastly ir1creases for PoisRon's ratios above 0.4. (3.28)

Table 3.2 gives several values of the Vp and Vs velocities for various types of soils
(soft, stiff, interrnediate).
• for S-\vaves:
uVB-·
_ J
CY"JJZ
- --
p
_ .J Cxzxz
- -
p
Table 3.2: Typical values of the P and 8-wave velocities for different types of (3.29)
soils.
Vr (rn/s) Vs=
J\!Iaterial Vs (m/s) p
Clay 100-200 1500
Sand above >.vatcrtable 200-400 400-800 In the case of anisotropic inedia 1 the vvave velocity is thus depending of the direction
Sand belo>.v >.vatcrtable 200-400 1500-1700 of propagation. Its characterization 111ay he perforn1ed by using field rncasurernents (Aki
Marl 400-600 1500-2000 and R.iclmrds, 1980) or laboratory tests (Nguyen et al., 2008).
R.ock > 800 > 2 000
 2.D /3D-v;,ravc propagation 3.:-3 \;\lave propagation in unbounded 1ncdia 139
138

3.3 Wave propagation in unbounded media In the case of a pure plane wave, the prestlnre 1,vavc governed by E·q.(3.31) generates a
vibration oriented along the direction of propagation (Fig.3.4, top). The pressure wave is
thus polariiccl along the direction of propagation. Conversely, the shear wave governed by
3.3.1 Wave equations for plane waves
E·q.(3.32) generates a vibration in the plane perpendicular to the direction of propagation
Assurning the propagation of P and S' >Naves to take place along one direction only1 de- (Fig.3.4 1 botton1). The shear \Vavc is thus polarized perpendicularly to the direction of
noted ~ 1 , tl1e variables involved in the vvave equations only depend on x1. Equation (3.15) propagation (tangent plane). Thro types of shear vvavcs, having the sa1ne velocity Vs,
is then reduced to: may he distinguished:
a¢ a1/J3 a,h
1f = --!::'._] - -,- f 2 + - - f ; : i (3.30)
ao:1 . ax, ax, • SV-\vaves polarized perpendicularly to the direction of propagation but \vith the
induced rnotion located in a vertical plane 1
that is, in terms of displacements (Eq. (3.13)):
• SH-waves polarized perpendicularly to the direction of propagation but V\rith the
a2 'U1 1 a2 v.1 (3.31)
induced rnotion located in a horizontal plane (for horizontally propagating \Vaves).
ax! vi at 2 AH it vvill be evidenced later, for general 3D cases involving oblique incidences, the
polarization of SV-\vaves is not vertical but located in a vertical plane including the
[j'2'Uj direction of propagation (i.e. direction of polarization of P-vvaves). Sirnilarly1 the polar-
and (3.32)
Bxi ization of !)H-\vavcs is not horizontal but located in a plane which is perpendicular to
the (oblique) direction of propagation.
This situation of one-dimensional propagation is also encountered in the case of curvi-
linear n1edia for which the geornetry of the inedium constrains the direction of propa-
gation (see Chapter 2). Equation (3.31) then represents the equation of longitudinal P-wave
vibratior1s of a bean1 in which v-,re considered E = pV~ and equations (3.32) that of the
transverse vibrations of bearr1s in \\rhich \Ve have taken µS' = pSV5~ and neglected the
bending strains when compared to shear strains (ES>> µS').

"'Vile can also find these wave types in the caRe of a two or threc-dirnensional rncdium.
For instance, in the case of an iHotropic infinite solid: a. point Hource will generate a
spherical V\ravefield because of the Ryrnn1etry of the problem. Far avvay frorr1 the souTce, longitudinal direction of
the \vavefront may be considered, due to its expansion) as a plane (the plane tangent to polarization propagation
the wavefront of the spherical wave, Fig.3.3).
transverse
polarization
-

1
'--'--4

* 3 SH-wave
Fig. 3.4: Plane pressure waves (top) and shear waves (bottom).

Fig. 3.3: Nearly plane waves generated at large distances from a spherical
source.
 2D /3D-\vavc propagation 3.3 \\Tave propagation in unbounded rnedia 141
140

3.3.2 Plane monochromatic waves They arc expre::;sed as:

Solutions in terms of potentials iw iw .

- +-f,B /\l- (3.39)
-u = --fpAl
Vp' Vs ' -
Tl1c expression of potcntiah> ¢and ?f· for a plane 1nonochrornatic 1,vavc at frequency w, is
obtained as the solution of the equations of propagation (3.18) and (3.19). We look for Under this forrn, v>re notice the P-vvave) -,,vith velocity Vp, leading to a displacc-
solutions of the form 7i(1\1)e-i,wl, point 1\1 belonging to a pla.nc of fixed direction2 . \iVe 1ncnt along the direction of propagation L; the 8-'\vave vvith velocity Vs leading to a
thus obtain: displacement in the plane perpendicular to the direction of propagation (Fig.~1.4). The
(3 33) latter motion may Le decon1posed into a cornponent located in a horizontal plane and
a component in a vertical plane. Denoting AsH and Asv the amplitudes of the related
clisplacerncnts, Eq.(3.39) yields:
(3.34) . 7.W
u811 = A sH f s = Vs (3.40)
where Lis the norrnalized 1vavc vector di:dined, frorn the v;.ravc vector "Ii_, by:

1vith: Ci= I', 8) (3 35) (341)

and Eqs (3.33) and (3.34) involve the following product: Proof. The d.isplacen1ent vector for tranvcrsc waves is given by the second tcrn1 of Eq.(3.39). For
w w S.ff-\vaves, ·we look for a horizontal cornponent of If_.
lsY 1.:JC= vi L.:i: = Vy (l,.r + lyy + lzz) (j = P, 8) (3.36)
Denoting!;_ the horizontal unit vector in the plane perpendicular to I aud (f 1 ,f2 ,f2) a Cartesian
orthonormal fnnne, we get the following relation::i:
\~Then the cornponcnts of vector L arc all real, it corrcspondr.; to the vector of the
-t.f.3 = o ; tI = o 11£11~1 (3.12)
cosine directions of the propagation axis for the body wa1Jcs \vi.th velocity Vp and Vs. It
Neglecting the n1ultiplicative scalar tern1s, the hori>0ontal co1nponcni: of If_ \vrites:
Hhould be 11oticed that the polarization of the S-v,ravc along B corresponds to the plane
perpendicular to L (i.e. !.B = 0). A~(!iAL).£

thus, developing and taking into ctccount the relations giving /,1 and tz obtained from Eq.(3.42):
V\Then a co1nponent of L for instance lz, is purely irnaginary and the other componentf> -ly -lx
~,i2= ~ (3.44)
arc real, the solution potentials (Eqs (3.~J~l) and (3.34)) thus represent a wave propagat- ti=
y lx + ly V l,r; + ly
ing in the ( x - y) plane with an exponentially decaying an1plitudc along z. Such 1..vavcs
are called surface 'WaveB and -,,vill be studied later in this chapter (§3.S.2). A= -(Bylz - Bzly)ly + (Bzlx - Bxlz)lx
(3.4S)
Jli+l.~
Con.sidcring that {.jj_ = 0, we have -B 11 lzl'!J -B:clzla, = Bzl; and including in Eq.(3.45), the following
relation is derived:
Solutions in terms of displacements
A= -=Bee"= (3.46)
The displacerr1ents associated to the potentials ¢ and '!£ are derived frorn Eq.(3.15). Jl~+l~
fron1 which \Ve obtain expression (3.40). The sauie type of proof stnnd.s for the calculation of Asv
Denoting:
fp = iw (lxx
exp [ Vp + lyy + lzz - Vpt) l (3.37)
by choosing to project Jl. on the vector !!:. defined by J:l = I/\!;_.

l
Keeping the general n1eaning of the previous relations, it is possihle to chose the
global coordinate sy::;tcrn in order to have the direction of propagation included in the
and iw (l,,,x
.fs =exp [ Vs + lyy + lzz - Vst) (3.38)
(x - z) plane. Thus, it comes that ly = 0 (Fig.3.5) and that relations (3.39) and (3.40)
can be Vi'ritten:
2 As already discussed in Chapter 2, solutiont; of the forrn u(s - ct) are equiv::ilcnt to solutions of
'Ux Apfplx + A.w fslz
the forrn u(ct - 8). :However, considering solutions of the fonn n(8 - ct) leads to spectral cornpoucnts
·u As1ds (3.47)
11,~
estiru1:ttcd by inverse Fourier transform illsl;ead of direct Fourier t:ransfornrn for solutions of 1.he fonn {
n(ct - s). Ar.fplz -Asv fsl,,
142 2D /3D-\vave propagation 3.3 Wave propagation in unbounded rnedia 143

Equations (3.47) show that, in the case of plane waves) the displaccrncnt along Oy
is uncoupled from displacements along ();r and Oz. It is only due to the propagation of
shear, Vi.raves vvith horizontal polarization, which arc called SH'-waves. 1~he displacernents
along directions x and z result from the propagation of P-waves as well as S-v,raves with
vertical polarization, which arc called SV-vvaves (Fig.3.5). Further1nore, if the direction
of propagation is vertical, parallel to Oz (lx = ly = 0, l2 = 1) 1 the problc1n bccon1cs
one-dimensional: each component of the motion is UI1couplcd fron1 the others:
zk: x I\
Asvfs '' \ \
AsHfB (3.48) ' \

Ar fr ~
Rsv ~
x-z plane y-u_i_plane
x y
Fig. 3.6: Reflection-refraction of plane waves at a plane interface: SH-waves
(left), P /SY-waves (right).

In the follo\ving 1 vve analy7'e in great details the reflection-refraction of a plane S'H 1
P and SY-waves respectively.
z u~
The variables related to the incident wave are denoted \Vi th an (i) index 1 those re-
Fig. 3.5: Polarization of P, 8V and 8H-waves displayed in the (x - z) and lated to the reflected waves with an (R) index and those related to the transmitted (or
(y - 1'_!_) planes. refracted) '°'raves "\O.rith a (T) index.

T'he displacements of the incident VI.raves can be vvritten under the general form:

3.3.3 Reflection-refraction of plane waves at an interface 11-n ~ An!l(n) exp [ ~ ( x z~n) + z z~n) - V,,t) l (3.49)

General form of the various waves relation in v,rhich index (n) identifies the various wave types, (lxJz) a.re the coRine
We noV\r consider an incident plane "\Vave (Fig.3.6) propagating across an interface between directions of the propagation axis (norrnali:ted wave vector). Vector d_ is the unit vector
two half-spaces having the following rncchanical properties: (A1, P,1, p1) and (A2, µ2 1 P2). giving the direction of the displacernent, that i::; the direction of polarization of the wave.
'l'his incident "\Vave "\Vill generate reflected waves propagating in the inedium frorn '°''hich These different paran1eterR are detailed hereafter for various wave types.
the incident wave is originated, and refracted v,raves crossing the interface and propagating
in the second medium. In the general case, Fig.3.6 (right) 1 we can dernonstrate that an Incident 111ave: zt,) = sin f)i l~'i,) = cos f)i
'
incident "\vavc gives rise to:
• t"\vo reflected VI.raves,
• SH-wave:
{ V'.9 ~ V'.9,
QJSH) = {0, 1, 0}

• t\.vo refracted "\Vaves.

The amplitude, the incidence and the nature of the reflected and refracted '°'raves are
• P-wave: { Vp = Vp,
d(P) --
-i
{ Slll
. eP
i : 0, COS f)i
p}

: I''
easily obtained V\rriting the continuity of the displacement vector 1!:: and the stress vector
! = q_.n_ at the interface. • SV-"\vave:
{ V'.g
fliSV)
~Vs,

= { C08 fJfV, 0, - Rin (/fV}

 2D /3D-svavc propagation 3.3 \A/ave propagation in unbounded media 145
lM

Reflection-refraction of a SH-wave between two media Continuity of traction. ()nly the :ry and yz components of the strain tensor are
non zero. Its trace is zero and the stress tensor has only non ~ero con1poncnts along xy
Reflected and transmitted waves. V'lc seek a solution under the for1n of a reflected and yz. T'he vector normal to the interface being 11. = §.2 , the equations of continuity for
SFJ-wave) \vith arnplitude Rsr1, and a transrnitted (or refracted) SII-v,rave, with an1pli- the stress vector g-11. thus involve the con1ponent Clyz:
tude TsH (Fig. 3.6 1 left). The directions of propagation and polarization of these waves
(i) (R) _ (T)
are defined as follows: CJ,yz + (Jyz - (J'yz (3.54)
Vs= Vs,
• R.cflected SH-\vave: iSH) =
-R
{0 ' LO}
,
yiclrling: (3.55)
{
LV/H) ={sin en, 0, - coseR}
Starting from Eq.(3.55), this equation can he rewritten under the forn1:

• Transrnitted SH-vvave:
(3 56)

Amplitudes of the reflected and transmitted waves. Equations (3.53) and (3.56)
The displace1nentt-> related to the inl'.itlcnt 1 reflected and transmitted waves are thus allo\v the detern1ination of the a1nplitudes of the reflected wave RsR and the trans1nitted
given by: \vave TsH as functions of the a1nplitude of the incident \vave AsH:
u~'i =Asllcxp [~;, (xsine,+zcose,-Vs,t)] RsH µJ Vs, cos e, -112 Vs, cos OT
AsH = /11 Vs2 cos e, + /12 Vs1 COR er
v,~R) = [~w
1
RsIIcxp (xsineR - zcosOR - Vs,t)l (3.50) (3.57)
s, 1SH 2111 V:.o,' 2 cos ()1
As11 µi Vs 2 cos Bi + µ2 Vs cos Br
,,(T) =
Y
TsR exp [~
Vs2
(:r sin Or + z cos eT - V:s, t)] in whichµ; anrl Vs, arc linked hy Eq.(3.20).
1

the other cornponcnts (.T and z) being zero for all vvaves.
Denoting xs = V:s12 /Vs 1 the shear v.rave velocity ratio and assuming that PI ~ p2, the
amplitudes of the reflected and transmitted waveR can he derived as f\1nctionR of xs and
Equations of continuity the incidence angle oi only:

Continuity of displacement. Choot->ing the origin z=O at the interface, the equa-
RsH cos (Ji - xsJ 1 - x~ sin 2 (Ji
tions of continuity for dit->placements require:

(3.51)
AsH COS (},i + XS' v 1- X~' sin 2 (},i (3.58)
TsH 2 cos oi
Including relations (3.50) into Eqs (3.51) and writing that, for z = 11, the derived AsR cos Oi + xsJl - x~ sin 2 Bi
identitieR are valid for all values of x and t, \Ve obtain the folhnving relationR, to be
satisfied by () j and V:'h,:
sin(] R. sin Gr
since Eq. (3.52) yields cos 2 O~ = 1 - x;, sin 2
Oi·
(3.52)
V:s'1 V'.:'>2 The normalized amplitudes RsH / AsH and 1:'>H / AsH are displayed in Fig.3.7 aR func-
One recognizes in relations (3.52) the Snell-Descartes law.s from classical optics. tions of the incidence angle (},i for tvvo values of the velocity ratio xs = V:.s 2 /V'.:s1 : xs = 0.8
(top, softer upper medium) and xs = 2.0 (bottom, stiffer upper medium).
I'he equation of continuity for displacernent thus leads to a first relation bet\veen RsH
and TsH
AsH + RsH = TsR (3.53)
146 2D /3D-\vavc propaga.tion 3.3 VVave propagation in unboundc<l rnedia 147

than unity. Conversely, for a stiffer upper 1nediu1n (xs = 2.0, botton1), the trans-
1.0
------------ ------- ---- --- n1ission coefficient is lo-vver than one.
- ................ IT,,1/AsHI • the incident SH-wave is fully refracted (Rs1-I=O) -,,vhen the follo-vi.ring condition is
',,
x=0.8 '' fulfilled:
'' (3.59)
0.5 '
''
'' A specific con1binatiou of the inechanical properties of Loth rnedia thus leads to no
'' reflected wave (xs = 0.8, top). Removing eT in relationR (3.52) and (3.59), this
0 .............. .
' expression becomes:
2 2
Jl·I)
µ2 cos 2Bi . 2f),i
+ Slll =
Vs, ) (3.60)
( ( Vs,

• if the tcrn1 (17s 2 /Vs,) sin ei is larger than unity, cos Br is then purely irnaginary and
the transrnitted \O.rave is expressed as:

-1.ooL_ _ _1~0---2~0--~30--~40--~5-0_ _ _6~0---7~0---s~o---=90 (3.61)

:
in which Xs correspundR to:

v
:

I x=2.o I ~
ii - -- -w-
xs . 2g i
Xs2 sin 1
ii - (3.G2)
jl critical Vs,
~ 1
angle
,,~x
The trans1nitte<l wave iR then an interface 1uave v,rhich a1nplitude exponentially
I: \ decreases vvith the distance to the interface. Furtherrnore, Eq.(3.58) Rhows that
I !
'
\
\
the RsH / AsH ratio corresponds to two conjugated complex variables, since cos BT
I \
I R51/AsH I is purely in1aginary ; the an1plitude of the reflected wave is equal to the an1plitude
/
/ '' of the incident wave with a ;r /2 phase shift (xs = 2.0, bottom).
''
------- ---
/
/
/
'' • If /J.2 = Vs 2 = 0, the wavefield propagates in a half-Rpace (medium 1) and is reflected
'' on the free-surface (this caRe will be examined in the next sections).
'''
' '
' ', I Ts1/AsH I Reflection-refraction of a P-wave at an interface
', '
-- --
0
J -----
' '
Reflected and transmitted waves. Since P-\vaves have two components of displace-
111ent (along x and z), the continuity conditions for both displacement and traction thuR
0 10 20 30 40 50 60 70 80 90
lead to four equations (instead of t\vo for SH-waves). If1 as in the caRe of SH-v,rave,
incidence angle
we only consider t\vo unknovvns Rp and Tp, the systern will be overdetermined. An
incident P-\vave actually generates reflected and refracted P-waveR as well as reflected
Fig. 3.7: Amplitudes of the reflected (solid) and transmitted (dashed) waves and refracted SV-v,raves (Fig.3.6, right). The equations of displacem.ents associated to
at a plane interface for two values of the velocity ratio X = Vs 2 /Vs1: X =0.8 the various waves types are given by Eqs (3.49) considering the adequate pararr1eten.;,
(top), x =2.0 (bottom). that is:

The curves dit->playe<l in Fig.3. 7 lead to the follov,ring rernarks: • Reflected P-wave:
• if the upper n1edium i8 softer (xs = 0.8 1 top), the transmission coefficient is larger
148 2D /3D-wave propagation 3.;) Wave propagation in unbounded media 149

Equations of continuity
• Ilefiectcd SV-vvave:
Continuity of displacement. At the interface (z = 0), the equations of continuity
for displace111ent can be written under the follovviug forn1:

(i-P)
lf,j
+ u.i(R-P) _,
r
(Jl-SV) _
uJ - u.i
(T-P) + uJ(1'-SV) j = x, z (3.68)
d(P) = { . () . ·OP}
• Refracted P-v.,ave: -T sin eP
T' ,coo T

L)/:l = ric}~l These relations must be valid for all values of x and t. It thus leads to the conditions:

Vs= Vs, sinB,F sin B!j,V

• Ilefra.cted SV-1.vave: d_~'JV) = { - cos ()¥V, 0, sin (J~V} Vp1 V:?2

{ (sv) -_ {sm
. gsv sv} Starting frorn the t>yste111 of equations (3.68), v.re obtain tv.'O equations relating the
l
-T T , 0 , cos 81
a1nplitudes R.i and T.i (j = P, SV) of the reflected and transrnitted vvaves to the an1plitude
The displacen1ents due to the different 1,va:ves arc such as u:~j) = 0 (Vj) and: of the incident wave Ap:

• incident v.rave, lf_(i-P): along x: (Ap + Rp) sin Bf+ R 8 v cose'ft = Tp sin Bf, -Tsv cose'fv
(:J.70)
f ·u1i-P) = Ap sin Of exp [ ;;,, (x sin er + z cos Of. - Vp, t) l (3.63)
{ along z: (Ap - Rr) cos Of + Rsv sin g~v = Tr cos ej, + Tsv sin BCfv

Continuity of traction. For each wavefield (j = i-P, R-P, R-SV, T-P, T-SV),
111~i-P) = Ap cos er exp [ ~;, (x sin Of+ z cos er - Vp, t)] the stress vector is vvritten as follov.rs:

(3 71)
• reflected \\raves, g(R-P) and y,_(R-S'V):
. (i-P) + O"kz
(R-P) + (Jl-SV) _ (T-P) + O"kz
(T-SV)
,,~n-P) = Rp sin&}; exp [ ~;'. (x sin Oh - z cos Hf; - Vp, t)] t h at u;: crkz O"kz - crkz k =x,z

l u~Tl-P) = -Rpcoserexp [~(xsinB:h- zcos8ft-Vp t)]

Vp-i
1

l
(3.64) It thus leads to the systcrn of equations:

J along x (Ap - Rp) sin 2Bf - Rsv Xr cos Wj\v

Tp ~~ xs sin 2Bf - Tsvx1xs cos 2e!j,v
=

·u)1_1-sv) = Rsv cos e~V exp [ ~;, (x sin e~V - z cos e~V - Vs, t)
(3.65) l along z. (Ap+Rr)x1cos20~v -R 8 Vsin28~v =

1·u~R-SV) = Rsvsine)';v exp [~w (xsinB~v - zcose~v -Vs,t)]

s,
Tpx2xs cos 2B!j,F + Tsvx 8 sin 2e!j,v

• tra.r1srnittcd (or refracted) ·vvaves 1 u(T-P) and Y:.(T -SV):

v.~T-P) = Tp sin oj, exp [ ~:, (x sin e.j:'. + z cos ej, - Vp, t) l and:

1 ·u~r-P) = 1? cos e!j exp [~(x sin Of,+ zcosOj, -

VP2
Vp, t)]
-Tsvcos()~,v- exp [~(xsinBfv +zcosB!fV -Y.s t)]

!
11(T-SV) = 2
x Vs2
(3.67)
(T-SV)
?Lz =Tsvs1n eSV
T exp [ Vs
· 'lW ( · nSV
xs1noT . gSV
+zcor:; T -
VS 2 t)]
2
 -I

2D /3D-wave propagation :).;) VVave propagation in unbounded 111edia 151

150

Amplitudes of the reflected and transmitted waves

1.2,---~----~-~------------
Variable incidence: Considering a variable incidence, we obtain the follo\'iring syr-;-
tcn1 of four equations \vi th four unknowns:

- sinB,f - cose~v . gP - cos (J~',V

1.0
I x~o.s I

l
SlllT w
~

e,r
(
sine!j,v
cos - sin BftV cosB!j.
-x1xs cos 2eiv
"
QJ
·~
0.8
sin 2trf x1 cos2e~v x1
X'
XS sin 2(}P
T s"
-x.1 cos28~v sin 2esv
R x2xs cos 2ef1' xs sin2B¥v (3.74)
QJ
0
u

( l~'"(
0.6
Rp sin Of
"
l
0
·~
w
Rsv cos Of w

Tr sin2Bf
·~

sw 0.4
Tsv x1 cos28%v ""'
;,
~

Conversely to the case of SH-"\vaves, for an incident P-v.rave, tvvo critical angles ean 0.2
--- --- --- ---------- --
be defined:
-- --- ---
• cr·itical angle 1: the transn1itted P-v.;ave becorncs an interface \.vavc if:
1
sin ()1:' ~-· <1 (that is, when Vp, ::C Vp,)
ic·1· 1 XP - 1.6
x~2.o
ITp/Apl
w 1.4
~
• critical angle 2: the transmitted SV-wave becornet-> an interface ·vvave if: • ,-, I TsviAr I
"
QJ
·~
u 1.2
first
critical
second
critical : I "-'
sin&[,., = Xi :<; 1 (that is, when Vs, 2' Vp,) ta angle angle !/ \
~ xs \

~rI
QJ
0 \
u 1.0 \
For two values of the velocity ratio x, VlC then obtain the results displayed in Figs 3.8 \

and 3.9 corresponding to various cases with or \vithout critical incidence. J\_s in the case "
0
·~
w
w
0.8
/
~
/!
\
\
\
\
of SFI-waves, the trans1nission coefficient of P-wavcs is larger than unity when the upper §
w 0.6 ,- ------ \
\
mcdiurn is stiffer (and less than one in the opposite case). /

""'
H
0.4
I
I \
\
\
,,
~
\
Normal incidence: Considering a norrnal incidence at the interface (Of=O), we , \

can easily show tha.t relations (3.74) become: 0.2 ,, / \

\
,' \
, ,' \
Rsv =Tsv = 0 0
0 10 20 30 40 50 60 70 80 90
Rp p2 Vp, - p1 Vp,
(3. 75) incidence angle (P)

Tp 2p1VP, Fig. 3.8: Amplitudes of transmitted P and SY-waves at a plane interface for
Ap P2 Vp2 + P1 Vp1 au incident P-wave and for two values of the velocity ratio x ~ Vp, /VP,: x ~0.8
Equations (3.75) show that, if the incidence of the P-v..Tave is perpendicular to the (top), x =2.0 (bottom). -
interface, there is neither refiected ,5\7-\\rave nor transmitted s~v-,vavc. \Vhen the two
n1cdia in contact. have the sarne m,echanical impedance (=pVp ), no reflected P-wave
appears for normal incidence, which is in agreerncnt \vith intuition.
 2D /3D-wavc propagation 3.:) V\Tavc propagation in unbountled media
152

Reflection-refraction of a SV-wave at an interface

The same method as in the previous section leads to sirnilar results for an incident SV-
\vave:
(3. 76)
00
.µ
x=o.s
'I'hc solution can be derived hy considering a system of four equations "\vith four ~

unkno\vns shnilar to systcrn (8.74): ·~

·~
'"u 0.8
:±:;
. BPR et; "u0
cosefv- cose~v sin

}
Slll 0.6
~
er;
( Bk 0
sin BfV - cos -sinGfj,V cos ·~
00
00
xs cos 2e~v - xs sin 2()P
- COS 2HfV _.l sin2BP
x1 R X2 T
·~

s
00 DA
x1 cos 2o~v -xs sin 2(}~v -x2xs cos 2e~v ,- - -, I Tp/Asv I
- sin 2(}'.sv
' cos e:;r1.i
(3.77)
~
,.,"'
.µ / ----- --- --
}~'w(
Rsv

l
I
' 0.2
- /

}
/
Rp - sin(lfV
Tsv
Tp
cos 28,{JV
- r::dn2esv 0
-- --
0 10 20 30 40 50 60 70 80 90
' 4
As illustrated in l,"ig.3.9 (bottom), in the case of an incident SV-wave 1 \Ve may have
at least one and up to three different critical angles: 3.5
• critical angle 1: the reflected P-vvavc becornes an interface Vlrave if: x=2.o I TsviAsv I
00
.µ
~
3
sinB,sv = -1:_::; 1 (that is, when Vp, ?: Vs, (!ilways true)) .2'u
'"'' 1 x1 ·~ third
:±:; 2.5 first second i critical
• critical angle 2: the transrnitted SV--,,vave beco1nes an interface wave if: '"u0 critical
angle
critical i
angle i
angle
~ 2
sin eSV = -1:_ <1 (that is, when V:s, ?: Vs, ) ·~
0
00
"":
lcr2 XS -

• critical angle 3: the trans1nitted P-wave become8 an interface wave if:

·~
00

s
00
'1
1.5
""· ,.I
i/-,
i \
\
\

1 .tJ"' : I
:I
\
\
i,

3.3.4
sinBfV
er~
= - - :S 1 (that is, when Vp, ?: Vs,)
XPXl

Plane waves in layered media: vibration isolation

0.5
-/
1!
1; - 1
I
---"'! ! ' ' ...
dTP/Asvl ----- ---
''

----
0
0 10 20 30 40 50 60 70 80 90
Vibration isolation by screens or trenches: simplified analysis incidence angle (SV)
As already discussed in Chapter 1, vibration isolation rnay be performed by screens or
trenches in the Hoil. UHing previous theoretical results, we shall now consider a sirnplified Fig. 3.9: Amplitudes of transmitted P and SV-waves at a plane interface for
analysis involving an infinite vertical layer of constant thickness h embedded in a free- an incident SV-wave and for two values of the velocity ratio x = Vs 2 /Vs'i:
space. As depicted in Fig.3.10, the m.echanical properties of the layer arc different from X =0.8 (top), x =2.0 (bottom).
that of the soil. l::'or sake of sirnplicity1 the free-surface is disregarded and the efficiency
of the isolation systern is studied aR a function of the velocity ratio between the soil and
the layer.
154 2D/3D-wave propagation :-::L:-3 \Vave propagation in unbounded media 155

Transrnittivity of a plane SH-wave through a single layer

µ,p µ',pl µ, p
Displacements and tractions. As depicted in Fig.3.10 1 we shall IlO\V consider an
infinite layer of constant thickness cn1bedclcd in an infinite elastic hon1ogencous isotropic
rnedium. The propagation of a harmonic plane v.,rave across Huch a layer can Uc described R'stt
easily. Only the continuity of clisplaccrnentt.; and tractions at the interfacetl arc needed.
As shov-rn in Fig.3.11 1 \\TC fin.;t consider the case of a norrna1ly incident plane SH-wave
(amplitude Ascr ): there are two reflected (amplitudes RsH and R~H) and two transmit-
T'sH
ted waves (amplitudes Tsn and T~H ).

h
µ,p µ', p' µ,p
z~
x
vibration isolation reduced
system vibration Fig. 3.11: Propagation of a normally incident plane SH-wave through a single

_l\j\/\r --- ----- --- .... --1\JV\r infinite layer.

The equations of continuity arc thus written:

h
• fir.st interface:

- displacement:
Fig. 3.10: Simplified analysis of vibration isolation by screens and trenches. (3.80)

- tract'ion:
For such a plane SII-v..rave, the cxpreHsion of the displace1nent field is the following: µAsH - - = µ ,T~'H
- - - µR8n - - - /L ,R~H
-- (:J.81)
Vs Vs V~ Vg
• second interface:
(3. 78)
displacernent:

T~Hcxp ( iw
V~h) +R~ffcxp ( -VJ/ iw )
iw ) =Tsncxp ( vs'r (:J.82)
-V\rhere Vs is the shear wave velocity.
traction:

Equations of continuity. Considering the expression of the diHplacement field, Eq.(3.78), [

V~, TsHexp
JL 1 (. )
iw
V::'.th
1
-RsHexp (-~~,h)]
1,w
. µ 1'.sH exp
= ,-,-
'S
(iw- h)
Vs
(3.83)
the only non !:';Gro ccnnponent of the streHS tensor is:

Du (3.79) \i\Te shall also consider the approxirnation pc:::: p 1 ) leading to a simple expre~Hion of the
ayx =µ,DY
;r ratio tf,;;J using the shear vvave velocity contraHt xs as follows:

(3.84)
 2D /3D-Virave propagation 3.3 \!\Tave propagation in unbounded rncdia 157
156

The folluviring syHtcrn is thus derived:

R,sH T~H .R~H __
1.0 /'.~:.~~-:'-·
l+------- 0
AsH Arn AsH ! .'/ :'/
i f :· .'I
RsH T~H R~H
1- - - -xs--+xs--
=
0 0.8 !//// ········· ...
As11 AsH AsH ·····.
(3.85) r·: .1 parameter a
0.6 '!~.::
e+ -
Tf3H
- + e- -
R's,11
- = Ts11 ,,,,
:.:., 0.0
AsH Arn 11::1 0.25
;;:,:1
T,5,-11- e .- -
e+ - R's'H
- Ts11 0 4 ..;,, 0.5
Arn AsH xs . r.:: 0.75
=t:
I·;:
0.9
where e+ =exp ( ~ h), e- =exp (- ~ h) and the transmission coefficient Tis: a.2 .
=t:
0.95
0.99

TsH TsH
= - - exp (iwuc-h) a.a~~~~~~~~~~~~~~~~~~~~~

a
AsH 's 2 3 4 5
velocity ratio
Transmission coefficient. The solution of Eqs (3.85) leads to the expression of the
transmission coefficient Tsr-r= Fig. 3.12: Transmission coefficient of a plane SH-wave through a single layer:
4xs influence of the velocity ratio.
(3.86)
TsH =
(1 + xs)2c- - (
1 - xs )?-e+

\vhcre xs is the flhear vvavc velocity ratio such as:

From Eq.(3.89), the influence of the velocity ratio on the transnrission coefficient is

xs = v:
V:'
(3.87) obvious. Since the velocity ratio and its inverse a.re directly combined in Eq.(3.89), the
transrnission coefficient may be the same if the layer is stiffer than the infinite 1nedium,
VS = xs V 9 > Vs (xs > 1), or if the layer is softer than the infinite mediurn, = v;.
Influence of the velocity ratio on transmission. The transn1ission coefficient TsH xsVs < Vs (xs < 1). The variouR curves tlisplayed in Fig.3.12 correspond to different
iR complex \ra1ucd and includes one ter1n related to the arnplitude and another one related values of a (0.0, 0.25, 0.5, 0.75, 0.9, ll.95 and 0.95 resp.). The maximum value of the
to the phase. Both terms are influenced by the propagation through the layer. In order transrnission coefficient, ITs11 I = 1, is reached for a velocity ratio xs = 1 (leading to
to assess the amplitude reduction through the layer, the modulus of the transmission v~, = Vs'). Starting from xs = 1, the transmission coefficient decrea8e.s for both a
coefficient wi11 no'\v be derived. decreasing or increasing velocity ratio. This decay is rnuch faster for small values of a:
the minimum values of ITsHI are 0.8 for xs = 2.0 and 0.6 for Xs = 3.0. The velocity
Considering the con1plex exponentials e+ and c- under a simple for1n: c+ = a+ ib ratio has a strong inf:iuence on the transmisHion coefficient. Nevertheless, for values of a
and c- = a - ib 1 the n1odulus of the trans1nisRion coefficient can be written as: close to 1, the transn1iRRion coefficient is not far frorn 1 and there is nearly no decay. The
pararneter a also has a significant influence on the transrnission coefficient. It is inainly
IT<s11. I_
-
2xs
y'4a2x~+b2(1 +x~)2
(:>.88) related to the layer thickness h and '\vill be diRCUHsed in details in the next paragraph.

1'aking into account the fact that a2 + U2 = 1 1 the previous expression can be furth:r Optimal layer thickness. The transmission coefficient is now displayed as a fnnction
sirnplified: of the layer thickness/wavelength ratio for three different values of the velocity ratio:

ITsHI = [ a + 1 ~a' (:s + xs) ']-~

Xs = 1.5, xs = 2.0 and xs = 3.0 (Fig.3.13). For a very thin layer, the transmission
2
(3.89) coefficient is nearly 1. The minirnurn transmission ir.; fur a layer thickness equivalent
to a quarter of the wavelength h = A'/ 4 with A' = VJ,/ f (f: frequency). Whereas
1vhere xs is the velocity ratio and a= cos( {f, h). for h = A' /2 (parameter a = 1 in Eq.(3.89)), the transmission coefficient is constant
s l'T:'JHI = 1. In such a caHe, the incident ¥.rave it; totally transmitted through the layer
 2D /3D-wave propagation 3:~1 Wave propagation in unbounded rncdia 159
158

Vi.rhatcver the velocity ratio. As shuvvn in F'ig.3.13, the trans1nission coefficient is periodic Influence of incidence on transmission. For a SH-v.ravc 1vith a non normal inci-
"\vith period A' /2 and the isolation efficiency iH sin1ilar for thicker layers. dence (Fig.3.14), the expression of the displace1nent field is now as follov.rs:

The tra.nsrnission coefficient is thus mininnur1 for an optimal layer thickness corre-
sponding to a quarter Vi'avclength. l?or this thickneHs, related to a = 0 , Eq.(~-L89) may (3.92)
be sirnplified in the follo\.ving form:

. 2xs (3.90)
ITsHI = f(xs) = -.--, ¥there Vs is the shear vvavc velocity a.nd f) the angle of incidence.
l +xs
The only non zero con1ponent of the traction at the interface iR the follovving:
'I'his expression is useful for applications in vibration ir.;olation (rninirnu1n transmission
coefficient). As already rnentioned prcvioutily, the isolation efficiency is the sa1ne for a (3.93)
stiff layer with a velocity ratio xs and a soft layer with a velocity ratio 1/XS· It is due
to the fact that: The Snell-l)escartes equation can be Virritten:

f(xs) = f (_1_)
xs
(3.91) sin 8 sin 8'
(3 94)
Vs V's
HoVii'ever, the corresponding optiinal layer thicknesses, h =A' /4, are different in both The same type of equations as in the case of nor1nal incidence are obtained. The ratio
cases since the wavelength is larger for the softer layer. p/'Vs / p V/,· is nevertheless replaced by the following ratio:

µ'Vs cos8 1 cor.;8 1

--, - - " " x s - - = - -
xs J1-x-sm e
2
5
. 2
(3.95)
p,V8 cos8 co88 cosB
1.0
This ratio vvill then he denoted Xs.

0.8
µ,p µ', p' µ, p

0.6

OA xs=l.5
R'stt __ e~: L'sH
xs=2.0 T'sH eT.
xs=3.0
0.2

h
o.o'--~-_J_ ___ l _ __ __,___ _ _~---"----~

0.0 0.25 0.5 0.75 1.0 1.25 1.5

layer thickness/wavelength ratio z~
x
Fig. 3.13: Transmission coefficient of a plane SH-wave through a single layer:
influence of the layer thickness as a fraction of the wavelength. Fig. 3.14: Propagation of a plane S.H-wave with oblique incidence in a single
infinite layer.
 r
]60 2D /:JD-wave propagation 3.3 \i\rave propagation in unbounded rnedia 161

Equation (3.95) is expressed as a function of the incidence angle e. A critical incidence Displacement and traction. As shown in Fig. 3.16) for an oblique incident _P-ivave
angle is obtained for sin (;l = 1/ XS· In Fig.3.15, for three different velocity ratios (xs=l.5 1 the vvaveficld is converted into tvvo vvave types: .P--vvave and SV-wave. The displaeenicn~
xs=2.0 and xs=3.0), the transmission COefficicnt increases belOViT the Critical incidence fields are the follovving:
(ranging frorn 40 to 20 degrees resp.) 1 reaches its rnaximnrn value ITsHl=l, and decreases
very rapidly above the critical incidence. As shov,rn in Fig.3.15 1 the velocity ratio has a • incident P-wavc:
strong infiuence on the results:

• the trans1nission cocIT1cient for a nor1nal incidence ranges from 0.92 for xs=l.5
(i-P)_A
11.:r; -
.
p cos e,ir exp ['Vp
w (x cos (Jip + z Sln. er - l
Vrt)
down to 0.6 for xs=3.0, (i-P) Q
(:J.96)

• above the critical incidence, the transrnission coefficient decay is much faster for
11.y =
.· or
u.z( i- P) -- A ps1n [ zw p . p
-i exp Vp (xcos(}'i +zs1nei -Vpt)
l
the largest velocity ratio values.
\Vhere Vr is the .P wave velocity and ef the incidence angle of the P-'v.rave.
• resulting (reflected) SV-wave:
1.0 ....,------- ..
.(_-;

,,
-'?"" \
\\
"
"'-, (SH-wave J
~
'1
Q)
-~
0.8 --- I
\I
".
" ',
"Q)
-~

:to I ',
(3.97)
I
\ \ -15
"Xs- ·
0
"'1 0.6 I ',
I Xs=2.Q \
0 I •
·~
w Xs=3.Q I \
w
·~
sw 0.4 \ \, where Vs iH the shear v,rave velocity and (}~v the incidence angle of the SV-wave.
'1
\\ "·..
"',.. ',, '··...
~
0.2 ',
·· ...... "
'•,
' 7c,µ, P A',µ', p' 7c,µ, p
o.o L__J_ __J__ __[__--1::::::-::==d.-~i-=.:--=.o~-~-
'............
--- •,

o. 10 20 30 40 50 60 70 80 90
incidence angle (deg.)

Fig. 3.15: Transrnission coefficient as a function of incidence for different

velocity ratios: xs=l.5, x.s=2.0 and xs=3.0 (single-layer case, SH-wave).

Transrnittivity of a plane P-wave through a single layer

z~
For a norn1a.l incidence, the results \vith P and SV-Vlraves arc identical to the previous x
ones. In Fig.3.15, the ,tra.nsrnission coefficient estirr1ated for a norrnally incident SH-
wave (ITl~0.6 for x~:J.O) will be the oamc for a normally incident P-wave (velocity ratio Fig. 3.16: Propagation of a P-wave with oblique incidence in a single infinite
nevertheless expressed in Vp ). However, the velocity ratio is different for a P-\vavc (the layer model.
velocity involving a,\.+ 2µ ter1n instead ofµ only for shear waves).
 ' I

162 2D /3D-wave propagation 3.;) VV"'ave propagation in unbounded inedia 163

l
Considering the expression of the displaccrncnt field (P-·v:..rave)) the traction at the • second interface:
interface is the follo-\ving:

cosB{'u~"-P) +" i:\ (;;;p cose{'u.~i-P) + {:,, sinBzu~i-P)) - first com,ponent of displacement:

!
2iµ;Jp
(J(i-P) "n = l) (3"98) Tp cos8f,efa, -Rp1 cos8-k,cp, + T~,vsinH!j,YeJv,
1

~ - " ( w " eP (i-P) + w eP (i-P))

+ T sv sin (3"104)
" esv
+ R 'sv sin R/ e-817 , = T'p cos eP T c+v
" esv
'l/.l Vr 8111 i Ux \/p COS i ?Lz
rep+ 8

Si1I1ilar expressions arc also obtained for other \vavefields. third component of displacement:

Equations of continuity. The Snell-Descartes equations can be written: Tpr sin8f,ct, + R1-" sin8:k,ep, -T~'V coHB~Y et17 ,
+- T + (3"105)
" eP sin &!j.'! sinesv +R'sv COR esv - -- T p Slll
R' eSV' " errep . sv cos, esv
T Esv
" eP
8111 i sin ()~_v SID T' _ _T_
(3"99)
Vp Vs VJ, V:'s Vs - first com,ponent of traction:
'Ihe equations of continuity yield:
Tfc,
V' (/\'' + +v
+ 211' cos 2 ePr' )er, R'p (/\'' + 2µcos
' 2 eP ) -
·n, er,
• first interface: p p

+ T~v I sv +
V.' /L Dln 2(} 1 , e S'V' -
R'sv
Y.' µ,
f • eS17 -
Hln 2 RI e SV' = (3"106)
- first co'!nponent of displaccrnent: s s
Tp (
-V :\+21tcos,2 Oy + Tsv " sv +
r) ep+--µsmWT c8 F
p
(Ap-Rp ) case,+ 1''SF81Il
." esv
R --
r Vs !
(3J 00)
(Tfc, - R'p) cos e!j,, + (T~F + R~v) sin e:jY third cornponent of traction:
third cornponent of displacernent:
,T~ ._ 2eP
µ V' sin
+ ,R'p . eP -
r1ep, -11, - 1 D1n2 n/ep,
(Ap + Rp) sin er+ RsF cos O~v = P Vp
(3"101)
. p' + R'
(T 'P ) Slll
" aPvT' - (T'SV - R' " esv
. SV ) COS 1 T,[,v , , sv + / R'sv sv - _
T' -11, - , - cos 2(}T, e SV' - µ -V:' cos 28 R' e SV' - (3"107)
v:g s
Tp " eP + Tsv sv +
µ-s1n2 rep-µ--cot->2HT e v 8
Vy Vs

vvhere ej, ej, ej and ejr are the same notations as e+ and e- for SH-v..rave but involving
(3"102) V; for index j and VJ for index j' Ci = P. S)"

Transmission coefficients. For aJ1 incident P-wave, Fig.3.17 gives two types of re-
sults: ITPI transmission coefficient for P-waves (top) and ITsvl transmission coefficient
for SV-waves (bottom). Three curves are displayed in each graph corresponding to three
th'ird component of traction: difl'erent velocity ratios: Xp=l.5, Xp=2.0 and Xp=3.0. For a nor1nal incidence (Br =0),
Ap " P Rp " P Rsv esv
ITsvl is always zero. For an oblique incidence, as shown in Fig.3.17 (top), the value.s of
µ-sm28" -µ-sm28R-µ--cos 2 R =
Vp ' Vp Vs
ITrl exhibit very different variations:
, T~ . eP ,T~,v sv • for the largest velocity ratio (xp=3.0), the transn1ission coefficient ITPI i8 alwayt->
µ, Vj, Slll 2 T' - µ, V~ cos 28 r' (3.103)
belovv 0.6,
Ii'p . P
-µ 1 - 1 Slll 28 R' -
, R'sv
µ --,- COS
sv
28 R' • for the lcnvest velocity ratio (xr=l.5), the transmit->sion coefficient ITPI iH above 0.6
Vp V8 for incidences up to 45 degrees 1
164 2D /3D-wavc propagation 3.3 \Vavc propagation in unboun<lc<l rnc<lia 165

• for xp=2.0; the trans1nission coefficient ITp I is generally in bet\VCCil the previous
cases. Its is nevertheless lower than the xp=3.0 case betv.reen 30 and 50 degrees
but higher than the Xp=l.5 case for all incidences above 50 clegreefl.
It is even more co1nplicatcd for ITsv I, the transn1ission coefficient of S'V-1.vaveH (Fig.:i.17,
bottorn). For instance, the lowest velocity ratio (xp=l.5) leads to the lowest trans1nis-
sion coefficient for inci<lenceH belu\v 40 degrees. Since the amount of transmitted P-waves
is larger in this ca.Re, the contribution of converted SV-vvavcs is rnuch lu1,ver. To investi-
gate the optirnal configurations, both transn1ission coefficients ITP I and JT~rF I should be
sirnultancously considered.

Transrnittivity of a plane SV-wave through a single layer

x,,=1.5 P-wave
Displacement and traction. As shown in Fig. 3.18) for an oblique incident S'V--,,v:-_tvei
0.8 --------,,x,=2.0 ··...\ JTl'I the wavefield is converted into two Vilave types: P-vn:1ve and SV-wave. The <lisplacement
fields are the following:
'\\
0.6 \
• incident SV-vvave:
\

I
I (i-S'V)
ux-' ·
=Asvs1nB,jS'V
exp [iu.,•
Vs(xcosBjS'V +zs1nOi
· $V
-Vst) l
0.4
\\............ '\.~ ..................... ..... (i-SV) Q (3.108)
··... Uy =

0.2
. ..... .....

''
'' ·u1i-SV) = Asv cos ef'V exp [ ~ (x cos efV + z sinefV - Vst)]
··---~~'
--~!t
-,,vhere Vs is the shear vvave velocity and B,fv the incidence angle of the SV-vvave.
0 ·0oL.----110_ _2J._O_ ___J30--4J._O_ _J5_0__6_l_0_ _7L0__8_L0_-=90
1.0 ~-----------------------, • resulting (reflected) P-wave:

P-wave ['W (x cos

..
u:r(R-P) -R
- p cos.gP
R. exp Vp gP
R - z. .sin
· gP
R - Vpt )]
0.8 x,.=3.0 I Tsvl
v.£R-P) = Q (3.109)

0.6
(R-P)
Uz = R ps1nBHexp
. P [ iw ( P
Vp xcosBR . P
-zs1nBR -Vrt) l
-,,vhere Vp is the P wave velocity and ef the incidence angle of the P-wave.
0.4

Equations of continuity. At the second interface, the continuity conditions for both
0.2 displacernents and tractions are the same as in the case of an incident P-v·:ave. One must
rewrite the continuity conditions at the first interface only1 thus:

30 40 50 60 70 80 90 • first interface:
10 20
incidence angle (deg.)
first cornponent of d'isplaccrnent:
Fig. 3.17: Transmission coefficients: ITPI (top) and l'Twl (bottom) vs incidence
for different velocity ratios: Xp=l.5, Xp=2.0 and Xp=3.0 (si11gle-layer case, - Rp cos fJ RP+ (A sv + Rsv ) sin Hi'v =
incident P-wave). (3.110)
(T}o - R'p) co' Bf, + (T~v + R~v) sin Bf,''/
 2D /3D-vvave propagation 3.3 \iVave propagation in unbounded lnedia 167
166

A.,µ, p A,',µ', p' A.,µ, P 1.0 r ,·,

------- " :. I
I \
\ : SV-wave
~
<I r
r
'\ A".
I' ··-1._ • ••r'
I \'
I Tsvl
r
-~
CJ 0.8 \ J ',.... I
:1
i:S"CJ
\ \
:1
0
\
\
\
... , :1
I
I
I
"<I 0.6 I
I I
I \
I
0
-~
w
\
I
I ".. xs=L5
w \ I
I
\ I
-~

sw OA \
\ I
I
I
'x -2 · o"\
I S-
<I
,.,"' xs=3.0"
~ \
""
zk: 0.2 \
\

'' ' '

x
0.0
0. 10 20 30 40 50 60
--70 ----
80 90
Fig. 3.18: Propagation of a SV-wave with oblique incidence in a single i11finite 3.0
layer model. /\
I \
I I SV-wave
2.5 I \
~
<I I I ITPI
CJ I I
-~
I \
third corrtponent of displaccnient: "
-~

tt:i 2.0 I
I I
\
CJ
. p
Rp sm8R + (-Asv + Rsv) cos&.; =
sv 0
"0<I I
I I
\ . ,--,
'

(T)o + R'p) sin ef, - (T~v - R~v) cos 8¥Y

(3.111) 1.5 I
I I: ,.
-~
w I :,
w : \
'§ \ \x
' s=1.5
- first cornponent of traction: w 1.0 \
<I \

"',., "
\

A Rp p Rsv sv ~ " xs=2.0 ',

\

SVµsinW.SV + -(,\ + 211cos' HR) - --µsin WR = 0.5 ~"

Vs ' Vp V:s r
- xs=3.0 '' '

T'
--1i (A' + 2µ' cos
2 p
Br,)
TS'v 1 sv
- -,-/I sin WT' (3.112) 0.0
-~-::"-.
r
r
·"
r
,.- -- ' '
--" -:::.::.:..
•,

Vp V3 0. 10 20 30 40 50 60 70 80 90
R'p (,\, + 2p,, cos 2 on,
+-, ,,p ) - R~v ' · 2nSV
-,-p sin un,
incidence angle (deg,)
Vp V8
Fig. 3.19: Transmission coefficient: ITsvl (top) and ITPI (bottom) vs incidence
third cornponent of traction: for different velocity ratios: xs=L5, xs=2.0 and xs=3.0 (single layer case,
incident SV-wave).
Asv
-11-- cos2Bi
sv - µ,-
Rp s1n2un
. nP -11-Rsv 2nSV
- - cos uR. =
Vs Vp Vs
/lj, . r ,T.~v , sv
µ - , Slil 2()'T, - µ -,- cos 2By1 (3.113) Transmission coefficients. For an incident SV-wave, Fig.3.19 gives two types of re-
Vp Vs sults: l:Z:wl transmission coefficient for SV-waves (top) and ITPI transmission coefficient
1 R'p 0p 1 R~'V 2nSV for P-waves (bottorn). For a norrna.1 incidence (efV =0) 1 ITPI is alvirays zero. For an
-µ -V' Slll 2 R' - µ - - COS uR'
p V:S, oblique incidence, the values of 17:s'vl sho-\v very strong variations (Fig.3.19, top). It iH
due to the fact that, for an incident SV-\0.rave, the P-\vavc incidence is larger than the SV-
 2D /3D-wave propagation 3.3 \i\Tavc propagation in unbounded media 169
168

wave incidence Rnd a critical angle thus appears. It leads to an inhornogeneous P-v,ravc
(i.e. interface Virave) propagating vertically and a very la.rgc trantnnission coefficient ITPI
for P-"\vaveH (Fig.3.19 1 bottom). Depending on the velocity ratio, the rnaxi1nurn value
of !Tri ranges from l.3 (xs~3.0) to nearly 3.0 (xs~2.0). Such transmission coefficients
(x~) than between the infinite medium and the first layer (xs). It should lead to a lower
transrnission coefficient than the one-layer case.
1 0 solve the two-layer case) the equations of continuity for displacements and tra.ctions
1

larger than 1 are nevertheless valid fron1 an energetic point of vie¥.' since the refracted rnust 110\V be v.rri.tten at all three interfaces (Scrnblat 1 1989):
P-wave dues not actually propagate beyond the layer Rincc its direction of propagation is • first interface:
vertical (exponential decay in the horizontal direction). For applications in vibratory iso-
11
lation, such conclusionR (local "arnplificatiun but strong Hpatial decay) should be taken displace1nent: AsH + RsH ~ T~ 11 + R~H
(3.114)
into account. { traction: AsH - RsH = xsT~H - xsR~H

Transmittivity in the m11ltilayered case • second interface:

Various configurations. The case of an infinite single layer is very si1nple to solve displacerncnt: Cr+T'
- SH + el-R'SH +Tn
= e21 811 -R"
+ P21 "SH
since it is governed by tvvo parameters only: the velocity ratio x and the layer thickness (3.115)
h. In the n1ultilayered case 1 several velocity ratios and layer thicknesses have to be { traction:
considered. One has to deterrninc the optimum layers arrangernont depending on the
isolation level required. In this section, different rnultilayered cases are compared to • third interface:
the one-layer case. As sho\vn in Fig.3.20) "-re shall especially conr.;ider two alternative
displacement:
configurations involving two layers. (3.116)
{ traction:

one-laye:· 1~,;1 • two-layers

• where et= cxp(iw~~)) ei = exp(iwh 1 ~h 2 ) 1 e~i_ = cxp(iw~~) and e+ = exp(iwh 1 tcih 2
)

(similarly for c;-, e2 and e21 ).

VOr the r.;econd interface, we obtain a factor l/x~ due to the velocity ratio between
identical both layers which iR larger than in the one-layer case. The definition of the transmission
coefficient Ts H being now the follo-,,ving:
two-layers
er _ TsH +
' S HA
--e
SH
Fig. 3.20: Various multilayered configurations: one-layer (top left), two-layers
(top right) and identical two-layers (bottom). its expression is obtained by solving the equations of continuity at the three interface.s:

er 4xHfi(xsl + iJeiei
'SII ~ -,-~~~~~~~~~~~-'-~c__o._"---~~~~~~~ (3.117)
(1- x~)(x~fi(xs) - l)c~+e 21 + (1 + xsl2(x~fi(xs) + l)c~i
Two-layers case. In the t1,vo-laycrs case (Fig.3.20), we consider a norrnal.ly incident
.. ) _ (xs + l)e1_ - (xs - l)ei
SH-wave in a free-space including two infinite layers of constant thickness: vvith f 1 ()(S - (xs + l)e 1 + (xs - l)ci
(3 118)
• the first layer having a shear wave velocity V1 , the velocity ratio with the free-space
E'or a normally incident SII--,,vavc, the transmission coefficient TsH is rninilnum for
is xs ~Vi/Vo. Its thickness is denoted h1,
the follovii'ing layers thicknesses: h 1 = A 1 /4 and h 2 = A. 2 /4. It may then be expressed
• the second layer having a shear 1,vave velocity V2 , the velocity ratio -,,vith the free- under a simplified for1n:
space is chosen as X~s = l/xs leading to V2 = l/V1 . Its thickness is denoted 2x~
ITsHI = g(xs) = ~ + , 1 (3.119)
h2· Xs
If the first layer is stiffer than the infinite rne<lium (larger velocity), the r.;econd one will The function g(xs) has the same property as the fonetion f (xs) (defined by Eq. (3.91) ),
be softer (lower velocity). The aim is then to have a higher contrast bet-,,.veen both layers that is: g(xs) = g(1/xs). The two-layers case leads to the same results with the first
170 2D /31)-\vavc propagation 3.3 VVave propagation in unbounded n1edia 171

layer having a velocity ratio xs (1/xs for the second layer resp.) or the first layer having Identical two-layers case. In the 1'identical t\vo-layer8ri case (F'ig.3.20, bottom), '"' e
a velocity ratio 1/XS' (xs for the second layer resp.). Furthermorc 1 the velocity ratio consider tvvo identical layers crnbccldccl in a free-space. 1'he t\vo layers a.re separated by
between both layers being 1/x~) the isolation efficiency should be inuch larger in the an infinite layer having the sarnc properties as the free-space. 'l'he main objective is now
t\vo-laycrs than in the one-layer case. Considering the expressions of the functions f to assess the efficiency of a configuration involving t-,,vo layers vvith identical properties
and g (Eqs (3.91) and (:Ul9)), this velocity ratio directly appears when comparing both and to perforrn a cornparison with the two-layen.; case Rtudied previously.
function8:
g(xs) = =f J(x~)
(:J.120) (~)
Xs
The equations of continuity vvill novv be considered at the four interfaces:

• first interface:
JiOr a norrnally incident SH-wave and a given velocity ratio, it n1canR that the iHola-
tion efficiency in the one-layer case i8 rnuch sn1aller than in the t\vo-layers caHe. The f displace1nent: Arn+ Rsu = TfH + R~H
and g functions arc displayed in Fig.3.21 to a.sse88 the infinence of the velocity ratio in (3.121)
{ traction: AsH - Rs11 = xsT~H - xsR~'H
both cases.

T·he reRnlts of the t-,,vo-layers case for an oblique incidence are displayed in Fig.3.22. • second interface:
They correspond to the follovving layer thicknesseR: hi = Al/8 ct h2 = A2/8 (where 1\.i displacement: ,+
f~11 T'SH + ell
- R'Sil +y"
=el SH + e_l-R"SH
are the wavelengtht> in both layers). T'he cornparison vvith other caseR will be di::;cusscd (3.122)
in the follovving. { traction: ,+ T'SFf -
f-11 ell
e-; T"8II
- R'SH -- xs - .:.LR"
xs SH

• third interfa.ce:
1.0 /

/
/
displacen1ent: P2+y"
SH + C2,-R"SH= e12· SH + el2
+ T'" - R"'811
/ (3.123)
I
{ traction: ,+yu -R SI-I
C2 SH - e2
11 + T SH
-- Xs'e12
111 . - R"'
- ·xsc12 SH
"
0.8 ···---·!- "
/·
I: • fourth interface:
I

I
I g(x). displacement:
0.6 --·. ___ (
I
. -- . _,_. -- ' . -... --- ---
..
(3.124)
I { traction:
I
I
I
I -,,vhere c+ = cxp(iwh 1 ). e+ = exp(iwhl+h 2 ) e+ = exp('iwhi+h 2+h 3 ). p+ = exp(iwfu.)
0.4 11 V1 · .l2 V1 ' 13 V1 , 1 Vo '
I
e,+ _ (:, h,+h1).an d e+ -_ exp (··
LW h,+h,+h.') l· f.or e 111
.. ·1 ar.y
(s1n11 - P--12 , 1: 13
- , e 1-- cin
. d c--)
I 2 - exp iw v2 3 170 2 .
I
I
I T·he expression of the transn1ission coefficient Ts11 for the identical t\vo-layers caBe i8
0.2 , __
I the follo\ving:
I
I

I
I
2xs h(xs)-1
( ) ,--- ,+ _ + - (
c12c11e 1 c2 e3 1 + h(xs)--
1-xs)
01.L~~.L_~~'--~~'--~~'--~~'--~~'--~~'--~~
11 xs - xs 1 + xs (3.125)
0 0.5 1.5 2 2.5 3 3.5 4 T5H= (Xs + 1 )e13 e+ - (XS - 1) f2 (Xs )' .
e12 e]'3
velocity ratio

Fig. 3.21: Comparison between the one-layer (dashed) and the two--layers
~'./~:\~~~ (xs + l)ei-ei + (xs - l)e;
with f 2 (xs ) = Mxsl+xs (xs - l)e2-e+ + (x.S + l)e-2
(3.l 26)
(solid) cases for a normally incident SH-wave: functions f and g vs velocity f1(xs)-xs · 1 2
ratio X· The results of the identical tvvo-layers are diHplaycd in Fig.3.22 for the follovving valuet-i
of the layers thiekne8Ses: h1 = A,/8, h2 = A2/ 4 and h, = A3/8 respectively.
172 2D /3D-\vavc propagation 3.3 V\Tave propagation in unbounded media 173

Comparison of the rnultilayered cases

Incident SH-wave. The resultH shovvn in Fig.3.22 assess the better efficiency of
the two-layers case when compared to the one-layer and the identical t\vo-laycrs ca,t-;es.
Only for very large incidenceR does the two-layers case give larger transmission coeffi-
cients. For very s1nall incidences, the identical tvvo-layers case is more efficient than the 0.6 one-layer
one-layer case. A.t other incidences there is a slight difference bct\veen both cases.
TP two-layers
id. two-layers
1.0
' \I 0.4
I
I - ----
SH-wave I
+' I I . " ·"
'1
'"
-~
u
0.8
I
I one-layer "' I
I
I
w:1
"--< '
:'"
' I two-layers
I
\
'"
0 I
I :' \ I
I
0.2 \
u 0.6 \
'1 I
I ', I
',I
id. two-layers \
.sw \

' '..... .... ____ _ -----

I ',I
w
,,.
I,•
/,' \1
sw
-~

04
I
I 0.0
'1 0. 20 40 60 80
oj

'""
+'
~\ "·
0.2 I ' \I
\ 0.8 I one-layer
\
.,_ ...............
'' ' ---- +'
~
I
I
I
I
Tsv two-layers
o.o 0 _ I I
20 40 60 80 '"
-~
u
I I --------
incidence angle (deg.) s 0.6 I
I I
I
id. two-layers
'"
0
u
I
I
I

I
Fig. 3.22: Transmission coefficient vs incidence for various multilayered con- ~ I
0 I
figurations (incident SH-wave).
-~
w 0.4 I
:·· I
w
-~ I
I
\ I
\ ,.. ·" ---..
sw I
/
··\
I
.. -- ---- ·"
'1 I
oj
0.2 I
I \
\
.
'""
+' I \
Incident P-wave. \\Te a.lso considered the n1ultilayercd cases for an incident P- I

wavc. A.s shown in Fig.3.23, the t\vo-layen.; case is much rnore efficient than both other
"
-"
\
''
'
.... ___ .... ----
cases (except for large incidences). The identical tvvo-layers case has a con1plex behaviour 0.0
since, for large incidences 1 it leads to lcnver transmission coefficients for P- and SV-waves 0. 20 40 60 80
than the one-layer case. For small incidences, it leads to close results for SV-vvaves and incidence angle (deg.)
lo\ver or larger results for P-waves depending on the incitlence range.
Fig. 3.23: Transmission coefficients ITPI (top) and ITsvl (bottom) vs incidence
for various multi.layered configurations (incident P-wave).
174 2D /3D-wavc propagation 3.4 Spherical vvaves 175

3.4 Spherical waves

3.4.1 Wave equation
1
u,.(r,w) =A ( - r' -iw )
+ rVr [-iw
exp Vp (r- Vpt) l (3133)

Frorn a practical point of vieVir, the case of plane \VaveH studied in the previous sections is
The quantities f(.) or A are determined through the initial and the boundary con-
very useful since, for the far-field, it is poRt:>ible to assume a plane v.ravefront. However, for
ditions of the problem. T·hus, in the caHe of a Hpherical cavity of radius R loaded by
the analysis of near-field propagation, this approxirnation is no longer valid. The solution
then depends on the nature of the displacerncnt (or stress) field induced by the source. In a radial pressure, p(t), we get D",.,.(R, T) = -p(t). Knowing the displacement field, we
this section, vvc consider the case for 1,vhich the enritted \vavcs have a spherical synunetry. n1ay derive the stresses originating in the rnedium considering its constitutive laVir (linear
elasticity herein). \\Te recall that, for spherical syn1rnetry, the only non zero cornponents
Under this assun1ption 1 the only difiplaccrncnt corr1poncnt is 1 due to the syrnrnctry, of the stress tensor arc O"rr, o-ee and cr'P'P" 1'hese variables have the follo\ving expression:
the radial one denoted 'Ur· Considering a spherical coorclinatc system, the equations of
3·u,. "Ur
propagation are thuH reduced to: D"rr = (>. + 2µ)- + 2>.-
0r r (3.134)
{ Bur Ur
(3.127) aee ="Vi'=.\-[) + 2(.\ + 1i.)-
r r

The equation governing the potential is expret-isc<l under a sirnilar forrn: 1'he err'( co1nponent corresponds to the radial effect vvhereas the o-oe and o-'Pr.p cornpu-
nents are related to purely tangential effects.
32¢ 2 3¢ 1 32 ¢
-+--=--- (3 128)
3r2 r 3r Vfi [)t2
Considering these results, Vire can for instance deterrn.ine the eigenrno<lcs of the entire
B~arth assuming a homogeneous elastic Hphere:\.
Choosing the new variable cp = r¢, Eq.(3.128) may be simplified:

(3.129) 3.4.3 Geometrical damping

Equations (3.132) and (3.133) show that the wavefront, location of the pointR with con-
3.4.2 Solution wavefield stant ¢, corresponds to a sphere centered at the source. 1'he vn-ivefront expands during
The previous equation corresponds to the Vira:ve equation of longitudinal \vaves leading propagation and the arnplitudc of displaccrnent decreases because of the conservation of
to a general solution of the form: energy. 1'he expansion of the vvavefront thus induces an arnplitudc decay; this decay,
similar to the effect of damping, is called geon1etrical darnping. It is irnportant to notice
( t) 1 that it is not related to da1nping that could be due to the dissipative properties of the
¢(r, t) = ~ = - [f(r - Vpt) + g(r + Vpt)] (3.130)
inediunr. In this section, the n1edium is non dissipative but the geometrical da1nping
r r
leads to a spatial arnplitu<le decay only due to the expansion of the wavefront.
In this equation, the two termH correspond to waves propagating away from. the source
and vvaves converging to the source respectively. In a free-space, no wave reflection occurs
As shown in Eq. (3.133), the amplitude decay in termH of displaeernent has t\VO different
and no \vave converging to the source may appear: the g(r + Vpt) is thus zero. This
forms:
condition is knovvn aH the So1nmerfeld radiation condition. 'l'he displacement is now
given by:
• a fast decaying ter1n (1/r 2 ) corresponding to near-field (i.e. nea.r ::;uurce) effects,
3¢ 1 1 ,
-ur(r, t) = -[) = ---,f(r - Vpt) + -r f (r - Vpt) (3.131)
r r~ • a slovver decay (1/r terrn) corresponding to far-field effects.
In the case of a monochron1atic \.vave, expressionH sirnilar to Eqs (3.130) and (3.131)
. 3 R,ealistic Earth models are available from various teams. The corresponding ai1alyse8 gcnor1:1lly
arc derived:
[•iw
¢(r,w) = -A exp -(r - Vpt)
r Vp
l (3.132)
involve both rotating and self-gravitating assumptions \vhich lead to much rnorc cornplicutcd equations
(Chaljub et al., 2003).
 2D /3D-wave propagation 3.5 VVaves in a homogeneous or heterogeneous half-space 177
176

Waves in a homogeneous or heterogeneous half- From Eqs (3.1:16) and (3.138), we obtain that, at the free-surface z=O:
3.5
space J t/il = 11AsH ~cos ei exp [ ~ ('.r sin ei - \!st)]

3.5.1 Surface waves: SH case

As depicted in Fig.3.24) v.re shall 11ovv consider the case of a plane monochron1atic EJH-
l I(R) = -µflsH~ coseR exp[~(xsineR-Vst)]
(3.139)

\va.ve reflected at the free-surface of the soil (assurr1cd aH a homogeneous half-Hpace). The Introducing this expressions in Eq. (3.137) and writing that this condition 111ust remain
incident rl_isplacement field reads as follo\vs: valid at every tirne t and every location x, ·vile thus derive:

u~1 l = AsH exp[~ (xsine; + zcosRi - \!st)]

(3.135)
Asu = Rsu (3.140)

The displacen1ent resulting fro1n. the superposition of the incident and reflected ¥.raves
{ uUl = u(i) = O
x '
ha.s the follovving expression:

(i) +u'lj(R) = 2AsHCOS

'Uy= 'U'(j
· ·
(wzcosei) exp -(.rs1nH'i-
Vs
[iw .
Vs
i~s,t) l (:J.1'11)

free surface The reflected ¥.'ave has the same amplitude as the incident Vi,rave and the angle of
reflection is equal to the angle of incidence. At a depth z, the total amplitude of the
1notion is:
wzcosGi)
l·uy I = 2AsH cos ( V'i (3.142)

3.5.2 Surface waves: P / SV case

Fig. 3.24: Reflection of a plane SH-wave at a free-surface.
Accounting for the uncoupling of SH-waves 1 the reflected \Vave inust be defined by a
pure SH-wave related to a.n antiplanc displaccrnent such as:
(:U.36)
\i\1e already mentioned that other vvave types than body waves rnay propagate in se1ni-
infinite n1eclia. Such \Vaves, called surface ·waves, correspond to the case \vhere one of the
cornponents of the nor1naliLJecl 1vave vector L characterizing the direction of propagation,
becornes purely irnaginary. These Viraves are characterized by an exponential decay of
displc-1cement c1\vay frorn the free-surface.
Form of the wavefield
Considering the case of a plane wave propagating in the x- z plane, such that lz is purely
irnaginary. In such conditions, the potentials ¢ and '!!!_ can be VVYitten a,s follovils:

The free-surface condition can Le derived considering that 1 for z=O, the traction is (3.143)
zero:
(3.137)
·iw (x - VRt)
~=Be"" exp [ VR l (:J.144)
At the free-surface (z=0) 1 the traction I= g-11 has the following cornponents: { lTxz,
>.vhcre VH_ i.s the velocity of the surface wave. The potentials given by the previous
azz}· Considering the constitutive la\o.r and the expression of the <lisplacement field,
lTyz,
equations must fulfil the corresponding wave equations. As a result, the constants a 1 b,
these cornponentt.; are reduced to: {O, O-yzi O} 1vhere:
VR 1nust Ratisfy the relations:
w2 w2
a2 _
(3.138) . - -v2R - -
v2P (3.145)
178 2D/3D-wave propagation 3.5 \iVaves in a hon1ogcneous or heterogeneous ha.lf-Rpace 179

w2 w2
b2 = v2R - V' (3.146) Table 3.3: Velocities of Rayleigh waves for various Poisson's ratios.
8
Poisson's ratio Vn/Vs
rrhc displacc1nent field may thus he derived under the follu\ving forrn:
l/=0.00 0.862
v=0.25 0.919
a,p aw2
Ux=---·-= ( -Ac"z-bB2e'z
iw 1 ) iw
exp [ -(x-Vnt) ]
a.TVR az VR l/=0.33 0.932
v=0.50 0.955
a'lj!1
uy=-----=
· az
a'lj!3
ax
( bB1--B3
iw
VR
) e In exp [ -(x-VRt
iw
VR
l] (3.147)

a,p a,P2
Uz=-.-+--=
az ax
( aAeaz+-B2cz
iw
VR
·iw
exp -(x-VRt)
VR
b·) [ l Solution wavefield
Considering Eqs (3.148) and (3.149), the displacements (3.147) are derived in the follow-
ing for1n:
Free-surface condition

7
~A ( eaz -
2
::/V~ cbz) exp [ ~. (.r - V.Rt)]

l
The free-surface condition implies the nullity of the traction in the z=O plane. It is thus Ux = b2
straightforward (cryz=O) that the displacerncnt u,y is always zero. Taking into a,ccount
the coru;titutive laVlr, the nullity of the traction, of cornponents er xz and a zz, leads to the
following relations:
_ A (,nx _ 2w /VR )n)
Uz - a c b' +w 2 /V~c
['W
exp VR (x - VRt)
l (3.152)

The \vavefields fulfilling the previous relations are called llayleigh waves 1 fro111 the

'7zz(z = 0) = . aux
\-
ax
auz
+ (.\ + 2µ)-
az
=A [( .\ + 2µ )a 2 - w'
~.\
Vn
l+ · w bB 2 =
2ip.-V
R
o (3.148)
name of Lord Jlaylci.gh who evidenced such waves first. The in1aginary ter1n 'i in the
expression of the horizontal di,splacernent showB that this component has a 90° phase
shift frurn the vertical di,splacernent. As depicted in Fig.3.25 1 the motion of the material
2
points follows retrograde ellipses in the x-z plane. One should notice that the amplitude
a·ux
CT,x (z = 0) = jJ ( az + a1lz)
ax
,
= 2·1Aa VR -
W ( 2
b
W )
+ VA B2 = 0 (3.149) of displace1nent decreases faster vvith depth for shorter -,,vavelengths A= 2KVH/w.

Equations (3.148) and (3.149) have a non trivial solution in A and B2 if, and only if,
Rayleigh waves
the determinant of the linear system is iero. Denoting /3 2 = p/(.\+2µ) and XR = VR/Vs,
vve obtain the following equation:

x1' - 8x'h + (2'1 - 16/3 2 )xh + 16(/32 -

·This equation i8 known as the Rayleigh equation. It is possible to show that it
has a solution Vn satisfying the inequality O<VR<Vs (Achenbach, 1973). This solution
1) =
depends on the f3 ratio that is on the Poisson's ratio of the medium since, considering
the relations given in Table 3.1, {3 2 = (1 - 21/)/(2 - 2v). In Table :J.3, the values of the
VR/Vs ratio are given for Poisson 1s ratios ranging frorr1 0.0 to 0.5.
It should be noticed that the velocity Vn in a homogeneous half-space does not de-
pend on the frequency w ; the R.ayleigh waves arc then non dispersive. AR it will be
discussed for rr1ethods of analysis of surface \vaves (§3. 7), this feature is no longer valid
in a layered medium.
To cstin1atc the velocity of R.ayleigh \vaves VR, a useful approximation is given by:
0.862 + L14v u
VR-- l+v 'S
o (3.150)
(3.151)
_,,,,,
Fig. 3.25: Wavefield due to surface waves showing the elliptic motion near
the free-surface.
For v=0.25, Fig.3.26 displays the variations of the horizontal and vertical <lisplace-
n1ents \vith depth. In this figure, the surface horiZJontal displaccrncnt is norrnalized to 1.0.
V\lhen the motion is perpendicular to the propagation plane (ux = Uz = 0 1 Uy f 0),
surface waves are called Love waves (Fig.3.27). '"rhe velocity of Love waves is equal to
that of shear waves (Vs). Waves having an exponential decay with depth do not onlv
occur at a free-surface) but rr1ay also appear at the interface, betvveen two rnedia with
cliff'er0nt inechanical properties. It is for instance possible tu have surface waves at the
180 2D /3D-wave propagation 3.5 \i'\Taves in a hornogenecnu; or lteterogencous half-space 181

amplitude/surface amplitude ratio

0 0.5 1.0 1.5 2.0
0
, I

0 .,..,... ................
.... ,,
·~

";; 0.5 ,,
,:;
....
bD @
/

///@
~ I
~
"
Q) 1.0 I
I

> I
" . I

~.... '
i f
i I
I
direction of
P< 1.5 :1
Q)
:1 polarization
"" I
I
2.0

Fig. 3.26: Horizontal and vertical displacemer1ts due to Rayleigh waves in an

elastic half-space (v=0.25). propagation

interface bet-,,veen a solid and a fluid or at the interface betv,recn t\vo solids (Aki and
Richards, 1980). Such waves, for \Vhich the n1otion is located in the propagation plane
( u 1 f 0, Uy = 0, Uz f 0), arc called Stoneley waves.
Fig. 3.27: Displacerne11ts due to Love waves in an elastic half-space.
3.5.3 Propagation of a plane SH-wave in a surface layer
V.lc h::-.tvc just sho-1,vn that, in the case of plane monochromatic waves, the horizontal
displace1ncnt in the direction perpendicular to the propagation plane wa.s uncoupled In each 1nedi111n, the total n1otion resulting from the superposition of the waveR prop-
frorn the other components of displacerr1cnt and rnay be \vritten as: agating in the upvvards direction (z > 0) and the do-1,v1nvard direction (z < O) is \vrittcn
choosing the origin of the z a.xis at the top of each mcdiu1n:

·u=Aexp [ -(xoin@+zcosB-Vst)
/W
Vs
l (3.153)
(3.154)

·To sirnplify the notation, the indices have been ornitted: the displacement li denotes where kz.,. is the vertical 1,vavcnurnber in medium n defined as:
the Uy cornponent of displacerncnt (for a propagation in the .T - z plane), and A iR the
amplitucle of the SH-wave. k - w cos f) n
Zn - v.8n (3155)

Layer over a half-space

fn(Y;, t) =exp [ ;;n (x sin en - Vsn t)l (3156)
Considering a layer of constant thickness overlying a half-space (Fig.3.28); we can calcu-
late the rnotion in the surficial layer and in the half-space. The solution corresponding The index n is used to identify each mcdiurn: n = 1 for the surficial layer and n = 2
to this configuration can be obtained by using the sa1ne rr1cthodology as in the previous f~r the half-space. An and A~ arc the arnplitudes of the v;,ravcs propagating in the direc-
sections (continuity of dir.;placc1nent and traction at the interface, free-surface condition). tions z > 0 and z < 0 respectively.
 2D /3D-wavc propagation 3.5 \i\Tavcs in a ho1nogcncous or heterogeneous half-space 183
182

Solution of the problem

Introducing the impedance ratio between both rncdia.:

(3.162)

the relations (3.160) and (3.161) allows to determine the amplitudes of the various
\Vaves:
A,= ~A1 [(1 + x)e-ik,,h + (1- x)e+ik,,h]
(:l.163)
el el A' 2 { A;= ~A, [(1 - x)c-ik,, h + (1 + x)e+ik,,"]
E·quations (3.16:1) are derived asHnrning the amplitude of the surface n1otion to be
kno-1,vn. It is generally an input data for the analysis of seismic motion. V\rhcn the Hurface
inotion is not represented by a monochron1atic v,ravc but by an actual time variation of
the displacement (or velocity1 or acceleration)) the solution may be found 1 for linear
.
F 1g. 3 . 28 : Pl ane SH-wave in a layer of constant thickness h over an elastic
constitutive laws, as the superposition of various rnonochro1natic waves (using direct and
half-space. inverse Fourier traru>forrns). It iH thuH interesting to consider the transfer function of the
layer.

Conditions at the interfaces Transfer function

The free-surface condition (§3.5.1) yields:
Ai =A~
T·he cornpatibility conditionH at the interface thu8 leads to:
• continuity of displacement u,(z = -h) = u2(z = 0),
I3y definition, the transfer function Letvveen t\vo pointR of the layer is the ratio betvvccn
the rnotion arnplitudes at these t\VO points expressed in the frequency dornain. If one of
(3.157) the points is chosen as the point where the displacernent is kno-\vr1 1 the product of the
transfer function by the F'ourier transforn11 of this displacernent represents the spectral
response at the second point. The inverse Fourier tranRform allo\vs the deter1nination of
the ti1ne-don1ain response.
that is:
Alternative definitions. Considering the points defined in Fig.3.29, the transfer func-
(3.158) tion of a wave through a soil layer n1ay he defined by various n1cans:

• Transfer function T(w) : it is defined as the spectra.I ratio betVileen the surface
frorn vrhichi the relation being valid for all .T and all L, vve derive: motion (point A) and the motion at the layer/half-space interface (point B),

sin02 (3.159) • T'ransfer function T(w) : it is defined as the spectral ratio bet\veen the surface
Vs2 motion (point A) and the n1otion at the .surface of the half-space ¥.rithout the layer
(point B 1 ); B' is kno\vn as the outcrop of the half-space.

