This page intentionally left blank

Dynamic Behavior of Concrete and Seismic Engineering

This page intentionally left blank
Dynamic Behavior of Concrete
and Seismic Engineering

Edited by
Jacky Mazars
Alain Millard
First published in France in 2004 by Hermes Science/Lavoisier entitled: Comportement dynamique des
bétons et génie parasismique © LAVOISIER, 2004
First published in Great Britain and the United States in 2009 by ISTE Ltd and John Wiley & Sons, Inc.

Apart from any fair dealing for the purposes of research or private study, or criticism or review, as
permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced,
stored or transmitted, in any form or by any means, with the prior permission in writing of the publishers,
or in the case of reprographic reproduction in accordance with the terms and licenses issued by the CLA.
Enquiries concerning reproduction outside these terms should be sent to the publishers at the
undermentioned address:

ISTE Ltd John Wiley & Sons, Inc.

27-37 St George’s Road 111 River Street
London SW19 4EU Hoboken, NJ 07030
UK USA

www.iste.co.uk www.wiley.com

© ISTE Ltd, 2009

The rights of Jacky Mazars and Alain Millard to be identified as the authors of this work have been
asserted by them in accordance with the Copyright, Designs and Patents Act 1988.

Library of Congress Cataloging-in-Publication Data

Comportement dynamique des bétons et génie parasismique. English

Dynamic behavior of concrete and seismic engineering / edited by Jacky Mazars, Alain Millard.
p. cm.
Includes bibliographical references and index.
ISBN 978-1-84821-071-4
1. Earthquake resistant design. 2. Structural dynamics. 3. Concrete--Plastic properties. 4. Buildings,
Reinforced concrete--Earthquake effects. I. Mazars, Jacky. II. Millard, Alain.
TA658.44.C653 2009
624.1'762--dc22
2009004399

British Library Cataloguing-in-Publication Data

A CIP record for this book is available from the British Library
ISBN: 978-1-84821-071-4

Printed and bound in Great Britain by CPI/Antony Rowe, Chippenham and Eastbourne.
 Table of Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Chapter 1. Dynamic Behavior of Concrete: Experimental Aspects . . . . . 1
François TOUTLEMONDE and Gérard GARY
1.1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1. Meaning of the word “dynamic” . . . . . . . . . . . . . . . . . . . . . 1
1.1.2. Reminders about dynamic experimentation . . . . . . . . . . . . . . 3
1.1.3. Identifying the behavior of concrete under fast dynamic loadings . 8
1.2. Tests in which the transient rate has little influence . . . . . . . . . . . . 10
1.2.1. Tests involving deviatoric behavior . . . . . . . . . . . . . . . . . . . 11
1.2.2. Tests with prevailing spherical behavior . . . . . . . . . . . . . . . . 17
1.3. Tests with transient phase conditioned interpretations . . . . . . . . . . 20
1.3.1. Tests involving mainly traction behavior . . . . . . . . . . . . . . . . 20
1.3.2. Tests implementing compression behavior . . . . . . . . . . . . . . . 25
1.4. Other tests. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.4.1. Tests adaptable to an energetic approach . . . . . . . . . . . . . . . . 29
1.4.2. Validation tests on structures requiring an inverse analysis . . . . . 30
1.5. Synthesis of the experimental data on concrete and associated
materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.5.1. Data on cement paste mortar and concrete . . . . . . . . . . . . . . . 33
1.5.2. Data available for reinforced concrete. . . . . . . . . . . . . . . . . . 40
1.5.3. Data about fiber-reinforced concretes . . . . . . . . . . . . . . . . . . 42
1.6. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
1.7. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Chapter 2. Dynamic Behavior of Concrete: Constitutive Models . . . . . . 55
Patrice BAILLY
2.1. Dynamics of concrete structures . . . . . . . . . . . . . . . . . . . . . . . 55
2.1.1. Macroscopic phenomena . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.1.2. Perforation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
vi Dynamic Behavior of Concrete and Seismic Engineering

2.1.3. Ejection of fragments . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

2.1.4. Loading range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.1.5. Loading path. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.2. Fast dynamics applied to concrete . . . . . . . . . . . . . . . . . . . . . . 62
2.2.1. Impacts and waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.2.2. Impact and shock polar curve. . . . . . . . . . . . . . . . . . . . . . . 66
2.2.3. Shock between two solids . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.3. Scabbing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.4. Effect of a shock wave on the structure of materials. . . . . . . . . . . . 69
2.5. Modeling types. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.5.1. Behavior description theoretical frames. . . . . . . . . . . . . . . . . 70
2.5.2. Integrating sensitivity to the strain rate . . . . . . . . . . . . . . . . . 71
2.5.3. Elasto-plasticity and criteria . . . . . . . . . . . . . . . . . . . . . . . 72
2.5.4. Damage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.5.5. Notion of a state law . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.5.6. Location limiter and time sensitivity . . . . . . . . . . . . . . . . . . 75
2.6. Models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.6.1. Elasticity-based model . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.6.2. Models based on the theory of plasticity . . . . . . . . . . . . . . . . 77
2.6.3. Models based on damage mechanics . . . . . . . . . . . . . . . . . . 81
2.6.4. Model coupling damage and plasticity . . . . . . . . . . . . . . . . . 82
2.6.5. Model coupling damage and mechanics of porous media . . . . . . 83
2.6.6. Model deriving from a hydrodynamic approach. . . . . . . . . . . . 84
2.6.7. Endochronic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
2.6.8. Discrete element method . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.7. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
2.7.1. Main features of the models. . . . . . . . . . . . . . . . . . . . . . . . 90
2.7.2. Contribution of distinct elements. . . . . . . . . . . . . . . . . . . . . 91
2.8. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Chapter 3. Seismic Ground Motion . . . . . . . . . . . . . . . . . . . . . . . . . 95
Pierre-Yves BARD
3.1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.2. Measuring seismic motions . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.2.1. Differences between seismological and accelerometer networks . . 96
3.2.2. Accelerometer networks . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.2.3. Accelerometer data banks . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.3. Quantitative characterization of seismic movements . . . . . . . . . . . 98
3.3.1. Time maximum values. . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.3.2. Spectral characterizations . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.3.3. Features of hybrid characterizations . . . . . . . . . . . . . . . . . . . 106
3.3.4. Caveats regarding differential motions . . . . . . . . . . . . . . . . . 107
3.4. Factors affecting seismic motions. . . . . . . . . . . . . . . . . . . . . . . 108
3.4.1. Spectral signature of the seismic source . . . . . . . . . . . . . . . . 109
3.4.2. Effects of propagation in the Earth’s crust . . . . . . . . . . . . . . . 111
 Table of Contents vii

3.4.3. Site effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

3.5. Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.6. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Chapter 4. Soil Behavior: Dynamic Soil-Structure Interactions . . . . . . . 125
Alain PECKER
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.1. Behavior of soils under seismic loading . . . . . . . . . . . . . . . . . . . 126
4.1.1. Influence of the nature of soils on seismic movements . . . . . . . . 126
4.1.2. Experimental description of soil behavior . . . . . . . . . . . . . . . 128
4.2. Modeling soil behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.2.1. An experimental description of elastic soil behavior (J d Js) . . . . 134
4.2.2. Linear visco-elastic models for medium strain domains
where Js d J d Jv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.2.3. High strain domain non-linear models where J t Jv . . . . . . . . . . 141
4.3. Linear soil-structure interactions . . . . . . . . . . . . . . . . . . . . . . . 143
4.3.1. Illustration of the soil-structure interaction effect . . . . . . . . . . . 143
4.3.2. Expression of a soil-structure problem . . . . . . . . . . . . . . . . . 148
4.3.3. Superposition theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.3.4. Practical modeling of the soil-structure interaction . . . . . . . . . . 154
4.4. Non-linear soil-structure interactions. . . . . . . . . . . . . . . . . . . . . 158
4.4.1. Geometric non-linearities and uplift of the foundations . . . . . . . 158
4.4.2. Non-linearities of behavior . . . . . . . . . . . . . . . . . . . . . . . . 158
4.4.3. Modeling the non-linear soil-structure interaction . . . . . . . . . . 159
4.5. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
Chapter 5. Experimental Methods in Earthquake Engineering . . . . . . . 165
Alain MILLARD, Pierre PEGON and Jean-Claude QUEVAL
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.1. The pseudo-dynamic method . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.1.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.1.2. History of the PSD method . . . . . . . . . . . . . . . . . . . . . . . . 168
5.1.3. The ELSA laboratory . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.1.4. Comparison with shaking tables . . . . . . . . . . . . . . . . . . . . . 170
5.2. The conventional pseudo-dynamic method . . . . . . . . . . . . . . . . . 170
5.2.1. Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
5.2.2. Implementation at ELSA . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.2.3. The sub-structuration method. . . . . . . . . . . . . . . . . . . . . . . 174
5.2.4. Illustration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.3. Continuous pseudo-dynamic method. . . . . . . . . . . . . . . . . . . . . 178
5.3.1. Continuous method principle . . . . . . . . . . . . . . . . . . . . . . . 178
5.3.2. Implementation at ELSA . . . . . . . . . . . . . . . . . . . . . . . . . 180
5.3.3. Sub-structuration for the continuous method. . . . . . . . . . . . . . 182
5.4. Final comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
viii Dynamic Behavior of Concrete and Seismic Engineering

5.5. Shaking table tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

5.5.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.5.2. Characteristics and performance of shaking tables . . . . . . . . . . 187
5.6. Laws of similarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
5.7. Instrumentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.8. Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
5.9. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
5.10. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Chapter 6. Experiments on Large Structures . . . . . . . . . . . . . . . . . . . 201
Patrick PAULTRE and Jean PROULX
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
6.1. Instrumentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
6.2. Dynamic loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
6.3. Data processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
6.4. Application to buildings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
6.4.1. The slanting tower at the Montreal Olympic Stadium . . . . . . . . 209
6.4.2. Reinforced concrete building . . . . . . . . . . . . . . . . . . . . . . . 210
6.5. Bridge application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
6.5.1. Pedestrian footbridge. . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
6.5.2. A mixed cable-stayed/suspension bridge . . . . . . . . . . . . . . . . 215
6.6. Application to large dams . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
6.6.1. Assessment of a response spectrum on the crown . . . . . . . . . . . 220
6.6.2. Study of foundation-ice-reservoir-dam interactions. . . . . . . . . . 222
6.6.3. Study of the effect of the water level inside the reservoirs. . . . . . 228
6.7. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
6.8. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
6.9. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
Chapter 7. Models for Simulating the Seismic Response of
Concrete Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Didier COMBESCURE, Nicolas ILE, Jacky MAZARS and Jean-Marie
REYNOUARD
7.1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
7.2. Different discretization families. . . . . . . . . . . . . . . . . . . . . . . . 234
7.2.1. Global modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
7.2.2. Semi-global modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
7.2.3. 2D and 3D fine models . . . . . . . . . . . . . . . . . . . . . . . . . . 239
7.3. Behavior laws for concrete. . . . . . . . . . . . . . . . . . . . . . . . . . . 240
7.3.1. Semi-empirical mixed models . . . . . . . . . . . . . . . . . . . . . . 240
7.3.2. Damage model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
7.3.3. Plasticity model for concrete . . . . . . . . . . . . . . . . . . . . . . . 243
7.3.4. Cyclic models for steel. . . . . . . . . . . . . . . . . . . . . . . . . . . 247
7.3.5. Taking construction layouts and second-order phenomena
into account . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
 Table of Contents ix

7.4. A few examples with their validation through experiments . . . . . . . 250

7.4.1. Application of the semi-global method to a four-storey
structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
7.4.2. Semi-global and local models applied to concrete walls . . . . . . . 253
7.5. Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
7.6. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
Chapter 8. Seismic Analysis of Structures: Improvements Due to
Probabilistic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
Jean-René GIBERT
8.1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
8.2. The modal method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
8.2.1. Data about the seismic source . . . . . . . . . . . . . . . . . . . . . . 274
8.2.2. Calculation of structural responses using the modal method . . . . 277
8.3. Criticism of the modal method . . . . . . . . . . . . . . . . . . . . . . . . 279
8.4. A few reminders about random processes . . . . . . . . . . . . . . . . . . 280
8.4.1. Definition and characterization of a time random process . . . . . . 280
8.4.2. Second order characterization . . . . . . . . . . . . . . . . . . . . . . 280
8.4.3. Response of a linear system to random stress . . . . . . . . . . . . . 282
8.4.4. Using stochastic equations . . . . . . . . . . . . . . . . . . . . . . . . 285
8.4.5. Extrema statistics in a stationary process . . . . . . . . . . . . . . . . 286
8.5. Improvements to the modal method . . . . . . . . . . . . . . . . . . . . . 292
8.5.1. Complete quadratic combination. . . . . . . . . . . . . . . . . . . . . 294
8.5.2. Peak factor effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
8.6. Direct calculation of the floor spectra . . . . . . . . . . . . . . . . . . . . 297
8.6.1. Representation of non-stationary processes . . . . . . . . . . . . . . 297
8.6.2. Adjusting a separable process from the ORS data. . . . . . . . . . . 298
8.6.3. Determination of the floor spectra . . . . . . . . . . . . . . . . . . . . 299
8.7. Creation of synthetic signals and direct numerical integration . . . . . . 301
8.8. Seismic analysis of non-linear behavior structures. . . . . . . . . . . . . 304
8.8.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
8.8.2. Main non-linearities of seismically-loaded structures . . . . . . . . 304
8.8.3. Notion of “inelastic spectra” . . . . . . . . . . . . . . . . . . . . . . . 305
8.8.4. Conventional method of stochastic linearization . . . . . . . . . . . 310
8.8.5. Random parameter stochastic linearization. . . . . . . . . . . . . . . 316
8.9. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
8.10. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
Chapter 9. Engineering Know-How: Lessons from Earthquakes
and Rules for Seismic Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
Philippe BISCH
9.1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
9.2. Lessons from earthquakes . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
9.2.1. Pathologies linked to overall behavior . . . . . . . . . . . . . . . . . 328
9.2.2. Problems linked to local under-design . . . . . . . . . . . . . . . . . 330
x Dynamic Behavior of Concrete and Seismic Engineering

9.2.3. Problems linked to construction layout . . . . . . . . . . . . . . . . . 334

9.3. The aims of anti-seismic protection standards . . . . . . . . . . . . . . . 336
9.3.1. Standardization of anti-seismic design . . . . . . . . . . . . . . . . . 336
9.3.2. Main objectives of anti-seismic protection . . . . . . . . . . . . . . . 337
9.3.3. Verification method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
9.3.4. Capacity-design method . . . . . . . . . . . . . . . . . . . . . . . . . . 341
9.4. General design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
9.4.1. Design principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
9.4.2. Regularity conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
9.4.3. Calculation of seismic action effects . . . . . . . . . . . . . . . . . . 346
9.5. Behavior coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
9.5.1. Using behavior coefficients . . . . . . . . . . . . . . . . . . . . . . . . 349
9.5.2. Structure behavior and behavior coefficients. . . . . . . . . . . . . . 351
9.5.3. Local ductility and behavior coefficients . . . . . . . . . . . . . . . . 352
9.5.4. Ductility classes and behavior coefficients . . . . . . . . . . . . . . . 352
9.6. Designing and dimensioning reinforced concrete structure elements . . 353
9.6.1. Regulations specific to reinforced concrete in seismic areas . . . . 353
9.6.2. Main types of reinforced concrete bracing . . . . . . . . . . . . . . . 354
9.6.3. Main frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
9.6.4. Reinforced concrete bracing walls . . . . . . . . . . . . . . . . . . . . 359
9.6.5. Detail designing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
9.7. Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
9.8. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
List of Authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
 Preface

The static and quasi-static behaviors of concrete have been the subject of so
many works that we often consider that they are quite well known and mastered as
far as modeling with a view to structure calculations is concerned. However, the
same is not true of concrete’s dynamic behavior, because of the complexity of
the tests needed to reach pertinent loading rates.

The subject matter of Chapter 1 is divided into two parts: it presents the most
widely used experimental techniques to study the dynamic behavior of concrete,
drawing attention to the difficulties in interpreting the results of tests designed to
identify its intrinsic parameters. It also offers a synthesis of properties that have been
published in the literature dealing with concrete (chiefly its traction and simple
compression strengths), as well as values for reinforced or fiber-reinforced
composites. An extensive bibliography enables the reader to refer to the relevant
original articles.

Dynamic loadings can generate non-linearities and a range of deteriorations in

concrete (failure from bending and/or shear, traction, mechanical spalling, tearing,
compression, compaction and hole perforation, etc.), all of which have to be
carefully modeled to enable prediction of the behavior of a specific structure under a
violent action. The variety of responses has generated several unique modeling
approaches. Depending on the phenomenon under consideration, we use either the
damage approach for cracking, the plasticity or viscoplasticity approach for shear,
or the still volume-pressure influence approach for compaction. The theoretical
contexts are discussed in Chapter 2, before the essential elements of several
“conventional” models are described, along with their strengths and weaknesses.

In Chapter 3, the subject matter turns to the particular category of dynamic

oscillations associated with earthquakes. As an introduction, Chapter 3 deals with
the way seismic movement measurements – which generate the data used for
xii Dynamic Behavior of Concrete and Seismic Engineering

structure reaction calculations – are made. Besides presenting the addresses of

databases of signals measured in different countries, this chapter also introduces the
concept of spectral representation, which plays a key role in engineering practice. A
geophysical interpretation of seismic movements in connection with subjacent
phenomena is proposed, which integrates the contributory effects of the site and the
topography of the environment around the structure.

Though typical practice involves calculating the reaction of a structure submitted

to an earthquake by considering its base to be totally embedded, the nature of some
soils, coupled with the exceptional character of some structures (like dams and
nuclear reactors), demands that the behavior of the structure is modeled in a
particular environment. This problem is called structure-soil interaction, and forms
the subject of Chapter 4. To solve this problem, it is necessary to have a model of
the soil’s behavior under cyclic loading. Different models exist, depending on the
nature and amplitude of the loading. After modeling, the interaction problem can
either be treated by superposition, by considering the soil and the structure
separately for linear cases, or globally for non-linear situations.

The difficulty of conducting structure tests on full-sized models led to the

development of experimental methods employing scale models. Vibrating tables,
which reproduce earthquakes on a small scale, were designed for this purpose. The
subsequent development of fast and powerful computers gave birth to the pseudo-
dynamic method, in which the purely dynamic effects of an earthquake are
simulated using calculations. These complementary techniques both have their own
advantages and disadvantages. The quality of the results they can produce depends
mainly on the quality of model implementations, which are described in detail in
Chapter 5.

Chapter 6 is concerned with experimental techniques on large structures.

Experiments play an essential role in obtaining realistic data about a structure’s
dynamic signature; mechanically-controlled vibration tests are not easy to
implement, but they are a crucial source of information. This chapter shows how an
excitation with rotating masses, coupled to adapted instrumentation and
measurement processing, gives access to a vast amount of key information
concerning natural modes, frequency damping, damage indications and coupling
effects between the structure and its environment. This forms an invaluable database
that model-makers need to calibrate their models, which is a pre-requisite of any
realistic analyses of the seismic response of an existing structure.

Chapter 7 examines the structure-modeling field as applied to the seismic

analysis of concrete buildings. The chapter focuses on approaches that allow
engineers to simulate reactions to the application of an earthquake by exploring the
non-linear field and collapse modes.
 Preface xiii

In this context, three model families are considered: global, semi-local and local
models. The first rely on empirical behavior descriptions, gathering phenomena at
the level of a single section or structural element with occasional brief discretization.
The second type of model works out global laws from phenomenological local
models, with discretization made at the multi-fiber or multi-layer beam level. The
third type of model is more sophisticated and takes the responses of a building’s
constituent materials and their bindings into account. These demand very thorough
discretization, their main disadvantage being that they are very time-consuming to
implement.

Validation analyses are proposed based on experiments carried out on shaking

tables or reaction walls. The results show that that modeling has reached such a
sophisticated stage of development that it allows complete experimental and
experiment-feedback analyses, and is therefore ready for transfer to everyday
engineering.

Chapter 8 introduces a quite promising analysis procedure: probability analysis.

It is clear that the uncertain nature of seismic loading must be taken into account for
the dimensioning of large structures. However, though sophisticated methods are
used (seismic movement correlation, structure-soil interaction, behavioral non-
linearity, etc.), the models remain “deterministic”.

This chapter shows that the determinist approach can sometimes lead to
erroneous predictions, and that better control of phenomena makes it necessary to
take into account the probabilistic character of the problem. Probabilistic seismic
analysis is a new field of research that should lead to significant advances in
paraseismic engineering.

Chapter 9 considers the craft aspect of engineering in the field of seismic

building analysis. The subject of this chapter, experiment feedback and regulations,
is important, as engineers are ultimately responsible for the safety of people as well
as buildings, despite the fact that building science is not a totally exact science.
Numerous theories can help analyze the behavior of structures, but the limits of the
problems faced by engineers in that field remain “fuzzy” (action characterization,
structure complexity, local behavior facts, soil-structure interaction, etc.).
Regulations can provide a framework that makes the analysis concepts reliable.
Experiment feedback gives an indication of those approaches that have worked and
those that have not, and considering this when developing regulations obviously
assists progress in safety control. This is the subject of this chapter, which also gives
an excellent account of the spirit in which the paraseismic design of various concrete
structures should be approached.

Jacky MAZARS
Alain MILLARD
This page intentionally left blank
 Chapter 1

Dynamic Behavior of Concrete:

Experimental Aspects

1.1. Introduction

1.1.1. Meaning of the word “dynamic”

As distinct from the term “static”, “dynamic” implies the influence of time. A
test is said to be “quasi-static” when the effects of time are present, but can be
neglected. For a structure test, and for any real test, the effects of time are typically
expressed in two ways:
– by forces of inertia resulting from the not equal to zero acceleration to which
the elements of structures are submitted;
– by the behavior of each elementary volume of the material depending on the
evolution in time of the elementary mechanical values (stress and strain) and
possibly of their time derivatives. This dependence is described by the generic name
of “viscosity”.

This distinction is strictly linked to the notion of elementary volume underlying

the definition of behavior. Actually, the fact that viscosity effects might be the
manifestation of inertial microscopic phenomena cannot be excluded. This remark is
important in the case of concrete, as considerations about material homogenity

Chapter written by François TOUTLEMONDE and Gérard GARY.

2 Dynamic Behavior of Concrete and Seismic Engineering

involve decimeter elementary volumes (but this is not the case with metals for which
the elementary volume is sub-millimetric).

Thus, to be quite clear, we will consider the dynamic behavior aspect as limited
to the description of the effects of time using elementary mechanical values and
excluding inertia effects.

From general physical and thermodynamic considerations concerning behavior

laws [MAN 67], we can deduce that the generalized mechanical variables Q (t)
(stress) and q(t) (strain) can be related in the following way:

t
Q (t ) G ª¬ H f ( q (W )); q (t ), q (t ), "º¼ [1.1]

where H describes the loading history. This formulation highlights the fact that these
values do not play a symmetric role. The instantaneous mechanical reaction depends
on the geometric history, its current value, and the values of its higher time
derivatives. Thus, it is not natural to consider stress velocity as a behavior variable.

If we limit our attention to formulations likely to be easily integrated into

calculation codes, the relation expressed in equation [1.1] can be re-written in the
following incremental form:

dV f ( d H , H , H.., D i ..) [1.2]

The values D i are internal parameters that take process history into account.
Their evolution has to be described as a complement to the relationship in equation
[1.2]. Their dependence on the history of the process explicitly results in their
loading and unloading paths being different. The values playing a part in equation
[1.2] are tensors. We can see the complexity of this relation. In most cases, the
simplifications carried out involve discarding strain time derivatives higher than 1,
and expressing the strain speed using a scalar value. Such simplifying assumptions
are justified for two types of reasons. Firstly, programming laws into codes will be
simplified by doing this. Secondly, an insufficient variety of dynamic tests is
available to identify more parameters. For this reason, from this point onwards, we
will refer to “strain velocity” without going back over the definition.

As far as strain velocity is concerned, it is standard practice to study its effects

on long time scales revealed through creep. Even though creep tests can clarify the
analysis of dynamic tests, we will not be considering them. The experimental
aspects of creep tests have no dynamic aspects, as typical strain velocities
 Dynamic Behavior of Concrete: Experimental Aspects 3

implemented are around 10-10 s-1, as compared to “static” test standard velocities that
range from 10-6 to 10-5 s-1, and the strain velocities reached during “hard” shocks on
civil engineering structures, which are usually in the range of 0.1 and 10 s-1.

These elementary considerations understood, it appears that a critical factor in

the experimental characterization of concrete behavior is discarding the inertia
terms. The problem is more delicate with concrete (a brittle material) than it is with
metals. As a matter of fact, the first manifestation of inertial effects on a sample
submitted to dynamic loading is the transient response observed when the time taken
by elastic waves to pass through the sample (the transfer time) is significant relative
to the test’s time duration. When studying this problem, the pertinent time-
dependent parameter is not the strain velocity (which, in any case, is not well-
defined in the transient phase), but the loading time relative to the transfer time. If
sufficient strain levels are reached in very short periods of time, the sample could
fail before a homogenous stress and strain state, measurable as an average, could be
reached. In fact, low amplitude traction strains (ranging from 100 to 150 x 10-6) lead
to material failure. Test analysis is generally difficult. For common sized samples
(centimeter scale), we cannot go beyond 1/s average strain velocities when
conducting a quasi-static test analysis. This feature of brittle materials can be
exploited advantageously, and is used in scabbing tests (see section 1.3.1).

This limitation is far less a problem with metals, where important local strains
arise, but do not cause failure. Such a situation can only occur in concrete if
particular conditions that guarantee mechanical field homogenity exist to prevent
cracking. This is the case when tests are conducted in strong confinement (under
which circumstances, concrete behavior is described by plasticity-type models). As
far as metals and most polymers are concerned, it is also important to take thermo-
mechanical coupling into account, due to the adiabatic feature of dynamic tests. This
effect can only be neglected when failure occurs under low strain for which the
dissipated heat remains low: with concrete, it can also be neglected in confinement
tests, since we can presuppose a low thermo-mechanical coupling.

1.1.2. Reminders about dynamic experimentation

1.1.2.1. Specificity of dynamic tests

As far as statics and dynamics are concerned, it is reasonable to consider sample
analysis in a separate section, along with the overall measures it involves (generally
carried out on the peripheral part of the material). This is the second aspect
mentioned in the introduction.
4 Dynamic Behavior of Concrete and Seismic Engineering

The first difficulties encountered in dynamic experimentation fall under the first
category mentioned in the introduction. They are linked to transient effects inside
the machine and the associated sensors: the balancing time of the machine and its
sensor array (elastic waves moving back and forth several times) are not negligible
relative to the length of the test. Thus, carrying out quality measures often requires a
transient analysis of the response of the machine itself. Hence, in a real situation,
characteristic testing times have to be compared with the acquisition chain and the
sensor pass-band. If the acquisition frequency is not far higher than the frequency of
transient signals, the observed result can be completely modified by the measuring
chain, and even average values can be wrong.

1.1.2.2. Hopkinson bar test

For average strain rates in excess of 50/s, because the transient effect inside the
test machine cannot be neglected, a way round the problem involves explicitly
taking wave propagation phenomena into account, using a bar system. Whilst the
transient analysis of three-dimensional structures is too complex to be taken into
account efficiently, using “one-dimensional” bars makes it possible, as we will now
explain.

1.1.2.2.1. A description of the bar test

To carry out a dynamic compression tests with Hopkinson bars [HOP 14] (also
called the SHPB (Split Hopkinson Pressure Bar) system, or Kolsky bars [KOL 49]:
named after the first person to use the system in its current configuration), a small
sample is placed between two identical long bars with a high elastic limit relative to
the tested material (Figure 1.1). Strain gages are glued to both bars. Due to a
projectile, a compression longitudinal elastic wave is induced into the input bar. Part
of this gets reflected at the sample-bar interface, whilst another part is transmitted to
the sample before inducing a wave in the output bar.

Figure 1.1. Hopkinson bar assembly

The waves at points A and B are determined by measuring and recording the
structurally-associated longitudinal strains. The need to know A, the incident wave
induced by the impact separator, and the reflected wave B, which depends on the
 Dynamic Behavior of Concrete: Experimental Aspects 5

reaction of the sample, arises because we need to find the optimal position of the
measuring point at the middle of the bar. On the other hand, considering the bar as
one-dimensional does not allow us to place the strain gauge too near an end. A
typical recording for a concrete sample compression test is shown in Figure 1.2.

Figure 1.2. Compression test on a concrete sample (40 mm diameter

aluminum bars): basic waves

Next, the waves have to be carried to the contacts between the sample and the
bar. Then we can calculate the stresses and displacements (by integrating the
velocities, which are directly accessible) on the corresponding faces.

The particulate velocities at the input and output faces can be written
respectively as:

Ve (t ) c ª¬H i (t )  H r (t ) º¼
[1.3]
Vs (t ) cH t (t )
6 Dynamic Behavior of Concrete and Seismic Engineering

The forces on the input and output faces are respectively:

Fe (t ) Sb E ª¬H i (t )  H r (t ) º¼
[1.4]
Fs (t ) Sb E H t ( t )

Measures on the two opposite faces of the sample allow estimation of strain field
homogenity by comparing the forces on each face (section 1.3.2, Figure 1.12). We
note that for this test, the assumption of homogenity in mechanical fields is
hazardous. As a consequence, the notion of average strain velocity is also hazardous.
In section 1.3.2 we will see the best way to use the available measurements. Thus,
we should stress that the Hopkinson Bar leads to overall values of loads and
displacements on both sides of the sample. All mechanical quantities are obtained by
making additional assumptions completely separate from the test facilities. These
have been widely reported in the literature [NIC 80].

1.1.2.2.2. Limitations of the conventional system

Accurate analysis of wave transport
To carry out a precise virtual wave transport between the measured points and
the sample (forward transport for the incident wave and backward transport for the
others), the three-dimensional feature of the bars need to be considered, and the
dispersal correction must be introduced using a signal treatment technique. This
parameter corresponds to signal modification during transport. An accurate time
calibration (to within a micro-second) is also necessary [ZHA 96]; it is especially
important for measurement of small strains, and thus for brittle materials such as
concrete.

Multi-axial characteristics of the test

The uniaxial characteristic of the test is also an approximation. Let us examine
this aspect in the case of compression. Whenever the material presents a Poisson
effect, the longitudinal strain comes with a lateral strain (as is the case in statics if
the support conditions are well controlled), which is opposed by radial inertial
effects. This causes an induced confinement. The confinement explains the obvious
sensitivity of concrete to strain velocity that is universally observed in dynamic
compression (see section 1.3.2).

Measurement duration
The proportionality between the mechanical values associated with a wave inside
a bar, on which the Hopkinson bar technique is based [1.3]-[1.4], only applies to a
 Dynamic Behavior of Concrete: Experimental Aspects 7

wave propagating in a single direction, which requires measurement of the incident

wave (propagating one way) separately from the reflected wave (which propagates
in the other direction). This limits the measuring duration to 'T  'T L / C,
C being the propagation velocity and L the length of the input bar. 'T is thus a
function of the length of the bars. Consequently, for a behavior test, the total strain
cannot exceed the product of the average strain velocity and 'T . For instance,
measuring duration will not exceed 400 μs (C| 5,000 m/s) for a 2 m long aluminum
bar, and the total strain will be limited to 4% for a test with a 100 s-1 average strain
velocity. Because of this limitation, even with concrete (for which high strains are
unlikely to be reached), the conventional Hopkinson bar system will not allow tests
at average strain velocities lower than 50 s-1. On the other hand, for reasons
explained in section 1.2.1, traditional machines used without specific precautions do
not give reliable results at lower velocities. Besides, their superior limit is not clearly
established and is determined to an extent by the material being tested (the test
piece). The machine must be used in a particular way; it varies between 1 s-1 and
about 10 s-1. However, a recent experimental technique using bars [BUS 02] that
covers this problem now exists.