(3.160) To use the first definition 1 it is necessary to kno-\v the in-depth rnotion >.vhereas the
second one involves the m.otion at a reference point B' (Fig.~i.29 right) 1 located at the
free-surface, which n1ay correspond to a bedrock outcrop (e.g. reference station).
• continuity of traction aW(z = -h) = o{;}(z = 0), that is:
4
0r inverse Fonricr transfonn depending on the harrnonic dependence chosen for the wavefield ( e+i<.:t
(3.161) or e-iwt).
 2D /3D-\vavc propagation 3.5 V\laves in a hornogcneous or heterogeneous half-t'lpace 185
184

B' outcrop 8

bedrock bedrock

Fig. 3.29: Points allowing the calculation of the transfer function through a
soil layer.

Ratio between surface motion and motion at the interface. For the systc1n 2
depicted in Fig.3.29 (left), the transfer function T(w) between the soil surface (point A)
and the interface with the half-space (point 13) may be expressed as:

'UA
Oc__ _ __i__ _ _~---'-----_L_--~---'----~

T(w) = - thus: (3.164) 0 2 4 6 8 10 12 14

UR
frequency (Hz)
Taking into account relations (3.163) :
1
Fig. 3.30: Transfer function T(w) of an elastic soil layer.
IT(w)I= -~~ (3.165)
cos (kz,h)
ThiR function is infinite for the follo\ving frequency valuer.;:
(2n - l)K Vs,
where: x=
Wn = (3.166)
2 hcosRr
(:J.168)
The eigenfrcquencies given by Eq.(3.166) represent the cigenfrcquencies of the tlOil
layer of thickness h. It it-> straightforvvard to check thir.; proposal by 8ceking the classical This alternative definition leads to a very different expret'lHion of the transfer function
periodic solutionti of the wave equation) under the follo\ving boundary conditions: zero since:
displacement at the base and z;ero stress vector at the free-surface.
• transfer function f'(w) depends on the features of both media (pararneter X),
Figure 3.30 displays the transfer function of a 20rn deep layer having a shear wave • the peaks have a finite amplitude (J /x), it is due to the fact that the energy is
velocity Vs=200 in/s, a rr1ass density p=2 t/rn3 and considering a norrnally incident partially absorbed by the refiected 1.vave A;
(radiative damping). It is no in.ore the
plane SH-wave (8 1 =0). It is obvious from Eq.(3.165) that the features of the half-space case for very large X values.
are not involved in this expression of the transfer function.
Keeping the sarne Rhear Virave velocity in the soil layer as in the previous ca.se (i.e.
Ratio between the surface motion and the outcrop motion. The transfer func- Vs 1 =2001n/s) 1 Fig. 3.31 displays a co1nparison of both tranHfer functions for varicnu; Hets
1
tion T(w) between the soil surface (point A) and the bedrock motion (point B : surface of properties for the bedrock (i.e. half-space):
of the half-space outcrop) is such that: • case 1 : Vs,=5000m/s, p2 =2500kg/m3 ,
(3.167) • case 2: Y;g,=2000m/s, p2 =2200kg/rn3 ,
186 2D/3D-wave propagation 3.5 V\Taves in a homogeneous or heterogeneous half-space 187

• case 3: Vs,=800rn/s, p,=2000kg/ni3. A recuTsivc n1ethod thus allows the deterrnination of the displacerncnt an1plitudes in
layer j as a function of the di8placcrncnts amplitudes at the free-surface:
The infiucnce of the velocity ratio bct1,.veen both n1edia clearly appears since, for case 3,
the inaximurn value of the transfer function is lower than 2. Conversely, for a very lcJ\V
x value, the transfer function T(w) leads to values very close to that of T(w) (case 1). (3.170)

10~~~~~~~~~~~~~~~~~~~~,,--~----,
VVe can thus expreRs the transfer function between layers j and k as follov,rs:
I T(co) I ~~-:'---':~ + gi(w)
I T(co) I i G)--- @ - - - - - @------- . ( ) _ fk(w)
TJ,k w - (3.171)
8 ~! V8'=5000m/s Vs2=2000m/s V 82=800m/s f1(w) + 9j(w)
3
p,=2500kg/m p,=2200kg/m" p2 =2000kg/m"
Comparison for the one-layer and two-layer cases

In order to esti1nate the influence of the soil stratification on the ainplification of seis-
6
111ic \Vaves, a comparison betvveen the one-layer case and several two-layers cases is nov.r
proposed. rrhe one-layer case (case 1) correRpon{h> to a single layer of thickness hi =20rn
and velocity Vs 1 =200rn/s, overlying a half-space. T·he transfer .function is displayed in
4 Fig.3.32 (it is similar to the curves displayed in Fig.3.31).

if
.
; r\
'''I;1I'\'Y·; : "
! I \'\'
:;
2
. 'I\
1/ --"
/'~
''
11--~ / '
''
·'!-~
/, ,,,' @
,
/
/

-~----/
/
,,.
'----, /
/
'
' 5
,,@
2 4 6 8 10 12 14
4 \,/;
1\.~
frequency (Hz)

Fig. 3.31: Transfer functions T(w) and T(w) of an elastic soil layer: modulus I ,'\'.,
vs frequency for various bedrock properties. 3 I ' \ ;1
I I \ \,
I I' \ ,,•\
I I \ \\
2 I II \ •\
..,,
3.5.4 Amplification of seismic waves in layered media
Transfer function between layers /
I
//
I \
,.
"\

In the more general case of a layered mediurn \Vith n layers) the relation (3.163) can be
generaliLJed un<ler the following for1n:

2 4 6 8 10 12 14
(3.169) frequency (Hz)

Fig. 3.32: Transfer function through an elastic soil layer (case 1) or through
µJPJ cosfJ.J
two elastic layers with different velocities (cases 2, 3 and 4).
\vherc: XJ =
µJ+lPJ+l cosB1+1
 2D/3D-wave propagation 3.6 Application 1: "\va.vcs in centrifuged n1odels 189
188

Jn the tvvo-layers cases (cases 2) 3 and 4), \Ve chose t\vo lOrn thick layers corresponding prototype (actual structure). The scaling factor 17J ir-; thnr.; v,rritten:
to the flarnc total layer thickness than in the previous case. The velocities in each layer
~ V.J.m,
are dcterrnincd in order to progressively increase the velocity ratio beb.vcen both layers 1.L'=- (3.172)
and sin1ultaneously keeping the san1e equivalent velocity (i.e. v, !v
2
=200rn/s as in case
'IJ,'p

1):
3.6.3 Calculation of the scaling factors
• case 2: 17s, =150m/s; Vs 2 =250rn/s,
Denoting ui the displacement con1ponents, l the length scale, CJij the co1nponents of the
• case 3 : V:s, =lDIJrn/s ; V:s,=300m/s, stress tenr.;or, p the 1nass density, t the_ti1ne, g,;, the cornponents of the gravity forces, vvc
consider the related scaling factors: fi, l, 0-, j5, f g. These factors can be calculated thanks
• case 4: Vs, =75m/H; Vs 2 =325n1/s. to the equations of mechanics. Considering for instance the equilihriurn equations that 1

'l'he results displayed iu l'ig.3.32 sho\11/ that, for an identical total layer thickness, the is:
two-layers cases lead to a stronger arnplification than the one-layer case. Further1nore, --·
Oi7;j
OiEj
+ p ( .9i - - o·ui)
[Jt2
= 0
the larger the velocity ratio between both layers, the stronger the arnplification. Finally1
the frequency of maxi1nu1u amplification decreases as the velocity ratio increases.
if these scales arc inodified 1 the first terrn is rnultiplicd by 0- [- 1 , the second one by pg
and the third one by j5 U t- 2 . To Ratisfy the equilibriu1n cquations 1 it implieR that:
3.6 Application 1: waves in centrifuged models
(3.174)
3.6.1 Historical summary
rib fulfil the "rheological scaling\ it is necessary that P=l (san1c n1aterial) and 0-=l
For Vi.rave propagation proble1ns, the inain interest of reduced scale experiments (e.g. (same behaviour). Frorn Eq.(3.174), \Ve thur.; derive: gl=l. In centrifuge experi1nents,
centrifuge) is to allov.r the analysis of con1plex wc.tvcfields for s1nall experimental configu- g is chosen in order to have 9 = gm/ gp = JV. The scaling factor for the length scale is
rations under well controlled conditions. It is useful for two types of problems: seismic thus: Z=l/ N. The 1nodel di1nensions are related to the prototype (actual) dirnensions
wave propagation or vibrations induced in soils. In 1869, E. Phillips presented at the as follovvs: Lrn, = lp/N. Centrifuging a scale model at Ng sirnulatcs the behaviour, at lg,
french Acadcm.y of Sciences the idea of reduced scale experiments on centrifuged rnod- of an actual structure N tilnes larger. For dyna1nic problems, the scaling factors of the
cls (Phillips, 1869). Phillips proposed to use the centrifuge forces to study problems various mechanical parameters arc synthcth-;ed in Table 3.4.
for \.Vhich gravity is an essential governing factor of equilibriu1n. The first scaling lavvs
established by Phillips (see hereafter) were rediscovered and finally applied in the early
30's simultaneously in USA anJ. in Soviet Union. Dut it was rnainly in the 70's, thanks
to researches perforrr1ed in Great Britain, that centrifuge 1nodelling v.ras "\vidcly developed. Table 3.4: Scaling factor for several mechanical parruneters.

inechanical scaling value

rrhe rnain experi1nental applications concerned problc1ns of structural statibility (ap- parameter factor
plications to dams and crnbankrr1ents) 1 of slope stability and of bearing capacity of foun-
length 1/N
dations. Afterwards, experi1nents on soil liquefaction under seisrr1ic loadings vvere devel-
oped and set the path for dynarnic centrifuge testr.;. ti1ne t l/N
stress iJ 1
velocity v 1
3.6.2 Equivalence principle mass density l
p
Ce11trifuge ter.;tr.; allov,1 simulation, at a reduced scale, of the response of actual gcotechnical force F 1/N2
structures (Luong, 1993). Generating artificial gravity forcer.; rnakes the reduced scale N
frequency f
rnodel equivalent, fro1n a 1nechanical point of vie\V 1 to a large actual structure under
acceleration a N
natural gravity. This equivalence is governed by the la"\VS of mechanics under the form
of scaling factors. For a rr1echanical para1ncter tJJ, the scaling factor W represents the
ratio between 1.Dm, i value of 'W for the reduced scale rnodeli and Wv the value of 'W for the
 2D /3D-"\vave propagation 3.6 A.pplication 1: >Naves in centrifuged rnodels 191
190

3.6.4 Dynamic experiments in the centrifuge
l\ilany geotechnical centrifuges are available in different tearns around the world (1\dalier
and Elgamal, 2002; Arulanandan et al., 1982; Brennan and Madablmshi, 2002; Chazclas
et a.l., 2001; Cheney et al., 1990; Luong, 1993; J\ig et a.l., 2004; Takahashi and Takemura,
2005). A.s Rhown in Fig.3.33, various types of experim.ental devices are used. The related
tnaxirnurn acceleration level and reduction scale are thus different from one device to the
other. Generating dynamic or seismic loadings is also a crucial issue in such experirnents
(Chazelas el al., 2001; Cheney et al., 1990; Coe et a.l., 1985; Luong, 1993).
rrhe accelerogra1ns displayed in Fig.:i.34 correspond to dynarnic expcrirnents per-
forn1cd in the Can1bridge University geotechnical centrifuge (Brennan and J\!Iadahhushi,
2002). A shaking table is installed at the bottom of the soil container and a shear wave
is generated. The up-,,vard propagating shear VI.raves allo-,,v the analysis of the propagation
phenomena as Virell as of dynamic soil structure interaction.
'fo have realit.;tic earthquake sirnulations in the centrifuge, it is necessary to apply a
dyna1nic loading to the -,,vhole soil rnodcl. It is for instance possible to design shaking
taLles efficient even under large artificial gravities (Finn) 2007). To retrieve actual seiH-
rnic excitations V1.rith such devices 1 it may be neceHHary to consider non rigid containers
to allow the Roil rr1odel to dcforrn during the dyna1nic excitation (e.g. equivalent shear
beam system).

bJJ ________ J. ___ _

?!?. 0 .. !.. ···--·--··--·--·-

-5~~----c:-----ie,-f'-;-~---'-~~__J_~~_j_~~_[_~~_j_~__J

10.-~~;-~i-+-,+-~~-,---~~~~~~~~~~~~~~~

·r-==-············ ...................... .

-10L-~~f-i'il-/~-'-j-~~--'--~~_j_~~-L~~-'-~~-'-~~~

20.-~~+-i<c-+~,-'-~~~~~~~~~~~~~~~~~~

bJJ 1o
~

Fig. 3.33: Geotechnical centrifuges at several research centers: LCPC-Paris

(top left), University of Cambridge (top right), C-Core-St-John (bottom left)
and PARI-Kanagawa (bottom right). bJJ o
0~
-1 o
-20:-:--~---:-":-::-~---.,._c_._~~_L_~~--'--~~_j_~~-L~~_[_~~_J
0.66 0.67 0.68 0.69 0.70 0.71 0.72 0.73 0.74
Dynamic and seismic loadings
time (s)
'I'o perform dynamic tests in the centrifuge (e.g.: seismic loadings or vibrations), it is
necessary to apply a dyna1nic loading at the bottorn of the inodel) at the free-su1-face Fig. 3.34: Shear wave propagation in the Cambridge University geotechnical
or on a structure founded on the soil (Brennan and 1'vladahhushi 1 2002; Chazelas et al., centrifuge (Brennan and Madabhushi, 2002).
2001; Cheney et al., 1990; Luong, 1993). The first method is very interesting to analyze
seisn1ic \.Vave propagation in soill'l, but it raises important technological difficulties.
192 2D /3D-\vavc propagation 3.6 Application 1: vvaves in centrifuged models 193

3.6.5 Examples of dynamic centrifuge experiments

300
Centrifuge experiments by Semblat and Luong (1998) (,"'"'
200
rrhc experirncnts presented hercjn \Vere perforn1ed in a French geotcchnical centrifuge 5"' 100
(CEA-CESTA) on sand by Semblat aml Luong (1998). For all the tests, the soil model
\.vas centrifuged in order to reach artificial gravity forces equivalent to lOOg \Vhere g stands
-~
~
"
0
0

for the natural gravity conf-ltant. ~ ""'

Q)

Q)
-100
u -200
u
In the fran1cvvork of these expcrirnents, the <lin1ensions of the scaled rno<lel are as
followH: length Lm=l.30m, height hm=O.'.J5m and width lm=0.80m (see Fig.:J.:15). The
"' -300

scaling factor for the length scale being [ = lm/lp = 1/100, the dirncnsions of the marl.el 3.0 4.0 5.0 6.0 7.0 8.0 9.0
correspond to an actual prototype of tli1nenr.;ions: length L.p = 130rn, depth hv = :35m
time (ms)
and width I,, = 80m.
Fig. 3.36: Typical seismic signal 1neasured in the centrifuge with an earth-
'Tv.ro excitation sourceR ·v;,rcrc uRed by Sernblat and Luong (1998): an earthquake sin1- quake simulator (Semblat and Luong, 1998).
ulator and a. Rurface wave generator (falling inass).

falling mass Surface wave generator

ci-~--~-"'--~ at surface lOOg
artificial The classical devices URe<l to analyze >,vave propagation in the centrifuge (explosives,
•7 •9 gravity elcctrornagnetic vibrators, etc) do not allo\v an easy control of the signal duration. A
•8 simple idea \VaR proposed (SernLlat and Luong 1 1998) to generate nearly-spherical surface
waves having a short duration. The cxpcrirnenta.l a.rrangcrncnt consists in a pneurnatic
system dropping, during the flight, a. mass on the soil Rurface. The system is thus called

z~x
drop-ball arrangement. Figure 3.35 giveR a schematic of the experimental arrangement:
a spherical ball, initially at rest, is pushed by a pneumatic system and finally falls on the
Roil surface. Ati tihov1ln in Fig.3.37, the <lynan1ic loading has thus a short <luration.

Fig. 3.35: Schematic of the centrifuge experiments performed by Semblat Several authors studied in great details similar systems inainly dedicated to surface
and Luong (1998). waves (Cha"elas et al., 2001). The work of Abe et al. (1990) for instance presents the
various features (in tin1e and frequency domains) of a systen1 generating seis1nic waves
Ly using a spherical IIH1SS falling on a plate. They rna<lc a paran1etric study considering
different rnasscs, densities 1 falling heights, etc. They analy,.,ed in details their influence
Earthquake simulator on the generated 1,0,ravefield.

Earthquake simulators are useful to analyze seisn1ic 1..vave propagation or dynarnic Roil-
:;:;tructure/soil-pile interaction (Finn, 2007). It n1ay he a shaking table or an explo.Ri~e
systcni. The latter \Vas used by Semblat and Luong (1998). \i\lith such a systen1, it ~s
possible to f'lirnulate earthquakeR of various durations and intens!tics. Nevertheless, _as it
vvas detailed previously, the scaling factor for the tirne scale is t = 1 / N for a centrifuge
rnodel at 1Vg. 'T'he excitations mnst then he a N times faRter than for an actual earth-
quake. A typical seisrnic signal n1ea8ured vvith such an earthquake sirnulator is displayed
in Fig.3.36 (Scmblat and Luong, 1998).
Two wave types
Thanks to the earthquake simulator, the \vavcfield generated in the soil model may be
assurned as plane \Vaves. VVhereas, for the drop-ball arrangerncnt, \Ve nearly have a point
l:lource leading to a spherical 1,0,ravefield (Sernblat and Luong, 1998). The geo1nctrical
featureR of the wavefield are funda1nental to analyze the geometrical, as \vell as phyRical,
attenuation of the waves in the soil. 1'hey change the basic ass111nptions for further
theoretical or nu1nerical analyses of the observed phenomena.
 2D /3D-wavc propagation 3.6 Application 1: wave8 in centrifuged 111odcls 195
194

100 .,
~ m.s
w 0 surface
~
s -100 200
acceleration
~
-200
·fil
h
100
.-"I -300
u
u
u -400
0.0
" -500
-100

3.0 4.0 5.0 6.0 7.0 8.0 9.0

time (ms)
0

Fig. 3.37: Typical signal measured in the centrifuge with a drop-ball arrange-
-200
ment (Sernblat and Luong, 1998).
-400
in-depth
3.6.6 Analysis of the threedimensional wavefield -600 acceleration
m.s
.,
The acceleration measurements performed by Semblat and Luong (1998) were made
in the three direction::; of space X, Y and Z. The threedi1nensional measurements of
acceleration allo·~.v the identification of the acceleration vector at the point considered. Fig. 3.38: Accelerations measured in the X Z plane for surface and in-depth
For the earthquake s'imulator1 the acceleration vector describes a very ccnnplex path which sensor (drop-ball experiments), from Semblat and Luong (1998).
is very difficult to characteriie. Conversely, for the drop-ball tests, the acceleration vector
follovvs a rather simple path.

Drop-ball experiments Identification of the wave polarization

Considering the symrnctry of the problem 1 the Y transversal cornponcnt should be zero.
It is nearly the case even if sorne very srr1all accelerations are found along this axis (Sern-
blat and Luong, 1998). The experimental findings indeed shoVi.r that the rnass does not
exactly fall along the X-axis corresponding to the sensors aligne1nent (it is probabl:'l due
to the Corioli8 forces). rrhc acceleration arnplitude along Y-axis is nevertheless very small
Semblat and Luong (1998) compared the angles of tbe acceleration vectors, e,,
and the
angles of incidence at the different senHors, denoted e~. The values of the (Ji angles are
deter1nined frorn the acceleration curves in the X Z plane (Fig.3.38). The locations of
the sensors allov,r the determination of the incidence angle8 (}~. The values of the () 1 arc
given in Table 3.5. , i

and may be neglected. Only the measurements along X (horizontal) and Z (vertical)
\vill finally be taken into account.
The drop-ball experirncntH leading to simpler signals, they will be chosen for the
analysis of the acceleration vector polarization. For this type of loading system, the
excitation is short enough to avoid \Vave reflections to be combined with direct waves. It
is thus possible to estirnate the acceleration amplitude along the X and Z axes and to
deduce the acceleration vector in this plane. As shown in Fig.~1.38, the components of
the acceleration vector along X and Z allovvs the calculation, for sensor i, of the angle
with the horizontal axis, denoted e,
(Scmblat and Luong, 1998). Tbis calculation will be
ma.de in the next paragraph.
From Table 3.5, the angles B;, are found to coincide with the angles e~, for the in-depth
sensors (even numbers). It means that, in the XZ plane, for all in-depth sensors, the
direction of the acceleration vector coincides with the d'irection of incidence. The accel-
eration vector having no transverse cornponent along Y) the drop-ball arrangcn1cnt is
found to generate pressure ~Daves at depth (i.e. P-vvaves). A.t the free-surface, the angles
(},i and e:, do not coincide: Heveral wave typeR are present (including surface \vaves).
Figure 3.39 displays, for all sensors, the acceleration vectors (maximum acceleration)
deduced frorn these results. The directions of the acceleration vectors for the in-depth
sensors obviously coincide vvith the incidence directions at these sensors. The vectors
 20 /:~D-\vavc propagation 3.6 Application 1: 1vaves in centrifuged n1odels 197
196

phase shift and l:J..f the bandwidth in the resonance curve.

Table 3.5: Values of the theoretical angles of incidence at the various sensors
(Semblat and Luong, 1998).
In-depth sensors Rurface senr.;ors Preliminary simulations from the experiments
#2 e~ = 52° #1 B\ = 16° Geometrical damping. For a spherical 1.vave, \Ve 111ay consider that, far a1,vay frorn the
#4 e~ = 32° #3 e(, =7.2° source, the geornetrical darnping is inversely proportional to the distance to the source.
#6 e~ = 22° #5 e; = <l.7°
We shall thus disregard the l/r 2 near field tern1 obtained in §3.4. Only considering the
#8 e~ = 17° #7 e~ = 3.4°
geometrical dan1ping 1 the con1parison of the acceleration an1plitude8 at two points .Afi
#10 0( 0 = 13° #9 0~=2.7°
and Nf; n1ay be \vritten as follows:

(3.176)
lengths shovv that the acceleration arr1plitude fastly decreases (V'ig.3.39).

Simplified material attenuation. Fro1n the centrifuge experiments perforn1ed by

llivi8re and Luong in the 80's (R.ivi8re, 1983), the attenuation factor a 1.vas aHs111ned to
be constant:
A(x) =Au cxp(-ax) with: a= 1.10 (3.177)
This estirnation of attenuation is rough but it allOVi'S a direct analysis of the pro-
pagation process in the soil. For the drop-ball experiments performed by Sen1blat and
~. _,£······ ..~·· .. .JO Luong (1998), the wavefield being spherical, we have to take into account the geon1etrical
damping together with rnateria.l damping. Comparing the acceleration arnplitudes at tv.ro
points J11i and Jl1j leads to:

ri
Aj =Ai- exp[-a(r; - r.,)] (3.178)
r .J

Fig. 3.39: Acceleration vectors in the X Z pla11e showing pressure waves at
depth and various wave types at the free-surface (Semblat and Luong, 1998).
3.6.7 Modelling propagation in dissipative soils
Attenuatior1 and dissipative phenomena
To assess the attenuation properties of the soil in the centrifuge experirncnts, it is nec-
essary to rnodel vvave propagation in attenuating soils and to identify the dissipative
properties of the soil used for the tests. The definitions of attenuation given in Chap-
ter 2 correspond to vr:trious approaches. The governing para1neters of cyclic, vibratory
or propagation testH are linked by the following relations:
This expression allovvs) from a signal measured at point Jl/[i to calculate the accelera-
ti?n at point lvlj. The r.i/rJ ratio takes into account the geornetrical damping (expansion
of the wavefront) 1vhereas the cxp[-ct(rj -ri)] factor stands for the material attenuation
bet\veen 1V1i, and A~i. It is also necessary to account for the propagation tern1 considering
a constant velocity (directly estirnated from peak to peak delays).
:Figure 3.40 gives an example of sirnulations for drop-ball experiments (Semblat and
Luong, 1998), a being assumed constant as suggested by Riviere and Luong (RiviCrc,
1983). These simulations involve both geometrical damping (spherical 1..vaves) and rna-
terial attenuation. As Hhown in Fig.3.40, the agreen1cnt betv.rcen the rneasured and the
sirnulated signals is not very good. It is thus necessary to investigate rnore complex
n1odels in order to retrieve the experi1nental results in terrr1s of attenuation.

_l_ = ai\ = '5_ = Jl!h =tan¢= /C,..f (3.175) Spherical wave in a viscoelastic medium
Q 1f 7T 1\1R fr
1vhere Q is the quality factor 1 a the attenuation factor 1 A the vvavelength, 6 the logarith- ~r~pa~ation and mechanical behaviour. 'I'o sin1ulate propagation phcnornena in
rnic decrement, 1\![H and lvf1 the real and imaginary parts of the cornplex modulusi ¢the d1ss1pat1ve soils, it is necessary to V\rrite the \Vave equations as well as the constitutive
 T

2D /3D-v.rave propagation 3.6 Application 1: vvavcs in centrifuged rnodels 199

198

rviaxwcll 1 _Kelvin-Voigt and Zener. In the one-dirnensional case. the solution of the V\rave
equation in vi.scoelastic rnedia waH (.see Chapter 2)5: ,

measuredA8 ii'(x,w) = Uo(w)exp[ik*(w)x] (3.180)

A~ simulated V\rhere 'U* is the spectral transfor1n of the displacernent) k* is the con1plcx vvavcnu1nber
depending on w and expressed a.s a function of the complex modulus l\II(w) under the
from A, .,
/' ..... form:
2
(k*(w)) 2 = __PUJ_ (3 181)
M(w)
The real part of the cornplex wavcnu1nbcr k*(w) correHpond.s to the real 1JJavenumber
k and itH im,ag1:nary part to the attenuation factor a 6 . The expressions of the cornplex
,.------- modulus !i,1(w) corresponding to the three viscoleastic 1nodels arc given in Chapter 2.
.-·· '•, ._ .. -·
The solution (3.180) of the wave equation in viscoelat'ltic rnedia i.s now co1nbined with
4.4 the solution (3.179) of the spherical "\vave equation (only keeping, for .sake of simplicity,
3.9 4.0 4.1 4.2 4.3
the l/r term). We thus obtain the radial displacement solution of the wave equation for
time (ms)
.spherical vvaves in viscoelastic n1cdia:

Fig. 3.40: Time domain simulation with a constant attenuation factor for
ii*(r,w)
n exp [ik*(w)r]
= - (3.182)
drop-ball experiments: modulus of the acceleration vector at se11sors #8 r
(measured and simulated from signal at sensor #4), from Semblat (1995).
\vhere B is determined by the boundary conditions of the problem.

Finally, the attenuation iH expressed through two different terms:

behaviour of the n1aterial. The propagation pheno111ena are accounted for using the equi-
• the geometrical dam,ping accounting for the spherical propagation is inversely pro-
libriun1 equation. As shO\.VIl in Chapter 2, the constitutive equations allovvs rcplacen1ent
portional tor (simplified solution in far field), but independent of frequency,
of stress co1nponcnts by strain corr1ponents in the cquilibrhnn equation. The differen-
tial equation thus obtained generally allows the deterrnination of the solution. For soils 1 • the m,ater1:al darnping characterized by the imaginary part of k*(w) and thus de-
the constitutive la¥.rs adapted to wave propagation considering weak rnotions genera.Uy pending on frequency.
involve linear viscoelastic behaviour.
Expression (3.182) fully characteriLJes the propagation of longitudinal spherical -,,vavcs
in a linear viscoelastic soil. Considering the cxpcrirnental results obtained in the cen-
Equilibrium equation in spherical coordinates. For the drop-ball experirnents
trifuge by Sernblat and Luong (1998), it is thus possible to Himulate the propagation of
perforrncd by Scmblat and L·nong (1998) 1 the directions of the acceleration vector being
spherical vvaveH in the soil and to identify the rnechanical par.arneters of the viHcoelastic
identical to the incidence directions 1 the vvavefield propagating at depth is spherical and
models.
only radial con1ponents have to be considered. It is thus possible to use the solution of
the spherical ·wave equation dcterrnined in §3.5.2 1 that is, for a monochromatic wave:
3.6.8 Simulations for drop-ball experiments
1
ur(r,w)=A ( ---,+-V
rp
r~
iw) exp ['iW-V (r-Vpt)l
p
(3.179) Computation for a transient wave
The measurements performed hy Sembla\ and Luong (1998) in the centrifuge experiments
Viscoelastic models. In order to solve the problem of propagation in a dissipative lead to 3C-accelerograrns. The solution in tern1s of spectral cornponents of acceleration)
medium, tl1e soil behaviour is supposed to be linear viscoelastic. Such rnodcls are often . 5.S~nce a e iwl harmonic dependence has been chosen in the present chapter, the sign in this equation
used to analyhe \Vave propagc_ition in dissipative soils. Indeed, they allow linking of is d1flerent fr?rn that in the equivalent expression in Chapter 2 (e+iwl l1annonic dependence).
. d epend ence, the complex 'Wavenumber is expre8se<l as k*(w) = k(w) + io:(w)
Gpor a e -1.wt h armon1c
the stress tern1s to the strain terms considering the complex spectra of the rnechanical
(theeig>.r is (rn
1 erentf"rom f·h ·
, at given · ClW-pter 2 for a e + iw t lw..rrnonic dependence).
1n
variables. \file shall consider three linear viscoelastic models already studied in Chapter 2:
 2D /3D-wave propagation 3.6 Application 1: \vaves in centrifuged models 201
200

a*(r,w), may indeed be written under a similar [orn1 as the solution in displacen1ent given For the J'viaxvvell and Kelvin-Voigt rnodels 1 the Laine constant~ A,µ. being identified
by Eq.(3.182). It is due to the fact that, in the frequency domain, the complex spectra (e.g. \Vave velocity) 1 the values of r7 should be estimated considering the experin1cntal
2
in displacement and acceleration arc directly proportional: a*(r,w) = -w u*(r,w). This results. rrhe agreement bct\vcen the measured and simulated signals v.rill be assessed
expression being valid for all frequency values w, it is possible to transforrn the signals in fro1n the first acceleration peak. In the follo\.ving, only the sin1ulations obtained \Vith
the tirnc do1nain as long as the HnpcrpoHition principle is fulfilled (linear behaviour). 1'he these t\vo n1odels arc discussed. The si1nulations cornputed \.vith the Zener rnoclcl \vill be
7 presented later.
acceleration a(r, t) in the tirne dornain is thus obtained '\Vith the follcJ\ving expression :

a(r, t) = - 1 r+= a'(r,w) exp(-iwt)dw

21f . ~oc'
The cornplcte expression of acceleration 1 as a function of
(3.183)
To con1pare the viscoelastic sinrula.tions to the measured accelerations, we shall choose
a reference signal fron1 \Vhich the acceleration at a distance r will be simulated. T·he Him-
ulations arc co1npa.red afterv.rards to the acceleration signals given by the senRor located
T and t, is then: at the sa1ne distance r from the reference sensor. The 'values of viscosity for different

a(r, t) = -1- 11+= - Bw

2
.
- - exp [ik'(w)r] .
cxp(-iwt)dw (3.184)
rnodels are tuned in order to retrieve the amplitude of the f£rst acceleration peak. After
the first peak, the direct and reflected waves are co1nhined. This is the reason vvhy it is
27T r ·-= r more difficult to identify the pararneters frorn the signals n1easured for the earthquake
simulator experiments (Semblat and Luong, 1998).
The si1nulations are thus performed using the cornplex spectra in the frequency do-
main. The tirne domain solution is co1npnted afterwfilds considering Eq.(3.184). The values of 'r} leading to the best fit for the first acceleration peak are the following:

For the 1.Jax\vell and Kelvin-Voigt models 1 the para1neters characterizing the be- • ~=150000 Pa.s for the Maxwell model,
haviour are the Young's m,odulus E and the viscosity TJ. F'or the Zener rnodel 1 the param-
eters are three instead of t1.vo: tv;To inoduli (E and Ev) and the viscosity 77. • r;=lOOO Pa.s for the Kelvin-Voigt model.

Simulations from experimental results
The sin1ulations are pcrforrned for the drop-ball cxperi1nents (Sernblat and Luong 1 1998)
since the vi.ravefield at-depth corresponds to pure longitudinal waves. In the X Z plane,
the acceleration vector is characterized by its inodulus arn = J
a~+ a~. It is possilJle
to sirnulatc the propagation at several distances from the source considering the first
measured signal as the reference signal used for the viscoelastic simulations.
Considering the acceleration moduli for sensors 'i and j, that is a1n (r,i, t) and Um (r j, t),
respectively located at distances Ti and T.J fro111 the source, the ratio of the acceleration
amplitudes in the frequency dornain 1 a:n (r.J, w)/ a,;n(ri 1 w) can be easily written thanks to
expression (:-3.182), adapted to acceleration corr1ponents under the follovving form:
(3.185)
Jf the acceleration is kno\vn at dit-;tance ri from the source, the acceleration at distance
Tj can be computed by using a vir.;coelastic siinulation. Frcnn expression (3.185), the
sin1ulated acceleration a~~m (Tj 1 t) is 1vritten: in the frequency don1ain, under the forrn:
1'he acceleration signals obtained with these viscosity values are displayed in Fig. (:-:~.41).
The acceleration signal (1nodulus of the acceleration vector in the XZ plane) given by
sensor #4 is the reference signal. The accelerations corr1puted at a distance r = r 6 - r 4
fron1 the reference sensor arc con1pared to the acceleration rncasurcd by sensor #6.
From the curves displayed in Fig.3.41 1 the values chosen for 'r} lead to simulated
acceleration curves very close to the rncasured acceleration. The sirnulations perfor1ned
are satisfactory, but their respective features arc different:
• the M a.'Lwell rnodel leads to an acceleration signal very close to the measured signal
since the sirn'Ulation fits the acceleration arnplitude but also the shape of the .first
pea.k,
• the Kel'uin- Voigt model gives a corTect siniulation in an1plitude hut it does not
exactly .fit the shape of the measured signal. The higher frequency cornponcnts of
the r.;ignal vanish whereas the lo\.\rer frequency cornponents are v.reakly attenuated:
the first acceleration peak tends to srnoothen during the propagation.
To precisely assess this frequency tlcpendence of attenuation, the analysis of the variationR
of k* as a fu11ction of frequency \vill be n1a.de in the follcnving.

a°''mh,t)
rn.
= 2_ ;+= a~,(r;,w)r;
2w. -= rj
exp[ik'(w)(rJ-ri)]exp(-iwt)dw (3.186) Optimization process for the Maxwell and Zener models
For the 1Iax\vell and Kelvin-Voigt rnodels, assurning the velocity to be slightly depen-
7Inversc Fourier transforn1s have 1.o be u8cd for a e-iwt harrnonic dependence (instcA.d of direct
dent on the viscosity, the para1neter identification is easy since only the viscosity has
transfonns ill Chapter 2).
 2D /:JD-wave propagation 3.6 Application 1: \Vaves in centrifuged rnodcls 203
202

300 measured
Maxwell
C'
Kelvin
w "
0
7.5
~~ "
;.,
Q)
200 u
0
·~
~ 7.4
·~ sensor!#4 "
""'"" 7.3 Y\,,,=144 kPa.s

""
8 100
"
o<
7.2

7.1

130 135 140 145 150 155 160

0.35 0.36 0.37 0.38 0.39 0.40
viscosity (kPa.s)
time (ms)
Fig. 3.42: Determination of the optimal viscosity for the Maxwell model:
Fig. 3.41: Accelerations corr1puted with linear viscoelastic models: modulus
minimizatio11 of the quadratic error (sensors #4 and 6).
in the XZ plane (drop-ball experiments), from Semblat and Luong (1998).

the quadratic error iH no-vv defined by the functional Jzen(Ev, ry) as follovvs:
to be eHtimated. Furtherrnore 1 the identification process is a.n amplitude fitting rnethod
between the peaks of the sirnulated and measured signals. For the Zener inodel, the value
vvith (3.188)
of the elastic modulus being fixed, two parameters still have to be determ1:ned. The es-
timation of these pararncters is inore cornplex than for :rv:Iaxwcll and Kelvin-Voigt 1nodels.
The functional J Zen (Ev, Tl) has heen calculated to determine the optimal (Ev, T/) pa-
'I'o estimate the optirnal mechanical pararr1cters for various types of rnodels, it is rameters. The curve displayed in Fig.3.43 gives the Jzcn(Ev, Tl) for Ev E [ O; 50 MPa] and
necesRary to quantify the quality of the viHcoclastic simulation. It is for instaJ1cc possible T/ E [130; 160 kPa.s] (simulations between sensors #4 and 6). The minimum quadratic
to e.stirna.te the quadratic error betvveen the si1nulation a~1n and the rncasured signal error is obtained for Ev= 25.5 J'vIPa and ry= lt16 kPa.s. The optimal vit:;cosity 17opt deter-
a.m,(rj)· Considering a ti1ne interval [t 1 ; t 2 ] corresponding to the first acceleration peak 1 mined for the Zener rr1odcl is very close to that of the ]\!Iaxvvcll 1nodel.
the functional J (T/) defines the quadratic error for the Maxwell model under the following
form: Another type of "objective fnnction' 1 may be considered to optimize the rheological
parameters. It is for instance possible to co1npare 1 in the frequency domain, the attenu-
ation values Q- 1 . The expression of such an objective function could take the following
(3.187)
forn1:
•Wo
J - -1
rrhe 1ninimization of Jlvf ax ( TJ) leads to the optimal viscosity T}opf,. The curve displayed
· - J,
0
j w,
- -1
(Qe,p(w) - -1
Q,im(w)) 2
dw with: (3 189)

in Fig.3.42 gives the w1lucs of J,,Ia,(T/) for T/ E [130; 160 kI'a.s] (calculation performed
for sensors #4 and 6). As shovvn in Fig.3.42, the optirnal viscosity is T/opt= 144kPa.s.
The fit between the fi.Tst peak of the sinrulated and rneasnred signals is then very good.
'To estimate the optirnal Ev and TJ parameters for the Zener rr1o<lel (see Chapter 2),
3.6.9 Influence of frequency on attenuation factor
Fro1n expression (3.181), the complex V\ravennrnber k* is a function of w. Its real and
in1aginary parts characteri2Je the phase-shift and the attenuation respectively. These two
 r
204 2D /3D-wavc propagation 3.6 Application 1: ¥.raves in centrifuged modelf> 205

3_0,~-~--~-~--~-~--~--

Maxwell
"' 2.5
3 /
---- -----------------
I
,_, I
0 I
H
H I

'ro°
..; 1.5
I
I
I
7.2
iiro I constant cx=l. l
~---------- ------------------------------------
~ 7.0 1.0
ry I
I
6.8 160
--.. - ~--,
Kelvin-Voigt
.. _____ _ ......____:::~-.:
_ ______ .. ______ ---- 0.5
-------------- ... :::~·
0

5.0 10 15 20 25 30 35
40
50 130 frequency (Hz)

Fig. 3.44: Attenuation factor as a function of frequency for the Maxwell

Fig. 3.43: Optimal mechanical parameters for the Zener model: minimization and Kelvin-Voigt models (Semblat, 1995). Comparison with the constant
of the quadratic error (Semblat and Luong, 1998). attenuation factor estimated by Riviere and Luong (Riviere, 1983).

para1neter8 depend on frequency but this dependence is different frorn one model to an-
other.

~- • • • • • • •
Figure 3.44 displays the attenuation factor a as a function of frequency for the l'viaxvvell
and Kelvin-Voigt n1odcls. The T/ values conHidercd corre,'3pond to the previous Hirnula-
tions (TJMax=l500()() Pa.s ct 1)ge1=1000 Pa.s).
• • • • • • • •

The attenuation factor estimated by Riviere (1983) is in the range of the values IC!ES~I IT,~0.29 msl IT,~0.69 msl
obtained in Fig.3.44 vvith viscoelastic rnodcls. The attenuation factor derived f:ron1 the
viscoelastic simulations ranges fro1n 0 to 2 rn- 1 in the conf::lidcred frequency band. The
attenuation factor appears to be larger than the value proposed by RiviCre (1983) which • • •
vvas previously seen to under-estirnate the dissipation (Fig.3.40). The curves displayed in
:F'ig.3.44 clearly shovv why, for the Kelvin-Voigt rr1odcl, the higher frequency con1ponents • • •
are n1ore attenuated than lower frequency co1nponents.
IT,~0.99 ms! IT,= 1.29 ms I

3.6.10 Numerical modelling of centrifuge experiments

Several nu1ncrical results obtained by finite elen1ent sirnulations arc shuvii'I1 in Fig.3.45. Fig. 3.45: Numerical modelling (FEM, CESAR-LCPC) of wave propagation
They illustrate the main features of the propagation process in the centrifuged model: in a centrifuge model (Semblat and Luong, 1998).
wave type, reflections at the botto1n of the rnodcl, etc. Considering the spherical source1
these computations were performed using a =isyrnmctrical model (CESAR-LCPC FEM/
BEM code), from Scmblat and Luong (1998).
 T

2D /:JD-wave propagation 3.6 Application 1: waves in centrifuged models 207

206

The re8ults displayed in Fig.3.45 correspond tu four different tirnc steps:
• time T1 : beginning of the propagation process, the various wave types can not be
distinguished yet,
• i'irne 72: clear separation of the longitudinal \\raves (the fastest) from the shear
\Vaves and the surface \vavcs,
3.6.12 Analysis of dispersion
A wavelet method was considered by Scrnblat and Luong (1998) to analyze the propa-
gation in variou8 frequency bands. It is thus possible to esti1nate the phase velocity by
con1paring, for different sensors, the signals analyzed in different frequency bands (see
Fig.:l.47). Various filtering techniques arc presented in (Scherbaurn, 2006) and (Mari et
al., 1999) and several issues are briefly discussed in Appendix C.

• time T?,: beginning of the reflection of longitudinal waves at the hot.torn of the
rnodel,
• t'irne T4: the direct longitudinal and shear waves keep propagatir1g in the rnodel
\Vhereas the reflected P-\.vave propagates towards the free-surface.
As tihovvrn in Fig.3.39 1 only longitudinal 1.vaves arc charactcriz;ed by in-depth accelcro1n-
eters. Conversely, the experimental resultH and the rnuncrical si1nulations (Figs 3.39 and
3.1!5) show that various wave types arc present near the free-surface.
For the centrifuge tests performed by Semblat and Luong (1998), a geometrical dis-
persion due to the n1odel boundaries is characteribed at lo\v frequencie8 1,vhereas a lovver
physical dispersion due to the dissipative properties of the soil is observed at higher
frequencies. As sho1,vn in Fig.3.4 7, the direct and reflected waves are well separated at
high0.;r frequencies but not at low frequencies (these frequencies arc given at the inodel
scale and a factor 1 /JV must be considered to derive the prototype frequencies).

1342 Hz
3.6.11 Removing reflections by homomorphic filtering
I'he ho1non1orphic filtering technique, described in the appendices (§E), is novv consid- 1613 Hz
ered to re1nove reflections on the container boundaries frorn the direct wavefield. For the
acceleration signals measured in the drop-ball experin1ents 1 Fig.3.46 shows that horno- 2154 Hz
1norphic filtering allow8 to re1nove waves reflected at the botto1n of the model. It is even
possible for signals having long durations (see §E). 3958 Hz

5762 Hz

7566 Hz
4.0 5.0 6.0 7.0 8.0
" time (ms)
·. ,.•,

after Fig. 3.47: Acceleration signals filtered in different frequency bands given at
.: :.filtering the model scale (Semblat, 1995) .

~.,before
filtering

4.40 4.80 5.20 5.60 6.00 6.40

time (ms)

Fig. 3.46: Echo removal by homomorphic filtering (Semblat and Luong, 1998).
 2D/3D-wave propagation 3. 7 Application 2: Spectral Analysis of Srn'face \llave::i a.nd in ::iitn te::its 209
208

3. 7 Application 2: Spectral Analysis of Surface Waves Con::iidering Fig.:-3.48 and the results frorn §3.5.2, the expression of the vvaveficld is:

and in situ tests for -h-<;z-<;O

(3.191)
'larious typet-i of in situ testH arc available to identify the dynan1ic properties of soils.
for z -<; -h
The Spectral Analysis of Surface \i\laveR (or SASH!) investigates the dispersive nature
of surface waves in layered soils (Achenbach, 1973; Aki and Richards, 1980; Bodet ct
al., 2005; Foti, 2000, 200:3; Lai et a.l., 2002). As it will be shown in this section, it is vvith, sirnilarly to Eq.(3.146), the expression of b,t:
possible to link such dispen.;ive properties to the gcotcchnical features of the site under
investigation. (3.192)

3. 7.1 Dispersion of Love waves in a single-layered half-space and V,0., = If!: (3.193)

Model and wavefields

Free-surface and continuity conditions
As depicted in Fig.:1.48 1 v.,rc consider a surface (Love) wave propagating horizontally, along The free-surface condition yields:
the x axis, in a soil layer (thicknet-is h) overlaying an elastic half-space. The V\ravefielcls
in each mediun1 are only polarized along the y axis and fulfil the follov,ring equations of Duy
propagation:
~-
Dz
= 0 '* Ai =A 11 (3.194)

2 2 2
U uy = f!J_ ( 8 11y +8 uy) for - h -<; z -<; 0 The continuity equations at the interface (z = -h) are thuR as folluv,r::i:
Dt2 P1 Dx 2 Dz 2
(3.190)
8 , - (8a2+a2
2
u,u
uy - /12 8 2 2
uy) for z -<; -h
2A1 cos(-ib 1 h) = A 2 e-"'"
2iµ.1b1A1 sin(ibrh) = µ,b2A2e-b,h
(3.195)
Dt' P2 .T z
The following equation is finally derived:

(3.196)
x

Dispersion law
From the previous results, the equation of di::ipersion of a surface (Love) wave is the
following (Aki and Richards, 1980):

tan [wh (3.197)

Vl.'here Vn(w) is the velocity of surface \Vaves (depending on frequency).

Each term of the previouR equation is displayed in Fig.3.49 as a function of hb 1 . A

Fig. 3.48: Dispersion of surface waves in a single-layered half-space: schematic graphical method may then be used to determine the roots of Eq.(3.197) in terms of
eigenvalues (Aki and Richards, 1980).
of the pro bl em.
210 2D /3D-wave propagation 3. 7 Application 2: Spectral Analysis of Surface VVavcs antl in situ tests 211

Application for a given soil layering

I Soil model and number of modes. 'Vile shall now consider a Ringle layer \Vith the
I
I
left hand term following pararneters: h=501n, Vs\ =400m/s and Vs 2 =8001n/s. Applying the previous
I
I method, for three given frequencies (5Hz 1 lOHz and 15Hz), we obtain various numbers
l of eigenvalues displayed in Fig.3.50. The number of eigenvalues ranges from 2 at 5Hz
I right hand term
l to 4 at 15Hz. The curves being dra:vi,rn as functions of hb1, it is necessary to find the
I
roots of Eq.(3.197) ancl compute the corresponding velocity valueR (Love waves) for each
exact roots
0 0 0 0 0 0 frequency.

Velocity values for each mode. Considering the curves displayed in Fig.3.50) one can
compute the eigenvalues of Eq.(3.197) at each frequency. The roots arc firstly determined
in ter1ns of hb1 and then converted into surface wave velocities using the expression of bi)
h
Eq.(3.192). All these values are detailed in Table 3.6 for /=5Hz, lOHz and 15H". The
n1unber of inodes depends on the frequency level:

• for f=5Hz: only two modes arc obtained with velocities close to that of the layer
0)
0)
(n=O) and that of the half-space (n=l) respectively,
• for /=lOHz: a third mode (n=2) is reached at this frequency and mode n=l now
leads to a velocity in bct\vccn Vs 1 ancl V:s::n

Fig. 3.49: Dispersion of surface waves (Love waves) in a surficial layer: graph- • for f=l5Hz: a fourth mode (n=3) is novv reached and modeR n=I and n=2 have
ical comparison of both terms of Eq.(3.197) to estimate its eigenvalues. velocities in between Vs\ and 'Vs 2 •

Table 3.6: Number of modes at three different frequencies and corresponding

Fina1ly1 the eigenfunctions corresponding to the previous eigenvalues are as follows hb1 values and surface wave velocities (h=50m, Vs, =400m/s and J;;g,=800m/s).
(Aki and Richards, 1980):
frequency rnode rnode mode mode
1 values n=O n=l n=2 n=3
1 ,(nl(x
"y
z t)
' ''
= 2A1 cos [w -V§, - - l
Vjl(wn)
z]
5 Hz: hb1 0.0462564 0.1080292 no no
velocity 430.56524 795.08442 no no
exp [v i(w ) (x - VR(wn)t)]
R Wn
-h s; z s; 0 10 Hz: hb1 0.0240927 0.0715505 0.1074094 no
velocity 407.64347 487.82181 781.99225 no

[
~1-1] (3.198) 15 Hi: hb1 0.0162638 0.0486718 0.0804589 0.1065728
= 2A1 cos w 17,f, - Vjl(wn) h
velocity 403.42933 434.27302 522.66928 765.41439

exp [-w 2(l ) - Vl' (h -

VR Wn 82
z)]
'1W
exp [ Vn(wn) (x - VR(wn)t) l z <(; -h
Dispersion curves for two soil models. V\Te shall nov.r consider tv,ro soil models and
compute the dispersion curves for each of the1n. Both 1nodels involve a single layer v;,rith
the following parameters:
As shown by Eq.(3.198), the eigenfunctions involve sinu::soidal oscilla.tion in the surfi-
cial layer 1vhereas they have an exponential decay in the half-space. • h=50rn, Vs, =400rn/s and V.s,=800m/s,
212 2D /~iD-wavc propagation 3. 7 Application 2: Spectral Analysis of Surface V\Taves and in situ te8tR 213

• h'=30m, V:s, =400m/s and Vs,=8110m/s.

8aa
The velocities in the layer and the half-space arc then identical and both configuration,s
are only differing through the layer depth.

28

m 20
24 f=5Hz
--- ------
s
,.. 16 left term
CJ
+'
+' 12 right term
A
M 8 1
ls root
·~
-':: 4 --- --- --- ---
~
~
CJ
0
---
-4 a.aaL.-~-~-~-~-~--~-~-~-~~
4.a 8.a 12. 16. 2a.
-8

28
24
\
\ f=lOHz 8aa """""""" ·······------------------- ----··-··-··-··--··-··-

' -- '
" " , n=l
\
'
m 20
\
\ left term '
n=2' ' ,
s
,.. 16
n=O
'
CJ
+'
+' 12
l"t root

---
right term Gaa "" .......................... _
i1
·~
-':: 4
~
CJ
8
-2~d;Qor --- --- --- 4aa
---
~ 0 3rd root
-4
-8
2aa
28
1st root f=15Hz
24 •\
: \
00 20 : ' left term a.a~-~-~-~-~-~-~-~-~-~-~
8,.. 16 : ' a.
:
. '' 4.a 8.a 12. 16. 2a.
CJ : '' frequency (Hz)
+'
+' 12 .
:
'
'
right term
2nd ;~o1
A
8
·~
M
-':: 4
- - - _- l2r~ :':_ot
- ___ 4~hroot Fig. 3.51: Dispersion curves (Love waves) between 0 and 20Hz for the two
~
~ 0
---- soil models: h=50m (top), h'=30m (bottom).
-4
-8
a.a a.a2 a.a4 a.a6 a.a8 a.1a Applying the previous method, vve shall con1pute, for each soil model, the roots of
hb, Eq.(3.197) for various frequencies in the range [ll,20Hz]. As shown in Fig.3.51, the mnn-
ber of rr1odes increases vvith frequency: new rnodes progressively appear at sorne ,specific
Fig. 3.50: Number of modes at three different frequencies: 5, 10 and 15Hz frequency values. Con1paring both soil inodels 1 the nu1nber of modes is larger (in the
(h=50m, Vs., =400m/s and Vs,=800m/s). sa1ne frequency range) for the deepest layer. For a given rnode, the velocity is close to
the half-space velocity Vs 2 =800m/ s at 101,v frequencies and converges to the layer vcloc-
 2D /3D-wave propagation 3.7 Application 2: Spectral Analysis of Surface Waves and in situ tests 215
214

ity Vs, =400m/s for higher frequencies (Fig.3.51). The surface wave velocity decreases
rapidly for h=501n, \vhercas for h=30n1, the decay is much slcn·ver (wider frequency range).

Considering these theoretical results 1 it will novv be possible to characterize the var-
ious rr1odes at so1ne specific site thanks to experin1ental rr1ethods. A way of separating
different modeH is for instance to record them at various distances. Lower order rnodes
are related to large distances (long wavelengths) V\rhereas higher order 1nodes can be
characterized on s1nall distances (short \vavelengths). It ir.; thuR possible, using inversion
techniques, to find out what are the soil profile characteristics at the t:>pecific site consid-
ered. This is the basis of an investigation rr1ethod called the Spectral Analysis of Surface
Wa·ue8 (Bode\ et al., 2005; Foti, 2000, 2003).

3.7.2 Dispersion of surface waves in a heterogeneous half-space

Fig. 3.52: Dispersion of surface waves in a heterogeneous half-space:
Heterogeneous half-space schematic of the problem.

\lilc shal1 now consider a heterogeneous half-space having its mechanical properties de-
fined as functions of the vertical coordinate z: ),(z), µ.(z) and p(z) (Fig.3.52). Surfaces
Free-surface and continuity conditions. The displacements and tractions rriust be
waves propagating horizontally, along the x axis) will be investigated (Aki and Richards,
continuous at any interfaces z = zo. The z dependence of the stress component a beino·0
1980). rewritten as: yz

<Tyz = L2(VR, z, w) exp [ ~ (x - VRt)] (3.202)

Theoretical analysis for Love waves equations (3.200) and ('.l.201) thus yield:
Model and wavefields. For Love waves, the expression of the vvaveficld is: 3Lr L2
3z p.(z)

(3.199)
3L 2
az = [wVjiµ.(z)-w-p(z)
2
, l Lr
(3.203)

Uz = 0 that is, under a matrix form (Aki and Richards, 1980):

1
The related stress components arc the follo¥ring:
1 11(z)
3 { L } - ( O (3.204)
O":rx = Clyy = O"zz = Clzx = 0 oz
-
L2 - w2 •
2
v,:;µ.(z) -w p(z) 0
<Tyz
3Lr [iw ]
=fl. az exp VR (x - Vl1t) (3.200)
The boundary conditions at the fTee-suTface and at infinity lead to:

""" =
iw [iw
VRµ.L1 exp VR (x - VRt)
] lim Lr= 0
z~-;-oo
(3.205)
L2(z=0)=0
T·he equation of inotion is thus:
Sin1ila.rly to the case of Love waves in a single-layered half-space 1 cornbining Eq.(3.204)
2
3 ( µ.(z)-
3Lr) - -11(z)Lr
w (3.201) and Eq. (3.205) leads to an eigenvalue problern requiring a numerical Holution inethod (e.g.
-w 2 p(z)Lr =-;---
:i
3z 3z v~ Runge-Kutta algorithm).
 2D (JD-wave propagation :). 7 Application 2: Spectral A.nalysis of Surface V\laves and in situ tesfa.; 217
216

Theoretical analysis for Rayleigh waves We thus obtain the following matrix system (Aki and Richards, 1980):

Model and wavefields. For llaylcigh -,,vavcs, the expression of the \vaveficld iR (1\ki
and Richards, 1980):

w 1
iw (.~ -VRt)
'Ux = R1(VR,z,w)exp [ VR l w
()

>.(z)
VR µ(z)
0

1 n,
0 0
(3.206) VR >.(z) + 2µ.(z) >.(z) + 2µ(z) R2

1n, )
Uy= 0

'Uz=iR 2 (VR,z,w)exp [ VR(:r-VRt)

iw l w 2 4µ(z)(>.(z)+1;(z))
VI~ ,\(z) + 2µ.(z)
- w
2
p(z) 0 0
w ,\(z)
VR ,\(z) + 2µ.(z)
R3

w
0 -w 2 p(z) {)
VR
(3.209)
The related stress components are the follo\ving: The buuntlary conditions at the free-su1'face and at infinity lead to:
Jim R1 = 0 and Jim R2 = 0
z~-= z~-=
(3.210)

The eigenvalue problern thus obtained needs some nun1erical rnethud of solution. Lai
(1998) proposed a cornprehensive presentation of the various approaches.

Solution methods for the eigenvalue problem

(3.207)
Propagator-matrix method. The Propagator-1natrix method vvaR originally intro-
duced in seisrnology by Gilbert and Backus (1966). A special forrnulatiou of this 1nethod
was proposed by Thomson and Haskell (Thomson, 1950; Haskell, 1953). In this formula-
tion, the solutions of eigenproblen1 (3.209) are found considering the R,ayleigh dispersion
equation (Eq.(3.150)) at v;i,rious frequencies. The roots of the lla.yleigh dispersion equa-
tion give the \vavenumber corresponding to the rnodes of propagation of Rayleigh Vi.rave~
at each frequency. The eigenfunctions rnay thus be co1nputcd yielding the dependence of
the waveficld vvith depth (Lai, 1998). Haskell also propo8cd son1e asynrptotic expansionR
for short and long wavelengths. Nevertheless, the original Thornson-I--Iaskcll 1nethod is
not. fully stable at higher frequencies (Schwab and Knopoff, 1970).

Free-surface and continuity conditions. The displacements and tractions rnust be
co11tinuous at any interfaces z = zo. The z dependence of the stress ccnnponent CTzx and
a zz yields sirnilar notations as in the case of Love VI.raves:
Reflection and transmission coefficients method. Kennet (19711) proposed a n1e-
tho<l considering the reflection and transmission coefficients for a 1nultilayered inedium.
The n1ethod is very efficient and vvas further irnprovcd by other researchers (HiRada. 1
1995). Furthcrrnore, it also gives a direct insight of the various interferences occuring in
surface soil layerR.

(3.208) Dynamic stiffness matrix method. Starting from the ThomRon-I-Iaskell approach,
l(ausel and RoesRet (1981) derived a finite elerncnt forn1ulation using layer Rtiffness rnatri-
ccs sirnilar to those considered in structural dynarnics. It thus allowR the use of standard
 11
218 2D /3D-wave propagation 3. 7 Application 2: Spectral Analysis of Surface Waves and in situ tests 219

structural analysis techniques such as substructuring. This formulation simultaneously

solves the eigenproblem and the inhornogcneous elastodynarnic problem. 'rhe use of dynamic vibrator
closed forrrm was also proposed by Kauscl (2000) leading to a very efficient method force
('Thin Layer Method'). sensor

Other nurnerical methods. In addition to the finite clement n1ethod, other numeri-
cal rnethods may also be considered to solve the eigenvalue as well as propagation proh-
lem8: the finite difference incthod 1 the boundary clement rnethod or the spectral clement
rr1cthod. These rnethods 1,vill be presented in the next Chapter.

3.7.3 Steady State Rayleigh method

1'his rnethod consists in generating Rayleigh Vi'avcs at a given frequency f. Their wave
velocity, denoted VR, inay be \Vritten as: VR = f A where A is the wavelength. As
shown in Fig.3.53, for each frequency fi, it is possible to deterrnine 1\_,i using the relation
1\i = x/ki. 'I'he R,ayleigh wave velocity VR is known for each frequency. The shear wave
velocity Vs is close to that of the Rayleigh waves (Vs ~ VR). Fmthcrmore, Rayleigh
waves have a significant effect down to a depth of one v.1avelength. The velocity ·vs is
thus known for every frequency Ji, alknving the characterization of a so'il layer hav'ing a
mean depth of A;/2.

Heitz and Bonnet (HeitD, 1992) considered a nonlinear constitutive law to analyze
such harmonic tests. 1'hey could investigate the influence of the excitation level on the
characterization of the dynarnic soil properties. Fig. 3.53: Principle of the Steady State Rayleigh method.

3.7.4 Spectral Analysis of Surface Waves: experiments

Experimental set-up and principle
1'his rnethod represents an extenHion, and an improvement, of the harrnonic vibration of
a block foundation resting on the ground surface (Stokoe and Nazarian, 1985). In this
older method, the wavelength AR of a Rayleigh \Vave generated by the block foundation
vibrating at frequency f is rneasured \vith a geophone moved on the ground surface.
In a homogeneous soil profile, this velocity is frequency independent. AH shovvn in the
previous sections, in a layered profile, the wave propagation is frequency dependent and
the relationship between frequency and VI.rave velocity is callc<l the dispersion curve. This
phenomenon arises from waves with difi'erent \Vavelengths propagating in different lay-
en>: high frequency VI.raves (short vvavelengths) propagate near the layers interfaces; low
frequency waves propagate in HhalloV\r and deep layers. Altering the vibration frequency
allows to sample different depthH and to derive a Rayleigh wave velocity profile and, if
Poisson's ratio is knu\vn or estimated, a shear wave velocity profile.
Recorded signals are digitized and transferred into the frequency do1nain 1 the phase shift
D.¢ between tV\ro recorded signals is determined, frorn vvhich the propagation tirne and the
wave propagation velocity are computed. R,epeating these steps for different frequencies,
the dispersion curve can be established.
Analysis of dispersion from recorded signals
Computing the dispersion. Considering a large number of Hensors in the cxperilnen-
ta.l set-up (Fig.3.54), the medium is described in the time-distance plane (t - x) through
either wave velocitie8 or slo\vnesses (Foti 1 2000). The dispersion is not easily character-
ized in this plane whereas it is in the wavenumber-frequency plane (k - w). First, the
signal inust be corrected frorn geornetrical damping. It is then necessary to transforrn
the recordings frorn the x - t domain to k - w don1ain. 1'his can be done using the f - k
transforn1, also called the tvvo-ditnensional Fourier transform (Bodet et al., 2005):

The SAS\\T method takeH advantage of a rnore refined analysis of the records for the
layered profile and of an easier and faster implernentation. As shown in Fig.3.54, the
s(k,w) =.fl s(x,t)ei(kx-wt)d:rdt (3.211)

generation of Rayleigh waves is produced by an in1pulse loading at the ground surface. This cxpreHsion must be conHidered in a discrete form with:
220 2D /3D- \V8;vc propagation :). 7 Application 2: Spectral Analysis of Surface \\Taves and in situ tests 221

significantly different for both areas (line 8 and line 7) and correspond to an average fiat
layered medium at each location. Such J_ifferences illustrate the influence of the layers
geometry (depth, inclination, etc) on the reliability of the SASW technique.
source sensor J

Simulated data for three different profiles. Several synthetic data a.re no-1,v pro-
layer l
posed to con1pare different geotechnical profiles depicted in Fig.3.57. As shcnvn in
Fig.~1.58, the analysis in ter1us of frequency-slowness (in Pls/m") is very close to the
layer 2
frequency-wavenurnber transforrn. Foti (2000) sirnulated the data corresponding to the
three different profiles displayed in Fig.3.57. From the frequency-slo-1,vness curves, the
d, layer 3 dispersion Jnay be easily assessed and inversion procedures rnay lead to the corresp,ond-
ing velocity profile if unknown.

Fig. 3.54: Schematic of surface wave experiments (SASW).

line 8 line 7
Do11vn-dip Up-dip
• the frequency sarnpling related to the time sampling: 40 mm •J ":rr 1: inm 4Q nun
~s~urt~a~c!e~~~~~~~~~,iiii= ~ ~~~=:.~~~~~~~~-
l:lf = _1_ = _1_ (3 212) . · 4.2' mm' 3.1[mm PMMA
tmax 1\f1)~.t 5.5'mr~~r----l',_-L--------;-:AI
-c-0-ur-,1-·1n_g_lo_y_e_r_(_-1,.-ci-11"'1)_ _;;.;.;;..~6..,.15 mnl 4,65 mrn
• the \Vavenumbcr sarnpling related to the spatial sarnpling: average thickness at the base of each line
(includin~1 coupling)
otY:>et UrdlinHiers) afset :mJlirnet8cs)
(3.213)

T'he distance dij bet\veen the sensors is thuH an irnportant pararncter since it may induce,
as the tirne san1pling 1 spectral aliasing. It governs the n1inirnurn >Navelength to which the
dispersion n1ay be investigated (Foti, 2000). 'I'he nurnbcr of sensors is also i1nporta.nt
l:'lince the longer the distance the better the resolution in frequency. A. longer sensor
netvvork thus leads to a higher (spatial) resolution of dispersion.

Inversion of the dispersion data The dispersion phenornena being characterized,

inverse methods are needed to estimate the velocity profile of the layered Roil. However,
as sho1vn by Backns and Gilbert (1967), the inverse problern is nonlinear and there is
no unique solution. Various rnethods are available (Tarantula 1 1987; Hunaidi, 1998; Lai,
1998; Beaty ct al., 2002; Wa.thelct et al., 2004). One of the main limitations is that the
direct problern is generally assumed to be lD. As it 1vill be shown in the next paragraph,
this assumption is not always valid and the influence of 2D /31) effects rnay by significant.

Experimental results
Analysis in the lab on model structures. On model structures, Bodet et al. (2005)
investigated the influence of 21) effects on the analyHis of surface 1vavc propagation. They Fig. 3.55: SASW: experimental configuration for a PMMA clipping layer
considered a. dipping PMMA layer on an aluminum plate (Fig.3.55) and recorded the (top) and recorded signals (bottom), from Bodet et al. (2005).
wavefields at different distances from the source. As sho\vn in Fig.~1.56, the results are
222 2D/3D-wave propagation 3. 7 Application 2: Spectral Analysis of Surface \j\T;-:i;ves and in situ tests 223

ph<>se velocity (nvs) phase velocity (mis)

150 150~~~

1500 2000 2500 3000 'I ODO 1500 2000 2500 3000
o+;;.~...,...._~..,.~~..,..~~rn..~
A B

I
100 100

-
200 200 100

N 300
"'g;»
~ 2,00 Q

~·
"'"o<
"'"
ill
6-400 400
4~ ~
50
500 500

600 600

line 8 line 7 0
2.0 3.5 3.5
slowness (s/m) x1 o-3 slowness (s/m) x10- 3 slowness (s/m) x10-o

Fig. 3.56: SASW: example of measured dispersion curves in a PMMA dipping Fig. 3.58: Slowness curves for three different profiles (Foti, 2000).
layer (from Bodet et al., 2005).

:F'oti (2000) proposed a sin1plified procedure to invert the velocity profile frorn the
Profile A Profile B Profile c
shear wave velocity, m/s
frequency-slowness curves. A.s sho¥.rn in Fig.3.59) for three different lengths, the inverted
shear wave velocity, m/s shear wave velocity, m/s
velocity profiles arc 1vell correlated to the actual ones. Nevertheless, for profiler.; B and
300 400 500 250 350 450 300 400 500
C, since the velocity does not follovv a n1onotonic increase with depth) the inversion is
0 ~~~~---~ 0 0
less satisfactory. 'I'his is generally the n1ain difficulty 1.vhen perforrning velocity profile
5 inversions.
5 5

10 10 10
E E E
£ 15 :S- 15 Analysis in the field on actual soils. Figure 3.60 presents a dispersion curve con-
'5. 15 ~ ~

• • •
" 20
structed from the SASW method (Stokoe and Hoar, 1978). The theoretical curve is
" 20 " 20 co1nparcd to the experimental one and the 1.vave velocity profile if.; altered until a good
agree1nent is reached between both curves. The wave velocity profile computed at the
25 25 25 last iteration is deerned to represent the actual velocity profile. This procedure is illus-
trated in Fig. 3.60 (Stockoc and Hoar, 1978) which shows the evolution of the velocity
30 30 30 profile with the iterations and the n1easured dispersion curve. The final ¥lave velocity
profile has been con1pared to measurements from a nearby cross-hole test) validating the
Fig. 3.57: Description of the three different profiles considered by Foti (2000). results. However 1 the SA.S\iV method yields a stepped wave velocity profile as opposed
to the more continuous variation obtained from a cross-hole test; this arises from the
sarnpling inethod '°'rhich ir.; more global in the SAS\i\T rnetho<l.
224 2D /3D- \Vave propagation 3.'7 Application 2: Spectral Analysis of Surface \i\laves and in situ tests 225

Profile A Profile B Profile C paths at interfaces between layers. For a three layers soil profile (two soil layers overlying
shear wave velocity, m/s shear wave velocity, mis shear wave velocity, m/s a bedrock), the refraction paths are depicted in Fig.3.61 (top). As discussed earlier in
300 400 500 250 350 450 300 400 500 this Chapter, the critical incidence.s may be esti1nated from the Snell-Descartes law (i.e.
0 0 frorn the velocity ratio between the soil layers).

5 5 The slopes of the straight lines obtained by plotting the propagation time versus the
distance are equal to the \ii.rave velocities (Fig.3.61 1 bottom). Such plots are obtained
10 10 fron1 more con1plcx experirnental traces. Such signals recorded by (Gueguen et al. 1 2007)
E E
are displayed in Fig.3.62. ~rhe source should preferably give rise to shear waves, vvhich
£
Q
15 £
Q 15
is achieved by horizontally hitting a block laid on the ground. \i\'ith this technique, an
© ©
D D
alrnost pure shear \ii.rave propagates perpendicularly to the source. Interpretation of the
20 20
-real profile tests is done assurning that the wave velocity increases vvith depth. If this condition is
25
--a- lambda/3
-- iambda/2.5
~lambda/2
25
1 not satisfied, significant erron.; can affect the interpretation. Likewise, the presence of a
soft layer (1.vith a smaller wave velocity) entrapped betvveen two stiff layers cannot be
30 30 detected. Finally, in this rnethod 1 the \Vaves propagate near the layen:l interfaces; it is
therefore impossible to sarnple an average wave velocity for the layer under consideration.

Fig. 3.59: Approximated inversion for three different profiles (Foti, 2000).

500
••
"'
+'
·~

"0 400
-
IJ.
Experimrnlal Dispenlm1 Curve
Thsoretirnl Dispersion Curve, Profile 1
.:. TI!eoretirnl Dispersion Curve, Profile 2 A
IJ.

o Theoretical Diipersion Curve, Profik 3 A

"'"
>
>
300
· ·-. Prnfik 1
Profile 2
-Profile3
layer 2
v,,
"i< 200
H
~
"ii3 layer 3
100
;;',
~" 25
0
10 WO 0 1{)() 200 300 400 500
0' lN,,
wavelength AR (m) shear wave velocity Vs (m/s)
w

""'
t;2
Fig. 3.60: SASW: example of measured dispersion curves in soils (from Stokoe ~
Q) b
and Hoar, 1978). ""w
~

·~"'
til

;.,

3.7.5 Seismic refraction "w

+'
;.,
q::
'fhis method is 1,vell suited for prelirninary site investigations but its in1ple1nentation may distance
be difficult. Theory and exploration techniques arc \vcll kno-\vll and have been described
in details in textbooks (Richart et al., 1970). As shown in Fig.3.61, the method consists
in determining the propagation time of body 1,.vaves (P or S) generated by a surficial Fig. 3.61: Principle of the seismic refraction method: source and receivers
so11rcc to a linearly aligned net\vork of receivers. According to the soil profile beneath arrangement (top), first arrivals curve (bottom).
the reccivers 1 the \ii.raves propagate either directly tovvards the receivers 1 or along refracted
 T

226 2D /3D-wave propagation 3. 7 Application 2: Spectral J\_nalysis of Surface V\raves and in situ tests 227

60m down-hole test up-hole test

2.5m
1~~1---- Ig~1----------------
~··
~ Seismic source
' Horizontal geophones

offset (m) offset (m)

10 20 30 40 50 60 -60 -50 -40 -30 -20 -10
'

0.1 .. ..!. ~
>

~ ··~:~
0.2
_...,,,
'"' ""•
I<.,_ '

:§::o.3 )~

"E
F 0.4 .:-er;'
";;- ~/!'
t ~ Fig. 3.63: Schematic of down-hole and up-hole tests .

.
0.5 I • 1 ' Down-hole tests
) ' )'
•
0.6 The rr1easuTement of the wave propagation velocity is pcrforrned along one hole (Fig.3.63).
I
• 'l'he source is placed at the ground surface and the receivers are placed inside the ca8-
ing. T·hc use of adequate sources (horizontal irnpact of a block 'foundation) gives rise to
Source position : 60 m predon1inant shear waveH. In its simplest version, only one receiver is used and lo\vered
Source position : 0 m
in the hole at successive depths. More elaborate techniques take advantage o:f a set of
receivers clamped on the sa1nc fran1e vvhich allovv for a simultaneous recording of so1ne
Fig. 3.62: Traces measured in a seismic refraction test with a source located
input signal T·he accuracy in the determination of the wave velocity is better because the
successively at both ends of the network (Gueguen et al., 2007).
distances bet1,veen receivers are perfectly knovvn and because the signals are originated
fron1 the sarnc irnpulse. Velocities measured in a dov,rn-hole test correspond to values
along the vertical side of the hole, for a vertically propagating v,rave. In-plane variations
of these velocities cannot be assessed. On the other hand, with a sufficiently sn1all spac-
ing betv,reen the receivers 1 it is possible to sample layers with lower characteristics, even
if they are entrapped in between two stiffer layers. This is one of the n1ajor advantages
3. 7.6 In-hole tests of the n1ethod. Figure ~-L64 presents an exarnple of a dovvn-hole record showing the way
\Vave velocities rnay be estim.ated.
In-hole tests can be grouped into t\vo categories: tho8e requiring a prior drilling of bore-
hole(s) in which n1casuring devices arc lo\vered do\vn (cross-hole, do\VI1-hole) suspension
logging) and thoHe for \vhich the measuring device is lowered with the drilling tool (seis- Suspension logging
rnic cone). For the first category, tubing of the hole Vlrith a PVC casing is required and
the annular space between the casing and the hole rnust be grouted to ensure a good This n1ethod, for the measurements of P and EJ wave velocities, uses only one, preferably
coupling \Vith the surrounding soil. These operations are time-consurr1ing and costly; in uncasecl1 hole. The borehole is filled -with water or drilling n1ud. The receivers and the
addition, it is absolutely required to rneasure the holes inclinations. source are clamped on the sa1nc frarne and lov,rered down at -the Harne time (Fig.3.65).
 T
228 2D /3D-wavc propagation
'!rt~;: 0{!f)!'f'['{)f)/}f'£S,'1";;~;~-;:. ~ Vs ,
time
Fig. 3.64: Typical signals recorded in a down-hole test.
The source, with the capability of inverting the v:.rave polarity, is made of a horizon-
tally oriented solenoid v:.d1ich generates a pressure \.Vave in the fluid. V\rhcn hitting the
borehole wall, tl1is presRure -i.vave induces a radial displacerncnt in the soil vvhich gives
rise to radially and vertically propagating body \Vavcs in the soil. The propagation and
displacc1nent of the boreholes "\Vall in turn induce pressure \vavcs in the fluid "\Vhich are
recorded l>y biaxial geophoncs. The separating distance bctv;.reen tvvo geophoncs is 1 m
which perrnits an accurate resolution of the signal. 1'he fran1e vrith a total length of 7 rn
is suspended vvith a wire li11e fron1 the ground surface; this test set-up rnakes possible
the investigation at great depths, in exces8 of 100 n1 (Nigbor and I1nai, 1994).
3. 7 Application 2: Spectral Aualysis of Surface \i\Favet-> and in situ tests 229
D
upper geophone --+-:>D
lower geophone --t~:>tl
source --f--:>t::J
source --f--:>t::J
driver
Fig. 3.65: Suspension logging test set-up.
Seismic cone
This is the only in-hole ineasuring technique which does not require the prior <lrilling of a
borehole. 1'he equi.p1nent is sin1ilar to the one used for the cone penetration te8ts (CPT');
in addition to the gages for n1casuring the tip resistance, the skin friction, eventually the
excess pore pressure) the cone i8 provided vvith a piczo-resistive gage for the detection
of seisn1ic waves antl Vilith inclinometers for the measurernent of verticality (Campanella
and Ste1vart, 1992). The seis1nic 1vave is generated at the ground surface Ly any appro-
priate rnechanical device which favours the generation of shear waves. R.ecurding8 of the
signals are rnade at stops during penetration, every 0.5m or ln1. In order to compare
the arnplitudes of the signals at variou8 depths and to cornpute the material da1nping,
the source must be capable of generating reproducible signals. The test interpretation
is 8traightforward and resorts to the co1nputation of the propagation time of the VI.rave
bet\veen tvvo successive positions of the receiver.
T'hc major advantage of the test if>, besides its low cost and quick realization, to allovv
\Vithin the same test the determination of different parameters vvhich can be used to
characterize the site frorn a geotechnical standpoint: tip resistance 1 friction ratio, pore
pressure ratio, per1neability1 and vvavc velocitieH. The li1nitations of the test are t->irnilar
to those of the down-hole \est: good definition of the velocity profile only along the hole,
lhnitation of the sa1npled depth because of the limiting capacity of the source.
 T

230 2D /3D-wave propagation ~1. 7 Application 2: Spectral J\nalysis of Surface V\Taves and in situ tcst,s 231

Cross-hole test
4,---~~-~~~-~~-~~~~~~~-~~-
Principle of the test. Unlike the other in-hole techniqucs 1 the cross-hole tests require
the drilling of at least two cased boreholes equipped \vith PVC tuhings, grouted to the 3
soil. To i1nprove the test quality, it is better to use three, or rnore 1 aligned boreholes 2
(Fig.3.66), spaced by a few meters (typically :Jm to lOm). The impulse is imparted in the
emitting borehole and the generated waves are recorded in the receiving boreholes with
triaxial geophones located at tl1c same depth as the source. The source and the receivers 0
are knvcred in the holes and clarnped to the casing with pneurnatic or hydraulic packers.
-1

-2
source
-3
_4L...__~~~~~~~~~~~~~~~~~~~~_J
4,---~-,~~~~-~~-~~~-~~~-~~~

3 [ P-waves) horizontal
2 / component

ol---__.
refracted path -1
-2
-3
_4'---~~~~~~~~~~~~~~~~~~~~-

Fig. 3.66: Cross-hole test set-up: direct and refracted waves. 4,---~~-~~~~~~-~~~~~~~-~~-

3 [ S-waves) vertical
'l~he n1ost com1nonly used sources are mechanical harruners with a vertically oriented 2 / component
stroke irnparting a shear displace1nent to the borehole wall; this stroke gives prilnary
rise to ~hear waves propagating horizontally in the soil. This type of sources allo\vs an
inversion of the direction of the stroke which n1odifies the polarity of the shear waves; in 0 f---.IJ~-.-J
addition, it is potisihle to sum up the signals frorn different, reproducible 1 strokes. Use of -1
at least tvvo receiving holes pre8ent rnany advantages:
-2
• the propagation time is con1pute<l fro1n differential measurements between receiving -3
holeR, when vv~ith a t\vo holes set-up, it is cornputed bet\veen the source and the
-4;--~~~;--~---:::-::::-~~-:--::-::-~~~~~~~~~_J
receiver; the coupling bet\veen the emitting source and the casing being weaker than 0 0.01 0.02 0.03 0.04 0.05 0.06
the coupling between the geophones and the casing 1 the deter1ninatiun of the initial time (s)
tirne (tirnc zero) rnay be subjected to errors. \i\Then only two holes are requircd 1 it
is rr1andatory to have a geophone attached to the source; Fig. 3.67: Typical signals from a cross-hole test: emitter (top), horizontal
• identificatior1 of refraction paths in stiffer layers located at shallow depths below component at receiver 1 (middle) and vertical component at receiver 1 (bot-
the n1easuring depth is possible. As shcnvn in Fig.3.66, refraction path8 in stiffer tom).
layers arc a con1mon ,source of error in cross-hole teRts (Stakoe and Hoar 1 1978).
232 2D /3D-wave propagation 3. 7 Application 2: Spectral Analysis of Surface V\1aves and in situ tests 233

3 To estimate the Virave velocity in the soil, it is possible to con1pare the signnls at tViro
receivers. As shown in Fig.3.68, the ina.xirnun1 arnplitude is reached at t 1 =0.01620s for
2
receiver 1 and t2=0.02625s for receiver 2. The travel tin1e between both receivers is thus
L'ltd=0.01005s.
0
Frorn these signals, vve rnay co1npute the cross-correlation as displayed in Fig.3.69
-1
(see Appendix D for its theoretical expression). Frorrr this curve 1 the travel ti1nc is found
-2 to be 6.tc=0.01015s vvhich is very close to the travel ti1ue directly esti1nated fronr the
-3
tin1e signals. Nevertheless, the estin1ation by the cross-correlation is nrore reliable since
it also takes into account the shape of the signal and not only one single peak.
-4
4
t;t, Misinterpretation in cross-hole tests. Cross-hole tests are very efficient but several
3
types of errors inay lead to flignificant mi8interpretations of the test: poor estin1ation of
2 the arrival tin1c of shear 1.vaves, bad quality of the reference ti1ne (description of the
source), influence of refracted waves in stiff layers, variability of the soil propertieR 1 etc.