1.1.2.2.3. Difficulties inherent to dynamic measurements

The dynamic test facilities have numerous limitations, especially for stresses
other than simple compression or small strains. This limitation mostly affects low
strength stressed materials (impedance adaptation and high strain problems) and
brittle materials (low strain at failure).

The Hopkinson bar example illustrates the generic difficulties quite well. The
very short loading times do not enable us to carry out multi-axial dynamic loadings
easily, and it is not easy to synchronize loading with two (or three) orthogonal
Hopkinson bars. If synchronization is tricky in dynamics, it is all the more so when
piloting the test. Therefore, we cannot (for now) contemplate carrying out tests
under controlled multi-axial loading (deviatoric, for example), as is required in a
quasi-static mode. The need to control the loading and the difficulty in carrying out
dynamic displacement measurements limits the potential tests to a very small
number, which are described, along with their specific problems, in sections 1.2 and
1.3.

1.1.2.2.4. Compression tests with confinement

It is quite easy to superimpose quasi-static confinement on a dynamic
compression test. A cell in which a gas pressure confinement can be maintained
during the compression test is described in [GAR 99]. Some authors have proposed
a bi-axial loading scheme, where the secondary static stress is applied using a jack
[WEE 88]. For higher confinements (necessary if we want to study compaction of
8 Dynamic Behavior of Concrete and Seismic Engineering

concrete, for example), a metal cylinder can be used [GAR 99]. In this case,
confinement pressure is not studied, but can be measured during the test by
assuming the (most often elastic) response of the confinement ring is known (as in
an oedometric test). Another way to carry out high confinements involves using the
“plate on plate” test developed to study the high-speed spherical behavior of metals.
It is a plane strain-loading test, the inverse analysis of which is based on behavior
modeling. High confinement there is associated with very high strain speeds.

1.1.2.2.5. Traction tests

A conventional traction test can be carried out with a Hopkinson bar [REI 86]. If
we consider only global measures, the main difficulty is due to keeping the sample
in contact with the bars. To avoid having to resort to assemblies leading to
impedance failures, it is reasonable to glue the sample to the bars. Some authors
[TED 93] have had the idea of using the Brazilian test again in dynamics. In this
case we have to check that the conditions of strain homogenity are compatible with
the assumptions. Finally, the spalling test [DIA 97] allows an accurate measurement
of the average stress just before failure, but its interpretation is difficult as it is
between the classical traction test and the fracture test (toughness measurement).

1.1.3. Identifying the behavior of concrete under fast dynamic loadings

When identifying the dynamic behavior of concrete, we are confronted with a

series of typical problems for each high-speed behavior identification test. Some of
these problems are increased by the nature of concrete, which is the reason why we
prompt the reader to be very cautious when using experimentation signals or results.

Due to its structure in aggregates, where it is mixed with sand and hardened
cement paste, concrete can be a highly heterogenous material, and the size of a
representative sample is not always an easy thing to state. As far as statics is
concerned, a 2 slenderness cylinder, over five times as big in diameter as the
aggregates, is the lowest volume necessary to obtain stable properties representing
the material in these tests, particularly as far as strength is concerned, otherwise
“scale effects” will be observed. Such a constraint raises several types of problems:
– for standard concretes, in which the maximum size of aggregates ranges from
20 to 25 mm, the dimensions of test samples (diameter over 10 or 12 cm, mass over
5 kg) involve resorting to important energies, particularly for high speed tests, which
involves sometimes tricky technological arrangements;
– to avoid this difficulty, tests are often carried out on micro-concrete, mortar or
cement paste samples. Transposing these results to structure concrete requires a
critical analysis, mainly because the volume fraction of cement paste (generally
 Dynamic Behavior of Concrete: Experimental Aspects 9

considered as the viscous element of the composite) is not always constant. In the
same way, the propagation of waves disrupted by the module differs between
cement pastes, and aggregates can also be different depending on the composition of
the studied material;
– even if we managed to identify the intrinsic properties of the material on big
enough samples, for many structures, the “representative material point” size is
important compared with the dimensions of the smallest pieces (building shells
about 20 cm, bridge webs from 30 to 45 cm). Furthermore, significant stress
variations on the scale of the structure can be discerned over short distances of the
same order of magnitude as the dimensions of the test sample. What is then involved
is the application of continuous medium theory, which is based on the assumption –
generally not well verified – that the material point is infinitely small compared to
the structure;
– a problem (which occurs in statics too) that becomes crucial as far as the
dynamic interpretation of tests is concerned is that the sample is not submitted to an
homogenous state of strain and stress owing to its size, and has to be considered as a
structure submitted to transient loading.

Because concrete is a brittle material (like most geomaterials, concrete can only
withstand very weak extension strains and its apparent “failure” takes place for a
compression strength about 10 to 20 times as strong as its traction strength), most of
the time, in practice, while interpreting the tests, we must consider:
– that we are dealing with an elastic homogenous material (which implies the
size precautions referred to above): the assumption is necessary for relatively low
velocities or low strain levels, in continuity with the quasi-static field. It is not good
enough to interpret the totality of a test when the speed increases, since the
maximum stress is reached when localized cracking has been reached significantly
on only a part of the structure;
– that beyond the stage corresponding to localized cracking, the test sample can
be modeled as a cracked structure where damage concentrates in the crack area,
which corresponds to fracture models;

– that beyond a stage corresponding to a distributed deterioration (which

corresponds to the bonding material crumbling away), the material can be described
by combining damage and plasticity models.

Hence, at the material failure of the sample, the interpretation of the tests
requires different analysis models, regardless of whether we are mainly in a
deviatoric behavior with a possible extension direction allowing localized cracking,
or in a mainly “spherical” behavior, and depending on the stress peak being
identifiable or not.
10 Dynamic Behavior of Concrete and Seismic Engineering

It is important to note that because of the weak growths withstood by the

material at high velocities, experimental precautions have to be taken – especially
because of transient effects, in dynamic experiments where limit conditions are
difficult to control. In experiments where an “energetic” approach is privileged, this
aspect is also important: the inertia of the test sample cannot always be neglected
with regard to that of the test machine, and the energy dissipation through damage
on the support, or through contact with the impact separator can be important
compared with the energy supposedly dissipated by the “normal” cracking expected
in bending.

Finally, a delicate feature of concrete is its porosity: it has such a tortuous

network that water exchange times with the environment are quite long (about 10
years for the representative volumes considered above). We can consider the
hydration state of the sample as constant during dynamic tests, which is not the case
for shrinkage or plastic flow tests. However, important relative pore moisture and
mechanical state coupling, together with frequent cracking due to the stress levels
reached when desiccated, begins at the sample’s surface and/or their environment as
soon as they are fabricated. In at least one stress and velocity field ([DAR 95] [TOU
95a]), researchers have shown that the partly water-saturated feature of the porous
network explains the modification of apparent mechanical properties: these are
generally called “velocity effects” in the literature.

In following sections (1.2 to 1.4), we will detail the arrangements, test type by
test type, used to analyze the results and infer the indications and modifications
required to calculate and understand the behavior of fast dynamic loading concrete
structures. The actual and measured behaviors are summarized in a rational way in
section 1.5.

1.2. Tests in which the transient rate has little influence

In this chapter, we will deal with behavior identification tests that, for reasons
developed in section 1.1 can have a “quasi-static” interpretation.

Two test families can be distinguished. The first is derived from typical concrete
characterization tests and emphasizes growth or cracking failures. This is called
deviatoric behavior, and is the failure kind that is also, indirectly, the cause of
collapse observed in compression and even in biaxial compression. The second test
type corresponds to “volumic” behavior, which can seldom be observed in ordinary
structures, except in relatively confined areas where specific reinforcement by the
surrounding material ensures tri-axial confinement at high velocity: concrete areas
directly submitted to impact and those close to an explosive charge or perforating
projectile are examples.
 Dynamic Behavior of Concrete: Experimental Aspects 11

Combining both types of information in statics enables a definition of failure or

plasticity criteria closed on the hydrostatic compression axle, as opposed to the
“intrinsic curves” (Coulomb criterion), the validity of which is preferentially
ensured when an extension direction is possible.

1.2.1. Tests involving deviatoric behavior

1.2.1.1. High-speed press machines and traction tests

Because of the difficulties connected with carrying out dynamic tests, most
authors use privileged uniaxial tests. Owing to the basic feature, traction behavior
identification stands out, and has given rise to a great number of tests. In order to
ensure continuity in the geometry of test samples, by controlling the loading
application speed and considering its limited artifacts, a direct traction test on a
cylindrical specimen has become essential. This is detailed in [HOR 87], [REI 82]
and [TOU 95a]. With particular precautions, this test can actually be carried out on
conventional servo-controlled machines with load build-up speeds ranging from
about 0.05 MPa/s (which is the standard loading rate for standard identification
tests) to about 10,000 times this load, with identification at still higher speeds of the
order of 50 GPa/s possible on the same specimen type thanks to the modified
Hopkinson bar (SHB).

The necessary precautions particularly involve:

– choosing to glue the specimen in place with centering and a rigid (without
spherical pairs) mounting onto the press to limit looseness which is a source of
interfering moments;
– choosing aluminum hard supports to limit the transversal strain divergence at
both ends of the specimen;
– controlling the hydration state of the specimen [TOU 95a];
– choosing a not too important slenderness ratio (1 to 1.5) to limit potential
bending;
– gauge extensometer or extensometers fixed in the middle of the specimen to
avoid the deformations due to the glue joint;
– using specimens with adequately sized diameters considering the maximum
size of the aggregates, and if possible core cylinders for better homogenity of the
material and to avoid scaling effects [ROS 92a].

With a sufficient automatic control and oil flow unit, and potentially using a
preload to carry out high velocity tests, we can consider that the load build-up speed
12 Dynamic Behavior of Concrete and Seismic Engineering

is rather constant during the test. The propagation speed of the waves within
concrete – about 4,000 m/s, the standard size of specimens (10 cm) – and the
traction failure stress (4 to 10 MPa) limit the quasi-static interpretation of this kind
of test, results typically showing a divergence about 10% between the specimen’s
input stress and output stress.

The measurements typically carried out during this test are of the applied force
as a function of time, and of the average longitudinal strain at the center of the
specimen (extensometer gages or sensors for which we have to check that the inertia
will stay weak and the fixing will be ensured during the test). Taking into account
the small size and fixedness of the assembly, we can consider that there are no
differences between the measured force and the force applied to the specimen, so we
can assess traction uniaxial behavior by eliminating time. In such a test, the
specimen behavior corresponds quite well to brittle elastic behavior up to localized
cracking. Localization brings about loss of the homogenity of the strains, and an
almost instantaneous decrease of the load.

Going through these tests, which implies expressing the maximum measured
stress according to the “load build-up” parameter in a logarithmic diagram, typically
allows us to define a traction rate effect corresponding to the strength relative
increase.

1.2.1.2. High-speed press machines and compression tests

The second most conventional test that can be performed at high speed is the
compression test. It enables us to define a compression “rate effect” from the
measurement of the maximum strain reached [BIS 91]. The size of the test sample
necessary to free oneself from the size effect and to ensure the correct strain level
reached lead to strict constraints on press dimensions, unit power and the jack flow
rate. For this reason, a great number of the tests described in the literature were
carried out on mortar, cement pastes or micro-concrete [HAR 90]. As is the case in
traction, it has proved possible to look for a size compatible with the higher speed
test performed with Hopkinson bars [DAR 95].

As it is difficult to stop the jack when its speed has been stabilized, few test
reports have included extensometer measurements [BIS 91], measuring the load
obviously remains the main data. For standard size specimens (10 cm), considering
the wave propagation speed and the maximum stress reached, the load build-up rate
beyond which the sample cannot be deemed to be in a stationary process is about 10
times as important as it is in traction tests, which correspond to the strength ratio.
When expressed in terms of strain rate, the threshold is about 10 instead of 1 s-1
[MAL 98].
 Dynamic Behavior of Concrete: Experimental Aspects 13

The incidence of superfluous interference moments is generally less important

than it is in traction tests; however, the precautions to be taken to avoid restricting
transversal strains are as important as in statics, especially for specimens with
slenderness ratios below 2. To this end, we can mention lubricating the faces or
using aluminum. The quality of the stress transmission surfaces is essential to avoid
premature concentration of stresses.

We note that as in statics, and even for specimens that are simply laid, the
relative displacement (including interface crush and deformations at the ends of the
specimens) cannot result in a reliable indication of the strains of concrete in its
standard part, the error typically ranging from 30 to 100% [BOU 99]. As in statics,
failure obtained in compression tests begins with transversal extensions. The traction
rate effect results in an “inertial confinement”. However, the maximum stress is only
reached when the cracks parallel to loading direction meet, allowing either buckling
in the “small columns” formed inside the test body, or shear localization. As a
consequence, interpreting the strength evolution, where load build-up speed is the
only parameter, becomes complex.

We could not find any references to tests deriving from standard quasi-static
identification of multi-axial behavior with prevailing deviatoric behavior (bi-
traction, pure shear, bi-compression), at least not in areas where transient test
characteristic can be neglected. As a matter of fact, the most frequent cases of
dynamic multi-axial behavior identification use unidirectional loading with a
Hopkinson bar [GAR 98], [LOU 94] and [WEE 92], while confinement or loading
in the other direction is often “static”. These tests will be described in section 1.3.
Such a situation can indeed be explained by the difficulty in controlling and
synchronizing dynamic loadings, even at the “low” speeds reached by conventional
presses or jacks. Furthermore, taking the properties of concrete into consideration,
the regulations rarely take multi-axial behavior into account. As a result this lowers
the validation of high velocity dynamic models adapted to concrete, in situations
other than simple traction, uniaxial compression or compaction.

1.2.1.3. Tests with small plates or beams submitted to pressure loading

Considering the difficulty in carrying out dynamic loading with mechanical
application of the loads, some authors perform controlled loadings on mini-
structures (small rectangular plates, beams or small plates), using a pressure loading
generated by an explosion. The purpose is then to identify the bending behavior, the
bend-moment law being material information directly transposable to the calculated
structure, taking into account the similar nature of the tested material and the
geometric and energy similarities – called Hopkinson’s – on the load. Detailed
experimentation of this kind will be described in [BAI 87] and [BAI 88]. The
limitation on the energies that can be used in a laboratory forces the use of
14 Dynamic Behavior of Concrete and Seismic Engineering

centimeter thick test elements, therefore generally of mortar (or possibly fiber-
reinforced) rather than concrete.

The loading process links the level of the applied overpressure with its duration
and load build-up speed. However, interpreting the trial remains simple in so far as
the load build-up times can be considered as very short compared with the specific
period of the structure. Thus, we have what is called a pulse loading: the
overpressure time, which causes the structure to start vibrating, is very slightly
ahead of the latter’s peculiar period, which is then in a free-vibration system. As the
probable area of maximum strain and even failure is known, the relevant section can
be instrumented in a preferential way. Therefore, we can measure the traction by
bending the final strain and the final bend. Note that shortly before failure, the strain
of the compressed side is slightly inferior to that of the opposite side (the start of
non-linearity which could be representative of micro-cracking). Since the structure
is undergoing free vibrations, the deformations should be linked to the stresses
generated from the by-pulse loading, which implies that a dynamic analysis can be
used to calculate the moments to link to the bends within the scope of behavior law
identification. Nevertheless as long as we stay at moderate loading levels and deal
with the behavior just before a brittle material fails, an elastic analysis is
satisfactory. The divergence from elastic behavior can be identified “at a quasi-static
speed”.

1.2.1.4. Shock tube tests on plates

The principle of a gas pressure by-pulse loading can also be applied by resorting
to a uniform loading the value of which is controlled thanks to a tube used as a wave
guide and called a shock tube. Using such a device is quite conventional for testing
industrial equipment in the defense field. Using the device for structure elements
was developed more recently [TOU 93]. Using explosives is limited and the loading
profile as well as its spatial repartition is better controlled than open-air explosions.
In so far as the conditions at limits can also be well controlled, we can directly
access to the behavior of a bending plate, which represents “basic” data for the
structure designer [KRA 93] or a simple basic situation to validate a behavior model
[PON 95, SER 98a].