0.8

-1
0.6

0.01 0.02 0.03 0.04 0.05 0.06

time (s) 0.4

<I
0
-~
~
Fig. 3.68: Typical signals from a cross-hole test at two different locations:
receiver 1 (top) and receiver 2 (bottom).
~
"'
Q)
;.,
;.,
0.2

0
u
00 0
00
0
The inajor advantage of cross-hole tests \vith respect to other in-hole rncasuring tech- ;.,
u
niques is the possibility of obtaining rnechanical characteristics within a horiDontal plane -0.2
and not along a borehole. ()-1,ving to the geological depoRition of sedirncntary deposits, the
variation of soil characteristics is more pronounced in the vertical direction than in the
-0.4
horizontal one and the cross-hole test is therefore vvcll-suited. If the receiving boreholes
arc arranged along different azimuthal directions (usually tvvo rnutually perpendicular
diJ_-ections), it is possible to mea.sure the in-plane anisotropy of a layer; this possibility -0.6
0 0.005 0.01 0.015 0.02 0.025
rnight be interesting in highly tectonized rock deposifa:i.
time (s)
The cross-hole test, oViring to its high versatility, and despite its relatively high cost 1 is
Fig. 3.69: Cross-correlation of two cross-hole test signals showing the travel
the ino.st po"\verful used test for i1nportant structures. Depths as deep as lOOrn can easily
time (see Appendix D).
be reached in cross-hole testH. Typical signah:; fro1n a cross-hole test are displayed in
Fig.3.67: the horizontal (radial) component at the receiver (middle) is mainly composed
of P-wavcs propagating faster than S-"\vaves characterized by the vertical cornponent
(bottom).
234 2D /:3D-Virave propagation 3.7 J.\.pplication 2: Spectral Analysis of Surface \\Taves and in situ tests 235

The influence of shallow Htiff layers n1ay be large since \VaveH refracted hy such layers
rnay influence the interpretation of the cross-hole ter.;t (Stokoe and I-Ioar 1 1978). As
depicted in Fig.8.66, the refracted 1,vave rnay reach the receiver before the direct wave.
The interpretation of the test 1vould thus be biased because of a shorter estirnatcd travel
tin1e. The refractjon in a deeper stiff layer rr1ay he found under the follo1oring condition: ( µ=a.Z'' J
(3.214)
source reception
\Vhere x is the distance bet\veen both holes, H is the vertical distance between the stiff
layer and the bot.torn of the holes) V1 and V2 the \.Vave velocities in the soil and in the
stiff layer respectively. z
From thi.s condition, the admissible zone for crosR-hulc tests are detern1inecl and diR-
played in Fig.3. 70. The shaded area corresponds to the situations v,rhere refr·action may
11ot occur thus avoiding rnisinterpreta.tion of the cross-hole test due to such phenomena.
x
The variations of the soil properties 1vith depth rr1ay also influence the interpretation
of croHs-hole tests. Since tbe shear modulus increases with depth (faRter near the free-
surface); the v.rave path may not be linear. The ratio betvveen the actual velocity Vact 0 0.5 1.0 1.5
and the estirnated one Ve t rnay thus differ frorr1 unity.
8
Z/x

1.01--------===========1 Fig. 3. 71: Influence of variable soil properties on cross-hole tests.

no
refractidn Tu estirnatc the potential error, we assume a power law for the variations of the shear
modulus with depth:
domain of (3.215)
refraction
For such a dependency, the velocity error Vru:r/~sl iR diRplayed for several fJ val-
x
ues in Fig.3.71. For a classical experimental configuration (distance bet1veen two holes
} 0.5 x=lO rn) i the velocity error is less than 53 for a 1neasurement depth larger than 3 m
(!J~l/3) or 4.5 m (!l~l/2). Closer from the free-surface. the velocity error may thus be
significant.

3.7.7 Microtremor methods

The microtremor method consists in recording ambient vibrations due to hurnan activi-
Q'--~-'-~~~--'--~~~~~~'--~~~~~--'--~~~~~~
ties, to the "\Vind, etc (Bonnefoy et al., 2004; Cornou et al., 2003; Duval, 1996). As shown
0 5.0 10 15 20 in Fig. 3.72, both the horizontal and vertical components of the motion, denoted h(t) and
x/H v (t), are recorded. VVc then <lcter1nine the Fourier spectra of the time signals, denoted
H(f) and V(f). Tbe calculation of the spectral ratio H(f)/V(J) allows the character-
Fig. 3. 70: Admissible zones for cross-hole tests. ization of the fundan1ental rnode of the geotechnical structures (sharp peak around the
funda1nentaJ frequency Jo). This spectral ratio is called H/V m,icrotrernor ratio.
 11I!

236 2D /3D-wave propagation 3. 7 Application 2: Spectral Analysis of Surface VVavcs and in situ tests 237

In-situ testing techniques differ by the possibilities they offer and the information

spectral I / '". . HN
they provide. On the one hand, depths reached with geophysical tests pcrforn1cd frorn
the ground surface (seisrr1ic refraction, SAS\Al, <lo-,,vn-hole, seis1nic cone) arc lirr1itedi typ-
ratio
~
frequency
ically of the order of 50m; on the other hand, techniqueH in which the source iH lo\O.'ered in
the hole, rnaintaining a li1nitcd distance between the source and the receiver (suspension
logging 1 cross-hole), rnay sarnple larger depths, in excess of lOOm.

i V(f)
Fourier Fourier
Techniques involving only one hole (susperu~ion logging, dovvn-hole, seisrnic cone) give
only local inforn1ation on the soil characteristics, those of the soil in the in1n1ediate

~> spectrum vicinity of the hole; ho\i\rever 1 they allo\i\r an accurate definition of the profile. Scisrnic
spectrum
refraction or SAS\V tests provide global infor1nation on the 1ncdiun1, but do not allo\i\r
frequency an accurate definition of the soil profile. Cross-hole testH are a goocl con1prorniHe giving
average in-plane characteristics, at a scale of approxiinatcly lOm, while maintaining a
good accuracy for the definition of the soil profile.

t. . . .
.1 ..... .1. v(t)
stacked
vertical
stacked
horizontal t.
11J11.111l1J11.111l1. h(t) The n1ain li.n1itation of field tests, besides their inability, in the present state of prac-

~I+---~ ~~-~+1~~
component component
(x or y)
Fig. 3. 72: Principle of the microtremor method.
The rnicrotrernor ratio n1a.y be cornputcd for each of the horizontal con1ponentR of the
motion: H.T(f)/V(f) et IIy(.f)/V(f) (denoting Oxy the free-surface plane). An exam-
ple •.vill be shovvn in Chapter 5 to investigate the rnodal features of an alluvial deposit
(Scrnblat et al., 2000).
3.7.8 Conclusions on field tests
All the previouflly described field techniques require, for a reliable interpretation, an ac-
curate knovvledge of the soil profile: for the cross-hole teRts, it prevents fron1 inistakes in
the interpretation; for SASVV, it gives an initial guess for the iterative calculations; for
all the tests, it helps defining a meaHuring depth increment adapted to the Hoil profile. It
is therefore of para1nount importance 1 for an accurate interpretation of the tests, to have
a continuouH log of a borehole close to the test. For the down-hole or suspension logging
tests 1 it can be the testing hole; for the cross-hole test, it can be one of the holes.
tice, to induce large strains in the soils and therefore to test the soil behaviour beyond
its elastic range, lies in the fact that the characteristics arc evaluated under the stress
Oeld prevailing in the soil profile at the tirne of testing. lVIore than often, constructions
involve significant inodificationH to the stress field in the soil: these variations arc re-
lated to exca:vations 1 backfills; 1nodification of the hydraulic regime, stresH increase due
to buildings, etc. Since elastic soil properties (shear modulus 1 for inHtance) depend upon
the effective stresses Hupported by the soil, these rnodifications to the initial stress field
lead to significant variations of the soil rnechanical properties \i\'hich must be assessed. It
iH therefore necessary to evaluate the in-situ stress field at the tirnc of realization of the
tests and co1nplernent the tests Vilith laboratory tests in which variable streHH fieldH can
be i1nposed to the sarnplc. For projects of secondary importance 1 with a limited budget,
the variation \Vith Rtress of the Hoil properties can be assessed on the basis of the classi-
fication tests, together vvith experimental published correlations on sirnilar 1natcrials.
Finally) it is 1na.n<latory to co1uple1nent the in-situ geophysical tests -,,vith an accurate
description of the materials ancl \i\rith laboratory tests including, as a n1inimumi clas;;;ifi-
cation tests (\i\rater content, dry unit weight, grain siLJc distribution, Atterberg limits, ... )
and, better, tests characterizing the general soil behaviour (con1prcssibility tests, triaxial
teRts) and the cyclic and dynamic behaviour. In addition, the geophysical investigation
should give an estirnate of the horizontal and vertical effective stresses in the soil at the
tin1e of the survey to allovv a correct interpretation of the tests and an extrapolation of
the results to other HtreHs conditions.
Chapter 4

Modelling wave propagation

4.1 Numerical methods for wave propagation

4.1.1 Modelling wave propagation
AR discussed in the previous chapters, wave propagation problems are characterized by
various phenomena (Aki and Richards, 1980; Bourbie et al., 1987): dispersion 1 scatter-
ing, attenuation, \Vave conversion, etc. 'rhe governing paran1eters of all these phenomena
can not always be cstirnatcd cxpcrin1cntally. It is often necessary to pcrforrn 1no<lel ex-
periments (model materials (Bodet et al., 2005), reduced scale model (Chazelas et al.,
2001; Coe et al. 1 1985), etc). A8 shovvn in Chap.2, Hopkinson bar tests for instance
allo\i\r to control the fast dynamic response of a 1naterial through indirect measure1nents
of the waves at both sides of the specimen (Semblat et al., 1999; Zhao et al., 1997).
One rnay also consider nurncrical inodcls and/ or inverse rncthods in order to character-
ize the material and the v,raves propagating through it (Lai et aL, 2002; Foti, 2000, 2003).

The various numerical methods allowing modelling of vvave propagation (finite differ-
ences, finite clc1ncnts, boundary clements, etc) have different advantages and draV1.rbacks.
The finite clcrncnt n1cthod is for instance convenient to have a reliable description of the
n1aterial behaviour, but does not allow an easy modelling of wave propagation in ine-
dia having an infinite extension. Furtherrnore: the finite difference or the finite elernent
n1ethods induce a dependence of the propagation features on some algorithmic parameters
(time integration scheme, spatial discretiiation, etc) (Bamberger et al., 1980; lhlcnburg
and Babuska, 1995; Semblat and Brioist, 2000). Other numerical methods (boundary
ele1nents, spectral elerneuts) elirninate or lirnit such pheuornena (Bonnet, 1999; I3eskos,
1997; Faccioli et al., 1997; Komatitsch and Vilotte, 1998). It is thus necessary to discuss
the advantages and drawbackr.; of each numerical method.
240 2vlodelling \va.ve propagation
4.1.2 Numerical Modelling in Elastodynamics
10 analyz.e 3D clastodynamic problcn1s) various nurnerical rnethods are available:
• the finite elerncnt rnethod which efficiently deals vvith cornplcx gcon1ctrics and nu-
rnerous heterogeneities (Chammas et al., 2003): even for inelastic constitutive mod-
els (Dcl6pine Pi al., 2007). It has nevertheless several drawbacks Huch as numerical
dispersion (and darnping) (Derae1naeker et al. 1 1999; Thlcnburg and BabuSka, 1995;
Semhlat and Brioist, 2000) and (consequently) numerical cost in 3D elastodynam-
ics,
• the finite d4ffcrence ·m,ethod \Vhich is very accurate in clastodynarnics but is mainly
adapted to sirnple geometries and linear constitutive rno<lcls (Frankel and Vidale,
1992; Graves, 1996; Mocw et al., 2002, 2004; Saenger et a.l., 2000; Virieux, 1986),
• the botindary elcrnent rncthod \Vhich allo\vs a very good description of the radiation
conditions but tends to be lim.ited to 1veak heterogeneities and linear constitutive
models (Banerjee et al., 1988; Beskos, 1997; Beskos et a.l., 1986; Bonnet, 1999;
l)angla, 1988; S<inchez-Ses1na and Luz6n, 1995; Yokoi, 2003) 1
• the spectral element method \Vhich has been increasingly considered to analyse
2D/3D Vi.Tave propagation in linear media (:F'accioli et al. 1 1996, 1997; Kon1atitsch
and Vilotte, 1996, 1998),
• the Aki-Larner m,ethod which takes advantage of the frequency-wavenumber dc-
cornposition but is lin1itcd to simple geometries (Aki and Larner, 1970; I3ouchon
et al., 1989; Kawase, 1988),
• series expansions of u ave functions 1vhich give a semi-analytical esti1nation of the
1
scattered V..Tavefield for simple geornetries (Lee, 1984; .i\!Ioeen-Vaziri and Trifunac,
1985; Simchez-Scsma, 1983).
Each inethod has specific advantages and drcr~vbackR. It is consequently often more in-
teresting to cornbine two methods to take advantage of their peculiarities. One co1nm.on
inethod in elaHtodynan1ics is to couple FEM and BEl\iI allov..ring an accurate description
of the near fielrl_ (:FE:rvl n1odcl including ccnnplex geometries, heterogeneities and constitu-
tive behaviours) an<l a reliable eHti1nation of the far-field (I3E1vI rnodcl involving radiation
conditions), (Dangla, 1988).
In this chapter, t\vo n1ethods vvill he mainly discussed: the :Finite Elen1ent lVIethod
and tl1e Boundary Ele1nent 1\ifethod.
4.1.3 Time domain vs frequency domain
Nurnerical modelling of vvaves and vibrations in soils 1nay be pcrfor1ne<l either by tilne
domain integration or by frequency dornain solution rnethods. In the field of struc-
tuTal d:ynamics, 1nodal methods, including superposition 1 are often used but arc not fully
4.1 Numerical inethods for Vi.Tave propagation 241
adapted to analy~c \vavc propagation (Clough and PenLJicn 1 1993). Hcnvever 1 a sirnpli-
(ied inodal approach adapted to the analysis of geological Htructures will be discussed in
Chapter 5.
Tin1c integration rnethodR allow an easier irnple1nentation of time-dependent constitu-
tive lavvs as vvell as nonlinear constitutive la1vs. They consiRt in a numerical esti1nation of
the solution at various times starting fro1n the initial conditions and the time variations
of the loading. These n1ethods involve assu1nptions allu1,ving the approximation of the
t.i1ne derivatives of displacernent. For linear constitutive laVi.TS, the Rolution is obtained
by solving a linear set of equations. For nonlinear problems, it is generally necesHary to
irnplen1ent numerical iterative procedures at each time step. Many ti.Inc integration algo-
rith1ns can be considered (I-Iughes, 1987), Henne of the1n vvill be discussed in the follo1,ving.
Frequency do1nain Holution methods are ¥lidely used in Boundary Ele1nent approache8
(Bormet, 1999). They are very powerful in the field of dynamics/wave propagation since
the equation of motion may be easily <lerived in the frequency don1ain (tirne derivatives
expressed as si1nple products by a fa.ct.or ±iw). As shown in (Jhapter 2 1 they arc also
convenient in accounting for attenuation.
4.1.4 Actual or synthetic signals
Nu1nerical modelling of wave propagation pheno1nena raiHes the need for the esti1nation of
the dynamic properties of the inediurn, but also the features of the loading. Various-types
of loadings rnay be considered depenrl_ing on the ain1 of the analysis (design 1 parametric
study, etc):
• actual seisn1ic signals or vibrations recorderl_ at a site 1,vhich n1ay be used (directly
or after deconvolution) as a loading/input in a nu1nerical n1odcl,
• synthetic signals cornputcd fron1 standard Rpectra of the sitc 1
• purely synthetic signals determined frorn sin1ple analytical expressions.
Several types of wavelets are preHented hereafter; several others are detailed in ap-
pendix B.
Original Ricker wavelet
In the latter class of signah> 1 R,icker (1943, 1953) proposed con1plex synthetic wavelets
to moclel wave propagation due to a shot pulse. fie considered the \vave equation in
dissipative rnedia, as firRt proposed by Stokes:
V'
2 (
"+ Wo1 Bu)
--c-
iJt
(4.1) i' I
\vhere u is a co1nponent of the displacc1ncnt field J£ and wo is a constant.
242 1\!Iodelling wave propagation 4.1 N umer.ical in.et hods for >Nave propagation 243

• F'or R=lO: the wavelet is much t:nr1oother than in both previous cases.
• li'or R=l5: the vvavelet iH nearly .':lymn1etrical vvith a n1ain negative peak and two
secondary positive peaks. This shape is close to the simplified R,icker \vavelet
defined in the next paragraph.
l\/Iany various types of seismic \Vavelets vrere recently discussed by ]'\/Iavrocidis and
Papageorgiou (2003): they arc 1nainly due to Gabor, I3crlage) Klipper or Ricker (see
appendix B). In thi.':l paper) these authors also proposed their o\vn analytical model and
perform.ed extensive identification studieH from actual seiHmological data.

Simplified Ricker wavelet

In the Ricker class of vvavelets, sirnplificd R.icker \vavelets are often used in geophyHicH
............ ·········"··-·- - - - - - - and seismology (Bard aml Bouchon, 1985; Bisch et al., 1999). This family of Ricker
vn-1velets is derived from a gaussian and the signals consequently have a strong decay in
both tin1e and frequency dornains (sec Fig.4.2). At order 2, the expression of the Ricker
wavelet is as follows (Mavroeidis and Papagcorgiou, 2003):
2 2
,(t-t,) ) [ ,(t-t,,) ]
(4.3)
t
R2 () = A o ( 211" - 1 exp -1!" ')
tp2 tp
Fig. 4.1: Time variations of the original Ricker wavelets for various dimen- where t 8 iH the tirne related to the maximum amplitude of the \Vavelet and tp is the
sionless distances R=5, 10, 15 and 20 (Ricker, 1943, 1953). funda1nenta1 period of the Hignal

As sho\vn in Fig.4.2, it is also posHihle to define the Ricker \Vavelet breadth tb as

The expression of the seismic -,,vavclct propo8ed by Ricker (1953), in terrr1s of velocity, (Ryan, 1994):
is the follo\ving: (4.4)

2
·u=-RJo r= 2 [
{J cxp -R(3(l+{J
2
)- i sin (arctan(3)]
1 4
2
tb corresponds to the tiine delay bet\veen both secondary peakR of the 2nd order Ricker
\vavclct. There is consequently a significant difference between the fundan1ental period

- ,BTl d(3
(4.2)

[
cos R(3(1 + (3 2
)-
; cos (arctan(3)
1 4
2
tp and the wavelet breadth tb.

where R = w 0 r/c and T = w0 t with r the true radial distance, c the wave velocity and
t the time. R and T arc dimenHionles8 parameters called the nurncrical distance and
numerical time respectively.
The velocity ¥.ravclet defined in E·q.(4.2) is nuvv computed for various dirncnsionless
distancer-> R. The corresponding curves arc displayed in Fig.4.1:
• For R=5: the V>'avelet ir:; strongly asymmetrical \vith a very sharp positive peak
follovired by a sn1oother negative peak. This shape fits well the main features of
near field scisn1ic \vaves in anelastic 1nedia.
• For R=7.5: the wavelet is still asyrnrnctrical but a bit smoother than in the previous
case. 1'he shape variations from R = 5 to R = 7.5 arc related to the properties of
seismic vvaves in anelastic media.
The interest of the 2nd order Ricker \Vavelet is that it is v,rell localiLicd in both ti1nc
and frequency don1ain (sec appendix B). It is thus easy to control the signal duration
and its frequency content.
Double-1\IJ wavelet
Other types of synthetic Hignals are considered in the field of train induced vibration.
Such a synthetic signal may be used either for niunerical simulations or cxpcrirnental
models. For instance: the double-1\!f wavelet characterizing the excitation induced by the
boggie is depicted in Fig.4.3 (Al Shaer et al., 2008). I\ is defined as a combination of two
gaussians (Ricker signalH at order zero):
(4.5)
244 11odelling vvave propagation 4.1 N utncrical rnethorl.s for -,,vavc propagation 245

0.008

0.007

[order 0 J w 0.006
u
;,
'8 0.005
en
en
~
w
'1 0.004
0
·~
en
'1 0.003
w
s
·~

"" 0.002

0.001

0.
0. 0.02 0.04 0.06 0.08 0.10
[ order i J
time (s)

Fig. 4.3: Time variations of the double-M wavelet (dimensionless load F/Q)
considered to model dynamic loadings by trains (Al-Shaer et al., 2008).

Q denotes the load supported by the axle, v the train speed) t the tirne, d the distance
betvi.reen tv.ro sleepcrs 1 a a critical distance, L the distance bct1,veen two axles of a boggie,
X and Y are din1cnsionless variables depending on the elasticity inodulus of the soil.

Such wavelets have been identified from field n1casurernents (under and along tracks)
and some typical values of X and Y are proposed in (Al Shaer et al., 2008).

Synthetic seismograms

In the case of seir-nnic excitations, the con1putation of realistic accelerogra1ns usually re-
quires a <letailed knowledge of the source, path 1 and site effects.

[order 2
time (s)
Fig. 4.2: Time variations of a oth (top), pt (middle) and
Ricker wavelet and definition of the wavelet breadth tb.
Son1e empirical rnodels 1.vere proposed by various authors (Pousse et al., 2006; Sa-
betta and Pugliese, 1996). The method proposed by Pousse et al. (2006) models the
J
tin1e-clomain accelerogram based on the assu1nption that the phase is random and that
the tirne envelope can be represented by a lognormaJ distribution for P and S-vvaveR
cornbincd \Vith an algebro-exponential function representing the envelope of coda waves.
In addition, the frequency content of the signal is nonstationary and follovvs a rnodified
2nd order (bottom) w-square n1odel.
The wave envelope proposed by Pousse et al. (2006) is displayed in Fig.4.4 sbowing
the various parts of the signal: P-v.,raves) S-wavcs and coda waves. This method dependti
on four comrnon indicators in earthquake engineering.
246 J'viodelling \Vave propagation :1.2 The Finite Element Method 247

1.2 The continuity of tractions at any interface also has to be satisfied:

P-wave S·wave Coda
(4.8)
1.0
~
00
where [.]:s denotes the jun1p across a surface of discontinuity 2=, Il is the vector normal
+"
·~
to 2= and I is the stress vector.
'1 0.8
"»
;.,
T,, As already shcn..vn in Chap.3, considering the Principle of Virtual VVork (Salenc_:;on,
"
H
+"
·~
0.6 2001) 1 the global equilibriurn rnay be written for any virtual velocity field fj as:
,.0
;., T,
- f g : dUl)dn + f t_.[Ldn + f T.d [Lda. = f pg.U dn (4.9)
"""
fl.<
0.4

0.2
"\vhere
Jn Jn
4is the strain rate tensor (see Chapter 3).
.loo Jn

It may also be expreHsed under an equivalent forn1 involving virtual displacements U

10.0 15. 20.0 25.0 instead of virtual velocities (Salern;on, 2001):
5.0
time (s)
- ( g: ~(fl.)dD + { [_fl.drl + { ;r_d fl.da = { pg.{J.dD (4.10)
Fig. 4.4: Envelope of the synthetic seismogram proposed by Pousse et al.
.Jn Jn Jan Jn
with~ the strain tensor (see Eq.(3.10)).
(2006).

4.2.2 Weak formulation

These indicators (peak ground acceleration, strong-motion duration, Arias intensity, For an isotropic linear ehu~tic mediun1 in srnall strains, the constitutive la1,ov is written as
and central frequency, see Appendix D) were calibrated by Pousse et al. (2006) using the classical Hooke's la"\v:
Japanese data recorded by the K-net array.
Q. = C: c; = >-(trc;):ll. + 2flc; (411)
or, in terrns of components:
4.2 The Finite Element Method
(412)
The rnain interest of the Finite Elerncnt 1\tfethod in the field of dynarnics is that it can be
used to rnodel cornplex gcon1etries and constitutive laws, t>trong heterogeneities (Hughes,
The constitutive law may be recast into Eq.(4.9) to obtain the weak forrn of the
1987; Cnrnier, 1993). Furthcr1nore, it is vvell-adapted to sirnulate dynarnic soil-structure
equilibrium of the medium (Le Ta.llcc, 2007; Salern:;on, 2001):
interaction (Clough and Penzien, 1 993).
- Jr:,,Cc;(J!.) : =~ : ~(fl.)dn + Jnl [_fl.dll + ;·on T.,,_fl.da
4.2.1 Strong formulation (4.13)

For a bounded <lornain 0 (see F'ig.3.1), the strong formulation corresponds to the analysis
- Jur p
82
8t
1'.1,dn
2 -
= 0

of the local equilibrium in an exact way such as (Salern;on, 2001):

for any di8place1nent field 'f2 E S, S being the space of all kine1natically adn1issible field8,
and with the following initial conditions on Ji.: J!.(I, 0) = uo(±) and fl.(±, 0) = vo(±)
V'.g+ [_=pg in 0 (4.6)
(IE rl).
q__.n_ = T_d on an (4.7)
Instead of searching for the exact solution of the problem, we rnay try to esthnatc
where -V' is the gradient vector (Appendix A), Q.
-
the Cauchy stress tensor, [_the body an approximate solution by mini1nizing the left scalar ter1n in Eq.(4.1~1), denoted W(il),
forces 1 g_ the acceleration vector and T_d the surface forces. in the space of kinen1atically ad1nissible fields Q. Equation (4.13) is called the ~neak
.formulat'ion of the problem.
248 1\/Iodclling \Vave propagation 4.2 The Finite Element Method 249

4.2.3 Approximate minimization: Galerkin method

The n1ain idea of the Galerkin approxirnation is to minimize the left term in Eq.( 4.13) not
in the space of all kinc.rnatica1ly admissible fields S but in a finite-dirncnsion sub-space
(Bonnet and Frangi, 2007; Le Tallec, 2007; Sa1eni;;on, 2001). The '1ppruxirnation thus
conRists in defining the sub-space Sh, for instance:
n
5h = {'!lh = L a.iN,i + ]Jd) ai E ~n} CS (4.14)
·i=l

where Qh n1ay be expressed from any given displaccrncnt ifrl satisfying the boundary
conditionR and n functions 1'-li explicitly given in the space of the displacerncnts being
z.ero ·on the boundarieR Vi.rhere the displaccn1cnt is prescribed. The n scalar pararnetcrs
ai should be identified by the follcJ\ving rninirnizing process:

. W('")
nun 1£ , (4.15)
jJ/'ESh

·T·his formulation is called the serni-discretized variational formulation since the tin1e Fig. 4.5: Discretization of domain fl by finite eleme11ts .\.le and description of
derivatives are still continuous and also have to be approxirnated considering a tirne nodal displacements (2D case).
discretization (Hughes, 1987; Ilecldy, 2004; Zienkiewicz and Taylor, 2005).

4.2.4 Finite elements It is also possible to choose quadratic or even higher order shape functions (H.ughes,
1987; R.eddy, 20().cl; Zienkie\vicz and Taylor 1 2005). _F'or a quadrilateral element, linear
Discretization of the domain
and quadratic shape functions are compared in Fig.4.6. The linear and quatlratic shape
Through the Galerkin method, the space of all kine1natically achnissible displa.cement functions are defined as follows:
fields Sis approxirnate<l by the sub-space Sh. As sho\vn in Fig.4.5 (2D case), a convenient
vvay to do so consists in approxirnating the exact dornain Sl by a set of triangular or • Linear quadrilateral element (4 nodes):
quadrilateral areas .~lf'. (spatial discretization) such as:
Ni(r, s) = Hl - r)(l - s)
N2(r, 8) = ;j(l + r)(l - s)
(4.16) (4.19)
:l(l + r)(l + .s)
Such srnall geornetrical areas are called finite elements and their average dimension \Vill
be denoted .6.h in the follu\ving. In the 3D case, 3D elements rnay be considered: tetra-
1 N3(r, s)
N4(r, s)
=
= ,l(l - r)(H- s)

hedra; .pentahedra or hexahedra for instance (Bathe, 1995; IIughcs, 1987). • Quadratic quad1ilatcral clement (8 nodes):

Ni (r, 8) = t (1 - r)(l - s) (-l - r - s)

Various types of finite elements N 2 (r, s) = ,l(l + r)(l - s)(-1 + r - s)
Various interpolation orders. As shown in Fig ..4.5, we generally consider the nodal
N 3 (r, s) = .l (1 + r) (1 + s) (-1 + r + s)
diHplacerncnts to define the displacen1ent field in a given triangular finite element: N4(r, s) = .l(l - r)(l + s)(-1 - r + .s)
(4.20)
Ns(r, s) = %(1- r 2 )(1 - s)
(4.17)
N5(r, s) = i(l - s 2 )(1 + r)
The shape functions rnay be linear, that is in t'°'ro dirncnsions: N1(r, s) = ±(l - r 2 )(1 + s)
N, (.r,, .x2) = a, + b1J;1 + c1x2 l = i, j, k (4.18) N 8 (r, s) = .l(l - s 2 )(l - r)
 T
250 .Nfodelling vvave propagation 4.2 The Finite Element Method 251

As shown in Fig.4.6, each shape function i8 zero at all nodes but one. Because of
the higher degree of the shape functions in the quadratic case (bot.torn), it is possible to
retrieve faster variations in a quadratic clement than in a linear ele1ncnt. 'fhis rernark
vvill be particularly irnportant .for the Hirnulation of \Vave propagation.
Reference finite element. In addition to the interpolation of the displace1nent field
vvithin each finite element, it is convenient to consider a reference elcn1cnt 1 fixed nor-
mali:zcd geonretry, to perforrn the basic calculations alVirays the sarne way (I3athc 1 1995;
Hughes, 1987). As depicted in f~ig.4.7, the quadrilateral reference element is a square
and any quadrilateral geornctry 1nay be defined by a sin1plc transforn1ation bctv,,1een the
actual elements and the reference one. The Jacobian matrix of this transforrnation must
be calculated since the element integrals have to be transforrned the sarne way. For
each finite clcment 1 it will thus be necessary to interpolate both the displacernent field
and the geornetry. An easy 1vay to perform both interpolations is the consider the same
shape functions: such finite elerncnts arc called isoparametric elements (Reddy, 2004;
Zienkiewicz and Taylor, 2005).

s
s=+l

---:--···
r=+l
-1

-0.5 r
r=-1 -·-·;·----------·- ------·-·-·-\·-·-·-·-··--

1----=~+_j_J
___1-
s=-1 2

s +1
(-1 ,1) 4 3 (1, 1)

-1 +1
r

(-1 ,-1) 1 -1 2 1,-1

Fig. 4.6: Shape functions for a linear (top) and a quadratic (bottom) quadri- Fig. 4. 7: Intrinsic coordinates for quadrilateral finite elements: distorted
lateral fir1ite element. element (top) defined from the reference element (bottom).
 1
I

252 lVIodelling wave propagation 4.2 The Finite Elemen\ Method 25:1

the n1ass (clistribnted inass) leads to a consistent inass matrix (.Hughes, 1987; Reddy,
200L1). Alternative for1ns for thiH rnatrix \\ri.1.1 Le discussed aftcrvvards in the fran1cvvork
of elastodyna1nics.
linear
4.2.5 Time integration algorithms
The semi-discretized variational for1nulation only concerns the spatial discreti7'ation. It is
also necessary to consider a tin1e discreti,,;ation to approxin1atc the tirne derivatives. \i\Te
shall thus con8ider ·m.+l discrete tin1cs tn such as: Ln = nD..t, n = 0 1 m,. 'The ti1nc step D..t
is defined as f:l[ = lmax/rn where tnw.:r: is the 111axirnum ti1ne of the sirnula.tion. Starting
fTon1 the initial conditions, ({Uo}, {U-o} ) 1 U1nc-integration algorithn1s then proceed to the
co1nputations of the approximate solutions at each time step, narnely ({Un}) {Un}) {Un}).
quadratic • Various ti1ne integration schen1es arc available: explicit) irnplicit 1 ruultiHtep algorithrn)
and sub-stepping methods, etc (I-Iughes, 1987; Zienkie-\vicz and 1'aylor) 2005). Vle v;,rill
only discuss son1e of the classical time integration schenres and their propertieH (advan-
tages and draV1.rbacks) in the field of dyna1nics and \Vave propagation.

[p=3) [p=4] The central difference method

• • • The central difference inethod is a sirnple explicit tirne integration scheme (Bathe, 1995;
higher R,eddy, 2004). The velocity and the acceleration arc expret->Hed under a sin1ple for1n:
• order • • •
. 1
• • • {U,,+i} =
2
/::,./ ({Un+I} - {U,,_,})
-
,. 1 (4 22)
{Un}= /::,.t 2 ( {Un+I} - 2{Un} + {Un-1})
Fig. 4.8: Various types of finite elements at different orders p.
It thus yields a classical second order approximation of the acceleration vector. If we con-
::iider a lumped nrass 1natrix (see §4.3), the expression of the central di:ffercnce algorithrn
Different types of triangular and quadrilateral isopararnctric clernents) at various or- is fully explicit. 1\1evertheleas 1 Vile shall see later that this algorithm is only conditionally
dern p (Hughes, 1987), are depicted in Fig.4.8. Linear (top) and quadrntic (middle) finite st'1ble.
clernents are first shown for both the triangular and the quadrilateral ca.set:;. Higher order
ele1nents are also described in Fig.4.8: a 3rd order triangular elenrent (bottom left) and The Newmark method
a 4th order quadrilateral elen1ent (bottorn right). The accuracy of such finite ele1nents 'I'he J\Tewrnark method is a second-order accurate approach \videly used in the field of
1vill be considered in the folloVl.'ing in the case of \vave propagation sirnulations. dynan1ics (Hnghes 1 1987; Zienkievvicz and Taylor, 2005). It is a t\.vo-parameters algorithrn
( 10, -y) defined as follows:

Eq11ation of motion [M]{Un-1d + [C]{Un+d + [K]{Un+l} = {F,,+d

The estinration of all integrals in each finite elenrent (Bathe 1 1995; Ilcddy, 2004) yields {Un+l} = {Un} + L'.t {U,,} + "';
2
[ (1 - 2{J){Un} + 2;3{ Un+d l (4.23)
the equation of inotion in the following n1atrix forrn:
{[ln+d ={Un}+ f'.t [ (1 - -y){U,,} + "J{Un+l}]
[M]{U(t)} + [K]{U(t)} = {F(t)} (4.21)
The stability conditions of the algorithrn are the following:
1vhcre [J(] is the stiffness rnatrix deriving fron1 the cla:::>tic energy) [_)\,f} the mass 1natrix • unconditional stability:
derived fro1n the kinetic energy, {U} and {U} the vectors of unkncnvn displacements and 1
-2 <
- -y <
- 2{:J (4 24)
accelerations (respectively) and {111} is the vector of nodal forces. The classical form of
254 ]\
1
Iodelling vvave propagation 4.2 The Finite Element Method 255

• conditional stability: A,

(4.25)
'\ supers ability I
\ I
¥there wh is the approximated circular frequency for finite elen1ents of average size \ p 1 I
\ I
f:j,,h and rlcrit is the critical frequency, such as: \ I
\ I
\ I
\ I

(4.26)
\
\ I
I A,
\
\ I
\ I
\ I
\Vhere ~ is the darnping ratio.
limit of/',, sta: ility ,1'
stability \ ,' instability
Classical choices for the algorithm parameters are:
• average acceleration method: the para1neters are (3 = %- and 'Y = ~. The algorithn1 p=l \ ,' p>l
\ I
\ I
is thus in1plicit and is unconditionally stable. \ I
\ I
• linear acceleration n1ethod: the pa.ra1neters are .6 = and 1 i = ~. The algorith1n I

is thus implicit and is conditionally stable with Dcrit = 2yl3.

• central difference rnethod: chosing /3 = 0 and 'Y = ~ in the Nc>N1nark scheme leads
to the central difference method. The algorithrn is thus explicit and is conditionally
stable with rlcrit = 2.
For other values of the pararnctcrs (3 and "y', the stability conditions arc depicted in
Fig.4.9 with respect to A 1 and A, defined as (Curnicr, 1993; Hughes, 1987):
Fig. 4.9: Stability conditions of the Newmark time-integratior1 scheme
( Curnier, 1993; Hughes, 1987).
\Vhich corresponds to the Ne1,v1nark algorithn1 for et=O.

[ED+ ';° (1 + ~)] This 1nethod is often called the o:-HHT rnethod. It is unconditionally stable when
A 1 =1-~--~-=~ (4.27)
-~ < a < 0, "! = 1 - 220 and /J = (l~n) . Its tnain intere.st is that it can easily suppress
2

1 + 21En + 13n 2
spurious high frequency components (using algorithm,ic damping).
A2 = 1-
[2~n + n 2 ( 1 - ~)]
~---~-~~ (4.28)
1 + 21ED + ,rm 2 Another method of this type was proposed by Wood ct al. (1981). In this method,
with D = c/'[J.t. only the inertial tern1 is modified yielding:

Outside the triangle b < 1/2 and 210 - 1 < 0), the Newmark scheme is unsta·
hie (Curnier, 1993). In the triangle, it may be unconditionally stable b ::0 1/2 and
2/3 -1 ::0 0) or conditionally stable b ::0 1 /2 and 2.8 - 1 < 0). The corresponding ranges
for the spectral radius p (Curnier, 1993; R.eddy, 2004) are also given in Fig.4.9.
Multistep methods
A1nong n1nltistep rnethod8, the a:- method proposed by Hilber; Hughes and Taylor (I-Iughes)
1987) has the following form:
(4 30)
This method is unconditionally stable for aa :S ~' (3 2: ~ 2 :t, o:B + 'Y 2: ~·
It is also possible to consider higher order time-integration schernes using for instance
sub-stepping procedures. The accuracy of computed dynan1ic responses appears to be
i1nproved by such proceduTes (Fung, 1997). Various other tirne integration schemes have
been proposed in the field of dynamics: \i\Tilson, Park, Houbolt, etc (Bathe, 1995; Hughes,
1987; Zienkiewicz and Taylor, 2005).

[M]{Un+d + (1 + cv)[C]{Un+d - cv[C]{Un} 4.2.6 Spectral elements

+(1 + ct)[K]{Un+1} - cv[K]{Un} = {F(tn+a)} (4.29) The spectral ele1nent method uses the flexibility of the finite elen1ent n1ethod and the
high accuracy of spectral methods (Patera, 1984; Faccioli et al., 1996, 1997; Kornatit.':lch II

II
256 J'vfodelling vvave propagation 4.3 Numerical dispersion 257

and Vilotte) 1996, 1998). Spectral clcincnts are flub-parametric since they involve lower
order intcrpola.nts for the gcotuetry and higher order intcrpolants fur the field variable
(J\.1eza-Fajarclo and Papagcorgiou 1 2008). The higher or<lcr interpolants may he chosen
as Legendre polyno1nlal (J\1Ieza-Fajardo and Pa.pagcorgiou, 2'008) lea.ding to integration
points defined 1 in tl1e lD case, aR the roots of:

(4.31)

vvhcrc Pj_v(x) is the Legendre polynon1ial at order N.

They rnay Le also chosen as Chcbyshcv functions (Dauksher and En1ery, 1999) leading
to integration points defined as:

Xj = cosGj with: e:; = ;_; (j = 0, ... ,N) (4.32)

In the 2D or 3D cases, a tensorial interpolation operator should be introduced (Chaljub

et al., 2003, Komatitsch and Vilotte, 1998). -0.4~~~~~--,..L_ __,.L_ _i_~-~--L__i____
-1.0 -0.8 -0.6 -OA -0.2 0.0 0.2 0.4 0.6 0.8 1.0
_i\_s sho-1,vn in Fig.4.10 at order 4, such higher order interpolants yield a non regular
Fig. 4.11: Lagrange polynomials related to the Gauss-Lobatto-Legendre
node spacing in the spectral elernents (right) \vhen cornpared to the finite elernents (left). points at order 8 (Kornatitsch, 1997).
Considering the srnallcst node-to-node distance, it rnay thuR lead to smaller tirne step
than for a regular spacing (Delavand, 2007). Nevertheless, the accuracy of spectral el-
ements is excellent for vva:ve propagation sin1ulations, especially in terrns of num,erical
dispersion. 4.3 Numerical dispersion
rrhe spectral cle1nent inethod often involves higher order interpolants up to order 8 4.3.l Physical dispersion and attenuation
(Faccioli et al., 1996, 1997: Komatitsd1 and Vilotte, 1996, 1998). To illustrate the way
As discussed in Chapter 2, in a linear viscoelastic mediu1n, the one-dirnensional \Vavc
the field variables are interpolated in Ruch cases, the Lagrange polynornials related to the
equation rnay be derived in the frequency dornain under the following forrn:
Gauss-Lobatto-Legenclre points at order 8 are displayed in Fig.4.10.
i) 2 ?J, ' ( X 1 UJ·) pW *
iJx2 -1- JW(w) u (x,w) = 0 (4 33)
(a) -- (b) -
II
• • • • • • vvhcre u* is the Fourier transform of the <lisplace1nent, x the distance, w the circular
frequency, p the mass density and M(w) the complex modulus (Aki and Richards, 1980;
. • • • •• • •
13ourhie el al., 1987) .

The problem solution may thus take the form (Semhlat and Luong, 1998):
• • • •• • •
- - - u'(x,w) = Uo(w)exp[-ik"(w)x)] (4 34)

Fig. 4.10: Fourth order finite (left) and spectral (right) elements. 1,vherc k* (w) is the complex wavenurnber such as:

k"(w) = (w) - ia(w) (435)

cw
258 J'viodelling 1.vavc propagation 4.3 Nu1nerical dispersion 259

with c(w) is the frequency dependent wave velocity and u(w) the attenuation factor.
In addition to this first classical term of phase shift, thiti con1plex wavenun1bcr also
involves an irnaginary ter1n related to darnping. Both ter1ns (phase shift 1 damping) gen-
erally depend on frequency. The dependency between the phase velocity c( w) and the
frequency is related to the material dispersion in the n1ediu1n.
From a numerical point of vie\~T, both properties have their counterparts which arc
generally called numerical dispersion and nurnerical dam,ping (Hughes, 1987; Semblat,
1998; Stewart and I--Iughes, 1998). ::\urncrical dispersion nrakes the "\\'ave velocity depend
on the model features (time integration scheme 1 elen1cnt size, elernent type, etc). J\Tu-
merical da1nping corresponds to the sarne type of dependency for the "\vave a1nplitutle.
Both numerical errors are diRcussed in the following.
4.3.2 Numerical errors for wave propagation
The relative period error is different from one time-integration schernc to another
(Fig.4.13). The u-HHT method, Eq.(4.29), leads to some of the lowest period errors
(a=-0.3 and -0.05). For 1.vavc propagation prublen1s 1 the relative period error appears
through the estimation of .the IO.rave velocity. The propagation of a wave in a given nu-
rnerical schen1c depends on the clcrnent size, the integration scheme, the element type,
etc. This phcnornenon is called nurnerical dispersion with reference to physical disper-
sion (Eq.(4.35)) making the wave velocity dependent on frequency (Hughes et al., 2008;
Ihlenburg and Babuska, 1995; Semblat and Luong, 1998; Semblat and I3rioist, 2000).
It may thus be difficult to sirnulate wave propagation phenon1ena because in the fi-
nite difference or the finite elernent methods the rnunerical error may increase during the
propagation of the wave. As shown in Fig.4.6, higher order interpolants n1ay yield rnore
accurate solutions in the fra.rnevvork of the finite elen1ent or spectral elei:nent method.
The comparison of lower order and higher order finite elements vvith respect to nnn1eri-
cal dispersion will be discussed in the following.

In the field of structural <lynanrics, two types of numerical error occur (Hughes 1 1987). As
depicted in Fig.4.12, the relati'ue period error is linked to the esti1nation of the funda1nen-
tal period of tl1e structure. Algorithrn'ic damping correspondR to an artificial arnplitudc
reduction due to a purely numerical dan1ping. It is possible to analyze and quantify such
nrnnerical errors.

theoretical
x---)(---x---x---- numerical

'•
:----~:-::_-_r2ns
: ,' - : ""k,
: .L!' I '.
1':r : 'l./:

'I
,( I ',
'I I ' a=-0.05
0 ~'
l T
/ T-T0
/'
/
"'' 0.2
MIT
0.4

Fig, 4.12: Relative period error and algorithmic damping as defined by Fig. 4.13: Relative period error for various time integration schemes (Hughes,
Hughes (1987), 1987),
 T

260 ?-,ifodelling \.Vave propagation 4..:) Numerical di8per8ion 261

2.0.---~---~---~--~---~---
4.3.3 Theoretical numerical dispersion
One-dimensional analysis 1.5

The numerical solution of Eq.(4.33) n1ay take a form sirnilar to that of the theoretical
10
solution (Eq.4.34):
uh(,:,w) = Uu(w)exp(-ik 1'(w)x) (4.36)
0.5
vvhcre uh and kh are the approxirnatc displacement and wavernnnber respectively.
0
Several theoretical v,rorks concern the analvsis of the error in the 11u1ncri.cal estimation
of the approxin1atc vvavcnun1ber kh -,,vhcn cor~pa.re<l to the exact/theoretical vvavcnurnber -0.5
k (Ba.mherger et al., 1980; Hu et al., 1998; Hughes et al., 20118; Ihlenhnrg & Babuska,
1995). In the case of linear finite elements, Ihlcnbnrg and Bahu,ka (1995) proposed the -1.0
following relation:
l -K' - -1.5
coskh6.h= _ ; 2 (437)
l+ri
-2.0:------:----~--~-_[_-~---~--_J
where J{ is the norrnalized frequency such as K = k6.h = w6.h/ c. 0 2 3 4 5 6

normalized frequency K
Expression (4.37) shows that the numerical solution of Eq.(4.33) is related to propa-
gation pheno1nena only for normalized frcquencieH belo-,,v a given cut-off frequency Ko, Fig. 4.14: Theoretical error on the approximate wavenumber kh for linear
Fig.4.14 (Ihlenburg & I3abuSka, 1995). I-10-1,orcver, for such frequency values, the nurncrical finite elements and cut-off frequency Ko (Ihlenhurg and Babuska, 1995).
vvave propagates faster or slower than the theoretical solution. It is thus necessary to
analyze the nu1ncrical diHperHion of waves and to quantify the related nun1crical error.

For higher order elerncnts, the error on the wavenurnber is derived by Ihlenburg and wave type; the angle of incidence 1 the type and si~e of finite elements (triangleR 1 quadri-
Babuska (1995), if k6.h/p < 1, as follows: 1.aterals, etc). Some dispersion relations are proposed for various nurnerical schcn1cs by
Bambe1ger et a.l. (1980) or Ihlenburg and Babuska (1995). The analytical expressiorn
(~) 2P(k:6.h) "
2

lkh-kl= kC1 (4.38) of the phase and group velocitieR of the nurnerical waves arc calculated: for various 1vave
4y0!P 4 2p types (Ban1bcrger et aL) 1980). Sonic of them are discussed in the follovving.
where Cr is a constant 1 p the interpol.ant order, 6.h the elerncut size.
Considering linear finite clc1nent) the din1cnsionleHR nun1erical pha8e velocities for P
and S waves are found to be (Bamberger et al., 1980):
This analytical result shoVilS that the decrenBe of the phase difference iH even stronger
for larger p. Hughes et al. (2008) recently proposed a unified theoretical analysis com- 2
paring higher order classical finite elernents and NURI3S 1 approxirnations. Vj; = /3 . arccos [1- !::,.t 26 ( (l - {J1f3 2) + (1- /31)(1-/32)]
2rr H 6.t( 2(/31 + 2) (/32 + 2)
2 (4.39)
In the frarr1ework of the classical Finite Elerr1cnt JVIethod 1 the co1nparison of inter-
Vs¢ -- /3 arccos
[1
.- u
At2 7 - 5,81/32 - {J, - /32 - 6( (1 - (3i{J2)]
polants of various orders will he considered in the following through nuruerical exarnples 2KH6.t~ 2(/31+2)((h+2)
involving \i\rave propagation in soils.
where: /31 = cos(27rH cos@); /32 = cos(27rHsinB); Bis the incidence angle of the numer-
Two-dimensional analysis ical wave; H = 6.h/ A (with 6.h the element size and A the wavelength)·' 6.t = M 2"t·~
(6.t;
AHscssing numerical wave dispen;;ion is rather simple in the one-dirnensional case. In two
being the tin1e step considered in the co1nputation· 6.t* =
'
2"h
y1vz+vz'
Vp and
,.,
the vvavc v,,
I' s
dimensions, such phenomena are more difficult to analyze. As dernonstrated by Ba1n- 2 2
velocities of P and S-wave respectively); ( (v) = V'v+ffiv' = ;::: " (thus ii:<; ( :<; 1).
4V
berger et al. (1980), it is necessary to take into account the properties of the n1ediu1n 1 the F S <l a

1 non-uni.forn1 rational I3-splines.

262 1viodelling Vilave propagation 4.3 Numerical dispersion 263

1'hese relations fully characterize 2D nurr1erical dispersion and lead to sorne o:f the 1.5 ,--------f,_Q~-::-:-:---:;,__ _ _ _ _ __
curves displayed in Fig.4.15. These curves give the dimensionless phase velocities of P uadrilateral , ,
and S-waves for a tv.ro-dirnensional rnesh. The dirnensi.onle8S phase velocity is the ratio ,,
betv.recn the velocit:y of the nun1erica.l waves and the theoretical phase velocity. Fig-
1.4 P-wave 1.4 S-wave
,,,
ure 4.15 corresponds to quadrilateral (top) and triangular (middle and bottom) finite 1.3 1.3
--8~0°
,,
----- 8~15°
elements having a linear interpolation. These reHults allo\v the quantification of the nu-
rr1erical error as a function of the element size to -,,vavclcngth ratio for various values of ---·········· 8~30°
~,;;;:;~-:_.~:~:~·-.,
1.2 1.2 ,/',;:'/ '\.
the angle of incidence. -------- 8~45°

1.1 1.1
All the results displayed in Fig.4.15 arc given for a Poisson's ratio v=0.25 and a tirnc-
stcp ratio .6..t=l. Frorn Fig.4.15, it is possible to rr1akc several re1narks for such various
1.0 1.0
configurations:
• the nun1crical dispersion (and thus the nurncrical error) is stronger for a larger
cle1nent size (i.e., larger element size to wavelength ratios); 1.0 r----,,,.·.-,_,,,=,;::7:0::----------------------· 1.0

• the error is generally maxirnurn for a norrnal incidence (8 = 0) and minirnurn :for v '~~-:·~:-··-···-... ________ _

a 45° incidence (fur this specific numerical scheme) except for triangular elerncnts ·s
0
o.9
''
0.9
e P-wave ''
when < O;
~ 8>0
''
• for quadrilatera.l clements (top) and small values of the elerr1ent size, P-vvaves are "'
w 0.8 '' 0.8
C\l ''
rnore sensitive to the angle of incidence than S-waves. Whereas, :for triangular ~
elernents (middle and bottom), S-v.raves are strongly dispersed Vlrith either larger "" 0.7 0.7
or smaller -,,vavc velocities than the theoretical waves.

For triangular elerr1cnts (right angled triangles) 1 the influence of incidence is different 0.6 Triangular
for positive or negative angles (there is no symmetry). In the case of P-\vaves, there
is no n11rnerical dispersion for a 45° incidence (constant phase velocity). 'fhe angles 1.0 S-wave I

of incidence leading to the lowest numerical dispersion are different for P and S-\vaves 8<0
...,
>, 1.2

- -_·- _- -_=~-~ -'-~ -·_,_._'_·___,·~-·.: _

(Bamberger et al., 1980; Semblat, 1998). ·~
'-' 0.9
0 II
~
....--
,,,..$:::=:-::------- ii
Several other theoretical studies have been made on the nun1crical dispersion of V.'aves "':> 0.8
1.0 - -- I
an1ong which: "'
w
C\l S-wave
~
• Joly (1982) studied numerical R'1yleigh waves and demonstrated that their nu- 8>0
1ncrical dispersion is slightly larger for low order finite elements than with finite "" 0.7 0.8

differences. Nevertheless, the Finite Difference Niethod rnay generate artificial sur-
face wave components. 0.6 ~-_L__ _L___ _J__ _L__ _ 0.6 L.__ _L__ _L___ _J__ _L__ _

0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
• Cohen et a.l. (1994) performed extensive ana.lyseH on the accuracy of higher order element size/wavelength ratio
finite clements considering rnass condensation.

• l{im et al. (1999) proposed numerical models to sirnulate cross-hole tests in isotropic Fig. 4.15: Numerical dispersion for a 2D numerical scheme as estimated by
and anisotropic soils. They studied the influence of numerical dispersion on the Bamberger et al. (1980), Eq.(4.39), for v = 0.25 and /}.t = 1: P-waves (left) and
arrival-tirne esti111ation. S-waves (right), quadrilateral elements (top) and triangular elements (middle
and bottom).
26'1 f..1lodelling vvave propagation 4.3 Nun1erical dispersion 265

4.3.4 Time-step estimates for some simple cases strongly influences the nurncrical error. The coarse n1eshes lead to rnunerical reRults
under-estirnating the an1plitudeR but overestilnating the velocitieH (group and phase).
A8 .sho\vn by Hnghe8 (1987) for a plane pressure wave in a bar 1 the critical time step for It illustrate.s the practical meaning of numerical dispersion vvhich may be li1nited by
the central-di:ffcrence method, that is Newrn_ark n1cthod with fJ = 0 and J' =~'is: chooRing an elen1ent size adapted to the vvavelength of the proble1n. The clernent size iR
2 flh classically chosen as one tenth or one twentieth of the v,ravelengih. H(nvever, as sho\Vll in
flt < - - = - for a lumped mass rnatrix (4.40) Fig.4.17 in cases 3 (200 elements, flh./A=l/10) and 5 (400 elements, flh./A=l/20), the
- wfhr1:1: C
nun1erical error becomes significant beyond a propagation distance of approxirnatcly SA
flt < flh or lOA (respectively).
for a consistent ma,ss rnatrix (4.41)
- v'3c
1vhere c is the pressure wave velocity in the bar: c = V£7P.
ThuB 1 consistent n1ass matrices tend to yield smaller critical tirnc steps than lurnpcd
n1ass n1atrices.

For a rectangular 2D element (dirnensions b.h1 and b.h2)i Flanagan and Belytschko
(1981) have Rhovvn that, for the central-difference method, the critical tin1e-step iR:
Fig. 4.16: Finite element mesh and Ricker wavelet considered for the analysis
(442) of lD numerical dispersion.

1vhere Vp is the P-\vave velocity.

4.3.6 Influence of the geometrical arrangement
'l~he
choice for the critical time-step depends on the tin1e-integration scheme as Vi'ell as
on the element Rize antl type (H't1gheR) 1987). In the following paragraphs, the influence Since \Ve shall analyse later the accuracy of different finite element orders, v,re only con-
of the latter on nun1erical dispersion vvill be investigated. sider t\vo types of 2D spatial arrangements. The influence on nu1nerical dispersion of
these 2D elerncnt patterns) depicted in Fig.4.18 1 v.rill nov,r be investigated. The first one
iH a regular Ret of triangles called nhalf-square'1 pattern (a). The second one combineR all
4.3.5 Numerical dispersion for low order elements possible orientations to get a n1ore isotropic pattern called nt'lquare-diamond 1' (b). In the
To assess the numerical error for \Vave propagation prohlerns, we firstly consider a one- follo\ving, vve shall analyse the efficiency of both 2D patternR vvith reHpect to nun1erical
dimensional case ':.vith a linear elastic isotropic rnc<liu1n (no physical diRperRion). As dispersion for a particular problern (seisrnic waveR originated by a vibrating foundation).
depicted in Fig.4.16, a 2nd orrlcr Ricker wavelet (Eq.(B.3)) propagates through a lD
rnodel of variable refine1nent. rrable 4.1 gives the number of elements for each model and To assesR the infiuence of the Rpatial arrangement, one considers the particular prob-
the corresponding elc1nent size !:::.h. In the last line, the !:::.h/ 1\ ratio characterizes the lc1n of seisrnic waves generated by foundation vibration and propagating in a centrifuged
elernent size normalized to the vvavelength A. soil rnasH. The complete description of the experirnents i.s given in (Chazelas et al., 2001).
Figure 4.19 shows the foundation above the soil maHs and displays the force vs tirne curve
considered herein. The tirne step is !:::.t=(l.05s and the time integration algorith1n is a
Newmark non di'"ipative one (,CJ= 1/4, "( = 1/2). The clement size is flh=O.Olm. The
Table 4,1: Mesh refinement and element size to wavelength ratio for the
rnechanical properties of the soil are the following: rnass density p=l600kg/rn3 ) Young's
various lD cases studied.
ir1odulus E=l001'v1Pa and Poisson:s ratio v=0.24. The nurnerical analysiR is perfor1ned
Nurnber of elen1ents 50 100 200 :JOO 400 800 with the FEM/BEM corlc CESAR-LCPC (Humbert et a.l., 2005).
Element size !:::.h 2m lm 0.5 m 0.333 m 0.25 ID 0.125 m
flh/A ratio 2/5 1/5 1/10 1/15 1/20 1/40 1'hc two ele1ncnt patterns depicted in Fig.4.18 are co1npared with respect to nurnerical
dispersion considering the propagation problem of F'ig.4.19. The hori:.:-;ontal displacernent
along the Roil surface is displayed in F'ig.4.20. The square-diarnond pattern (SD) appearR
The Vi'aveforms cornputcd at different tirncs are diRplaycd in Fig.4.17 for each nu- to be 1nore accurate than the half-square pattern (HS) Mince the numerical dispersion is
rnerical n1odel detailed in Table 4.1. These curves clearly shov,r that the cle1nent size stronger with the latter (velocity increase).
266 J'vio<lclling vva.ve propagation 4.:1 Nurnerieal dit'lpersion 267

0.5 ' (b)

oaf11iA v

-0.5
0.5
l=:::±:====±:::::!===±====±:::::!======::::;
i ! /1 t\ 100 elements
/\ ' II • ' ' ·"·
0.0 E/1 -f-.-l.- -___".".".:'_\-~----l----\---_t-\__.,.....;:.:'.".'.\.: .... 1.....\.. l~~-1::;-;,,..~----i- \ ' \_:/·-'-;~/\-~-/-'-,v-----
\ I \/ ~.\ ../ \/ i... !'·.,: \.: -
:" Fig. 4.18: Types of 2D element pattern considered half-square (a) and
-0.5
0.5 l=::±====±::o===:±====b======::::; square-diamond (b)-
!r ~ {\ 200 elements

o.otf (\__ /\ • I\
--·-- --··----·-·------\TT---\.-- __/\ : \
\~ r· \/-----~· ..- -..\J , --~---:--.,-·-------···

·:
j_ F(t)
F(t)
\I '' \.)
-0.5:!==::!!::::!::=::!==:!::=::!~V=::!==::':::=::!==::':t\,='::':::=::!==::':::=::!:j::::::':::=::!==::':::=::!==::':::=::!:::::::
0.5 E
<D
300 elements
/' 'r
! \ ,, .: • ~ "'
o.otf {\_ ------~---·\t·:----'-
I I

------- - - --~' --\-f--;---\_____ .<~-\-/--;

: ,

--.,------ 0.60m
0
time (ms)
0 0.05 0.10 0.15 0.20 0.25

-0 .s :!==::!!::::!::=::!==:!::=::!\:!':U,=:::t==:!::=:::t=='"\:;:j/=:::t==::':::=::!=='"'±t·=:::':::=:::t==:!::=:::t==::':::=::!:::::::
0.5 Fig. 4.19: Model for the analysis of waves generated by foundation vibration
:r ,.,, ". 400 elements in the centrifuge (Chazelas et al., 2001).
ti {\_ - _____ _.;____Tt·r--->-- --- ---------~----\·!-: --~----
o. o ----- .....
f \ : \ I I : I \ -:'.~\ : __
~i
f___\ _____ _
I
\:/ \!/ ~:: 2e-07
I
-0.5:!=:::'l~<:::::::!:::='==:!'~'=:::t==!:::::!:::='±''='===:!::::::!:==~~~:::!::=':==::'::::::!:==<:::::::!::::::; ~

E 1e-07
0.5

LI'
0.0 ..... {L
......
I \ :'
-"'·1 1.'1
r
\
- ... _,
I\ : 1'
I \ : / \ '
' ', ,_ ··········--· ,,,_.
····--~r(·----
" "
~ i :\
: :: ___ " .
800 elements
-"s"'
~

0 .-- . .-,
\. I

I
\,
I. I '•' ": ""'
-a"'
-1e-07
-0.5+----+"-+-~~+-~l~'!/~~+-~~+\~!f~~+-~~+:~:!:!~~+-~~+-~-i>E~SAR-~_, ~
" I
'i3 -2e-07
0.0 5A 10A 15A 20A
"\ i
distance foumber of wavelengths)

Fig. 4.17: Analysis of the one-dimensional numerical dispersion (waveforms

OJ
1:1
0

0
..0
-3e-07
-~ -4e-07 \.
···'
/
/ ...l
SD pattern

HS pattern
I
at different times) for wave propagation problems computed by the FEM 'I
-5e-07
(lower order elements). 0 0.05 0.1 0.15 0.2 0.25 0.3 11

distance (m)

Fig. 4.20: Horizontal displacement along the surface at time t=O. 75ms : com-
parison of both element patterns (SD: square-diamond, HS: half-square).
T·he SD pattern being rr111ch rnorc 11 isotropic 11 than the IIS pattern 1 this result is
logical. In the follo\ving 1 \Ve shall conHider the square-diamond pattern to investigate the
nuinerical accuracy for diJierent finite element orders.
268 J'vlodelliug Viravc propagation 4.3 .f\Turnerical disper,':-lion 269

4.3.7 Influence of mass matrix formulation 0.3

The nu1ncrical dispersion also depends on the mass matrix forrnulation. As shovvn in 0.2
Fig.4.21, the numerical error it:> different for a consistent or a lumped inat->H matrix for-
1nulation (Hughes, 1987). Tn the previous one-dimenHional exarnple (Fig.4.17), the mass 0.1
,, "
,i
1natrix formulation is a consistent one. rro limit the numerical dispersion, it 1nay be ,,
interesting to con.sider alternative rnass rnatrix forn1ulations. T-,,vo formulations will be
~
'1 0 - ,,,
,. ............... .

novv l'.Ornpared: a lurnped forrnulation and a rnixcd consistent/lumped forrnulation. s'" ,:

,'/ j
'"
~ -0.1 ,, '
Tlrc efficiency of both for1nulations in tcr1ns of numerical dispersion 1,1;.rill novv be as- 0.
w I:
/ / _i
I
sessed. To derive a lurnpcd inass matrix, a simple n1ethod consists in replacing the ~ -0.2 /fj reference curve
I: I
diagonal term by the sum of the terrns of the line and the non diagonal terms by zero / ! _i consistent mass
-0.3 I;
values. This method is called the "sum of the row" method (HughcH, 1987; Reddy, 2004). I

Both types of :for1nu]ation (consistent and lun1ped) are now compared in tern1s of disper- f
/!/
,'I
lumped mass
-0.4 ':
sion tln·ough a onc-dirr1ensional cxa1nplc (propagation of a flicker \Vavelet). The curves \\,1:.' mixed matrix
displayed in Fig.L1.22 show that the consistent rnass 1natrix formulation overestimates the '~, ·)<~ />'
-0.5 L__~-~-~-~--'"'""'-'--~-~-~-~-__J
v;rave velocity (signal in advance on the reference curve) vvhereas the lu1nped inass inatrix 300 400 500 600 700 800
forrnulation obtained by su1nn1ing the rov.r ter1ns underestin1ates the wave velocity (sig- distance (m)
nal delayed frorn the reference curve). As already shown in Fig.4.21 1 both for1nulations
lead to nurnerical errors with equivalent levels but opposite effects (overeRtimation for Fig. 4.22: Comparison of the numerical dispersions obtained from a consiste11t
the consistent mass, underestimation for the lumped ma.sH). mass matrix (dotted-dashed), a lumped mass matrix (dashed) and a mixed
formulation (dotted).

1.2
consistent
Considering the results obtained 1vith both formulations 1 Lysmer (1970), as 1,vcll as
1.1
HanRson and Sandberg (1997), proposed a 1nixed formulation considering a linear cornLi-
1.0 .... __ _ nation of the consistent rnass rnatrix [M]c and the lun1ped mass nratrix [.1Vf]1 as follow.s:
--- --
0.9
[M]m ~ a[M], + (1 - a)[M]I (4.43)
8
"8 0.8
In Fig.4.22, the result obtained vvith the mixed for1uulation and a coefficient a=0.5
is displayed. The n1ixed forrnulation lecuh; to nu1nerical results very close to those of the
reference case. It is thus much better in terms of nu1ncrical dispersion than the consistent
0.7 and lu1nped forrnulationH. Similar formulations n1ay be considered for higher order finite
element' (Cohen et al., 1994, 2001).
0.6

4.3.8 Efficiency of higher order elements

0.5
0. 0.2 0.4 0.6 0.8 1.0
lD analysis
(2n-l)/2n,,
Nu1ncrical dispersion is also influenced by the order of the polynornial interpolants con-
Fig. 4.21: Comparison between a consistent mass matrix and a lumped mass sidered in the finite element formulation. Higher order finite clc1ncnts are known to be
matrix in the case of the identification of the eigenmodes of a cantilever beam very accurate for conrputatiuns in elastoplasticity (l'vlolenkan1p and Kay, 1997; Sloan and
(Hughes, 1987). Randolph 1 1982). For \\rave propagation, several theoretical v.rorks have proposed son1e
theoretical results to assess the numerical error in terms of dispersion (Bamberger et al.,
1980; Hughes et a.I., 2008; Ihlenburg and l3abu8ka, 1995). Higher order finite clements
 T

270 l'viodelling VI.rave propagation 4.~1 Nurnerical dispersion 271

tvvo tirncs larger vvill have tvvo times less ele1ncnts for an identical nun1bcr of nodes in
/\ /\ '1 " ,,
''
,':
_ __, 11 ~ II '
'•,
the direction propagation (Fig.4.23).

' : '
/\ :
f;
'' I JI I !I
0.5
''' : I I : I'
I J \ I !' I.
linear ' ... : I I :
p=l "' ,./\: ' / I: I 1 .--·.. :
""'B 0.0 ,,
--------~-~ ---1-: --\--«'------! ... i.: --:---~-,'-\.,...-.:/ \.1
\/ ~: J I I --\:

l -0.5
::,,
,,
'I,I I
I '
I
~ '
11 I
I
\I ~
~-
:\ i
:\:
direction of :1/
:v"
I

propagation ''' '

'' T3
quadratic -1.0
p=2
0.5

~
B
;.::::
0,
'I
higher • • 8
"' -0.5
order • p=4
• •
-1.0
' '' '
Fig. 4.23: Numerical dispersion for various types of finite elements: one- 0.5
'
' ' '
'

dimensional problem and arrangement of the elements. '

'
I :'l. : '1 '
: (l
f\ : '1

""'"'
/\ i i \
________/ l : i _\.,. __
I I 1 I I
/ ~ :: \
·-r·:-1 ------ ____ /
/\ :I\.---.\,_,.,.__ _
. : I.
--~-f-~
.B 0.0
--\Ti ,: r \:I, I!
seem to lead to a lov.rer numerical dispersion. } 11:
''I
1: J \1 i
"' \ii I '1 1:1
-0.5 1:1 ;:,
In order to compare the efficiency of finite elen1ents at different order of interpolation, \'/

a onc-diincnsional wave propagation problcrn is D.G\V studied. As shown in Fig.4.23, three

\ii\ri 1:1 1:1

different types of (triangular) elernents arc considered: three-noded linear elc1nents, six- -1.0
"• 1/ T15
0.0 25. 50. 75. 100.
noded quadratic clerncnts and 15-noded fourth order clements (I-Iughes, 1987). For a
meaningful cornparison, the number of elements in each finite elerncnt mesh is chosen in distance
order to keep the nnrnber of nodes in the direction of propagation conHtant for all element
types (Semblat and Brioist, 2000). It turns out that the proportionality of the number Fig. 4.24: Numerical dispersion for various types of finite elements: 3-noded
of elcrnents corresponds to the ratio bct\vecn the interpolants orden.; for the three ele- (top), 6-noded (middle) and 15-noded (bottom) elements. Waveforms at
ment typet:i. The three rneshes indeed involve 80 linear elements for the first one (linear different times and expected theoretical propagation delays (vertical dashed
interpolants), 40 elements for the second (quadrntie intcrpolants) and 20 clements for lines).
the third (4th order interpolants). A. mesh involving elen1cnts with an interpolant order
 --- --
272 Jlvfodclling Vi'avc propagation 4.3 1\1utncrical dispersion 273

rrhe reHults obtained for each mesh are diHplayed in Fig.L1.24. The different clcincnt Higher order elements in 2D
types are denoted 7'3 for the linear threc-noded ele1nents, 1'6 for the quadratic six-noded
As sho\vn by the rei:;nlts displayed in Fig.4.15, nu1nerical dispersion is also influenced
clcn1cnts and T15 for the fourth order 15-noded elements. Figure 4.24 shows that, for
by the 2D features of the 1nodel. For insta.nce 1 the angle of incidence of the vvave on
linear elernents (T~1), the nurnerical dispersion is very strong (the expected theoretical the inesh has a t->trong influence on the numerical error. For higher order elements 1 2D
propagation delays are displayed as vertical dashed lines). For the quadratic elernents nu1nerical dispert->ion vvill be discussed considering a model involving a topographic ir-
(1'6), the results are inuch better 1 but, at a certain distance, the curnulative error be-
regularity subn1itted to a seismic ex.citation. The goal is to esti1nate the influence o:f
corncs significant (third and fourth signah>). In the case of higher order elements (T15), nurnerical dispersion on the seii:;n1ic wavefield scattered at the surface for various ele1nent
the nurncrical dispen.;ion is very srnall. For an identical nurnber of nodes in the direction
types.
of propagation 1 the simulation involving such higher order elements thus gives very good
results. The accuracy of higher or<ler finite elernents in ter1ns of numerical dispersion
scen1s excellent. ]\;fore detailed results arc discussed in (Hughes et al., 2008; Se1nblat and linear quadratic 4th order - - - - -
Drioist, 20110).

scattered
wave

l l l l l l reflected t=0.419s t=0.519s

wave

incident
wave

(a) (c)
' / "' ' ' '/ '
/ ' / ' / /
• • • •
/

/ I/
/

' '/ ' / ' • • • •

'/ '/ ' / '/
/ ' /"'. / ' / '
'/ ' ./ ' / ' '/ /
•
/ ' / ' /'.
"'./ '.I/ './ './
t=0.619s t=0.719s
Fig. 4.25: Scattering of a plane seismic wave by a topographic irregularity to
analyze 2D numerical dispersion for finite elements at different orders: linear
(left), quadratic (centre) and fourth-order (right). Fig. 4.26: 2D numerical dispersion of the scattered wavefield for various finite
elements types.
274 Modelling vvavc propagation 4.4 Physical and nurnerical darnping 275

As shovvn in V'ig.4.25 1 the geurnetrical arrangement of the triangular finite element-H
is chosen in order to avoid any particular orientation for the propagation (Scmblat and
Brioist, 2001). 'The basic patterns of the model involve various types of triangular fi-
nite element types: linear 3-noded clements (a), qu;i,dratic 6-noded (b) and 4th order
15-nodcd (c). All these models involve the same number of degrees of freedom.
2D numerical dispersion being influenced by the wave incidence and the spatial ar-
rangement of the elerr1ents, it is interesting to estimate the effect on the estirnation of
the scattered vvavefield. The corr1puted scattered wavefield is displayed in Fig.4.26 for
various finite elerncnt types at different times. For the first time value) the effect of
numerical dispersion is very s1nall. It becomes significant for the second tirne value and
is strong for the t\vo last graphs. Following a vertical line starting from the top of the
hill, the effect of nun1crical disperHion is rather srnall for all element types. V\Thereas 1 for
otlier directions) the scattered wavefield and the surface \vaves are Htrongly influenced by
numerical diHpertiion (Sernblat and Brioist 1 2001).
'I'he scattered wavefield displayed in Fig.4.27 shcnvs that 2D numerical dispersion is
influenced by the numerical anisotropy of the model, that is by the simultaneous effects
of the wave incidence 1 the spatial arrangernent and the shape of the elernents. Several
areas corresponding to small or large nurnerical errors can be distinguished in Fig.4.26.
However, as shu~rn in (Sernblat and Brioist, 2001), the best nurnerical accuracy in tern1s
of 2D nurnerical dispersion is obtained \Vith the higher order clements.
4.4 Physical and numerical damping
As discussed in Chapter 2, the attenuating properties of soils are very important since
they govern the arnplitude variations during the propagation process. Various types of
methods arc available to model wave attenuation in soils: empirical n1o<lelt.;) rheological
1nodcls, constitutive models, etc. In this section, several approaches dedicated to the
nu1ncrical rnodelling of damping will be presented. The influence of the numerical scheme
will be also discussed since it may modify the \vave an1plitude through purely numerical
damping.

4.4.1 Rayleigh and Caughey damping

A convenient way to model darnping consists in the formulation proposed by R.ayleigh
(Clough and Penzien, 1993). The Rayleigh formulation considers a damping matrix [CJ
proportional to the mass [NI] and stiffness [K] matrices under the following form:

[CJ= a[M] + b[K] (4.44)

A more general forrnulation derived from the Rayleigh one has been proposed by
Caughey (Clough and Pemicn, 1993) and is expressed as follows:
1n-l

[CJ= [M] L Oj ([A·"f]- 1 [Kl)' (4.45)

j=O

For m=l. the Caughey formulation is equivalent to the Rayleigh one. The Rayleigh
and Caughey formulations arc \videly used in structural dynamics and allov,r an easy
con1putation through 1nodal rnethods. They indeed both lead to damping matrices that
are diagonal in the real eigenmodes base.

The Rayleigh formulation leads to a damping ratio ~ involving a first term propor-
tional to frequency and a second term inversely proportional to frequency (Clough and
Penzien, 1993), that is:
Fig. 4.27: 2D numerical dispersion at t=0.719s: comparison of the displace- ~=_a._+bw (446)
ments and the polarizations for linear and higher order (p=4) finite elements. 2 2w
As sho\vn in Chapter 2, the attenuation Q- 1 has a simple expression in terms of
damping ratio, yielding:
l a
Q- = 2~ = -+bw (4.47)
w
276 lVIo<lelling wave propagation 1.4 Physical and nu1nerical darnping 277

Generalized Maxwell model

[Rayleigh J [Caughey J Rayleigh damping
, ,, E
,
,, [CJ= a.[K]+ b.[M]
/
/
/

/
,,<~=bw/2
.. ,,,;..."'
\ / .
\ ,.."'
:
·.-'·.,,.."
.,
. /
_____ J==_a(_~?1_ __ _
,.
/· b
=aw+-
())2
ro
frequency

Fig. 4.29: Equivalence between Rayleigh damping and generalized Maxwell

Fig. 4.28: Damping ratio vs frequency for the Rayleigh and the Caughey
model (Semblat, 1997).
formulations.

As shovvn in Ii"·ig.4.28, for the R.ayleigh for1nulation (left), the dan1ping ratio i8 infinite
at zero fr·cquency and infinite at infinite frequency. It reaches a rninimun1 value in
an intermediate frequency range. For the Caughey formulation (Fig.4.28, right), the
variations of damping are rnorc cornplex depending on the number of terrns in the series
(Eq.(445))
4.4.2 Rheological interpretation of Rayleigh damping
To use the Rayleigh formnlation, it is necessary to have realistic values of both a and
b coefficients in Eq.(4.44). The experi1nental estirnation of these pararnctcrs raises the
need for a physical interpretation of the Rayleigh forn1ulation. As shovvn in F'ig.4.29 1 a
generalized J\!Iaxwell rno<lel leads to the san1e expression of attenuation as the Rayleigh
formulation (Semblat, 1997):
bet1vecn nun1erical and analytical results arc <lisplaycd in Fig.4.30. They shovv an ex-
cellent agree1nent bctvveen both approaches for lov,r to rnuderate darnping values. F'or
larger dan1ping values 1 the approximation <.J- 1 ':::::'2~ is no longer justified and the results
di,agree (Scrnblat, 1997).
4.4.3 Attenuation models for geomaterials
Experirnental results on geomaterials clearly shov,r that various factors infiuence the atten-
uating properties of soils: frequency, mean stress, nurnber of cycles, etc (sec Chapter 2).
:rviore con1plcx damping rnudels than the previous ones arc necessary to include such
various dependencies.
Viscoelastic models

-r Mr E(77r + r12) 1 7]2

QGM ~ l\IIR ~
R,ayleigh <larnping is thus equivalent to a generaliLJed Maxwell rnodcl as depicted in
:Fig.4.29. This equivalence is valid :for moderate values of the damping ratio~ (~<20%),
that is vvhen the approximation c2- 1 ':::::' 2~ is justified and rnatcrial dispersion is weak
(Scmblat, 1997).
To assess such an equivalence for various damping values, wave propagation sim-
ulations are pcrfor1ned considering the R,ayleigh for1nulation. They are cornpared to
semi-analytical results obtained using the generalized :l\.tlax\vell model and directly solv-
ing the ID wave equation in the frequency domain (sec Chap.2). The comparisons
(448)
Various attenuation n1odcls a.re already discussed in Chapter 2 and n1ay be considered
'li w + Ew for the nun1erical analysis with finite elements (DourbiC et al., 1987; Lucet, 1989; Sem-
blat) 19971 1998). It i.s nevertheless necessary to incorporate so1nc dissipative terms in
the variational forn1ulation through a fully reliable mechanical formulation. Another key
point is the accuracy and the cost o:f the computations. Since attenuation rnodels gen-
erally involve mcrnury variables, the con1putational cost may be large (depending of the
considered range for the relaxation tirne). Various approaches vvere proposed by differ-
ent authors ranging frorn Pade's approximants (Day and 1'vlinster 1 1984), coarse-grained
n1emory variables (Day, 1998) to parallel 3D viscoelastic finite difference :forrnulations
(Bohlen, 2002). Thanks to such models) it is possible to obtain satif;factory results in
the linear range (Emmerich and Korn, 1987; Scmblat, 1997; Semblat and Luong, 1998)
as well as in the nonlinear range (De!Cpine et al., 2007).
278 Modelling \\rave propagation 4.4 Physical and numerical damping 279

- - - Rayleigh ........... analytical (G.M.)

1.0

attenuation : Q"'
0.75
. ...--;.-
" / '
''
"Ci /
Saxena's law
~
"
;!:I 0.5
•3
I
I
''
'
s'" '' Hardin's law
I j
" •p·4/3 I
Stewart's law
0.25 .I ''
:S ! fp·0,5
'" I
" z
"Ci

1.0
Is~ 13 % l Fig. 4.31: Various damping profiles for wave propagation tests in the cen-
trifuge (see Chapter 3).
0.75

"
"Ci

'"'"
Nevertheles.s, a first attempt n1ay be to n1odel attenuation using R,ayleigh darnping
;!:I 0.5
but considering an inhornogeneous darnping profile (Idriss et al., 1974; Semblat and
'"s Brioist, 1998). To do so, it is neceHHary to define a darnping subrnatrix [C]c for each
" 0.25
finite element e as follows (Idriss et al., 197 4):

(449)

where [M]e, [CJ, and [K], arc the mass, damping and stiffness submatriccs (resp.) and
0.0 0.5 1.0 1.5 2.0
ae and be depend on the dan1ping propertieH of elcn1ent e at an average depth Zr-:.
time (ms)
It is thuH possible to simulate seismic \vave propagation fron1 a inhomogeneously
Fig. 4.30: Agreement between analytical results derived ~rom. the generali~ed damped soils. Scrnblat and Brioist (1998) performed numerical sirnulations considering
Maxwell model (symbols) and numerical simulations cons1derrng the Rayleigh
various da1nping profiles. As depicted in Fig.4.31, centrifuge teHts (already presented in
formulation (solid). Chapter 3) arc analyzed considering three different damping profiles:

• Saxena)s law: Saxena and Reddy (1989) proposed the following attenuation depen-
dence:

Other finite clc1ncnt for1nulations are also available such as fractional derivatives (4 50)
viscoelaHtic rnodcls as proposed in the fieldH of structural dynamics (Sorrentino and \Vhcre K is a constant, P is the rnean stress a.nd c is the Rtrain.
Fasana, 2007).
• Hardin)s lau;: Hardin (1965) proposed a specific relationship for sands:
Rayleigh damping for wave propagation problems
Q- I = 9 ,0.2 p-0.S
le:c (4.51)
The use of Rayleigh darnping for \Vave propagation proble1ns is possible, but it is rather
limited since it does not account for the coruplcx anel;-u;;tic properties of Hoils. where "/e:r is the shear Rtrain and P is the mean stress.
280 .!'vlodelling vvavc propagation 4.4 Physical and nu1ncrical da1nping 281
• Ste,wart's la,111: Stewart ct al. (1983) asfn1mes another kind of relation for rocks: The damping profiles derived froru theRe lavvs for the centrifuge expcrirncnts discussed
in Chapter 3 arc displayed in Fig.4.31 (right). The case of ho1nogcneou.s damping is also
(4.52) considered. It can be noticed that the darnping profile derived frurn Stev.rart's law leads
to very strong variations of attenuation with depth. This lct\V was proposed for rocks and
i.s probably not well-adapted for .soil.s.
Vill1erc ( is the crack density in the spccirncn.

F'F~l\1I sin1ulations considering the.se four darnping profiles were pcrforrned hy Se1nblat
.. ----· ··'·· surface
and Brioist (1998). The results are displayed in FigA.32 for both a surface and a deep
0.002
---1-- . ···---.
sensor Sl sensor location. The n1otion cornputed v.rith the Stev,rart 1s lavir is very small and this law
is not reliable for soils. Couversely1 the results derived from the Hardin's la\v see1n to
-:-_-:r-
~
'1 0.0
--/-'<
---(./,,.----------- .,,.:-·~.-·--
-:-~~. ~;.[,.;-~--~-~--==c;:;;:z:~:::s. __________
-~~---:_-_:,__.
::::~::~- be very interesting when cornpared to experi1nental results (Semhlat and Brioist, 1998).
s'" 1 /1,
_:~' :
F·inally, for the Saxena's la1,v 1 the numerical resultt-> are very close to the homogeneous
'"u -0.002 "•
I
I /~ : darnping case and do not alloi,v to retrieve the dan1ping to nrcan stres.s dependence ob-
"'w "'-•• 1
served experimentally (Semblat and Luong, 1998).
""_,,
'CJ -0.004
~
'

Saxena's law From these results, it nray Le concluded that, to sinrulate 1.vave propagation in soils,
I I

"'
~
-0.006 son1e sitnple nurnerical rnodels, involving rnore or less realit->tic damping profiles, may
'"'>" Hardin's law
be perforrned using the Rayleigh dan1ping assu1nption. Nevertheless, for rnore CC)lnplex
-0.008 Stewart's law situations 1 truly viscoelaRtic, or even elastoplastic, inodels are necessary (Day, 1998;
Delepine et al., 2007; Euunerich and Korn 1 1987; Sorrentino and Fa.sana. 1 2007).

-0.002 0.0 0.002 0.004

4.4.4 Numerical damping
0.0 --:;:-"";. ~;!_ ______ _ In addition to numerical dispersion previously discuRRed (and inainly depending on the
/
~
'1
spatial discreti:-~ation of the nrodel), time integration schernes alt->o involve nun1erical
s'" -0.001 mid-depth da1nping phenom.ena. Such phcnornena lead to an artificial arnplitude decay of the v,rave
'"u sensor 82 due to purely nurnerical errors (see Fig.4.12). This arnplitude decay is generally narned
"'w -0.002 algor·ithrnic or numerical damping (Hughes, 1987).
""
'CJ -0.003
Ciu boundary
<:l

cond~JJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJ
-0.004
'"
'"
>
-0.005
u(x=O,t)
-0.004 -0.002 0.0 0.002
horizontal displacement t
I
Fig. 4.32: Soil response in the centrifuge for various damping profiles: FEM
simulations using variable Rayleigh damping (Semblat and Brioist, 1998).
Fig. 4.33: Effect of numerical damping on the propagation of a steep signal:
schematic of the model.
 '
'

282 l\!Iodelling wave propagation 4.fi .i\ Iodelling wave propagation in unbounded media
1
283

Hughes (1987) and Curnier (1993) derived various results for classical time integration
schcrncs. Nurnerical da1nping is inore or less sensitive depending on the chosen algorithm.
The Ncwrnark type algorithms may, for some values of the controlling pararneters 1 in-
volve no numerical da1nping at all (for inHtance if the 2nd parameter of the Newmark
sche1ne is 0.5). If v,rell controlled, numerical damping inay allow the attenuation of spu-
rious oscillations at high frequencies.
As sho-\vn in Fig.4.34, nurncrical darnping attenuates the alnplitude of spurious os-
cillationH in the Reis1nic wave. Such oscillations arc artificial an<l are sirnply due to the
fact that the excitation includes higher frequency components. Such cornponent.s cannot
propagate correctly in a discrete unda1nped rnediurn, since the element size should be
extrernely srnall (Hughes, 1987). The observed oscillations are larger for quadratic ele-
rnents (bottom) since these elernents are inore sensitive to the fast variations of the wave.

As depicted in Fig.4.33, the influence of nun1erical darnping is now cstirnated conRid-
ering a step \Vavefront generated at the end of a one-dimenRional inodel (linear elements
or quadratic elen1ents). The :.>revvrnark algorithm conRidered herein involves either no
mnnerical damping (,8=0.25 and '/=0.5) or a significant numerical damping (/]=0.3 and
--y=0.6).
AR Rho>,vn through the curveR diRplayed in ~'ig.4.34) numerical da1nping allows re-
moval of spurious oscillations in the case of fast varying excitations (pulRe) Heaviside,
etc) or nonlinear \vavc propagation including, or generating, higher frequency con1po-
nents (Del8pine et al., 2007). Another solution to avoid thiR prohlern consists in using
excitations having slower variations, as is generally the case for actual signals.

Newmark (non dissipative) Newmark (dissipative) 4.5 Modelling wave propagation in unbounded media
~=0.25 and y=0.5 ~=0.3 and y=0.6
1'v1odelling Virave propagation in (scn1i-)infinite rnedia rnay be difficult depending on the
numerical n1ethod conRidered. AR it will be shown at the end of this Chapter, bound-
hJ ~.,,,_1::..,.- . . . . .<-1;-·-·~·:'.'\•••''' ary integral equation methodR allow an accurate description of wave propagation in

I
'
i
\
\
unbounded incdia (Bonnet, 1999). \\Tith the finite or spectral elernent 1nethod) it is nec-
..._ I ~ \ e.s.sary to avoid artificial \vave reflections on the metih houn<larie.s. VariouR techniques
I i \ vvhich solve thi.s problern will be discussed in the following: absorbing boundaries, infi-
I ~ \ nite elements and absorbing layers (Givoli, 1992; J\lleza-Fajardo and Pa.pageorgiou) 2008;
I i \ F'esta and Nielsen, 2003; 1\1odaresRi and Benzenati, 1992).
I i \
I i \
\ \ \
4.5.1 Absorbing boundaries in lD
I \ I\ \.I

As already demonstrated in Chapter 2 in the lD case (§2.3.7), it is possible to remove

¥.rave reflections at the n1odel boundaries by adding a dashpot having an adequate viscos-
ity depending on the mechanical properties of the medium (Fig.4.35). The parameters of 11
the abHorbing elernents rnust cornpcnsate the influence of the refiectecl wave (free-surface
condition). From Chapter 2, the optimal viRcosity of the dashpot is TJ = pc (p: mass I 1

denHity) c: wave velocity). As it vvill be Rhown in the next section 1 the analysis is n1ore
con1plicated in the 2D case.

distance

Fig. 4.34: Effect of numerical damping on the propagation of a step wave- I I I I I I I I I I I I I I I I I I I I 111111111111111111111111111111
fror1t: waveforms at different times for linear elements (top) and quadratic
elements (bottom) using a non dissipative (left) or a dissipative (right) New- Fig. 4.35: Absorbir1g elements in the one-dimensional case.
mark integration sche1ne.
 T
284 J'viodelling vvave propagation 4.5 l\iiodelling wave propagation in unbounded 1nedia 285

4.5.2 Absorbing boundaries in 2D 1'he co1nponents o:f the Cauchy stress tensor in cylindrical coordinates are thuR:

Du, 'Ur
(
u,,=>-+2µ-+>--
)
In 2D cases) the theoretical solution for the optimal absorbing conditions is not as simple Dr r
as in the lI) case (§2.3.7). A 2D example will now be considered to design absorbing crre =µ(Bue_ ?Ie) (4.53)
boundary conditions: a cylindrical domain is rnodelled by a quarter of a diHk discretized Dr r
with finite clements (plane strain). Bur Ur
aee = A -
0r
+ (,\ + 2µ)-
r
As shovirn in Fig.4.36, discrete elen1cnts are connected at the <lornain boundary. The From the equilibrium equations in cylindrical coordinates and assun1ing pure body
goal is to have a theoretical dctcrrnination of the optirnal features of the absorbing ele- wavefl at a distance from the source 1 the follu1,ving differential equations corresponding
rncnts in order to rninilnize or fully ren1ove spurious 1.vave refiections. to the t1,vo body v.rave types (P and S-v.ravefl) are obtained:

• for P-waves:
This exarnple vvas for instance proposed by Dubrencq and Pia.u (sec Bisch et aL) 1999) 8 2 ur l B'Ur 1
+ - -,- - - 2 Ur
---;::JT (4 54)
to analyse soil-structure interaction. The first step consists in deriving the equation of ur- rUr r
propagation for cylin<lrical waves (Bisch ct al. 1 1999). The n1ediurn is assumed elastic)
• .for S-waves:
linear and isotropic in the framevvork of s1nall strains. The dit->placement field is supposed D2 ue 1 Dne 1 1 iJ 2 ue
to depend on the radial distance r only. --+--- - -ue
2 = --- (4.55)
Br2 r Br r V 5? 8t2

The idea of the rnethod is to evaluate the stresH components at a certain radial distance
Re (n1eRh boundary) and to design adequate aLsorbing boundary conditions using springs
and dashpots. For that purpose) vve need to express the strefls con1ponentfl as .functions
of the diflplacement cornponents and their time derivatives. Fro1n the expression of the
stress components 1 Eq.(4.53), the spatial derivatives may be easily converted into time
derivatives as follu1,vs (see Chapter 2):

k=P,S (4.56)

At radial distance Re, the co1nponcntH of the stresfl tensor n1ay be easily derived as
functions of the displacement co1nponents and their time derivativeR:
• for P-waves:
,\ + 211 , ,\
O"rr = - -V-- 'Ur + RUr (4.57)
p e

• for S-waves:
µ , /J
uro = --ue- - u e (4.58)
Vs Re
v.rith Re the external radius of the cylindrical dornain, Vp and ~s the P and S \\rave
vclocitiefl (resp.).

The coefficients of the displacement tcrn1fl and of the time derivatives of displaccrnent
represent the lineic stiffr1eflses and lincic dampings:
Fig. 4.36: 2D mesh with dashpots and springs at its boundary (Bisch et al.,
1999). 21
• the - >.+
Vp 1,and - -1!:_
V:s· ter1ns are the long'itudinal and transverse viscoufl damping
coefficients respectively,
286 1.ilodelling \Vave propagation 4.5 l'vlodelling wave propagation in unbounded rnedia 287

• the }L ancl -tf;_ tern1s correspond to the lincic stiffnesses for each wave type. Secor1d approach
It is also necessary to dctcrn1inc the characteristic features of the springD and daHh-
In thiti case, the dornain of interest rernains bounded and the local variables 1 depending
pots in the fra1ne\vork of a finite elernent formulation (depending on the ele1nent type)
on the nu1nber of directions extending to infinity, range between -1 and + 1. \i\Te thus
and take into account the discretization of the boundary (I3isch et al., 1999).
transforrn the infinite gcon1etry of the real domain in a bounded dornain hy using rnapping
functions involving the reciprocal function of the considered decay. For this type of
For a given finite element discretization and excitation (P or S wave), it is thus
element, \VC thus keep the shape functions of the classical finite elements (see Fig.4.38).
possible to determine the optimal stiffness and darnping values to rc1nove the spurious
reflections at the don1ain boundaries. One should nevertheless notice that these optirnal
values arc dctcrn1ined assurning the radius of the cylindrical don1ain: Re, to be large and 0 1 2 3 00

the simplification of the stress components thus obtaincu (Eqs (4.57) and (4.58)) does
not account for the influence of frequency. These resultR appear as a generalization of

j
the lD results obtained in Chapter 2 (§2.3.7).

4.5.3 Infinite elements

The first researches on infinite elernents were proposed in the early 70's by Zienkievvicz
et al. (1985). The goal of this method is to artificially extend the medium towards 1 2 3
infinity, by coupling infinite elements at its houndaries (Chadwick et al., 1999). It allows
a reduction of the 1nodel size as \vcll as of the cornputational ti1ne. Niore recent works
r=-1
•
r=O
•
r=l
were also performed in the 90's (Autcnzio, 1997; Bettess, 1992; Laghrouche, 1996).
Fig. 4.38: Mapped infinite element (-1:<:;r:<:;1).
First approach of the method
The basic principle consists in keeping the functions :for the geometrical transforrnation
of the finite _elernents and to chose shape functions with a weighting function nrodelling
the decay at infinity. By thiH rneans: one tries to approxirnate the actual physical process.
The do1nain of interest, as v,rell as the local variable(s) characterizing the systern geon1ctry1 Comparison of both approaches
is thus extended to infinity (see Fig.4.37).
The major advantage of the first rncthod is its natural approach an<l itH easy in1plernen-
1 2 tation in a cornputer program (a sirnple multiplication bet\veen a polyno1nial function

··-.....
00
5 and the weighting function is only necessary). ItH main drawback is the co1T1putation of
the elerr1cnt matrices involving generalized integral along infinite directions.
r
'rhe second method leads to integrals over finite limits and involves polynornial r:;hape

3 4
··-.....
00

6
functionti. The n1ain difficulty is the calculation of the Jacobian of the trantiformation,
the integration being performed in local variables. The derivatives of the reciprocal func-
tions are thus needed. In order to avoid complex cornputations, it is necessary to cho8e
decay lav,rs having the most regula.r analytical expression as possible. For this rnethod,
classical integration methods such as Gauss quadrature may Le used (Chadwick et al.:
1999; Laghronche, 1996).
s
Both methods are actually equivalent since 1 through a change of variables, -,,o,re can
Fig. 4.37: Infinite elements and decay function (r--+ oo). show that the integrals cornputed by each rnethod arc identical. T'he Jacobian of the
11 i'.I
gcon1ctrical transforrnation indeed allovvs us to retrieve the same cxpre8sion (Bettess, 1:
1992).
288 1\.-fodclling wave propagation 4.5 l'vlodelling Vl.'ave propagation in unbounded 1nedia 289

.- - __________
- - -1_,_ 4 = _:;::_::::_
The infinite elements depicted in Fig.4.:l9 (top) may thus be related to standard finite
elen1ents through such transfor1nationR.

·------ 5 6
-----·----------·= fn the case of an infinite element along two directioru; (Fig.4.39 1 bottom), previcnu>
expressions may be easily generalized (Bettess, 1992).
3 =
• - - - - -2-e-----+--------=c:::_ Infinite elements in dynamics

Depending on the chosen n1ethod, the weighting functions in the shape functions or the
inapping functions depend on the type of decay considered in the analysis. The choice
of the decay is related to the physical problem studied: cylindrical geornetry1 spherical
geornetry) anelastic inedia, etc. Furthern1ore, these decay lavvs depend on the wave type
conRidered. The work of fv1edina and Taylor (1983) gives the analytical expressions o:f
the gcornetrical transformation functions. A da.rr1ping factor 1 denoted (, is taken into ac- :

count in these functions under the form of a co1nplex exponential exp( iE). The final forrn 1

of the functions accounts for the type of decay and the damping factor. 1'he inethod I

proposed by Medina and Taylor (1983) has been considered for the analysis of traffic
induced vibrationR in soils by Laghrouche (1996). The numerical reRults obtained by this
author show the efficiency of such a m.ethod for 'Nave propagation problerns.

Fig. 4.39: Schematic of a 2D infinite elements along one direction (top) or Some other '.vorks in the field of dynan1ics \Vere dedicated to the forrnulation of infinite
two directions (bottom). elements (Gerdes, 1998; Ycrli ct al .• 1999).

4,5,4 Absorbing layers (PML)

Two-dimensional case
·The abRorbing layers rnethod is different frorn the absorbing boundaries technique since
In 2D, the principle of the 1ncthod is the same as in lD and the 2D clcrnent may be the idea is llOV\T to avoid Hpurious reflections at the medium boundaries using layers having
infinite along one or t\vo directions. Figure 4.39 (top) displays a 2D infinite element specific attenuation properties. It is for instance the case :for the Perfectly Nfatchcd Layer
along one direction only. In this case, Chadwick et al. (1999) proposed a radially infinite method briefly presented in the following.
ele1T1ent with the folluvving mapping functions:
Perfectly Matched Layer method
M1 = ( -1-s
2s ) 1
( -2r(l-r) )
The method proposed by Basu and Chopra (2003) is presented for a semi-infinite rod on
elastic foundation subjected to an harmonic diRplacement at one end. The tirne-harmonic
lvl, = c+s)(
-.-
l - s
--r(l-r)
1
2
) displacement in the rod is governed by the following equations:

dCT kg
M, = ( -~
2s ) (l+r)(l-r) - - -u=-w 2 pu
l-s dx S
(4.59) er= EE
lvl4 = c+s)
l-s (l+r)(l-r)
E=-
dv,
(4.60)

dx
lvf5 = ( --2s- ) -r(l+r)
l
1- s 2 \Vherc er and E are the) scalar, axial stress and strain in the rod, E and S' are the Young'H

1\116 = c+s)
- - -'r(l + r)
1-s
J
2
n1odulus and the cross-sectional area of the rod 1 p the 111.ass density and k 9 the stiffness
per unit length of the foundation.
290 1vJ odelling \\'ave propagation 4.5 J\ifodelling 1,0,rave propagation in unbounded n1cdia 291

The space coordinate :r is then replaced by a stretched coordinate X such a8 (Basu

elastic medium Perfectly Matched Layer
and Chopra, 2003):
(4.61) ... -··"'
incident ·· ·· attenuated
where ,\ is the coordinate stretching function. It is assumed continuous and complex- wa~ wave
valued th118 leading to: .
dx
- = .\(x) (4.62)
dx Fig. 4.40: Principle of Perfectly Matched Layer formulations.
The systen1 of equations (4.60) can be rc\vritten as:

l dJ kg 2
--~ - -?J, = -W pH
.\(x) dx S This property is the main interest of PJ'vl:i'vl: it 111ay be used to rnodel VI.rave propaga-
tion in an unbounded domain. A Perfectly lviatchcd Layer (PJ'vliVI of finite length L) iH
a= Ee (4 63)
then considered (Fig.4.40). The paran1cters of the PJ'vlL n1ust be chosen in order to have
1 du sufficient attenuation in a finite distance.
c=---
.\(x) dx
The PlvII_j generally very efficiently avoids reflectionR of body 1vaves. Nevertheles8,
As depicted in Fig.4.40, for Perfectly Matched Medium (PMNI), the.\ functions may as discussed in the next paragraphs, the effect of the Pl\!IL technique is not optirnal for
be chosen 1vith identical values at the interface between two 1nedia (no reflected \Vave) grazing incidences and rr1ay even turn aR a drav1lback for surfa.ce ¥.raves.
and the solutions of Eq.(4.63) are similar to those of Eq.(4.60). For ao < 1, evanesccnt-
wave-type solutions are of the form (Basu ;.ind Chopra, 2003): Filtering PML formulation

u(x) =exp (-Jl - a6~) and u(x) =exp ( +J1 - a6.~) (4.64) As shown by _Festa ct al. (2005), denoting the coordinate change as:

whereas, for a0 > 1, Eq.(4.63) yields propagating-vvave-type solutions: (4.69)

a plane vvave <P(x, z, t) = Aei(wt-k"':i;-kzz) beco1nes in the Pl\IIL area:

u(x) =exp (-iJa6-l~:) and u(x) =exp ( +iJa5-1~) (4.65)
"'. ' - k ..,, :s
iP(x, z, t) = iP(x, z, t)e w

For a 0 > 1, the stretching function A rnay be chosen as:

1vhere I; is an arbitrary function decaying fron1 the inediurn/layer interface to the exter-
'( ) = 1 _ . f(x) nal boundary of the layer and thus leading to an exponential decay of <f>(x, z, t).
AX 'l~. (4.66)
v ao -1.
In the case of Rayleigh vv~aveR, we consider the plane wave decomposition as follows:

then: x
ro
= x - i
ro
~
a6 - 1
where: P(x) = ["
Jo
f(l,)dl, (4.67)
W(x, z, t) = e-i.(wt-k.,,,x) (Apc-wr/pz + Ase-wiJsz) (4. 71)
where i/.t = J,R - J,' (·i = P, S) anrl (Ar, As) are constants (see Chapter 3). i:I
vcith l, = f,/r 0 and f(l,) = J(rol,) = !(0 (f ie a real-valued continuous function).
fl
If the R.ayleigh 1.vave has sorne interaction 1vith the bottorn PJ\!IL in the model, the
The solution is thus: geo1netrical transformation yields:

u(x)=exp -F [ -(.7;)] ( .c:;--;x)

ro exp -·iyao-lro (4.68) (4.72)

For w -+ 0, the Rayleigh 1vave beco1nes a Lody VI.rave and the vertical wavenurnber is
corresponding, for .F'(-!r;) > 0, to a propagating wave attenuated in the x direction. controlled by L; which is an increasing function of depth (Festa anrl Nielsen, 2003).
292 lVIodelling wave propagation 4.5 l'vfodclling wave propagation in unbounded inedia 293

left right
PML PML

surface ~ ~
waves
anisotropic grazing " /
incidence ., :
medium
GFMPL

bottomPML

Fig. 4.41: Various configurations leadi11g to significant numerical errors in

the PML formulation. PML

Fig. 4.42: Comparison between the filtering PML formulation (middle) and
As depicted in Fig.4.41, the surface v.raves may then he arnplificd at depth leading to the classical PML formulation (bottom) leading to an artificial amplification
a significant numerical error. Festa et al. (2005) proposed an enhanced PML method: of surface waves (Festa et al., 2005).
the Filtering PNIL (FPML).

In the frequency domain, transformation (4.69) has a pole at w = 0. As proposed by

Festa et al. (2005), it can be moved along the imaginary axis as follows: This transforrnation thus retains the features of a surface wave at low frequencies
without spurious effects in the PML (Dclavaud, 2007). As shown in Fig.4.42, the filtering
E(x) PrvfL for1nulati.on (rnitldle) avoids the problen1 of artificial surface wave a1nplification
X=x+-.~~­ (4.73)
iw+u,.·e with the classical PML (bottom). Nevertheless, as depicted in Fig.4.41, there are still
where We is a constant. sorne nlunerical instabilities vvith the Pl\iJL for grazing incidences (Festa et al., 2005;
Komatitsch and Martin, 2007) or anisotropic media (Bccache et al., 2003). Meza-Fajardo
Using transforrnation (4.73) instead of (4.69), the body vvavc no\\r decreases in the and Papagcorgion (2008) proposed a multidirectional PML formulation to overco1ne this
PML, yielding: difficulty.
~ (kx
<l>(x, z, t) = <!>(x, z, t) exp -
2
w - iwwc E ) (4.74) M ultidirectional PML formulation
w w 2 +we2
The exponential decay is now frequency dependent: the layer corresponds to an elas- To deal "1.vith the nu1nerical in.stabilities of Pl'vfL for grazing incidences (Festa et al., 2005;
tic n1edium at lo-\V frequencies and to an attenuating layer at higher frequencies. Komatitsch and Martin, 2007) or anisotropic media (Becache et al., 2003), IVIeza-Fajardo
and Papageorgiou (2008) proposed a multidirectional PML formulation (Fig.4.43). In the
For surface "1.vaves propagating vertically1 the e-wryz tcrn1 in Eq. (4. 72) becorr1cs: classical PlVIL approach) the maxirnnm attenuation is ahvays in the direction of one co-
2 ordinate (generally x). As depicted in Fig.4.41, for waves entering the PML at grazing
exp(-wr1z) __, exp (-wr1z - E(z) ww, "1) + iE(z) 2
w r1) (4.75) incidences) the vvavcnumber kx is nearly zero (i.e.) vvavc vector nearly perpendicular to
2
w +we w +we2 the direction of attenuation) and the attenuation is thus minimal.
The first terrn ret;:..tins the exponential decay of surface v.raves and the second term is
also a decreasing term (tending to 0 for w ---t 0). The basic idea of lVIeza-Fajardo and Papageorgiou (2008) is hence to consider var-
ious laws of attenuation along the three space coordinates. The coordinate stretching
I
 T
294 l\Joclelling 1.vave propagation 4.5 J'viodelling wave propagation in unbounded 1nedia 295

approach is thus gcncraliLJed as follovvs:

classical PML M-PML
-- 2_ (x)
x ~ x + . a'f; x
iw .,

- 1 (z) (4,76)
'u=y+-a''Y
d • iw Y
1
Z = z+ ~a(x)z
' z
iw

"'l "' ' ~ ~ [""''""

where the functions ctlx) arc related to the damping profiles dl:i.:) through:

(4, 77)
a~xl(x) = d£xl(.r)
a1'l(x) = d~xl(x)

In the classical Pl\/IL approach, a:~x)=O and a~x)=O and the maxirnurn attenuation is
ahvays in the direction of the x-coordinate. For the JVI-PJ\IIL, similar expreHsions rnay be
obtained in layers having their damping profile controlled by the y or the z coordinates.
'The treatn1ent of corner regions is sirnilar to that of the classical PJ'v1L, the properties of
the overlapping layers being superi1nposed (J'vieza-Fajardo and Papageorgiou, 2008).

The lVI-PJ'vIL plane-wave solutions arc derived under the following form:

y, =A cxp(-g,;r) exp[-i([;;,;r - wt)]

T (4 78)
(x)
nzO!.z }

where T denotes the transpose of a vector/ matrix.

The real-valued vector Q can be interpreted as an attenuation vector sirnilar to the

case of viscoelastic materials (Chapter 2), In the M-PML, the attenuation docs not
depend on frequency and the for1nulation thus corresponds to a non dispersive mediurn.
Conversely to the classical PJ'vIL, the J\II-PJ'vIL allows an optimal choice of vector _g_: its
components may be related to those of the wave vector Ji.. The direction of 1naximum
attenuation n1ay thus be closely related to the direction of propagation. As shovvn by
Meza-Fajardo and Papageorgiou (2008), the M-PML is consequently more efficient for
grazing incidence than the classical Pl\!IL. 1\!Ic:ta-F'ajardo and Papagcorgiou (2008) also
proposed dispersion relations and reflection coefficients for J\II-PJ'v1L in the case of isotropic
media n1odelled by the spectral elerr1ent rnethod. They also defined a general procedure
for constructing stable JVI-PJ\IL terminations for any anisotropic media. The stability of Fig. 4.43: Snapshots of propagation of the velocity magnitude in orthotropic
this formulation is con1pared to that of the classical P1\!IL in Fig.4.43 for an orthotropic elastic medium at t = 35µs (top), t = 60µs (middle) and t = 400µs (bottom):
medium (Meza-Fajardo and Papagcorgiou, 2008), For the longest time (t=400µs), the classical PML (left) and M-PML (right), withp(y/x) = p(x/y)=0.15, as proposed
classical PML formulation is found to be unstable (FigA43, bottom left) whereas the by Meza-Fajardo and Papageorgiou (2008).
M-PML formulation remains stable (FigA43, bottom right),
296 J\!Iodelling \vavc propagation
4.5.5 Coupled approaches
In addition to the absorbing boundaries/layers, others technique.s rnay also be considered
to deal vvith vvave propagation in unbounded do1nains. It if.; possible to couple different
numerical methods (FEM/BEM, FEM/FDM, etc.) or scale models with various spatial
resolutions. Two approaches will be briefly discussed in the following: the FEM/BEM
coupling and the Dornain Reduction lt1-'ethod.
FEM/BEM coupling
'To deal \Vith \Vave propagation in unbounded media, it is possible to couple the Finite
Element Method and the Boundary Element Method (Bonnet, 1999; Dangla, 1988). As
it vvill be seen in the following, the Boundary Elernent J'vfcthod allows an exact descrip-
tion of the radiation conditions at infinity and is thus a very interesting alternative to
FE:V[ approaches.
Since the Finite Elernent ]\ilethod is very flexible and allov.,rs ll10delling of cornplex
constitutive lavvs as \vcll as realistic civil engineering structures, whereas the Bound-
ary Element Jlviethod does not, it is interesting to cornbine both rnethods. As shov.rn in
Fig.4.44, the propagation near the free-surface and the dyna1nic soil-structure interac-
tion is modelled by the Finite Element Method (near field). The propagation at long
distances, far .field) is preferably rnodellc<l by the Boundary Element JVIethod since the
radiation conditions are exactly accounted for thus avoiding such techniques as absorb-
ing boundariet>/layers. At the interface between both dornains, the degrees of freedom of
tbc FEM model arc condensed and coupled to those of the 13EM model (Bonnet, 1999;
Dangla, 1988).
FE model
-- --
(near field)
co co
I I
BE model
/ f---·---------1
(far field)
Fig. 4.44: Principle of FEM/BEM coupling in elastodynamics.
4.5 l\!Iodclling wave propagation in unbounded n1edia 297
Domain Reduction Methods
Another alternative is for instance the Domain Reduction Jv[ethod inveRtigated by Bielak
et aL (2003). The rna.in idea is to inoclel Vi.rave propagation using the finite-element
methodology but considering highly heterogeneous localiL'ied regions with large contrasts
in v.ravelcngthR. The target is then complex geological structures (sedimentary basins,
ridges) that are far away from the earthquake source. Bielak et al. (2003) proposed to
subdivide the original problem into t\vo sirnpler ones. The first is an auxiliary problern
that silnulates the earthquake source and propagation path effects forgetting the detalled
surface geology. The second problern models local arnplification in surface layers (sec also
Chapter 5). The basic idea consists in choosing a set of equivalent localized forces derived
from the first step as the input for the second problem. The dornain size in the second
step may thus be drastically reduced. If the subsurface geological Rtructurc is simple,
one can replace the finite-elen1ent rnethod in the first step \~rith an alternative efficient
method. The Domain Reduction Method, proposed in (Bielak et al .. 2003), is illustrated
in another paper by the sarne authors (Yoshirnura et al.) 2003) dealing with 3D problems
such as v.rave arnplification in the Los Angeles Basin. In this paper, as illustrated in
Fig.4.45, Yoshimura et al. (2003) also applied the method to the analysis of the seismic
response of a \Veathered hill. The Domain Reduction lVIethod may thus be conRidered at
various scales and in different frequency ranges.
X(m)
0 500 1000
o
o-
I I I
"'
0
~-
I I I
0 2 4 6 8 1012141618 (cm)
I OOOm Min=3.90Max=l1.2
I
Fig. 4.45: Response of a weathered hill modelled by the Domain Reduction
Method in the framework of the FEM (Yoshimura et al., 2003).
298 1v1o<lclling wave propagation 4.6 The Boun<lary Elernent J'vlethod 299

4.6 The Boundary Element Method Fundamental solutions

4.6.1 Interest of the method in dynamics 'I'he funda1nental solutions, in tirne-harmonic elastodyna1nics) are defined by a force of
unit amplitude applied at a fixed point 1L along a fixed direction k• .f,(.i;) = o(.i; - u)oik·
In addition to the Finite Ele1ncnt J\1cthod, such 1ncthods as the Boundary Elen1ent For unbounded domainR, the fundamental solution, denoted by ui(.£) = ut:(z_, y; w ), is
lviethod are particularly suitable to model 1..vave propagation problems. This method knovvn as the Heln1holt:6 fundarnental solution and is given by (E·ringen and Suhubi,
avoi<ls spurious reflections at the rnodcl boundarieD since it directly includes exact radia- 1975)•
tion conditions (Aubry and Clouteau, 1991; Bonnet, 1999; Dangla, 1988). Neverthele"8, k. . - 1 [ 1 a2 (eiksr - eikpr) - J
it is mainly adapted to media having a linear behaviour. As the Finite Elernent J\1ethod 1 U, (.i;,y,w)- - ,, -- + eiksr 6,k (4.83)
- 47rµ ks O.TjOXk r r r
the Boundary Elernent 1'v1ethod is also sensitive to the clcrncnt size considered 1 but there
are no sin1ilar cumulative errorR during the propagation process (nurncrical dispersion). where r 2 = (I - y) 2 and where kp = w) p/(>. + 2µ) and ks = w,/P[µ are the lon-
Furthermore, this method is suitable to inodel various seif;n1ic t;ources such as plane >.vaves gitudinal and transv8rse >.vave numbers respectively. 1'he stress tensor associated Virith
or extended as well as point sources. U;'(I,JL;w), defined Ly (4.81), is denoted by I:f,(I:,JL;w) while the streos vector applied
to the surface boundary of fl is T,k(I:,JL;w) = I:fj(I,JL;w)n1 .
4.6.2 Maxwell-Betti theorem
'I'he funda1nental solutions arc generally kno>.vn as the Green's functions of the prob-
The classical >.vay to derive the Boundary Integral Equations (BI E) fro1n the equations le1n. In the 21.) case, the fundan1ental solutions are defined as Hankel functions (Dangla,
of continu1un mechanics is not linked to variational approaches. The n1ain idea is to 1988; I3onnet 1 1999). A cc)lnpendi111n of fu11da1ncntal solutions in elastodynamics vvatl
consider the lVfaxwell-Betti theore1n as follows. For sake of si1nplicity, we Tu'ill firstly recently published by Kausel (2006).
Virrite the equations in the case of linear elastostatics. In that case, the principle of
virtual vvork leads to:
Integral representation formula
f g • "(fl)dfl - ;· (g n) fl.dS - ;· l_.fl.dO = 0 (4. 79) For Rake of simplicity let us consider no body force fron1 noVir on. Application of the
lo ao o
Maxwell-Betti reciprocity theorem, Eq. (4.80), leads to the following displacement integral
All clastostatic states { (T.!.1, g1, [' ), (T.!.2 , g 2 , [')} in n satisfies the Maxwell-Betti reci- representation at point 1L E R 3 (Bonnet, 1999), with r,, = 1 (JL E fl) or"= 0 (JL \i fl)•
procity theorem:

r [(g' .T.J.)."IJ.' _ (g 11).T.!. J ds = Jnr Lf.2.ji, -t.'.[\ J

.fen
2
.
1 1 2
dfl =o (4.80) (4.84)

·To derive the Boundary Integral Equations, the basic idea is to chose a particular
elastostatic state (Jl.2 , q_2 , f 2 ) corresponding to a si1nple reference problem (e.g. point
load in an infinite spac~)- The solutions of this proble1n are called fundamental solutions
and will be used to solve Eq.(4.80). For time-harmonic problems, The Maxwell-Betti
theoren1 leads to a similar equation. In the transient case) we shall nevertheless see
several differences afterwards (§4.6.9)
The integral representation forn1ula thus allo>.vs the deter1nination of the solution at
any interior pointy only knuvving the solution at the medium boundary 80. A co1nparison
vvith the Finite Elernent 211Iethod is proposed in Fig.4.46. As sho>.VIl in Fig.4.46c, the
Rolution Virill only be deterrnined at the model boundary.
Integral equation

4.6.3 Integral equations in elastodynamics Let y denote a fixed point on the boundary surface 80. As depicted in Fig.4.47 1 for
a gi\-:;en srnall c > 0, >.ve introduce a spherical shaped neighbourhood vs(Y) of y 1 called
Equation of propagation an cxcluHion neighbourhood (Dangla et al., 2005; Guiggiani et al., 1992.)~ The-do1na.in
n, (y) = n - v, (y) obtained by removing v, (y) from n is such that the point y is exterior
\Ale >.vill now consider isotropic elastodynamics for tirr1e-har1nonic problems of circular
frequency w. For any given body force distribution fi(;r;_) over 0, the governing equations +
to <'4(ul· Its boundary is {)fl,= (Dfl - e,) 8" where e, ={)fl n ·u" s, = n-n OVe-
which must be verified by any displacement and stress fields, u,(.i;) and CTij (.i;) are•
CTiJ = Auk,koiJ + µ( ui,.i + u,;,,;,) (4.81)
2
CTij,J + pw ui + fi = 0 (4.82)
where n'i,.i and CT(i,.i are condensed notations for the spatial derivative 8 / 8x,;.
300 J'viodelling wave propagation 4.6 The Boundary Element Method 301

(a) mechanical problem The classical form of the integral equation consists in taking the lirnit c _____, O in the
representation formula (4.84) taken for the domain '1 0 (Bonnet, 1999; Dangla et al.,
an 2005). The lirniting expret->r.;ion thus obtained is known as the Sornigliana identitu:

(4.85)

The notation f stands for the Cauchy principal value of a singular integral, i.e. the
lin1it:

(b) finite element method

in () ~ ~il."/i Jan ce) ()
(4.86)

The free term Cik(11) appearing in Eq.(4.85), is defined by:

(4.87)

It is found to be equal to l/26iJ:: \Vhen n is srnooth at 1/_·

(c) boundary integral equation method

an an-e
\ '•, an '
integra]'·· .....:----.... ,......-/
representation ~ in\erior•
formula points

(D.)

Fig. 4.47: Exclusion neighbourhood considered to apply the limiting process

and derive the integral equations (Bonnet, 1999).

4.6.4 Discretization and regularization principle

Fig. 4.46: Modelling a mechanical problem (a) by the Finite Element Method
(b), the Boundary Integral Equation Method (c) and the Boundary Element
Method (d).
Regularization of Boundary I11tegral Equations
In contrast to other discretization n1cthods) the Boundary Elcn1cnt 1\.1ethod involves sin-
gular integrals (Bonnet, 1999; Dangla et a.l., 2005). There are three different kinds of
singularities: weak singularity) strong singularity or hyper-singularity. Strong and hyper-
singular integrals have to be converted into regular ones in the regularL~ation of the BEJ'vf
formulations (Tanaka et a.l., 1994; Sladek and Sladek, 1996, 1998). From the point of
vie\v of nu1nerical integrations) one should devote great attention to the evaluation of
these integrals because standard integration quadratures fail in accuracy (Lachat and
302 J'viodclling \o.1a:ve propagation 4.6 The Boundary Element Method 303

Watson, 1976; Sladek et al., 1997, 2001; Manolis and Besk<m, 1988). Therefore each type interior problem exterior problem
of singularity has to be treated by appropriate techniques. ------- ''
''\
Most of the researches has dealt with strong (Bonnet and Bui, 1993) and hyper- \
\
singnlaritics. Son1c n1ethods have been proposed.in the literature to treat thet'le singular \
integrals (Sladek and Sladek, 1998; Niu and Zhou, 2004; Guiggiani, 1994; Guiggiani et I
I
al., 1992; Chen and Hong, 1999; Bonnet, 1999; Bui et al., 1985; Aubry and Clouteau, I
I
1991). It is notable that the regularization can be perforrr1ed either before or after the I
I
diHcretization, i.e. in the global or local (intrinsic) coordinate spacc 1 as obHerved in sornc I SR
I
papers mentioned above. J\ comprehensive review of I3EI'vl in dynarnic analysis has been I
I
proposed by Beskos (Beskos, 1997). I
I
I

------ /

Analytical method
Fig. 4.48: Description of interior and exterior problems for the derivation of
Dangla et al. (2005) proposed an analytical regularization in.et.hod taking advantage the corresponding Boundary Integral Equations.
of an am~urncd thrcc-nodcd triangle elerr1ent for the discretization of three-di1nensional
proble1ns. Thanks to this siinple shaped elernent, one can perforrn analytical evaluations
of the Cauchy principal value of the singular integrals. Such an approach has been
correspond to the eigenvalues of a certain interior problem (Bonnet 1 1999). Such frequen-
previously applied for 2D elastodynamic problems in (Dangla, 1988, 1989). The efficiency
cies are called ,fictitious eigenval·ues and rnay be avoided using indirect BIE forrnulationR
and accuracy of such a regularization rnethod Vl.'ill be illustrated nu1nerically hereafter in
(Bonnet, 1999; Pedersen ct al., 1995).
3D ela.stodynan1ics.

4.6.5 Wave propagation in unbounded media Radiation conditions

Exterior problems Using the for1nulation of the exterior proble1n 1 it is possible to account exactly for the
conditions at infinity considering the exterior of a bounded dornain f2. The radiation
The Boundary Integral Equation (4.85) has been established for a bounded domain 0. conditions may be derived as follows (Bonnet, 1999):
\\Te rr11\y consider the co1nplen1entary to Cl denoted !le, and try to write the integral equa-
1

tions for such an exterior problcni. To do so 1 a bounded sphere ~2R such as n C nR c !le tp - iwpVp1lp = o(n-1)
is considered (Fig.4.48). Thanks to a limiting process (R--+ +oc), it is possible to gen- { ts - iwpVs1ls = o(R- 1 ) (4.89)
eralize the Boundary Integral Equation if there are no souTces at infinity.
for longitudinal and transverse waves respectively (tk being the stress vector cornponentR
'The conditions on the decay of the waveficld at infinity is then: for each wave type).

"li,p-11 = o(l) "li,p-I. = o(R- 1 ) As it will be shuwn aft.erV\rards, it is then possible to accurately rnodel wave propaga-
(4.88)
{ lis-11 = o(R- 1 ) lis-I. = o(l) tion in an unbou11ded don1ain considering only the boundary 8f1 of the bounded domain
fl It is particularly interesting to n1odcl scisrnic 1.vave propagation or vibrations in the
for longitudinal and transverse vvavcs respectively, I being the tangent vector such as
environrnent.
'Il·I = 0.

An expression on !l is then obtained (Bonnet, 1999) which is close to that given
in Eq.(4.85) but with normals directed outward (Fig.4.48). Considering exterior prob-
le1n forn1ulations 1 \Vave propagation in infinite or serni-infinite media can thus be easily
rnodelled by Boundary Integral Equations. However 1 the integral equations associated
to exterior problems may have a non-unique solution for sorne specific frequencies \Vhich
4.6.6 Numerical Implementation
Both boundary and unknownti are discretized and interpolation techniques initially de-
veloped for the Finite Element Method are considered (13onnet, 1999). Using for instance
three-noded fiat triangles, the discretization of the gcornetry and the unknowns is thus
304 J'viodelling >.vave propagation 4.6 The Boundary Element Method 305

written, respectively, as follows (Dangla et al., 2003):

;i;({) =
3

LN,({h:' a(:i;) = L
3
Nk ({)ak (4.90)
t
P(co)
k=l k=l

,
vvith ;fk: the node coordinates, Nk: the linear interpolation functions and ak: the nodal
values of the diHplaccrnent or traction unknowns.
'
0 R
'Thus, the Ret of scalar equationH resulting fron1 the discretization of eqnations (4.85))
enforced at the nodes of the rnesh) has the foll<nving n1atrix structure:

[A]{u} + [B]{t} = O (4 91)

where [AJ and [BJ are fully populated non sy1mnetric matrices: {u} and {t} are the 'vec-
tors) containing the nodal values of ui('jj_) and ti('fL), respectively. Fig. 4.49: Cavity under harmonic internal pressure: model description.

'The incorporation of the boundary conditions consists in substituting the prescribed

nodal values of (u,, t,) into {{ u }; {t}} in Eq. (4.91 ). The columns of this matrix equation
arc reordered so as to have a rna.trix equation of the forrn: The validation of the numerical results is rnatle by conRidering the analytical solution
in terms of radial displacement u(r, w) given by Eringen and Suhubi (1975) as follows:
[K]{u} = {!} (4 92) P(w)R 3 (ikr - 1/r) exp (ikp(r - R))
·u(r,w)= (4.93)
where the vector {v} consists of the unknown components of {v.}, {t}. 4µr(l - ikrR - k~R 2 /4)

where kp and ks· are the longitudinal and transverse wavenurnbers.

The matrix [KJ contains the columns of [A], [BJ associated with those unknown com-
ponents while the right-hand side {f} results frorn the multiplication of the knovvn com- This equation may be re\vritten using norrnalized distance f = r /Fl, norrnalized fre-
ponents of {v.}, {t} by the corresponding columns of the matrices [A], [BJ. As shown in quency T)p = kpRf'rr (that is T)p = 2R/ Ap, Ar being the longitudinal wavelength) and
the follovving for unbounded n1cdia, the right hand side {f} n1ay also involve a contribu- 4p.u.(r.w)
tion due to an incident wavcficld. The results shov.rn hereafter have been computed by considering u(r, rJP ) = ( ). . It leads to:
PwR
Dangla et al. (2005) using the computer code CESAR.-LCPC (Humbert et al., 2005) of
the Laboratoire Central des Ponts et ChaussCcs (French Public \iVorks R,esearch Labora- (irrrr7r -1) exp(iKr;p(i' - 1))
tory, Paris, France). (4.94)
(1- im7p - Jr 2 ( 2 r7~/4)r 2

4.6.7 Validation and influence of the regularization with ( = ks/kp = .j(2 - 2v)/(1- 2u) = V3 (for u=0.25).

Spherical cavity under harmonic internal pressure
The first cxa1nple (Fig.4.49) concerns a spherical cavity of radius R in a full elastic
isotropic space undergoing an internal harn1onic pressure (Chaillat et al., 2008). The
cavity rnesh includes 320 triangular boundary elements (that is 162 nodes) and a special
generation process is considered to have a regular triangular mesh of the sphere starting
from an icosahedron (Edouard et al., 1996). Using the regularization method proposed
by Dangla et a.l. (2005), the displacement field around the cavity is computed at various
(norn1alized) frequencies.
Comparisons between numerical and analytical results
In Fig.4.50, the real part of the normalized radial displacement u(r, 7/P) defined by equa-
tion (4.94) is displayed vs norrnalized distance r for both analytical and nurr1erical solu-
tions at norrnalized frequencies 77p=O.Ol, 0.50, 1.00 and 2.00. For the nearly Rtatic case
(17p=O.Ol) aR well as the full dynamic cases, the agreernent bet\veen the numerical results
and the analytical ones is very good at all norrnalized <listanceH. From this first sin1plc
exarnple, the reliability and accuracy of the proposed inethod for analyL;ing propagation
in unbounded rnedia appear very good.
306 l\!Iodclling 1.vave propagation 4.6 The Boundary Element Method 307

0.25
2000) pern1its to v.rrite the Helmholtz Green's function as:
analytical numerical
I ::i
µ
'1
Q)
0.20

-----· 11,=0.50
11,=0.01 •••••
00000
cxp(ikllL - £1) _ ik 1.
•
IJ!... - _±
I - in1
471" L-+=
i F._EB
e
ikf.:fg (-·
L Q,
)
ro e
-ikf.yd~
- ii., (4.95)
sQ)
0.15
-·-·-·- 11,=1.00 •••••
""' 0.10
\
\
., . . . . . . . . 11, =2.00 where S is the unit sphere of R.3 and the transfer .function 9L (§.; z:0 ) is defined by:
"a00
·~

"""
\
\
., gL (§_; ro) = L (21 + 1)i' h)') ( k Ira l)I'e (cos(§_, ro)) (4.96)
"""N
Q)

·~
0.05
' \Q ~-<>·¢".A ,.~·
-·-· A.A
OS_lS_L

Oil . \ ·"' v.. ,,J' .o/·V V-·¢-~

0.0 ,,--------'.<l.-<:>------------~;•---------c~-----------
s <>-<>-<>· ()'
in terms of the Legendre polyno1nials Pe and of the spherical Hankel functions of the
first kind h~ ). r.rhen, the elastodyna1nic funda1nental solution (4.83) is easily seen to
;., ·.:... ¢•'6l, _,•' 1
0 Q.~ 4'"' Q .!It
'1 """' ''O .......
-0.05 admit representations of the form (4.95) with gL replaced with suitably-defined (tensor)
..... ...._ __ .... _.v'C>.... -0.... ,..,_g ....
..0---
-0_ .o_ ....0- ...0- transfer functions U~L and T,~L (Cha.illat et al., 2008).
-0.10 L-~-~-~-~-~~-===--'-~-~-_J
1.0 1.2 1.4 1.6 1.8 2.0 2.6 2.8 3.0
Formula. (4.95) with L = O(kslz: - z:0 1) is suitably accurate whenever Ii: - z:0 l/li:ol <;
normalized distance r 2/../5 (Darve, 2000). Guided by this consideration, a 3D cubic grid of linear spacing d

Fig. 4.50: Normalized radial displacement fi.(f,77p) (real part) vs normalized

embedding the Virhole boundary an
is introduced. The above condition is then auto1nat-
ical.ly satisfied whenever _± and y__ belong to non-adjacent cubic cells, the poles ;r.0 and 'l!_o
distance F: comparison between numerical and analytical results for normal-
being chosen as the cell centers, vrhile it rnay be violated otherwise. The one-level Fl\!LJVI
ized frequencies 71p=O.Ol, 0.50, LOO and 2.00 (Dangla et al., 2005).
basically consists of using decomposition (4.95) whenever ;r_ and y belong to non-adjacent
cubic cells (Chaillat et al., 2008). -

4.6.8 Advanced formulation: the Fast Multipole Method
1~·he standard BElvI forrnulation leads to fully populated non symn1ctric rnatrices in
Eq.(4.92). It is thus very difficult to model problems involving more than about 10 4
DOFs. It iR possible to overcome this difficulty by considering an alternative formulation
called the Fast Multipole Method (Fujiwara, 1998, 2000; Greengard et al., 1998).
As depicted in Fig.4.51, the interactions hetvveen remote points are con1puted once
for all at the centre of each cell. It leads to much 10-1,ver co1nputational costs and memory
requirements. If 22 and y_ belong to adjacent cells) the corresponding terrns are con1puted
by standard 13EM. Thanks to the FMM, it is then possible to compute problems involving
up to 10 6 DOFs on a single-processor computer (Chaillat et al., 2008).

The goal of the FJVIJ\.11 is to speed up the matrix-vector product con1putation required Standard BEM
for each iteration of the iterative solver applied to the BEJVI-discretized equations arising
from (4.92). 1\!Ioreover, the governing matrix is never explicitly formed, which leads to a
y,
storage requirement well below the O(N2 ) memory required for holding the whole BEM
1natrix. The FlvI-accelerated BE:l'vI therefore achieves substantial savings in both CPU
time and memory (Chaillat et al., 2008).

The main idea behind the Fl\11\!I is to reformulate the fun<larnental .solutions in terms
of products of functions of .:.r:. and of y. This allovvs to reuse integrations with respect Fast Multipole BEM
to y "'\Vhen the collocation point ;r_ is -changed) a strategy which is not feasible in the
traditional BEM, based on expression (4.83). Such fundamental solutions may be written
as linear con1binations of the Green)s function for the Helmholt;;, equation) for which such Fig. 4.51: Comparison between the classical BEM and the Fast Multipole
a reformulation is kn(J\.VIl from earlier works) e.g. (Nishirnura) 2002; Darve, 2000). The method.
position vector i: = 11-;i: can be decomposed as i: = (£0-£)+(110 -£0)-(110 -11) = i+z:o-ji,
vvhere ±o and }!_0 are two poles. Application of the Gegenbauer addition theorern (Darve,
808 J'vio<lelling 1.vavc propagation 4.7 Applications to \Vave propagation in soil 309

4.6.9 Elastodynamics in time domain results computed by the Boundary Elcrnent l\1ethod arc represented by sy1nboh~ (Sern-
For transient problcrns, the I3E.J'vI forrnulation rnay also be written in the tin1e do1nain blat and Dangla, 2005). Since this nu1nerical approach involves an exact description of
(Schanz and Antes, 1997; Takahashi et aL, 2003). Considering the scalar vvavc equation the radiation conditions, the analytical and rnunerical results are in very good agrecrncnt.
in time domain (Bonnet, 1999):
Such a comparison \vould be n1ore difficult vvith the Finite E1en1ent lVIethod since it
1 \vould raise the need for specific nurnerical t<?chniques such as absorbing boundaries or
6.1, - -ii+ F = 0 V(y, t) E rl x [O, T] (4.97) absorbing layers (see §4.5). The validation in co1nplcx configurations (geo1netry, charac-
c2
teristics of the excitation, etc) llH\Y thus Le difficult \Vith such rnethods.
For transient problerns, it is also necessary tu account for the initial conditions:
------ analytical solution o o o o o numerical solution
u(y, 0) = u 0 (y) and u(y, 0) = vo(y) (4.98) 1.8..----------, 1.4~-------~
1.0
The reciprocity theorem thus takes the follo-\ving form:

f [(q 1 * u 2 )(y, t) - (q 2 * u1 )(y, t)]dSy =

Jon
.l [(F 2 * u 1 )(y, t) - (F 1 * u 2 )(y, t)]drly
8 10 12 14

Jnr c~:, [1!;i(y)u

(4 99)
..ti
2.0~-------~ 0.86~-------ft-,~

+ 1
(y, t) - v6(y)u 2 (y, t)]dily ~
<:! 1.6
Q)

s \
8~180°

;f-s, 0.82
8=0°

,.
,.., G'
£/"

~
+ r 1
Jn c
, [u6(y)1i.'(y, t) - n6(y)1i2(y, t)]dfly
Q)
'-' 1.2 \ I
I '
f5
~
" 0.8
"a
\
(\)
\
I
I
p 0.7
f5
where q =\Ju.Tl_ and 11
* 11
denotes the convolution product.
\
'.,¢ 0.7
Gf5
0.4
6 8 10 12 14 6 8 10 12 14
T·his expression involves convolution products, instead of 1nultiplica.tions in the f're- distance distance
quency don1ain 1 and includes two additional integrals when con1parcd to Eq.(4.80) (J'viax-
\Vcll-Betti thcor<?1n in elastostatics or tirne-harmonic elasto<lyna1nics). To deal with these Fig. 4.52: Diffraction of a plane SH-wave by a cylindrical cavity: comparison
convolution products, Schanz and Antes (1997) propoRed a BEJ'vI forrnulation in the between analytical (dotted) and numerical results (symbols), (Semblat and
Laplace domain. rrhcy also developpcd a convolution quadrature method to approximate Dang la, 2005).
the convolution integrals numerically. Their forrnulations have also been generalized to
viscoclastodynarnics (Gaul and Schanz 1 1999).

3D unbounded medium
4. 7 Applications to wave propagation in soil
The second exa1nple deals Vilith the diffraction, by a spherical cavity, of a plane P---;,vave
4.7.1 Diffraction of a plane wave in unbounded media propagating along x axis and defined by:

2D unbounded medium !line= Uo exp [i(wt - krx)] 1'x with U0 = 1

The diffraction of a plane vvavc by a cylindrical cavity 1nay be fully characterized through
theoretical approaches. For such a problem, the cornpa.rison \vith the Boundary Elernent
1\ilcthod will evidence the capabilities of the method to 1nodel wave propagation in un-
bounded dornains (Sen1blat and Dangla, 2005). In the case of a plane SH-\vave, the
wavefield calculated analytically (Eringcn and Suhuhi, 1975) is displayed in Fig.4.52
(dotted line) for various radial directions around the cavity. In Fig.4.52 1 the 11u1nerical
The numerical results are firstly computed in frequency domain and con1pared with an-
alytical results (Dangla et al., 2005). They arc afterv;lards converted into time dornain
to characterize the scattered \Vaveficld.
The vvavefield con1puted around the cavity for various directions iR shovvn in Fig.4.53.
The boundary ele1ncnt rnesh of the cavity (2562 nodes) is generated the sarne way as in
310 1v1odelling wave propagation 4. 7 Applications to wave propagation in soil 311

the case of the pressurized cavity (Edouard et al., 1996). However, this mesh has been
refined since the wave:ficld has strong variations especially at gra:ting incidences. The
analytical solution in tern1s of radial displacernent Ur is given by Pao and Jlvfow (1973)
as well as Eringen and Suhnbi (1975).

The resultt.; are cornputcd for various azirnuthes ((Ji = ( i - 1) x 45°, 1 :s; i :s; 5) and
Fig.4.53 displays the real part of the radial displacement vs normalized distance r = r / R
(1 :S r :S 3) at two different normafocd frequencies Tfp=l.00 and Tfp=2.00 (Dangla et
al., 2005). The analytical results are plotted with lines (dotted for 1)p=l.ll0 and wlid
for 1)p=2.00) and the numerical results with symbols (circles for T/p=l.00 and bullets
for 71p=2.00). ·The agreernent Uctvvcen the numerical and analytical results is very good
for all azirnuthes at 17p=l.OO. For () 3 = 90°, some slight differences can be noticed at
17p=2.00 near the cavity wall. This is probably due to the fact that there is a grazing
incidence at this point.

In Fig.4.54, the numerical results arc also displayed as displacen1ent isovalucs on the
cavity \Vall and on an horit-Jontal area embedded in the domain. It can be noticed that)
at the grazing incidence, the interferences bet\veen the wavefield and the cavity are very
strong.

0 0 0 0 0 •••••
~,=1.00 (analyt.) ~,=1.00 (BEM) ~,=2.00 (analyt.) ~,=2.00(BEM)

0.2---------- o.s,--------....,
.µ 0.8 8;;;;;135° • 0=90°
~ OA
d
P.. 0.4 ~;i
w
'8 0
0
r;Jpi
o.., 0

""
~ -0.4
~ -0.8
• 0
<>oo
-0.3
-0.4
eo.
°'oeo-o .cl
o'

1.0--1-.5--2-.0--2-.5--3.0,1.0 1.5 ..--'--t7-.0"---- 2.5 3.j?' 1.0_ _1_._5__2_._0__2_._5_ _3.0

r """ 0=90° / r
2 0=135° : 8=45° 8;;;;;QO
',
'
:' , //

---',--,---;,!. __

..,
,. ' I / '-.
0 8=180° /' ',I.-" ' 8=0° 0
~ --------0 R -'---?
-1
...rl -1
-2 u

1.0 1.5 2.0 2.5 3.0 plane cavity ~ . , t"' 1.0 1.5 2.0 2.5 3.0
norm. distance x norm. distance X Fig. 4.54: Diffraction of a plane wave by a spherical cavity: wavefield com-
wave --"VV\.Alo- z
puted by the BEM for two normalized frequencies ~p=0.25 (top) and T/p=0.50
(bottom).
Fig. 4.53: Diffraction of a plane wave by a spherical cavity: comparison with
analytical results for various azimuthes at normalized frequencies 7/p=l.00
and ~p=2.00 (Dangla et al., 2005).
312 l\Joclelling wave propagation 4.7 Applications to 1.vave propagation in soil 313

4.7.2 Vibrations of a foundation

The exan1ple depicted in F'ig.4.55 deals \vith the analysis of the wavefield generated by a
vibrating foundation. The analysis is performed by the Finite Elcn1cnt 1\11.ethod and the
features of the rnodcl arc the following:

• a circular foundation (radius lrn) is sub1nitted to a dynan1ic vertical load whose

variations are given in Fig.4.55,

• the foundation being circular, the finite element rnodcl is axisy1nmetrical (radhu.;
8rn).

• the model includes 2600 quadrilateral elernents and the dynamic cornputa.tions ®
involves 100 tirnc steps (unconditionally stable/non dissipative Ne\vmark tilne-
integration sche1ne).
Rayleigh

The results displayed in Fig.4.56 correspond to two different times T1 (left) and T2
(right). Displacen1ent isovalucs arc given at the top and the corresponding wavcficld
is dra1vn as vectors at the bottorn. At tirne T2, the isovalue diagrams clea.rly shovv the
pressure waves propagating faster than the shear 1vave (V:s < Vp). In Fig.4.56 1 the results
in terms of displacement wavefield (bottom right) show that, as predicted by the theory
of surface waves in Chapter 3, the rnotion describes an ellipse close to the free-surface.
T'liis type of rnotion is due to R,ayleigh Viravcs the amplitude of Vi.rhich decreases rapidly
with depth.

F(t)

®
Fig. 4.56: Wavefield generated in the soil by the vibration of a foundation
(Finite Element model): displacement isovalues (top) and displacement wave-
field (bottom).

t
)
4.7.3 Vibration isolation using piles or trenches

As discussed in Chapter 1, vibratory isolation iH a crucial issue since annoyances due to

Sm +--+ traffic induced or construction works generated vibrations 1nay be strong. In Chapter 2,
lm FEM model (axisymm.) 1.ve analyzed the influence of a rigid layer (en1bcd<lcd in an infinite mediun1) on the
propagation and trarunniHHion of a plane wave. V\Te shall ncnv try to study numerically
Fig. 4.55: Vibration of a foundation and idealization of the problem. the efficiency of cylindrical inclusions c1nbeddecl in a free-space (2D model). The 3D case
will also be investigated for piles and trenches. 1'he inain goal is to assess and illustrate
the efficiency of various types of isolation t>yHtems.
314 J\!Iodelling \Vave propagation 4. 7 Applications to ¥.rave propagation in soil 315

Influence of cylindrical piles/holes 7/3 = 0.140, r74 = 0.166, 1)5 = 0.191. These results show that the efficiency of the isolation
systc1n is very srnall for 'T}1 and r72 (less than 12 to 253) and is rnuch better for higher
Various authors have studied the influence of piles as isolation barriers (Aviles and
frequencies. The area having the largest an1plitu<le reduction is located at the center of
Sanchez-Sesma, 1983; Liao and Sangrey, 1978). The piles may act as scatterers leading,
the isolation systern and strong eclge effects are found (1,vave scattered at the boundaries
if arranged adequately, to a reduction of the wave amplitude in the soil. The interest
of the isolation system). For 775, the lowest transn1ission coefficient is around 0.5 and the
of piles iH mainly in terrns of durability since systems involving trenches rr1ay be easily
geometrical arrangement in this case at this frequency appears to be very efficient. This
darr1aged.
frequency nearly corresponds to a quarter 1,vavelength configuration where the isolation
effect should be optimal (as shown in Chapter 3).
Description of the problem. In this paragraph) we shall consider a 2D 1nodel using
the Boundary Elernent Method. The influence of cylindrical holes on mechanical VI.rave
Several dirnensionless spacings) S /a) are now considered to asseHs the influence o:f this
propagation is investigated considering various configurations (Fig.4.57). The radius of
geornetrical pararneters on the efficiency of the iHolation system.
the cylindrical holes being a., the distance betV\reen two holes is denoted S' and various
Viilues of the dirnemionless spacing S/a arc corrnidered (8/a. = 0.5, 1, 2).

~0.587
0 ~ 0.6618

0 sx 00 0 ~0.7366
~ 0.8114
2alD 0 ; ; ; 0.8862
0 0 11111111 0. 961 0
0 0 0 111111111.0358
_ 1.1106
0 0 0 - 1.1854
+r 0 0 _ 1.2602
0 0 0 1.3350
0 0
0
0
S/a=0.5 S/a=l.O S/a=2.0

Fig. 4.57: Vibratory isolation with cylindrical inclusions and variable spac-
ings: model description.

Efficiency of the isolation system (2D). From the model depicted in Fig.4.57, var-
ious DEJ\11 sin1ulations vvere performed in the frequency dornain. A norrnali:6e<l frequency
17, related to the size of the cylindrical holes, is defined as follows:

2a
1J = - (4.100)
A
vvhere A is the v,ravelcngth in the Roil. Fig. 4.58: Vibratory isolation by cylindrical inclusions at various frequencies:
amplitude reduction factor at various normalized frequencies for a dimen-
The arnplitude reduction factor computed for a dimensionless spacing ~9/a=l is dis- sionless spacing S/a=l.
played in Fig.t1.58 for the :following normalized frequencies: 1]1 = 0.089) T/2 = 0.115,
316 lVIodclling wave propagation 4. 7 Applications to wave propagation in 8oil 317

As depicted in l-"'ig.4.59, at normali.~cd frequency r75=5.23 1 the transrnission coefficient - - 111 - - - - 112 ---------- 11:1 -·-·- T]4
is minin1um for S/a=0.5 (Tmin = 0.4), whereas it is never belovv 0.6 for S/a=l and 0.8 1.0l-----==========~~
for S / a=2. Since the nu111ber of cylindrical holes rcrnains consta.nt 1 it can also be noticed --- ---- ----------
in Fig.4.59 that the extension of the isolated area is smaller when the holes are closer.

The results are now displayed as a function of the distance to \the iSolation system
(Fig.4.GO). The amplitude progrest:>ively decreases beyond the isola\ion systcrn until it
reaches its minimum value. AfterVirards, far frorn the inclusions, . it inCrcascs until the
reductior1 factor reaches 1 (no influence of the isolation syste1n at long ciit'ltances). The S/a=0.5

1.0~'\'~-=:=~~~--=-=-==-=~=1
rnaxi1n11m efficiency is reached for S'/a=0.5 (top) vvith a reduction factor reaching 0.4
whereas it remains above 0.8 for 8/o.=2 (bottom).

The influence of frequency also appears in liig.4.60. ·The arnplitude reduction is larger ] 08 ~--~ -~-;;~-~-~-;.;:;;.~ =-~ ~-=-~------ ----------------
at higher frequencies since the radius to "\\ravelength ratio, a/ A, iH thus larger. The dis- ~ 0. 6 \ _.. / .,,.--=-:::.-----=----~- ---
tance at vvhich the maxilnurn arnplitude reduction iH reached also depends on frequency.
"' 0.4
S/a=l.O
1 : 0.4 / 2 : 0.5 I 3 ' 0.6
4: 0.7 / 5: 0.8 / 6' 0.9
.,...---··-········

0.4
) S/a=2.0
15 17.5 20 22.5
distance

Fig. 4.60: Amplitude reduction factor as a function of the distance to the

isolation system for various dirnensionless spacings (S'/a=0.5, S/a=l.O and
S/a=0.5; 11,=5.23 S/a=2.0) and different dimensionless frequencies (171 = 0.089, 172 = 0.115, 7)ci
S/a=l.O ; 11,=5.23 S/a=2.0 ; 11 0=5.23
0.140, 77'1 = 0.166, 1)5 = 0.191).
Fig. 4.59: Displacement isovalues for various dimensionless spacings S/a.

Efficiency of the isolation system in 3D. Kattis et al. (1999) also modelled an
isolation systcrn involving a row of cylindrical holes in the 3D configuration (free-space).
Using the Boundary Element l\!Iethod: they computed the surface displacement around
the isolation syste1n. As depicted in Fig.4.61, the efficiency of the system is good since
the arnplitude reduction factor is around 0.45 beyond the inclusions.
 T

318 lVIodelling 1vave propagation 4. 7 Applications to \Vavc propagation in soil 319

1.333
1.278
1.223
incident
D on
D

--
1.167
1.112
R-wave
1.057
1.001
0.946
0.891
0.835
0.780
0.725
0.669 0.0 2.0
0.614 x
0.559 1.5 1.5
0.503 0.98
0.448 1.0 1.0

0.5 0.5

Fig. 4.61: 3D wavefield scattered by a row of 8 cylindrical holes: numerical

BEM results from Kattis et al. (1999). I » 0.0 I » 0.0

-0.5 -0.5

For stiff 3D piles of square croSR-section, Gao ct al. (2006) modelled the propagation -1.0 -1.0 0.98
0.98
of Rayleigh waves by the Boundary Integral Equation Method. As shown in Fig.4.62
-1.5 -1.5
(top left)) they considered various configurations for the isolation syste1n involving from -0.5 0.0 0.5 1.0 1.5 2.0 -0.5 0.0 0.5 1.0 1.5 2.0
one to three pile rows. Their nurnerical results arc di.splayed in Fig.4.62 for each config- x x
uration. For the single pile row, the reduction factor is around 0.94 and the system is
actually not efficient. For t"\vo rows of piles, the reduction factor if; nearly 0.60 and the Fig. 4.62: Vibration isolation by pile rows: schematic of the problem and
efficiency is satisfactory. Finally1 for three pile rows, the arnplitude is reduced by t\vo runplitude reduction factor fo various configurations involving 1 to 3 rows
thirds and the influence of cornbincd pile rows appears to be beneficial. from Gao et al. ( 2006).

Vibration isolation using trenches • half-circle trench: dimensionless trench depth H/ An 0.596 and circle radius
R/ An = 0.596;
As depicted in Fig.4.6:), t\.vo problems involving trenches as v1rave barriers are considered:
a half-circle trench (left) and a rectangular trench (right). As shown in Chapter 1, such
• rectangular trench: dimensionless trench depth H /An = 1.19, trench width L/ An =
cases have already been studied experimentally by Richart et al. (1970). Some 2D 1. 79 and distance to the source R/ An = 2.97.
computations were performed by Segal et al. (1978). Banerjee et a.l. (1988) carried
out 3D sirnulations using the Boundary Element l\!Iethod. Their numerical results a.re
where An is the Rayleigh wavelength.
displayed in Fig.4.63 (bottom) and compared 2 to the experimental results from Woods
(1968). The folluvving geometrical pararneters arc chosen for the trench:
As shown in Fig.4.63, the BEl\i[ solutions predict the amplitude reduction factor
2 It
reasonably well. As discussed by Banerjee ei al. (1988), the lack of symmetry of the ex-
should be noticed that the experimental results are displayed using a single isovalue scale whereas
numerical results from Banerjee et al. (1988) correspond to (liiferent scales for the half-circle and the perirnental results reflects the infiuence of the inhomogeneous properties of the actual soil.
rectangul1:tr trenches.
320 1\.ilodelling VI.rave propagation 4. 7 Applications to wave propagation in soil 321

Ho\\rever 1 for the half-circle trench, the n1ain isolated area computed by the DEJ'vI is
found to be similar to that found experirncntally: it is bouncletl by tvvo radial lines at 45
degrees fron1 each encl of the trench.
experimental results, from (Woods, 1968)
4. 7.4 Traffic induced vibrations in railway tunnels
Dangla and van Hoove (1988) analyzed the influence of traffic induced vibrations in
raih..vay tunnels. Considering a single tunnel, the numerical results co1nputcd Ly the
Boundary Elernent J'vfetho(l are displayed in Fig.4.64 in terrr1H of displacernent iRovalues
(Sen1blat and Dangla, 2005). As shu\vn by these results 1 the Holution is strongly depen-
dent on the frequency content of the loading.

f=5Hz f=lOHz f=l5Hz

2.2 ;

f=20Hz f=25Hz f=40Hz

i--
j
numerical results, from (Banerjee et al., 1988)
0
L LI
Fig. 4.64: Traffic induced vibrations in a railway tunnel: displacement iso-
valt1es at various frequencies computed by the Boundary Element Method
(Semblat and Dangla, 2005).
111111 >1.5
111111 0.6-1.5
D o.3-o.s Dangla and van .Hoove (1988) al::~o perforn1ecl sorne experiments on site using a dy-
D 0.25-o.3 namic loading device called Dynaplaqne. rrhey n1casurcd the vertical displacernent at
D <0.25 the top of the hill above the raihvay tunnel. So1nc interesting comparisons bctvveen their
experim.ental and nun1erical results are proposed in Fig.4.65 (Semblat and Dangla. 1 2005).

Fig. 4.63: Vibration isolation by trenches: comparison between experimental

results from Woods (1968, top) and 3D BEM results obtained by Banerjee
et al. (1988, bottom).
:J22 J\iiodelling Viravc propagation

Chapter 5

0.01 - - - - - - - - - - - - - - - - - - - - - - - . . ,
Seismic wave propagation and
- - - - experiments
------model 1
amplification
············ model 2

I
I 5.1 Introduction
I
,-,
I ,. •'I
'I
I

I ' I'
T·he seismic ground inotion is strongly influenced by the source and pa.th effects, but
I : I ~
I : /
\ ! I '
propagation and an1plification of scisrnic ViraveR in surficial soil layers is alRo a key is-
I • .- •• ,\
!,• '.i. ,' I \ : I sue (Boore, 2004). As discussed in Chapter 3, the reflection and refraction of scis1nic
.r
' \
\•
,' I
,' I
\
\ •
:
",,,.
: I
I
... . waves at the interface bctvveen layers having different velocities may lead to larger an1-
:
,' I
'/
\'. '/
..:. :,
\
. . . ,,,"! :
~
plitudcs at the free-surface (Bard and Bouchon, 1985; Chavez-Garcia et al., 1999, 2000;
.'/ v "
o.o"-~-----'---_,_---------~ ' l{awase, 2003; J\!Ioeen-Vaziri and Trifunac, 1988; Pitilakis et al., 1999a; Scinchez-Sesma
0. 10.0 20.0 and Luz6n, 1995). Hence, it is necessary tu estimate, through cxpcrirnental as well as
frequency (Hz) numerical approaches, the arnplification factor at a specific site and the corresponding
frequency range. Characterizing the seismic ground rnotion is mandatory to assess the
Fig. 4.65: Vertical displacement induced by dynan1ic loadings in a railway tun- seismic risk in a particular area.
nel: comparisons between numerical and experimental results, from (Dangla
and van Hoove, 1988). As it 1vill be shown in this Chapter, the amplification factor rnay be estilnatcd ex-
perimentally through various tcchniquets. The analysis may also be performed through
different nurncrical approaches investigating the propagation process in complex geologi-
cal structures such as alluvial deposits. The Spectral Element 1V1cthod and the Boundary
Element Jlviethod are very useful techniques for rr1odelling seismic wave propagation and
amplification. They allow the analysis of various earthquake scenarios and the assess-
ment of the scisrnic hazard at a specific site.

Simple an<l extensive experimental n1cthods 1 such aH the m.easurement of ambient

vibrations, are also available. They give very interesting results allovving a simplified
characterization of alluvial deposits. Frorr1 a theoretical point of vicnv, it is also possible
to consider simplified approaches Huch as the modal rnethod to be discussed herein. It
leads to a fast and reliable eHtimation of the fundarnental frequency of sediincntary basinH.
Finally, the characterization of the soil properties directly on site or in the laboratory is
324 Scis1nic vvave propagation and a1nplification 5.2 Scitirnic VI.rave arnplification 825

also a key issue to quantify~ sei~nnic ground rnotion in alluvial deposits. lt is especially the PGA=l 70 cm/s'
case fur strong events for which variour.; (anelastic) constitutive pararneters arc needed.

SCT
"'j
5.2

5.2.1
Seismic wave amplification

Main governing phenomena

·-
-170
PGA=35 em/s'

UNAM

Site effects: some examples

"'~
-170

The propagation of seismic \Vaves in surficial geological r.;tructu.Tcs n1ay often lead to
strong mot.ion amplifications due to the velocity contrast betv.reen alluvial deposits and
the bedrock. Such phenorncna, nan1ed site effects 1 rnay considerably increase the irnpact
of an earthquake (Baise et al., 2003; Bard and Bouchon, 1980a,b, 1985; Kham et a.l.,
2002a; Raptakis et a.l., 2000; Sommerville, 1998). One of the first large earthquakes to
show the importance of this issue \Vas the 1'v1ichoacan 1985 earthquake in ?viexico (Fig.5.1)
(Fiih et al., 1994; Kawase and Aki, 1989; Singh et a.I., 1988). The maximum accelerntion
at Rixty kilometer8 frorn the epicentre V\ras around 0.28 g. It reached 0.2 g in the centre
of Mexico city located much farther ("!00 kilometers). As shown in Fig.5.l in terms of
velocity; the Peak Ground Velocity vn-18 even larger in the 1\!Iexico basin (SGT station)
than at the Carnpos station located nearby the epicenter. In the city ittielf, the inten8ity
variations v.rere very large. The thick clay deposit in the lake-bed area nearby the vol-
canic belt an1plified up to a factor 5 the seisrnic inotions \Vhen cornpared to the bedrock
(1naximum acceleration in time domain).
In the case of areas \Vith moderate t:>eismicity1 the site effect influence is also very
irr1portant. As depicted in Fig.5.2, for alpine valleyt:> 1 deep and narro\v alluvial deposits
may lead to co1nplex propagation patterns. In this figure, the 500n1 deep Grenoble batiin
(French alps) is schematized. The reference bedrock site is called OGNTU (top left) and
other stations are located at the surface of the deposit. 1'heRe stations arc part of the
French accelerornetric net-,,vork (RAP). T·he Grenoble basin strongly amplifies the seis-
mic rnotion due to rnultiple reflections and di:ffraction.':l at the basin edges (ChaljuL, 2006;
Delepine and Semblat 1 2006). An international numerical benclunark \Vas organized to
estimate seismic \Vave amplifica.tion in this alpine valley (Chaljub 1 2006).
Seisn1ic Vilave amplification may also occur in very large ba.sinti such as that of Los
Angeles, Fig.5.3 (Ko1natitsch et al., 2004). ThiH nurnerical 1nodel involving the Rpectral
elernent method allo\vs a detailed analysis of the arnplification process (rnain ainplifica-
tion areas, effect on duration, frequency range, etc) as well as of the source and path
effects. Other sites, such as the Ashigara valley in Japan, have also been extensively
studied (Kawase and Sato, 1992; Pitarka et al., 1994; Scherbaurn et al., 1994; Uetake
and Kudo, 2005; Zahradnik et al., 1994).
Campos Teacalco
Mexico
epicenter city
- 400 km
Fig. 5.1: Propagation of seismic waves for the Michoacan 1985 earthquake in
Mexico (adapted from P.Moczo, Comenius University, Bratislava).
Site effects and seismic risk
From these examples) it is apparent that the at:>Ressmcnt of seisn1ic wave arnplification is
a crucial isRue. It is not only nReful for the quantification of the seisrnic hazard but also
for seis1nic risk purpoties.
Such a coincidence occured during the 1967 Caracat:> earthquake which \Vas strongly
amplified by the alluvial deposit in the centre of the city (Duval et a.I., 1998). As it will
be discussed herein 1 the funda.rnental frequency of a basin is linked to the geometrical
configuration as vvell as the velocity contrast in the various soil layers. For Cara.cas, this
frequency was around 0.6 Hertz for the n1ost dam.agetl blocks (Delavaud, 2007; Sc1nblat
et al., 2002b). It has been shown that the three buildings that collapsed during the
earthquake \Vere structureR including more than 14 storeys leading to eigen frequencies
close to 0.6 Hertz. In the case \vhere the a1nplification of the seisn1ic rnotion inay be
large, it is then neceRsary to characterize the local seismic reHponse of the soil to scale
the featureR of the reference earthquake considered for the design of the buildings in this
area. The case of Caraca.':l will be discussed herein since it also involves a significant
influence of the topography in the propagation/ an1plification process.
326 Seisrnic 1,va-ve propagation and a1nplification 5.2 Seisrnic 1.vave arnplification 327

5.8 5.85 119 00'

5.65 5.7 5.75 118 30' 118 00' 117 30'W
~-~------IGJ.!::;;;!;;:=~A•===-~-,------~----,

. 01£ .... • ~:·. LKL

A ry

34" 30'N

s~
TA2

~:,~
CHF
-.,,-._
45.2 I 45.2
B S

L~~ SPF

34" OO'N 34· OO'N

45.15 45.15

Il c

33"30'N ~--~-------~------~----'----~----..L 33'30'1'1

119" OO'W 118" 30'W 11a· OO'W 11T30'W
100secs.