The innovation of this trial was that it generated loading by means of a well-
controlled air shock wave (Figure 1.3). By using the closed tube, for the same plate
with the same support conditions, it is possible to carry out quasi-static loadings by
slowly inflating the whole tube. As an example¸ a 35 m long tube, 66.6 cm in
diameter, was used to compile an important experimental database about concrete
and reinforced concrete plate bending parameters [TOU 95a].
 Dynamic Behavior of Concrete: Experimental Aspects 15

Figure 1.3. Skeleton diagram of the shock tube trial (from [TOU 95a])

Considering the inner diameters of the tube and the support area (82 cm), to
preserve the cylindrical symmetry of the test, the test sample is a “thin plate”
(thickness/span  1/10) 900 mm in diameter and 8 cm high. It has dimensions
compatible with the performances of the tube (allowing it to actually reach failure
requires using a concrete with aggregates that are not too small, or realistically
standard reinforcement (welded wire mesh), or fiber reinforcements). We can note
the particular care taken to achieve limit conditions close to those for an ideal simple
support, the circular slab being “pinched” between the humps of two massive
guides, a thin rubber-steel sandwich (a 3 cm wide ring) allowing absorption of
geometric defects and distribution of the clamping load. Its stiffness has been
measured, and control of the displacement and acceleration on the supports during
blasts enables analysis of the bending of the support slab under uniform loading on a
driven reference line.
16 Dynamic Behavior of Concrete and Seismic Engineering

Figure 1.4. Shock tube failure trials for a plain or reinforced

concrete slab (from [TOU 95a])

In addition to excellent loading control and a size adapted to a well controlled

trial on “realistic” concrete, the advantages of this test are the realistic
representativity (bending is obtained with maximum deformation speed typically
ranging from 0.01 s-1 and 1 s-1, which corresponds quite well to the “hard” shock
range) and geometric simplicity (radial symmetry is preserved up to cracking) which
make it possible to validate a calculation model as well as for comparing various
materials. The relative ease of interpretation stems from the fast loading building up
(about 10 Ps for a maximum deformation reached in about 1 ms) and from the
absence of pressure gradients on the loaded face. We can consider that the plate is
loaded instantly (vibration setting with a first deformation peak which is particularly
intense compared to static loading), but with a bearing constant loading, which
allows a stationary vibration rate to be set up before unloading. A “conventional”
modal analysis enables access to local stresses and strains, at least until cracking
starts.

In [TOU 95a], the details about the instrumentation implemented to characterize

strains in test samples in these types of trials are presented. We have seen that in a
series of plain or reinforced concrete plates with strength in the range of 35 to 120
MPa (Figure 1.4), we are able to show the progressive deterioration of the modal
response (frequency drop, increasing damping), the appearance of deflection,
plastification of the reinforcement, crack progression (which is sometimes delayed
with regard to the maximum strain rate) and the collapse mode type (shear
force/bending competition) the respective appearances of which can be justified by
limit analysis-inspired calculations [TOU 95a].
 Dynamic Behavior of Concrete: Experimental Aspects 17

1.2.2. Tests with prevailing spherical behavior

When loading has a strong tri-axial component, concrete undergoes a global

reaction resembling that of a coherent material, even when it has failed on a
microscopic scale, which is the case for confinements over 10% [GAR 99]. The
models used to describe this are generally plastic models (not necessarily standard
and usually coupled to deviatoric and spherical behaviors). In these cases, even high
strain gradients do not bring about failure or localization, and the concrete sample
can be analyzed as if composed of a homogenous material.

1.2.2.1. Slab-plate tests

In the standard case [ZUK 82], loading is caused by the impact of two identical
plates. The impact speed is known. On the unused slab, a rear face rate measurement
(usually made using laser interferometers) is conducted. Another version of the trial
involves applying the same type of loading (in plane strain), using an explosive. The
shock induces a plane shock wave propagating at a velocity D. Discontinuities of
material rate u, pressure P, volumic mass or mass volume V and inner energy E are
associated with this wave. Assuming the material speed (u0 = 0) and pressure (P0 =
0) initial conditions are zero, we can infer from the Rankine-Hugoniot conservation
equations that:

V V0 ( D  u ) / D
P Du / V0 [1.5]
E E0  P / 2(V0  V )

As the time of the shock is known (by contact measurement for example),
measuring the free rear face speed allows us to locate the moment when the wave
arrives and to measure D. It also allows us to calculate u. Thus, one test establishes a
relationship between P and V, and also between D and u: these are called “shock
polar curves”. To deduce a strain-stress uniaxial relationship from them, we will
have to make a hypothesis about the behavior model of the material.

For metals and high strength shocks, the elastic response is neglected, and we
assume that the plastic behavior is purely deviatoric (without any volume variation).
Strictly speaking, concrete behavior analysis should be different. Each test gives a
point on a curve. The “Hugoniot curve” links pressure to material physical speed
and the “ shock polar curve” links shock speed with material speed (objective
measurements). There is therefore no direct way of converting this to mechanical
values that geomechanical engineers are familiar with.
18 Dynamic Behavior of Concrete and Seismic Engineering

The plate-plate test is a relatively pure trial. However, it has to be interpreted, is

difficult to implement, and only can only inform us about concrete compaction
behavior at very high strain rates (above 105 s-1).

1.2.2.2. Hopkinson bar tests with strong confinement

This test was developed at the LMS in co-operation with LMT Cachan [GAR
99]. A cylindrical specimen is confined within a metal cylinder (Figure 1.5). It is
loaded using a large diameter (80 mm) steel Hopkinson bar, which allows the use of
test samples large enough in comparison with aggregate size to be adequately
representative of the material.

Figure 1.5. Confined sample for Hopkinson bar test (from [GAR 99])

The complete collection and analysis of the signals recorded on the bars
(described in section 1.1.2.2) allows the measurement of the forces and
displacements applied on both faces of the sample.

When the input and output forces are equal (which is the case shown in Figure
1.6) and we can assume an homogenous state of stress and strain, the stresses,
strains and axial strain speeds can be deduced.
 Dynamic Behavior of Concrete: Experimental Aspects 19

Figure 1.6. Static and dynamic volume-pressure relationship

Figure 1.7. Static and dynamic volume-pressure relationship

20 Dynamic Behavior of Concrete and Seismic Engineering

The behavior law of the metallic ring is known. A thick enough ring to remain in
the elastic field allows the application of strong confinements. Using a ring made of
material that enters the plastic field (brass for example) will enable controlled
confinement to be applied. Thus, measuring the transversal strain of the ring allows
the confinement to be calculated, after which we can calculate the values that are
usually dealt with in geomechanics. As an example, Figure 1.7 shows evolution of
the volume-pressure relationship compared to the same relationship obtained using a
static trial.

1.3. Tests with transient phase conditioned interpretations

1.3.1. Tests involving mainly traction behavior

1.3.1.1. Modified Hopkinson bar

As explained in section 1.2.1.1, traction behavior is essential for characterizing
the failure of brittle geomaterials like concrete, which is why adapted tests have
been designed to obtain this data for high speeds, and has been widely studied.

The design has been achieved, mainly thanks to modified Hopkinson bar
configurations in which the specimen is glued between the input and output bars,
where it is submitted to traction produced by a shock to a retaining shoulder at the
end of the input bar. The main results with this technique were obtained on the
device of the University of Technology in Delft [REI 86] and [ZIE 82] between
1980 and 1995. The tested specimens are typically core sampling specimens 74 mm
in diameter (the same diameter as the bars), with a 1 to 1.5 slenderness. The duration
and energy of the shock which generates the traction wave depends on the mass
used, hydrostatic pressure and the number of dampers inserted between the masses
whose fall is triggered and the lower input bar shoulder.

In practice, as we want the shock to be intense enough to cause specimen failure,

and the loading build-up rate to be constant during the trial, the device allows
loading rates ranging from 4 to 200 MN/s, about 100 to 1,000 above the rates
reached with conventional press machines with similar specimen geometries.

The analysis of specimen loading uses the transient analysis described in section
1.1. The quality of glueing interfaces and the nature of the aluminum bars
contributes to impedance compatibility between concrete and the loaded material, so
an important part of the wave is transmitted to the specimen and the obstacles to
transversal strains are limited. We have verified that the transmitted-wave signal
gives a precise measurement of the average stress developed inside the sample –
after conversion into stress and calibration in time.
 Dynamic Behavior of Concrete: Experimental Aspects 21

Figure 1.8. Direct traction tests on Delft University’s Hopkinson bar,

plain concrete and very high performance concrete

The simultaneous measurement of the strains on the specimen (Figure 1.8) is

made possible either by extensometers gages glued to the sample [TOU 95a] or by
pre-slotted fiber concrete (where the measurements concerns crack opening), by
gages fixed directly on the sample [TOU 99b]. For the speeds considered, the time
delay between stress and strain signals is about 220 Ps, whilst the space difference is
about 1 meter. The “suitable” loading time (from 0 to maximum load) ranges from
100 to 500 Ps, and sampling is carried out at 250 kHz. The excellent stress-strain
linearity obtained confirms the validity of the hypotheses. Nevertheless, considering
the time to go through the specimen (about 25 Ps, i.e. a difference about 1 MPa), the
rates reached limit the interpretation as far as sample homogenity is concerned.
22 Dynamic Behavior of Concrete and Seismic Engineering

14

Figure 1.9. Ispra Centre device for big sample testing

To improve understanding of the mechanisms of traction failure and crack

dynamic propagation, a specific device has been developed for effort transmission
and measurement and is included in the large-scale dynamic test equipment (LDTF)
at the European Research Centre Ispra [CAD 01]. To increase the capacity of the
shock transmitted to the specimen at that installation (20 cm-edge cube), the shock is
generated by the violent release of a tight cable. The device (a Hopkinson Bar
Bundle (HBB)) consists of a prismatic Hopkinson bar beam, each bar being
instrumented, which transmits the traction wave to the specimen. Potential helical
reinforcements at both ends of the specimen are eliminated. It is possible to follow
both the opening of a crack across the specimen and the loading transmission
remaining in the not yet broken ligament, by applying a simplifying hypothesis of
wave propagation and load transmission inside the breaking specimen.

Most of the significant results concerning high-speed traction concrete behavior

detailed in section 1.5 were discovered using this installation (Figure 1.9) on quite
large scales.
 Dynamic Behavior of Concrete: Experimental Aspects 23

1.3.1.2. Hopkinson bar Brazilian test

The test is an expansion of the Brazilian test, whose traditional analysis is based
on the assumption of brittle elastic behavior. We consider an elastic cylinder
compressed perpendicularly to its generators: compression is applied along two
diametric generators. A plane deformation elastic calculation shows that loading
causes practically constant traction maximum stress along the cylinder axle, at right
angles to the compression axle. We assume cylinder failure takes place when the
strain reaches the ultimate value. Carrying out this test in quasi-statics is not
obvious, as it requires strict respect for limit conditions and the ideal elastic model
(stiff supports among others). Nonetheless, this trial is easy to carry out and gives a
consistent order of magnitude for simple traction failure stress.

Extension to the dynamic situation is easy. Compression is applied using a

Hopkinson bar. If we want to analyze the results in the standard way, we suppose
that the situation is not too far from the quasi-static case. To do this, we have to
assume that inertial effects can be neglected. They can be neglected before failure
but, as is the case for simple compression, they cause an apparent increase in the
maximum load after failure, so consequently it is important to detect failure by
direct observation (using high-speed imaging), as it is for dynamic compression tests
where localization of strains with block development does not necessarily lead to
load drop immediately. We should also check that the mechanical fields are not too
far away from the fields we would have in statics at the same applied force value.
Thus, we have to verify that failure occurs at a time when input and output forces
are quasi-equal. Such a situation will only happen when loading is slower than the
homogenizing time (typically the time for the elastic waves to cover the diameter of
the sample several times).

Achieving all these conditions simultaneously is difficult, but as we saw in

section 1.1.2.2, the Hopkinson bar provides us with information about the loads and
displacements applied to the sample all the time. Assuming this data is accurate, we
can then carry out a numeric simulation of the test (assuming brittle elastic
behavior), which gives a more precise assessment of failure strength [TED 93].
However, this hybrid approach (calculation-test association) is that of a structure
trial, and is better suited to model validation than directly determining a behavior
parameter.

1.3.1.3. Scabbing test

The scabbing test is a test with a fundamentally transient analysis. Actually it is
based on analyzing wave propagation inside a bar made of the material itself.
Concrete, a brittle material has a uniaxial compression strength that is clearly
superior to its traction strength.
24 Dynamic Behavior of Concrete and Seismic Engineering

By using an assembly like the one in Figure 1.10, we induce a compression wave
(propagating to the right in the figure), which is reflected at the free end as a traction
wave [BRA 99], [DIA 97].

Figure 1.10. Scabbing test diagram

The compression pulse produced by the impactor is measured via a strain gauge
glued to the bar. The elastic properties of the bar and the sample being known, we
can infer the shape of the pulse induced inside the sample. We can also glue a gauge
on the specimen to measure it directly. The compression wave thus produced has a
lower amplitude stress than the compression concrete failure stress. The opposite
amplitude reflected traction wave is sufficient to cause failure in the sample at a
specific position. By applying the principle of elastic wave superposition, we can
infer the stress value at the failure point. The analysis is easy because the pulse is
short compared to the propagation time inside the sample. This is why we use short
impactors and long specimens. Making specimens respecting homogenity conditions
is therefore delicate.

The accuracy of the test analysis can be improved by additional information such
as the failure instant, which can be obtained by high-speed imaging. In some cases,
we can observe successive failures in the sample, analysis of which gives redundant
measurements of failure stress.

This trial also gives accurate and reliable measurements of limit conditions, and
the loading parameters are well-mastered. Fine interpretation still remains difficult
as it is one-dimensional (as far as wave propagation is concerned), whereas failure
has to propagate in the transverse direction. Moreover, the characteristic
phenomenon is quite local. High strain gradients do not allow easy measurement of
the strain rate characteristic of the test. This speed is usually taken as the strain time
derivative near the failure point; for a one-dimensional wave, this derivative is
proportional to the deformation spatial derivative.

For concrete, a very marked increase of failure stress with strain rate has been
observed [BRA 99]. Between 1 and 100 s-1, failure stress can be multiplied by as
much as a factor of 10. The physical interpretation of this result still has to be more
closely examined.
 Dynamic Behavior of Concrete: Experimental Aspects 25

1.3.2. Tests implementing compression behavior

1.3.2.1. Hopkinson bar trial

As explained in section 1.1.2.2, the Hopkinson bar allows an accurate
measurement of the forces and displacements applied on a both faces of a sample,
especially in compression. Particular precautions pointed out give access to the weak
strain area in the case of concrete. Figure 1.11 shows an example of the forces
measured on each face, as well as the rates applied to each face of the sample
(Figure 1.12) and the associated displacements (Figure 1.13).

For this test, the specimen is initially 40 mm in both length and diameter. Its
relative density is 2.25 kg/m3, with a largest aggregate diameter of 8 mm. It is
loaded via an aluminum Hopkinson bar, 40 mm in diameter. The 1.3 m long
impactor is projected with a speed of 14.5 m/s. When observing the speeds to be
measured, we notice that the specimen absorbs little of the available energy, since
the loading bar speed is roughly equal to the initial speed of the impactor at the end
of the test, i.e. when the sample has failed. The induced displacements are very low,
as the displacement associated with the force peaks is below 1 mm. The post-peak
phase observed on the loads says a lot about the existence of inertial confinement.