Jan. 1999 Laffrey Earthquake

45.1 (Grenoble valley) 45.1 Fig. 5.3: Seismic wave amplification in the Los Angeles area: numerical sim-
ulations by the Spectral Element Method (Komatitsch et al., 2004).

contrast betvveen the soil layers (vertical heterogeneities) rnodifies the rnotion amplifi-
cation at the !Yee-surface. As depicted in Fig.5.4 (left), it is mainly a lD effect. The
km geo1netry of the deposit is also a governing factor (Fig.5.4, right). It may be character-
45.05 45.05 ized Ly an average depth or by a detailed forrn for a sediinentary basin. In the latter
0 1 2
case (lateral heterogeneities), the Heismic VI.raves are trapped in the basin leading to a
epicenter larger arnplification than thTongh the lD assumption. In the case vvhere only the av-

5.65 5.7 5.75

"' 5.8 5.85
erage depth of the layers is conRidered (one-dimensional case), the arnplification factor
can be estimated theoretically (Chapter 3). The estimation of the seismic gruu11d mo-
tion in the codes is rnainly perforrned through such simplified approaches (Pitilakis et
Fig. 5.2: Velocities (N-S component) recorded at various locations in the city al., 1999b). Whereas, when the lateral heterogeneities are large, it is necessary to in-
of Grenoble during the 1999 Laffrey earthquake (data: French accelerometric vestigate seisrnic VI.rave propagation in tvvo or three din1en.sions (Bard and Bouchon, 1985).
network, www-rap.obs.ujf-grenoble.fr, reference: (Lebrun et al., 2001)).

5.2.2 Experimental characterization

Site effects: lD, 2D or 3D analysis
To characterize seisn1ic wave arnplification experimentally, it is necessary to perforn1
As already shown in Chapter 3, the amplification of seismic 1vaves in an alluvial deposit rneasuren1ents of the ground n1otion at the considered alluvial site, denoted 3(i), and to
is strongly influenced by the mechanical properties of the layers. Indeed, the velocity compare it to unamplified rnutions. It is thus mandatory to choose a reference stat'ion,

I
ii
328 Scis1nic vvave propagation and ainplification 5.3 Seis1nic ¥.rave arnplification in layered 1nc:dia 329

spectral
ratio
frequencY
2D
,.-----~./
i R(f) Fourier
Fourier ,l·\
s'"(r)
Fig. 5.4: Seismic wave amplification: main phenomena in lD (left) and 2D/3D
(right) geological structures.
~spectrum spectrum ,..• -................... .
frequency frequency

denoted R, located in a quiet area free of topographic irregularities as Vi'ell as sedimentary

deposits (e.g., an outcropping bedrock). The 1ncasuremcnt perforn1ed at R is then used aB
j r(t) reference signals
s"'(t)
signal at site s"
~14---~
a reference to which the signals measured at 3(i) can be con1pared. Figure 5.5 describes
(i~1,2J
the methodology for such analyses: the Fourier spectra of the time signals r(t) at 11 time
and s(i)(t) at S(i) arc co1nputcd in order to cstin1ated the s1:te/referencc spectral ratio
3(i) (!) / ll(f). When the reference station is well chosen, the site/reference ratio allows
the assessrnent of the a1nplification phenomena in the frequency don1ain (Theodulidis
and Bard, 1995). One can either characterize the amplification due to a relief (Fig.5.5, ~~
site 3(1)) or that due to an alluvial deposit (Fig.5.5, site 3l 2 l). As shown by Chavez- 0 ~ ---------------

Garcfa et al. (2002), the choice of the reference bedrock station is crucial to estirnate the
an1plification factor.
The n1ea8urements performed at R and 3('i,) rnay involve the horiiontal co1nponents
of the ground motion or the vertical ones. For three-dimensional iT1easure1nents, three
different site/reference spectral ratios are hence available: (3£'l (!)/Rx(!), 3~i) (!)/Ry(!) Fig. 5.5: Amplification of the seismic motion through site/reference spectral
ratios.
and s£il (!)IR"(!)) allowing the analysis of the ground motion amplification for each of
these components.

ConRidering the definitions given in Fig.5.5, 'UA corresponds to the ground 1notion at
5.3 Seismic wave amplification in layered media the alluvial site S wherea8 UB' is the reference rnotion at the reference site R.

5.3.1 From transfer function to time-domain response Since Ruch transfer functions correspond to complex-valued spectral ratios, it is pos-
As discussed in Chapter 3, in the case of a har1nonic vvave 1 the amplification of the seis1nic sible to consider the ti1ne-don1ain amplification through inverse Fourier transforn1s. If
we consider the tirne-domain response at the surface of the layer, namely uA(t), it may
motion in a single layer mediun1 inay be quantified through a lD transfer function:
be related to the response at fl' through the follo¥.ring relation:
(5.1)
UA(t) = _1_
2K -=
J+. = T(w)u[dw)eiwtdw = _1_
2K -=
j+= (cosp1h+ixt->1np·
u];,(w). . ) ciwtdw
1h
(5 2)
wfh
\vhere: kz. =- The una1nplified wavcfield 'IJB1(t) corresponds to the double of the incident vvavefield
i Vsi:
u.inc (due to the free-surface condition). Knovving the transfer frn1ction f(w) and the
/I1p1 cos 81 Fourier transfor1T1 of the reference rnotion u1p(w) (or) alternatively, the incident n1otion
and: x= ----
µ2p2 cos 82 if the free-surface effectfl arc rernoved), it is thnR possible to compute the amplified
330 Seismic \Vave propagation and amplification 5.3 Scis1nic wave amplification in layered rnedia 331

wavcfield u,1(t) through Eq.(5.2). As discussed hereafter, the time-dom'1in analysis of

2
the amplification process may be pcrfor1nc<l directly using this 111cthod. incident
1.5
Ricker
5.3.2 Amplification in single-layered media 11s 0.5
wavelet
As already studied in Chapter 3 (Fig.3.31), a single layer medium is considered first. It '"
u Of----<

corresponds to a 201n deep layer having a shear Vi.rave velocity Vs=200 m/s, rnasH density "'
P..
w
-0.5
p=2 t/m3 . A vertically incident plane S.H-\vave is considered. The ground response is ~ -1
computed using expression (5.2) and considering a 2nd order Ricker vvavelet 1 with cen- -1.5
tral frequency JR, as the incident vvavcfield (Hee definition in Chapter 4 and Appendix B).
-2~====:=;:==::::::::===:::::::===:::::::====:;
2

1.5
Ricker wavelet at maximum amplification
§
The central frequency of the Ricker wavclct 1 fn, is first chosen at the fundamental fre- s 0.5

quency of the layer: JR= .fs- This case correponds to the rnaxi1num arnplification. As '"
u Of----~.

already proposed in Chapter 3, three different layer/bedrock velocity ratios {x1, X2, X3} "'w
P, -0.5

are considered (sec values in Fig.3.31). rrhe incident ¥/avefield is displayed in Fig.5.6 ~ -1

(top) together with the amplified \Vavcficlds for each velocity ratio. For the largest ve- -1.5

locity ratio x1, the maxirnun1 arnplitude iH increased hy a factor of 2. ThiH value inay be
compared to the maxirnun1 spectral amplification (around 6) obtained, in Chapter 3, at -2 ~====:===::::::::====:::===:::::======:;
2
the fundamental frequency of the layer (Fig.3.31). Even if the central frequency of the 1.5
Ricker wavelet coincides \Vith the funda111cntal frequency of the layer, the time-don1ain <J
arnplification is much lo-vver than the spectral one. This is due to the fact that the Ricker s
Q)
0.5
¥.ravelet has a short duration \vhereaH the spectral analysis assumes sinusoidal signals vvith Q)

infinite durations. This difference is thus crucial v,rhen comparing spectral a1nplifications
iil 0
P, -0.5
and tin1c-dornain amplification. w
~ -1
-1.5
Comparing the an1plificd signals displayed in Fig.5.6, it Hhould he noticed that the
infiuence of the velocity ratio is huge: the lo\ver the velocity ratio the smaller the am- -2 ~====::::::===::::::::===:::::::===:::::::===
2
plification. These results also sho¥.r that the amplification proceRs in time-domain leads
1.5
to different signal durations. Finally, the tin1e-dornain an1plification is 1nuch lov,rer than
the Hpectral one but the effect in tirne-don1ain also concerns the rnotion duration. In §
the following, such conclusions Virill also sta.nd for actual seisrnic rnotions in geological s
Q)
0.5

structures. Beauval et al. (2003) for instance evidenced the ground motion lengthening u Of----~·

at the Volvi E·uroseistest site. "'

P..
w
-0.5

~ -1

Ricker wavelet at fR = f.,/2 -1.5

Another Ricker wavelet is now considered at a lo\vcr central frequency: JR = f 8 /2. ~~-----~-----~-----~-----~-----~
1.0 20 3.0 4.0 5.0 6.0
As shown in Fig.3.31 (Chapter 3) 1 the spectral arnplification is not rnaxin1um at this time (s)
frequency and the tirne-dornain amplification should follow the same trend. The lower
frequency Ricker wavelet is displayed in Fig.5. 7 together with the amplified signals. Com-
pared to Fig.5.6, the amplification in the tin1e-dornain is significantly lower and the du-
Fig. 5.6: Amplification of a Ricker wavelet (!R J,) in a single-layered
medium for various velocity contrasts (x1, x2, X.3)·
ration is increased less than in the previous ca.Re. A lower spectral arnplification thus
lead to a lower tirne-dornain arnplifieation and a Hmaller duration increase. i
332 Seismic vvave propagation and amplification 5.4 An1plification due to the topography 333

2
5.3.3 Amplification in multi-layered media
incident
1.5
Ricker As discussed in Chapter 3, the case of rnulti-layercd 1ncdia may be easily derived fron1
1:l
Q) wavelet the case of single-layered media. The transfer function between layers j and k is easily
s
Q)
0.5 expressed as:
u 0
aj T k(w) = fk(w) + Yk(w) (5.3)
0. -0.5 J, fj(w) + YJ(w)
00
-~ -1
"" -1.5
If the time-don1ain n1otion is knovvn in one layer j) it is then pom.,ihle to cornpute
-2
the time-donrain response in any other layer k using the transfer function bct-\veen layers
j and k, Tj_k, and the Fourier transform of the motion in j, nj(w) (see Eq.(5.2)). The
2
analysis 111ay be perforn1ed for any kind of seis1nic rnotion. Ho\vever, as it 1,vill be discussed
1.5
+-'
in the follovving, the an1plification estimated frorn lD analyses is often rnuch lo\ver than
<I the actual one due to specific 2D or 3D effects (Fig.5.4).
Q)

s
CJ
0.5
u 0
aj
0.. -0.5 5.4 Amplification due to the topography
00
-~ -1
"" -1.5 5.4.l Main phenomena and simplified analysis
-2
Topographical site effects
2

1.5 As evidenced for various actual sites (lVla et aL 1 2007; Paolucci, 2002) 1 the influence of
+-'
<I topography on seisrrlic vvave propagation rna.y be significant. It is the case for crests,
CJ
s
CJ
0.5 hills, canyonR and any type of steep topographic irregularity (Boore, 1972; Gaffet and
u 0 13ouchon, 1989; Kouoh-Billc et al., 1991). As depicted in Fig.5.8, a simple interpretation
~
" -0.5 of topographic site effects rnay be to consider the conRtructive interferences due to various
"'
00
'8 -1 wave paths focusing a.t the sa1ne point. The an1plification proceRs ma.:y then be influenced
-1.5 by different parameters: the \\rave type (through \\rave conversion effects), the geornetry
-2 (narnely the aspect ratio) and possibly the heterogeneities vvithin the profile.
2

1.5
+-'
<I
(J)

s 0.5 . . . . . . . . .+
/ I
"u 0 / I
"
~

-~

""
"'
00
-0.5

-1 ¥ ........................ 1 :I

-1.5 interferences :
-2 I
I
1.0 2.0 3.0 4.0 5.0 6.0
I
I
time (s) I

Fig. 5.7: Amplification of a Ricker wavelet (! 11 = f,/2) in a single-layered

medium for various velocity contrasts (x1, x2, x::i)· Fig. 5.8: Constructive interferences due to a simple topographic irregularity.

I
334 Seis1nic \Vave propagation and amplification 5.4 Arnplification due to the topography 335

Simplified analysis for 2D topographies By integration along the .r-axis, it is then possible to evaluate the displaccn1cnt field
produced by the total force distribution:
Basic principles. Bouchon (1985) proposed a simplified inethod to atJse.ss the influence
+= 1L/2
of 2D topographies on the scis1nic n1otion. In the case of plane SH-v,raves, he considered
1 1
a simplified 2D topography (x, z = ((x)) as depicted in Fig ..5.9. u(.x,z) =
2µ
i L
n=-=
L Q(xo,zo)kcxpikz,,lz-((xu)lexpikx,(x-xo)dxu
-L/2 Zn
(5.8)

11' z
11

- - - - - - " ' " " " " " " " " " " " " " .... , .......... """""""-----~

Fig. 5.9: Simplified topographic model proposed by Bouchon (1985).

Consiclering the notations given in Fig.5.9 1 the following time-harmonic displacement

field may be written:
u(x, z) =exp i(kxx + kzz) (54)
The resulting incident (shear) stress along the surface is then:
~--incident stress
(5.5)

Simplified method. The simplified method proposed by Bouchon (1985) consists in

finding the force diRtribution along the surface \Vhich cancels the incident strer:;s. The
displacement field produced at (x, z) by a line force Q(xo, zu) acting in the direction
normal to the (x, z) plane (i.e. SH-wave) can be written:

v(x,z;xo,zo)=
Q(xo-Zo)
:
4Kiµ
1+= l . .
-expikz(z-((.ro))expikx(x-xo)dk
-= k 2
(5.6)
residual stress
1 iteration
2 iterations

To transfor1n this integral into a discrete summation) Bouchon (1985) introduced a

periodicity in the surface shape: ((x + nL) = ((x) with L the periodicity interval and n "
:~
t~~o-g~~~a-p-h-ic_p_r_o_f_il-e----------------
an integer. It yields:
.. .............
Q(xo. zo) ~ l . .. ··l>
v(x,z;xo,zo) = . ·L L. kexpikxJz-((xo)lcxpikx"(x-xo) (5.7)
2 iµ, n=-= 'Zn
Fig. 5.10: Geometry of the topographic profile considered (bottom), final sur-
face displacement (top) and evolution of the residual surface stress (middle)
as investigated by Bouchon (1985).
336 Seitnnic \lii'avc propagation and arnplification 5.4 An1plification due to the topography 337

Discrctizing the s11rface (interval ,6.;r) 1 the discrete Fourier tranHform over the x coordi- Ah = ah/ ah, ff for the hori7'ontal one (v,rherc aJ and a.i,ff denote the jth acceleration corn-
nate arises and the solution may thus be estimated by inverse discrete :BOurier transform: poncnt at each point of the ground surface and ff the similar variable for the free-field
case). The topography aggravation factors computed by Bouckovalas and Papadi1nitriou
M M l (2005) are displayed in Fig.5.11. They show that the amplitude is reduced at the bottom
L Clm L ~expikzJz-((rnL'>x)I
11
of the slope and is enhanced at its top.
rn=-Ai v=-}d (5.9)

x exp ( i 27f
JVpj ) exp ( 'i 211"
Nprn )
r--- ~
0
2.0

1.5
(slope 10°)
::r::
T'l1e linear systern of equations (5.9) leads to the force distribution vvhich cancels the
stress produced by the incident "\Vavefield along the surface. A fast iterative procedure
"';§"
u
1.0
Ah
~,..,,.__~
: v: "."
•

was also proposed hy Bouchon (1985). (-) ,;c-B---f (+) "~ 0.5

x=O 0.0
Approximate solution. The iterative procedure proposed by Bouchon (1985) consistH 2.0
in choosing the follo-\ving force distribution as a preliminary assun1ption: § [slope 45°) (slope 90°)
1.5
·.g
Qm = 2y't.7n + 1 aom (5.10) u ~jA,{~
;·
-
~s 0.5
1.0 .~ ~· - ~ ~~-' ~

vvhere aom denotes the surface incident stress at Xrn. and trn is the surface slope.
" A.
0.0 ic-::C:::±e:.,__:::..r::::::;:~::::::id
This distribution creates a stress discontinuity across the surface vvhich cancels the -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
incident t>tress field. I-Iowever, the radiated stress field yieldt> a non-zero resulting stress distance from crest (km) distance from crest (km) distance from crest (km)
at the surface. Equation (5.9) is then used to evaJuatc the residual stress a1"" a.t the
surface points. Hence, a new force distribution is defined using these values: Fig. 5.11: Amplification of seismic wave around a crest for various inclinations
of the topography (Bouckovalas and Papadimitriou, 2005).
(5.11)

The new ret>idual stresH is evaluated and the process is continued until it bccorncs Douckovalas and Papadin1itriou (2005) perfor1ned extensive paran1ctric studies to as-
negligible. 'I'he resulting displacement field being cornputed 1 the effect of surface period- sess the infiuence of such topographies on the seismic ground motion. Frorn these results,
icity may be removed using the method proposed by Douchon and Aki (1977). they derived the follcnving expression for the an1plification of the horizontal con1ponent:

Some results con1puted by Bouchon (1985) using this iterative procedure are displayed (5 12)
in Fig.5.10. The residual stress for each iteration is given: it is nearly zero at the third
iteration. Bouchon (1985) also investigated the case of a triangular profile for \Vhich the
where functions F, G 1 H and J arc chosen as:
residual stress it> found to be large around the <liscontinuitieH of the surface slope.

5.4.2 Amplification by crests and hills

Amplification at a hill crest G(J) = I' + 216
I3 + 0.02 (5.13)
For step-like slope topographies, so1ne analyses "\Vere perforrnc<l in the late 6(J1s in the case
of earthbanks (Idriss et a.l., 1967). Recently Douckovalas and Papadimitriou (2005) stud- 1
ied in great details such topographies in order to propose code provisions. As depicted
H(~) = 1+09~
in Fig.5.11, they computed the seismic ground rnotion around step-like slope topogra- J(N) = 0.225
phies for various geon1etrical features. They assessed the effect of the slope in terms of
topography aggravation factors defined as A.v = a11 /av,ff for the vertical component and where H is the crest height, A the vvavelength, I= i/90 the norrnalized incidence, ~ the
338 Scis1ni.c wave propagation and amplification 5.4 An1plification clue to the topography 339

damping ratio in the soil and 1V the number of significant excitation cycles.

In (Bouckovalas and Papadimitriou, 2005), the amplification of the vertical component

Av,max is analyzed as '1.rell. For both components, the variations of Ah,niax and Av,rna:i:
are displayed in Fig.5.12 vs tvvo para1nctcrs: the diincnsionless height H /A and the
normalized incidence I= i/90. Bouckovalas and Papadimitriou (2005) compared their ------·-- ---~

predictions to the code provi.sions from European EC-8 and frorn the French PS-92. 0.1
T-11cy concluded that the 20 to 40o/c. increase of the horizontal ground rnotion et'ltimated J\ m= 2: 1.10
numerically is similar to the provisions of both codes. Nevertheless 1 their numerical (for H/A. 2: 0.03 and i 2: 10°)
results lead to a significant parasitic vertical component (F'ig.5.12) Virhich is not accounted
------·-···-- ··-··· ---·····--·····--······---········---·····-·····---···-····
for in the scisrr1ic codes. Douckovalas and Papadimitriou (2005) also proposed distances
:Erorn the crest to the free-field that are larger than in the codes.
0. 01 LLL~LLL~LLLLLLLLLLLLLLL...LL.LLLLJ..LlLLL..J...LLLLJl..Lu...LJ

Amplification by a hill
1 '"""'""""'
Approximate fundamental frequency. Paolucci (2002) proposed a simplified method
using the Rayleigh approximation to estin1ate the fundarnental frequency of hills defined
by two different slopes (Fig.5.13). This method will be presented later in the case of
alluvial basins.

For steep hills and plane SH-waves, the fundarnental frequency as derived by Paolucci 0.1
(2002) has the following form:
J\ .mox ;> 0.10
fo = _ Vs sin(01 + 82)
0 38
R ) r/(o,
1 + ln 2 --2
+e,)] 2 ln(n2 /R .)
J (5.14)
(for H / A 2 10 -o.4(1-i 1o36))
· L cos 82 . [ ( Ri (Rz/ Rr)2 - 1

It n1ay be expressed as a function of the sloper:; on both sides of the hill, nan1ely II/ Li 0.01 ~~~~~LLL~~LLI.~~~LLLLLLLLI..LLl.LLl..LJ

and II/ L2. From thifi expression: Paolucci (2002) found that, for aspect ratios beloV\r 0.5 1
0 0.2 0.4 0.6 0.8
the fundarnental frequency may be approximated as follows: I=i/90°
- Vs Fig. 5.12: Aggravation factors for the horizontal (top) and vertical (bottom)
fo ::= 0.7- (5.15)
L motion components with respect to the topographical features: normalized
In the case of plane SV-waves, the approximate method proposed by Paolucci (2002) height H/A and slope I (Bouckovalas and Papadimitriou, 2005).
yields the funda1nental frequency as:

Jo = 0 38 Vs sin(81 + 82) / "

Through this formula, Paolucci (2002) found that the fundamental frequency in the
· L cosB2 V(Rz/R 1 ) 2 -1
SV case is 303 larger than in the SH case. As also demonstrated by G6li et al. (1988),
x {(1<2 +1) (3+~) (81+82) (5.16)
these theoretical results show that the fundamental frequency for both SH and SV-waves
corresponds to 1vavelengths that are significantly longer than the hill 1vidth. Paolucci

µ 1
+ ( 1 + ") [ (1< 2 -1) sin2B +sin2B + 1<(cos2B2 - cos2B1)
2
2
l}- 12
/
(2002) also studied actual steep topographies and one of these examples will be shown
later ori.

1 R2 Simulations for a triangular hill. As shown in Fig.5.8 1 the arnplification by a hill

vvhere: K=---ln-
Or + 82 Rr may be due to the focusing of various upcon1ing waves. In Chapter 4, we considered
340 SeiHn1ic >..vavc propagation an<l a1nplification 5.4 Arnplification due to the topography 341

''
'' ~ 2.0

H R, "
~ 1.0
P<
s
"'
L, L,
L

Fig. 5.13: Geometrical description of the theoretical topography proposed by

Paolucci (2002).

a triangular hill to discuss the accuracy of numerical rncthods (Fig.4.25). The sarnc
exarnplc is IlO\V used to analyze seis1nic wave scattering (vertically incident plane \Vave) 0.1
by a hill in terrns of topographic site effect. The displacement field iH co1nputcd by the
o.os ti.me (s)
finite elen1ent rnethod and displayed in :F'ig.5.14 vs tirnc and distance. The arnplification
00
is rnaximum at the hill top. At other locations, the radiatc<l vvavefields lead to longer
rnotion duration. Conversely, a slight arnplitude decrease rnay be noticed at the base
of the hill. As previously shown, the interaction depends on the ratio bct1.veen the hill Fig. 5.14: Amplification of the seismic motion at the surface of a topographic
dimensions and the IO.ravelength. Sin1ilar effects \Vill be sho\vn later in realistic cases. irregularity: FEM results.

Gcli et al. (1988) studied the influence of the topography combined with th'1t of a soil
layering along the slope (i.e. ridge subsurface layering). 1'he forrr1er lead to significant a
amplification/dcarnplification phenomena (simila.r to that depicted in Fig.5.14) whereas
the latter may genernlly be neglected.

5.4.3 Amplification by canyons

The case of canyons is close to the case of alluvial valleys. Topographic effects are
nevertheless don1inating since the Roil is generally assu1ned hornogcneous. However, as
it will be discussed at the encl of thiH Hection, such purely geornetrical effects may be
combined vvith the influence of the velocity conHtrast.

2D canyons Fig. 5.15: Triangular canyon considered by S:inchez-Sesma and Rosenblueth

Triangular canyons. SAnchez-Sesrna and Rosenblueth (1979) studied the response of
car1yons of arbitrary shape under plane \vaves (Fig.5.15). They formulated the problem
in ternJ.R of a. Fredholrn integral equation and 80lved it numerically. They analyzed
the influence of the angle of incidence () as well a8 the canyon slope h/a. 'Itiangular
and cosine-shaped canyons \Vere considered. sanchez-Sesrna and Ror.;enblueth (1979)
discussed the results for given values of the norrnalized frequency TJ defined as:
ka 2a
rJ=-;=1: (5.17)
(1979) to analyze the scattering of plane SH-waves.
where k is the wavenun1ber, a is the canyon half-vvi<lth and A the wavelength.
As depicted in Fig.5.16 in the case of a triangular canyon at normalized frequency
17 = 0.5, larger incidences lead to stronger arnplifications on the upstream part of the
canyon. Conversely, the arnplification on the left part of the canyon is lower for increasing
incidence angles.
342 Seisrnic wave propagation and a.1nplification 5.4 Amplification due to the topography 343

3.0 , - - - - - - - - - - - - - - - - - - - - - - - -

3.0 -:.... \ ,.-----~-..... :::':-. /""-.>-"·-----.

____ 1 / : /
/.

.
' \ 8=0° ,. ..... •• I/ I.
.· .. - '
\

I •
\
\
·················7!........ . ,:'
,' \
······· :-...,... ····· ..
: ,.. ··············
I • I I ······· '
···...... \'\ \
'1'1J
s . I
I \
I I
I
\'.\ h=0.4a 1/ , / \

.
2.0 / ;• \ /,:
I I
·c, ' , 0.53a ; / .,..,'/
/
8=90° ··.~·
Q)

""' I
I ;
I ------ / \

'
\'\.:.: -~-~.~//
~ , I I
<'< '. I \
w
'8 ' .......... x, ,_ -: \\
\
1.0
\2a!
..............

o.oL_--~2L.0_ ___.1_ _ _ _ 1L.O_ _L__o~.-o-__L_ _1L.0---1_ _2~.0-- 0.0 L_--::":---'----c-L__L__L__L__L__L__JL__

-2.0 -1.0 0.0 1.0 2.0
normalized distance x/a normalized distance xla
Fig. 5.16: Surface motion for a triangular canyon (slope 45°) and an oblique Fig. 5.17: Surface motion for a triangular canyon and various slopes h/a
SH-wave, Sanchez-Sesma and Rosenblueth (1979). (plane vertical SH-wave), from Sanchez-Sesma and Rosenblueth (1979).

vvhere H~ ) is the Hankel function of the second kind and, denoting

2
,Jk the Bessel functions
Another interesting issue is the location of the maxirnum arnplification: it is around of the first kind, the follovving expressions:
x/a = 1.75 for 0 = 0, x/a = 1.0 for 0 = 45 (tha.t is the canyon slope) and x/a = 0.75
.!1 (ka)
for 0 = 90. For plane SH-\vaves, the n1aximnrn arnplification may thus be located along ao = -2 (2 )
the free-surface or along the canyon slope. A.s it will be seen in the follo\\ring, the time- H 1 (ka.)
don1ain analysis leadH to 1nore complicated results. (5 19)
,. . ka.Jo(ka) - .11 (ka)
bO = ' i i Slll "'( 2
kaH6 l(ka) -Hi')(ka)
In Fig.5.17, the influence of the canyon Rlopc h/ a is no-\v studied. The wavelength is
kept constant (A= 5h) and a vertical incidence is considered. For the steepest slope, a and for n f 0:
strong dcan1plification of the ground rnotion is found at the bottom of the canyon. The
n1axirnu1n a1nplification is above 1.25 (free-surface effects being removed) and is located n ka.hn-l (ka.) - 2nhn(ka.)
an= - 4( - l) cos 2Tl/'('--=-~-'-----'---'--
near the basins edges. kaHJ~-l (ka) - 2nHl~ (ka)
(5.20)
bn = 4i(-1)" sin(2n + l)cv kahn(ka) - (2n + l)hn+1 (ka)
'ka.HJ~(ka)- (2n+l)HJ;,l+ 1 (ka)
Circular canyons. For t1V~o-dilnensional canyons and plane S~FJ--..vaves, Trifunac (1973)
derived the closed-form solution in terms of Hankel and Bessel functions. He obtained After determination of these parameters, the total wavefield n1ay be deterrnined:
the scattered \Vavefield under the follo\ving forrn:
(5 21)
= where 1t; and u~ are the incident and the refiected wavefields respectively.
u: = 2..)anHl;} (kr) cos 2n8 + bnHl~+l (kr) sin(2n + 1)8] (5.18)
n=O
344 Seis1nic 1.vave propagation and an1plification
From this closed-form solution, Trifunac (1973) studied the scattering of plane SH-
wavcs and conclu<lcd that the surface topography has prornincnt effects when the \Vave-
lengths are short co111pared to the radius of the canyon. He found that the amplification
factor is always ler.;s than tvvo.
3D canyons
Vertical plane waves. ~1D canyons are nov.r considered for various wave types. Dif-
ferent canonical :-3D rnudels were studied in the fra1ne\vork of the QS'IfA 1 research
project considering various nun1erical methods (see http: I /qsha. obs. ujf-grenoble. fr
for data and results). As depicted in Fig.5.18, the scattering of a plane P-\vave by a sen1i-
sphcrical canyon is first considered. The Fast J\1ultipole forn1ulation) for1nulated in the
framework of the Boundary Element Method (see Chapter 4), is used for the computa-
tions in the frequency domain (Chaillat et al., 2008).
semi ·spherical
canyon 4f yR
tI
plane
P-wave
IIIll
Fig. 5.18: Diffraction of a plane P-wave by a semi-spherical canyon: schematic
of the problem (Sanchez-Sesma, 1983).
In Fig.5.19, the FJ'vI-BEl'vI results) con1puted for a norrr1ali;.-;ed frequency rt = 0.50)
are compared to those published by Sanchez-Sesma (1983) and Reinoso et al. (1997).
They are displayed as a function of the distance along the canyon surface, denoted s,
normalized with respect to the canyon radius. T'l1c agreement bet1veen all theRe results in
tcrrns of both vertical and horizontal con1ponents is excellent. A. strong dca1nplificc_ttion
is observed at the centre of the canyon. A. slight a1nplification is found near the canyon
edges.
Oblique plane waves. The case of oblique plane 1vaves is now considered. Various au-
thors analyzed the scatteril1g of plane oblique waves due to irregular topographies such as
canyons (Eshraghi and Dravinski, 1989; Reino8o et al., 1997; l\!Iossessian and Dravinski,
1990). For an oblique P-vvave) Reinoso ct al. (1997) computed the surface dif.;placements
1 Quantitative Seismic Hazard ARRessrnent
5.4 Amplification due to the topography 345
FMM
o a Sanchez-Sesma
3 • • Reinoso
0
vertical
horizontal 1
--- --------· - . ij--ij· ""fjJ "-fSJ ·o--- -E>-· ··o--. ·o- ----
0 , __
0 2 3
abscissa s/R
Fig. 5.19: Diffraction of a plane P-wave by a semi-spherical canyon: compar-
ison of the FM-BEM results with previous results by Sanchez-Sesma (1983)
and Reinoso et al. (1997) at normalized frequency "i/p=0.50 (Chaillat et al.,
2007).
at a norrnalized frequency r;=O.fi. Their resultR are diRplayed in Fig.5.20 (top) in tern1s
of vertical and horizontal displacements. These curves arc strongly assynunetrical witl1
a larger vertical displace1ncnt on the left side (incident wave) but a 8Inaller horizontal
con1ponent due to lower vvavc conversion at this side.
For an oblique SV-wave, R,einoso et al. (1997) computed the surface displacen1ents
at the nor1nalized frequency 77=0.5. rrheir results are displayed in Fig.3.20 (bottorn) in
ter1ns of vertical and hori;,ontal displaccn1cnts. These curves are strongly assy1nmetrical
with a very sharp peak. The amplification of the vertical coruponent on the left side of
the canyon (incident wave) is larger than for an incident I'-vvave.
5.4.4 Amplification on actual topographies
lVfany different areas in the vvorld are affected Ly topographic site effects. In sorne ca888 1
sn1all cities are located in such areas and may suffer fron1 ground rnotion a.rnplification
(Paolucci, 2002) or induced landslirlcs (Dourcleau and Havenith, 2008). We shall recall
some results for two specific sites: CaracaR) Venezuela and a very steep site in Italy
(Ba.gnoregio). It will be al8o shown that topographic site effects rnay also influence the
stability of natural slopes such as in Las Colinas, Salvador (Bourdeau, 2005).
346 SeiHn1ic Virave propagation and amplification 5.1 Amplification due to the topography 347

I P-wave) f=0.25Hz
en 3.0 vertical Am,,=1.57
~
"
"
""s
0

+'
.:Q) 2.0
s
Q)

"
"'en ----
0. --
-~

"" 1.0 - .- .... horizontal

---
f=0.35Hz
- :
-------··- - ----- .. --
Am,,,=l.57

y/R
-- -· :
0.0
-3. -2. -1. 0. 1. 2. 3.
9.0

8.0
vertical [sv-wave)
7.0
en
~
"" 6.0
""s
0

+'
.:
Q)

s
Q) 4.0
"
~
"'w 3.0
"'
-~

"" 2.0
horizontal
-- -- -
1.0 --- - - --- -- -- -
---------- -- -- --- x/a -- -
0.0 --
-2. -1. 0. 1. 2.

Fig. 5.20: Surface motion for a semi-spherical canyon computed by Reinoso

Fig. 5.21: Topographic amplification of seismic waves around the city of Cara-
et al. (1997) for an oblique P-wave (top) and an oblique SV-wave (bottom)
cas, maximum runplification values given by Amax (vertical scale exaggerated
at the normalized frequency T/ = 0.5.
by a factor of 2).
348 Seitnnic wave propagation and arnplification 5.4 An1plification due to the topography

Amplification in Caracas

Description of the site. The city of Caracas 1 Venezuela, is founded on a very deep
alluvial basin surrounded by high mountains. The analysis of seismic site cffcct8 in Cara-
[
cas has been pcrforrned by various tcarns through experin1cntal (Duval et al. 1 1998) as 2.25
well as nurncrica.l approaches (Delavaud 1 2007; Papageorgiou and I<iln, 1991; Semblat
et al., 2002a 1b). Site effects in Caracas arc dorninated by the amplification in the deep
alluvial ba8in, whereas topographic effects are only significant at very lu\v frequencies
2.0

I1
I "
I I
I
I
,,
'
I \

(Scrnblat et al., 2002b). ""

·~
0
+-'
1.75
t\ 1' f
"'
u
4:1 1.5 I
f
\
I,
/
/ '-
'fI
"
I\
·~

°'s"' 1.25
I
,/ \ /
f \
- ,,....
\ I
I I
\I

Topographic amplification. The topography is modelled by the Boundary Element. -- I

1.0 -"------~~-:-··:···~. L
u ..... ·----------·-··-···-··-
1\Tethod considering a lOkn1 long geological profile. The topographic arnplification of
·~
..0
~
'-- - -...- '
"'
I
plane SIJ-v.raves \Vas cornputed for several lu\v frequency values: 0.25, 0.35, 0.45 and
""'
0.75 \/\.~--- . _..:··,,,
0.53 Hz. The numerical results arc displayed in Fig.5.21 all along the profile with a fixed OJJ
0
~ 0.5 point 1
" ,.. _.-,
',•
grey scale for the arnplification level. At very lo\v frequencies (f=0.25 Hz) 1 the topo- 0
graphic an1plification is pron1inent Vlhen con1pared to that due to the alluvial Lasin. 'I'he
+-'
point 2 :'"- -.
0.25
largest arnplification is found at the top of the highest inountain and reaches A1; 01JO=l.57, point 3
the free-surface effect being re1noved. For f=0.35 Hz, the an1plification is larger in the o'---~~~~~~~~~~~~'--~~~~~~~~~~~__J
""
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3
basin (out of scale) than along the topography. For the highest frequency (!=0.55 Hz),
the rnain topographic an1plification area is located at the top of a smaller and narrower frequency (Hz)
mountain, due to the shortest -i,vavelength in this case.
Fig. 5.22: Amplification and deamplification at different locations along the
topography around Caracas.

Comparison with previous results. In order to compare with previous reHults de-
rived for sirnplified geon1etries, \Ve selected three points along the topography (Fig.5.22):
at the top of the tvvo rnountains (points 1 and 2 resp.) and in bet-i,veen at the lowest
local height (point :J). The topographic amplification at these three points is displayed
in Fig.5.22 as a function of frequency. For point 3, the ground rnotion is generally <lea1n-
plified with values of Atopo ranging fron1 0.5 to 1.0. At the top of the 1nountainsi the
amplification is Rignificant since Atopo=l.64 at f=0.43Hz for point 1 and Atopo=2.12 at
f=0.78Hz for point 2. Paolucci (2002) abo found topographic amplification factors larger
than 2 for very steep sites.
'The present resultH are in agreen1ent with the conclusions of the canonical topogra-
phies: the ground motion is arnplified at the top of the mountains/crests vvhereas it is
deamplific<l at the botto1n locations. As it can be seen in Fig.5.22, the free-surface effect
being removed 1 the arnplification inay even be larger than 2 is sorne location8. Ncvcr-
thelest>, in the case of Caracas, the amplification factor in the basin is much larger than
the topographic effect (Se1nblat et al., 2002b). As concluded by experimental studies,
the maxirnurn a1nplification areas in the city of Caracas are located at places and for
frequency ranges where the buildings 1,vere n1ostly damaged during the 1967 earthquake
(Duval et al., 1998).
Amplification for steep narrow hills (Italy)
Description of the site. Rovelli et al. (2001) and Paolucci (2002) investigated to-
pographic site effectH for various steep italian sites. One of those is called (~ivita di
Bagnorcgio and is located on a steep isolated hill (Fig.5.23). The hill is about 110 m
high and -i,vafl hit by a strong earthquake in 1695 (local intensities ranging fron1 IX to X).
The enorTnouR damage in Ci vita rnay suggest a significant topographic an1plification. For
this site, Paolucci (2002) perfor1ned nurnerical simulations using the Spectral Elernent
Method.
Topographic amplification. For an E\iV crosR-section 1 the results computed by Pao-
lucci (2002) at four receivers are displayed in Fig.5.24: a R,icker wavelet vvas considered
for thnc-dornain analysis (top) and the spectral arnplification \Vas also estimated (inid-
dle). Ftorn the displace1nent tirne-histories, the topographic effect is found to increase
both the amplitude and the rnotion duration. In the frequency dornain, a pron1incnt
peak at about 1 Hz appears in all transfer functions at the hill top. Paolucci (211112)
co1npared this fundarnental frequency of the hill to the values computed by a sirnplified
350 Seismic wave propagation and amplification 5.5 A1nplification of seismic waves in 2D alluvial basins 351

:trntl :~r1 }Ntf1 tttif1

012345
time (s)
012345
time (s)
012345
time (s)
-012345
time (s)

~~ ~!~ ~~ ~~
0.0 1.0 2.0 3.0
frequency (Hz)
4.0 5.0 0.0 1.U 2.0 3.0

frequency (Hz)
4.0 5.0 0.0 1.0 2.0 3.0

frequency (Hz)
'1.0 5.0 00 1.0 2.0 30

frequency (Hz)
40 5.0

500

Fig. 5.23: Picture of Civita de Bagnoregio (Italy) where topographic site ] 450
effects were studied by Paolucci (2002). <:I
0
·~
+' 400

"'
>
Q)
~ 350 CD
Q)
111odal approach (this rnethod will be presented herein for alluvial basins).
300
0 100 200 300 400 500 600

Comparison with previous results. Paolucci (2002) computed the 3D topographic
amplification all around the hill as illustrated in Fig.5.25: the displacernent peaks from
the numerical analysis (EV\.T polarization) >Nith f~; =l Hz) are depicted with dots of
variable sil'Jc. As already noticed in the previous sections, the larger values occur at
the top of the hill. However, Paolucci (2002) showed that for other frequencies (e.g.
fp=2Hz) the a1nplification due to resonance is srnallcr, while that due to the convexity of
the profile beco1ncs prornincnt. Sirrrilarly to previonR rer.;ults 1 Dignificant dearnpli:fication
is found at the base of the hill. These results are in good agreen1ent with the theoretical
formula proposed by Paolucci (2002) for such concave wedges.
Effect on slope stability
The previous paragraphs only considered the amplification of the seismic motion at the
top of crests or hills. The infiuence of this phenornenon n1ay also induce instabilities
in slopeH. The analysis of both phcnornena was for instance perforrned by Bourdeau
(2005) for the Las Colinas site (Salvador) and by Bozzano et al. (2008) for the Salcito
distance (m)
Fig. 5.24: Analysis of topographic site effects in Civita de Bagnoregio (Italy):
results from Paolucci (2002) by the Spectral Element Method.
5.5 Amplification of seismic waves in 2D alluvial basins
Amplification of seisrnic 1vaves may be due to soil layering or topographic irregularities.
As previously depicted in Fig.5.4, the amplification process may strongly differ between
the lD (horizontiil layering) and the 2D/3D cases (Moczo and Bard, 1993). The influence
of dipping layers or alluvial basins having sin1ple geometrics will be discussed in this
section (2D) as well as in the following one (3D). The case of actual alluvial deposits will
be investigated later.
I
I

landslide (Italy). For the Las Colinas site, a large landslide was observed after the 2001 ''
earthquake. ·The numerical back-analysis of this earthquake (Bourdeau) 2005) evidenced 5.5.1 Amplification by wedges I',
the fact that the landslide v.ras induced by the a1npli.fication of the seisrr1ic n1otion at 11
the top of the slope. Such phenornena are very cornplex since they also depend on the Sorne analytical rcr::;ults \Vere derived by various authors for simple geometrics (S&.nchez- [i

nonlinear constitutive laws of the various inaterials) their initial stress state 1 their water Sesn1a, 1985). The case of semi-infinite layers is investigated firr::>t. Alluvial basins \.Vith
content) etc. finite gcon1etrical extents will be discussed in the next Rection.
3G2 Seisn1ic wave propagation and alnplification 5.,) Amplification of seiRm.ic -i,vaves in 2D alluvial basins 353

6.0

Peak amplification 5.5

0.5·0.75
0.75-1.0 5.0
1.0-1.25
• 1.25·1.5 1050
• > 1.5
::§
'1
0
·~

'"
>
QJ

al 1000
3.0

2.5

- 2.0

1.5
(Bourdeau, 2005)
1.0

Fig. 5.26: Topographic amplification at the Las Colinas site (Bourdeau, 2005).

Fig. 5.25: Analysis of topographic site effects in Ci vita de Bagnoregio (Italy):

...
results from Paolucci ( 2002).
... ... ... ... .. ....
image ... ...
Infinite dipping layer
nl~~.......... ·
...
Description and solution of the problem. S'1nchez-Scsma and Velazquez (1987)
for instance analyzed the case of au infinite dipping layer depicted in Fig.5.27. For
plane SH-waves, they derive<l, fro1u gcornctrical considerations, the solution under the
following form:
M
harrr1onic excitations: __'!!_ = L EAf-j (-1 )j exp( -ikx cos ej)
Vo j=O
(5.22)
Fig. 5.27: Seismic wave propagation in a dippir1g layer: description of the
transient excitations: __'!!_ =
Vo
tEA1-j(-l)jf
j=O
(t - T~
v8
COS(}j) problem (Sanchez-Sesma and Velazquez, 1987).

v,rhere ()j = [J\'-(~~;r)]K, A1 = .iV2l and En, is the Ncurnann factor defined as:
Spectral runplification. The .spectral amplification Vila.s co1nputed by S<i.nchc~-Ses1na
if n = 0 1 and Vela~quez (1987) for various angles 21Jv. The results displayed in Fig.5.28 for N =3)5, 7
(5.23)
if n ::> 0 lead to the following conclu.sions:

Dravinski and l\!Iosscssian (1987) al.so studied the scattering of various \.Vave types by • the first arnplification peak is located at a distance x = A for ]\T =3 and at x -:: : '. 2A
<lipping layers of arbitrary shape. for N~7 (A: wavelength),
354 Seismic wave propagation and amplification 5.5 Am plification of seismic waves in 20 alluvial basins 355

1 2 3 4 5

6 ....
.... ....'
... . '
:
\
:>o
--> 4 I
I
I
I 51-- - - - - -
I
: I
I
2

4 1------~

4 8 12 16 20
kx/n

Fig. 5.28: Seismic wave propagation in a d ipping laye r: amplificatio n for 31-- - - -
various angles 7f /2N (Sanchez-Sesma and Velazquez , 1987) .

• the spectral amplificat ion is larger (up to a facto r of 2) fo r smaller angle8, 2 •- - - -

• the m aximum amplification m ay be fou nd fa r away fro m the edge of the layer (e .g.
a i x = 6A for N =7).

Ii should nevert heless be noticed t hat the a bove spectral a mplification is comp uted for
a non at tenuating med ium. Sanche:i:-Sesma a nd Velazquez (1987) also p roposed some 1 1-- -
results in the frequency domain for various qualit y factor value8. The iufl.uence of a tten-
ua tion is strong since, in t his case, t he maximum amplification is reached close to t he
layer edge. T ime-domain results will now he investigated in t he damped case.
0 2 4 8 10 12
Time- domain amplification . Considering a 2nd order Ricker wavelet as the incident
wavefield (characteristic p eriod tp) , t he surface motion was computed at several points
F ig. 5.2 9: Se is mic w ave propagation in a dipping layer: t ime-d omain a m p li-
by Sa.n chez-Sesma and Velazquez (1987). The d ip ping angle iH 7f / 14 and t he layer is
fication for N=7 and Q=20 (Sa nchez- Ses m a a nd Velazq u ez, 198 7) .
assumed viscoelast ic wit h a quality fac tor Q = 20 . The locat ions of t hese points and the
computed time-hist ories are displayed in Fig.5.29.

From Fig.5.29, the amplification is found to be large near t he layer edge (point 2),
whereas t he amplit ud e at other lo cations is only slight y increa..'ied with respect Lo t he
in cident m otion . H owever. t he duration of the seismic m otion is much larger at points 3
ot 5. T hese effects are du~ io t he two-dimensional geometry of t.he layer aud may not be
fonnd in one-d imensional layer s.
Wall-layer syste m
As d epicted in F ig.5.30, t he wall-layer Hyst.ern correspond8 to a semi-infini te horizontal
layer (constant d ept h ) . It was firstly studied by Sanchez-Sesma ct al. (2000), hut recent
result s wer e al5o proposed by P aolucci and Mostabilini (2006) .
356 Seismic wave propagation and amplificatio11 5.5 Ampliflcatio11 of seismic waves in 2D alluv ial basins 357

w w
The transfer function derived by Sanchez-Sesma et al. (2000) has the following form: x

4 ~ (-1r ( 2 . nmz
V(w) = -
7r
L.,
2Tn + 1
n.,,. -w2 exp(-ikmx)
. )
n?m1- W2 cos-V:*
<;
(5 .24)
rn=O "

(2m + 1)7rVs Fig. 5.31: Seismic wave propagation in a rectangular basin: description of
wh er e: 12m= ----~ the problem (Bard and Bouchon, 1985; Rodriguez-Zuniga et al., 1995).
2H

x Fundamental frequency. Bard and Bouchon (1985) analyzed the ant i-plane cigen-
... modcs of the rectangular basin to estimate its modal features. For anti-plane motion ,
they derived cigcnmodes having the following form:

u(:i:, z, t) = cos(T'mz) siu [k,.(x + w)J exp(2i11Jt) (5.25)

(n + 1)7r
w here: kn = - - - -
2w
Fig. 5.30: Seismic wave propagation in a wall-layer system: description of the
2
and: Im= ( m + l )7r
problem (Sanchez-Sesma et al., 2000) . (5 .26)
2H
4 21·2
k2 + ~ 2 - _7r_.-
Paolucci and :Mostabilini (2006) computed the transfer funcLion for various values n Yrn - y:2
8
of the normali:ted d istance, x/ A, and the normalized frequency,/ / fo , where fo is the The resonant frequencies arc therefore (Bard and 13ouchon, 1985):
[undarnental frequency of the ·1D layer. They found t hat the peak responses occur close

Lo the layer edge at normalized distR.nces np to 0.5 or 0.7. The correspondiug frequencies
me slightly higher t han the lD fundamental frequency due to the lateral confiucrncnt or
the layer (Paolucci and Mostabilini, 2006). These authors also proposed some aggravation
factors for both the wedge and wall-layer cases. These aggravation factors were estimated
from response spectra with a maximum of about 1.5 for the most severe case ( Paolucci
and ~ostabilini, 2006).
5.5.2 Theoretical basins
Since all uvial deposits often have limited geometrical extent, theoretical cases in volviug
t wo edges will now be investigated. The amplification of seismic waves in such alluvial
basins lead s to complex propagation patterns clue to the influence of both edges.
Rectangular basin
Several authors analyzed the case of rcctangnlar basins considering plane SH-waves. As
depicted in Fig .5.31, the rect angular soft layer is crnbedded in a rigid half-space (depth
H , width 2w). Bard and Bouchon (1985) analyzed the modal features of s uch a basin
a nd Rodrfguez-Z1iiiiga et al. (1995) derived the exact transfer function.
= 1HV(2m + l )2 + (n + 1 )2 ~:
fn ,rn (5.27)
with: - v;.,
f H - 4H
The eigcufrequencies defined by Eq.(5.27) depcud on t he vertical aspect ratio Kv
H/w. For the same depth, the fundamental frequency of a narrower basin is t hen larger
than t hat of a wider ha.sin.
The r elated mode shapes are depicted in Fig.5.32 in terms of vertical interferences
(index m) and horizontal inter ferences (i ndex n). There are symmetric and antisymmetric
modes. Jn the case o r P or SV-wavcs, which corresponds Lo in-plane motion, the verLical
and horizontal displacements arc coupled and the eigenfrcquencies may not be expressed
i11 a simple form . Bard and Bouchon (1985) proposed to identify the dependence of the
fundamental frequency on the aspect ratio by identification from parametr ic studies on
sine-shaped basins. Choosing a basin half~width L equivalent to 2w, the identification
is very good for a nti-plane motion. For in-plane motion, Bard and Bouchon (1985)
considered a simi lar assumption and derived the following approximat ions:
(5.28)
358 Seii.;rnic wave propagation and amplification 5.G Amplification of seismic waves in 2D allnvia l basins 359

w w m=O m=l m=2

······················· .............................................:.-. ~_.:;.;:·=~-.

P -waves
n=O

···········-·············--· --············--····-··-··· n=l

SV-waves
Fig. 5 .32: Eigenmodes of a r ectangular basin as proposed by Bard and Bou-
chon (1985) .

JJV = f~v vl + (2.~H) 2

(G .29)

where Jf;. and Jfl are the l D fundamental frequencieH. SH-waves

From these relations, it can be noticed that the ratio between the 2D a nd th e lD Fig. 5.33: Basin modes for various wave types as proposed by Bard and
fundamental frequencies is larger for S V-waves. The influence of t he vertical aspect Bouchon (1985).
ratio K ,, = H / L is then larger for horizontally polari&ed waves (i.e. ve rtically incident
SV-waves) . As depict ed in Fig.5.33, the inflnence of the basin width is much lar ger in
the case of shear waves. From their wor k, Bard and 13ouchon (1985) defiued lwo differ-
respectively, () the wave incidence, and denoting:
ent ranges fo r the asp ect rat,io: for small as pect ratios (between 0 and 0.3), where JD
propagation and lateral propagation are combined, whereas for larger aspect8 ratios, 2D 27rVs2
resonance dominates. Wo = - -
4h
Wn = (2n + l )wo
Exact solution. For plane oblique SJI-waves in rect angular basins, Rodrfguez-Z{miga v.2
w=w 1 - ___.§_]_ sin2 ()
et al. (1995) derived t he exact so lut ion in t erms of harmonic displacement u(a:, z, w) : v2S:1 (5.31)
00 k' = __w_ _
U 4 { (-l )n 1
Uo =; L 2n + 1 w 2 - w2 x V.51 sin()
n =O n
(5 .30) k - Jw2 -w;.
[
- ·k'
w 2 e '· '" - w -
.? (
cos k 1a coskn:i; - 1.. sm
_ k 1a-sin
_- k,,x)]
- -coswz
-- } n - Vs2
n cos kna srn k,.a Vs 2
From these expressions, t he basin response m ay be computed. Rod rfguez-Z11.ii.iga et
where u 0 is the displacement at the ba.5e of the rectangular depos it, h t he basin depth, al. (1995) also analy~ed t he case of a cylind rical 3D basin having its vertical cross-section
a it s half-widt h , Vs, and V52 arc th e shear wave velocit ies in the bedrock a nd the basin, identical to Lhe previous 2D rectangular model (i.e. axisyrnrnetrical arom1d the vertical
360 Seismic wave propagation and aruplifica.tion 5.5 Amplification of seismic waves in 2D alluvial ha.'li ns 361

axis) . They found that the difference between the 2D and the 3D models is large at the
centre of the basin. Near the basin edge, the difference is much less.

Sine-shaped basins
Bard and Bouchon (1985) studied 2D sine-shaped basins and derived a simple relation
to discriminat e between 2D resonances and lD resonances of geological profiles:

(LII) " = _v_C_

0.65
v ---J (5.32)

Considering the velocity contrast between the basin and the bedrock, denoted C,,, as
well as the shape ratio H / l (H: depth; £: width), their empirical relation defines the
values (Cv, H / L) leading Lo 2D resonances and t hose leading to lD resonance cornbine<l
to lateral propagation (Fig.5.34). For a shape ratio of 1, the velocity contra.'lt should be
larger t han 2 to reach 2D resonance. Fig. 5 .35 : Cylindrical basins with variable horizontal shape ratios "'h = L/ H.
To derive this relation, Bard and Bouchon (1985) compared the motion p hRSe for var-
io us cases and defined as 20 resonances t he situations were the phase is constant along
a sufficient part of the basin surface. In such cases, t he 2D geometr y ha.<; a significant Otherwise, lD resonance dominates and lateral propagation due to the finite geomet-
effect on the modal features of the basin. rical extent of the basin is observed. Jn this case, the (undamental frequency of the basin
is not significantly .influence by its 2D geometry. :>Jevertheless, as it will be discnssed
in the following for actual shallow basins, the influence of lateral propagation on the
amplification level may be la,rge.

5.5.3 Cylindrical basins

0.8
Tn addition to the case of rectangular and sine-shaped basins, the influence of the basin
geometry will now be investigated in the case of 2]) circular basins (i.e. cylindrical with
2D r esonance a circular cross-section). The case of 3D theoretical basins will be discussed in the next
section.

As depicted iu F ig.5.35 (Lop left), we consider in this paragraph a homogeneous elas-

tic circular basin (half-width L equivalent to the depth H ) having a shear wave velocity
lD resonance Vs, laying over a homogeneous elastic half-space with Vs 2 > Vs 1 • The propagation of
+ lateral propagation plane vertical SH-waves is analyzed using the Boundary Element Method in the fre-
0.2 quency domain (Bonnet, 1999, Dangla, 1988). Similarly to previous studies, the analysis
is perfor med considering dimensionless variables such as the amplification factor A , the
velocity contrast (V:<>, > -J2Vs 1 herein) and finally the depth/wavelength ratio (i.e. H/ A)
which is, for a constanl depth, proportional to the freq uency.
2 3 4 5 6 7
velocity contrast The amplification factor at the free-surface is depicted in Fig.5.36 a.'> a fun ction of the
norma.li:ted distance x/ L) for different values of the depth /wavelength ratio H/ A. The
Fig. 5.34: Limit b etween lD and 2D resonances as proposed by Bard and maximum amplification factor is about l. 7 for normali11cd frequency values H / A= 0.34
Bouchon (1985). and 0.87. For these frequencies, the maximum amplification is found at the centre of the
ba.<;in. Conversely, for t he normalized frequency H / A= 0.62, the maximum amplification
3G2 Seismic wave propagation and amp lification 5.5 Amplification of se i.smic waves in 2D alluvial basins 363

,- - H/A=0.16; A=l.1 5 H/A=0.34; A=l.72

,/r-~\
I I \\
,'I \\ HJA=0.16
,' I \\
I I I I
1.50 I I I\ H//\=0.34
....
.,
0 /I
, I
\I \,
H//\=0.62
(.)

Jl I
/ I
I I"
\ \
i::: I
, I I I I H/A=0.87
.....
...,0 1.25 /.··1 -. ··'··.'.
I

ell
(.)
i: I ·. \ ";

-s
I I I
~
I
j I I ~
0. I , I
.\ . .. ~..... I I ( "/ '
..·>~,.- -
-'> · ·.. H/A=0.62; A=l.56 H/A=0.87; A=l.83
C<l
\-->. / ' \ /,' I j \\
1.00 ~4--~'"7--....-- \ I \ .; ,
\ -·;. I 1- '-- \" \ / /
\ ''- ' ... ,;<'\
',..,,/ ">~.----:.~~.. ~--:'.: - -.::--~·· ',//
0.75'--'--'--'---'----'--'-....__..__.__.__.__.___.___.__._....__..__.__.~~~~~~~~~~

-3L -2L -L 0 L 2L 3L
normalized distance x/L

Fig. 5.36: Amplification A at the s u rface of a circular basin for vari-

ous depth/ wavelength ratios: H f A=0. 16; A=l.15 - H f A=0.34; A= l. 72 -
H/A=0.62; A=l.56 - H/A=0.87; A=l.83. Fig. 5.37: Amplification of the seismic motion in a cylindrical basin for various
depth/wavelength ratios (H/ A) w ith A d e noting the maximu m amplificat ion
facto r .
appe~trn a L Lhe bas in edges . T hese resu!Ls m ay a lso be analyzed at depth. As shown
in F ig.5.37, t he maximum amplification values ar c reached at t he free-surface except
for t he normali;~ed frequency H/ A=0.87. In this case, t he ampl ification at the surface direction of propagalion of the wave a nd t he vertical d irection, in t he layer and the
reaches 1.7 (Fig.5.36) whereas the maximum is above 1.8 at dept h (Fig.5.37). For higher
frequencies, depending on the bas in geomeLry, focusing effects may occnr and mod ify the
1
bedrock respectively (8 is calculated frolll fh (Chapter 3)) .

motion amplification. The rnotiou amplification clue to the single layer soil is deLerm ined from Eq.(5.33)
using the :mme properties as fo r the cylindrical lmsin and consid criug two different t hick-
5.5.4 Cylindrical b a sin vs horizontally layered soil uesses:
In order Lo assess t he 20 effects in the ground moLion amplification, the nu merical r esults • the layer t hickness H1 is identical to t he maximum depth of the cy lindrical basin,
for a cylindrical basin arc compared to the case o f a si ngle layer soil. As aheady shown in t hat is H1 = H ,
C hapLer 3, for a plane S H -wave, t he amplification in a single layer soil can be quantified
t heoretically t h rough t he following t ransfer function: • the layer thickness Hi is smaller t han t,he maximum depLh of the cylindrical basin,
that is Ht = H/ ./2.
- 1
!T1 2(w) I = (5.33)
J
-----;=== = = == =
cos2 kz, II + x2 sin 2 kz 1 H
In Fig.5.38, the amplificat ion obtained for t he one-layered soil with t wo d ifferent
thick nesses is compared to t hat reached in the cylindrical basin . The horizont al scale
corresponds t o t he normalized frequency H / A.
With: and x- =
kz I = cos 81
w cos 81 (5.34) /¥!I
1µ1
V.s-1 p2µ2 cos 82
where w is t he circular frequency, (V.s-1, µ 1 ) and (V82 , µ2) are t he velocities and shear
moduli in the basin and the bedrock respectively, 81 and 82 are the angles between the
364 Seismic wave propagaLiou aud a.mplifica.tiou 5.5 Amplification of seismic waves in 2D alluvial basins

1.7 1.7
circular
.. .... ·····
···..
,• ...... .:. ·("".--- ---
, ....
1.6 lD (H) 1.6 : .......· .·'"'··..... / ·.
1.5 lD (H /.fi) 1.5
/_/_,. ./-. .'.>< ·····../ //
1.4 / ' y
. . . -·-·- . 1.4 · · (lD}· ;-0./;--·~
.-·· . .:_. .~.·. .<,'........
, /,~·~- ~<-.<\ . ...... \.: .. . ..... .... .
/
'
1.3 , /
/ /

'' .§ /./ .. //\ \

''
"-"'.~ /.~.·./_',.:-~<·/
/
, /
' ' ,,.: '· 1.3 : //// ',, . . . .. ..

1 ,,.s;:t:::>/' ' __ , __/\\ \\,'

/
1.2 ~
/

/
/
/
":: ' ' 12 ,."" \. / - ......
/
: ',
1.1 ,,,.,/ '
/

:: ' ',

1.0
! ', ------- ............ /\
1.0 K"=o.50 :
Kh=0.75 :
K,.= 1.50
Kh=2.00
·.•.."" /
,,,· ··.....
" , _.... .............
0.9 [ I-l/A=0.25] [H!A=0.35 ] 0 .9
_

·... ...........··· -.. . . . ....

K,,=l .OO j K 11=3.QQ

0.1 0.2 0.3 0.4 0.5 0.6 0 ·8 ~o~.1----~ 2--; 0 · 2 5 1--0~.3-----o~.4-----

o .- o~.5----~
o.~6~

H/A ratio H/A. ra tio

Fig. 5.38: Amplification for a sin gle layer soil and a cylindrical b asin (Hi = H Fig. 5 .39: Amplification factor at t h e centre of e lliptical basins as a func-
and II1 = H / ,/2) as a func tion of the norm alized frequency H /A. tion of the normalized frequency (II/ A) for various h orizontal shape ratios "'h
(K.1,=0.5 , 0 .75, 1.0, 1.5, 2.0 and 3.0) : compar iso n w ith the l D case.
1

The main differences between both types of alluvial deposits a.re the following:
5.5.5 Elliptical basins with varia ble shape ratio
• Lhe two cases involving a single layer soil (H1 = H and H1 = H / ../2) lead to t he
same maximum amplifi cation value (1.41) since it only depends on the mechanical Previous results show a significant amplification increase in the case of a cylindrical basiu
properlies of the layer (the corresponding frequency nevertheless depending on the because of 2D effects infinencing seismic wave propagation. rn order to assess the 2D
layer thickness) , effects for ground motion amplificat ion , variou8 shape ratios are now considered. As
depicted in Fig.5.35, the horiwntal shape ratio is defined as "'h = L / H and is equal to 1
• the amplification obtained at t he centre of the cylindrical basin ( l. 73) is significantly for a circular basin . The various cases considered herein correspond Lo t he following shape
larger than in t he single layer cases. For the cylindrical basin, a 2D effect hence ratio values: K1i=0 .5; "'"=0.75; K1t= L.0; n,1i=l.5; 1,-,h=2.0 and 11:1i=3.0. Starting from the
strengthens the amplification of seismic waves when compared lo a horizontal layer circular case, two narrow basin cases and t lll'ee wide basin cases arc investigated . The
having the same depth, amp lification-frequency curves at the centre of the basin are compmed for all shape ratios
in Fig.5.39.
• the frequency of the amplification peak is larger for the cylindrical basin (H / A= 0.35) From Fig . 5 . ~·rn, the influence of the shape ratio on t he ground motion amplification
than for the one-layered soil with thickness Ht = H (H / A= 0.25 a.'> classically ad- may be characterized as follows:
mit ted). Conversely, for a thickness JJ1 = H/ ../2, the one-layered case leads to
Lhe same frequency as the cylindrical ba.c;in (F ig.5.38). In terms of fundame ntal • the shape ratios below 1 lead to lower amplification levels,
frequency, the cylindrical basin is then close to a one-layered soil with a t hickness
• the normalized frequency related to the maximum amplification decreases when
II1 = H j ./2. It is due to t he fact Lhat Lhe cylindrical basin is i:;tiffor than the
t he shape ratio increases (>vcaker 2D effects),
one-layered soil (finite geometrical extent) .
• for the largest shape ratio (i1:1i=3.0), the maximum amplifica.lion (1.45) is close Lo
366 Seismic wave propagatiou and amplification 5.6 Amplification of seismic waves in :3D alluvial lmsins 367

Table 5.1: Comparison of maximum amplifications and related normalized

frequencies for elliptical alluvial basins with variable shape ratio.
Shape ratio Amax H / A for Amax A=3.o
K.1i=0. 50 1. 67 0.;)0 · ·············· ··········· · • ··•···· • ··· ········· ··,...--------~

/'i,h=0 .75 1.73 0.40

/'i,h =l.00 i.n 0.35
K.h =l. 50 1.64 0.31
1.0 1 - - - - - - -
n:-1,.=2.00 1.56 0.27
K1i.=3 .00 1.45 0.26

that obtained for the one-layered soil (l .41) and the r elated normalized frequency
(0. 26) is slightly different, from that of the l D case (0.25 for H 1 = H) .
For each shape ratio, the maximum amplification and the related normalized frequen-
cies are gaUlCred iu Table 5.1. In addition to these two parameters, the shape ratio
0.1 1.0
may also influence the location of the maximum amplification area.s (basin edge effects,
period (s)
foc using effects, etc) . ln t he following, 3D theoretical cases wi ll firstly be discussed and
actual alluvial basins will then be consider ed.
Fig. 5.40: Spectral aggravation factor as proposed by Pitilakis et al . (1999) .

5.5.6 2D/1D aggravation factor

As shown by previous results, t he influence of 2D effects may b e large. For actual basins,
it is possible to consider codes provisions, mai nly based on l D analyses, or to record and
compute the 2D ' vavefield in the geological profile. Since it may be costly in lerms of
geotechnical investigation and numerical requirements, some authors proposed simplified
methods to account for 2D effects. Pitilakis et al. (1999) proposed an aggrnvation factor
in order t o increase the amplifica tion level from lD analysis to 2D predictions. These
authors analyzed seismic vvave amplification in the Volvi basin (G reece) and performed
numerous lD and 2D analyses. From t hese compar isons, they fo und that t he 2D effects
are significant and proposed the following 2D/ 1D aggravation factor, denoted A:
3
Since it is a well documented site studied by many authors (Chavez-Garcia et al.,
2000; J\!lakra et al., 2002; Pitilakis et al. , 1999; R.iepl et al., 1998; Semblat et al. , 2005),
wave amplification at the Volvi Euroseistest site will be investigated in the following.
5.6 Amplification of seismic waves in 3D alluvial basins
The amplification of seismic waves is now investigated in the case of 3D alluvial basins.
Various theoretical bas ins have been proposed by several authors: the sem i-spherical
basin (Sanchez-Sesma, 1983) , the semi-ellipsoidal basin (Clouteau , 1990; 1fossessian and

A = 1+2(~)
Dravinski, 1990) , the moon-valley model (Sanchez-Sesma and Luzon 1995), etc. Some
for TE [O ,Tt]
r esults showing the infiuence of 3D effects arc discussed in this section.
A = 3.0 for 1'E [T1,T2] (5.35)
5.6.1 Semi-spherical basin
A= 1 + 2 (Tr; for T E [T2 , Tmax]
Description of the problem
As depicted in F ig.5.40, in the period range [T1 , T 2 ], the aggravated spectral amplifica- Numerous papers have investigated t he 3D wave diffract.ion by a semi-spherical canyon
tion is hence t hree t imes the lD spectral amplification. The aggravation factor decr eases (Lee, 1978; Liao et al., 2004; Yokoi, 2003) or 3D seismic: wave amplification by sur-
rapidly for shorter, and longer, periods. T his proposal is very interesting since it accounts face heterogeneitiei-; (Dravinski , 2003; Komatitsch and Vilotte, 1998; l\!Ioczo et al., 2002;
for 2D effects in a simple way. The aggravation level, and related period range, should Sanchez-Sesma, 1983; Sanchez-Sesma and Luzon, 1995). The first example considered
probably be assessed for various basin geometries and different soil layering. in t his section corresp onds to a semi-spherical alluvial basin in an elastic half space. As
depicted in F ig.5.41 , the basin radius is denoted R and a plane P -wave is considered.
368 Seismic wave propagation and amplificatiou 5.6 Ampl ification of seism ic waves iu :~D alluvial basins 369

semi-spherical
basin gR 5

...., 4

plane r ~
al
s
P -wave
r r r r r al
C.)
ell
.-
c.
en
.....
3

Fig. 5.41: Amplification of a plane vertical P-wave by a semi-spherical basin: "'C

2
model description.
horizontal
.... --...... .
, ,, ..
...
,,
Several results have been pu blis hed for the case of a semi-spherical alluvial basin ,,
, ......
,
(Drnvinski, 2003; Lee, 1984; Sanchez-Sesma, 198~~). The results of Sanchez-Sesma (1983) 0 -· ---- -------------------
are derived using a series expam;ion method. For the semi-spherical basin and the half- 0 0.5 1.5 2 2.5 3
spacc, Sanchez-Sesma chose the mechanical parameters as follows : normalized distance
• shear moduli: ttR. / ftF, = 0.3, Fig. 5.42: Vertical and horizontal s u rface motions computed by Sanchez-
• mass densities: PR/ PE = 0 .6, Sesma (1983) for a semi-spherical basin at normalized frequ ency 17p = 0.50.

• Poisson 's ratios: VR. = 0.30 and VFJ = 0.25.

where subscript R refers to the alluvial basin and subscript E to t he half-space. frequency for the 3D semi-spherical basin is surely much higher.

Reference solution
Similarly to the previous sections, lhe results were computed by Sanchez-Sesma (1983)
in the frequency domain a t a given normalized frequency corresponding to t he diameler-
to-wavelength ratio 'f}p = 2R/ A p where Ap is the P wavelength in the alluvial basin.
Tn Fig.5.12, the amplification of the seismic motion computed by Sauchez-Sesma
(1983) at the free-surface is displayed vs normalized distance x/ R (normalized .frequency
'T/P = 0.50) . The amplification factor a l Lhe centre of the semi-spherical basin for the ver-
tical component is 2.82 (i.e. 5.64 in aruplitnde to be compared to 2.00 for the half-space).
T his amplification value is larger t han for the one-layered case or the 2D circnlar basin
(§5 .5.3) since, for the semi-spherical basin, focusing effects are very sLrong (Sanchez-
Sesma, 1983; Dangla et al., 2005).
As shown in Fig.5.42 , a horizontal motion component due to wave conversions also
appears. At this frequency, Lhe m aximum horizontal motion is reached at approximately
x/ R =0.5. It should be noticed that the normalized frequency 'T/P = 0.50 corresponds to
the fundam ental frequency of t he ·1n case (the wavelength being Ar = 4R \vith R the
depth of the basin). As already shown in t he 2D cir cular case (§5 .5.3), the fuudamental
5.6.2 Sine-shaped basin
As also considered by Bard and Bouchon (1985) in the 2D case, Sanchez-Sesma (1983)
computed the amplifi.cated wavefield in the case of 3D axisyrnmetrical sine-shaped basins
(ver tically incident P-wave) . The numerical results are plotted in Fig.5.43 (solid line)
for nonnali?~ed frequency T/p = 0.50. AL the centre of l he basin, t he vertical component
is amplified by a factor close to 3 (when compared to free-field), that is very close to the
amplification factor obtained for the semi-spherical basin (dashed line). T he amplitude
of the transverse component is also equivalent to t he previous case.
However, the spatial variation of the amplitude is much faster than in the semi-
spherical case. This is certainly due to the fact that Lhc effective width for the sine-
shaped basin is smaller than its real width due to very small dipping angles at its edges .
As suggested by B ard and Bouchon (1985), the a.clual width of the sine-shaped basin
must be r educed by a factor of about 2. Such a scaling factor has been applied to t he
resnlts of the sine-shaped ba.<iin to plot the dotted cnrve in Fig.5.43. It appears Lhat,
for normalized frequency 'f/p = 0.50, these scaled results are very close to the r esults of
370 Seismic wave propagation and amplification 5.6 Amplification of seismic waves in 3D all uvial basins 371

k .,ahr=l
·'·._,·., IU. 1
sine-sh a ped
5 >· ·.\

\vertical semi· spherical

".\
4 ·.,..
··~
.,., sine-shaped
\ (scaled)
\.
3 \· ..
\ .........~caled x2
-......::~~-+'·········· -- - ·~·~ .... _________ _
2 ·· ·············
h orizontal
..... ·.:.: :..:.·.;;.~. ~ ...
'~·~·.:...:-.:_.. .. .
..;..·'"'·.:..: ·..;..·'"'·.:..:· ............. .
~---..;;;..o~·..;.;··,:.,;·:._='··=.:.....,,_,..,_~_

0.5 1.5 2 2.5 3

normalized distance
Fig. 5 .43 : Surface motion computed by Sanchez-Sesma (1983) for a 3D sine-
shaped basin at normalized frequency 17P = 0.50: comparison with the semi-
spherical case (dashed) and the scaled sine-shaped case (dotted).

t he semi-spherical basin . The equivalence suggested by Bard and Bouchon (J 985) for 2D
sine-shaped basins hence appears to be valid in the 3D case.

5.6.3 Semi-sphe rical basin a nd oblique incidences

Using the FM-accelerated BEM, Chaillat et al. (2009) computed the seismic wavefield
for an oblique incident SV- wave. Such r esults were also proposed by Mossessian and
~
Dravinski (1990) and comparisons between both numer ical methods were performed by . ~-=----
Chaillat et al. (2009). O< I U; I <4.38 0< I Ui I <5 .3 1

I n the case of an oblique incident P-wavc (e = 30° ), the nurnerical results from Fig. 5.44: Amplification of a plane oblique P-wave (B = 30°) in a semi-spherical
Chaillat (2008) arc plotted in Fig.5.44 for two different normalized frequencies kpa/Ti= l. 0 basin at normalized frequencies 1.0 (left) and 2.0 (right): jU,v l top, IUY I middle
(left) and kpa/7i=2.0 (right). The three motion components are displayed: IUxl (top) , and IUzl bottom, from (Chaillat, 2008) . See h ttp://qsha .obs.ujf-grenoble . fr
for data and additional results.
IUyl (middle) and IUzl (bo t tom). The spectral amplification is found to be large in the
basin but, due to the oblique incidence, the propagation phenomena arc very different
along the x and y axes.
372 Seismic wave propagation and amplification 5.6 Amplification of seismic waves in ~~D alluvial basins 373

5.6.4 Semi-ellipsoidal basin

topography
The case of semi-ellipsoidal basius, which arc non-axisymroeLric, has mainly bceu inves-
tigated by :Mossessian and Drnvini;ki ( 1990) and Clouteau (1990). They compnted the
seismic waveficld in the time domai n as well as in the frequency domain . Their results
are very clifTercnt when considering the major or the minor axis due to Lhe asymmetry of
the basin (and possibly the incidence if it is oblique). Chaillat et al. (2009) recently morl-
elled this problem using the Fa.st Multipole Method and comparing with Mossessian and
Dravinski (1990) . For a plane oblique SV-wave (B = 30°) at normalil~ed frequencies 0.25
and 1.0, the seismic wavefield computed IJy ChaillaL el al. (2009) is displayed in Fig.5.45.

.. .. __ ....
I Uyl
-1 .0
y
x/a
-1.5
-1.5 -1 .0 -0.5 0. 0.5 1.0

20

O< I U yI <9.53 O< I UyI <14.42

------------ , ,.- ,,,,,.,,,,,,..::-
.... ~ .,,,,~,,..............
- - - - - - - - - ---- .... -,\\\,/flf1- '"'l'~ ,;11
---- _... __... __
.... ... ___
/ ... , ' --___ .,...-...... ... _.,.,,_::
.... -
......
..........,,,.,,, ....... __
,,, ............. -
_............ .....___
__ ...._

Fig. 5.45: Amplification of a plane oblique SV-wave (8 = 30°) in a semi- ::::::::::=~~:~~~~~:~1~~;~~~~i~:~~:i:::::::?::::;;:::~~~:~:::::::::~::

___________.,. ... .......... ... __ ..... _. ___ __ ... __..._ ....... ,--------- .... _...
-,,,,,,,~,. ,,~-""',. - ~

e llipsoidal basin at normalized frequencies 0.25 (left) and 1.0 (right) , from
(Chaillat et al., 2009).

5.6.5 Moon-valley model

oo~~~~~__.,_,
~J~~~~-:'.,-~~~~---..,~~~~~-::".::--~~~~~
T he moon-valley model is also a typical 30 non-a.xisymmetric model: a large semi-sphere 5 10 15 20 25
is truncated by a smaller semi-sphere to generate a moon-shaped surface (Fig.5.46) . T his time (s)
model was for instance sLudicd by Sanchez-Sesma and Luzon (1995). For such a basin,
St upazziui (2004) computed the 3D Reismic wavefield in t ime-domain by the Spectral Fig. 5.46: 3D moon-valley geometry (top) studie d by various authors (e.g .
Element Method. His results are d isplayed in F ig.5.46 (bottom) for an oblique (B = 30°) Sanchez-Sesma and Luzon, 1995) . Numerical results (bottom) computed in
p lane SH-wave. The seismograms show t hat interesting propagation and amplification time domain by the Spectral Element Method (Stupazzini, 2004) .
phenomena occur in such a non a.xisyrnmetrical basin. Stupazzini (2004) also compared
his results with those computed by Sanchez-Sesma and Luzon (1995).
374 Seismic wave propagation and amplification 5.7 Modal approaches to analyze site eficcts 375

5.7 Modal approaches to analyze site effects
5.7.1 Amplification of the seismic motion and resonance
T he amplification of the seismic motion in alluvial basins is mainly d ue to the resonance
of sur ficial layern at peculiar freq uencies. Site effects may t hus be analy11ed in terms of
vi bratory resonance of a sur face geological struct ure (Bard and Ilouchon, 1985; Dobry et
al., 1976; Paolucci, 1999) . Dobry et aL (1976) proposed several closed-form solutions to
estimate the fundamental frequency of simple geological structures (two-layered, multi-
layered soils, etc). ln this section, the fundame ntal frequency of a geological structure
is determined through a simplified modal approach based on Rayleigh 's approximation
(Paolucci, 1999). A full modal character ization of an act ual site will also be proposed.
5.7.2 Various types of modal approaches
The use of modal approaches to characleri11e geological structures is widely spread (Pao I u-
cci, 1999; Zhao, 1996). These ap proaches generally yield t he fundamental frequency of
geological structures account ing for their geometry (Bard and Bouchon, 1985 ; Paolucci,
1999; Semblat et al., 2003a,b) or t he heterogeneities in the deposit (Dobry et al. , 1976;
Hadj ian , 2002; Pecker, 2005; Zhao, 1996). Starting from several simple assump tions, it
is hence possible t o estimate the fundamental freq uency of various alluvial d eposits. It
is nevertheless difficult to directly compare the various frequencies in terms of amplifica-
tion. The main assum ptions considered in these modal methods will be compared in the
following.
Ji and Yi denote the order i Bessel and Weber Bessel functions (resp.) . Dobry et al.
(1976) also proposed a generalization of these results for two-layered and multi-layered
soi ls. Using this method, more recent results (Hadjian, 2002) proposed an iterative ap-
proach to estimate t he fundamental period o f multi-layered deposits.
In Lhe case of arbitrary variations of the shear modulus 11, with dept h z, closed-form
solutiom; proposed by Pecker (2005) arc p resented in Chapter 2 (§2 .12) .
Methods for 2D or 3D deposits
To characterize the resonance of alluv ial basins, Paolncci (1999) proposed a simplified
method involving the Rayleigh approximation. This method may be used for any type of
2D or 3D deposit. It consists in minimizi ng the ratio between t he strain energy and t he
kinetic energy in order to estimate t he fundamen tal frequency. This method is described
in details in §5. 7.3 and is used t o estimate the modal features of an actual alluvial deposit.
5.7.3 Simplified modal method
The modal method considered herein aims at estimating t he fundamental frequency of
a geological structur e. 'Nave propagation is a complex phenomenon leading to a strong
amplification at various frequencies (Sernblat el al., 2000) b ut, as a practical goal, it is
interesting to determine the fu ndamental frequency thr ough simplified approaches. As
shown by Paolucci (1999), the Rayleigh method allows a fast and reliable estimation of
t he fundamental frequency of a geological structnre.

Methods adapted to lD fillings T his method considers t hat the displacemen t of an elastic system in one o f its eigen-
Dobry et al. (1976) were t he first ones to propose a full modal characterization of geolog- rnodes may be approximated by that of a one degree o f freedom system. 'vVe shall
ical structu res. For a homogeneous or inhomogeneous horizontal layer (linear variation investigate herein t he first eigcnrnode characterized by t he frequency w 0 = 27r fo. Denot-
of the shear mod ulus) or a rnultilayercd soil, they proposed closed-form solutions for t he ing W the elastic energy of t he system and JC its kinetic energy, the conservation of the
fund amental period of geological structures. For a horizontal layer with t hickness II and total euergy of an elastic system implies that W max = JCrni.x· The displacemen t uk (Jd., t )
shear modulusµ varying linearly with t he depth z, they assumed l he following expression corresponding Lo t he t ime-harmonic vibrations at freqnency w 0 may be wr itten as:
for µ.(z) :
µ(z) 2 1 - J<2 (5 .39)
- - = I< + z (5.36)
µff H w here± <leuotes the space coordinate, i = .;=I, t Lhe time and 'ifJk(±) t he modal s hape
where f{ = 11Jf;,
11 0 and /lH denoting the mod ul us at the top and at the base o f the layer. along direction k. The kinetic energy may hence be calculated:

To estimat e the period of the layer, denot ed by T, Dobry et al. d erived the p rop-
ert ies of the homogeneous equivalen t layer writing T = ~~, where ·v;,,1 = /?1j- at the
JC (t) = 1~
1
0
2
P(±) ( ;; ) dD = - w5 e 2iwo l k~ p(J<) 1f;k (x_) dfl (5 .40)

equiva lent depth Zeq defined by:

2 J<2 thus: JCmax = mFJC(t) = - w5 Jn/ 21 p(Jd.) 'ifJk(;r.) dfl (5 .41)
H
Ze q = (~
H)
(1 _ J<2) _
1 - 1<2
(5 .37)
The elastic energy W hence reads as follows:
where a 1 is the first root of equation :
(5.42)
376 Seismic wave propagation and amplification 5.7 Modal approaches t.o analyze site effects 377

where £ij = ~ (u;,_7 +uJ ,i) i8 the strain tensor and aij = A£u0ij + 2/i,£.;j the s tress tensor ob-
Table 5.2: Various modal methods and types of deposits possi bly investigated.
tained t hrough Hooke's law with J .;j the Kronecker symbol,>. andµ t he Lame coefficients.
Authors homog. hetcrog. lD 2D 3D excitation
As for Kma~" W reaches its maximum value when le2 iwotl = l. We thus obtain: Dobry et al. (1976) yes yes yes no no no
Hadjian (2002) yes yes yes no no no
r ajt(;J;.)£jt(;f) dn Ba.rd & Ilouchon (1985) yes no yes yes no no
2 r:n'--- -- - -
"-./"" (5.43) Paolucci (1999) yes yes no yes yes
f p(;i;.)'lfi~(;f) dn
110
Wo = Pecker (2005) yes yes yes no no no
.Jn Semblat et al. (2003b) yes yes (yes) yes (yes) (yes)
The fundamental frequency of t he model, Eq.(5.43), leads to the exact fundamental
frequency of the system when Lhe actual mode shape is known. However , since the exact
solut ion is gener '.llly unknow n, the value of w 0 m ay be est im ated through an adequate
experimental surveys as well as numerical analyses (SernblaL et al., 2000). This shallow
approximation, '¢k, of 1/Jk· This approximation must fulfill the geometrical compatibili-
ha.sin will be also considered in the next i;ection to investigate seismic wave propagation
ties and t he boundary conditions. Nevertheless, Paolucci (1999) demonstra ted that the
and amplification. The basin is assumed homogeneous and its vibratory resonance is
second condition may not be fully satisfied and that the modal shape may be chosen in a
analyzed for anti-plane motion (SH-waves). T he mechanical properties of bot h media
wide variety of functions only satisfying the geometrical compatibili ties . One may then
(basin and bedrock) are the following:
estimat e the fundamental frequency using the following expression:

1 O'j1(;&.)€j1(Ji.) dr?
• basin: p=2000 kg/m 3 , µ =180 MPa, v=0 .2, that is Vs=:rno m/s;

• bedrock: p=2300 kg/m3 , /J.=4500 MPa, v= 0.2, t hat is Vs= l400 m / s.

w~ S in p(;f)'l/Jx
-2
(5.4tl )
(;r) dfl where p is the mai;s dens ity, µ is t he shear modulus, v the Poissou's ratio and Vs the
.n
shear wave velocity.
Various modal shapes may be chosen in order to minimize the approx imate funda-
mental frequency. Ai; depicted in F ig.5.4.7, t he basin/bed rock interface may be described by two s inu-
soidal func tions as follows :
5. 7.4 Featu res of the various modal methods
•west ern deepest part of the basin: f(x,z) = (h1 + l).cos(2 .710- x+ l.55)
3
The main features of the various modal methods are gathered in Table 5.2. For each
modal method, Table 5.2 details the types of deposits possibly investigated by t he • ca.stern shallowest part of the basin: g(x, z) = (h2 + 2). cos(2.810- 3 x - 1.3)
method. The simplified modal approach proposed by Dobry et al. (1976) and Hadjian Amongst the admissible modal shapes, the followi ng ones arc chosen (Paolucci, 1999) :
(2002) may be considered for one-dimensional homogeneous and heterogeneous geological
structures. As presented in Chapter 2 (§2.12), it is even possible to find closed-form sol u-
tions for ar bitrarily inhomogeneous l D profiles (Pecker , 2005) . Two-dimensional alluvial
- ( x, z ) =cosrn(
'!f;2 - l - f (:c, z)) s111 x)
. 2 8 +1(n+l)n( 1 + - cost (2m+l)n -z (5.45)
2 · · 2 a 2 h
deposits may be investigated by the method proposed by Bard and Bouchon (1985). The
method proposed by Paolucci (1999) m ay be used for homogeneous and heterogeneous where f (x, z) is the function given previously; r ? 1 and l ;::: 1 are real parameters,
2D or 3D alluvial basins. The method proposed by Scmblat et al. (2003b) is adapted s=O,l,. .. is an integer, rn and n denotes the orders of the modes along the vertical and
to any Lype of homogeneous or heterogeneous deposi t (lD , 2D or 3D) and, furthermore , horizontal directions (respectively) .
the seismic excitation may be accounted for through a simplified modal superposition
approach. In this case, the inequal ity giving the fundamental frequency is the following:

5.7.5 Fundamental frequ ency of a geological structure

w
2
_.l ~') f,' )']
µ [( \
< llllil ---==----=-------=~-
(
8
dxdz
(5.46)
To use the simplified m odal method, we consider a shallow geological profile located in o- r ,s.t fnp·~~(x,z)dxdz
the centre of Nice, France. This alluvial site has been extensively studied through various
378 Seismic wave propagation and amplificatiou 5. 7 Modal approaches t o analyze site effects 379

I 5. 7. 7 Simplified modal method vs experimental spectral ratios

'· I
v.
I \
g(x,z) The fundamental frequencies of the Nice basin estimated by the simplified modal method

-20 I
I
I
\./ h2 v
are now compared to the frequencies obtained from the peaks in the experimental spec-
t ral ratios (Semblat et al., 2000) . The vertical lines displayed in F ig.5.48 represent the
f(x,z) I
I
' v
' fundamental frequencies for the deepest and shallowest parts of t he basin and for various
,..-._
s
..._, h1 ~/ I
shear moduli (111; 11,2; J.1.3) .
A..., -40 I
I

0..
Q) I
I For the deepest part of the basin, the fundamental frequency h estimated for {t3
'"Cl
I
I
basin profile (solid line) is in very good agreement with the experimental amplification peaks. For the
,I
shallowest part of the basin, the frequency value obtained for shear modulus µ 1 (dotted-
-60 dashed line) is close lo the second amplification peak in the experimental spectral ratio.

T hese comparisons show t he interest of the simplified modal method to estimate the
modal features of 2D (as well as 3D) alluvial deposits. Through this homogeneous model,
-80 L-~~~~---'~~~~~---'-~~~~~~~~~~~~~~
0 500 1000 1500 2000 t he influence of the wave velocity in the basin, in terms of fundamental frequency, may
distance (m) be easily assessed . Jn the following, t he numerical results obtained by the Rayleigh ap-
proximation will also be compared to t he frequencies corresponding to the maximum
Fig. 5.47: Description of the homogeneous geological profile for the simplified amplification computed by the Boundary Element Melhod through the analysis of seis-
modal approach (Semblat et al., 2003a). mic wave propagation in the profile (Semblat el al., 2000) .

50 ' ' . '

West East
...... " _ ... __ ' ' .. ' . .
··········-f-·····-]-···+···i····[...r··H· ··T ···:········ ···r.. ··:+······+·····!···+··!···[·-r+···················
Table 5.3: Fundamental frequencies of a shallow basin (Nice, France) obtaine d ···········r··-----1-·····r···-1··-·r···r·-1·· .. ···r····:··--·--- ··· 1···· · :· r ··· ····-i· ··- ·· t ··-- -~ ···i-··t··1 .. t··· -- ·--------··---
by the simplified modal approach (Semblat et al., 2003a). 20 ----·------L ...... J. .....L... J.... L ' I : ....... ..1.. : : .. ___i .. . ..L..L.J. ·...... ·······-·····
Shear F\tndamcntal freq. Fundament al freq. <l)
l l l :: I : : : : !
u 10 ...........l.......L ...L...!....' I I
modulus western pa rt eastern par t ~
Cl) :::::::::::~:::::::j::::::~:: : :j:::: _.:j: :::1
... : I
µ1 1.50 Hz 2.86 Hz i-< ·······----~---···· ;· ··· ·- ; .. ·-~---· .:~::~: :::i::::::::::.. .J .... ::~:: : :: {··-·
µ2 1.23 Ilz 2.34 Hz ~ 5 ::::: ::::::~:: :::::r::: ::~:: :: r~: :--~-- ~- ---•----. :-------. ---•··· --~-!----- --+---
Cl)

1.07 Ilz 2.02 Hz i-<

..._
-----------~ -------1------ ~ ----~ -- --r--1--1- ---1-----:-------- ---1-----:-r---· ·-· ~··· ·
µ3 .. -.•...•. ·~ ....•.. ~- •.•..~ ----~ -- ;·---~ --~-- ~- ---1-.•. ·i ... -.. -. ..-f- -.-·{-~ ---.... ~-. --
<l)
~
....... : : : : : ' : : ' ': ' modal m ethod

,TIU TL Tl I ---- ~:~2µ,IJ

en 2

i 1
:
.:::j::::~:::~ ::j:: ~: ::=:== :::;:::::::: ::=:==:::::~:::::::~:::: µ,=µ / 2
5.7.6 Modal estimation of the fundamental frequency --r::::1: :::~:::~::1::~: :::.:::::?:::::::: :::.:::::i:f:::::::r:::. ~.-,.......,--..,-....,·,....,....·- --·
0 . 5+-~--.-~-,-_,..._,..._,...._,_,.........~~~-+-_.._-i---r~t--+-;t--+-t-t--~~---+

0.2 0.5 2 5 10 20
For the Nice basin, the results arc computed by the Rayleigh approximation for the first frequ ency (Hz)
mode considering different values of Lhe shear modulus: µ 1 = 180 l\IIP a (i.e. Vs , =300 m /s),
µ2 = 120 l'vIPa (i .e. Vs,=245 m/s) and µ 3 =90 ?vIPa (i .e. V.93 =212 m /s) . The values of Fig. 5.48: Fundamental frequencies of a shallow bas in (Nice, France) esti-
the fundamental frequency are given in Table 5.3 for the deepest part (West) and the mat ed by the simplified modal approach and comparison with experimental
shallowest part (E ast) of the basin. These values range between 1.07 and 1.50 Hz in the spectral ratios (Semblat et al., 2003a) .
deepes t part and 2.02 and 2.86 Hz in the shallowest part. T hey will be compared to the
peaks of the site/ reference spectral raLios in t he next paragraph.
380 Seismic wave propagation and amp lification 5.8 Amplification in shallow basins (e.g. Kice) 381

5.8 Amplification in shallow basins (e.g. Nice) 50+-~~'--~'---

_,~
.-----~-
! :
.-~
~--~
............_
: --~
i
_-._-
_-
.._-_-_.-~~:-~~~~~--'~~+-+-'--'-~~~~+
!

5.8.1 Analysis of site effects in Nice (France) 20 ·--··--··--·-·-··--+--+··-+-·-r--1'- i

l l :1
10 -····-·---··+-·-·····-! ........l. --:c : c ::J.
Measurements at the alluvial sit e ~::::_.:::::::x~~~-J --
5
The French Riviera is located in a moderate seismicity area and an important seism ic : I :
--i ····················--···············-········ ........ I . ;-·- ·-l· .. -- ¥ •. ; •...•••••••••••
event is mainly expected in the Genoa gulf (magni Lude 6 at 30 km from the coasL). In
t he ciLy of Nice, the geological st.ructme was studied in great details by Lhe seismic risks
2 ··········-·--····.L........-...... 1 ! ! ·----···························-L ....... .l.. .----~------l l ~
tea.JU of the CETE Mcditerranee (Duval, 199G) . As in many other ci Lies, the highest
dern;ity of popu lation is found in the areas located on alluvial deposits.
··=·~·"·-:-J::::~~.1, - -~: :,:,_~::.,:_ J_•i.-~i.; =:._
- -
--:::::::::::::::::::::t:::::::..::::::.... l:: : : : :+~: : (N ort:h-'.South}::::
!- ···················t····..·-··· · ··········· · ~··· ·· ···· · ·····- 1-· - ·· ·~-- - ·- ··· ···•·· -·
0 . 5+-~~~~...--;----T~-+-...,--~~~~...-~--T~~~-..--~~--..--,--~~~--r-

.
10 -+- ,_-__ -+--~-.----~.........-..-. ~
__-_ - ..-....-...+-. ....-...- -.+,---~---..,-...-....-...-...-....-__-____.,..,______-___....,_~,___-._-___-____-___- - - - r -

=:=1 J==F~iS~:\~=~;~,~!~:\~,:S:\fr~: A
A first experimental survey (1984) allowed the delerminat ion of the regional seismic
ris k and the charncteri:tatiou of various geological areas having very difTereut seismic
responses. T he seii:nnici ty was recorded during one year (1992) at four siles where t he
seismic amplificatio n may he important (Duval, 1996). Experimental measurements were s :

~ : ;J~~~~1:q·~1~~:~~tt;-~~~·~~
performed to characteri7.e site effects in the southern part of the area of Nice central
station. As depicted in Fig.5.49, t his area corresponds to a shallow alluvial deposit
located in an ancient North-South valley. Eastward, the Cimiez hill is due to a bedrock
outcropping.

West alluvial site ref. bedrock East

/
~
0.1-+-~~~~----+~-~-+~~~~~~~~-~---r--+-+-+--r-~~~--r

0.2 0.5 2 5 10 20

~ ~
frequency (Hz)