Figure 1.11. Hopkinson bar compression test input and output loads
26 Dynamic Behavior of Concrete and Seismic Engineering

Figure 1.12. Hopkinson bar compression test input and output rates

A one-dimensional transient elastic calculation simulates the test quite well. The
incident wave being known, we calculate the reflected and transmitted waves during
the first 45 Ps, after which the result of the calculation suddenly deviates from the
measurements. From this, we can infer that failure takes place after 45 Ps at the
latest. This instant is more or less synchronous with the output load peak. For this
calculation, the apparent elastic modulus is 7.8 GPa. This “modulus” is quite weak
and can probably be explained by concentrated strains at the interfaces between the
specimen and bar surfaces: these cannot be neglected in statics either when
considering the specific strains on a specimen. If we only consider this phase of the
trial, we notice a time shift between the maximum of the load, equal to about half
the 45 Ps.
 Dynamic Behavior of Concrete: Experimental Aspects 27

Figure 1.13. Hopkinson bar compression test. Displacement at input and output faces

These values taken into account, it is certainly not reasonable to suppose the
mechanic fields are homogenous and deduce a stress–strain relationship from them.
Nevertheless, if this simplified analysis is done to obtain an order of magnitude of
the strain associated to the stress peak and corresponding strain rate, we obtain the
results shown below (Figure 1.14). Depending on the way the stress is calculated
(without homogenity, there is no reason to consider the average effort more than the
output load), the stress peak is reached for an overall “strain” (average relative
displacement between the input and output faces) ranging from 0.75% to 1%. In
fact, as indicated, failure probably occurred 20 Ps before, i.e. maybe for a half as
low “strain”, the average strain rate is about 200 m/s without being really accurate.

A transient analysis can be carried out within a one-dimensional frame ([GAR

96], [GAR 98], [ZHA 96]), by using a “simple” negative strain-hardening elasto-
visco-plastic model. This approach gives results in accordance with the
measurements. However, it is insufficient, because it considers the sample as the
material and it does not take into account structure effects or inertial confinement. It
makes more sense to use a three-dimensional model and to simulate the test using
finite element dynamic calculation. This approach was developed in the GEO
network [BAI 99].
28 Dynamic Behavior of Concrete and Seismic Engineering

Figure 1.14. Approximate average behavior

At last, the same type of trial can be carried out in confinement, using
confinement cells coupled to a compression Hopkinson bar system [GAR 98]. It is
important to use pressure confinement with gas and a large enough chamber inside.
Using an incompressible fluid actually leads to interference confinement, because it
acts as a strain limiter. Typical results [GAR 98] show that the confinement effect is
the same kind as the strain speed effect to which it is added. Therefore, we can infer
that the main effect of strain rate in a concrete dynamic compression test is a
structure effect linked to the inertial confinement. This result was confirmed by
digital simulations, which are developed in [BAI 99].

1.3.2.2. Direct impact tests (shock cannon)

The Hopkinson bar test gives information on high-speed concrete compression
behavior, however the practical limitation of transversal dimensions (less than 10 cm
in general) works against good representativity of the concrete material. That is why
“block bar” devices that carry out direct impacts on a concrete cylinder have been
developed ([BIS 95], [DAR 95]). The impactor is guided and propelled either by
direct falling or a compressed-air canon. The quality of the contact between the
 Dynamic Behavior of Concrete: Experimental Aspects 29

impactor and the sample face is essential for a good repartition of loads. In practice,
the surface of the impactor and that at the rear of the reaction system can lead to a
limitation of the transverse strains at both ends of the specimen, upon which it is
necessary to use a specimen with sufficient slenderness (2 or more). If an efficient
anti-helical reinforcement system is used, we observe prismatic failure
corresponding to the cutting of angular sectors [MUR 86].

The rates reached are important, ranging from 1 to 100 s-1. The strains can only
be measured by extensometer gages glued to the sample. An artifact is possible due
to confinement, which is different at the heart and at the periphery of the specimen.
Measuring the loads also requires particular precautions, considering the inertia of
the impactor. It is generally inaccurate to consider the acceleration measurement on
the impactor. Bischoff and Perry developed an ultra-flat pressure cell to minimize
the reflections of waves therein, the cell being inserted between the specimen and
the assumed motionless reaction body at the back of the specimen.

The results obtained with these devices (mainly the maximum average stress
reached) are along the same lines as those obtained on Hopkinson bars. Guidance
defects (centering loads) and surface evenness (contact hard spots) could be the
cause of dislocations when the results are in a continuity with those obtained with
presses. Besides which, the immobility of the reaction device has to be verified,
otherwise a correction for inertia becomes necessary. Anyway, at the speeds
reached, the direct transformation of the maximum effort recorded into “failure
stress” can only be considered as conventional, as the analysis reveals inertial effects
and the “inhomogenous” divergent feature of the axial and radial stresses inside the
specimen.

1.4. Other tests

1.4.1. Tests adaptable to an energetic approach

In all that has been said so far, we have noted the difficulties linked to finely
identifying the high-speed dynamic properties of concrete. Considering these
experimental difficulties, and the necessity to identify calculation parameters simple
enough for engineers, some means have been developed to enhance – in a
comparative way – the energy absorption properties of some concretes (especially
fiber reinforced concrete). Basically the approach involves adapting the resilience
test, which is standard for metals, to concrete, and which corresponds to a dynamic
bending loading, the load being applied via a pendulum ram impact testing machine
(Charpy test).
30 Dynamic Behavior of Concrete and Seismic Engineering

Interpreting the test relies on the fact that the drop energy (potential energy of
the pendulum, turned into kinetic energy) is partially transmitted to the specimen as
strain energy, and can even be completely consumed in the event of the specimen
failing without resilience of the ram. Depending on the type of impacted sample
(prism slotted or not), the failure energy determined in this way is incorporated in
the hypothesis of a brittle-elastic behavior, by determining either a bending traction
strength, the “global” energy of the failure per unit area, or an energy restitution
critical rate. Employing the result of the “pre-dimensioning” test involves
determining (by calibrating in comparison to thickness of well-known materials) a
material thickness as a function of the shock energy to be absorbed, in applications
where the concrete wall has to withstand bending impacts. Whatever the case, if this
type of test allows a comparative analysis, it can be subject to artifacts, due to the
energy dissipation sources which are not taken into account (movement of supports,
frame and sample, local dissipation at the impact spot, local deterioration of the
concrete and heating). The respective masses of the specimen, pendulum and frame
are such that the inertia of the test body can rarely be neglected in shock.

However, the use of a pendulum system is based on a good control of the

initially transmitted energy, which is why a traction device adapted from a
pendulum has been developed ([BAN 91], [BAN 96]) so as to carry out traction or
pull-out tests on fiber-reinforced concrete. The strain rate reached with this type of
device is about 0.1 s-1. The load measurements carried out are difficult to interpret,
as the presence of the sample acts as a divergence from a no-load measurement. A
significant part of the energy seems to be dissipated by frame vibrations.

1.4.2. Validation tests on structures requiring an inverse analysis

1.4.2.1. Falling mass tests

Generating shocks using falling mass devices, for example with pendulum
systems, is a relatively simple and economical solution to produce high-speed
loadings on concrete test-bodies, which is also useful in typical loadings, where
concrete distortions on plates essentially operate in bending. These are no longer
behavior identification tests of a material, but tests of representative structure
components. These devices have been widely used ([KRA 93], [KRA 96], [MIN
87]). One advantage – theoretically at least – is that it enables a “static” test with
similar limit conditions and loading application geometry to be carried out.

Nevertheless, correctly interpreting this kind of test requires a careful inverse

analysis, if we want to identify details of the dynamic behavior of the concrete
material (possibility of parametric studies with varying properties of the constituent
materials) or of the reinforcement [DAN 01]. Actually the generally sought
 Dynamic Behavior of Concrete: Experimental Aspects 31

operation is a bending mechanism, though with some inherent complexities [KRA

93]. We need to consider:
– operation as a slab which cannot be reduced to a beam operation;
– plate operation (diaphragm effect);
– possible local mechanisms depending on the shock range and energy (local
crush, punching or transverse load failure on the supports);
– sensitivity to the support conditions (problem linked to the dynamic unilateral
support conditions);
– problems linked to the resilience of the impactor;
– sensitivity of the failure mode to singularities (corner effects).

As an example of the extra information that can emerge if the above points are
considered, an important sensitivity of such tests’ overall results to the “nose-
shaped” impactor [BIN 01] has been shown, with a part of the impact energy being
consumed when starting a penetration mechanism [WAT 02]. When a complete
inverse analysis can be conducted, notably with well controlled loading and limit
conditions, the competition between the different damage and collapse mechanisms
can be highlighted according to the impact speed, which was especially the case for
reinforced concrete walls submitted to the impact of a pendulum in [MIY 91a] and
[MIY 91b].

As regards validating complex systems [SAT 95a] comprising a concrete

structure (reinforced or pre-stressed) and/or supports interacting with the ground,
tests where impact is achieved by means of a falling mass are often used [PER 01],
the ability to control and modulate the incident energy (blocks ranging from a few
tens to several hundreds of kilos, falling heights up to 30 m – limitations due to the
sizes of cranes) must be taken into account. Re-calculating the test is often difficult
because of the frequent presence of dissipative materials (ground, granular materials,
energy dampers). Research into systematic empirical interpretations worthy of note
(especially with a view to dimensioning rock fall devices to protect transport
infrastructures) includes [LAB 96], [MIK 95], [MON 98] and [SAT 95b].

1.4.2.2. Block fall tests

If we want to collect exploitable information about the participation of the
different materials and components in the overall strength, tests on structure
elements require an inverse analysis, which is often complex in dynamics. However,
in so far as the load and the test body are close to the “real” situation, some tests are
used directly, to confirm and/or compare various technical solutions. In a manner
complementary to block-falling tests, tests where the concrete element itself is
32 Dynamic Behavior of Concrete and Seismic Engineering

submitted to a shock are used. This is especially the case when validating
radioactive waste containers ([LAE 94], [VEC 88]). In order to anticipate the result
of this kind of test, simplified analysis methods have been developed ([MAR 87],
[SER 98a]), and these allow parametric study when designing and developing
prototypes. This experimental configuration allows good control of energy during
the shock, and confirmation of the hypothesis when the impacted area is motionless
and dimensionally stable. On the other hand, fitting structures with the required
instruments for this is generally expensive and difficult (on-board accelerometers);
consequently, identifying the analysis should be based on a posteriori observable
cracking state.

1.4.2.3. Explosion resistance test

For a number of protection structures, overall dynamic loading is more
significant than a local impact. Adapted large-scale experimentation is carried out
via pressure loading caused by an explosion. Control of such loading and its
similarity rules is rather good, subject to limit conditions and simple structure
geometry, which limits uncertainties linked to reflections. In this kind of experiment,
instrumentation can be quite complex (numerous pressure sensors, gauges,
displacement sensors and accelerometers). The most important factors are the data
acquisition speed, and the qualities of activation and filtering. Many experiments
have been carried out on typical structures, including structural walls and slabs
([GRO 90], [KR 96]), vaults and tunnels [KRA 89] and hot caves. Controlling the
mechanical limit conditions is the main difficulty for obtaining a precise inverse
analysis in such cases; therefore, this kind of test is often used for validating
simplified regulations and abacuses for the dimensioning of protection structures.

Another category of tests carried out using blasting charges is aimed at

characterizing the “compaction” behavior of a material, the latter being closest to the
blasting charge in a highly confined stress state due to the pressure resulting from
the explosion nearby on the one hand, and to the rest of the surrounding structure not
yet hit by the loading wave. Such tests are transient, and the analysis requires re-
calculation (made easier by the semi-spherical symmetry of the problem). The data
is generally adapted to a volume-pressure limit curve interpretation, characteristic of
the areas submitted to strong tri-axial compressions. A recent description and
interpretation of such tests can be found in the works of the GEO network [BAI 99].
 Dynamic Behavior of Concrete: Experimental Aspects 33

1.5. Synthesis of the experimental data on concrete and associated materials

1.5.1. Data on cement paste mortar and concrete

1.5.1.1. Looking for a consistent interpretation

When analyzing experimental results, the multiplicity of experimental data,
obtained under different stress conditions, and the difficulty in interpreting them
with frequent necessary re-calculation and transient analysis, coupled with the
variety of observation scales used makes a clear presentation of the characteristics of
concrete’s dynamic behavior rather difficult.

We propose an interpretation based on Rossi’s initial ideas [ROS 91],

highlighted by many experimental programmes, the validity range of which is being
discussed. The underlying idea is essentially linked to the fact that cement materials
that have had evaporable water removed do not show any evolution of strength
when stressed at rates ranging from 10-6 to 10 s-1 [TOU 95a]. This fact has been
verified on mortar and concrete, in both traction and compression ([DAR 95], [HAR
90], [ROS 92b], [TOU 95a]), within a domain where the quasi-static interpretation
of the test results is valid and thus allows a “conventional” interpretation of the
behavior of the material.

We can infer from this that the “sensitivity” of the concrete material to stress rate
is (within the considered domain) linked to the presence of free water inside the
porous material [TOU 99a]. This fact makes the control of its hydration state crucial
when its properties are being identified at high speeds, which is rather difficult to
implement. Part of the variability of the results observed in literature can be
explained by partial drying of specimens ([COW 66], [KAP 80]), in addition to self-
stresses linked to drying, which are superimposed to the initial mechanical state of
test bodies, the influence of which is all the more important due to their small size.
34 Dynamic Behavior of Concrete and Seismic Engineering

Figure 1.15. Apparent strength of concrete in traction

dynamic loading (from [TOU 95a])

The free water present in the porous volume is stressed like a viscous fluid by the
(fast) motion imposed on the sides of the skeleton, which in the ideal case of a film
between two walls is known as the Stefan effect. The consequences of this can be
observed in high-speed traction or compression tests as an increase (low relative
value) of stiffness, and a more significant increase in the strength, called the rate
effect. The macroscopic stress increase can then be interpreted by partition between
the stresses borne by the skeleton and viscous stresses borne by the fluid (Figure
1.15). Things progress as if these viscous stresses cause pre-stress in the skeleton
and delay either its traction failure or the failure in the extension direction induced
by loading when the latter is not purely tri-axial. The partition and its effect on
material failure are at the root of the elasto-plastic viscous strain-hardening model
developed by Sercombe [SER 98b].

For higher-rate tests (over 1 to 10 s-1), even when the hydration state is well
controlled, the transient character of the test and the failure phase of the specimen
take precedence over the rate effect linked to the nature of the material, qualitatively
at least [WEE 98]. The relative increase in “strength”, compared to the static
reference value, can exceed a value of 2, even for specimens in which free water has
been eliminated [ROS 96]. In fact, we can notice that a dynamic failure mechanical
analysis (which takes critical crack propagation inside a material with non-zero
inertia into account) is consistent with the experimental observation, which is that
the relative strength increase (dynamic increase factor (DIF)) evolves with the strain
 Dynamic Behavior of Concrete: Experimental Aspects 35

rate in a logarithmic diagram, along a 1/3 slope straight line ([CHA 98], [KIP 80]),
beyond a certain threshold (typically 1 to 30 s-1, depending on geometry and
loading) which corresponds to the limit beyond which the test has to be analyzed as
a transient state ([REI 91], [WEE 89]). An analogous model taking local inertia into
account [BAI 94] also justifies the “double state” obtained experimentally if the
transient character of the failure is interpreted as a local property.

The latest results obtained on quite large-size concrete samples [CAD 01] are
consistent with these two basic mechanisms causing the strength increases observed
during high-speed dynamic tests, with the participation and viscosity of water,
beyond a specific threshold, and the participation of inertia on both sides of the
failure origin.

1.5.1.2. Effect of the structure of the cement paste

The crucial part played by free water in determining the sensitivity of concrete
behavior to loading speed in a transient state has led to speculations about the
relevance of conventional parameters used to describe the dependence. Actually, it
appeared that the conventional definitions ([COL 88], [BIS 91], [MAL 98]) of
compression or traction DIFs, as well as those of ultimate strain or Young’s
modulus, have led to values varying according to the static properties of concrete,
including compression strength ([COL 88], [JAW 87], [ROS 95]). This is apparently
responsible for the wide discrepancies in the diagrams used to describe strength
evolution (Figures 1.16 and 1.17), and interferes with them being taken easily and
reliably into account in a regulation context.

Figure 1.16. Concrete compression strength. “Rate effects” (from [BIS 91])
36 Dynamic Behavior of Concrete and Seismic Engineering

If, as a first approximation, we assert that in the rate domain (where inertial
effects can be neglected), absolute traction strength increases observed in concretes
of various strengths (35 to 120 MPa [TOU 95a] and 230 MPa for tests on high
efficiency concretes [TOU 99b]) are not a function of the strength (Figure 1.18),
with a careful identification of the specimens to avoid hydration gradients, we have
proposed that this characteristic should be used [TOU 99a], rather than the relative
increase [ROS 95]. A reasonable order of magnitude is obtained with a rounded
down value of + 0.7 MPa/log10 (loading rate in MPa/s) to represent the increase in
strength, for a common or high-efficiency concrete. In the same way, a
+ 0.9 GPa/log10 increase of the Young’s modulus (loading rate in MPa/s) can be
adopted as a first approximation. From these values, using classical hypotheses
about concrete behavior in other stress states, we obtain a reasonable order of
magnitude for the compression strength increase (about + 6MPa/log10 with loading
rate in MPa/s).

Figure 1.17. Concrete traction strength. “Rate effects” (from [BRA 99])

More accurate identification of the really influential parameters has been

obtained from data for which the amount of hardened cement paste, the size of the
biggest aggregates and the water/cement ratio (control of micrometric porosity) were
used as the main variables of the experiment surface. The fact that the whole amount
 Dynamic Behavior of Concrete: Experimental Aspects 37

of hydrated calcium silicate (CSH) and the relative compactness of the aggregated
skeleton [LAR 00] (a characteristic value of the part of the cement paste and its
defects with regard to the part of the defects due to the paste-aggregate interfaces
and the compactness defects of the skeleton) could constitute two significant factors
to improve the accuracy of the description of rate effects as a function of the
composition of concrete has been brought about [TOU 95a]. Within the limit where
the strength is not first controlled by various defects, the sensitivity of the strength
of the material at loading speed is mostly controlled by the amount of free water
present in the nanopores of the material, which are intrinsic to the porosity of
hydrates and become saturated when the outside relative humidity exceeds 50%.

-3
1E-05 10 -4 10 10 -2 10 -1 11 10 100

Figure 1.18. Traction strength variation according to loading rate (from [TOU 95a])

1.5.1.3. Description of a plasticity criterion in terms of evolution

The previous description from monotonous testing is based on the evolution of
the maximum load applied to the specimen. For the purposes of dimensioning by
extrapolation from static calculations, on verifying the section strength or the
ultimate time, concrete dynamic strength (especially in compression for reinforced
concrete pieces) is the main problem, and the ultimate time can be used to calibrate
the oscillator that corresponds to the structure element under consideration,
according to several codes or recommendations ([COL 88], [COL 86]). With such
an approach, the DIF concept becomes interesting, despite the fact that the value
obtained is linked, amongst other things, to the considered concrete and the stress
rate.
38 Dynamic Behavior of Concrete and Seismic Engineering

For more comprehensive calculations, strength evolution is insufficient, as the

strain speed value is not constant and cannot define the behavior characteristics
alone. In addition, strain data during characterization experiments (modulus, limit
strains) are not abundant and are sometimes contradictory because of difficulties
with measurements and interpretations (like localization, the effects of which have
been described when mobilizing inner inertial loads), as shown by [COL 88], [BIS
91] and [SER 98a].

A “unified” use of results published in the literature can lead to a description of

the dynamic behavior of the material through the “evolution” of its static behavior.
The advantage of this approach is that it covers a priori (using a conventional three-
dimensional interpretation) all possible stress states, whilst also taking advantage of
the (rather rare) validations of criteria in strongly tri-axial stress domains. However,
from this perspective, using viscoplasticity or damage models with “standard”
gradients has limitations, because experimental data coincide rather badly with the
calibration of viscosity aimed at mastering numerical regularity problems ([GEO
98], [TOU 95b]).

This is why it has been necessary to explore more complex modeling by

extension of a damageable elasto-plastic model, thanks to a strain-hardening
variable with the same nature as a viscous strain [SER 98B]. This inner variable
corresponds to an extension strain (Figure 1.19), in so far as the rate effects are
principally linked to the deviatoric component of the stress state, the intervention of
confinement delaying localization of failure in the potential growth direction, the
same as in statics [KON 01]. The methodology for identifying the parameters of the
model from a relatively low number of well controlled empirical data points (direct
traction tests at various rates) have been detailed and validated by traction
simulations, compression and shear tests on specimens, and bending tests on slabs
[SER 98b]. These simulations have allowed it to be validated within the studied
domain, for testing hypotheses about the kinematic nature of viscous strain
hardening (Figure 1.20). The validation of such an approach should be continued
using high-confinement tests ([GON 90], [GRA 89], [MAL 91]). As for the problem
of falling containers, which was used as a basis for this development, the importance
of the various sophistications of the model (damaging, taking viscous strain-
hardening into account) was verified by a sensitivity study, which essentially
showed the behavior of the studied structure was governed by both local
compression of concrete at the impact point and propagation of induced tractions
within part of the structure where confinement was weak [SER 98a].
 Dynamic Behavior of Concrete: Experimental Aspects 39

Figure 1.19. Visco-elasto-plastic model with viscous strain hardening [SER 98a]. In this
p e
1D diagram, the total strain H is the sum of the 3 scalar terms H , x and H

Figure 1.20. Visco-elasto-plastic model with viscous strain hardening [SER 98a].
The plasticity criterion is translated into viscous strain hardening

Note that such modeling is below the localization of strains integrated into the
local description of the behavior. For the purpose of the study behind the
development of the model [TOU 99b], this limit seemed acceptable because
calculating the structure was not supposed to reveal any fracturing, and the
calculated strain rates appeared to be limited to about 1 s-1.

For justifications based on “ultimate limit state” calculations including possible

localization of strains in part of the structure, the question of integrating inertial
effects into the local behavior description can be raised [MAL 98]. We have to be
aware of the theoretical difficulties caused by taking non-linearities into account,
40 Dynamic Behavior of Concrete and Seismic Engineering

and difficulties in ensuring validation on the scale of the structure, as it has already
been enhanced during test interpretation.

1.5.2. Data available for reinforced concrete

1.5.2.1. Dynamic behavior of reinforcements

To design structures, the dynamic characteristics of the steel framework are as
crucial as those of concrete; indeed, the ductile character of a failure due to
accidental dynamic strain can be investigated, owing to failure due to R-bars.
Technical data concerning high-speed behavior of reinforced concrete framework
proved to be rare, and information about pre-stressed frameworks was non-existent.
A recent synthesis is available in [MAL 97]. The results are exploited to determine a
relative increase in traction strength (DIF) or elastic limit. Owing to the idealization
of framework behavior that is general in calculations and to the fact that no elasticity
modulus variation seems to be revealed at the strain rates considered, rate effects on
the elastic limit and strength are sufficient to ensure that a consistent behavior
description is obtained using standard calculation methods for reinforced concrete
sections.

The expression proposed by Malvar for describing the relative increase in steel
strength as a power of the strain rate (value imposed in the monotonic identification
tests) is consistent with the usual descriptions given for concrete ([MAL 98] for
example). For about 1/s, relative increases ranging from 10 to 50% of the elastic
limit can be expected, depending on the nature of the steel considered. Taking into
account the mainly one-dimensional feature of strain in frameworks, it must be
possible to directly calibrate a strain-hardening elasto-plastic model for frameworks
from this data – with the physical meaning of the variable driving the increase of
strength still to be determined.

1.5.2.2. Identification of steel-concrete adherence in dynamics

The R-bar-concrete adherence enables us to consider the strains of the
surrounding framework and concrete as identical, and is at the root of the operation
and calculation of reinforced concrete structures. The permanence of this property in
dynamics and the evolution of limit shear stress are basic questions; however,
experimental identification of adherence properties is relatively complex, because
measuring techniques used at the interface disrupt the phenomenon itself. In
practice, most of the sliding strength of reinforcement is provided by setting locks
on concrete, which disconnect when traction strength is reached in the transverse
direction. Adherence can then be considered to be directly linked to concrete traction
strength, as stipulated in most calculation regulations. Nevertheless, transverse
 Dynamic Behavior of Concrete: Experimental Aspects 41

confinement, which delays disconnection of the rods and activates their gearing with
each other, also has a significant influence [MAL 92].

In a reinforcement dynamic pull-out strain, we can expect some inertial

confinement to be brought about because the concrete around the reinforcement is in
traction, and traction strength increases activation. A description of the evolution of
adherence limit shear with regard to speed, which uses a DIF with the same form as
the equivalent coefficient for traction strength, has therefore been proposed [VOS
82]. It is based on pull-out tests carried out for an assembly using the Delft modified
Hopkinson bar [REI 82]. We note that for smooth steel, the adherence increase is not
significant, which confirms the proposed mechanisms.

1.5.2.3. Repeated stresses

The behavior of specimens submitted to repeated dynamic stresses is difficult to
access experimentally, especially when it is important to achieve significant damage
from the first impact. In practice, this information seems to be important if the
dynamic situation is not only considered as a fortuitous action, but as likely to be
subject to “replicas” (successive shocks between structure elements during
earthquakes, for example) or even as a servicing action (which could be the case
with rock fall protection works). As for behavior in traction repeated dynamic
stresses, it appears from [REI 82] that the “rate effect” displayed is reduced in
comparison to the DIF obtained for a single shock. However, the accumulation of
traction cycles would cause an apparent reduction of strength that could be of the
same order, even at low speeds. As regards numerous indeterminate repeated
shocks, avoiding crack propagation once they have started seems out of the question.
This leads us to rely on the cracking limit obtained in statics where we have to
guarantee the structure must not be cracked; otherwise calculations have to take
probable concrete cracking into account.

The main problem with repeated dynamic stresses concerns the operation of the
R-bar framework, especially if the action causes a variation of R-bar stress sign.
Without confinement, the effect of limit shear during a shock causes concrete
damage around the reinforcement according to orientation, and the effect of
alternating vibrations or additional shocks can keep on damaging the surrounding
concrete fast and irremediably, which reduces adherence and damages the anchorage
of the reinforcement. Therefore, it is advisable to cautiously take into account
the “adherence increase” due to rate effects, and to associate them with constructive
arrangements that will enable them to stay effective – transverse confinement, for
instance.

Implementing such arrangements has particular relevance for earthquake

resistance fields. Because of identification difficulties at a local level, it is often
42 Dynamic Behavior of Concrete and Seismic Engineering

necessary to resort to experiments on structure components (see [PAU 02] and [TOR
88]). In fact, at this level it is possible to underline the effectiveness of transverse
reinforcement, thereby ensuring the confinement of concrete beyond cracking,
allowing loads to travel after redistribution via the creation of “joints”. The amount
of experimental work in this field and its complexity will not be dealt with here.

It should be noted that the difficulties arise not because of the dynamic character
of stresses (some Hz frequencies with regard to the specific frequencies of the
elements – in the 100 Hz order of magnitude) or the dynamic reaction of the
material. The difficulties are linked to the intense and repeated feature of stresses
(incursions into the plastic field, stiffness damage, crack spreading), and even to the
interaction between the loading frequency and the specific frequency of the whole
structure (whether sound or progressively damaged).

1.5.3. Data about fiber-reinforced concretes

1.5.3.1. Post-cracking mechanisms and “rate effects”

Thanks to metallic fibers, the favorable influence of diffused reinforcement in
dissipating energy during a shock has been empirically proven [ROS 98], but is
difficult to quantify. The sensitivity to pealing caused by shocks during transport of
some prefabricated fiber reinforced concrete pieces, compared to corresponding
concrete or reinforced concrete pieces is admitted by professionals. Using fiber
reinforced concrete for particular applications where absorption of energy is
important has been shown to be interesting with regard to conventional reinforced
solutions [HAN 92]. Most of the time it was proved globally through experiments
allowing interpretations in terms of energy. However, complete and documented
experimental data on the dynamic behavior of such materials in traction for
characterizing after-peak behavior and revealing the contribution of fibers [ROS 98]
are quite rare.

The existing data [KÖR 88, TOU 99b] highlight major aspects of the rates of
behaviors of these materials: the increase in the linearity limit corresponding to
matrix traction strength, a phenomenon which can be directly compared to rate
effect of all cement materials in direct traction; also, a stress increase in the after-
peak phase (with regard to the load obtained in statics during this phase), the
increase being all the less important as a widespread range of crack openings is
considered. In other words, the relative increase of the absorbed energy and its peak
value is lower than the relative strength increase of the matrix, and it is even weaker
if we take crack openings into account.
 Dynamic Behavior of Concrete: Experimental Aspects 43

Such observations are consistent with the rate effect being a function of the
cement phase only. Before cracking occurs, fibers have no influence, and the
linearity-limit increase (matrix cracking) is similar to that observed in a non-fiber
reinforced concrete. After cracking however, the anchoring provided by fibers on
both sides of the crack assists in the maintenance of strength. As for reinforcement
anchoring, the strength of concrete mini-connecting rods confined by the presence of
the other fibers shows a rate effect, as the anchorage is limited by the strength of the
concrete around the fibers. Moreover, the wider the crack opening, the more micro-
cracked the concrete where anchoring has to take place will be, which limits the
effect of the viscous mechanisms underlying the strength increase.

A beneficial “synergy” could be observed by comparing the shock strength of

reinforced and fiber reinforced concrete pieces to those of fibered or reinforced
pieces. Considering the limited crack openings permitted by fibers for a given load,
once cracking has started, some confinement seems possible around the
reinforcements, which is not the case for reinforced concrete where adherence has
been damaged because of alternating dynamic stresses.

1.5.3.2. Anisotropy and its consequences

a) Opening rate 0.15 μm/s b) Mean opening rate 3.2 m/s

Figure 1.21a and b. Direct traction characterization tests on very high performance
concrete, at low and high rates, on pre-slotted specimens. Direction A specimens.
Low efficiency of the fibers (from [TOU 99b])
44 Dynamic Behavior of Concrete and Seismic Engineering

a) Opening rate 0.15 μm/s b) Mean opening rate 2.8 m/s

Figure 1.22a and b. Direct traction characterization tests on very high performance
concrete, at low and high rate, on pre-slotted specimens. Direction C specimens.
High efficiency of the fibers (from [TOU 99b])

Quasi-static reference
value Variation with speed
(0.05 MPa/s) in MPa/u. log.
in MPa
Traction strength 8 + 0.8
eq. 1 mm threshold stress 7 + 0.5
Young’s modulus 52,000 env. + 450

Table 1.1. Calculation characteristics for very high performance

concrete containers, derived from the traction characterization
(“rounded off” values used for calculation) (from [TOU 99b])
 Dynamic Behavior of Concrete: Experimental Aspects 45

Quasi-static value
(interpolation for
Trend of evolution
Value Direction 0.05 MPa/s) and
with rate
standard deviation

Young’s modulus
A 50.9 GPa (2 GPa) + 0.32 GPa/u.log.
C 53.9 GPa (2.5 GPa) + 0.38 GPa/u.log.
A+B+C 52.2 GPa + 0.31 GPa/u.log.
Maximum stress
(unslotted
specimens)
A 5.96 MPa (2 MPa) + 0.82 MPa/u. log.
C 11.17 MPa (3 MPa) + 0.77 MPa/u. log.
A+B+C 8.52 MPa + 0.65 MPa/u. log.
Maximum stress
(slotted
specimens)
A 4.29 MPa (2 MPa) + 0.70 MPa/u. log.
C 16.35 MPa (5 MPa) + 0.73 MPa/u. log.
A+B+C 9.65 MPa + 0.40 MPa/u. log.
Equivalent
threshold stress
(1 mm opening)
A 3.75 MPa (1.6 MPa) + 0.53 MPa/u. log.
C 13.99 MPa (3.8 MPa) + 0.62 MPa/u. log.
A+B+C 8.06 MPa + 0.45 MPa/u. log.

Table 1.2. Traction characterization of very high performance concrete and rate effects.
Results (means): A, B and C are the three perpendicular directions of the sampling
within an L piece deemed representative for the project (from [TOU 99b])

Let us recall that observing the mechanisms referred to above and obtaining
characteristic properties suited for the calculation of fiber reinforced concrete
structures demands respect for the strict regulations concerning possible anisotropies
in the behavior and constitution of the material, which may arise due to the
manufacturing mode of the structure [ROS 98]. It was thus possible to enhance
(Figures 1.21 and 1.22) rate effects (strength absolute increase), according to the
46 Dynamic Behavior of Concrete and Seismic Engineering

traction maximum stress or equivalent plastic stress absolute increase, which are
comparable as they are linked to the cement matrix for the different directions
considered in a fiber concrete piece.