Fig. 5.50: Experimental measu rements at the s hallow a llu via l s it e (Nice) :
~~~~~~~~~~~
standard spectral rat ios from w eak seismic motion (top) and H / V spectral
rat ios from microtremor recordings (bot tom) (Semblat et al. , 2000).
Fig. 5.49: Shallow alluvial deposit in the centre of Nice (French Riviera).

Spectral ratios
From the experimental smvey, the site/reference spectral ratios may be computed. Com-
paring the measurements performed at different siles, the spectral ratios show the influ-
ence of the layer depth on the peak frequency (Duval , 1996). An interesting mcLhod using
am bient noise (rn.icroLremors) has also been p erformed (see Chapter 3). The stabil iLy of
ambient noise has been st udied in the time and frequency domains t hrough spectral ratios
between the horizontal and the vertical motion components (denoted H/V) . It has been
established fro m comparisons between standard spectral ratios (i.e. site/reforence ratios)
that the H/V spectral ratio allows lhe determination of the fundamental frequency of
t he soil.
The real eart,hquake, as well as microtremor, recordings clearly show a ground mo-
tion amplification a t the cenLre of the bas in between 1 Hz and 2 Hz (Duva.I, 1996).
These results are displayed in Fig.5.50 through the spectral ratios computed from weak
seismic moUon measurements (top) and II/ V spectral ratios from microlremor record-
ings (bot tom). The measurements are performed at t,he al luvial site (ALS site) and
at the reference bedrock station (REF site) . The fundamental frequency estimated
from microtrernor recordingH iHvery close t,o that determined by standard spectral ratios
(Fig.5.50). Howe ver, the H/ V spectral ratio is much less t han the site/ reference ratio
and decreases at higher frequencies . Above the deepest part of the basin, t he amplifica-
tion factor obtained from site/reference spectral ratios (F ig .5.50 Lop) reaches a maximum
value of 20 around 1.5 Hz.
To investigate time-domain amp lification, several accelerograms recorded in t he cen-
tr e of Kice ar e rlisplayed in Fig .5.51. For a seismic event of similar magnitude (feb. 2001,
M4.6 ear Lhquake), the discr epancy in amplitude is very large. The 'Alsace-Lorraine'
(ALS) alluvial sit e leads t o t he largest amplitudes and lime d urations.
:382 Seismic wave propagation and amplification 5.8 Amplification in shallow basins (e.g. Nice)

NROC-*
\
I
5.0 h=32m1 \
NLIB I
\
\
I
I
\
\
4.0 I
\
(J) I
\
"Cl I
\
NPOR
-s
;::l I
.;.;> \
• rl
3.0 I
0.
I
I
'
\
\
co /
I \

2.0 / '
''
''
1.0

0.0
0.0 1.0 2.0 3.0 4.0
frequency (Hz)

5 10 15 20 25 30 35 40 Fig. 5.52: lD transfer function for the deepest (solid) and the shallowest
time (s) (dashed) part of the Nice basin.

Fig. 5.51: Accelerograms recorded at various locations in the city of Nice dur-
ing the feb.2001, M4.6 earthquake (from CETE Mediterranee, Nice, France).
• for the deepest part (H=64 m): fo=l.2 Hz,

• for the shallowest part (H'= 32 m) : .f6 = 2.3 Hz.

5.8.2 Amplification from lD transfer functions
To analyze seismic wave amplification in this shallow basin, a simple estimation may be
performed assuming a single layer soil over an elastic half-space. For the analytical and
numerical analyses, a 2D East-v\Test profile is now considered. T he average mechanical
properties of the materials are chosen as follows:
• alluvial deposit : p1 =2000 kg/m3 , µ 1 = 180 MPa, thus Vs, =300 m/s,
• bedrock: p2 =2300 kg/m3, µ2 = 4500 l'vfPa, thus Vs2 =1400 rn/'6.
Considering a plane vertical S H-wave, the closed-form of the transfer function, de-
rived in Chapter 3, is recalled through Eq.(5.1). In its deepest part, the alluvial deposit
is assmned as a constant depth layer with H =64 rn, whereas the shallowest part is mod-
elled by a H' = 32 m deep layer . These values correspond to the maximum depth in each
part of the basin (Fig.5.49).
Computing the transfer function from Eq.(5.1), the curves corresponding to depths
H and H ' are displayed in Fig.5.52. The maximum amplification factor is below 6 for
both depth values and t he related frequencies are respectively:
From t hese lD results, the fondamental frequencies arc in good agreement with the
frequencies estimated fro m the maximum values of the spectral ratios (Fig.5.50). Con-
versely, the lD spectral amplification of the ground motion is much smaller t han the
experimental one (nearly 20). It is hence necessary to consider a more realistic model to
quantify site effects in the basin. A 2D model involving the Boundary Element Method
(see Chapter 4) will be studied in the following.
5.8.3 2D model of the geological profile
T he East-West geological profile depicted in Fig.5.49 is considered but the exact 2D ge-
ometry of t he basin is now modelled t hrough a Boundary Element l'vfodel. The alluvial
deposit is still assumed homogeneous and the mechanical properties arc the same as pre-
viously. The incident wavefield is a plane vertical SH-wave. In Fig.5.53, the Boundary
Element mesh is depicted (top) for the solution of the integral equation and additional
interior points are chosen (bottom) for the a posteriori computation of the seismic wave-
field within the basin and bedrock (integral representation formula, see Chapter 4) .
384 Seismic wave propagation and amplification
West Boundary element mesh East
• 1•~m::'":,.•::":•:•""::m111111ir.u111:11::::u;~:::::::1111\\1111111111•~
I n •itftt..ttnu--1~-..t+11Htt. HY I
~~---------=~~~~-~~~~--------'.=~· ________2o~o_d'.~'~~~~-~
////s~l~acvc
Fig. 5.53: Boundary E lement mesh of the alluvial deposit in the centre of Nice
for a plane S H-wave (top) and additional interior points (bottom) (Semblat
et al. , 2000).
The model involveB the fuudamcntal solution (Green's function) of the free-space
fo r Lhe deposit a.nd the hill (hounded media). It involves the fundamental Bolution of
t he half-space to accurately model t he bedrock a.s a sub-domain of a half-spa.ce. The
com putation is performed using Lhc FEM/BEM code CESAR-LCPC (HumberL el al.,
2005) .
5 .8 .4 Amplification of a plane SH-wave
Comidering the model depicted in Fig.5.53, the Boundary Element Method allows Lo
compuLc the wavefield along the boundary and int erfaces. The amplification factor is
derived afterwards removi ng the free-imrfacc effect. The isovalues of the amplification
factor arc given in F ig.5.54 fo r different frequencies. T he amplification is large at the
s urface of t he depo;;it since it reaches a illaximum value of 15.0 at frequency 1.6 Hz. The
largest amplification is found in the deepest, part of the deposit. However, for the highest
freq uency (Fig.5.54, bottom) , the am plification factor in the shallowest part of the basin
Hignificantly increases. The resu lts displayed in Fig .5.54 clearly show that, for the two
highest frequencies (1.6 and 2.0 H'./,), the maximum amplification occurs across several
small areas. It corresponds t o the fact that, at these frcquencieB, the wavelengths arc
mnch shorter than at lower frequencies.
T he 3D curve diBplayed in Fig.5.55 shows the variations of the amplificaiion factor at
the surface of the deposit a.5 a function of distance and frequency. It is then p ossible to
assess the amplification level at each frequency as well as the location and extent of the
5.8 Ampl ification in shallow basins (e.g. Nice)
f=0.4 Hz -A,,,.,=1.1 5
f= l.2 Hz - A,,,,.=6.5
1-==
f=l.6 Hz - A.n,,,=15.0
, ~---
f=2.0 Hz - A.n••=6.0
1-==
Fig. 5.54: BEM simulations of the amplification process for a p lane vertical
SH-wave in a shallow geologica l profile located in the centre of N ice (Semblat
et al., 2000).
related maximum amplification area. In F ig.5.55, the amplification factor remains low
for frequencies below 0.8 Hz. Above thiB frequency, a signi ficant amplification appears in
the deepest part of the basin (West) . For higher frequencies, several high amplification
a reas arc assessed in the west ern part of the deposit. Above 1.5 Hz, the amplification
factor in the shallowest par t of the deposit (East) incrcase8 progressively. Between 2.0
and 2.5 IIz, the amplification decreases in the deepe8t part and becomes even larger in
the Bhallowest parL.
T he rcBults from F igB 5.54 and 5.55 show that the ground motion amplification reaches
a rnaxinmm value of 15.0 at approximately 1.6 Hz. These results arc in good agreement
with the freque ncy values previously estimat ed from the lD approach. In the 2D numeri-
cal simulations, the fundamental frequency is nevertheless slightly larger tha n that given
by t he trausfcr fun ction. It can be easily cxplaiucd since the lD approach considers a
horizon Lal layer havin g a constant depth equivalent to the <leepe8t local part of the basin,
386 Seismic wave propagation and amplification 5.8 Amplification in shallow basins (e.g. Nice)

undamped -------· Q-1=0.02 ---------·--- Q-1=0.04 - - - - Q-1 =0.06

2.0

1.0
'
' ' ........ ____ .,,..., /
amplification
----· <2.0 8.0
0.5
- - 3.0 -11.0
1.0 2.0 3 .0 4.0 5.0
c:J 5.0 ->15.0 frequency (Hz)

0 500 1000 1500 2000 Fig. 5.56: Overall amplification factor in the centre of Nice for various atten-
distance (m) uations (Semblat et al., 2000) .

Fig. 5.55: Amplification computed at the surface of the Nice basin as a func-
tion of distance and frequency (Semblat et al. , 2000).
thus leading to an underestimated fundamental frequency. The 2D amplification level
estimated numerically by the Boundary Element ·M ethod is nearly three times larger that
than obtained in the l D analysis. Even for a shallow basin, the numerical model allows
the detailed description of the geometry of the deposit leading to t rapped surface waves
and , consequently, to stronger site effects . These values appear to be in good agreement
with the experimental ones displayed in F ig.5.50.
5.8.5 Influence of attenuation
In the BEM formulation, a Zener viscoelastic model is introduced to analyze the influ-
ence of attenuation on the amplification process (Scmblat et al., 2000). The undamped
amplification is compared to several damped cases in Fig.5.56 (Q - 1 =0.02; Q- 1 =0.04 and
Q- 1 =0.06). The influence of attenuation is found to be significant since the reduction
ranges from 20% to 50% depending on the attenuation Q- 1 . This result is consistent
with previous remarks on t he effect of trapped surface waves propagating from one edge
of the basin to the other. When trapped (due to the velocity contrast between the basin
and the bedrock) , these waves propagate through an attenuating medium and are simul-
taneously amplified and attenuated. This phenomenon may even be more complex for
strong seismic motion, i.e. larger attenuation in the soil (Gelis et al., 2008; Delepine et
a,l., 2007).
5.8.6 lD and 2D amplification vs experimental results
The numerical results arc now compared to the experimental spectral ratios computed
from real earthquake measurements (weak motions). Since computations were also per-
formed for P and SV waves, Semblat et al. (2000) considered the site/reference spectral
ratios along the three directions of space: vertical, North-South, East-\Vest. The ge-
ological profile being oriented along the East-West direction (Fig.5.49), the vertically
incident P -wave corresponds to vertical motion, SH-wave to antiplane motion along the
North-South direction and the SV-wave to an East-vVcst polarized motion.
The numerical and experimental results for the North-South component are displayed
in Fig .5.57. The numerical re~mlts correspond to the maximum overall amplification, de-
fined in (Semblat et al., 2000), and for the experimental results both the mean value
(dotted line) and t he standard deviation (shaded area) are depicted. The comparisons
for the three motion components are given in (Semblat et al., 2000).
For the NorLh-South component (Fig.5.57), the 2D results are computed for the shear
modulus value /J, 3 = µi/2 = 90 NIPa, that is Vs0 = 212 m /s. The agreement with the
experimental results is very good in terms of amplification level. The two main amplifi-
cations obtained experimentally, in the deepest part of the deposit (above 1.0 Hz) and
in the shallowest part (above 2.5 Hz), arc retrieved with the 2D BEM model. However,
the exp erimental spectral ratios arc lower than the numerical amplification factor at low
frequencies (below 0.6 Hz).
388 Seismic wave propagation and amplification 5.8 Amplification in shallow basins (e.g. Nice) 389

50 10~--.-~--.-~-,-~--.--~--.--~-.--~-.--~-.-----:::..--~.-~.,--~-,-,
.. 5
9
·-· · · -- · --- 8 4
,-.. 7 ,-..
w. 20 . . . .. . .... . --- 0

z
..__,
6
5
0
3 ..__,
0..-<
s:1
0
......
_...,
C1i
10 ...
. .. ·-·
...
..
---
4
3
2 0 --
IN
u 2

-s°'
~
· ..-<

C1i
5
···· ··· -··-· . ,
1
o L_~~_J_~~~----'-~--'-~--'-~~~----'~---'~~'----~~~L__J
10 ~-.--~--.-~~~---,-~---.~--,-~----.~--.--::.
0
5
(tj 9
;:..,
_..., 2 ., . . . 8 ,-..
u(j)
4 i:::1
7 C1i
°'
(jJ
~ .. '.
6 3
(j)

__§
5
lD 4 2 0..-<
0.5 ~--·::·_
··-~····_·~...;...·.·_·.·;._.··...;..
·. . .;'. .....;... " ----r---r--;---r--r--t---r-1~-----t
. "...;.. 3
a
2 IN
0.2 0.5 2 5 10 20 1 Gelis et al. (2008)
frequency (Hz) OL_-'--~_J_~~~----'-~~~----'-~--'-~--'-~~~---'~---'~~L-J
0
20
1000
Fig. 5.57: Comparison between the weak motion measurements (dot- 0
ted/shaded) , the lD transfer function (dashed) and the 2D BEM results
(solid) for the North-South component in the centre of Nice. s
'-'
-20
800

600
'5p.. -40
~ -60 400
Conversely, the discrepancy between the lD transfer fu nction and the experimental 200
-80
results is large. The frequency of maximum amplification is not exactly found by t he lD
analysis and the amplification level is even strongly underest imated in some freqnency - 100 L_~~6~
00~---'-~8-0L0~-'--~10~0-o~~-1-2~00~~~14~0-o~-'--1-6~
0-0 0
ranges . In this shallow basin, the lD estimation of seismic wave amplifica tion is not
distance along profile (m)
reliable wheu compared to experimental spectral ratios . A 2D aggravation factor will be
discuHsed in t he next section.
Fig. 5.58: 2D/1D a ggravation factor vs distance and frequency for the Nice
profile: numerical estimation for a vertically incident wave (top) and mean
5 .8. 7 2D /lD aggravation factor value for several computations at various incidence angles (middle) (Gelis et
al., 2008) .
GClis ct al. (2008) also studied the amplification of seismic waves in the centre of Nice.
They computed the seismic wavefield using the Finite Difference Method (Saenger et al.,
2000). As shown in Fig.5.58, the 2D/ 1D aggravation factor is estimated for a vertically
constitutive model. From these computations, they observed deamplification above 2 Hz
incident wave (Lop) and for various incidences (middle). The maximum aggravation
and a shift of the resonant frequency to lower frequencies. Such results show t he influence
factor .is found around 3, that is close to the ma,"Ximum valne proposed by Pitilakis et
of combined geometrical effect s (2D/ 3D) and consLitutive effects (nonlinearities) .
al. ( 1 99~l ) . The results involving various incidences (middle) lead t o larger aggravation
factors in a certain frequency range (4 to 6 Hz) which is similar Lo that suggested by
Pitilakis el al. (1999) . In t he Hofter part of the basin (1200-1400 m ), the aggravation 5.8.8 Comparison with the simplified modal method
facto r may even reach an approximate value of 4 (Fig.5.58). In their paper , C elis et The 2D fundamental frequencies of the alluvial basin were previously computed by the
al. (2008) also proposed several 2D simulations fo r strong motions using a nonlinear simplified modal approach (§5. 7.3). Several value of the shear modulus of t he deposit
390 Seismic wave propagation and amplification 5.8 Amplification in shallow basins (e.g. Nice) 391

were considered: f,L1 =180 MPa (i.e. Vs=300 m/s), µ 2 = 120 M Pa (i.e. Vs=245 m/s) and
µ, = 90 MPa (i.e. Vs =212 m/s) . The numerical results obt ained by t he Rayleigh method 60
are compared to t he freq uencies leading to t he maximum amplification in the 2D IlEM
simulations. The comparison is shown in Table 5.4 for the deepest par t (left ) and the
shallowest part (right) of the basin. 50

Table 5.4: Comparison between the peak frequencies from the 2D BEM sim- 40
ulations and the fundament a l frequencies estimated by the simplified modal 3
<1l
method. s
.....
....., 30
Shear Freq. west ern part Freq. eastern p ar t
modulus BEM modal BEM modal
µ1 1.35 Hz 1.50 Hz 2.42 H'.11 2.86 Hz 20
µ2 l.30 H'.11 1.23 Hz 2.13 Hz 2.34 Hz
µ3 1.13 H'.11 l.07 Hz 1.75 Hz 2.02 Hz
10

From these results, it may be concluded that the fundamental frequencies obtained
from t he Rayleigh approxima tion are in good agreement with that observed in the 2D ""' 0'--'-'-----'---'-_._-'-.._____..---'-_._-'--1-...1-.L_U--L-L.__..L__J__~
BEM computations. As it can be seen in Fig .5.57, the ID analysis leads to significantly reference O 500 1000 1500 2000
(Feb.2001) distance (m)
lower frequencies. For a homogeneous basin model, the Rayleigh approximation is hence
a reliable mean to estimate the fundamental frequency of 2D / 3D alluvial basins.

5.8.9 Time-domain simulations
Frum the frequency doma in results comp uted by t he Boundary Element Method (Scmblat
et al. , 2000), t he time domain response of the ba.5in may be easily derived considering t he
Fourier transforms at each point. T he Feb. 2001 earthquake is chosen as the seismic input
and the results arc computed for a duration of 60 s. As shown in Fig.5.59, the largest
amplification is fo und in the deepest part of the basin (Kham, 2004). The influence of
trapped surface waves may thus be considered a.5 very large. The duration increase is a lso
very impor tant in this part of the basin. The time histories a lso show t hat the shallowest
parL of the basin mainly amplifies higher frequency components.
5.8.10 Comparisons with a deep alluvial basin
Fig. 5.59: Simulated wavefield in the a lluvial basin for the Feb. 2001 earth-
quake in the centre of Nice (Kham, 2004).
areas also appear at depth in t he Caracas basin. As shown in Fig.5.60, focusing effects
strongly govern t he variations of the seismic motion with depth:
• in the case of Nice (top), the ground motion is maximum at the free-surface and
regularly decreases with depth ,
• in the case of Caracas (bot tom ), the ground motion may be large at the free-surface
as well as in some deep areas where focusing of seismic waves may lead to significant
amplifications.

To compare the amplification in t he shallow deposit in Nice to deep and irregular al-
luvial deposits, the case of Caracas (Vene?.uela) is now considered (Duval et al., 1998) .
As shown previously, t opographic site effects may be significant at t his site while the
stratification may even lead to larger amplifications. The irregular Caracas basin leads
to a very different amplification/ frecp1ency dependence than the shallow Nice deposit
with many large amplification peaks (Semblat et al., 2002a). Furt hermore, the maxi-
mum amplificat ion is found at the surface in the case of Nice whereas large amplification
392 Seismic wave propagat ion and amplificat ion 5.9 Ampl ification in a. deep basin (Volvi) 393

5.9 Amplification in a deep basin (Volvi)

5.9.1 The Volvi EuroSeisTest

Location and main features

~
20
0
·.-< 15
+-'
In this section , we shall consider a deep irregular basin located in Volvi, Greece. The Volvi
crj 10
<:..) EuroSeisTest is a E uropean test siLe crea ted through grants of the European Commis-
~ 5
...--<
sion in the framework of the research programme ' Global Change and Natural Disasters"
~ 0 (J ongrna.ns ct al., 1998; Makra et al., 2005; Riepl et al., 1998). Several europcan research
s
crj programmes (e.g. EuroSeisRisk) investigated site effects and soil-structure interaction
0
t hrough this test site (http://euroseis.civil.auth .gr/) .

?' 20
T he EuroSeisTest is located in an alluvial valley a.L 30km north-cast of t he city of
~
~ 30 Thcssaloniki in Greece (Vig.5.61) . I t is an active sismotectonic area where the large 1978
~ 40 Thcssalon iki earthquake occurred. T he basin is 6km long and more than 200m deep.
"!:_,/
Permanent and t emporary sensors arrays a.re used on the test site to measure act ual
50 2000
1500 (earthquakes) as ·well as artificial dynamic loadings.
500 1000
0 dist a nce (m)

~ 15
0
....,0 10
......
crj 41 .0
<:..)
5
-s
..<;::::. .
~

ro
0

Greece

Volvi
EuroSeis ---+---===----'
Test
39.0
2500 ~
1000 1500 2000 http://euroseis.civil.auth.gr/ ~_2_2~.0---~~-.__~~-E~a_st_,
distance (m)
Fig. 5.61: Map of the Volvi area showing the EuroSeisTest (Semblat et al.,
Fig. 5.60: Surface and in-depth amplificatio n for a shallow (Nice, top) and a 2005 ) .
deep (Caracas, bottom) deposit (Semblat et al., 2002a).

Geotechnical and geophysical analyses of the site conditions are numerous and it is
known in details . These analyses wer e performed at the test sit e or in the lab . Conse-
quently, the soil properties are well-known and fully reliable. The geological structure is
depicted in Fig.5.62.
394 Seismic wave propagation and amplification 5.9 Amplification in a deep basin (Volvi) 395

Experimental results in the Volvi basin
Several recorded a.ccclerograms are displayed in Fig.5.62 for the Arnea M5.8, 1997 earth-
quake. The amplification is found to be large in the deepest part of the basin (e.g. TST
station ). The PRO station being the reference station, there is nearly no amplification
at that loca.Lion (see (Chave:-1-Garcia ct al., 2002) for a discussion on this critical point).
At the STC and STE stations, the amplification is also significant and is mainly due to
basin edge effects. As it will be discussed hereafter, the influence of large faults in the
Volvi graben may also inftuence the propagation process.
The seismogram recorded at the T ST station during the Arnea M5 .8, 1997 earthquake
is displayed in Fig.5.6:3 (top). Ch{1vez-Garda et al. (2000) computed the time-frequency
transform of this record, Fig.5.6:3 (bottom) . When compared to the same plot for the
PRO reference sta.Lion (Chavez-Garcfa el at., 2000) , the influence of site amplification at
the T ST station is obvious:
• the largest peaks in the low frequency range (i.e. close to the fundamental frequency
of the basin) appear between 20 and 25 s leading to a longer motion duration than
at the PRO site (where the energy is mainly locali<1ecl between 15 and 17 s),

• between 15 and 17 s, only the higher frequency components (':::'3 Hz) are large at
the T ST site,
~
u
,.., • comparing the Peak Ground Acceleration between the T ST and PRO reference
'""'
>-< station leads to an amplificaLion level of nearly 3.
Q)
:>

..s::::
_....,
;::i
0
(/)

300
;::
_....,
>-<
0

250
z 2.09E-05
a0 <lq <la
~ 80
8 8
d
8
~~ 8d 4.0
s"'
0 d <;>
2 s s s
2(X)

150
3.0

,,....__ 'N
1m
s
.._, ES
<ll ;;...
<.;
"'O 50 >:: 2.0
~ 1
· '
.,..., ;:;
""" O'
"""
Cil 0
1'
~

.50 1.0
0 .0

-1m F1

-150 O.O '-'--10.L.0~-~~~-1-5"-

.0~-~~~-2-0'-.0~-~~~...,2:-5'-.0~-~~

NNW SSE time (s)

·2CXlo.m o.50 i.oo l.50 2.m 2.50 3.oo 3.50
distance (km) Fig. 5.63: Accelerogram (top) measured in g at the TST station (Volvi basin)
during the Arnea M5 .8 , 1997 event and its time-frequency transform (bot-
Fig. 5.62: Measurements from the Volvi test site (Arnea M5.8, 1997 event) tom), (Chavez-Garcia et al., 2000).
showing the amplification p rocess in the basin (Pitilakis et al., 1999).
396 Seismic wave propagation and amplification 5.9 Amplification in a deep basin (Volvi)

5.9.2 Sim p lified and complete mod els o f the Volvi basin
Variou s mode ls fo r t he Volvi b asin
For the Volvi t est sit e, several geotechnical models have been proposed. In th.is section ,
we shall choose one of them and der ive two numerical models: a simplified one with only
two soil layers and a complete one involving six soil layers. The main goal is to investigate
the influence of th e lmowledgc of the local geology on site effects computations.
T he basin model considered hercafLcr was proposed by LGIH (Eng. Geology Lahora.-
tory, Hyd rogeology and Geophysica,I Prospecting) from the University of Liege (Belginm),
(Jongrnans et al., 1998) . This gcotedmical model is depicted in Fig.5.64 (top) with a
correction giving an horizontal free surface but respecting layers dept hs as proposed hy
P.Y. Bard (LCP C/ LGTT, University of G renoble). Other geological models were also
proposed by Raptakis et al. (2000) a.ud Ch{wez-Carcia et al. (2000) and were nsed for
2D analyses of site effects.
As suggested in the work of Makra et al. (2002), we firstly considered a simplincd
model wi th only two layers on an elastic bedrock. As depicted in Fig.5.64 (bottom), the
various layers of the gcotcchnical model are corubincd to derive a simplified two-layer
model. The m echanical proper ties of the simplified model arc given in Table 5.5 and
are estimated as mean velocity values of t he velocities of t he detailed profile (Scrnblat et
aJ , 2005). The comp lete model (actual layering) directly corresponds to F\g.5.64 (top)
with six different layers on an elastic bedrock. The mechanical properties of t he com-
plete b asin model arc detailed in Table 5.6. T he purpose is now to analy;1e seismic wave
amplification for the Volvi basin by t he Boundary ElemenL Method and compare both
models for s ito effects and especially basin effects .
5.9.3 SH wave amplification in the Volvi basin
Amplification in t h e basin
[n this section , the seismic excitation is a plane SH-wave with vertical incidence. For
the simplified n10<lcl, awplification valnes est,imatcd through DEM simulations are given
in Fig.5.65 for varions frequencies. T he maiu features of the amp lification patterns are
as follows:

s
..._.. • For 0.6 Hz: the largest amplification occurs iu the deepest part of the basin and
this case seems to correspond to the fundan:icutal mode of vibration of the basin.
...., 100
,.q
p, Nevert heless, the maximum amplification factor is not very high (i.e. below 3).
<l.l
'"O 150
® • For 0.8 Hz: two areas of large amplification appear along the free surface in Lhc
central part of the basin wil h a rather high maximum val ue (9.5).
200 bedrock
0 • For l .O Hz: maximum amplificat ion is reached at the free surface but the main
amplification area (9.5) is shifted to the right in the intermediate part of t he basin.
50
• For 1.2 Hz: amplification areas also concern Lhc left m edium depth part of Lhe
s
..._.. basin and the maximum val ue is slightly lower (7.5) .
...., 100
,..q
• For ·1.8 Hz: both free surface and deeper areas reach large amplifications (8.3). The
0.
<l.l
.-0 150 largest am plification corresponds Lo the cxtrewe left of the medium depth part of
the basin.
200 be drock • For 2.4 Ilz: wit h such wavelengths, the shallow right part of the basin shows
0 1000 2000 3000 4000 5000 6000 large amplifications wit h both surface and deep processes. There is nearly no
dist a nce (m) amplification in the deepest central part of the basin.

F ig. 5.64: Geotechnical mod els of the Volvi b a sin: comple te mod el (top ) a nd
simplifie d m odel (bottom) as p rop osed by Se mblat et al. ( 2005 ).
Basin effects fo r the simplified a n d complet e mo de ls
Six points are chosen a long the Volvi profile and are marked in Fig.5.66. They are
located at 1000111 from one another and these areas correspond to var ious basin depths .
In F ig.5.66, the amplification/ frequency curves arc displayed for each of t hese locations.
398 Seismic wave propagation and amplification 5.9 Amplification in a deep basin (Volvi) 399

Table 5.6: Properties of t he soil layers for the complete model of the Volvi
basin (Fig.5.64, top ).
Soil layers mass density Young's modulus P oisson's ratio
layer 1 1700 kg/ m 3 180 MPa 0.33
layer 2 1800 kg/ m3 300 MPa 0.33
f,=0.6Hz, A,=2. 9 f2=0.8Hz, A2 =9.5 layer 3 1800 kg/ m3 300 MPa 0.33
layer 4 2000 kg/ m 3 530 MPa 0.48
layer 5 2200 kg/ m 3 1200 MPa 0.47
layer fi 2:300 kg/ m 3 3300 MPa 0.49
bedrock 2600 kg/ m 3 4200 MP a 0.19

f3=1.0Hz, Aa=9.5 f ,=l.2Hz, A,=7.5 • Point 2 (d=1380m): as shown in Fig.5.66, the maximum amplification is a.round
9 and is reached for frequency l.9Hz. This rather high frequency value can be
explained by the moderate depth in this part of the basin.

• Point 3 (d=2380rn): there is a large amplificatiou at 0.8Hz reaching the maximum

value of 11. This point is in the deepest part of the basin but amplification remains
nevertheless significant in the medium frequency range (f=l.9Ilz) due to the influ-
ence of t he shallowest left part of t he basin which is accurately modelled with the
f.,=l.8Hz, A,=8.3 f 6=2.4Hz, /\;=7.3 complete model (larger amplification).

• Point 4 (d=3380m): this point is in the central part of the basin and the ampli-
Fig. 5.65: Amplification values in the Volvi basin estimated numerically (sim- fication factor reaches its maximum in a lower frequency range. T he maximum
plified basin model) at various freque n cy: frequency values fi and r elated amplification level is close to the previous ones and the corresponding frequency
maximum amplifications Ai (Se mblat et al. , 2005) . is .identical. For this point, there is also large amplifications around 2.0Hz and
above 4Ht;: differences between simp lified and complete models are enhanced in
t he ruediurn frequency ranges, clue to the inflnence of the subsurface layers.
Table 5.5: Properties of the soil layers for the s implified model of t h e Volvi • Point 5 (d=4380m): the rnax:immn amplification (11) corresponds to frequency
basin (Fig.5 .64, bottom) . 0.8H:6. For this location, amplification for the simplified model is quite low (less
Poisson's rat,io than 6) for other frequencies whereas amplification for the complet e model is clearly
Soil layers mass density Young's modulus
stronger (around 8) for the medium frequency range. This difference may be due
layer 1 2100 kg/ rn3 677 MPa 0.280
to lateral heterogeneities, particularly strong for t he complete model. T he amplifi-
layer 2 2200 kg/ m 3 3595 MP a 0.453
cation pat tern at 0.8Hz shown in Fig.5.65 leads to maximum amplification in this
bedrock 2600 kg/m3 4390 MPa 0.249
particular area.

• Point 6 (d=5380m): for this point, the basin depth is slight ly smaller than in point
2 but the soil layers are softer. We get significant amplification values for the
Solutions for the simplified model are compared to those of the complete model. From simplified model and t he maximum (9.0) is reached at frequency 2.8Ht; whereas for
these curves, we can make the following comments: the complete model, there is no significant amplification.

• Point 1 (d=380m): for t his location , there is almost no amplification (at the scale
of the maximum amplification curves). The hasin depth is small aud t here is no
very soft soil layer at t his point.
400 Seismic wave propagation and amplification 5.9 Amplification in a deep basin (Volvi) 401

Comparison of amplification vs frequency and distance for both models (SH

wave)
As shown in F ig.5.67, the complete model (6 layers) leads to a larger number of ampli-
fication peaks on both edges of t he basin near faults Fl (South) and F4 (North). These
peaks of large amplification are especially located in three different areas:

simplif.
• f E[2-3Hz] and x=4000rn,
model
• .f E[3-4H:i:] and x=2500rn,
complete
model • f E[4-5Hz] and .T=l500m.
The transfer functions for each basin model are displayed in Fig.5.67 as isovalnes vs
distance and frequency. T he largest amplificat.ion corresponds to the same frequency
Jo :::'.0.8Hz for both models. It is located in two different areas of the basin. This
phenomenon is not due to the near surface layering since the simplified model also shows
200.- a 'double fundamental mode" (Makra et al. , 2002). This observation is also made by
0. 1000. 2000. 3000. 4000. 6000. Chavez-Garcia ct al. (2000) who suggests that it corresponds to the contribution of
distance (m) surface waves.

Fig. 5.66: Amplification/frequency cur ves for various locations along the
Volvi basin surface: simplified (solid) and complete (dashed) models (Semblat
et al., 2005).
T he amplifica tion values est imated numerically by the Boundary Element Method
are close to exper imental ones (Pitilakis ct al., 1999). ~When compared to lD analysis
of amplificat ion (Makra ct al., 2002), 2D estimations lead to larger values since the
actual amplification process is strongly infinenced by lat eral heterogeneities (sec following
sections). These effects are clear in Lhe previous numerical resulLs (Fig.5.66) and we will
iuvcstigate later t heir influence on time domain response and seismic signal duration .
5.9.4 Comparisons between simplified and complete models
As described in Tables 5.5 and 5.6, t he simplified model is a two-layered basin over
an elastic bedrock whereas the complete model incl udes 6 d ifferent soil layer:; over the
bedrock. The purpose of the comparison is to assess the infl uence of the soil layering
description on seismic wave amplification analysis including basin effects. To compare
the si mplified basin model to the complete one, we will analyze both freqnency and Lime
domain results.
Comparison for a Ricker wavelet (SH- wave)
To fully understand the influence of the basin model description of both vertical and hor-
izontal heterogeneities, the results computed in the frequency domain are transformed as
time domain responses at t he free surface. For both models (simplified and complete) ,
we consider an upward propagating SH-wave described by a Ricker wavelet whose spec-
trum is cent ered at l Hz . From the frequency domain BEM results presented in previous
sections, the t ime domain seismic waves are computed along the free surface. The time
histories are displayed in Fig.5.68 along Lhc basin for both types of models.
From the lime domain solutions derived from the simplified model (Fig.5 .68, top), the
amplification process appears clearly. The effect of lateral heterogeneities (basin effects)
is obvious since wave reflections on basin edges occur. The amplification of the first
arrivals also shows the influence of the velocity contra..'lt in the central (deepest,) part of
the basin. Seismic wave amplification in the simplified Volvi basin is then influenced by
both vertical (soil layeri ng) and lateral (basin effects) heterogeneities.
402 Seismic wave propagation and amplification 5.9 Amplification in a deep basin (Volvi) 403

6 When compared to solutions derived from the complete Volvi model (Fig.5.68, bot-
tom) , the amplification of the first arrivals is larger than for the simplified basin model
(simplified model) (since the magnitude scale between two following traces is twice that of the latter model).
5 The soil layering being described more precisely, the amplification due to velocity con-
trast is thus larger. Furthermore, the direct Ricker wavelet is combined with refiected
,...... a.ncl refracted waves to give a more complex wavefield (Fig.5.68) . It is especially the case
N
:r::
._._.,, 4 on both left and right sides of the deepest part of the basin. It is possibly due to the
>-,
u
i:1
Q,)
am combination of vertical and lateral heterogeneities influences. Since the velocity contrasts
are described more precisely in t he complete model, the lateral wave propagation in each
;::i 3 layer is made easier and the global amplification process enhanced. Concerning the signal
O"
Q,)

~
duration, it is significantly increased showing once more the combined influence of basin
effect and soil layering.
2
The influence of the soil layering of the basin on the amplification process as well as
on the signals duration raises the need for a very detailed knowledge of the soil properties
and layers geometry. T his is a key point to have reliable predictions of surface seismic
CD motion in alluvial deposits .
0
Comparison for a real time acce le rogram (SH-wave)
6
A real earthquake is now considered and the surface motion is computed in time domain
( complete model ) for both basin models. The June 1994 earthquake (.i\,f =3) is presented but other com-

ID@
5 putations for the May 1995 Arnaia earthquake were also performed and are discussed
by Kham (2004) . T he incident seismic mot.ion is chosen as the reponsc at the PRO
,...... station (North). As shown in F ig.5.69, the signals arc then computed at all other station
N 4 locations along the free surface for both the simplified and complete models. As previ-
::q
._._.,,
>,
u
~
am ously discussed by Cha.v ez-Garcfa et al. (2002), the quality of the reference station is
very important and PRO is t he best one in the area but is still located on a very thin
(].) 3 alluvial layer approximately at point 1 of Fig.5.66. In Fig.5.69, the measurements made
;::i
O"
Q)
at the different stations are given (top) for the sake of comparison. June 1994 signals are
~ filtered above 6Hz. The sea.le of the signals is identical for all cases. Slight differences
2 arc observed between measured PRO signals and computed ones because PRO is not a
perfect bedrock station (Ch{wez-Garcfa et al., 2002) .

As shown in Fig.5.69, both models lea.cl to a large amplification of the signals as

distance along the surface (m)
Fig. 5 .67: SH wave amplification vs frequency and distance for the simplified
(top) and complete (bottom) models of the Volvi basin (Semblat et al., 2005).
well as an increase of their duration in the central part of the basin. This trend is in
good agreement with the measurements which show a large amplification at the centre
of the basin . The complete model generally gives larger amplification (closer to the mea-
surements) since it describes more precisely the soil layering near the free surface. It is
especially the case for the GRE station for which the influence of near surface geology
appears significant.
404 Seismic wave propagation and amplification 5.9 Amplification in a deep basin (Volvi) 405

june 1994 earthquake recordings

00[~ ~ ~ 001~
30

(j

25 ~

,..-,_ 20
-3
Q)

·.s.s 15

10

20 25 Q)
u
i:::
Q)
( complete model
i...
<8
20 ~

15
3Q)
15
s
.....
.....,
10
E10
' .;::l
:

5 ;;- ~ ' :'=

tI
:

5
0

1000 2000 3000 4000 5000 6000

distance (m)
Fig. 5.68: Time domain responses for the simplified (top) and the complete
(bottom) Volvi basin mode ls unde r a lHz R icker excitation (Semblat et al.,
2005) .
Fig. 5.69: Accelerograms along the Volvi basin for the june 1994 earthquake.
Comparison with the simplified (top) and the complete (bottom ) basin mod-
els solutions for the bedrock station (PRO) signal (input) (Semblat et al. ,
2005).
406 Seismic wave propagation ancl amplification 5.9 Amplification in a deep basin (Volvi) 407

5.9.5 SV-wave amplification in the Volvi basin

T he amplification of SV-waves in the Volvi basin is now corwidercd. Figw:e 5.70 fi rsL
shows the transfer function of t he complete model for a vertical plane SV wave. The horizontal motion
amplification of both horiwntal and vertical motions is displayed as isovalues vs distance 5
( alo11g the free surface) and frequency.

-
,......_
The main amplification areas of the horiwntal motion of the basin under SII wave N 4 Ux
:::r::
.._..
excitation (Fig.5.67) arc recovered in the case of SV-wave (Fig.5.70). Their locations
>.
are identical in both distance and frequency: the fundament al resonance mode is locaLed C)
i:::
at the centre of the basin (x=3500m and fo = 0.8Hz); a large number of amplification Q)
;:J
3
p eaks also appears on both sides of t he model between faults F4 a nd F3 (.1: E[l000-2000] O'
Q)

m) and faults F2 and Fl (x E[4000-5000] m). The corresponding frequency ranges are ~
f E[l-3] Hz and f 2::4Hz respectively. 2

In Fig.5.70 (top), three main amplificaLion areas appear at frequencies lllz, 2Ilz and
3Hz (resp.) all along the basin width. T hey correspond to the contribution of the main
surface geological structures at lower dep t hs for larger frequencies. Thus , the lhird am-
plification area at 3Hz is linke<l to t he resonance of sup erficial layers well described by the CD
complete model. This superficial rei;onancc was not so strong in Lhe case of S JI-waves
6
(Fig.5.67).

The transfer function of the vertical motion (Fig.5. 70, bottom) gives even more in-
5
( vertical motion )
t eresting results, since it shows 2D effects due to wave conversion at the layers interfaces
as well a s surface waves generation. 1\vo particular areas can be noticed:

-
• The first amplification area, ranging from frequencies 1 and 2Hz and dist ances 'N' 4 Uy
2500m and 4500m, confir ms 2D effects involving the deepest part of Lhe basin. :::r::
.._..
>.
C)
•c,
• The secon d area, localed at large frequencies (f 2::3Hz) and at the basin edges, is i::: 3
Q)
associat ed to the generation of surface waves. ;:J

'
O'
These a reas also correspond to the cases where the discrepancy between the simp lified ~
Q)

2
. ,,.,
and the complete model is the largest (SH wave excitation). This is a good reason Lo
consider the complete model rather than the simplified one to investigate such detailed
amplification processes.
-
5.9.6 2D /lD aggravation factor
CD
1000 2000 3000 4000 5000
Makra ct al. (2005) also stud ied t he arnplificaLion of seismic waves in Volvi. They distance along the basin (m)
computed the seismic wavefield at various station loca tions using the Finite Difference
Method (1'.foczo , 2002). As shown in Fig.5.71 , t.he 2D/ 1D aggravation factor is fou nd Fig. 5. 70: Transfer function for the complete Volvi basin model under SV-
t o b e nearly 3 for stations FRlitf, TST, GRB and GRA. This valne is close to the wave: amplification of horizontal (top) and vertical (bottom) motions (Sem-
maximum va lue proposed by P itilakis et al. (1999) and that estimated in the case of blat et al. , 2005).
Nice by Celis et al. (2008), Fig.5.58.
408 Seismic wave propagation and ampli fication 5.9 Amplification in a deep basin (Volvi) 409

different numerical models: a very simplified one not taking into account the topmost
10.0
STE STC alluvial layers and a more complete model including a detailed description of the surface
soil layering. The main purpose of this comparison is to discuss the influence of Lhe
knowledge of the geological structure of the site on Lhe quality of site effects computations.
T he main conclusions of Lhe analysis arc as follows:
1.0
• Both models give reliable representation of 2D basin effects and the related seis-
mological processes: an increase of the amplification of the seismic motion (when
compared to a lD analysis) and a duration lengthening in the central part of the
basin. I3oth models give a correct estimation of t he main resonance of the ha.sin,
in both amplitude and frequency.
10.0
FRM TST
• T he influence of superficial soil layers shown by the complete model (LGIH) is
- Model I
sometimes significant but mainly influences the high frequency content of the signals
- Model II
- - - Model III as well as (trapped) surface waves propagation.
1.0
----· Model IV • The comparison between SV-wave amplification and SH-wave analysis also shows
······-·Model V significant differences between both approaches: the S H model is rather simple but
does not represent the infl uence of Lhe basin complex geometry (especially around
t he grabcn edges) which is quite important on the loca l amplification of the seismic
motion. These aspects, due to the scattering of SV-waves, are taken into account
10.0 in the SV-model, but lead to larger computational costs.
• T he amplification spectra, commonly used, do not fully describe 2D site effect s
and give on ly a partial view of Lhe amplification processes (for instance, complex
1.0 scattering phenomena or spatial variability of the seismic mot ion due t o surface
waves). An accurate description of the subsurface lit hology thro ugh complete geo-
logical models is necessary to make such detailed analyses.
0.1~~~~_._~~...._~_.____._~~~~~~~~~~~

0.1 1.0 0. 1 1.0
period (s) period (s)
F ig. 5 . 71: 2D /lD aggravation fact or for the Volvi EuroSeisTest at different
station locations (Makra et al. , 2005).
Conversely, for t he STE and STC st ations (Fig.5.71 , top) , ~vlakra et al. (2005) found
no 2D / lD aggravation. Since these stations are located on very shallow deposits, lD
amplification is mainly involved at these sites. Despite their location, the infiuence of
basin edge effects is also surprisingly small, it is probably due to the fact that the STE
and STC stations are "beyond" the F1 fault (i.e. outside the Volvi graben).
5.9. 7 Conclusions on site effects in Volvi
The cornplcxity of the geotechnical models depends on the available field data as well
as the required accuracy a nd reliability for the tar get parameters to characterise seismic
wave amplification. The seismic response of the Volvi basin has been studied using two
• As discussed by R iepl et al. (1 998) , the influence of the azimuth of the source on
the local amplification of the seismic m.otion seems to be significant. This issue
was not discussed .in the previous simulations since we wanted to avoid additional
focusing effects in both numerical models (due to source location) for the sake of
easier comparisons. Further analyses (Pedersen et al., 1995) involving local sources
rather than plane waves (considered herein) wi ll be needed probably for both 2D
and 3D models.
• Such parameters as the frequency content of the seismic event is also import ant to
make the choice of Lhe most sui table (simplified vs detailed) model.
• The 2D/ 1D aggravation factor was shown to be significant in the deepest part of
the basin (Fig.5. 71 ). The level of aggravation found by Makra et aL (2005) for
Volvi is close to that found for the Nice basin by Celis et al. (2008).
• Finally, t he amplification level is found to be large in all cases since we only consid-
ered weak motions (small earthquakes) . In the case of strong earthquakes, the soil
nonlinearity may lead to some deamplification in the surficial soil layers (Bonilla,
2000; Delepine et al., 2007; Gelis et al., 2008; Henze et al., 2004; Kwok et al. , 2008) .
410 Seismic wave propagation and amplifioition 5.10 Wave-structme interaction 4ll

5.10 Wave-structure interaction 4.0

5.10.1 From soil-structure t o wave-structure interaction D ~

0
Wave propagation in soils may be analyzed in the case where civil engineering strnctures ·~ 3.0
c.J
arc located above the soil. As already dbcussed in Chapter 4, it is then necessary to ~

account for the coupling between the structure and the soil (Fig.5.72). Indeed, when the SH wave ~

HHH
i...
2.0
structure moves under a. seismic excitation, it may radiate waves that induce perturba- ~
(fJ
\
\
7 \
lions in the ground motion when compared to the free-field case. Consequently, it is no t ~ I \
t'll I \
always possible to prescribe, at the base of the structure, the free-field seismic motion in ....,
i...
1.0 I
I \
\
order to simulate its efTect on t he structure. In the frequency range of earthquakes, this /base \
\ I
I
I I
perturbat ion may be neglected for smal l or light structures but not for large or rnasHive I
I
' ' ...J I

oues. 0.0
0.0 0.25 0.50 0.75 1.00 1.25 1.50
frequency (Hz)
DO Fig. 5. 73: Soil-structure model and numerical results in terms of d isplacement
DD
DO transfer function at the top and base of the building (Semblat and Dangla,
DO DD 2005).
DO DD

T he d isplacement transfer functions at the base and at the top of the building arc
plotted in F ig.5.73 wit h respect to frequency for a vert ically incident wave. For very low
freqnencies (long wavelengths), the seismic waves are not sensitive to the infiuence of the
build ing and the amplitude of the free-field is retrieved (superposi tion of the incident
and refiected waves, that is 2).

In F ig.5. 73 , specific frequencies for which the base displacement is zero can be no-
ticed. They correspond to the resonant frequencies of the building for the fixed base
condition. ~cvcrthclcss, as shown in Fig.5.73 , the top displacement at these frequencies
does not reach itH maximum. This result characterizes dynam ic soil-structure interac-
tion: the maximum top displacement is reached at frequencies also depending on the
Fig. 5. 72: Various types of proble m s involving wave-structure interaction.
soil proper ties. It is thus mandatory to account for this phenomenon to a nalyze the
dynamic response of the buildi ng. It is even possible to derive the frequencies leading Lo
the maximum top displacement from the analysis of soil-strncture interaction.

5.10.2 Seismic analysis for buildings

The analysis of dynamic soil-structure interaction may be performed through numerical
methods. The first example proposed herein to illustrate t he problem of Hoil-strucLure
interaction iH depicted in Fig.5. 73. A 50 rn high and 30 m wide building is fouuded on
the soil. The soil and the building arc hit by a harmonic plane S H -wave. The b uilding
and the soil arc assumed as homogeneous isotropic elastic bodies and arc modelled by
the Boundary Element Method (Semblat and Dangla, 2005).
T he BEJVI Holution has been previously computed at the free surface and the interface
between t he Hoil and the building. T he wa.vcficld in Lhe soil is now estimated using the
integral representat ion formula (see Chapter 4). The displacement isovalues are displayed
in F ig.5 .74 in the case of one building (top) or three identical buildings (bottom). The
soil-Htructurc interaction at low frequencieH (left) is found to be small when compared
to higher frequencies (r ight) . In the case of three buildings, the interaction between
the buildings (and through the soil) may be significant since there arc some important
difTcrences in the buildings motions (esp. at f=0.2Hz and 0.4Hz).
412 Seism ic wave propagation a.nd amplification
f=0.2 Hz f=0.4 Hz f=0.6 Hz
F ig. 5.74: Displacement isovalues in the case of one (top) or three buildings
(bottom) (Semblat and Dangla, 2005).
5.10.3 Seismic interaction with underground structures
At Lhc local scale, one may also consider the interaction between an underground cavity
(lunnel, mines, etc) and a seismic wave propagating through the soil. The damages ob-
served on under ground structures after strong earthquakes assess the vulnerability of such
structures especially in their cross-section (K u rose, 2000). For underground structures,
the current seismic codes arc mainly based on equivalent static approaches in which the
complex wave propagation phenomena are not taken inlo account.
The interaction between a plane S H-wave and a tunnel will now be simulated by
the Bow1dary Element :vlethod (see Chapter 4). Consideri ng a homogeneons half-space,
the interference patterns (displacement. field) displayed in Fig.5 .75 illustrate the inter-
action between the seismic wavefield and the tunnel for two different incidence angles.
The wave/tunnel interaction may be strong a l certain frequencies influencing both the
displacement and tractions a long t he tunnel wall, but also the seismic wavefield at the
free s urface (Fig.5 .75). The wave/tunnel interaction may indeed increase the displace-
ment, field in some areas but also, as depicted in Fig.5.75, lead to some shadow areas
with a ver y low dispacement amplitude (above the tunnel for B = 90° or on the right
for fJ = 60°). As shown by Kurose (2000) or Kham ct al. (2002b), the interact ion is
governed by the ratio between the dimensions and depth of the tum1el on one hand, and
the wavelength of the seismic wavefield on the other. The numerical solutions computed
by the Boundary Element Method and displayed in Fig.5. 75 arc compared to closed-form
solutions in (Kham et al., 2002b ).
5.10 Wave-strnctnre interaction 413
0=90° f=5Hz f=20Hz f=40Hz
f=20Hz f=40Hz
Fig. 5 . 75: BEM simulations of wave/ t u nnel interaction: seismic w a vefield for
two inciden ces (8 = 90°, top , and (} = 60° , bottom) and different frequencies
(Se mblat a nd Dangla, 2005).
5 .10.4 Seismic interactions at the local and global scales
From soil-structure to site-city int eraction
The analysis of wave/structure interaction may be performed at a local or global (i.e.
large) scale (Bard et al., 2005; Boutin and Roussillon, 2004). Indeed, for alluvial basins
located in densely urbanized areas, the buildings may infinence the propagation of seismic
wave near the free surface. Furthermore, as previously evidenced by several authors
(Gucgucn et al., 2000; Kham et al., 2006; Wirgin and Dard, 1996), the vibration of a
structure at the free surface generates a radiated wavcficld into the soil. The dynamic
soil/structure interaction will be firstly analyzed at the local scale of a single building. At
the global scale, the infinence of surface structures, such as buildings, on the propagation
of seismic waves will then be analyzed in the case of an entire alluvial basin. Dynamic
soil-structure interaction may thus be investigated at various scales:
• the local scale (e.g. single building) corresponds to classical dynamic soil-structure
interaction (SS I ) and involves the building foundation and a par t of lhe neigh-
bouring soil;
414 Seismic wave propagalion a nd amplification 5.10 Wave-si ruci ure int eraction 415

• t he intermediate scale concerns one or several build ings as well as the soil at i he
geotechnical scale (layered soi l) . Waves generated by the vibrat ion of lhc structure
may have a significant influence on the motion of the neighbouring structures (sec
Fig.5.74);
• the global scale represents an entire city and the geological p rofile (e.g. alluvial
ba,<>in) and the interaction is thus considered as site-city interactiou (SCI) for
.ins lance (Scmblat et al., 2008). It may lead to significant modifications of Lhe
seismic waveficld due to the dynamic response of the numerous buildings invol ved
in the city.
_,.,,...,,.---- ------- ~--~-
/ the struct ur e to t he soil.
Local seismic int eraction
At the local scale, beyond the classical soil-structure interaction, soil-structure-soil i11-
teraclion (SStSI) may appear as a link between the dynamic response of the structures
and the seismic waves radiated into the soil. This case is illustrated in F ig.5. 76 where,
considering dynamic experiments performed at Volvi EuroSeisTest (Greece), t he three-
dimensional seismic waveficld radiated by a v ibrating building is com puted by the Finite
Element Method (sec Chapt er 4) . As in t he case of sei:;;mic wave amplification, t he ve-
locity cont rast between t he soil layers governs t he propagation of t he radiated surface
wave:,;. From the experimental and numerical resulls obtained by Gueguen el al. (2000),
the wavefi.eld radiated by the structure may be trapped in the surficial soil layors and
hence be propagated along large distances. As illustrated in Fig.5. 76, the directivity of
the three-dimensional wavefield also characterizes the way the energy is transmitted from

G lobal seismic interact ion

Seismic interaction between structures at large scales. At an intermediate scale,
Lhc structure-to-structure through t he soil interaction may also be significant. It may be
---... assessed at the sea.le of several strucLmes: the wavefi.eld radiated into the soil bv some
---- ·--..._
---- ----- ·-- of these structures may influence the motion of neighbouring structures (Fig.5.74). T he
level of :;;uch interactions depends on the dynamic fealures of t he structures, as well as
'1\ =0.120s the fu ndamen tal frequencies of the soil layers (Bard ct al., 2005 ; C lo uieau and Aubry,
-----. 2001 ; Lombaert et al., 2004; Tsogka and Wirgin, 2003) . For this type of interaction,

_.-r·---·------.. . . . . _ __.. -----~ ---...

detailed results ar e proposed in (Bard et al., 2005), (Gueguen et al. , 2000) and (Kham
et al., 2006) .

Seismic interaction at the scale of an alluvial bas in. At a global scale, ihc
seismic interaction between an alluvial basin and a large number of buildings may be
considered (Bielak et al., 1999; Ch<ivez- Garcia and Cardenas-Soto, 2002). As depicted in
Fig.5.77, the shallow alluvial ba.sin, located in l\'ice and studied earlier in this chapter,
is combined with a building network of variable density. The site-city interaction was
studied for various urban configurations and soil layering by Kham et al. (2006) and
Semblat et al. (2008) .

oooooo ooooooo

Fig. 5.76: FEM simulations of dynamics experiments performed at Volvi

1-==
EuroSeisTest: displacement wavefield generated by the vibration of a b u ilding
mock-up at different times T 1 =0.095s, T2 =0. 120s, T3 =0.145s and T 4 = 0.170s Fig. 5. 77: Model for the analysis of site-city interaction in a shallow a lluv ial
(Bard et al. , 2005). basin (Semblat et al., 2008) .
 T

4.16 Seismic wave propagation and amplification

M odifications of the se ismic ground motion. The inftnence of the buildings on the
seismic gro und motion is analyzed by the Boundar y E lement Met hod in the frequency
do main (see Chapter 4). The ·u rban seismic wavefield is compared Lo the free-field ground
motion studied earlier in this chapter, as well as in (Scmblat et al., 2000). The pertur-
batiorn;, t l p, arc then defined as the difference between the t otal waveficld, Ut, and the
free-field motion Uff: 1Lp = Ut - ·u 11 . The numerical results are displayed in Fig.5.78 in
terms of perturbations for a given city configuration. The influence of the building net-
work is found to be siguificant. As shown in (Kh am el al., 2006), the radiated waveficld
is larger when Lhe eigenfrequencics of the building8 are dose to t he fundamental frequen-
cies of t he alluvial deposit . Furthermore, complex directi vity elTects may be observed
in Fig.5. 78. They are due to simple or multiple interactions belwecn the structures and
Lhe alluvial basin (Sernblat et al., 2008). At Lhe global scale, t he dynamic inLcractiou
between the soil/site and the surface constructions may then be signi ficant especially in
dense urban areas. As shown by (Scrnblat et al., 2008), it may lead to detrimental as
well as beneficial effects 011 the ground motion alo ng Lhe basin surface.

20
Appendices

15

Q)

.§
~
10

Fig. 5.78: Influence of site-city interaction on the amplification of the seismic

ground motion in a shallow a lluvial basin (Nice, France): perturbations nl'
for a given city configura tion (Sernblat et al. , 2008).
Appendix A

Several operators in mechanics

A.l Vectors and product of vectors

In this book 1 vectors are denoted as underlined characters as follows:

1l or ui ('i = x) y) z) in com.poncnts (A.l)

The product of tv,ro vectors 14_ and Q is \:VTitten:

(A 2)
vvhere the summation on repeated indices is considered.

It yields a scalar and corresponds to the classical inner-product.

A.2 Tensors and product

2nd order tensors are denoted as double-underlined characterH as follows:

Q_ or D"ij (i,j = x, y, z) in components (A.3)

The product of a. 2nd order tensor g and a vector n. is defined as:
(A.4)

A.3 Gradient and Laplacian

A.3.1 Definitions
The gradient is denoted \7 and is such as:
3 3 iJ
'i7 = u"xfz + -3 y fy. + -3 z fz (A.5)
420 Several opera.tort" in n1echanics

The Laplacian is denoted t:. and is defined by:

(A.6)

A.3.2 Examples
'The gradient of a scalar variable u is a vector defined as follovvs: Appendix B
D·u Du Du
\7(u) = -D
x '
!';., + -Dy fy + -Dz fz (A.7)

T·he Laplacian of a scalar variable u is a scalar expressed as:

Synthetic wavelets
(A.8)

The product of the gradient b:y a second order tensor g yields a vector: B.l Ricker wavelet
The Ricker wavelets are defined by considering the derivatives of a Gaussian at various
(A.9) orders (l'v1avroeidiR and Papagcorgiou, 2003; Ryan, 1994).
• The R,icker vvavelet at order 0 (i.e. gaussian) iH thus:
'I'his expression is used in the cquilibriun1 equation (Chapter a). 2
Ro(t) = -A exp [ -7r 2 (t-t,)
•
]
(B.l)
2
The gradient of a vector 1f yield8 a second order tensor 1 \Vritten under inatrix form as
2 tp
follows: • At order 1 (first derivative of Ro(t)), tbc Ricker wavelet is:
Bux Dux Bu:c
2
Dx Dy Dz
R 1 (t) -- -
A 1T -
t-t, [ 2 (t-t.,)
- exp -Jr 2
]
(B.2)
Duv Du
__ Y Duy tp tp
V'(ll) = \711 = (A.10)
[}.y Dy Dz
• AH already detailed in Chapter 4, the 2nd order Ricker wavelet (second derivative
Duz Buz Buz
of Ro(t)) is defined as:
OX oy oz

Si1nilar expressions in cylindrical or spherical coordinateH rnay be foun<l in (Salengon, R 2 (t) =A (27r 2 (t~;,)
2

-1) exp [-7r2(t~;,)'] (B.3)

2001) or (Fung and Tong, 2001).
\vhere ts gives the time corresponding to the inaximum a1nplitude for the 0th order and
2nd order R.icker vvavelets and tp the fundarnental period of the 2nd order wavelet.

The time variations of these R,icker Vi.ravelets are displayed in Fig.B.1 and the related
spectra in Fig.B.2. As shovvn in F'ig.B.1 for the 2nd order Ricker wavelet (botto1n), it is
also possible to define the wavelet breadth tb ' " (Ryan, 1994):

tb =
/Gt
P ~
__ 0.78tv (BA)
7r

tb corresponds to the tirne delay between both secondary peaks of the 2nd order Ricker
wavelet.
 Synthetic v.ravclcts D .1 R,icker wavelet 423

18
0.5
16
[order 0) 14 [order 0)
0.4
CJ 12
"CJ CJ
"CJ
~
"
•M

0.
0.3
"
.µ
•M
~
10

s 0.2
p,
s 8

" " 6

0.1 4
2
0.0 0
0.5 16
0.4
14
[ order 1 j
0.3 [order 1 j
0.2
CJ
"CJ CJ
0.1 "CJ
"
.µ
•M
~ 0.0 "
.µ
•M

s""
~
p,
-0.1
s
"' -0.2 "
-0.3
-0.4 2
-0.5 0
30
0.4

0.2 25 [order 2 j
0.0
CJ 20
"CJ CJ
"CJ
"
.µ
•M
~
p,
-0.2
"
.µ
•M
~ 15
-0.4
s p,
s
" -0.6
"' 10
-0.8 [order 2 j 5
-1.0
-4.0 -2.0 0.0 2.0 4.0 00 0.2 0.4 0.6 0.8 1.0 1.2 1.4
time (s) frequency (Hz)
Fig. B.1: Ricker wavelets at various orders: Oth order (top), 1st order (mid- Fig. B.2: Fourier spectra of Ricker wavelets at various orders: Oth order
dle) and 2nd order (bottom), also see Chapter 4. (top), 1st order (middle) and 2nd order (bottom).
424 Synthetic wavelets B.3 1favroeidis & Papageorgiou vvavclct 425

B.2 Gabor wavelet B.3 Mavroeidis & Papageorgiou wavelet

The Gabor wavelet is defined as follows (Mavroeidis and Papagcorgiou, 2003): The wavelet recently proposed by l\.1avroeidis and Papageorgiou (2003) is such as:

4Jr f t 2 ) cos(27rfpt + v)
f(t) =A exp -~
2 2
(B.5) f(t) = 2
A (2rcf )] cos(27r J,,t + v)
[1 +cos ---;; t (B.6)
(
1,vhere: A is the arnplitude,
where: A is the an1plitudc, fp is the fundamental frequency)
fp is the fundarnental frequency, v i.s the phase angle,
v is the phase angle, I' characterizes the oscillatory character.
1 characterizes the oRcillatory character. The l\ 1avrocidis & Papagcorgiou vvavelet and its Fourier spectrurn are displayed in
1

The Gabor vvavelet and its Fourier spectrum are displayed in Fig.B.3 for various I' Fig.B.4 for J,, = 0.2Hz, ')'=4 and v = 0. Mavrocidis & Papageorgiou (2003) performed
values (with fp = lH" and 7/ = 0). rnany seisrnic Viraveform identifications using their wavelet and comparing with the Gabor
wavelet.

1.0
0.8
0.6
0.5
0.4 ............ .
'-~ ··,, ,•,. ~ .
0.2 -········· ·····~~.~~·~·~~\- ·····-'·····--,:---~--·..,
" ""'"
0 ·-
~\ (
', \". .. . ""{".
. I•
. ...:... ,;_.
,'/-:-.. ~
~"
·~
p,
0
....

', \ l I; Ei
"'
:~: · · · · · · · · · · · · · · · · \:~ f ·············••\~f? .
-0.5

-0 8-2 -1.5 -1 -0.5 0 0.5 1.5 2 -1.0

-10 -5.0 0.0 5.0 10
time (s) time
180
160 500
140 f,=L v=O
400
120 y=2
""""" 100 y=4 """'"
~
·~
300
~
·~
~ 80 y=6 ~
p,
p,
Ei 60
Ei 200
"' 40
I : "'
/
I :
I '
: "" \ 100
20
0.0
,
/
": \
""··' ' ..... 0.0
J \
0. 1.0 2.0 3.0 4.0 5.0 0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
frequency (Hz) frequency (Hz)

Fig. B.3: Gabor wavelet (top) and its Fourier spectrum (bottom) for fv = lHz Fig. B.4: Mavroeidis & Papageorgiou wavelet (top) and its Fourier spectrum
and 7/ = 0. (bottom) for fv = 0.2Hz, ')'=4 and u = 0.
426 Synthetic \¥avelets B.5 Kiippcr vvavelet 427

B.4 Generalized Rayleigh ·wavelet B.5 Kiipper wavelet

The generalized 11ayleigh 1vavelet is defined as followH (:t'vlavrocidit'l and Papageorgiou, The Klipper \Vavclct is defined as folln1vs (?viavroeidis and Papageorgi.ou, 2003):
2003):

·
f(t) = A(-l)k exp
(·
[i (v + ~)]
2wf t)k+l
(B 7) f(t)=A sin m- [ ( Jrt)
T m+2
Jrt)]
- - -ms i n ( (m+2)-
T
(13.8)
i+--+
where: A is the arnplitudc,
vvhere: A is the amplitude,
T is the duration,
f~) is the fundarnental frequcncy1
rn controls the nun1bcr of half-cycles.
v is the phase angle,
k controls the number of lobes". 11
The KUpper 1.vavelet and its Fourier spectrun1 arc diRplaycd in Fig.B.6 for various rn
values (with T = 5s).
The generalized Rayleigh vvavclct and its Fourier spectrum are displayed in Fig.B.5
for vario11s k values (1,vith fp = lHz and u = 0). 2.0

0.8 1.5 ----------- . - - ... '. - - - - - - - - - - - .... '- --- --- . - .... _,-•.

0.6 -, I/
- --- - -
'
:\
~ \" . 1.0 ---- .. ·····-_-.-.;:-~~<~.. :--.,<·-- ---- - --····- --' -----····\;/_,..__ ~~~
0.4 ·-:--/-- ---~\-· CJ 0.5 --- -----~·~.-':. ___ \ ___ --"' -----.'·· --- /-', __ :-,,,._
: I \ "d ,-;.'/ \ \\ II ". ',...._
0.2 . --- ····:-r·
0.0 L--.:-'.-_---- -----~----- ..\-------+
CJ
"d 0 i,,,.,.,..,._..,~~~'-- - ' . .. .. ... . l "
.µ
·~
~

\.,_ '\ //
·-r·-----------\ ---------- ___ ,....... ·-· ------~

\.... ://
"
:;;
~ -0.2 - ····-·'-------··'···-·---~-~\ -- - ....... __ / __ s"" -0.5 ... ... ... . . . . ... . . . . .. . .

~ , \ I :
"' -1.0 - .. ... - - - - - - - - '. - - - - .. ... -\,- - - - ~ - '\:/·" / /
s -0.4 . - ... ". ,_ - _\_ -
\
: -- .. / .. :
I
- " " ""

--- ----- -· .. :~.<--,·---"'- . . . _,,, ' ,'\ /

"' -0.6
\ -1.5

-0.8 - - .... _:_ - - - - . - " " ~ "_:::::: ::: ::::::~:~ ~ (.~ - -2.00
2 3 4 5
-1 01 time (s)
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 300
time (s)
140
,, 250
,... ,,.···\,
,\
I \
I :
, \ ,
.. T=5; O<t<T
120 I \ \
I I 200 I : \ .. m=2
I I f,=L v=O CJ I I
100 I \ "d I
I m=3
-k=2
CJ
80
I I
I --- k=3 "
.µ
·~
150 I
I
I
\
m=4
"":;;" ~

s
~
100 I
I
I \
\
~ 60 I \
~
s \
\
"' 50
I
I \
\
40 ~--->.,
"' I \ \
I \ ,' \ I '. ' \

20 I \ ,' \I ',,' \ ..., ....-----

I
I '' ' 0.4 0.8 1.2 1.6 2.0
0.0 frequency (Hz)
0 2.0 4.0 6.0 8.0 10.
frequency (Hz)
Fig. B.6: Ktipper wavelet (top) and its Fourier spectrum (bottom) for T = 5.
Fig, B.5: Generalized Rayleigh wavelet (top) and its Fourier spectrum (bot-
tom) for fp = lHz and v = 0,
428 Synthetic wavelets B. 7 Mori et wavelet 429

B.6 Ormsby wavelet The following vvavclct types \Vere mainly proposed as analyzing \vavelets for tin1e-
frequency signal processing (Daubechies, 1992; Jaffard ct al., 2001).
The Ormshy wavelet is defined as follows (Ryan, 1994):

f(t) =A { [
f 1C fl .
. 4-fs
sinc 2 (7rf1l) - -7r-j23- sine 2 (7rf3t)
f4 -h
l B.7 Morlet wavelet
(13.9) The analyzing wavelet proposed by Morlct et al. (1984) may be defined as follows:
- [ Jrfjf sinc 2
(Jrj,t) - Jrg sinc 2
(7rfit)]}
12 - 1 h-fi
where: f1 and f2 define the lo-Vi.rer range frequency band,
J(t) =exp (-t;) cos5t (B.10)

fs and f4 define the higher range frequency band.

Various scalings inay be applied to thiR wavelet. The J\1orlet 1vavelet and its Fourier
T'l1e normalized Ormsby wavelet and its Fourier spectrun1 arc cliHplayed in Fig.B. 7 for Hpectrun1 arc displayed in F'ig.B.8.
various higher frequency values h and /4 (with fi =5Hz and f2=10Hz). For the Ormsby
vvavelet, the spectrun1 is very fiat in the frequency band of interest.
1.0
,:
f,=40Hz, f,=45Hz . . . . . (' 1.. 0.5
0.8 .. : ....••.. J

0.6 f,=20Hz, f,=25Hz

--------
:
I

I
1 ""'"
~
;:J
0
""'"
·~
0.4 .......... · . . . . . . . . . . '............:. . . . . . . . / ~
p,
~
;:J
·~
0.2 .. /.
I
s
~
p,
s 0
-0.2
.......... ~ -.......... :........... :- \ . . . I.
I
" -0.5

"' -0.4
.......... '
. I I
...........
. I
·........... .' .. \ ..

- 1 ·0 c_-~-3~.o,---_2~.~o---~1'c.o--~oec.o,---c>1.~o--"2'c.o---03ec.o,--~
_:_ -\/
-0.6
-0.8 L_--~--~--~--~--~--~--~----' time (s)
1400~----~---~----~----~------,
-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2
time 1200
7000

6000" I . 1000
I

5000
I
I
I
f;=40Hz, f,=45Hz ""E'" soo 1- I

""'"
·~

I
£;=20Hz, f,=25Hz
cas 600
.
~
"' 4000" I

s
·~
~

3000
I
I
I
I
" 400

"' 2000 I
I
200
)
.

I \
1000 I . 0.0 0 2.0 4.0 6.0 8.0 10 .
I
0.
0.0
\ frequency (Hz)
0 20 40 60 80 100
frequency (Hz) Fig. B.8: Morlet wavelet (top) and its Fourier spectrum (bottom).

Fig. B. 7: Ormsby wavelet (top) and its Fourier spectrum (bottom) for fr =5Hz
and h=lOHz.
430 Synthetic \\ravclets B.8 Meyer wavelet 431

B.8 Meyer wavelet

Other types of a.naly:;>;ing vvavelets have been propoRed from the theoretical \vork of J\1eyer
(Jaffard et al., 2001). The decornposition of a signal on a set of vvavelets raiseH the need
for the definition of orthogonal 'Wavelet bases. The set of l'vieyer \Vavelcts 'r/;jk is defined
in the frequency do1nain as follows:

'</!jA(f) = 2~'</J (2if - k) (B.11)

\vhere: ,P(f) =exp (-if) hi/) (B.12)

vvith: if f E [0, 2; l 1(f) = 0

if f E [2".
;) . 3
4Kl 1U) = e (~~ -1)
if f E [ 4w 8ff
3 ' ;)
l 1U)=e(2-!~)
(B.13)

if f E [s;' +oo [ 1(f) = 0

One of the best set of e fonctions was found to be as follows (Jaffard et al., 2001):

J{ .x"(l
en (f) = ".",--1 - x)"dx
~-_c__­ (13,14)
fo xn(l - x)ndx

Considering Eq.(B.12) for n = 1 (i.e. B(f) = 3/2 - 2f 3 ), the 1(f) function and the
Fourier r::;pectru1n of the l\/Ieyer vvavelet arc displayed in Fig.B.9 (top and middle resp.).
The l\1Ieyer Vv'avelet in tin1e domain, con1puted using the inverse Fourier transfor111 1 is also
plotted in Fig.B.9 (bottom).

When compared to the 1forlet v,ravelct, the shape of the l\!Ieyer wavelet leads to
lo\ver arnplitnde secondary peak.s. One of the rnain intereHt of l\1eyer vvavelets is that it
leads to an orthogonal base of vvavelets which is very interesting for decomposition and
synthesis purposes (Jafford et al., 2001). D;;ubechies (1992) also proposed various types
of v.ravelets.

Fig. B.9: Meyer wavelet: 1(f) function (top), Fourier spectrum (middle) and
time-domain signal (bottom) for n=l.

!t

II ,1
432 Synthetic \Vavelcts

B.9 Double-J\;f wavelet

In the field of train induced vibrations, the doublc-1\!l wavelet characterizes the excitation
induced by the boggie on the track. It is defined by two gaussians (Ilickcr signah;; at order
zero). Al Shaer ct a.l. (2008) proposed the following combination:

QY [ (:'-"''
F(t,Q,v) = - - x "'
2
+x ,,,_o-LJ'
,, l (D.15) Appendix C
\vhcrc Q i.s the load supported by the axle, v is the train speed, t is the tin1c 1 d the
<liHtance between t-1,vo Hleepers, a a critical distance, L the distance between t>.vo axles of
a boggic, X and Y arc <lirnenHionless variables depending on the clcu.;ticity modulus of
the soil.
Spectral analysis and filtering
The doublc-1\!I wavelet proposed by Al Shaer et a.l. (2008) is depicted in Fig.B.10.

C.l Fourier transform

0.007 The mathc1natical propcrtieH of the Fourier transforrn have not been discussed yet. The
analysis of signals in the frequency domain is very often rnade using Fourier transforrns:
Q) 0.006 arnplitude .spectra, tranRfer functions, cohcrency1 filtering, etc (SchcrLaurn, 2006).
",.,
"'
~
00
00
Q)
0.005

0.004
C.1.1 Definitions
.s"
00
.For an integrable function x(t), the Fourier transforrn x*(w) exists and n1ay be \\rritten
<I under the following form:
Q) 0.003
s • direct transforn1:
'8 0.002
x*(w) =
r+= x(t) exp(-iwt)dt
J_= (C.l)
0.001
• inverse transforrn:
0. L_~"---~--~---~---="'----~__J
0. 0.02 0.04 0.06 0.08 0.10 x(t) = -l 1+= x*(w) exp(iwt)dw (C.2)
time (s) 21T' -=
The Fourier transform x* (w) is generally called the complex spectr11m, of the Rigna1
Fig. B.10: Double-1\!I wavelet (dimensionless load F/Q) considered to model
x(t). The function x*(w) being complex, it is often studied in terms of amplitude and
dynamic loadings by trains (Al-Shaer et al., 2008).
phase (i.c modulus and argument) through the arnpl-it,ude spectrum Ix* (w) I and the phase
spectrum arg (x'(w)).

As discussed in Chapters 2 and :1) choosing a e+iwt tirne harrnonic dependence is

compatible vvith such definitions. Conversely, if a c-iwt harrnonic dependence is chosen,
one must use a direct transforrn to calculate x(t) fron1 x*(w) and an invenc;e transforn1
to compute x'(w) from x(t).

These definitions 111ay be usccl in the caRe v,rhere the signal has a. continuous amplitude
and a continuous tirnc base. The experimental signals arc digitized Vl.rhen recorded: they
434 Spectral analysis and filtering C~.l Fourier transform 435

are at the san1c tin1c sarnpled (discrete time) and quantified (discrete arnplitude). The C.1.4 Fourier transforms of synthetic wavelets
analysis of experimental signals in the frequency do111.ain is thuR perforn1ecl thanks to the
For synthetic wavelet.s, the theoretical Fourier transforn1s (direct or inverse) rnay be de-
discrete Fourier transform (Mari et al., 1999; Schcrbaum, 2006).
termined using Eqs (C.J.) and (C.2). As shown in Fig.C.l, the theoretical amplitude
spectrun1 is dctcrrnincd against po.sitive as well as negative frequencies.
C.1.2 Main properties
Denoting F the Fourier transfor1n, its n1ain propertie.s arc the following (Scherbaum,
2006): 4.0

F
• Involution: x* (t) --+ x(-f) Q) 3.0
F
'+'
""'
• symrnctry: x(-t) --+

F
x*(-f)
" 2.0
"1
~
x* (f)e±iwt()
• advance/delay: J;(t±to) --+
s
• derivation:
d"x(t) F
--+ (iw f)"x* (f)
"' 1.0
dtn
F dnx* (!)
(-2i1rt)n.r,(t) --+
0.0
df" -£max 0.0
• convolution: x(t) * y(t) _,
F
x* (f)y* (!) frequency
F
x(t).y(t) --+ x*(f) * y*(f)
Fig. C.l: Theoretical Fourier (amplitude) spectrum of a Ricker wavelet.

\i\Then computing nurnerically the Fourier transforrn of a discrete signal (e.g. Fast
C.1.3 Usual transforms Fourier Transforrn (FFT))) the values corresponding to the negative frequencies are
The Fourier transfor111s of the following usual firr1ctions are (Scherbau1n 1 2006): tra1rnlated to the right part of the spectrum (fictitious higher frequencies). As depicted
in Fig.C.2 for a 2nd order Ricker vvavelet (tp = 0.5s) 1 the arnplitude spectrum (top) is
F obtained under a different forrn 1.vhen compared to the theoretical Fourier spectrurn dis-
o(t) --+ 1
played in Fig.C.1. The first half of the nun1erical transform corresponds to the Hpectral
F
1 --+ o(f) con1ponents of the signal het1.veen 0 and +f max 1.vhereas the second half is related to the
F spectral components between - !max and 0 (periodic signals). In Fig.C.2, the real part
e±iwt --+ Ii(! 'f fo) (middle) and the imaginary part (bottom) of the Fourier spectrum are alw plotted.
F 1
sign(t) --+
iw Ttanslating the second part of the spectrum to the left of the first part, the real and
F Tsin Jr JT irnaginary parts of the Fourier spectrum arc as depicted in l<'ig.C.3 (Ricker wavelet vvith
IIr(t) --+
1rfT tp = 2.0s). As sho1.vn in this figure, the real part of the spectrurn is even whereas the
sin Jrfot F imaginary part is odd.
--+ -}cIIfo(f)
"fat
Ar(t)
F
--+ T (sin
1rfT
"tr)" C.1.5 Wave propagation in viscoelastic media
F
In order to sin1ulate vvave propagation in viscoelastic rnedia, it is convenient to \.Vork
e-1T't2 e-1T'f2
--+ in the frequency dornain and to use the exact expressions of the complex modulus and
vvavenurnber (Hee Chapter 2). The analysis rnay be only performed for the first half of
the Fourier spectrum (fron1 0 to !max)·
where ITT denotes the step function and Ar the triangular function of duration T.
436 Spectral analysis and filtering C.1 Fourier transfor1n 437

As sho-,,vn in Fig.C.3, the real part of the spectrurn if; even vvhereas the imaginary
part is odd. The Fourier spectrum of the propagated t->ignal rnay thus be cornputed using
the fin·;t part of the spcctrun1 of the original signal. The second part of the spectrurn of
4.0 ( mod~lus) the propagated signal may then Le computed by taking the conjugate of the values of
the first part. The inverse Fourier transforn1 finally yields the propagated signal in tirne
'
' don1ain.
3.0 '''
"""'
"
'

'''
µ
·~ 20.--~~~~~~~~~~~~~~~.--~~~~~~~~~~~~~~~.
~

''
s"' 2.0
'' 15 [real part)
" ''
10
''
+fmax :' -fmax ~ 5.0
''
~
~ o.01r--""""

4.0 ~"' -5.0

( real part )
-10 even
2.0
-15
"""' -20L--::;:::==:::;::;:::;=::::;;;::====3.._L~-~~-~~--.......J
µ
" 0.0
;.:::; fictitious higher frequencies
s"' 20r-~~~~~~'r--~~~~~~~.--~~~~~~~~~~~~~~,

" -2.0 15 (imaginary part)

10 odd
-4.0
~ 5.0
B
~ O.Or--~"'

(imaginary part) "'

~ -5.0

-10
-15

~186 18.8 19D 19.2 19.4 19.6 19.8 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
frequency (Hz) frequency (Hz)

Fig. C.3: FFT of a Ricker wavelet with tp = 2.0.s at lower frequencies and
translated 11 negative 11 frequencies showing the even real part (top) and the
0. 5.0 10. 15. 20. odd imaginary part (bottom).
frequency

Fig. C.2: FFT of a Ricker wavelet with tp = 0.58: modulus (top), real part
(middle) and imaginary part (bottom).
438 Spectral analysi8 and filtering C.2 Filtering 439

C.1.6 Fourier transforms of actual signals C.2 Filtering

\i\le have chosen a record from the Nice Feb.2001 earthquake to compute the f'ourier C.2.1 Classical filters
spectrum and briefly illustrate filtering techniques. The accelerogran1 recorded at the Various filtering techniques are available: Buttcrworth 1 Chebychcv) Desr.;el, etc (Boore,
ALS alluvial site (sec Chapter 5) is displayed in Fig.C.4 (top) '1nd its amplitude spectrum
2005; Mari et al., 1999; Scherba.um, 21106). For instance, the Butterworth filter is designed
(i.e. modulus of the Fourier transform) is plotted at the bottom. Most. of the large
to have a frequency response as flat aH possible in the bandpass of interest. T'lre nth order
spectral cornponents are found bet\veen 1 and 8 Hz. Bntter\vorth filter is defined Ly the following gain:
2
G2(w) - Go
0.15 - l+ ( )2n
~
We
(C.3)

0.10
0 T·hc gain of Buttcrvvorth filters at v·arious orders is displayed in Fig.C.5 as a function
w
s
~

0:
of frequency.
0
·~
+'
ro
~'"
(l)
(l)
0.9
u
u

" -0.10
0.8

0.7

-0.15 0.6
10 15 20 25
time (s) '1
·~
0.5
14 "'
bJJ

0.4
12
8 0.3
"'"
+'
u
(l)
10
P<
0.2
w 8.0
(l)
0.1 ·.. ····- .... ···· .....
"""
- --·--- ·..: · --..: · ··
+' 6.0
·~

"'8"' 4.0
0
0 0.2 0.4 0.6 0.8
frequency
1.2 1.4 1.6 1.8 2

2.0
Fig. C.5: Gain vs frequency for low-pass Butterworth filters at orders 2, 3, 4
o.oo. 4.0 6.0 8.0 and 5.
frequency (Hz)

Fig. C.4: Accelerogram (top) and its amplitude Fourier spectrum (bottom)
recorded at the ALS site in Nice, France (Feb.2001 earthquake). C.2.2 Filtered signals: examples
A8 depicted in Fig.C.6 1 the accelcrogra1n recorded at the Af_,S station duTing the Feb.2001
earthquake (Fig.C.4, top) is filtered irning va.rious Butterworth filter' with J, = 8.0 Hz.
440 Spectral analysis and filtering C.3 Hilbert transforn1 and envelope curve 4'11

The influence of the type of filter on the final wavefor1n of the filtered signal appears to The Hilbert trnrnform J-i (s(t)) of s(t) allows to separate the information related to
be significant (Doore, 2005; Scherbaum, 2006). phase and amplitude. The complex function S(t), defined as S(t) ~ s(t)+ih(t), represents
the complex envelope of the signal (this function is often called the analytical signal).
0.04

0.03
- - no filter C.3.2 Envelope curves
---- n=2
Fro1n the Hilbert transform and the con1plcx envelope of the signal S(t) 1 its envelope
" m
O.Q2 n=4
·············· n=S
~ curve may be easily dctcr1nincd (Farnbach, 1975). The infor1nation related to amplitude
] :; '
/v~
0.01 are given by the modulus of this function I S(t) I which defines the envelope cvrve. The
'1
0
... ..
i\ inforrnation related to the phcu.;e are taken fro1n the argun1cnt of S'(t). ·The envelope curve
_,_,
·~ 0.0 , -
charactcrize.s the propagation fro1n the energetic point of vie1,ivr and allowH the esti1nation
"',.,
CJ
-0.01
~!
of the group velocity.
Cl
u
u ~.
-0.02
"' A typical envelope curve is propoHed in Fig.C.7 for a filtered signal recorded dur-
ing centrifuge dyi:1arr1ic experi1nents (see Chapter 3). The peaks in the envelope curve
-0.03 -
indicate the various \Vave packets characterizing the 1,vay the energy of the signal is prop-
-0.048 agated.
.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0
time (s)

Fig. C.6: Accelerograrn recorded in Nice (France) and filtered using various envelope
Butterworth filters. curve

-- ...
-/
---. ,/ ·-.

C.3 Hilbert transform and envelope curve ;:, -- --- -- -- -- -- --

C.3.1 Definition
The Hilbert transform J-i of a signal B(t) is defined as follows (Aki and Richards, 1980;
Scherbaum, 2006):
. .
1
J-i (s(t)) = -
l+oc s(T)-dT (C4) "-·'
7f. -= T - t 4.0 5.0 6.0 7.0 8.0 9.0

'The practical computation of the Hilbert transfor1n is made directly considering the time (ms)
complex spectrum of the signal, s*(w), and its spectral Hilbert transform, J-i (B'(w)):
Fig. C. 7: Typical envelope curve computed by the Hilbert transform from I
I
11'

J-i(s*(w)) = -isign(w)s*(w) (C.5) accelerogrruns recorded during centrifuge tests (Sernblat and Luong, 1998).

where sign(w) is the sign of w.

111
To co1nputc the IIilbert transfor1n, one firstly considers the ti1nc signal s(t). Its
Fumier transform s* (w) is then multiplied by a step hmction (sign( w)) to get the fre-
quency domain Hilbert transform. Using an inverse Fourier transform 1 the Hilbert trans-
11t.
form J-i ( s (t)) is finally computed.
·11..

I
I
'I

il i I
l1lli
Appendix D

Propagating waves: duration,

velocity, echoes

D.l From acceleration to displacement

D.1.1 Integration of accelerograms
It is often neceHRary to integrate accclcrograrns to obtain velocities or displacements.
When cornputing such integralR 1 it iH n1andatory to ca.refully account for the integration
co11Rta.nts (Boore and Bo1nmer 1 2005). ThiR problem iH crucial t>ince the direct integration
of acceleration signals leads to spurious ramps in resulting velocities or displacen1ents (see
Fig. D 1, top).

D.1.2 Baseline correction

One -,,vay to correctly co1npnte the integrated 8ignah; consistR in perforrning baseline cor-
rections: the trend observed in an integrated signal is corrected by chooHing an adequate
baseline corrections. As shovvn in Fig. D.1 1 various baseline corrections methods are
available to correctly integrate acceleration signals: n1ulti-scgment baseline corrections
rnay for instance be performed in sorne ca,ses (Doure and Dornn1cr, 2005).

D.1.3 Spectral methods

As discuascd by Boore and Bon1mer (2005) 1 filtering techniqueH (e.g. lovv-cut filters to
reduce lcJ\v-frequency noise) arc also an interesting 1nean to perform accuTate integrations
of accclerograms.
444 Propagating 1.vaves: duration 1 vclocit:y, echoes D.3 Signal duration 445

6 no baseline adjustments field of earthquake engineering. Hov;,rever, several other irnportant features of the recorded
4 signals arc crucial for the detailed characterization of the ground motion (number of
cycles, duration, etc).
2
Loma Linda. 551C: Velocity (cm/sec)
0 0.1
0.5
;;---
w
0 ~

fit from 10.3--12.5; tvo = 10.33s

$ 0
-0.5 .s"
~

~
"',.,
Q)
-0.1
0 Q)
'-'
""'
-1 fit from 12.5--22.2; tvo = 12.565 sec
-0.2 ~----7 PGA=-0.2028m/s 2
-2
0 10 20 30 40 50 60 70
1.5
0
1.0
-0.5 fit from 22.2--23.2; tv 0 = 22.01 sec
s:$
0.5
-1 ~

"s
Q)
0
0.1 Q)
'-'
0 "'w
~ -0.5
-0.1 ·~

fit from 23.2--45.0; tvo = 23.335 sec

""' -1.0
~-------i> PGD=-1.197mm
0 10 20 30 40 50 -1.5
0 10 20 30 40 50 60 70
time (s)
time (s)
Fig. D.1: Various baseline corrections to integrate accelerograrns and avoid
unrealistic ramps (Boore and Bommer, 2005). Fig. D.2: Peak Ground Acceleration (top) and Peak Ground Displacement
(bottom) of a seismogram recorded during the Feb.2001 Nice earthquake
(RAP, CETE-Med).

D.2 PGA, PGV, PGD

Peak ground rnotion values are often used in various fields of application in ter1ns of
acceleration (PGA), velocity (PGV) or displacement (PGD). As shown in Fig. D.2 for
the Fcb.2001 Nice earthquake, the PGA (top) '1nd the PGD (bottom) are very easily
cstirnated fro1n. the maxin1u1n values in the curves. It is an irnportant feature of the
ground motion since it 1nay he used for the con1parison of ground inotions due to various
earthquakes as vvell as for the determination of the responRe spectra \Vidcly used in the
D.3 Signal duration
The peak ground rnotion (PGA, PGV or PGD) is not Rufficient to fully characterize the
features of a scisrnic VI.rave or vibration induced in the soil. 'l-le may have a large peak but
very few secondary peakfl or, conversely, numerous inter1nediate amplitude peaks bnt no
very large ones. It is thus interesting to estilnatc the signal duration, to investigate the
\Vay the energy of the signal is distributed in tirnc.
446 Propagating v.raves: duration, vclocity1 echoes ]) .4 Estimation of wave velocity 447

The duration d of a signal s(t) is defined from its energy E:.,(t): D.4 Estimation of wave velocity
t \i\ravc velocity is the key pa.ra.rnetcr to investigate wave propagation in solids. However,
E:,(t) = s 2 (t)dt (D.1) as already discussed in Chapters 2 and 3, various typet> of velocities rnay be considered
In
.Q
and the estirnation on actual recordingR is often difficult due to such phenornena as re-
VariouR definitions of duration may be considered (Dornrncr and l\fartinez-Pereira, flections, attenuation, dispersion, scattering, etc.
1999). As suggested by Trifunac and Brndy (1975), the signal duration may be chosen
as the tirnc interval [t 5 %; t 95 3L t 5 3 and t 95 3 being the tirncs such as: E.~(t53) = 5%Emax
and £ 8 (t 95 3) = 95%Ema:i:· This signal duration is often called the 'rrifunac duration or D.4.1 Peak to peak estimation
the 5%-95r7c. duration (Bon1mer and J'viartiucz-Pcreira, 1999).
A si1nple method consists in directly estin1ating the velocity on the Rignals recorded in
the experin1cnts. The peak to peak estirnat'ion is the sirnplcst 1nethod to esti1nate the
Jn l~'ig. D.3, a seismogran1 recorded duTing the Feb.2001 Nice earthquake is considered
¥.rave velocity, but it is necessary to have a good sinlllarity of the peak shapes in each
to co1npute the signal energy. Estimating the Ei%Emax and 95%Ernax thTeshold 1 the
signal. \Vhen dispersion is significant 1 such a sin1ple procedure is generally difficult to
duration rnay be easily deter1nined: d = 22.Ss.
apply because the peaks shape 1nay be distorted during the propagation process.

C' Figure D .4 displays several accelerogran1s recorded during centrifuge experin1ents

w
~
0.1
E (Chapter 3). The peak to peak estimation may be easily performed, but the peak shape
'1 0 nevertheless changes during the propagation process.
0
·~
+'

"'"'
Q)
~
Q)
-0.1

u
u -0.2
"' 0 10 20 30
O.D3
T
fo s( t) 2 dt 95%

0.02

d=22.5s
0.01

5%
---------------------------------------
10 20 30 40 50 60 70 0.0 1.0 2.0 3.0
time (s) time

Fig. D.3: Definition of the 5-953 duration for a seismogram recorded during Fig. D .4: Direct determination (peak to peak) of the propagation delays on
the Feb.2001 Nice earthquake (RAP, CETE-Med). time signals recorded in centrifuge experiments (Chapter 3).
448 Propagating VI.raves: duration, velocity1 echoes D.5 Detection of reflected vvavcs and echoes 449

D.4.2 Cross-correlation function D.5 Detection of reflected waves and echoes

The cross-correlation function allo-\vs a more global analysis of the transmisi:lion of a 1.vave
through various paths or processes. Denoting x(t) a.nd y(t) the input a.nd output signals D.5.1 Autocorrelation function
respectively, the cross-correlation fnnction R:ry(T) is defined under the following form ·The autocorrelation function Rxx (T) iR defined under a form siruilar to that of the cross-
(Mari et al., 1999; Scherbaum, 2006): correlation function, that is:

R,,(T) = lim -
T
T--+=
]1T x(t)y(t + T)dt
0
(D.2)
Rxx(T) = lim -J lT x(t).x(t + T).dt (D.3)
T-+oo T. O
'fhe cross-correlation function leads to the propagation delays of a 1.vave through var-
ious paths. It al.so allo\vs to characteri~e the silnilarities bet\veen t\vo signals. Figure D.5 This function is equivalent to a cross-correlation function bet\veen the signal and it-
displays a crosH-correlation function co1nputed from signals recorded during cross-hole self. It allov,rs to sturly the sin1ilarities between different parts of the signal. It is thus
testH (Chapter 3). The delay bet¥.reen tvvo signals is directly estimated fron1 the abscis.sa possible to evidence son1e periodicities or the influence of reflected waveR or echoes due
of the largest peak. to propagation along rnultiple paths (Scherban1n 1 2006; Sernblat, 1995).

The peak to peak estirnation gives very poor inforni.ation about the propagation pro-
cess. '!"'he cross-correlation function is a mueh powerful tool Lut it rnay be influenced by
the 1,vave reflections on the inedium boundaries or sornc heterogeneities in the rnediurrL
In such cases, it iH mandatory to analy?;e this specific problen1.
0.8
0.6
Since the crosR-correlation and the auto-correlation are defined as convolution prod-
ucts, they rnay be easily cotnputed in the frequency dornain using the Fourier transfor1n.
D.5.2 Real cepstrum
'l'he cepstru'm is a powerful tool often used in the fields of acoustics, image processing,
speech syntheHis or inechanical engineering (Kunt, 1986; Sessarego, 1990; Tubach, 1989;
Ulrych, 1971). The real cepstrum of a signal is defined as a double (uirect and inverse)
Fourier transforn1 under the foll(nving forn1:

C,(t) = F-r [log I F(s(t)) II (D4)

0.4
As the autocorrelation function 1 it allUVi-'S the detection of echoes or periodic excitations
"'
0
·~
+' in eon1plex signals (Sen1Llat and Luong, 1998). AR it will be shovvn later, it is possible
"'
~
Q)
;.,
0.2 with Huch techniques to cornbine the complex cepstrun1 vvith ccpstral 1JJindowing tech-
;.,
0
niques to remove echoes from a signal. In certain cases (delay of the reflected \vave clo.':le
u
w 0 to the pseudo-periods of the signal), the direct and reflected \Vaves n1ay be cornhined.
w The cepstral rnethod is powerful to analyze such cornplex situations.
0
;.,
u
-0.2

D.5.3 Time phase

-0.4
The time phase i8 computed by using the Hilbert transform, h(t), of the signal s(t).
'_fhis function allo1,vs to separate, in the signal, inforn1ation related to the a1nplitude and
-0.6 infor1na.tion related to the phase. If the reflected wave arrivals cannot be pointed out
0 0.005 0.01 0.015 0.02 0.025
in the tirne signal, it rnay be detected in the tin1c phaHe diagram (Scn1blat, 1995). The
time (s)
derivative of the tin1e phase also allovvs an efficient detection of echoes or reflections.
Fig. D.5: Cross-correlation function for two signals measured during cross-
hole tests (see Chapter 3).
Appendix E

Echo removal by homomorphic

filtering

E.l Basic principles

The hornornorphic filtering o:f signals re~n1lts fron1 a generalization of the theory of linear
syste1nR. The rnain goal is to change the vectorial space in order to trans.forrn convolution
operators in addition operators. It is thus necessary to find a hornornorphism associating
the convolution to the addib:on operalor. The convolution in tirne becoming a multiplica-
tion in the frequency don1ain (sec appendix C), v;,re can use the logarith1n to relate both
operators. The homomorphism thus results from the cornposition of two transforrnations
as follcn.vs:
Fourier complex
transform logarithm
* operator = x operator
= +operator
This transformation replaces the convolution relation by a simple addition. the ho-
j

rnornorphic treatment is widely used in speech processing (Tubach, 1989) since it allows
the characterh.,ation of the noise source.

In scisn1ology, Ulrych (1971) used the hornomorphic deconvolution to analyze the at-
tenuation and dispersion of elastic 1,vaves separating the sources terms and those due to
the propagation. Its 1nain advantage is that no preliminary assun1ptions on the propa-
gation process are needed.

E.2 Computation of the complex cepstrum

The definition of the real cepstrum Cr is given in §D.5 through Eq.(D.4). In this expres-
sion, the logaritlun is real valued since it is co1nputed frorn the rnodnlnt'l of the Fourier
transforrn, narnely the amplitude spectrurn. To compute the cornple.T cepstrum C 1 the
 Echo re1noval by homornorphic filtering E.3 Removing echoes thanks to ccpstral filtering .453

corr1plex logaritlun is considered. The con1plete procedure is described in Fig.E.1. E.3 Removing echoes thanks to cepstral filtering
J)enoting :F the F'ouricr transform and Log the cornplex logarith1n, the expression of E.3.1 Principle of the method
the con1plcx cepstrurn C(T) of a tirne signal s(t) is the following:
Removing echoes or reflections can be perfor1ncd thanks to the complex cepstrun1 defined
C(T) = :;::- 1 [Log(F(s(t)))J (E.l) previously. As depicted in Fig.E.2, the complex ccpstrurn displays several peaks due
to the different arrival tirne.s charactcriLJing the various propagation paths (reflected or
If s*(f) is the Fourier transform of s(t), the con1plex logarithm iH directly expreH.sed scattered 1vavcs). To remove such vv~aves, it is possible to remove the correspontling peaks
as a function of its real and i1naginary parts: in the cornplex cepstrun1 thanks to a weighting \Vindow ( ccpstrnl 1Dindow). The inverse
cepstral transform then allovvs to retrieve the time signal free of reflected 1vavcs (Se1nblat
Log[.i'(f)] =Log I s'(f) I +iArg[s'(f)] (E.2) and Luong, 1998; Ul1ych, 1971).
This relation raises the pro!Jlem of the computation of the phase spectrum Arg[ s* (!) J
since it is generally defined in the range [-11, 11]. Furtherrr1ore 1 considering the properties E.3.2 Homomorphic filtering: summary
of the logarithrn, the phaRe spectrum rnust fulfil the follo\i\ring relation:
The different steps of the analysis are the follo1ving:
if s*(f) = s;(f)s~(f), Arg[s*(f)] = Arg[si(f)] + Arg[s;(f)] (E.3)
• time signal s(t),
To solve theHe problems) it iR necessary to have a continuouR phat-ie Rpectrnrn. The
phase iR then called the unwrapped phase. T'hiH procedure is strongly sensitive to the • truncate the initial part of s(t) fort< 0,
accuracy of the data. Tribolet (1977) proposed a phase un\i\rrapping algorithm considering
the expression of its Ueriva.tive. Denoting sR and s; the real and irnaginary parts of the
• rnultiply by a decreasing exponential to H1nooth the phase,
complex spectrun1 s*(f), the derivative of the cornplcx spectrun1 is the folloVi.ring:

·' (!) d,;(f) - "*~(!) "(f') • compute the Fourier transform s·• (f) = 8'[, (!) + i sf(!)
d(Arg[s*(f)]) SR rlf df 81.
(E4)
df I s'(f) I' • cornpute the derivative of the phase:
l)sing this expression and considering several other steps (sec synthesis belo\i\r) 1 it is
possible to compute the complex ccpstrum C(T). d(Arg(s*(f))
(E.5)
df
time complex
signal 11111.+ cepstrum • lovv-past> filtering of the derivative (\i\rit:h 1r1ean value correction)
s(t) e(tl
• esti1nation of the continuous phase, or nn:wrapped phase) Arg(s* (J)) by integration 1

Fourier inverse • computation of C'(f) = log(I s*(f) I)+ iArg(s'(f)) Fourier tramform of the com-
transf. Fourier plex cepstru1n,
s'(t) tr an sf.
complex • con1putation of the complex ccpstru1n C(T) by inverse F'ourier transforrn,
logarithm
Log(.)
C1,ift ( T)
11
• cepstral windo\ving (also called liftering 11) of the co1nplcx ccpstru1n
W(T)C(T)
Fig. E.1: Procedure to compute the complex cepstrum of a signal.
• liftered time signal: sllJt(t) = :;::- 1 [exp[F(CliJt(T))]]
454 Echo rernoval by hornornorphic filtering E.3 Removing echoes thanks to cepstral filtering 45G

logarithm of the Figure E.3 displays an acceleration signal (dashed) recorded during centrifuge exper-
complex spectrum ® iments (Chapter 3). As shown in Fig.E.2 (step 4) for a synthetic signal, the reflected
v,rave arrivals can be detected and re1noved (step 6). The same signal after hornomorphic
signal with unwrapped filtering (Fig.E.3, solid) is nearly free of any reflected wave.
echoes phase
This technique seems to be very efficient for the signals rneasured in centrifuge tests
(Fig.E.3). This n1cthod is very interesting since it can be applied to longer signals, even
I I I I if the re:Hcctcd \vavcs arc cornbinecl vvith the direct waves (e.g. }i'ig.E.2).
0.0 25000

0.0 1.5 3.0 ms 0.0

~'l><Oe . :<;>4'
"\' '!;"\'"\''
0.0 10000
v.:<:>41-
after .
' .,.'
,'"•

fft of the : : filtering

@ complex cepstrum ®
'"-.before

filtering ® after echo

removal
filtering

······· 4.40 4.80 5.20 5.60 6.00 6.40

0.0 25000 time (ms)

Fig. E.3: Removing reflected waves by homomorphic filtering for an actual

signal recorded during centrifuge experiments (Semblat and Luong, 1998).
0.5 2.0 ms

0.0 10000

Fig. E.2: Procedure to remove echoes by cepstral filtering for a synthetic

signal (Semblat and Luong, 1998).

E.3.3 Application to actual measurements

As already shown in §D .5, the (real or complex) cepstrurn allows the detection of reflected
wave arrivals. If the peaks corresponding to the reflections arc rcn1ovcd by 1nultiplying
the cornplex cepstrum by a 1vcighting windovv, the reflections vvill be rernove<l fron1 the
flignal itself after inverse cepstral transforrnation.
 !
Ij
i

References
References

[1] Abe S., Kobayashi Y., Ika.wa T. (1990). Seismic characteristics of the weight-
dropping source, .Ju·urn.a.l Phys. Earth, 38, pp.189-212.

[2] Abraham 0., Chanmrns R., Cote P., Pedersen H., Semblat J.F. (2004). Mechanical
charactcriLJation of heterogeneous soils vvith surface ¥.raves: experi1nental validation
on reduced scale physical models, Near Surface Geophysics, 2, pp.249-258.

[3] Abramowitz M., Stcgun I.A. (1970). Ha.ndbook of mathematical .functions, 9th edi-
tion, Dover PuLlications Inc. 1 N e\v York.

[4] Achenbach J.D., Wm!e propagation in elastic solids, North-Holland, 1973.

[5] Adalier K., Elgamal A.W. (2002). Seismic response of adjacent dense and loose
saturated sand columns, Soil Dynamics and Earthquake Engineering1 22(2), pp.115-
127.

[6] Agarwal T.K., Ishibashi I., Anisotropic ehuitic constants of granular assembly fron1 ~.·i [

vvave velocity 1ncasurcrnents, Advances in JV!icrornechardcs of Gran'ular JVfaterials

(H.H.Shen), 1992.

[7] Aki K., llichards P.G. (1980). Qcuantitative Seismology, Freeman and Co., Ncw-
York.

[8] Aki K, Larner K.L. (1970). Surface Motion of a Layered Medium Having an Irreg-
ular Interface due to Incident Plane SI-I Waves, Jo'Urnal of Geophysical Research;
75, pp.1921-1941.

[9] Al Shaer A., Duhamel D., Sab K., Foret G., Schmitt L. (2008). Experimental
settlement and dynarnic behavior of a portion of ballasted rail\.vay track under
high speed trains, .Journal uf Sound and Vibration, 316(1-5), pp.211-233.

[10] Andersen L., Clausen J. (2008). Impedance of surface footings on layered ground,
Computers and Structures, 86, pp. 72-87.

[11] Andersen L., Nielsen S.R.K. (2005). Reduction of ground vibration by means of