Nevertheless, the absolute values of these strengths and stresses proved to be

widely different according to the direction (Tables 1.1 and 1.2), due to preferential
orientating of fibers during manufacturing [TOU 99b]. Therefore, the problem of a
potential taking into account of the anisotropy as far as fiber-reinforced structure
modeling is concerned remains in dynamics as well as in statics. The questions
related to the dispersion of properties are also the same.

1.6. Conclusion

The accumulated knowledge available for understanding and describing the

high-speed behavior of concrete material remains at a complex overall stage and
leaves both the structural engineer and the mechanic dissatisfied. This can be
explained by several factors: experimental difficulties in accessing the intrinsic
behavior of materials in dynamics tests, difficulties linked to the heterogenity range
of the concrete “material”, its sensitivity to the water environment, its brittleness as
a geomaterial which involves crack propagation effects within the specimens, and
the wide range of materials actually corresponding to the generic term “concrete”.
Besides this, we also have to note that part of the difficulties reflecting on
mechanical modeling problems are also present in the usual quasi-static field, even
if a standardized corpus valid for engineering common needs often avoids having to
ask too many questions.

After recalling the different experimental techniques that allow us to explore

concrete dynamic behavior, and after taking a few precautions, we described the
main established facts, i.e. the noticeable increase in strength and slight apparent
increase in the Young’s modulus, which can be explained by the viscosity of the
interstitial water present inside the nanopores (the finest pores within the cement
hydrates). This viscous inner phenomenon is inherent to porous solids, and can be
observed separately in direct traction tests over a standard range from 10-6 to 1 s-1. It
also explains the rate effects induced in other stresses (compression, adherence, fiber
concrete behavior) reasonably well. At higher rates, interpreting the tests involves a
transient analysis of the loading and failure phases of the specimen, as inertial
phenomena (in terms of measured loads) that oppose critical crack propagation
become predominant.

Different empirical description levels of the mechanisms have been developed,

together with the underlying theoretical support and its potential limits: the DIF,
which can vary widely depending on the rate and strength of concrete, absolute
 Dynamic Behavior of Concrete: Experimental Aspects 47

traction strength increase, which enables us to calibrate a visco-elasto-plastic model

with viscous strain-hardening, and application of dynamic failure mechanics when
high-rate failure propagation is at stake. The choice of the behavior description level
seems to be consistent with a clear choice of other modeling hypotheses: speed
range of the considered strains, unique or repeated characteristic of the dynamic
stress, presence or absence of uniaxial behavior, presence or absence of
reinforcements, possibility or impossibility of limiting the justification validity in a
domain where concrete can still be considered as continuous, etc. Depending on the
circumstances, we will naturally turn to a different description level with a more or
less important integration of chance in the properties of the material.

After taking these considerations into account, as well as considerations dealing

with dimensioning an important number of structures and increasingly considering
fortuitous situations ([PER 01], [TRO 01]), we underlined a few shortcomings. First
we have seen how difficult it is to access the material under high tri-axial stress.
However, it is a crucial problem for direct impacts, areas close to an explosive
charge or in cases of potential penetration. As for validating “viscous strain-
hardening” hypotheses, and in order to better model the combined effects of inertial
confinement [UNO 02], more complete information in the field would be quite
helpful as well.

While understanding the mechanisms in the case of a single dynamic stress can
be considered as correct and reliable for calculations, the problem of repeated
impacts remains difficult, both for validating the potential progressive damage
predictions models are likely to supply, and accurately taking into account the
evolution of adherence for reinforced concrete structures. This field seems to
represent very important stakes for engineers, in the frequent cases of fortuitous
dynamic stresses.

Finally, the increasing range of materials coming under the definition of

“concrete”, including fiber reinforced concretes, high and very high-performance
concretes and ultra-efficient fiber-cement composites will demand more diversified
validation of the indications and mechanisms highlighted in this synthesis.

1.7. Bibliography

[ALE 94] ALEXANDRE M., “Etablissement du cahier des charges d’un conteneur de haute
intégrité pour l’entreposage de longue durée des déchets B en vrac”, Rapport de la
commission pour les questions scientifiques et techniques relatives à la gestion des
déchets radioactifs auprès du conseil scientifique du CEA, March 1994.
48 Dynamic Behavior of Concrete and Seismic Engineering

[BAI 87] BAILLY P., BROSSARD J., DESROSIER C., RENARD J., “Approche
expérimentale du comportement et des caractéristiques dynamiques des mortiers armés de
fibres”, Cinquièmes journées universitaires de Génie Civil, Bordeaux, 1987.
[BAI 88] BAILLY P., Effets des explosions sur les constructions. Chargement et réponse de
la structure, Thesis, University of Orleans, June 1988.
[BAI 94] BAILLY P., “Une modélisation d’un matériau fragile avec prise en compte d’effets
dynamiques ”, C.R. de l’Académie des Sciences de Paris, Series II, vol. 318, pp. 1–6,
1994.
[BAI 99] BAILLY P., “Comportement des ouvrages en dynamique rapide”, Recueil de
communications du groupe de travail Géomatériaux Environnement Ouvrages (GEO),
Aussois, France, 1–3 February 1999 (Ouvrages et Actions accidentelles project).
[BAN 91] BANTHIA N., TROTTIER J.-F., “Deformed steel fiber – cementitious matrix
bond under impact”, Cement and Concrete Research, Vol. 21, pp. 158–168, 1991.
[BAN 96] BANTHIA N., MINDESS S., TROTTIER. J-F., “Impact resistance of steel fiber
reinforced concrete”, ACI Materials Journal, Vol. 93, N° 5, pp. 472–479, 1996.
[BIN 01] BINDIGANAVILE V., BANTHIA N., “Machine effect in the drop-weight impact
testing of plain concrete beams ”, 3rd International Conference on Concrete under Severe
Conditions: Environment and Loading, CONSEC’01, Vancouver, BC, Canada,
Proceedings Banthia, Sakai and Gjørv (Ed.), E&FN SPON, Vol. I, pp. 589–596, 18–20
June 2001.
[BIS 91] BISCHOFF P.H., PERRY S.H., “Compressive behavior of concrete at high strain
rates”, Materials and Structures, 24, pp. 425–450, 1991.
[BIS 95] BISCHOFF P.H., PERRY S.H., “Impact behavior of plain concrete loaded in
uniaxial compression”, J. of Eng. Mechanics, Vol. 121, N° 6, pp. 685–693, 1995.
[BOU 99] BOULAY C., LE MAOU F., RENWEZ S., “Quelques pièges à éviter lors de la
détermination de la résistance et du module en compression sur cylindres de béton”,
Bulletin des Laboratoires des Ponts et Chaussées, N° 220, pp. 63–74, March–April, 1999.
[BRA 99] BRARA A., Etude expérimentale de la traction dynamique du béton par écaillage,
PhD Thesis, University of Metz, 1999.
[BUS 02] BUSSAC M.N., COLLET P., GARY G., OTHMAN R., “An optimization method
for separating and rebuilding one-dimensional dispersive waves from multi-point
measurements. Application to elastic or viscoelastic bars”, J. Mech. Phys. Solids, Vol. 50,
N° 2, pp. 321–350, 2002.
[CAD 01] CADONI E., ALBERTINI C., LABIBES K., SOLOMOS G., “Behavior of plain
concrete subjected to tensile loading at high strain-rate”, in Fracture Mechanics of
Concrete Structures, de Borst et al. (Ed.), Swets & Zeitlinger, Lisse, The Netherlands,
pp. 341–347, 2001.
[CHA 98] CHANDRA D., KRAUTHAMMER T., “Rate-sensitive micro mechanical model
for concrete”, in ACI SP-175 Concrete and Blast effects, ACI International, W. Bounds
(Ed.), paper SP 175-14, pp. 281–305, 1998.
 Dynamic Behavior of Concrete: Experimental Aspects 49

[COL 86] COLLECTIF, Tm5-1300, Structures to resist the effects of accidental explosions,
Department of the Army, the Navy and the Air Force, technical manual, NAVFAC P-397,
AFM 88-22, 1986.
[COL 88] COLLECTIF, CEB, “Concrete Structures under impact and impulsive loading”,
Bulletin d’information, N° 187, Lausanne, Switzerland, 1988.
[COW 66] COWELL W. L., “Dynamic properties of plain Portland cement concrete”, Techn.
report R447, DASA 130181, US Naval Civ. Eng. Lab., Port Hueneme, California, 1966.
[DAN 01] DANCYGIER A.N., YANKELEVSKY D.Z., “The role of concrete in resisting
projectile penetration into concrete barriers”, 3rd Int. Conf. on Concrete under Severe
Conditions: Environment and Loading, CONSEC’01, Vancouver, BC, Canada, Proc. by
Banthia, Sakai and Gjørv (Ed.), E &FN SPON, Vol. I, pp. 612–619, 18–20 June, 2001.
[DAR 95] DARVE F., HICHER P.Y., REYNOUARD J.M., Les géomatériaux (3 volumes),
Hermès Science Publications, Paris, 1995 (particularly Vol. 1 Ch. 1, 5, 3; Vol. 2 Chs. 1
and 4, Vol. 3 Paragraphs 2.2.2, by Bailly, Gary, Klepaczko, Hamelin, Rossi,
Toutlemonde).
[DIA 97] DIAMURAYA M., KOBAYASHI H., NONAKA T., “Impact tensile strength and
fracture of concrete”, Journal de Physique IV France 7, Colloque C3 (Supplement of
Journal of Physics III), pp. 253–258, 1997.
[GAR 96] GARY G., ZHAO H., “Measurement of the dynamic behavior of concrete under
impact loading”, Proceedings of 2nd ISIE’96, Beijing, China, Chinese Journal of
Mechanics Press, Jin, Shi and Ding Editions, pp. 208–213, 1996.
[GAR 98] GARY G., BAILLY P., “Behavior of a quasi-brittle material at high strain rate.
Experiment and modeling”, European Journal of Mechanics, A/Solids, 17, N° 3, pp. 403–
420, 1998.
[GAR 99] GARY G., BAILLY P., GATUINGT F., “Testing concrete at high strains and high
rates of strain”, Proc. 3rd Int. Symp. on Impact Eng, Singapore (December 1998),
published in Impact Response of Materials & Structures, OUP, pp. 241–246, V.P.W.
Shim, S. Tanimura, C.T. Lim (Ed.), National University of Singapore, Singapore, 1999.
[GEO 98] GEORGIN J.-F., Contribution à la modélisation du béton sous sollicitation de
dynamique rapide. La prise en compte de l’effet de vitesse par la viscoplasticité, Thesis,
INSA de Lyon, 203 pages, 1998.
[GON 90] GONG J.C., MALVERN L.E., “Passively confined tests of axial dynamic
compressive strength of concrete”, Experimental Mechanics, pp. 55–59, 1990.
[GRA 89] GRAN J.K., FLORENCE A.L., COLTON J.D., “Dynamic tri-axial tests of high-
strength concrete”, J. of Engineering Mechanics, Vol. 115, N° 5, pp. 891–903, 1989.
[GRO 90] GROS J.-F., Etude du comportement dynamique de dalles de béton sollicitées par
des charges explosives, ENPC, 1990.
[HAN 92] HANUS F., FAURE J.-C., PEIFFER G., “Béton léger hautes performances –
utilisations dans les structures devant résister aux explosions ”, in Les bétons à hautes
performances. Caractérisation, durabilité, applications, Y. Malier (Ed.), Presses de
l’ENPC, pp. 651–665, 1992.
50 Dynamic Behavior of Concrete and Seismic Engineering

[HAR 90] HARSH S., SHEN Z., DARWIN D., “Strain-rate sensitive behavior of cement
paste and mortar in compression”, ACI Mat. Journal, Vol. 87 N° 5, pp. 508–516, 1990.
[HOP 14] HOPKINSON B., “A method of measuring the pressure in the deformation of high
explosives by the impact of bullets”, Phil. Trans. Roy. Soc., A213, pp. 437–452, 1914.
[HOR 87] HORDIJK D.A., REINHARDT H.W., CORNELISSEN H.A.W., “Fracture
mechanics parameters of concrete from uniaxial tests as influenced by specimen length”,
Proc. of Int. Conf. on the Fracture of Concrete and Rocks, Houston, Texas, USA, Shah
and Swartz (Ed.), pp. 138–149, 1987.
[JAW 87] JAWED I. et al., “High strain rate behavior of hydrated cement pastes”, Cement
and Concrete Research, Vol. 17, pp. 433–440, 1987.
[KAP 80] KAPLAN S.A., “Factors affecting the relationship between rate of loading and
measured compressive strength of concrete”, Magazine of Concrete Research, Vol. 32
N° 111, pp. 79–88, 1980.
[KIP 80] KIPP M.E., GRADY D.E., CHEN E.P., “Strain rate dependent fracture initiation”,
International Journal of Fracture, Vol. 16 N° 5, pp. 471–478, 1980.
[KOL 49] KOLSKY H., “An investigation of the mechanical properties of materials at very
high rates of loading ”, Proc. Phys. Soc., B62, pp. 676–700, 1949.
[KON 01] KONO S., WATANABE F., KAJITANI A., “Stress-strain relation of confined
concrete under dynamic loading”, in Fracture Mechanics of Concrete Structures, de Borst
et al. (Ed.), Swets & Zeitlinger, Lisse, The Netherlands, pp. 585–592, 2001.
[KÖR 88] KÖRMELING H.A., “Experimental results of plain and steel fiber reinforced
concrete under uniaxial impact tensile loading”, Delft University of Technology, The
Netherlands, Report 5-84-8, July 1988.
[KRA 89] KRAUTHAMMER T., FLATHAU W.J., SMITH J.L., BETZ J.F., “Lessons from
explosive tests on RC buried arches”, Journal of Structural Engineering, Vol. 115 N° 4,
pp. 809–825, 1989.
[KRA 93] KRAUTHAMMER T. (Ed.), Structural concrete slabs under impulsive loads,
Proceedings of the Workshop held in Trondheim, Norway, Forsvarets Bygningstjeneste
Forsøkskontoret, Fortifikatorisk Notat Nr 211/93, 14–16 June 1993.
[KRA 96] KRAUTHAMMER T., JENSSEN A., LANGSETH M. (Ed.), Precision testing in
support of computer code validation and verification, Workshop report, Norwegian
Defence Construction Service, Fortifikatorisk Notat Nr 234/96, 1996.
[LAB 96] LABIOUSE V., DESCOEUDRES F., MONTANI S., “Experimental study of rock
sheds impacted by rock blocks”, Structural Engineering International, IABSE (3),
pp. 171–176, 1996.
[LAR 00] LARRARD F. de, “Structures granulaires et formulation des bétons”, Etudes et
recherches des Laboratoires des Ponts et Chaussées, OA34, 2000.
[LOU 94] LOUW J.-M., KOCH B.C., KOEN F.J., “Direct shear strength of concrete under
impact”, in Structures under Shock and Impact III, Madrid, Spain, 1–3 June, Bulson (Ed.),
pp. 147–154, 1994.
 Dynamic Behavior of Concrete: Experimental Aspects 51