barrierH or soil in1provcrnent along a railway track, Soil Dynamics and Earthquake
Engineering, 25, pp.701-716.
 T
460 REFERENCES
[12] Anderson D.G. (1974). Dynamic mod·ulus uf cohesive soils, Ph.D. Dissertation,
University of J'viichigan.
[13] Arulanandan K., Canclini J., Anandarajah A. (1982). Simulation of earthquake mo-
tions in the centrifuge, Journal of Geolechnical Engineering {A.8.C.E), 108(GT5),
pp. 730-742.
[14] Assirnaki D., Kauscl E., Whittle A. (2000). Model for dynamic shear modulus and
dan1ping in granular 80ils, Jal of Geotechnical and Geoenvironrnental Eng., 126(10),
pp.859-869.
[15] Aubry D., Clouteau D. (1991). A regularized boundary element method for strati-
fied media, 1st Int. Conf. on Mathematical and Numerical Aspects of Wave Propa-
gation {SIAM).
[16] Autenzio 0. (1997), J~'ronliEres absorbantcs et ClCrnents infinis en dynamique, ]\!las-
ter thesis, LCPC/University Paris 6.
[17] Aviles J., Sancbez-Sesm'1 F.J. (1983). Piles as barriers for elastic waves, Journal of
Geotechnical Engineering (A.S.C.E), 109(9), pp.1133-1146.
[18] Backus G.E., Gilbert J.F. (1967). Numerical applications of a formalism for geo-
physical inverse problems, Geophys. J. Roy. Astr. Soc., 13, pp.247-276.
[19J I3adri :~vI.) l\.1ooney H.JVI. (1987). Q n1easuren1ents fro1n cornpresHional seis1nic waves
in unconsolidated sediments, Geophysics, 52(6), pp.772-784.
[20] Baise L.G., Dreger D.S., Glaser S.D. (2003). The effect of shallow San Francisco bay
sedi1nentH on 1,vaveforrns recorded during the l\1vv 4.6 Bolinas California Earthquake,
Bulletin of the 8eismological 8ociety of America, 93, pp.465-4 79.
[21] Balaban M.M., Green A.E., Naghdi P.M. (1970). Acceleration waves in clastic-
plastic n1aterials, International Jo'urnal of Engineering Sciences, 8, pp.315-335.
[22] Bamberger A., Chavent G., Lailly P. (1980). Etude de schemas n'Umeriq'ues pour
les eq11,ations de l}Clastodynaniique linfaire, llesearch report No.Lil, INRJA, France.
[2:J] Banerjee P.K., Ahmad S., Chen K. (1988). Advanced application of BEM to wave
harriers in 1nulti-layered thrcc-di1nensional soil rnedia, Earthquake Engineering and
Struct·ural Dynamics, 16, pp.11141-1060.
[24] Bard P.Y., Cha"elas J.L., Gueguen P., Kham M., Semblat J.F. (2005). Asse8sing
and 'm,anaging earthquake risk - Chap.5 : Site-city interaction 1 Eds: C~.S. Oliveira 1
A. Roca and X. Goula, Springer, 375 pages.
[25] Bard P.Y., Riepl-Thomas J. (2000). Wave propagation in complex geological struc-
tures and their effect.Hon strong ground 1notion; Wave motion in earthqnake eng.,
Kausel & Manolis eds, WIT Press, Southampton, Boston, pp.37-95.
REFERENCES 461
[26] Bard P.Y., Bouchon IV!. (1985). The two dimensional resonance of sediment filled
valleys, Bulletin of the Seism,ological Society of America, 75, pp.519-341.
[27] Bard P.Y., Bouchon M. (1980a). The seismic response of sediment-filled valleys.
Part 1. The case of incident SH waves, Bulletin of the Seisrnological Society of
America, 70, pp. 1263-1286.
[28] Ba.rd PY., Bouchon M. (1980b). The seismic response of sediment-filled valleys.
Part 2. The case of incident P and SV vvaveR, Bulletin of the Seismological S'ociety
of Arneri:ca, 70, pp. 1921-1941.
[29] Basu U., Chopra A.K. (2003). Perfoctly matched layers for time-harmonic elasto-
dyna1nics of unbou11ded do1na.ins: theory and finite-clcrncnt irnple1nentation, c~om,­
puler Methods in Applied Mechanics and Engineering, 192, pp.1337-1375.
[30] Bathe K.J. (1995). Finite Element Procedures, Prentice-Hall, Englewood Cliffs.
[31] Beaty K.S., Schmitt D.R., Sacchi M. (2002). Simulated annealing inversion of multi-
1nodc Rayleigh -Vi.rave diRpersion curves for geological structure, Geophysical Journal
International, 151, pp.622-631.
[32] Beauval C., Bard P.Y., Moczo P., Kristek J. (2003). Quantification of frcquency-
dependent lengthening of seismic ground-rnotion duration due to local geology:
applications to the \Tolvi area (Greece)) Bulletin of the E!eism,ological Society of
America, 93, pp.371-385.
[33] Becache E., Fauqueux S., Joly P. (2003). Stability of perfectly matched layers,
group velocities and anisotropic 1,va.ves, Jal Cornpnta.tional .Phusics) 188(2), pp.399-
433.
[34] Beresnev I.A.) Nikolaev A.V. (1988). Experimental investigation8 of nonlinear seis-
n1ic effects, Physics of the Earth and Planetary Interiors, 5(\ pp.83-87.
[35] Beskos D.E. (1997). Boundary elements methods in dynamic analysis: Part II
(1986-1996), Applied Mechanics Reviews (A8ME), 50(3), 149-197.
[36] Beskos D.E., Dasgupta B., Vardoulakis LG. (1986). Vibration isolation using open
or filled trenches, Computational lvlecha.nics, 1, pp.43-63.
[37] Bettes P. (1992). Infinite elements, Penshaw Press, Sunderland, U.K.
[:J8] Bielak J., Loukakis K., Hisada Y., Yoshimura C. (2003). Domain Reduction Method
for ·Three-Din1ensional Earthquake l\todeling in Localized Regions; Part I: Theory,
Bulletin of the 8cisrrwlogical 8ociety of America., 93(2), pp.817-824.
[39] Bielak, J., J. Xu, and 0. Glrnttas (1999). Earthquake ground motion and structural
response in alluvial valleys, Jal of Geotechnical and Ci'eoenvironmental Eng., 125,
pp.413-423.
462 REFERENCES
[40] Disch P., Langeoire A., Prat M., Semblat .J.F. (1999). Ouvrages en interaction
(Ernploi des Elements Finis en Genie Civil) 1 Chap. 'l: Nlodrilisation des O'U'urages
en zone sismique - propagation d 1ondcs, pp.467-562, Hcrn1Cs 1 Paris, France.
[41] Blangy J.P., Strandenes S., Moos D., Nnr A. (1993). Ultrasonic velocities in sands-
rcvisitcd, Geophysics, 58(3), pp.344-356.
[42] Bode\ L., van Wijk K., 13itri A., Abraham 0., Cote Ph., Grandjean G., Leparonx
D. (2005). Surface-wave dir.;pertiion inversion lirnitations fron1 laser-Doppler exper-
iments, .Jo·urnal of Environmental and Engineering Geophysi:cs, 10(2), pp.151-162.
[43] 13ohlen T. (2002). Parallel 3D viscoelastic finite difference seismic modelling, Com-
puters and Geosciences, 28, pp.887-899.
[44] Bommer LT., Martinei-Pereira. A. (1999). The effective duration of earthquake
strong rnotion 1 Journal of Earthquake Engineering, 3, pp.127-172.
[45] Bonilla L.F., Archuleta R..J., Lavallee D. (2005). Hysteretic and Dilatant 13clrnvior
of Cohesionless Soils and 'I'heir Effects on Nonlinear Site R,esponse: Field Data
Observations and Modeling, Bv.lleti:n of the Seismological Society of America., 95(6)
pp.2373-2395.
[4()] Bonilla F. (2000). Computation of Linear and Nonl-inear Site Response for Near
Ii'ield Ground l\!Iotion, Ph.D., University of California at Santa Barhara.
[47] Bonnefoy-Claudet S., Cornou C., Orhnberger M., Wathelet M., Bard P.Y., Cotton
F., Fiih D. (2004). H/V ratio and seismic noise wavefield, GeophyBical llesearch
Abstracts, 6.
[48] Bonnet M., Frnngi A. (2007). Analyse des solides deformables pa.r la. methodes de8
elements finis, Ecole Polytechnique EditionH.
[49] 13onnet M. (1999). Boundary integral equation methods for BolidB and fluids, Wiley,
ChicheRter.
[50] Bonnet M., 13ui H.D. (1993). Regularization of the displacement and traction BIE
for 31) elastody11amics uRing indirect method8) in Advances in Boundary Elem,ent
Techniques, Kane JH, Maier G., Tosaka N. and Atluri SN (eds), Springer-Verlag,
Berlin, pp.1-29.
[51] Bonnet M., Maier G., Polizzotto C. (1998). Symmetric Galcrkin boundary element
methods, ASME Applied Mechanics Reviews, 51(11), pp.669-704.
[52] Bonnet G., Heitz .J.F. (1989). Effets non lineaires en dynamique des sols: essais in
situ, Revue Ji'ran9aisc de Geotechniquc, 46, pp.19-30.
[53] Boore D.M. (2005). On pads and filters: processing strong-motion data, B-ullctin
of the Seismological Society of America., 95(2), pp. 745-750.
REFERENCES 463
[54] Doore D.1.lf., Bon1mer J.J. (2005). Processing of strong-rnotion accelerograms:
needs, options and consequences, Soil DynamJcs and Earthquake Engineering, 25,
pp.93-115.
[55] 13oore D.M. (2004). Can site response be predicted?, .Journal of Earthquake Engi-
neering, 8(1), pp.1-41.
[56] Boore D.~VI. (1972). A note on the effect of sirnple topography on sci81nic SH \vaves,
Bullelin of the Scisrnologica.l Sociely of Arneri:ca., 62, pp.275-284.
[57] Bouchon M., Campillo M., Gaffe\ S. (1989). A Boundary Integral Equation-
Discrcte Vla.venumher Representation J'viethod to Study Wave Propagation in J'viul-
tilayered :rvledia Having Irregular Interfaces 1 Geophysics, 54 1 pp.1134-1140.
[58] Bouchon M. (J 985). A simple complete numerical "'Jlution to the problem of diffrac-
tion of SII vv11ve8 by an irregular surface 1 Journal of the Acoustical /)ociety of Anicr-
ica, 77, pp.1-5.
[59] 13ouchon M., Aki K. (1977). Near-field of a seismic source in a layered medium
Vv'ith irregular interfaces 1 (Jeophys. J. fl. Astr. Soc. 1 50, pp.669-684.
[60] Bouchon M. (1973). Effects of topography on surface motion, Bullet·in of the Seis-
mological Society of America, 63, pp.615-622.
[61] 13ouckovabs G.D., Pa.padimitrion A.G. (2005). Numerical evaluation of slope to-
pography effects on seisrnic ground motion) Soil Dyna'mics and Earthquake Engi-
neering, 25(7-10), pp.547-558.
[62] Bourbie T., Coussy 0., Zinszner B. (1987). Acov.sti:cs of porous med-ia, Technip-
lnstitut Frangais clu Pet.role, Paris 1 France.
[63] 13ourdean C., Havenith H.B. (2008). Site effects modelling applied to the slope
affected by the Sunsamyr earthquake (Kyrgy01stan, 1992), Engi:neering Geology,
97, pp.126-145.
[64] Bourdeau C. (2005) . .Effets de site et mouvernents de versant en zones sisrniqucs:
apport de la modClisation num.erique, PhD the8is, Ecole N ationale Superieure des
JVIines de Paris 1 France.
[65] Bourdin 13., Luong M.P., Morlier P. (1989). Cara.ct6risation des non linearites ap-
parues lors deR essais de vibrations forcees en centrifngeuse 1 Revue F'ranr;aise de
Geotechnique, 47, pp.39-59.
[66] Boutin C., Roussillon P. (20111). Assessment of the nrbani7'ation effect on seismic
response, Bulletin of the Seismological Society of America, 94(1), pp.251-268.
[67] BoZhano F., Caserta A., Govoni A., Marra F., Martino S. (2008). Static and dy-
narnic characterization of alluvial deposits in the Tiber R,iver Valley: N e\v data
for assessing potential ground motion in the City of R.ome, Journal of Geophysical
Research, 113, B01303.
 -,
I
464 REFERENCES
[68] Dragov A.M., Lomunov A.K. (1994). Mechanical properties of oome polymers and
composites at strain rates of 1000/s, Jouma.l de physique, 4(8), pp. 337-342.
[69] Bravo M.A., Sanchez-Sesma F.J., Chavez-Garcfa F ..T. (HJ88). Ground motion on
stratified alluv.ial depositR for incident SH vvaves, Bulletin of the Seismological 51 0-
ciety of America, 78, pp.436-450.
[70] Brebbia C.A., Telles J.C.F., Wrobel L.C. (1984). Boundary element techniques,
Springer.
[71] Brennan A.J., l\!Iadabhut'11i S.P.c;. (2002). Effectiveness of vertical drains in tnitiga-
tion of liquefaction, Soil Dynamics and Earthquake Engineering, 22(9-12), pp.1059-
1065.
[72] Brennan B.J. (1980). Pulse propagation in media with frcqueney-clcpenclent Q,
Geophysical Research Letters, 7, pp.211-213.
[73] Brothers P.W., Sinclair G.B., Segedin C.IVl. (1977). Uniform indentation of the
elastic half-space by a rigid rectangular punch, International Journal of Solids and
Slmct·ures, 13, pp.1059-1072.
[74] Bui H.Il., Loret B., Bonnet M. (1985). Regularisation des equations integrnles de
PClastostatiqne et de PClastodynarnique, C1urnptes-Rendus Acadtim,ie des Sciences
Paris (11), 300, pp.633-636.
[75] Bui H.D., Introduction aux problCrnes inverses en rnCcaniq'ue des m,atf.riav,x, Ey-
rolles, Paris, 1993.
[76] Can1pauella R. G., Ste1.vart Vl. P. (1992). Seisrnic cone analy8is using digital signal
processing for dynarnic site characterization, C1anadian (Jeotechnical Journal, 29(3),
pp.477-486.
[77] Campillo M. (2006). Phase and correlation in 'rnnclom' seismic fields and the rc-
constrnction of the Green function, Pure appl. geophys., 163, pp.475-502.
[78] Carcione .J.M., Kosloff D., Kosloff R.. (1988). Wave propagation simulation in a
linear viscoelastic rnediurn) Geophysical Journal International, 95, pp.597-611.
[79] Cascante G., Vandcrkooy J., Chung W. (2005). A new mathematical model for
resonant-column n1easurernents including eddy-current effects Canadian Geotech-
1 effects, Bulletin of lhe Sc·ismologica.l Society of America, 89, pp.718-7:l2.
nical Journal, 42 pp.121-135.
1
[80] Chadwick E., Bettcss P., Laghrouche 0. (1999). Diffraction of short waves modelled
using ne'°'r mapped -,,vave envelope finite and infinite elernents, International Journal
for Numerical A!ethods in E'ng., 45, pp.335-354.
[81] Chaillat S., Bonnet M., Semblat .T.F. (2009). A new fast multi-domain BEM to
n1odel seisn1ic '°'rave propagation and amplification in 3D geological structures, Geo-
physical Journal International (to appear).
REFERENCES 465
[82] Chaillat S., Bonne\ M., Semblat J.F. (2008). A multi-level fast multipolc BEM
for 3-D clasto<lynarnicD in the frequency domain, Com,pnter IVIethods in Applied
Mechanics and Engineering, 197, pp.42:J:J-4249.
[8:J] Chailla.t S. (2008). Fast m.ultipole method for 3-D elastodynamic bu·undary integral
eq'Uations. Application to seisrnic 1JJa1Je propagation, PhD thesis, Ecole Nationalc
des Ponts et ChaussCcs.
[84] Chaljnb E. (2006). Spectral element modeling of 3D wave propagation in the alpine
valley of Grenoble, France, i'-Jrd International ,S'ym,posiurn on the Effects of Surface
Geology on Seisrnic Motion, Grenoble, France.
[85] Chaljub E., Capdeville Y., Vilottc .J.P. (2003). Solving Elastodynamics in a Fluid-
Solid IIeterogeneons Sphere: a Parallel Spectral Element Approxirnation on Non-
Conforming Grids, Jal of Cornp·uta.tional Physics, 187(2), pp.457-491.
[86] Charnmas R, Abraham 0., Cote P., Pedersen II.A., Semblat .J.F. (2003). Charac-
terization of heterogeneous soils using surface '°'raves: ho1nogenization and nun1eri-
cal modeling, International Journal of Geomechan'ics {ASCE), 3(1), pp ..15-63.
[87] Chantawarangul K. (1993). Numerical sim.nlations of three-dimensional granular
assembLies 1 _PhD Thesis, University of \~Taterloo 1 Canada.
[88] Chavez-Garcia F., Raptakis D., Makra K., Pitila.kis K. (2002). The importance of
the reference station in n1odelling site effects up to larger frequencies. The case
of Euroseistest, 12ih European Conference on Earthquake Engineer1:ng, London,
1Jnited l{ingdoni.
[89] Chavez-Garcia F.J., Cardenas-Soto M. (2002). The contribution of the built en-
vironment to the 11 free-field' 1 ground rnotion in :tviexico city) Sor:l Dynamics and
Earthquake Eng., 22, pp. 773-780.
[90] Chavez-Garcia F.J., Raptakis D.G., Makra K., Pitilakis K.D. (2000). Site effects
at Euroscistest-II. Ret>ults from 2D nurnerical 1nodelling and co1nparison with ob-
servations, Soil Dynamics and Earthquake Eng., 19(1), pp.23-39.
[91] Chavez-Garcia F.J., Stephenson W.R., Rodriguez M. (1999). Lateral propagation
effects observed at Parkv;ray, New Zealand. A case hiRtory to cornpa.re lD vs 2D
[92] Chazelas J.L., Abraham 0., Semblat J.F. (2001). Identification of Different Seismic
\~raves Generated by Foundation Vibration in the Centrifuge: Travel Time, Spectral
and Numerical InvestigationH, F'ou.rth International Conference on Recent Advances
in Geotechnical Earthquake Engineering and Soil Dynamics, San Diego, California,
USA.
[93] Chehab A.G., el Nagga.r M.H. (2003). Design of efficient base isolation for hammers
and presses, S'oil Dynarnics and Earthquake Engineering1 23, pp.127-141.
466 REFERENCES
[94] Chen J.T., Hong H.K. (1999). Review of dual boundary element methods with em-
phasis on hypcrsingular integrals and divergent serieR, Applied J"\!fechanics Re1Jiev1s 1
52(1), pp.17-33.
[95] Cheney J.A., Brown R.K., Dhat N.R., Hm 0.Y.Z. (1990). Modeling free-field con-
ditions in centrifuge models, Journal of Geotechnica.l Engineering, 116(9), pp.13,17-
1367.
[96] Chou P.C., Hashemi J., Chou A., Rogers H.C. (1991). Experimentation and fi-
nite element simulation of adiabatic Rhear bands in controlled penetration in1paet 1
International Journal of Impact Engineering, 11(3), pp.305-321.
[97] Chow Y.K., Yong D.M., Yong K.Y., Lee S.L. (1990). Monitoring of dynamic com-
paction by deceleration ineasurementt~, Cornputers and Geotechnics, 10) pp.189-209.
[98] Chow Y.K., Yong D.M., Yong K.Y., Lee S.L. (1992). Dynamic compaction of loose
sand deposits, Soil.9 a.nd Foundations, 32(4), pp.93-106.
[99] Chow Y.K., Yong D.M., Yong K.Y., Lee S.L. (1992). Dynamic compaction analysis,
.!u"Urnal of Geotechnical Engineering (A.S.C.E), 118(8), pp.1141-1156.
[100] Chung R.M., Yokd F.Y., Drncvich V.P. (1984). Evaluation of dynamic properties
of sands by resonant column testing, Gcotcchnical Testing Journal, 7(2), pp.60-69.
[101] Clough R.W., Pen"ien J. (1993). Dynamics of st·mct·arcs, Mc-Graw Hill, New-York,
USA.
[102] Clouteau D., Al-Hussaini T., Degrande G., Hunt H. (2001). Stochastic modelling
of traffic induced vibrations, 1st Albert Caquot International Conference, Paris 1
France.
[103] Clouteau D., Aubry D. (2001). Modifications of the ground motion in dense urban
areas, Journal of Cornputational Acoustics) 9 1 pp.1659-1675.
[104] Cloutcau D. (19911). Propa.ga.tion d'ondcs dans des milieux heterogenes. Application
ala tenne des 01t1Jrages sous si1:.sm,es 1 PhD thesis 1 Ecole Centrale de Parisi France.
[105] Coe C.J., Prevost J.H., Scanlan 11.H. (1985). Dynamic stress wave reflec-
tions/attenuation: earthquake simulation in centrifuge soil rnodcls, Earthquake
Engineering and 8tructnrnl Dynarnics, 13, pp.109-128.
[106] Cohen G., .Joly P., Roberts J.E., Tordjmarr N. (2001). Higher order triangular finite
clernents with mass lu1nping for the wave equation, SIAM Journal on Num,erical
Analysis, 38(6), pp.2047-2078.
[107] Cohen G., .Joly P., Tordjman N. (1994). Higher-order finite elements with mass-
lurnping for the ID \Vave equation, F1:nite Elenicnts 'in A.nalysis and Design, 16(3-4),
pp.329-336.
REFERENCES 467
[108] Collino F., Joly P., Millot F. (1997). Fictitious domain method for urrnteady prob-
lc1ns: application to clectrornagnetic scattering, Journal of Coniputational Physics,
138(2), pp.907-938.
[109] C~ornon (].,Bard P.Y., Dietrich lVI. (2003). Contribution of dense array analysis to
the identification and qnantification of basin-edge-induced \Vavcs; B'ulletin of the
Seismological Society of America, 93(6), pp.2604-2623.
[110] Critescu N. (1974). Dynamic plasticity, North-Holland.
[111] Curnier A. (1993). Methodcs ffumcriques en mecanique des solides, Presses Poly-
techniques et UrllversitaireH Ron1andes 1 Lausanne) Switzerland.
[112] Dangla P., Semblat .J.F., Xiao H.H., Delepine N. (2005). A Simple and Efficient
Regularization Method for 3D BEM: Application to Frequency-Domain Elastody-
namics, Bulletin of the Seismological Society of America, 95(5), pp. 1916-1927.
[113] Da.ngla P. (1988). A plane strain soil-otructure interaction model, Earthq"Uake En-
gineering and Structural Dynarnics, 16, pp.1115-1128.
[114] Dangla P (1989). Couplagc Clements finis equations integralcs en Clastodynamique
ct 'interaction sol-strnctnrf\ PhD Thesis, Ecole Nationale des Fonts et Chaussees,
Paris, France.
[115] Dangla P., \lanhoove J.P. (1988). Transrnission des vibrations par les tunnels fer-
roviaireR: validation experimentale cl'un rnod8le de calcul, International Conference
on Afeasuremcnts and Testing in C'iv'il Engineering) L·yon, France.
[116] Darve E. (2000). The Fast Multipole Method: Numerical Implementation, .Jo'U'(nal
of Cornvutational Physics, 160(1), pp.195-240.
[117] Das B.M. (1983). Fundarnenta.ls of soil dynamics, Elsevier, New-York, USA.
[118] Daubechics I.D. (1992). Ten lectures on Wavelets, SIAM.
[119] Dauksher W., Emery A.F. (1999). An evaluation of the cost effectiveness of Cheby-
shev spectral and p-finite element solutions to the scalar 1.vave equation, Interna-
tional Journal for Numerical Methods in Engineering, 45, pp.1099-1113.
[120] Dauksher W., Emery A.F. (1999). An evaluation of the cost effectiveness of Chcby-
shev spectral and p-finitc elcrnent solutions to the scalar 1.vave equation) Interna-
tional Journal for Numerical Methods in Engineering, 45, pp.1099-1113.
[121] Davies E.D.H., Hunter S.C. (196:J). The dynamic compression testing of solids by
the rncthod of the split Hopkinson pressure bar, Journal of .Nlcchanics and Physics
of Solid, 11, pp.155-179.
[122] Day S.J'vl. (1998). Efficient simulation of constant-Q using coarse-grained me1nory
variables, Bulletin of the Seismological Society a.f America, 88(4), pp.1051-1062,
1998.
468 REFERENCES
[123] D'1y S.M., Nlinstcr J.B. (198,1). Numerical simulation of attenuated wavefields using
a Pa.de a.pproximant method, Geophysical Journal of the Royal Astron. Society, 78,
.pp.105-118.
[124] Delavaud E. (2007). Numerical modelling of wave propagation in com.ple.T geological
rn.ed£a, PhD Thesis (in French), University Paris VII/Institute of the Physics of the
Earth 1 Paris.
[125] Delepinc N., Bonnet G., Semblat J.F., Lenti L. (2007). A simplified non linear
rnodcl to analyze site effects in alluvial dcposits 1 4th International Conference on
Earthquake Geotechnical Eng.) Thet->t->a.loniki 1 Greece.
[126] Delepine N., Sernblat J.F. (2006). Site effects in a deep alpine valley for various
seism.ic sources) Srd International Symposium on the E'ffects of Surface Geology on
Seisrrl'ic Motion, Grenoble) France.
[127] Dcraemaeker A., Babuska I., Bouillard P. (1999). Dispersion ancl pollution of the
FEJ\.1 solution for the Helrr1holtz equation in one, t~vo and three din1ensions, Inter-
national .Journal for Numerical Methods in Engineering, 46, pp.471-499.
[128] Dobrin M.D., Lawrence P.L., Sengbusb R.L. (1954). Surface and near-surface waves
in the Delaware basin, Geophysics, 19(4), pp.695-715.
[129] Dobry R., Oweis I., Urzua A. (1976). Simplified procedures for estimating the fun-
damental period of a soil profile, Bulletin of the Sefom.ological Society of America,
66, pp.1293-l'.l21.
[130] Dormieux L. (1989). Determination de caracteris\iques dynarniqucs chm sol a!'aide
d)un essai pression16trique cyclique) Revue Franr,;aise de GCotechnique) 46, pp.31-41.
[131] Dowding C.H. (1985). Blast vibration monitoring and control, Prent.ice-Hall, 297
p.
[132] Dravinski M. (2003). Scattering of elastic waves by a general anisotropic basin. Part
2: a 3D model, Earthquake Engineering and Structural Dynamics 32, pp.653-670.
[133] Dravinski 'JI., Mossessian T.K. (1987). Scattering of Plane Harmonic P, 8V and
R.ayleigh Waves by Dipping Layers of Arbitrary Shape, B11lletin of the Seismological
Society of Anierica1 77) pp.212-235.
[134] Dria.d-Lcbeau L., Lokmane N., Semblat J .F., Bonnet G. (2009). Local amplification
of deep ruining induced vibrations, Part 1: Experhnental evidence for site effects
in a coal ha.sin, Soil Dynarnics and Earthq'uake Engineering, 29(1)) pp.39-fiO.
[135] Drnevich V.P. (1977). R.esonnant. column testing. Problems and solutions, Sympo-
sium, on Dynarnic Soil and Rock Testing 1:n the Field and Laboratory for Seismic
Studies, ASTM, Denver, USA.
[136] Drnevich V.P., Richart F.E. (1970). Dynamic prcstra.ining of dry sand, .lo-urnal of
the Soil Mechanics and Foundations Di·uision (A.S.C.E), 96, pp.453-469.
REFERENCES 469
[137] Duval A.M., Mencroud J.P., Vidal S., Bard P.Y. (1998). R.elation between curves
obtained frorn rnicrotremor and site cffcctt> observed after Caracas 1967 earthquake,
1.llh European Conference on Earthquake Engineering 1 Paris, France .
a
[138] Duval A.M. (1996). Determination de la reponse d'un site a-ux seism.es /'aide du
bruit de fond. Evaluation expfrim,entale, PhD theRis University Paris VI, Etu<lcs ct
Recherches des LPC, GT62, LCPC, Paris, France.
[139] Edlinger M.L., P.Grnt.acos, P.Montmitonnet., E.Feldcr (199:J). Finite clement. anal-
ysis of elastoplastic indentation vvith a deforn1able intleuter) Ev,ropean Journal of
Mecha.nirs (A/Solids), 12(5), pp.679-698.
[140] Edouard S., Lcgrns TI., Lefevre F., Eymarcl R. (1996). The effect of small-scale
inhomogeneities on ozone depletion in the Arctic) Nature 1 384, pp.441-447.
[141] Ekcvid T., Wiberg N.E. (2002). Wave propagation related to high-speed train:
J\_ scaled boundary FE-approach for unbounded don1ains 1 Com,puter Methods in
Applied Mechanics and Engineering, 191(36), pp.8947-3964.
[142] }~1111nerich H. 1 Korn1\1. (1987). Incorporation of attenuation into tilne-don1ain co1n-
putations of scisrnicwave fields, Geophys·ics, 52(9), pp.1252-1264.
[143] Eringen A.C., Suhubi E.S. (1975). Elastodynamics (vol. II-linear theory), Academic
Press.
[144] Eshraghi H., Dravinski M. (1989). Scattering of plane harmonic SH, SV, P and
11ayleigh waves by non-a.xisym1netric three-dimensional canyons: a wave function
expansion approach 1 Earthquake Engineering and Structural D:ynarnics 1 18, pp.983-
998.
[1'15] Faccioli E., Maggio F., Paolucci R., Quartcroni A. (1997). 2D and 3D elastic wave
propagation by a pseudo-spectral domain decon1po8ition method 1 Journal of Seis-
mology, J, pp.237-251.
[146] Faccioli E., Maggio F., Q1mrteroni A., Taghan A. (1996). Spectral-Domain De-
co1nposition J\/Iethods fo.r the Solution of Acoustic and Elastic V\Tave Equations 1
Geophysi:cs, 61, pp.1160-1174.
[147] Fiih D., Suhadolc P., Mueller St., Panza G.F. (1994). A hybrid method for the esti-
mation of ground 1notion in sedimentary basin8: quantitative 1nodcling for J\1exico
city, Bulletin of the Seismological Society of America, 84(2), pp.'.l83-'.l99.
[148] Farnbach .T.S. (1975). The complex envelope in seismic signal analysis, Bulletin of
the Seismological Society of America, 65(4), pp.951-962.
[149] Felice C.W. (1985). The response of soil to impulse loads using the S.H.P.B tech-
niqv.e, PhD Thesis, University of Utah, Salt Lake City, USA.
[150] Felice C.W., Gaffney E.S., Brown J.A. (1987). Dynamic high stress experiments on
soil, Geotechnical Testing .Journal, 10(4), pp.192-202.
470 REFERENCES
[151] Felice CW., Gaffney E.S., Brown J.A. (1991). Extended split Hopkirrnon bar anal-
ysis for attenuating 1natcrials, Journal of Engineering _Afechanic.c; ( ASCE), 117(5),
pp.1119-1135.
[152] Festa Ci. 1 Vilotte J.P. 1 Delavaud E. (2005). Interaction between surface \vaves and
absorbing boundaries for Vila:ve propagation in geological basins 21) nun1erical
simulations, Geophys£ca.l research letters, 32(20), pp.L20306.l-L20306.4.
[153] Festa G., Nielsen S. (2003). PML absorbing boundaries, Bulletin of the Seismolog-
ica.l Society of America, 93(2), pp.891-903.
[154] Fiala P. 1 Degrande G., AuguRztinovicz F'. (2007). Numerical rnodclling of ground-
borne noise and vibration in buildings due to surface rail traffic, Journal of Sov,nd
a.nd Vibration, 301(3-5), pp.718-738.
[155] Finn L. (2007). Earthquake Engineering in Canada: A Selective ()vervie\v Dth
Canadian c~onference on Earthq'Uake Engineering, Ottawa, (jc_tnada.
[156] Flanagan l).P. 1 Belytschko '.f. (1981 ). A uniform strain hexahcdron antl quadrilat-
eral '°''ith orthogonal hourglass control, International Jo'Urnal for Num,erical Meth-
ods in Engineering, 17, pp.679-706.
[157] Foti S. (2003). Small-strain stiffneHs and darnping ratio of PiHa clay fron1 surface
vvave tests 1 Gliotechnique, 53(5), pp.455-461.
[158] Foti S. (21100). Multistation Methods for Geotcchnical Charo.cteriza.tion using Sur-
face T,Vaves 1 PhI) thesis, Politecnico di Torino, Italy.
[159] Franc;;ois S. 1 Pyl J,,., Jlvfasourni I-I.R,., Degrande c;. (2007). The infiuence of dyna1nic
soil-~tructure interaction on traffic induced vibrations in buildingH, Soil Dynarnics
and Earthquake Engineering, 27(7), pp.655-674.
[160] Frankel A., Vidalc J. (1992). A Tlrree-Dirnensional Simulation of Seismic Waves
in the Santa Clara Valley, California, from a Loma Prieta Aftershock, Bulletin of
the Seismological Society of America., 82, pp.2045-2074.
[161] Fujiwara H. (1998). The fast multi pole method for the integral equations of seismic
scattering problernH, Geophysical Jo11,rnal International, 133 1 pp. 773-782.
[162] Fujiwara H. (2000). The fast n1ultipole inethod for Rolving integral equations of
three-dimensional topography and basin problems, Geophysical Journal Interna-
tional, 140, pp.198-210.
[163] Fung Y.C., Tong P. (2001). Cla.ss£co.l and Compv.tatfona.l Soli:d Mechanics, World
Scientific Publishing.
[164] Fung T.C. (1997). Unconditionally stable higher-order Newmark methods by sub-
Htepping procedure, Computer Methods in Applied M'echanics and Engineering
147, pp.61-84.
REFERENCES 471
[165] Futterman W.I. (1962). Dispersive body waves, Journal of Geophysical Research,
67(13), pp.5279-5291.
[166] Gaffet S., Bouchon M. (1989). Effects of two-dimensional topographies using the
discrete 1,vavenurr1ber-houndary integral equation inethod in P-SV cases, Journal
of the Acoustical Society of America, 85(6), pp.2277-2283.
[167] Gmnbin M. (1985). Dix am de consolidation clynamique, Ann.ales de l'Insti:tut Tech-
nique du Batiment et des Travaux Publics (I. T.B. T.P), 433.
[168] Gao G.Y., Li Z.Y., Qiu Ch., Yue Z.Q. (2006). Three-dimensional analysis of rows
of piles as paMRive barriert-i for ground vibration isolation, Soil Dynamics and Earth-
quake Engineering, 2(), pp.1015-1027.
[109] Gaul L., Scharrn M. (1999). A comparative study of three Loundary clement ap-
1 proaches to calculate the transient response of viRcoelastic Holids vvith unbounded
do1nainH 1 Co,mp'uter Methods in Applied Nlechanics and Engineering, 179(1-2) 1
pp.111-123.
[J 70] Geli L., Bani P.Y., Jullien B. (1988). The effect of topography on earthquake
ground rnotion: a revievv antl ne\O.r results, Bulletin of the Seism,ological Society of
America, 7, pp.42-63.
[171] Gelis C., Bonilla L.F., Regnier ,T., Bertrand E., Duval A.M. (2008). On the use
of a Saenger's finite difference stencil to model 2D P-SV nonlinear baHin response:
application to Nice, France 1 Seismic Risk 2008: Earthq'uake in North-West Europe,
Liege, Belgiunl.
[172] Gelis C., Leparoux D., Virieux J., 13itri A., Operto S., Grandjean G. (2005). Detec-
tion of underground cavitieH: dispersion diagram analysis for synthetic finite differ-
ence data, Journal of En1Jironrnental and Engineering Geophysics, 10(2), pp.111-
122.
[173] Gerdes K. (1998). A summary of infinite element formulations for exterior
Heln1holtz problcrns. Computer Methods in Applied JVIechanics and Engineering,
164(1), pp.95-105.
[174] Gilbert F., Backus G. (1966). Propagator matrices in elastic wave and vibration
proLlems, Geophysi:cs, 31, pp.326-332.
[175] Gil-Zepeda S.A., Montalvo-Arrieta .J.C., Vai R., Sanchez-Sesma F ..J. (2003). A 11
hybrid indirect boundary element-<liHcrete wave nu1nber n1ethod applied to sirnulate
the seis1nic response of stratified alluvial valleys, Soil Dynamics and Earthq'Uake
Engineering, 23(1), pp.77-86. II
[17()] Givoli D. (1992). Numerical Methods for Problems in £nfinite domains, Elsevier. [,
1
[177] Gladwin M.T., Stacey F.D. (1974). Anelastic degradation of acoustic pulses in rock,
Physics of the Earth and Planetary Interiors, 8, pp.332-336. !Ii
I
II
Iii
lq
472 REFERENCES
[178] Graves R,.\,T. (1996). Sirnulating seismic wave propagation in 3D ela8tic rnedia using
staggered-grid finite differences, Bulletin of the Seism,ological Society of America,
8(), pp.1091-1106.
[179] Greengard L., Huang .J., R.okhlin V., Wandzura S. (1998). Accelerating fast mul-
tipole methods for the IIchnholtz equation at lovv frequencies. IEEE Comp. Sci.
Eng., 5(3), pp.:12-38.
[180] Groby J.P., Tsogka C., Wirgin A. (2005). Simulation of seismic response in a city-
likc cnvironnrent, Soil Dynamics and Earthquake Engineering, 25(7-10) 1 pp.487-
504.
[181] Gucguen P., Cornou C., Garambois S., Banton J. (211117). On the limitation of the
H/V spectral ratio using seis1nic noise as an exploration tool: Application to the
Grenoble valley (Fr-ance), a small apex ratio basin, Pure and Applied Geophysics,
164(1), pp.115-l:J4.
[182] Gueguen P., Bard P.Y., Oliveira C.S. (2000). Experimental and Numerical Analysis
of Soil Motions Caused by Free Vibrations of a Building Model , Bulletin of the
Seismological Society of America, 90(6), pp.1464-1479.
[183] Guelin P., Nowacki W.K. (1981). R.cmarques sur Jes ondes d'acceleration dans un
continu 6la.stopla.stique avec hyHtt3r8si.s 1 Arr;hh} Mer;h, 36(1), pp.49-66.
[184] Guiggiani M. (1994). Hypersingular formulation for boundary stress evaluation.
Eng. Analysis Boundary Elements, 13, pp.169-179.
[185] Guiggiani M., Krislmasamy G., Rndolphi T.J., Rizzo F.J. (1992). A general al-
goritlnn for the numerical solution of hypersingular boundary integral equatioru.;,
ASME .foumal of Applied Mechanics, 59, pp.6114-614.
[186] Habib P. (19()7). Sur un mode de glissement des massifs rocheux, Comptes Rcndus
de l'Acadern:ie des Sciences, t.264, Serie A, pp.151-153.
[187] Hadjian A.H. (20112). Fundamental period and mode shape of layered soil profiles,
So-ii Dynamics and Earthqv.ake Engineering, 22(9-12), pp.885-891.
[188] Haskell N.A. (1953). The dispernion of surface waves in multi-layered media, Bul-
letin of the Seismological Society of America, 43(1).
[189] Hansson P.A., Sandberg G. (1997). Mass matrices by minimintion of modal errors,
International Jo1Lrnal for Nv,merical l\1ethods in Engineering, 40, pp.4259-4271.
[190] Hardin B.O. (1965). The nature of damping in sands, Jal of the Soil Mechanics
and Fo·unda.tion Division {ASCE), 91(SM1), pp.63-97.
[191] Hardin B.O., Drnevich V.P. (1972a). Shear modulus and damping in soils : mea-
surcrnent and para1neter effect.s, Journal of the Soil Mechanics and l 1"oundalions
Di-ui8-ion (A.8.C.E), 98, pp.603-624.
REFERENCES 473
[192] Hardin 13.0., Drnevich V.P. (1972b). Shear modulus and damping in soils: design
equations and curves, Journal of lhe ,5oil Mechanics and Foundations Division
(A.S.C.E), 98, pp.667-691.
[193] Hardy C., Baronet C.N., Tordion G.V. (1971). The elasto-plastic indentation of
a half-space by a rigid sphere, International J onrnal for Nv,m,erical Jifethods in
E'ngineering, 3, pp.451-462.
[194] Hashash Y.M.A., Park D. (2001). Non-linear one-dimensional seismic ground mo-
tion propagation in the J'vli8sissippi en1bayrnent, Engineering Geology) 62(1-3),
pp.185-206.
[195] Haskel N.A. (1953). The dispersion of surface waves on multilayered media, Bulletin
of the Seismological Sodcty of America., 43, pp.17-34.
[196] Hata S., K.Tatcyama (1991). Theoretical approach to impact soil compaction
through plastic wave propagation, Jo11rnal of TerraTnechanics 1 28(4), pp.349-358.
[197] Haupt W.A. (1977). Surface-waves in non-homogeneous half-space, Proceedings of
DMSR 77, 1, Karlruhe, Germany.
[198] Heicrli W. (1962). Inelastic wave propagation in soil columns, Jo·urnal of the Soils
Mechanics a.nd Foundations Di1Jision {A.S.C.E), 88(6), pp.33-63.
[199] Heih .J.F. (1992). Propagation d'ondes en m·ilieu non lineaire. Appli:cations au
ginie parasisrnique et a la rer;onnaissance des .c;ols, PhD The8i.s, University Joseph
Fourier, Grenoble, Fra.nce.
[200] Henze F., Archuleta R., l3onilla F., Day S., Doroudian M., Elgamal A., Gonzales
S., Roehler M., Lai T., Lavallee D., Lawrence 13., Liu P.C. (2004). Estimating
site-specific strong earthquake rnotion, Soil DynamJcs and Earthq'uake Eng., 24,
pp.199-223.
[201] Hisada Y. (1995). An efficient method for computing Green's functions for a lay-
ered half.space with sources and receivers at close depths (Part 2)) Bulletin of the
Seismological Society of America, 85(4), 10811-1093.
[202] Huang A.B., Ma M.Y., Lee J.S. (1992). A micromechanical study of penetration
teHts in granular rnaterial, Advances in Micromechanics of Granular Materials
(H.H.Shen), Elsevier.
[20:1] Hu N., Guo Z., Katz I.N. (1998). Bounds for eigenvalues and condition numbers in
the p-version of the finite element method, Mathematics of Computation, 224(67),
pp.1423-1450.
[204] Hughes T ..J.R.., Reali A., Sangalli G. (2008). Duality and Unified Analysis of Dis-
crete Approximations in Structural Dyna1nics and \iVave Propagation: Comparison
of p-method Finite Elements with k-mcthod NUB.BS, Cornp·uter Methods in Applied
Mechanics and Engineering, 197(49-50), pp.4104-4124.
174 REFERENCES
[205] Hughes T.J.R. (1987). Linear stat·ic and dynamic finite element analys·is, Prentice-
Hall, E·nglcwood Cliffs, NeVlr Jersey.
[206] Humbert P., Fc"arrn G., Dubouchet A., R.crnaud D. (2005). CESAR-LCPC : A
cornputation soft\vare package dedicated to civil engineering uses, Bulletin des
Laboratoires des Fonts et Cha'ussf-es, 256-257, pp.7-37.
[207] Hunaidi 0. (1998). Evolution-based genetic algorithms for analysis of non-
destructive surface wave ter;ts on pave1nents 1 JVDT f3 E International1 31(4))
pp.273-280.
[208] Iai S., Morita T., Kanieoka T., Matsunaga Y., Abiko K. (1995). Response of a
dense sand deposit during 1993 Kushiro-Oki earthquake, !3oils and ! 1'oundations,
35(1), pp.115-131.
[209] IdriSR I.M., Seed H.B., Scrff N. (1974). Seismic response by variable clamping finite
elements, Jo'Umal of the Geotechnical Engineering Division (ASCE}, pp.1-13.
[21 OJ Idriss I.M., Seed H.B. (1967). Response of earth banks during earthquakes, Jo'UT-nal
of the Soil Mechanics and Foundations Di·uision, ASCE, 93(SM3), pp.61-82.
[211] Ihlenburg F., Babuska I. (1995). Dispersion analysis and error estimation of
Galerkin finite ele1nent rr1ethods for the H.ehnholt,., equation, International Journal
for Numerical Methods in Engineering, 38, pp.3745-3774.
[212] Ishihara K., Towhata I. (1983). Sand response to cyclic rotation of principal stress
directions as induced by wave loads, Soils a.nd Foundations, 23(4), pp.11-2().
[213] Jaffard S., Meyer Y., Ryan R.D. (2001). Wavelets, tools for Science and Technology,
SIAM, 256 p.
[214] Jennings P.C. (1970). Distant motions from a building vibration test, Ballcti:n of
the Seismological Society of America, ()(), pp.2037-2043.
[215] Joly P. (1982). Les ondes de Rayleigh num.eriques, Research Report No.107, INRIA,
Rocquencourt, Ftance.
[216] .Tongmans D., Pitilakis K., Dcrnanct D., R.aptakis D., R.iepl .J., Horrent C. (1998).
E·uroSeistest: deter1nination of the geological structure of the Volvi basin and val-
idation of the basin response, Bnlletin of thr: SeL'3m,ological Societu uf Arnerica) 88,
pp.473-87.
[217] Jongmans D. (1991). Near-source pulse propagation: application to Q-determi-
nation, Geophysical prospecting, 39, pp.943-952.
[218] Ju S.H., Lin H.T. (2004). Analysis of train-induced vibrations and vibration reduc-
tion sche1nes abovo and belo\v critical Rayleigh speeds by finite ele1nent method,
Soil Dynamics and Earthquake Engineering, 24, pp.993-1002.
REFERENCES 475
[219] Jumin P. (1992). Contribution de la penetration dynamiq·ue des picm; 0. la recon-
naissance des sols) application au battage des ancres et au pifzoc8ne, PhD Thesis,
University Bordeaux I.
[220] KarlstrOm A., BostrOrn A. (2007). Efficiency of trenches along raihvays for trains
moving at sub- or supersonic spcc<l.s, Soil Dynam,ics and Earthquake Engineering,
27, pp.625-641.
[221] Kattis S.E., Polyzos D., Beskos D.E. (1999). Modelling of pile wave barriers by
effective trenches and their screening effectiveness, Soil Dynarnics and Earthquake
Eng·ineering, 18(1), pp.l-10.
[222] Kauscl E., ASRimaki D. (2002). Seismic simulation of inelastic soils via frequency-
dependent n1oduli and da1nping) Journal of Engineering Mechanics {ASCE) 1
128(1), 34-47.
[223] Kausel E. (2006). Fundamental Sol'Utions in Elastodynamics, A Cornpendiurn, Cam-
bridge University PreHs 1 260 p.
[22"1] Kausel E·. (2000). The thin-layer rnethod in scisrnology and earthquake engineering,
in Wave Jvfob:on in Earthquake Engineering) E. Kauscl and G. J'vianolis (Editors),
WIT Press, Southampton, UK, pp.193-213.
[225] Kauscl E., R.oesset J.M. (1981). Stiffness matrices for layered soils, Bulletin of the
Seismological Society of America, 71(fl), pp.1743-17()1.
[226] Kawase H. (2003). Site effects on strong ground motions, in Iniernat-ional Handbook
of Earthquake and Engineering Seisrnology, Part B) \V.HJ(. Lee and II. Kana1nori
(eds.), Academic Press, London, pp.1013-1030.
[227] Kawase H., Matsushima S., Graves R.W., Somerville P.G. (2000). Strong motion
simulation of Hyogo-ken Nanbu (Kobe) earthquake considering both the hetero-
geneous rupture process and the 3-D basin structure, 12th VVorld Conference on
Earthquake Engineering, Auckland) N e\v Zealand.
[228] Kawase H., Sato T. (1992). Simulation analysis of strong motions in the Ashigara
valley considering one- and two-dimensional geological structures, Journal of the
Physics of the Earth, 411, pp.27-56.
[229] Kawase H., Aki K. (1989). A study on the response of a soft basin for incident.
S, P and Rayleigh waves \Vith special reference to the long duration observed in
Mexico City, Bulletin of the Seismological Society of America, 79, pp.1361-1382.
[230] l{a\vase H·. (1988). Time-don1ain re8ponHe of a sernicircular canyon for incident SV,
P, and Rayleigh \VaveH calculated by the discrete vvavenun1ber boundary element
method, Bulletin of the Seisrnologi:cal Society of America, 78, pp.1415-1137.
[231] Kennett 13.L.N. (1974). Ileflections, rays, and reverberations, Bulletin of the Seis-
mological Society of America, 64, pp.1685-1696.
476 REFERENCES
[232] Kham M., Semblat J.F., Bard P.Y., Dangla P. (2006). Seismic Site-City Interaction:
l\tfain Governing Pheno1ncna through Sin1plificd Numerical lVIodels) Bulletin of the
Seismological Soc£ety of America., 96(5), pp.1934-1951.
[2:l:J] Kham M. (2004). Propagation d'ondes sismiques dans les bassins scdim.entaires:
des effets de site a l 'interaction site-ville, PhD thet>is 1 E·cole N ationale des Fonts et
Chaussecs: France.
[234] Kham M., Sernblat .T.F., Xiao H.B., 13ouden-Romdhane N., Dangla P. (2002a).
Nu1nerical rnodelling of site arnplification vs experin1cntal data frotn Tunisia, 12th
European Conference on .b'arthquake Engineering, Lon<lon, UK.
[235] Kham NJ., Scrnblat .T.F., Kurose A., Xiao H.B., Dangla P. (2002h). Wave/cavity
interaction: analytical and I3EJ\i[ approaches, t2th European Conference on Earth-
quake Engineering, London, UK.
[236] Kim S.M., Liao S.T., Roesset J.M. (1999). Simulation of the cross-hole method in
iRotropic and anisotropic n1c<lia, International Journal for Numerical and Analytical
Methods in Gcom.echanics, 23, pp.J 101-1119.
[237] Klepaczko J.R, Gary G., 13arberis P. (1991). Behaviour of rock salt in uniaxial
cornpreRsion at n1ediu1n and high strain ra.tes 1 .Arch. Mechanics, t13(4), pp.499-517.
[238] Kjartarnson E. (1979). Constant Q wave propagation and '1ttenuation, Journal of
Geophysical Research, 84, pp.4737-4748.
[2:19] Koerner RM., Lord A.E . .Jr, McC'1he W.M., Curran ,J.W. (1976). Acoustic emis-
sion behaviour of granular soils, Journal of the Geotechnical Engineering Division
(A.S. C.E), 102(7), pp. 761- 773.
[240] Kolsky H. (1949). An investigation of the mechanical properties of material at a
very high rate of loading, Proceedings of the Physical Society B, 611, pp.676- 701.
[241] Kokusho T. (1980). Cyclic triaxial test of dynamic soil properties for wide strain
range, So£ls and Foundations, 20, pp.45-60.
[2,12] Komatitsch D., Martin IL (2007). An unsplit convolutional Perfectly Matched
Layer irnprovcd at grazing incidence for the seis1nic wave equation, Geophysics,
72(5), pp.155-167.
[24:1] Komatitsch D., Liu Q., 11-omp .T., Siiss P., Stidham C., Shaw J.H. (2004). Simula-
tions of Ground J\1otion in the Los Angeles I3asin ba.sed upon the Spectral-f]len1cnt
Method, Bulletin of the 8eism.ological Society of Am.eri:ca, 94(1), pp.187-206.
[241] Komatitsch D., Vilotte J.P., Vai R., Castillo-Covarrubias J.M., Sanchez-Sesma
F.J. (1999). The .spectral ele1nent 1nethod for Gla.stic v.rave equations - Application
to 2D and 3D seisrnic problen1s, IntC'rnational Journal for Nurnerical 1\1ethods in
Engineering, t15 1 pp.11:-39-1164.
REFERENCES 477
[245] Kornatitsch, D., Vilotte J.P. (1998). The Spectral Element Method: An Efficient
Tc)ol to Sin1ulatc the Seis1nic R.etiponsc of 2D and 3D Geological StructureR) Bulletin
of the Seismological Society of America, 88(2), pp.368-392.
[246] Komatitsch D., Vilotte J.P. (1996). Simulation of Elastic Wave Propagation in
Ccnnplex 1vicdia lJsing an Explicit Parallel Spectral Element lvfethod 1 Arnerican
Geophysical Union {AGU) Fall Meeting, Abstract #S41D-15, San Francisco, USA.
[247] Kouoh-Bille L., Sanchei-Sesma F ..J., Wirgin A. (1991). Reponse resonante d'une
n1ontagne cylindriquc a une ondc sis1niquc SI-I, Cornptes-Rendus Acadernie des Sci-
ences Paris, :Jl2(1I), pp.849-854.
[248] Kramer S.L. (1996). Oeotechnical earthquake engineering, Prentice-Hall, Engle-
wood Cliff, N ..J.
[2119] Kunt J\11. (1986). Traitem,ent n'urnerique des signaux, Traite d'ClectricitC, Pret·mes
Polyteclmiques Romandes, pp.293-324.
[250] Kurose A. (2000). Effets des seismes sur !es ouvrnges soutcrrnins, PhD thesis, Ecole
Polytcchnique, Palaiseau, France.
[251] Kwok A.O., Stewart J.P., Hashash Y.M.A. (2008). Nonlinear ground response anal-
ysis of Turkey fi'lat shalln\v stiff soil site to strong ground motion, Bulletin of the
Seismological Society of America., 98(1), pp.331-343.
[252] Lachat, J.G., and J.O. Vlatson (1976). Effective nu1nerical treatn1ent of boundary
integral equations: a formulation for three-dimensional elastostatics, International
Journal for Nurncrical Methods in Engineering1 10 1 pp.991-1005.
[2,53] Lade P.V. (1977). Elasto-plastic stress-strain theory for cohcsionless soil with
curved yield surfaces, International Journal of Solids and Structures, 13 1 pp.1019-
1035.
[254] Laghrouche (1996). Sim,ulation num,irique de propagations d 1ondes dans les sols -
a
Application l1isulation vibrato£re, PhD thesis, 'University of Nantes 1 France.
[255] Lai C.G., Rix G ..T., Foti S., Roma V. (2002). Simultaneous Measurement and In-
version of Surface VVave Dispersion and Attenuation Curves, Soil Dynaniics and
Earthquake Engineering, 22(9-12), pp.923-930.
[256] Lai C.G. (1998). Sinmlta.neous inversion of Rayleigh phase velocity and attenuation
for near-surface site characterization, PhD I'hesis, Georgia Institute of '"rechnology.
[257] Lebrun 13., Hatzfclcl D., Bard P.Y. (2001 ). Site effects study in urban area: exper-
in1cntal results in Cirenoblc, Pure and Applied Geophysics, 158, pp.25t13-2557.
[258] Lee V.W. (1978). Displacements near a three-dimensional hemispherical canyon
subjected to incident plane waves 1 Il.eport No. CE 78-16, University of Southern
California. 1 126 p.
478 REFERENCES
[259] Lee V.W. (1984). Three-dimensional diffract.ion of plane P, SV & SH waves by
a hen1isphcrical alluvial vc1lley1 Soil Dynarnics and Earthquake Engineering1 3,
pp.133-144.
[260] Lefeuve-Mcsgouez G., Peplow A.T., Le Houedec D., .Jones D.V. (1998). Ground
vibrations due to high speed rnoving har1nonic load, 12th Engineering lvlechanics
Conference, San Diego, 17-20 Jlviay.
[261] Lcfcuve-Mesgoucz G., Peplow A.T., Le Houeclec D. (2002). Surface vibration clue
to a sequence of high speed rnoving harmonic rectangular loads, Soil Dynamics and
Earthquake Engineering, 22, pp.459-473.
[262] Le Tallec P. (2007). Mecaniq-uc des miliw.r, continus, Ecole Polytechnique,
Palaiseau, Ftance.
[263] Liao W.I., Teng T ..J., Yeh C.S. (2004). A series solution and numerical technique
for vvave diffraction by a three-dirnensional canyon, Wave ivfotion, 39 1 pp.129-142.
[204] Liao S., Sangrey D.A. (1978). Use of piles as isolation Larriers, Journal of the
Geotechnico.l Division (A.S.C.E), 104(9), pp.1139-1152.
[265] Light.hill M.J. (1965). Group velocity, Journal of the Institute of Mathematics and
its Applications, 1, pp.1-28.
[266] Liu H.P., Anderson D.L., Kanamori H. (1976). Velocity dispersion due to anelas-
ticity: Implications for scis1nology and rnantle composition, Geophys. J. R. Astron.
Soc., 47, pp.41-58.
[267] Lockner D.A., Walsh J.B., Byerlec .J.D. (1977). Changes in seismic velocity and
attenuation during deforn1ation of granite, Journal of Geophysical Research, 82(33),
pp.537 4-5378.
[268] Lokmane N., Bonnet. G., Semblat J.F., Driad L. (2003). Seismic loading due to
raining : vvavc arnplification and vibration of structures 1 Geophysical Research Ab-
stracts (EGS), 5, p.14204.
[269] Lombaert G., Clouteau D., Ishinwa 0., Mc"hcr N. (2004). The city-site effect: a
fu:.::z;y substructure approach and nun1erical sirnulations, 11th International Con-
ference on Soil Dynamics and Earthq'uake Engineering, University o:f California. at
Berkeley, USA.
[270] Lombacrt G., Degrande G. (2003). The experimental validation of a numerical
1nodel for the prediction of the vibrations in the free field produced by road traffic,
Jon·mal of Sound a.nd Vibration, 262, pp.309-331.
[271] Lomnit" C. (1957). Linear dissipation in solids, Journal of Applied Physics, 28,
pp.201-205.
REFERENCES 479
[272] Loret B., Simoes F.M.F., Martins J.A.C. (1997). Growth and decay of accelerntion
1vavcs in non-associative elastic-plastic fluid-staturated porous media; Jnternaliona.l
Journal of Solids a.nd Structures, :l4(13), pp.158:1-1608.
[273] Lore\ B., Luong IVLP. (1982). A clouble deformation mechanism model for sand,
Fourth International Conference on N1trnerical Methods in (ieom,echanics, Edrnon-
ton1 Canada.
[274] Lucct N. (1989). Vitesse ct attenna.tion des on.des elastiq-ucs soniques et -ultra.-
8oniques dans les rochcs .so11,.c; pression de confinerncnt1 PhD T·hesis, 1Jniversity
Paris VI, 264 p.
[275] Luong M.P. (1993). Centrifuge tests - A clynamic approach, So·il Dynamics and
Geotechnical Earthquake Engineering, pp.19-48, Balkema, Rotterdam.
[276] Luong 1vI.P. (1986). lVIcsure des propr.i6t6s d.ynamiqucs des sols, Re1Jue P'ranr;.aise
de Geotechnique, 37, pp.17-28.
[277] Luong M.P. (1986). Characteristic threshold and infrared vibrothermogrnphy of
sancl, A.S. T.M Geotechnical Testing Journal, 9(2).
[278] Luong J\1.P. (1980). Stress-strain aspects of cohesionless soils under cyclic and tran-
sient loading, International Sym,posium on Flo1Lc; under Cyclic and Transient Load-
ing: Soils under cyclic and transient loadi·ng) BaJkc1na, pp}J15-324 1 Svvansca.
[279] Lysmer J., Waas G. (1972). Shear waves in plane infinite structures, Journal of
Engineering Mechanics Di:vision (ASCE), 98(EM1).
[280] Lysmer ,). (1970). Lumped ma8' method for Rayleigh waves, Hulletin of the Seis-
mological Society of America, 60(1), pp.89-104.
[281] Ma S., Archuleta R.J., Page M.T. (2007). Effects of large-scale surface topography
on ground motions) as de1nonstrated by a study of the San Gabriel inonntains 1
Los Angeles, California, Bulletin of the Seismological Society of America, 97(6),
pp.2066-2079.
[282] MacCartor G.D., Wortman W.R. (1990). Analysis of nuc:lear Salmon free-fic:ld
ground motion data for non linear attenuation, Journal of Geophysical Research,
95(B13), pp.21805-21821.
[283] Makrn K., Chavez-Garcia F ..J., Raptakis D., Pitilakis K. (2005). Parametric anal-
ysiR of the seis1nic re.:iponse of a 2D sedirnentary valley: irnplications for code irn-
plementations of con1plex site effects, Soil Dynamics and Earthquake Engineering,
25(4), pp.303<Jl5.
[284] Makra K., Raptakis D., Chavec-Garcia F.J., Pitilakis K. (2002). How import.ant is
the detailed kr1owledge of a 2D soil structure for site response evaluation?, 12th
l!7uropean Conference on Earthquake Engineering, London 1 UK.
480 REFERENCES
[285] Manolis G.D., 13eskos D.E. (1988). Boundary F:lerncnt Methods in Elastodynamics.
Urnvin-H.yn1an (Chapman & IIall), London, UK.
[286] Mari .LL., Glangeau<l F., Coppens F. (1999). Signal processing for geologists and
geophysicists) Technip Ed., Paris, France 1 480 p.
[287] Masoumi H.R., Degrande G., Lombaert G. (2007). Prediction of free field vibrations
due to pile driving using a dynan1ic soil-structure interaction formulation 1 Soil
Dynamics and Earthquake Engineering, 27(2), pp.126-143.
[288] l\i[atasovic N., Vucctic 1\,1. (1995). GeneraJiz-;c<l cyclic-degradation-pore-pres.sure gen-
eration model for days, Journal of Gcotechnical Engineering (ASCE), 121(1),
pp.33-42.
[289] Matsushima S., Kawasc l:L (2006). Re-Evaluation of Near Source Strong Ground
l\.fotion During the 1995 I{obe Earthquake Considering The Heterogeneity of Source
and 3-D Basin, Third International Symposium on the Effects of Surface Geology
on S'eismic 1lfoiion, Grenoblc 1 France, paper No.141, pp.617-625.
[290] Mavko G.M. (1979). Frictional attenuation: an inherent amplitude dependence,
Journal of Geophysical Research, 84(B9), pp.4769-4775.
[291] Mavroeidis G.P., Papageorgiou A.S. (2003). A mathematical representation of near-
fault ground n1otions, Bulletin of the S'cism,ological Society of Arner1:ca1 93(3),
pp.1099-1131.
[292] May T.W., Dolt B.A. (1982). The effectiveness of trenches in reducing seismic
rnotion, Earthquake Engineering and Structural Dynarnics, 10, pp.195-210.
[293] Mayne P.W., .Jones .J.S. Jr (198:l). Impact stresses during dynamic compaction,
Journal of Geoiechnical Engineering (A.S.C.E), 109(10), pp.1342-1346.
[294] Medina F., Taylor R.L. (1983). Finite element techniques for problems of un-
bounded domains, International Journal of Numerical Methods in Engineering,
19, pp.1209-1226.
[295] Meek J.W., Wolf J.P. (1992). Cone models for homogeneous soil, Journal of
Geotechnica.l Engineering (ASCE), 118(5), pp.667-685.
[296] Mella! A., Modaressi H. (1998). A simplified numerical approach for nonlinear
dynamic analysis of multilayered media, 11th European Conference on Earthquake
Engineering, Balke1na ed., Paris, France.
[297] Menard-Soltraiternent (2002). Mesure de vibration lors de trmmux de cornpactage
dynarnique, Report by lvienard-Soltraiten1ent and 1neaHnre1nents Ly SEGG on a
dyna1nic compaction site in Huelva (Spain), Noiay, France.
[298] Menard L. (1974). La consolidation dyrmmique des sols de fondation, Annales de
l'I7'BTP, 320, pp.194-222.
REFERENCES
[299] Mestat P. (1993). Lois de cornporterncnt des geornateriau.T et rnodelisation par la
rnCthode des tlfments finis 1 Etudcs et recherches dcti labora.toircs des ponts et
chaussees, S6rie GCotechnique GT.52, Laboratoire Central dc8 Pants et Chauss€es 1
Paris, France.
[300] Meunier J. (1974). Contribution 0. l'eiude des ondes ct des ondes de choc dans les
sols, PhD T'hcsis 1 University of Grenoble, France.
[301] Meza-Fajardo KC., Papagcorgiou A.S. (20118). A Nonconvolutional, Split-Field,
Perfectly l\tJatched Layer for \i\rave Propagation in Isotropic and An.if.;otropic Ela8tic
Media: Stability Analysis, Bulletin of the Seismological Societ:y of America, 98(4),
pp.1811-1836.
[302] Mikhailenko B.G. (1999). Spectral Laguerre method for the approximate solution
of time dependent problems, Applied Math. Letters, 12, pp.105-110.
[303] Mizutami H., Geller lLJ., Takeuchi N. (2000). Comparison of accuracy and effi-
ciency of tiine-domain .schc1ne8 for calculating synthetic seis1nograms, Physics of
the Earth and Planetary Interiors, 119, pp. 75-97.
['.l04J Moczo P., Kristek J., Halada L. (2004). The Fin'ite Difference Method for seismol-
ogists: an introduction (SPICE project), 150 p., Con1cnius University, Bratislava,
Slovakia.
[305] Mocw P., Kristek J., Vavrycuk V., Archuleta R.J., Halada L. (2002). 3D Heteroge-
neous Staggered-Grid Finite-Difference l\iiodeling of Seisrnic 1\.J:otion with Volun1c
Harn1onic and Arithn1etic Averaging of }-::;lastic lVIodnli and DensitieH, B'Ulletin of
the Seisrnologi:cal Society of America., 92(8), pp.3042-3066.
[306] Moczo P., 13ard P.Y. (1993). Wave diffraction, amplification and differential motion
near strong lateral discontinuities, Bulletin of the 5\::isrnological Society of Anierica,
83, pp.85-106.
[307] Jlvfodaressi .i\11. 1 Denzenati I. (1992). An absorbing boundary ele1nent for dyna1nic
analysis of tvvo-phasc rnedia 1 10th World Conference on Earthquake E'ngineering 1
Madrid, pp.1157-1163.
[308] Moeen-Vaziri N., Trifunac M.D. (1988). Scattering and diffraction of plane SH-
v;,ravcs hy t\vo-dimensional inho1nogeneities 1 Soil Dynarnics and Earthquake Engi-
neering, 7(4), pp.179-200.
[309] Moeen-Vaziri N., Trifnnac M.D. (1985). Scattering of plane SH-waves by cylindrical
canals of arbitrary shape, Soil Dynamics and Earthquake Engineering, 4(1), pp.18-
23.
[310] Molenkarnp F., Kay S. (1997). Comparison of convergence of 8-node and 15-node
elen1cnts 1 Int. S'ympos'ium on Numerical lvlodels in Geomechanir:s (iVUll;JOG), 415-
419, .i\/IontrCal, Canada.
482 REFERENCES
[311] J\1orlet J., Grossmann A. (1984). Decomposition of Ha.rely functions into square
integrable wavelets of constant shape, S.J.A.M Journal Math. Anal., 15, pp.723-
736.
[312] Nlosscssian T.K, Dravinski M. (1990). Amplification of elastic waves by a three
dirncnsional valley, Earthquake Engineering and Structural Dynarn'ics, 19, pp.681-
691.
[313] Nliiller G. (1983). Rheological properties and velocity dispersion of a medium with
power-law dependence of Q on frequency, Jo-urnal uf Geophysics, 54(1), pp.20-29.
[314] cl Naggar l\II.H., Novak M. (1994). Non-linear model for dynamic axial pile response,
Journal uf Geotcchnical Engineering {A.S.C.E), 1211(2), pp.3118-329.
[315] Nagy Ci., J\/Iuraka1ni II., Hcgcrnicr G.A., Florence A.L., Experi1nental and analytical
study of the dynarnic response of low-porosity brittle rockr.;~ Journal of Geophysical
Research, 98(1312), pp.22081-22094, 199:l.
[316] Ng C.W.W, Li X.S., Van Laak P.A., Hou D.Y.J. (2004). Centrifuge modeling of
loot>e fill e1nbankment subjected to uni-axial and bi-axial earthquakes, Soil Dynam,-
frs and Earthqv.a.ke Engineering, 24(4), pp.305-318.
[317] l\guyen T.L., Scrnblat J.F., Rciffstcck P., Lenti L. (2008). Anisotropic Velocity
and Attenuation for Shear \i\Favcs in a Layered Soil by Resonant Cohunn 'rests,
Seismological Research Letters, 79(2).
[318] Nigbor R.L., lrrmi T. (1994). The suspension P-S velocity logging method, in Gcu-
physica.l Characterization of Sites, !SSMFE, R.D. Woods ed., 13alkema.
[319] Nishirnura N. (2002). Fast n1ultipole accelerated boundary integral equation rneth-
ods, Applied Jl1echanics Revie'ws, 55, pp.299-324.
[320] Niu Z.R., Zhou ILL. (20114). The natural boundary integral equation in potential
problems and regularization of the hypcrsingular integral, Computers and Struc-
tures, 82, pp.3L5-323.
[321] No-v1acki \i\T.K. (1978). Stress 1Daves in non elastic solids, Pergamon Press.
[322] Nowacki W.K, Raniecki B. (1987). Theoretical analysis of dynamic compacting of
soil aroun<l a spherical source of explosion, Archi'u. Mechanics, 39(4), pp.359-384.
[323] Pao, Y.H., Mow C.C. (1973). Diffraction of elastic waves and dynamic stress
concentrations, Crane, R,ussak & Con1pany lnc.
[324] Paolucci R, Mostabilini L. (21106). Non-dimensional site amplification functions
for basin edge effectR on seismic ground rnotion, Third International Symposium
on the Effects of Surface Geology on Se·isrnic Motion (ESG), Grenoble, France.
[325] Paolucci R., Spinelli D. (2006). Ground motion induced by train passage, Journal
of Engineering Mechanics, 132(2), pp.201-210.
REFERENCES 483
[326] Paolucci R., Maffeis A., Scandella L., Stnpai"ini M., Vanini M. (2003). Numeri-
cal prediction of loV\'-frcqucncy ground vibrations induced by high-t:>peed trains at
Lcdsgaard, Sweden, Soil Dyna.mies and Earlhqua.ke Engineering, 23, pp.425-433.
[327] Paolucci R (2002). Amplification of earthquake ground motion by steep to-
pographic irregularities, Earthqv,ake Engineering and Structural Dynamics) 31,
pp.1831-1853.
[328] Paolucci R. (1999). Shear resonance frequencies of alluvial valleys by Rayleigh's
n1cthod, Earthquake Spectra, 15, pp.503-521.
[329] Papagcorgiuu A.S., Kim J. (1991). Study of the propagation and amplification of
scisrnic vvaves in Caraca8 Valley \Vith .reference to the 29 July 1967 earthquake: SH
waves, Bulletin of the Seismological Society of America, 81, pp.2214-2233.
[330] Patera A.T. (1984). A spectral element method for fluid dynamics: laminar flow
in a channel expansion, Journal of Comp'ufational Physics, 54, pp.468-488.
[331] Pecker A. (20118). Dynamique des struci'ures et des onvrages, Ecole Nationale des
Pants ct Chairnsees, 220 p. (http://www.cnpc.fr/).
[332] Pecker A. (2005). 1\!Iaxirnurn ground surface motion in probabilistic seisn1ic haLlard
analyses, Jo·umal of Ea.rthqua.ke Engineering, 9(1), pp.187-211.
[333] Pecker A. (1984). Soil dynamics (in French), Presses de !'Ecole Nationille des Pants
ct Chauss8e.s, Paris, France, 259 pp.
[334] Pecker A., Dupas J.M. (1981). Etude experimcntale du comportement d'une vase
sous chargcmcnt cycliquc, Re1Jue Franr;aise de Geotcchniq1.Le, 5.
[:J35J Pedersen H.A., Campillo M., Sanchez-Sesma F.J. (1995). Azimuth dependent wave
arnplification in alluvial valleys) Soil Dynamics and Earthquake Engineering1 14(4),
pp.289-31111.
[336] Phillips E. (1869). De 1'6quilibre des solides 6lastiques, Com.ptes Rendus de
l'Acadimie des Sciences 1 torne 68) Paris, France.
[337] Piau M. (1975). Ondes d'acc6leration dans !es milieux 6lastoplastiques viscoplas-
tiqnes, Journal de Mecaniq=ue, 14(1).
[338] Pietzsch D., Poppy W. (1992). Simulation of soil compaction with vibratory rollers,
Journal of Terramechanics, 29(6), pp.585-597.
[339] Pitarka A.S., Takenaka H., Suetsugu D. (1994). Modeling strong motion in the
Ashigara Valley for the 1990 Odawara, Japan, earthquake, Bulletin of the Se=ismo-
logica..l Society of America, 84(5), pp.1327-1335.
[3411] Pitilakis KD., Raptakis D., Lontzetidis K, Tika-Vassilikou Th., Jongmans D.
(1999a). Geotechnical and geophysical description of EURO-SEISTEST using field,
laboratory tests and rnoderate strong-motion recordings, Journal of Earthquake
Engineering, 3, pp.381-409.
 REFERENCES
484
[341] Pitilakis K.D., Raptakis D.G., Makra K.A. (1999b). Site effects: recent considera-
tions and design provisions, 2nd International Conference on Earthquake Geotech-
nical Engineering, Lisbon, Balkerna ed, pp.901-912.
[342] Foran C ..T., Heh K.S., Rodriguez J.A. (1992). Impact behavior of sand, Soils and
Foundations, 32(4), .pp.81-92.
[343] Poussc G., Bonilla L.F., Cotton F., Margerin L. (2006). Nonstationary Stochastic
Simulation of Strong Ground lVIotion 'rime Histories Including NatuTal Variability:
A.pplication to the K-Net Japanese Database, Bulletin of the Seism,ological Society
of America, 96(6), pp.2103-2117.
[344] Prevost .J.H. (1985). A simple plasticity theory for frictional cohesionless soils, Soil
Dynamics and Earthquake Engineering, 4(1), pp.9-17.
[345] Pyke R (1979). Nonlinear soil models for irregular cyclic loadings, Ju·urnaJ of the
CJeotechnical Engineering Division (A.S.C.E), 105(6), pp.715-726.
[346] R.aptakis D.G., Chavez-Garcia F.J., Makra K., Pitilakis K.D. (2000). Site effects
at Euroseistest-L Deterrnination of the valley structure and confi·ontation of ob-
servations vvith lD analysis 1 Soil J)ynarn'ics and Earthquake Engineering, 19(1),
pp.1-22.
[347] Reddy J.l\. (2004). An .Introduction to Nonlinear Finite Element Analysis, Oxford
University Press, Oxford, UK, 488 p.
[348] Reinoso E., Wrobel L.C., Power H. (1997). Three-dimensional scattering of seismic
waves fro1n topographical structures, Soil Dynarnics and Earthquake Engincering1
16, pp.41-61.
[349] Richart F.E., Woods R.D., Hall J.R. (1970). Vibrations of soils and fo·undations,
Prentice-Hall.
[350] Ricker N. (1943). Further developments in the wavelet theory of seismogram struc-
ture, Bulletin of the Seismological Society of America., 33(3), pp.197-228.
[351] Ricker N. (1953). The form and laws of propagation of seismic wavelets, CJeophys·ics,
18, pp.10-40.
[352] Riepl J., Bard P.Y., Hatzfeld D., Papaioannou, Nechstcin S. (1998). Detailed eval-
uation of site response estirnation methods across antl along the seclirr1entary val-
ley of Volvi (EuroSeistest), Bv.llctin of the Seismological Society of America, 88,
pp.488-502.
[353] R,iviCre J. (1983). Interaction sol-structure lors d'un sfisrne ; sim,ulation en cen-
trifugense, PhD Thesis) lJnivcrsity Bordeaux I, France.
[354] Rocher-Lacoste F., Lenti L., Lassere! M., Saulc R. (2007). Cornparntive vibration
behaviour of vibratory and impact driven prefabricated concrete pileR 1 8th Int.
Conference on the Application of Stress- Wave Theory to Piles 1 Lisbon, Portugal
REFERENCES 485
[355] Rodriguez-Zuniga J.L., Sanchez-Sesma F.J., Perez-Rocha L.E. (1995). Seismic re-
sponse of shallo-vv alluvial valleys: the use of simplified rnodcls, Bulletin of the
Seismologica.l Society of America, 85(3), pp.890-899.
[:J56] Rovelli A., Scognamiglio L., Marra F., Caserta A. (2001). Edge-Diffracted 1-Sec
Surface \iVaves Observed in a Srnall-Size Intramountain Basin (Colfiorito, Central
Italy), Bulletin of the Seismological Society of America, 91(6), pp.1851-1866.
[357] Ryan H. (1994). Ricker, Ormsby, Klauder, Butterworth - A choice of wavelets,
Canadian Society of Exploration Geophys·icists Recorder, 19(7), pp.8-9.
[358] Sabctta F., Pugliese A. (1996). Estimation of response spectra and simulation of
nonstationary earthquake ground rnotions, Bnlleb:n of the Seismological b'ocicty of
America, 86(2), pp.337-352.
[359] Sadd M.H., Qiu Lun, Boardman W.G., Shukla A. (1992). Modelling wave pro-
pagation in granular rnedia using elaHtic net-virorks 1 International J onrnal of Rock
Mechanics and J\!Iining Science, 29(2), pp.161-170.
[360] Saenger E.H., Gold N., Shapiro S.A. (2000). Modeling the propagation of elastic
\VaveH using a modified finite-difference gric1 1 VVave Motion, ~11, 77-92.
[361] Salengon J., Handbook of continnurn rnechanics: general concepts 1 therrnoclasticity,
Springer, 2001.
[362] Sanche,-Sesma F ..T., Vai R., Drctta E., Palencia V.J. (2000). Fnndarnentals of
elastic Viravc propagation for site a1nplification studieR) Wave Motion in Earthquake
Engineering, E. Kauscl and G. J'vianolis (Editors), V\TIT Press, Southampton) UK,
pp.1-36.
[363] S{111chez-Sesma F ..T., Lm6n F. (1995). Seismic Response of Three-Dimensional
Alluvial Valleys for Incident P, S and Rayleigh \\Taves, Bulletin of the Seism,ological
Society of America., 85, pp.260-284.
[364] S'1r1chcz-Sesma F ..T., Ramos-Martinez .T., Campillo M. (1993). An indirect bound-
ary elerncnt n1ethod applied to siinnlate the scisrnic response of alluvial valleys for
incident P, S and Ilayleigh waves, Earthquake Engineering and b~tr'uci:ural Dynarn-
ics, 22, pp.279-205.
[365] Sanchez-Sesma F.J., Campillo M. (1991). Diffraction of P, SV Rayleigh waves by
topographic features: a boundary integral formulation, Bulletin of the Sei81nolog'ical
8ociety of Amer·-ica, 81, pp.2234-2253.
[366] Sanchez-Sesma F.J., Velazquez S.A. (1987). On the seismic response of a clipping
layer, Wa·uc Motian, 9(5), pp.387-391.
[3Ci7] Sanchez-Sesma F.J. (1985). Diffraction of elastic SH waves by wedges, B·ulletin of
the Seismological Society of America., 75(5), pp.1435-144().
486 REFERENCES
[368] Sancbez-Scsma F ..J. (1983). Diffraction of elastic wa.ves by tbree-dimensional sur-
face irregularities, Bulletin of the Seismological Society of Arne'T"ica, 73(6), pp.1621-
1636.
[369] Sanchez-Sesma F ..J., Rosenblueth E. (1979). Ground motion at canyons of arbitrary
shape un<lcr incident SH waves, Earthquake Engineering and Structural lJynarnics,
7, pp.441-:150.
[370] Saxena S.K., Reddy K.R. (1989). Dynamic moduli and damping ratioR for Monterey
11°0 sand by resonant cohnnn tests) Soils and Foundations, 29, pp.37-5:1.
[371] Scham M., Antes H. (1997). A new visco- and clastodynamic time domain bound-
ary clerncnt forrr1ulation) G'ornputational ~Mechanics, 20(5), pp.452-459.
[372] Scherbaum F. (2006). Of Poles and Zeros: Fundamentals of Digital Seismology,
Springer, 280 p.
[37:1] Scherbilum F., Pa.lmc C., Langer H. (1994). Model parameter optimization for site-
dependent si1nulation of ground 1n.otion by simulated annealing: re-evaluation of
the A.shigara valley prediction expcrirncnt, jValural Hazards, 10(3), pp.275-296.
[374] Schnabel P.D., Lysmer J., Seed H.B. (1972). SHAKE: A comp"Uter program for
carthqna.ke response analysis of horizonlally-la.yered sites, Report No. EERC-72/12,
Earthquake Engineering R.cscarch Center, 'University of California at Berkeley,
Berkeley, CA, USA.
[375] Schofield R. (2004). Seismic measurements at the Stateline Wind Project
and a prediction of the scis1nic signal that the proposed )/Ia.iden V\rind
Project would produce at LIGO. LIGO-T020104-00-Z LIGO Project,
http://www.ligo. cal tech. edu /docs /T/T020104-00. prlf.
[:176] Schwab F., Knopoff L. (1970). Surface-wave dispernion com]Jutations, B·ullci'i'a of
the Seismological Society of America, 60, pp.321-344.
[377] Scott R.A., Pearce R.W. (1975). Soil compaction Ly impact, Geotechnique, 25(1),
pp.19-30.
[378] Seed H.B., Wong R.T., Idriss I.M., Tokimatsu K. (1986). Moduli and damping fac-
tors for dyna.rnic analyses of coheHionless soils, Journal of Geotechnical Engineering)
112(11), pp.llll6-1032.
[379] Seed H.B. (1974). Landslides d·uring earthquakes due to soil liquefaction, Terzaghi
Lectures, Arnerican Society of Civil Engincers 1 pp.193-261, New-York.
[:J80] Sego! G., Lee P.C.Y., Abel J.F. (1978). Amplitude reduction of surface waves
by trenches, .lo-umal of the Engineering Mechanics Dfoision (A.S. C.E), 104(3),
pp.621-641.
REFERENCES 487
[381] Sernblat .J.F., Kbam M., Bard P.Y. (2008). Seismic wave propagation in alluvial
basins and influence of site-city interaction 1 Bulletin of the Seisrnological Society
of America, 98(6), pp. 2665-2678.
[382] Scmblat .J.F., Leroy Y. (2007). Sismologie ct Tectoniqv.e, Ecole Polyteclmique,
Palaiserrn, France (http://www.polytechnique.fr/).
[383] Sernblat J.F., D'1ngla P. (2005). Modeling wave pro]Jagation and soil-structure in-
teraction: Finite clc1ncnt and boundary element approaches, Bulletin des Labora-
toires des .Pants et Chaussir:s, 256-257 1 pp.163-178.
[384] Semblat .T.F., Kham M., Parara E., Dard P.Y., Pitihkis K., Makra K., Raptakis D.
(2005). Site effects: basin geo1netry vs soil layering, Soil Dynarnics and Earthquake
Engineeri:ng, 25(7-10), pp.529-538.
[:J85] Semblat .J.F., Paolucci R., Duval A.M. (2003a). Simplified vibratory characteriza-
tion of alluvial basins, G'omp!_e_s-Rendus Geoscience, 335(4), pp.365-370.
[386] Scmblat .J.F., Duval A.M., Dangla P. (2003L). Ylorlal superposition method for
the analy8iH of seismic 'vave an1plification) Bulletin of the S'eism,ological Socir:ty of
America., 93(3), pp.ll44-115:J.
[387] Semblat J.F., Dangla P., Kham M., Duv'11 A.M. (2llll2a). Seismic site effects for
shallo-vir and deep alluvial basiru~: in-depth n1otion and focusing effect) /)oU lJynarn-
ics and Earthquake Engineering, 22(9-12), pp.849-854.
[388] Semblat J.F., Duval A.M., Dangla P. (2002b). Seismic Site Effects in a. Deep Allu-
vial Dasin: Numerical AnalysiH by the boundary element rnethod 1 Coniputcrs and
Geoter-hnics 1 29(7), pp. 573-585.
[389] Semblat .J.F., Drioist J.J. (2001). Higher order finite elements for lD and 2D analy-
sis of nun1crical Vi.rave dispersion 1 10th International Conference on Com,puter Meth-
ods and Advances in Geornccha.nics (JACMAG), Tucson, USA.
[390] Semblat .J.F., Duval A.M., Dangla. P. (2000). Numerical analysis of seismic wave
a.rnplification in Nice (France) and cornpa.risons with 8Xperiments, Soil Dynarnics
a.nd Earthquake Engineering, 19(5), pp.347-362.
[:l91] Sembht .J.F., Drioist J.J. (2000). Efficiency of higher order finite clements for
the analysis of Heis1nic \vave propagation, Jo'Urnal of Sound and 11{bration, 231(2),
pp.460-467.
[392] SemL!at .J.F., Luong M.P., G'1ry G. (! 999). :JD-Hopkinson bar : new experiments
for dyna1nic testing ou soils, Soils and Fo'Undalions, 39(1), pp.l-10.
[393] Sernblat J.F., Luong M.P. (1998). Wave propagation through soils in centrifuge
experiments, .Jo·urnal of Earthq·uakc Engineering, 2(1), pp.147-171.
[394] Scmblat .J.F. (1998). Amortissement et dispersion des ondes points de vue
physique et rnunerique, lleV'UC Franc;aiBe de GCnic Ci'vil1 2(1), pp.91-111.
488 REFERENCES
[395] Semblat J.F., Brioist J.J. (1998). Wave propagation in non-homogenesouly damped
inediu1n: nu1nerical Vf-l expcriinental approach, 11th European Conference on Earth-
quake Engineering 1 Parir:;, France.
[396] Semblat .J.F. (1997). Rheological interpretation of Rayleigh damping, Journal of
Sound and Vibration, 206(5), pp.741-744.
[397] Semblat .J.F. (1995). Sols sous sollicitations rlynarniqu.es et transita.fres : reponse
dynarri'iquc aux barres de Hopkinson 1 propagation d 1ondes en rnilicu centrifuge, R,ap-
port Etudcs ct RccbercheR des Labora.toircs des Ponts ct Chaussecs, G'l'60, 206 p. 1
LCPC, PariR, France.
[398] Scmblat J.F. (1989). Precontra.fritc des sols et sollicitations dynarniq·ues, Master
thesis, ENTPF~, Vaulx-en-Vclin, France.
[399] Sessarego J.P., Sagcloli J., Degoul P., Flandrin P., Zakharia M. (1990). Analyse
temps-frequencc de signanx en n1ilicux diHpcrsifr;. Application ii. P6tude des ondes
de Lamb, Jov.ma1 d'Awustique, 3, pp.273-280.
[400] Scxtos A.G., Pitilakis K.D., Kappos A.J. (2003). Inelastic dynamic analysis of
RC bridges accounting for spatial variability of ground inotion, site effects and
soil-structure interaction phenorncna. Part 1: l\Iethodology and analytical toolR,
Earthquake Engineering fj Structural Dynamics, 32(4), pp.607-627.
[401] Shen R.H., Satake M., l\fohrabadi M., Chang C.S., Campbell C.S. (1992). Aclvances
in rnicro1nechanics of granular materials, Proceedings of the second US/Japan sem-
inar on niicrornechanics of granular rriaterials, E1sevier, 461 p.
[402] Sheng X., Jones C ..J.C., Thompson D.J. (2006). Prediction of ground vibration
from trainH u~ing the ·vi'avenurnbcr finite and bounrl_a.ry ele1nent rncthods, Journal
of Sound and Vibration, 293, pp.575-586.
[4o:J] Shibusawa S., Oida A. (1992a). Transient load test system for evaluating dynamic
Roil behavior hierarchies by differences in the period of motion, J ourn,al of Ter-
ramecha.nics, 29(2), pp.161-171.
[404] Shibusawa S., Oida A. (1992b). Dependency of observation parameters on soil
dynan1ic parameterH in unconfined i1npact co1npression tests, Journal of Terram,e-
chanics, 29(3), pp.289-306.
[405] Shukla A., Dam.ania C. (1987). Experirnental investigation of vvave velocity and dy-
namic contact stresses in an asscn1bly of disks 1 Journal of E.rper'irnental J11echanics 1
27(3), pp.268-281.
[clO(i] Shukla A., Zhu C.Y., Xu Y. (1992). Dynamic stresses in granular assemblies
with rnicrostructural defects, Journal of Engineering Mechanics (ASCB), 118(1),
pp.190-201.
[407] Sieffert J.G., Tison L., Thiel G. (1990). Rccherche d'un critere d'cfficacite en conrs
de pilonnage intcnsif, llevne Franr;aise de Giotechnique, 51, pp.61-67.
REFERENCES 489
[408] Sinclair G.B., Follarnbcc P.S., Johnson K.L. (1985). Quasi-static normal indenta·
tion of an clasto-plastic half-space by a rigid Hphere (IL R.csults) 1 International
Jov.ma.l of Solids and Strnctv.rcs, 21(8), pp.865-888.
['109] Singh S.K., Mena E., Castro R. (1988). Some aspects of source characteristics of
the 19 Scpten1ber 1985 l'viichoacan earthquake and ground motion amplification
in and near l\Jexico City frorn strong rnotion data, Bulletin of the Sei.sm.ological
Society of America, 78(2), pp. 151-477.
[410] Skipp B.0. (1997). Ground vibration-codes and standards, Proceedings of the Con·
fcrence on Ground Dynamics and Man-m.ade processes, lru>titution of Civil Engi-
neerfl, London, UK.
[411] Skipp 13.0. ed. (1966). Vibmtion in civil engineering, Symposium of the Inter·
national Association for Earthquake Engineering (London 1965), Bntterworths,
London, UK.
[412] Skoglund G.R., Marcuson W.F., Cunny R.W. (1976). Evaluation of resonant
cobunn test devices, Journal of Geotechnical Engineering Division (ASCE),
102(GT11).
[413] Sladek V., Sladek J., Tanaka M. (2001). Numerical integration of logarithmic
and nearly loga.rithrnic singularity in BElvis, Applied Mathem,atical ivfodelling, 25,
pp.901-922.
[414] Sladek V., Sladek J. (1998). Singular integrals and boundary elements, Comp-utcr
]\![ethods in Applied Mechanics and Engineering 1 157, pp.251-266.
[415] Sladek V., Sladek J., Tanaka M. (1997). Evaluation of l/r integrals in BEM
for1nulationH for 3-D proble1ns nRing coordinate 1nultitransfor1nations, Engineering
Analysis with Boundary Elements, 20, pp.229-244.
[416] Sloan S.W., Randolph F.M. (1982). Numerical prediction of collapse loads using Fi·
nite elcrnent methodR, International Journal for Nv,m,erical and Analytical ]\![ethods
in Geom,echanics, 6 1 pp.47-76.
[417] Sommerville P.G. (1998). Emerging art: earthquake ground motion, ASCE
Geotechnical Special Publications, Dakoulas et al eds, l, pp.1-38.
[418] Sorrentino S., Fasa.na A. (2007). Finite elernent analysis of vibrating linear systemR
v.rith fractional derivative viscoelastic n1odels 1 Journal of Sound and Vibration, 299,
pp.839-853.
[419] Stephenson RW. (1978). Ultrasonic testing for determining dynamic soil moduli,
lJynamJc (i\;otechnical Testing 1 Arnerica.n Society for Testing and 1Iaterials (ASTJVI
STP 654), pp.179-195.
[420] Stewart J.R., Hughes T.J.R (1998). A tutorial in elementary finite clement error
analysis: a Rysterna.tic presentation of a priori and a posteriori error cs-ti1nates,
Computer Methods in Applied Mechanics and Engineering, 158, pp.1-22.
 490 REFERENCES
[421] Stewart R.R., Toksoz M.N., Timur A. (1983). Strain dependent atteuuation: ob-
servations and a proposed 1nechanis111 1 Journal of Geophysical Research1 88(Dl),
pp.546-554.
[422] Stokoe K.H., Nazarian S. (198fi). Use of Rayleigh waves in liquefaction studies, in
IVleas'Urernent and use of Shear Wave Velocity for E'ualuating Dynamic Soil Prvp-
erties (ASCE), RD. Woods ed., New York, USA, pp.1-17.
[423] Stokoc K.H., Hoar R.J. (1978). Variable affecting in situ seismic measurements,
Proceedings of the (}onference on Earthquake Engineering and Soil Dynarnics,
ASCE, Pasadena, lJSA.
[424] Stoll R..D. (1979). Experimental studies of attenuation in sediments, Journal of the
Acoustical Society of America, 66(4), pp.1152-1160.
[425] Studman C.J., Moore M.A., Jones S.E. (1977). On the correlation of indentation
experiments, Jo·umal uf Physics (D: Applied Physics), 10, pp.949-956.
[426] Stupazzini M., A Spec/;ral clement approach for SD dynamic soil-structure interac-
tion problems, PhD Thesis, Politecnico di Milano, 2004.
[427] T'1kahashi A., T'1kerrmra J. (2005). Liquefaction-induced large displacement of pile-
m1pporte<l vvha.rf, Soil Dynarnics and Earthquake Engineering; 25, 11, pp.811-825.
[428] Takahashi T., Nishimura N., Kobayashi S. (2003). A fast DIEM for threc-
<lirnensional elaHtodync.lmics in tirne do111ain, Engineering Analysis u;ith Boundary
Elements, 27, pp.491-506.
[429] Tanaka M., Sladek V., Sladek J. (1994). Regularization Techniques Applied to
Boundary Element Methods. Applied Mechanic8 Reviews, 47, pp.457-499.
[430] Tarantula A. (1987). Inverse Problem Theory, Elsevier, Amsterdam.
[431] Taylor P.W., Larkin T ..J. (1978). Seismic site response of nonlinear soil media,
Jmrrnal of the Geoiechnica.l Eng·ineering Di·uis-ion (A..S.C.E), 104(3), pp.369-393.
[432] Taylor T., Fragaszy R.J., Ho C.L. (1991). Projectile penetration in granular soils,
Ju·ur-n.al uf Geotechnical Engineering (A.S.C.E), 117(4), p]J.658-672.
[433] Theodulidis N.P., Bard P.Y. (1995). Horizontal to vertical spectrnl ratio and ge-
ological conditions: an analyHi.s of strong motion date from Greece and Tahvan
(SMATIT-1), Soil Dynamics and Earthq11.ake Engineering, 14(3), pp.177-197.
[434] Thomson W.T. (1950). Transmission of elastic waves through a stratified solid
medium, Journal of Applied Physics, 21, pp.89-93.
[435] Transvib, (2006). Intemationa.l Syrnposi11.m on Vibratory pile driving and deep soil
'vibratory cornpaction, H. Gonin, A. Holeyn1an, F. llocher-Lacoste ed.s) LC.PC)
Paris; France.
REFERENCES 491
[436] Tribolet J.M. (1077). A new phase unwrapping algorithm, IEEE Transactions on
Acoustics, Speech and Signal Processing, 25(2), pp.170-177.
[437] Trifunac M.D., Drady A.G. (1975). On the correbtion of seismic intensity scales
1.vith the pcakR of recorded Htrong ground motion, Bulletin of the Seisrnological
Society of America, 65(1).
[438] Trifunac M.D. (1973). Scattering of plane SH waves Ly a semi-cylindrical canyon,
Int. Journal of Earthquake Engineering and Structural Dynamics, 1, pp.267-281.
[139] Tsogka C., Wirgin A. (2003). Seismic response of a set of blocks partially embedded
in soft wil, Cornptes-Rendus Mecanique, 331(3), 217-224.
[440] Towhata I. (2008). Geotechn·ica.l Earthquake Engineering, Springer, 684 p.
[441] Tubach J.P. (1989). Analyse spectrale de la parole, in La pa.role et son traitcrn.ent
autom,atiquc, :l'viasson Ed.) Paris, France, pp.265-:-~0l.
[442] Uetake T., Kudo K. (2005). Assessment of Site Effects on Seismic Motion in
Ashigara Valley, .Japan, Bulletin of the Seismological Society of America, 95(6),
pp.2297-2317.
[443] Ulrych T.J. (1971). Application of homomorphic deconvolution to seismology, Geo-
physics, 36(4), pp.650-660.
[444] Veyera G.E., Ross C.A. (1995). High strain rate testing of unsaturated sands using
a S.H.P.B., Third International Conference on Rr:cent Advances in GcotechnicaJ
Earthquake Engineering and Soil Dynarnics 1 St-Louis, USA.
[445] Virieux J. (1986). P-SV Wave Propagation in Heterogeneous Media: Vclocity-
Stress Finite-Difference Method, Geophysics, 51, pp.889-901.
[446] Wathclct M., .Jongrnans D., Ohrnberger M. (2004). Surface wave inversion using
a direct Hearch algorithrn and itH application to an1bient ·vibration rneasnrernents)
Near Surface Ceoph:ifsics, 2, pp.211-221.
[447] Whitman RV. (1957). The behaviour of soils under transient loadings, Proceedings
of the 4th International Conference on Soil Mechanics and Foundation Engineering,
vol.I, p.207, London, UK.
[448] Wirgin A., Bard P-Y. (1996). Effects of buildings on the duration and amplitude
of groun<l 1notion in l\/Iexico city, Bulletin of the Seisrnological Society of America,
86, pp.914-920.
[449] Wolf .J.P. (2003). The scaled bovndary finite element method, Wiley, Chichester,
UK.
[4511] Wood W.L., Dossak M., Zienkiewicz O.C. (1981 ). An Alpha modification of New-
rnark's n1ethod, International Jo'urnal for Nnrnerical l\llethods in Engineer'ing1 15,
pp.15G2-1566.
492 REFERENCES

[451] Woods R.D. (1978). Measurement of dynamic soils properties, Proceedings of the
Conference on Earthquake Engineering and Soil Dunarnics 1 ASCE, Pasadena, 1JSA.

[452] Woods R.D., 13arnet.l N.E., S'1gesser R. (1974). Holography-A new tool for soil
dynamics, Jo"Urna.l of the Geotechnica.l Engineering Division {A.S.C.E), 100(11),
pp.1231-1247.

['15:J] Woods R.D. (1968). Screening of .mrfa.ce wa"Ues in soils, Technical report IP-804,
University of J'vfichigan, USA, 57 p.

[454] Yerli H.R., Terncl 13., Kiral E. (1999). Multi-wave transient and harmonic infi-
nite elements for tvvo-dirncnsional unbounded do1nain problcn1s, Cornp'Utcrs and
Geotechnics, 24(3), pp.185-206.

[455] Yokoi T. (2003). The higher order Born approximation applied to improve the
solution of seismic response of a three-dirnensional canyon by the Indirect Boundary
Method, Physics of the Emth and Planetary Intcr·ior, 137, pp.97-106.

[456] Yoo T.S., Selig E.T. (1979). Dynamics of vibratory-roller comp'1ction, Journal of
the Geotechnical Engineering D'i'Vision (A.S.C.E), 105(10), pp.1211-1231.
Index
[457] Yoshimura. C., 13ielak J., Hisada Y., Fernandez A. (2003). Domain Reduction
:Yiethod for Three-Din1ensional :B~arthquake J'viodeling in Localized Regions, Part
JI: Verification and Applications, Bulletin of the Seismologica.l Soci:ety of America,
93(2), pp.825-840.

[458] Zahradnik J., Moczo P., Hron F. (1994). 13lind prediction of the site effects a.t
Ashigara valley, Japan, and its co1nparison \Vith reality, Natv,rnl Hazards, 10(1-2),
pp.149-170.

[459] Zhao H., Gary G., Klcpaczko J.R (1997). On the use of a viscoelastic split I-Iopkin-
Hon pressure bar 1 Interrrational Journal of Irnpact Engineering, 19(4) 1 pp.319-330.

[460] Zhao J.X. (1996). Estimating modal parameters for a simple soft-soil site having a
linear <liHtribution of shear vv~avc velocity 1vith depth, Earthquake Engineering and
Str"Uct"U.ral Dynamics, 25(2), pp.163-178.

[461] Zienkiewicz O.C., Taylor R.L. (20115). The Finite Element Method for Solid a.nd
Structural Nlechanics, }~].sevier Butter\vorth-Heinemann, 631 p.

[462] Zicnkicwicz O.C., Bando K., Bcttcss P., Emson C., Chia.m T.C. (1985). Mapped
infinite elernents for exterior v.ravc problc1ns, International Journal for JVumerical
Methods in Engineering, 21(7), pp.1229-1251.
 Index
Absorbing boundaries
absorbing layers (PML), 289, 291, 292
l)o1nain llc<luction lVlcthods, 297
filtering PML, 292, 29:J
infinite Eilc111cnfa.;, 286 -28H
local absorbing condition (JD), 70, 283
local absorLing condition (2D), 28'1, 285
rnultidircctional Pl\dL, 293-295
Anisotropy, 137
Attenuation, 79, 83, 91, 92, 125, 277, 29:J,
386
Attenuation factor, 86, 125, 197, 204, 258
Beam vibrations
cone rnodcls, 120
eigenn1odes (axial), 62
eigenn1odcs (torsion), 73
norn1al force, 68
Principle of Virtual Work, 53
13last
I-Iopkinson bar tests, 103
mines (in), 33
quarries (in), :J4
Boundary Ele1nent J'vlethod
examples, 344, 348, 383, 397
Fast Multipole Method, 3116
FEM/BEM coupling, 296
forn1ulation, 298
frequency dornain 1 298
fundarnentaJ solutions, 299
integral equation, 299
integral representation, 299
introduction, 240
radiation conditions, 303
regnlarb1ation, 301
time dornain, 308
unbounded rnedia 1 :)02
valid'1tion, 304
Causality, 77
Centrifuge experin1cntR
dynarnic loading, 192
general, 190
geon1etrical da1nping, 197
numerical inodel) 204
scaling factors, 188, 189
simplified attenuation, 197
viscoelastic n1odcls) 197
viscoelastic sirnulations 1 199
Cepstrum, 449, 451
Co1npaction
dynamic, 24
vibratory, 28
Co1nplex n1odnlus, 257
Cone n1odels, 119
Constitutive equation, 59, 72, 75, 112 1 133,
137, 247
Creep function, 77, 82
Cross hole, 230, 234, 235
Cross-correlation, 233, 448
Cyclic tests, 122
Dan1ping
Caughey, 275
damping profiles, 279
da1nping ratio, 94, 95, 101, 126 1 275,
276
fractional derivatives, 278
frequency band-width, 101, 126
geometrical, 76 1 175 1 197, 199
logarithn1ic decrement, 101 1 125 1 126
material, 76, 197, 199, 277
 INDEX INDEX 497
496

FEM/BEM coupling, 296 efficiency, 45, 314, 317 eigenfrequcncies 184

1
R.aylcigh, 275, 276, 278 holes (by) , :J 14-317 transfer function 1 183, L84, 186
Discrete V\Tavenu1nbcr iviethod, 24() higher order elen1ents, 256 1 260, 269,
272, 273 identical t\vo-layers case, 171 Plane vvavcs in half-space
Dispersion (1..vave) interface wave, 168 P- SV case, 177
centrifuge experiments (in), 207 infinite clernents, 286
introduction, 240 rnultilayered case, 168 SH case, 176
dispersion curve, 213, 220, 222, 223 nnmerica.l sirnulations, 44
isopara1netric elernent, 251 Plastic waves, 12:i
group ·velocity, 87 optimal layer thickness, 157 Polarization, 139 143
linear acceleration rr1ethod, 254
Love waves, 208, 209, 214 piles (by), 314, 318, :ll9 Principle of Virtua.l V\Tork
mass matrix, 264, 268, 269
XCQ model, 93 single layer (P), l(j() general, 54, 131, 247) 298
rr1ultistep methods, 254
phase velocity1 87 single layer (SH), 154, 159 rate of work (external forces), 56, 132
Rayleigh waves, 216, 217 Nev.rmark n1ethod, 25:-3
numerical anisotropy, 275 single layer ( SV), J65 rate of vvork (internal forces), 5'1-56,
rheology (due to), 87, 258 trains (for), 46 l:l2
nurnerical dan1ping, 255, 258, 275, 281
Do\vn hole, 227 transmission (multilaycrcd), 169, 172 rate of vvork (quantities of acceleration),
nu1nerical dispersion, 257-260, 262, 26.c!,
Dura.tion 1 445, 1146 tra11s1nission coefficient (I'), 163 57, 132
265, 270, 273, 275
quadrilateral ele1nent, 249, 252, 262 trarirnr1ission coefficient (S.H), 156, 158,
Eigenfrequencies 160 Quality factor
lD heterogeneous soil, 117, 209, 213 Rayleigh damping, 275, 276, 278
trans1nission coefficient (S'V), 167 constant Q rnoclel, 91
2D basins, 378, :J79 reference elen1ent, 251
trenches (by), 42, 313, 318, 320 definitions, 125
alluvial basinf>, 357 shape functions, 248-250
spatial discretization, 248, 253 exa.rnple, :-354
rer.;ouant bar, 122 Kolsky test, 123 Hardin's la\v, 279
resonant eolurnn, 99 stability conditions, 254
tirr1e discrcti.,,;ation, 253 nearly constant Q rnodel, 92
Eigcnrnocles Lame corntants, 133 rheology (from), 79, 81, 83
lD axial (barn), 63 tirne steps, 264 J_.1aplacian, 419 rise ti1ne, 124
lD heterogeneous ooil, 116, 117, 214, tin1e-integratio11 scherr1cs, 253 Loading history1 77 Saxena's la"\v, 279
374 triangular ele1nent 1 252, 262
spectral ratio, 124
lD torsion (bars), 7 4 Vireak forn1ulation 1 24 7 Maxwell-Betti theorem, 298
Ste"\~n1rt's law, 280
alluvial basins, 359 Fourier transform, 78, 84, 140, 183, 200, Modulus
resonant eoltnn11 1 99, 100 219, 235, 4:J3, 435 -438 cornplex 1 78 1 81
Reflection- refraction at an interface
Equilibriun1 relaxed 1 7n
P-wave, 147, 150
lD rnediurn, 57 Galerkin method, 248 resonant column (fro1n) 1 99
SH-v.rave 144, 145
3D (displacements), 134 Gradient, 41Y unrela,-xed, 79 1

3D 111.edium, 133
Filtering
examples, 207, 440
filters, 43:l, 439
homomorphic, 206, 451-455
Finite Difference Method, 240
Finite Elernent 11ethod
a-HI-IT method, 254
average acceleration rnethod, 254
Caughey darnping, 27G
central difference rnethod 1 253, 254, 264
critical frequency) 254
Domain R,ednction J'v1ethods) 297
equation of rnotion, 252
Group velocity, 87
I-letcrogencmrn soil profile, 112, 2H, 374
Hilbert transform, 440, 441, 449
Hopkinson bars
3D device, 108
axial strain, 107
axial stress, 107
experimental device, 104
radial streBs, 110
strain in specirnen, 107
stress in specirnen, 108
"\vave reflection, 69, 104
Isolation (vibration)
SV-vvave 1 152
Nnisances (vibrations), 20, 32, 33, 37 1 39 arr1plification, ~i23
critical incidence, 147, 150, 152, 225
equation of continuity, 144, 149
Oedometric test, lo:J, 108, 110
heterogeneous bar, 68
Particle velocity, 23, 27, 28, 35, 37, G2, 107 r~Iopkinson Lars, 104
Peak (;round ruotion) 444 interface v.rave, 147, 150, l,52
Phase velocity, 86, 87, 207 plane "\vaves. 142
Pile driving R,elaxation function, 77, 82
dynan1ic pile driving, 21 Resonant colun1n
longitutlinal "\va.vcs, 64 actual test, 99
vibratory pile driving, 22 darnping, 101
>Nave reflection, 21, 23, 65 equivalent linear rnodel, 94
Plane vvaves in a surface layer experimental results, 102
h'I-I-..~vaves, 180 principles, 97
498 INDEX JNJJEX

specirncn n1otion 1 98 topography aggravation factor, 336, 339, single layer (SV), 165 \i\.Ta.ve vector, 140
v. ave equation, 98
1 356 V\.Tave veloeit;y
Ricker "\vavclct transfer function, 328 Ultrasonic v;.rave8, 122 Vp /V:s ratio, 135
original, 241 triangular canyonH, 340-342 1J nhounded 1nedia, 28:1 11) longitudinal vvaves, 60, 1:)8
simplified, 243 Volvi basin, :J66, 393, 394, 396 398, Up hole, 227 anisotropic 1ncdiun1) 1~17
400, 406 esti1nation 1 447
Scattering l'.rall-layer rnodel, 355 Variational for1nulation) 248, 253 heterogeneous soil, 113
cylindricc:Ll cavity, 308 Site-city interaction, 413, 414, 416 Velocity ratio, 145, 150, 152, 155-157, 159, pressure \\raves, 104 1 107, 135
spherical cavity, 309 311 Snell-Descartes law, 144, 149, 152, 159, 162, 169 Rayleigh, 178
Seis1nic cone; 229 225 Vibrations shear v;.raves, 72) 76) 98, 135, 154, 185
Sehimic refraction, 22d Soil-structure interaction) 119 blast, 35 typical values, 60, 72, 136
Shocks Spectral Analy8i8 of Surface \i\TaveH, 208, cornpaction, 28 \!\lave-structure interaction) 410
dyna1nic compaction, 24-26 218 foundation, 3:12 Waveficld
Hopkinson bar tests, 103 Spectral Elernent l\!Iethod, 240, 255 1 3 -19
1 isolation, 45 P--,,vaves, 135, 139, 195 1 2L15
Site effects Spectru1n pile driving, 24 ,5-vvaver>, 135) 245
2D aggravation factor, 366, 388, 406 arnplitucle, 433 road, 36 SH-waves, 139
2D resonance, 358, :-H:iO complex, 433 trains, 41, 46, 47, 321 S11-Viraves, 139
Ashigarn valley, 324 Fourier, 433 -,,vind turbines, ~11 body \¥aves, 135
Caracas basin, 325 phase, 433 Viscoelasticity displacen1ents 1 1113

Caracas topography, 348 Standard spectral ratio, 328, 380, 381 constitutive law, 76 rnonochro1natic v;.raveH, 140
circular canyons, 342 Steady State Rayleigh Method, 218 Kelvin-Voigt model, 80, 89, 200 plane \Vaves, 138
Civita di Dagnoregio, :-350-352 Strain level, 19, 102, 103, 126, 128 Ma""vell model, 80, 89, 200 polarization, 139 143
cylindriml basin, 361, 363, 364 Strain rate, 21, 26, 103, 108, 110, 112 relaxation time) 80 llayleigh, 179
dipping layer, 352-:-355 Strong :for1nnlation) 246 rheological nrodelsi 79 spherical, l:J8, 174, 193, 197, 198
elliptical basin, 365 Super-seismic, 41 viscosity, 201 1 203 surface waves, 140
I! Zener model, 80, 82, 84, 200 uncoupled, 134, 142
cxpcrirnental charactcri,,,ation, 327 Surface waves
Grenoble baain, :-324 dispersion, 208, 2H, 216, 218 Wavelength, 43, 99, 126, 157, 158, 218, 220,
Kobe basin, 50 Love, 208, 210, .214 I \,\Tave equation 261, 264
Las Colinas (landslide), 353 Love wave8, 179 lD solution ('L'Cial displacement), 61, Wavelets
Los Angeles basin, 324 71 double-M, 243, 4:J2
Rayleigh, 43, 177, 178, 216
1\.1exico basin, 324 Suspension logging, 227 lD solution (axial stress), 62, 66, 7l Gabor, 424
lD 8olution (torsional rnotion), 73 Generalized Rayleigh wavelet, 426
moon-valley model, 372, 37:1
Tensors, 419 ID Rolution (torsional stre88), 73 Kupper, 427
Nice basin, 376, 380, 382 384, 387
Tirne discretization) 248 cone rnodels 1 120 l\ilavroeidis & Papageorgion, 425
rectangular basin, :-356-:-358
heterogeneous soil; 114 Meyer, 430, 431
reference station, 381, 394, 403 Traffic
double-lVI loading, 243 longitudinal vibrations (beams), 59, 138 Morlct, 429, 430
serni-clliptioidal basin, ~172
moving loads, 40 plane v;ravcs, 138 Ormsby, 428
semi-spherical haHin 1 367-371
resonant colurnn, 98 Ricker (damped pulse), 241., 242
serili-spherical canyo11, 344 raihvays, 39
Ricker-Stokes, 241 Ricker (simplified), 243, 421-423, 436
shape ratio, 358, 360, 361, :J65 road) 36 1 38
Transfer function, 183 187, 328, 329, 33;1, shear vibrations (bea1ns), 75 triangular 1 88
sine-shaped (3D), 369, 370
spherical, 17 4 V\.Tavenumber
spectral a1nplification, 330 382
ton.;ional vibrations (beams), .71 cornplex, 85
time-domain amplification, 3~10 T'ransrnittivity
uncoupled) 135 dispcrsion 1 87
topography (2D), 3:l4 1nultilayered· casc 1 168
viHcoela8tic 1ncdium, 85) 257 V\Teak forrnulation, 24 7
topography (crest), 336 single layer'(P), 160
viscoelastic solution, 86 199 \i\tind tnrbineR) 3]
topography (hill), 338 single layer (SH), 154, 159 1
498 fNDEX INDEX 499

spccin1cn motion, 98 topography aggravation factor, 336, 339, single layer (SV), 165 \i\Fave vector, 140
vvave equation, 98 356 \Vave velocity
R,ickcr v,ravclet transfer function, 328 Ultrasonic \Vaves, 122 Vp /V,, ratio, 135
original, 241 triangular canyons, 340 342 Unbounded rnedia, 283 lD longitndinal 1..vaves 1 60, 138
I
simplified, 243 Volvi basin, 366, 393, 394, 396-398, '' Up hole, 227 anisotropic rnedhun, 1:~7
400, 406 esti1na.tion, 447
Scattering Vl.'all-layer n1odel, :155 Variational forn1ulati.on, 248 1 253 heterogeneous soil, 113
cylindrical cavi\y, 308 Site-city interaction, 413, 414, 416 Velocity ratio, 145, 150, 152, 155-157, 159, pressure 1,vaves, 104, 107, 135
spherical cavity, 309 311 Snell-Descarteslaw, 144, 149, 152, 159, 162, 169 Rayleigh, 178
Seismic cbne, 229 225 Vibrations .shear ¥.raves, 72, 7G, 98, 135, 154, 185
Seisrnic refraction, 224 Soil-structure interaction 1 119 blast, 35 typical values, 60, 72, 136
Shocks Spectral Analysis of Surface \~Taves, 208, cornpaction,. 28 V\Tave-structurc interaction, 1110
dynarnic cornpaetion, 24-26 218 foundation, 312 Wavcfield
Hopkinson bar tests, 103 Spectral Element Method, 240, 255, 349 isolation, 45 P-vvaves, 135 1 139, 195, 245
Site effects Spectrun1 pile driving, 24 S-wavcs, 135, 245
2D aggravation factor, 366, 388, 406 amplitude, 433 road, 36 S~H--1vaveR, 139
2D resonance, 358, 3611 con1plex) 433 trains, 41, 46, 47, 321 SV-waves, 139
Ashigara valley, 324 Fourier 1 !1:1;1 VI.rind turbines, 3:1 body 1vavcs, 135
Caracas basin, 325 pbase, 433 Viscoelasticity displacen1cnts, 143
Caraca.s topography, 348 Standard spectral ratio, 328, 380, 381 conRtitutivc law, 76 rnonochrornatic 1vaves 1 140 ·
circular canyons, 342 Steady State Rayleigh Method, 218 Kelvin-Voigt model, 80, 89, 2110 plane ¥.'aves, 138
Civita di Bagnoregio, 350-352 Strain level, 19, 102, 103, 126, 128 Maxwell model, 80, 89, 200 polarization, 139-143
cylindrical basin, 361, 363, 364 Strain rate, 21, 26, 103, 108, llO, 112 H relaxation tin1e, 80 Rayleigh, 179
F
<lipping layer, 352-355 Strong for1nulation, .246 ',I' rheological 1nodelR 1 79 spherical, 138, 174, 193, 197, 198
elliptical baRin, 365 Super-seismic, 41 I viscosity, 201, 203 surface Vi'aves, 140
experin1ental cha,ractcrizatibn 1 327 Surface v.,raves Zener model, 811, 82, 84, 200 uncoupled, 134, 142
Grenoble basin, :-324 dispernion, 208, 214, 216, 218 Wavelength, 43, 99, 126, 157, 158, 218, 220,
Kohe basin, 50 Love, 208, 210, 214 I V\Tave equation 261, 264
Las Colinas (landslide), :J5:l J,' lD solntion (axial displacement), 61, Wavelets
Love waves, 179 '
Los Angeles basin, 324 Rayleigh, 43, 177, 178, 216 71 cluuble-M, 243, 432
Mexico basin, 324 Suspension logging, 227 lD solution (axial stress), 62, 66, 71 Gabor, 42,1
rnoon-valley model, 372, 373 ID solution (torsional rnotion), 73 Generalized Rayleigh wavelet, 426
Nice basin, 376, 380, 382-384, 387 ·Ten8ors, 419 lD solution (torsional t>tress), 73 Kupper, 427
rectangular Lasin, ~-3;)6-358 Tin1e discretization, 248 Ii
i' cone n1odels, 120 1\tfavroeidis & Papageorgiou, 425
Meyer, 430, 431
heterogeneous soil, 114
reference station, 381, 394, 403
serni-cllipsoi<lal basin, :172
Traffic
double-M loading, 243 I
,1 longitudinal vibration.s (bearn8) 1 59, 138 Morlet, 429, 430
!
serni-spherical basin, 367--371 n1oving loads) 40 plane \vaves, 138 Ormsby, 428
senii-spherical canyo11, 344 raihvays, 39 resonant colnn1n, 98 Ricker (clamped pulse), 241, 242
sh'1pe ratio, 358, 360, 361, 365 road, 36, 38 Ricker-Stokes, 241 Ricker (simplified), 2'1:J, 421-423, 436
sine-shaped (3D), 369, 370 Transfer function, 183-187, :128, 329, 333, Rhear vibrations (beams), 75 triangular 1 88
spectral amplification, 330 382 spherical, 174 \::Y'avenumber
ti1ne-don1ain amplification, 330 Transmittivity torsional vibrations (bean1s), 71 complex, 85
topography (2D), 331 multilayered, case, 168 uncoupled 1 135 dispersion, 87
topography (crest), 336 single layer'(P), 160 viscoelastic mediun1, 85, 257 VVeak for1nulation, 247
topography (hill), 338 single layer (SH), 15'!, 159 viscoelastic solution, 86 1 199 \Vind turbines, 31