[MAL 91] MALVERN L.E., JENKINDS D.A., TANG T., MAC LURE S., “Dynamic testing
of laterally confined concrete”, in Micromechanics of Failure of Quasi-brittle Materials,
Elsevier Applied Science, pp. 343–352, 1991.
[MAL 92] MALVAR L.J., “Bond of reinforcement under controlled confinement”, ACI
Materials Journal, Vol. 89 N° 6, pp. 593–601, November-December 1992.
[MAL 97] MALVAR L.J., “Review of static and dynamic properties of steel reinforcing
bars”, ACI Materials Journal, Vol. 95 N° 5, pp. 609–616, 1997.
[MAL 98] MALVAR L.J., ROSS C.A., “Review of strain rate effects for concrete in tension”,
ACI Materials Journal, Vol. 95 N° 6, pp. 735–739, November-December 1998.
[MAN 67] MANDEL J., “Application de la thermodynamique aux milieux viscoélastiques à
élasticité instantanée nulle ou restreinte”, C.R.A.S, 264, A, pp. 133–134, 1967.
[MAR 87] MARTI J., GOICOLEA J., KUMAR R., ALDERSON M., “Impact analysis of
transport flasks”, Int. Conf. on Nuclear Containment, Cambridge, UK, 6–8 April 1987.
[MIK 95] MIKAMI H., TAMURA T., SATO M., KISHI N., “Shock absorbing performance
of a three-layered cushion system using RC core slab reinforced with AFRP rods”,
Proceedings of the Concrete under Severe conditions CONSEC ’95 Conference, Sapporo,
Japan, Sakai, Banthia, Gjørv (Ed.), E & FN Spon, pp. 1633–1643, 2–4 August 1995.
[MIN 87] MINDESS S., BANTHIA N., YAN C. “The fracture toughness of concrete under
impact loading”, Cement and Concrete Research, 17, pp. 231–241, 1987.
[MIY 91a] MIYAMOTO A. KING M. W., FUJII M. “Non-linear dynamic analysis of
reinforced concrete slabs under impulsive loads ”, ACI Structural Journal, Vol. 88 N° 4,
pp. 411–419, 1991.
[MIY 91b] MIYAMOTO A. KING M. W., FUJII M., “Analysis of failure modes for
reinforced concrete slabs under impulsive loads”, ACI Structural Journal, Vol. 88 N° 5,
pp. 538–545, 1991.
[MON 98] MONTANI STOFFEL S., Sollicitation dynamique de la couverture des galeries de
protection lors de chutes de blocs, Thesis N° 1899, Ecole Polytechnique Fédérale de
Lausanne, Switzerland, 1988.
[MUR 86] MURIA VILA D., Contribution à l’étude du comportement au choc des bétons,
Thesis, INSA de Lyon, 1986.
[NIC 80] NICHOLAS T., “Material behavior at high strain rates”, Mechanical Properties at
High Rates of Strain, pp. 277–328, Institute of Physics, London, 1980.
[PAU 02] PAULTRE P, MITCHELL D., “Seismic response of high-strength concrete
structures”, Proc of 6th Int. Symp. on High Strength/High Performance Concrete, Leipzig,
Germany, König et al. (eds.), Leipzig University, pp. 457-471, June 2002.
[PER 01] PERROTIN P., MOMMESSIN M., MOUGIN J.-P., TONELLO J., “Etude
expérimentale du comportement d’une dalle pare-blocs” , XIXes rencontres universitaires
de Génie Civil, COSS’2001, Lille, pp. 69–75, 3–4 May 2001.
[PON 95] PONTIROLI C., Comportement au souffle des structures en béton armé, PhD
Thesis, Ecole Normale Supérieure de Cachan, 1995.
52 Dynamic Behavior of Concrete and Seismic Engineering

[REI 82] REINHARDT H.W., “Concrete under impact loading. Tensile strength and bond”,
Heron, Vol. 27, N° 3. Delft, The Netherlands, 1982.
[REI 86] REINHARDT H.W., KÖRMELING H.A., ZIELINSKI A.J. “The Split Hopkinson
Bar, a versatile tool for impact testing of concrete”, Materials and Structures, 19, pp. 55–
63, 1986.
[REI 91] REINHARDT H.W., WEERHEIJM J., “Tensile fracture of concrete at high loading
rates taking account of inertia and crack velocity effects”, International Journal of
Fracture, Vol. 51, pp. 31–42, 1991.
[ROS 91] ROSSI P., “Influence of cracking in the presence of free water on the mechanical
behavior of concrete”, Mag. of Conc. Res., Vol. 43, N° 154, p. 53–57, 1991.
[ROS 92a] ROSSI P., WU X., LE MAOU F., BELLOC A., “Effet d’échelle sur le
comportement du béton en traction”, Bull.Liais. Labo. P. et Ch., 182, pp. 11–20, 1992.
[ROS 92b] ROSSI P., VAN MIER J. G. M., BOULAY C., LE MAOU F., “The dynamic
behavior of concrete: influence of free water”, Materials and Structures, Vol. 25,
pp. 509–514, 1992.
[ROS 95] ROSS C.A., TEDESCO J.W., KUENNEN S.T., “Effects of strain rate on concrete
strength”, ACI Materials Journal,.Vol. 92, N° 1, pp. 37–47, 1995.
[ROS 96] ROSS C.A., JEROME D.M., TEDESCO J.W., HUGHES M.L., “Moisture and
strain rate effects on concrete strength”, ACI Mat. J., Vol. 93 N° 3, pp. 293–300, 1996.
[ROS 98] ROSSI P., Les bétons de fibres métalliques, Presses de l’ENPC, Paris, 1998.
[SAT 95a] SATO M., NISHI H., SUGATA N., KISHI N., “Full-scale impact test of PC rock-
shed with shock absorbing system”, Proceedings of the Concrete under Severe conditions
CONSEC ’95 Conference, Sapporo, Japan, Sakai, Banthia and Gjørv (Ed.), E and FN
Spon, pp. 1623–1632, 2–4 August 1995.
[SAT 95b] SATO M., NISHI H., NAKANO O., KISHI N., MATSUOKA K.G., “Full-scale
test on impact resistance of PC girder”, Proceedings of the Concrete under Severe
Conditions CONSEC ’95 Conference, Sapporo (Japan), Sakai, Banthia and Gjørv (Eds.),
E & FN Spon, pp. 1644–1652, 2–4 August 1995.
[SER 98a] SERCOMBE J., “Modélisation du comportement du béton en dynamique rapide.
Application au calcul des conteneurs à haute intégrité”, Etudes et recherches des LPC,
OA 30, 1998.
[SER 98b] SERCOMBE J., ULM F.-J., TOUTLEMONDE F., “Viscous hardening plasticity
for concrete under high rate dynamic loading”, Journal of Engineering Mechanics, ASCE,
Vol. 124 N° 9, pp. 1050–1057, September 1998.
[TED 93] TEDESCO J.W., ROSS C.A., KUENNEN S.T., “Experimental and numerical
analysis of high strain rate splitting tensile tests”, ACI Materials Journal, Vol. 90 N° 2,
pp. 162–169, 1993.
[TOR 88] TORO R. del, Comportement des nœuds d’ossature en béton armé sous
sollicitations alternées, Thesis, Ecole Nationale des Ponts et Chaussées, Paris, 1988.
 Dynamic Behavior of Concrete: Experimental Aspects 53

[TOU 93] TOUTLEMONDE F., BOULAY C., GOURRAUD C., “Shock-tube tests of
concrete slabs”, Materials and Structures, Vol. 26, pp. 38–42, 1993.
[TOU 95a] TOUTLEMONDE F., Résistance au choc des structures en béton. Du
comportement du matériau au calcul des ouvrages, Research report of the LCPC, July
1995.
[TOU 95b] TOUTLEMONDE F., PIERMATTEI E., ULM F. J., ROSSI P., “Modélisation du
béton en dynamique rapide. Résultats d’une première expérience à l’aide d’un modèle
d’endommagement avec gradient”, Bulletin de liaison des Laboratoires des Ponts et
Chaussées, 198, p. 39–52, July–August, 1995.
[TOU 99a] TOUTLEMONDE F., ROSSI P., Discussion of the 95-M73 article published in
the November-December 1998 volume of ACI Materials Journal, p. 735 “Review of
strain rate effects for concrete in tension”, L.J. Malvar and C.A. Ross, published
discussion in ACI Materials Journal, Vol. 96 N° 5, p. 614–615, September-October,
1999.
[TOU 99b] TOUTLEMONDE F., SERCOMBE J., TORRENTI J.-M., ADELINE R.,
“Développement d’un conteneur pour l’entreposage de déchets nucléaires : résistance au
choc”, Revue Française de Génie Civil, vol. 3, n° 7–8, p. 729–756, December 1999.
[TRO 01] TROUILLET P., “Truck impacts on French toll-motorways bridge’s piers”, IABSE
conference Safety, Risk, Reliability – Trends in Engineering, Malta, 2001.
[UNO 02] UNOSSON M., Constitutive equations for concrete materials subjected to high rate
of loading, Linköping studies in science and technology, Thesis N° 936, Linköpings
[VEC 88] VECCHIO F.J., SATO J.A., “Drop, fire and thermal testing of a concrete nuclear
fuel container”, ACI Structural Journal, pp. 374–383, July–August 1988.
[VOS 82] VOS E., REINHARDT H.W., “Influence of loading rate on bond behavior of
reinforcing steel and prestressing strands”, Materials and Structures, Vol. 15 N° 85,
pp. 3–10, 1982.
[WAT 02] WATSON A.J., “Loading from explosions and impact”, in Dynamic Loading and
Design of Structures, A.J. Kappos (Ed.), Spon Press, 2002.
[WEE 88] WEERHEIJM J., REINHARDT H.W., “Biaxial loading for dynamic concrete
response”. J. de Physique, Coll. C3 (Suppl with No.9, Vol.49), pp. 733–738, 1988.
[WEE 89] WEERHEIJM J., REINHARDT H.W., “Structural response in impact tests on
concrete”, 4th Int. Symp. on the Interaction of Non-Nuclear Munitions with Structures,
Panama City Beach, Florida, USA, pp. 17–21, 1989.
[WEE 92] WEERHEIJM J., Concrete under impact tensile loading and lateral compression,
PhD Thesis, Delft University of Technology, November, 1992.
[WEE 98] WEERHEIJM J., “Prediction of dynamic tensile strength”, in ACI SP-175
Concrete and Blast Effects, W. Bounds (Ed.), paper SP 175-14, pp. 215–239, 1998.
54 Dynamic Behavior of Concrete and Seismic Engineering

[ZHA 96] ZHAO H., GARY G., “On the use of SHPB technique to determine the dynamic
behavior of the materials in the range of small strains ”. Int. J. Solid. & Structure, vol. 33,
pp. 3363–3375, 1996.
[ZIE 82] ZIELINSKI A.J., REINHARDT H.W., “Stress-strain behaviour of concrete and
mortar at high rates of tensile loading”, Cem. Conc. Res., vol. 12, pp. 309–319, 1982.
[ZUK 82] ZUKAS J.A., Impact Dynamics, John Wiley & Sons, 1982.
 Chapter 2

Dynamic Behavior of Concrete:

Constitutive Models

2.1. Dynamics of concrete structures

2.1.1. Macroscopic phenomena

We can distinguish two structural response levels to a dynamic load: the local
level and the overall level.

2.1.1.1. Local effect of an impact

This is the stress and strain state created inside the impact area. This state
develops simultaneously with loading and for very high strain rates. Studying the
local effect requires taking the inertia effects, wave strain propagation and tri-axial
stress states into account. Whenever failure criteria are reached during this phase,
scabbing and shear phenomena with chip projections will take place.

2.1.1.2. Overall effect on a structure

If loading is globally distributed on the structure, or if, in the case of an impact,
local failure conditions have not been fulfilled, then the whole structure responds
dynamically. The inertia effects of the structure are such that the reaction
immediately follows loading. If loading results from an explosion or a soft shock,

Chapter written by Patrice BAILLY.

56 Dynamic Behavior of Concrete and Seismic Engineering

the response is initially of transient type, then a quasi-stationary vibration regime

develops if the structure stress is exerted for a period of time that is long relative to
its natural frequencies, as is the case during an earthquake. The strain rates reached
are lower than those observed at the location of an impact. The behavior of the
structure can be described by the strength of materials theory, i.e. the structure can
be modeled by means of plates and beams subjected to traction, compression or
bending. However, the strain modes can be completely different to those observed
under static loading.

2.1.1.3. Analyses of the failure elementary phenomena

We consider a reinforced concrete wall receiving a dynamic loading on one side
(Figure 2.1). The observed phenomena are the following:
– concrete compaction near the loaded side;
– scabbing on the non-loaded side;
– failure of the structure under shear or bending.

Here we describe the mechanisms separately; actually, they are linked and
simultaneous.

Shear
Bending

Compression
Compaction Scabbing

Figure 2.1. Fracture mechanism

2.1.1.4. Compaction of the loaded side

In the first microseconds and in the case of high-intensity loading (impact from a
solid or strong pressure), the loaded side is subjected to high pressure and the
superficial concrete is reduced to powder. A volume reduction caused by the porous
structure of the material collapsing occurs. The stress state is close to that of a uni-
axial strain and to approximately model that behavior, we resort to the so-called
 Dynamic Behavior of Concrete: Constitutive Models 57

“plate-plate” test [LEV 96], which gives the pressure level necessary for dynamic
loading compaction.

2.1.1.5. Scabbing caused by wave reflection on the non-loaded side

Scabbing of reinforced concrete elements results from fracture due to traction
perpendicular to their free surface. This phenomenon is linked to wave propagation
inside the material. Typically, for a 20 cm thick concrete wall, the wave arriving
time onto the “free open surface” is about 0.6 Ps. From experience, we can notice
that the surface ejected by direct scabbing corresponds to the cover between the last
layer of the reinforcement and the free surface.

2.1.1.6. Bending of the structure

The wall’s bending phenomenon is slower, as it depends on the structure more
than on the local behavior of the material. It important to note that shear failure of
slabs due to shear loads are possible in dynamics, whereas it is extremely rare under
static loading.

2.1.1.7. Influence of the reinforcements

Reinforced concrete reinforcements have an effect on the development of macro-
cracks. The cracks, which separate blocks during failure, preferably meet with the
reinforcements (Figure 2.2). Within structures specially designed to withstand
impacts, so-called “lacing” reinforcements limit chip projections.

Figure 2.2. Shape of fragments on a plate submitted to a shock tube (from [PON 95])

2.1.2. Perforation

Perforation of a concrete wall generally reveals three areas (Figure 2.3):

58 Dynamic Behavior of Concrete and Seismic Engineering

– crater formation; this is the extremely damaged zone with fragment ejection on
the impacted side;
– “tunneling”; the projectile progresses through the wall while causing big shear
strains and compaction inside concrete;
– scabbing; at the output of the projectile, traction stresses lead to an ejection of
fragments.

40
crater

20 tunneling
Epaisseur (cm)

scabbing
-20

-40
0 20 40 60 80
S ection 270° - 90° (cm )

Figure 2.3. Concrete slab perforated by a projectile (from [BUZ 98])

2.1.3. Ejection of fragments

When the dynamic loading causes projections of fragments of the concrete

structure, we can classify them into three types.

The scabs are formed due to concrete traction failure according to a process
which will be detailed in section 2.3. If we assume a perfect brittle failure, we can
estimate the ejection speed from equation [2.1], which also gives usual values for
concrete:

V r 10 MPa ½
°3
Vr U ceVe U 2, 200 kg / m ¾ Ÿ Ve | 1.5 m / s [2.1]
ce 3, 000 m / s °¿
 Dynamic Behavior of Concrete: Constitutive Models 59

The ejection of a block follows bending or shear failure. The order of magnitude
of the ejection speed is closely linked to the speed of the structure element at the
moment of failure, therefore to its transient dynamic response. The formation of
small rubble and dust type fragments is linked to the characteristics of the dynamic
failure of the material. If the failure of the structure is accompanied by a blast wave,
the latter will carry the fragments along. The effect of that blast dragging along is
inversely proportional to the fragment size.

2.1.4. Loading range

2.1.4.1. Pressure level

The main feature of dynamic loading is that it involves very high-pressure levels,
completely out of the typical range of structures under quasi-static loading. The
shock of a motor vehicle against a concrete structure, for example, can generate a
several hundredths of a MPa pressure inside the latter. The consequence on a
material such as concrete is compaction, i.e. an irreversible volume strain. Figure 2.4
shows the behavior of a compacted concrete (the concrete called MB 50 has been
the subject of various studies within the frame of the “GEO” laboratories network
[BAI 99]).

Figure 2.4. Hydrostatic compression tests of MB50 concrete.

Mean stress-volume strain diagrams (from [BUZ 99])
60 Dynamic Behavior of Concrete and Seismic Engineering

2.1.5. Loading path

Dynamic loading induces strains and stresses which are described mainly locally
as waves. The propagation and reflection of the waves lead to stress tensor
components that vary within a wide range. For impacts from a solid, the normal
stress components can range from a tensile value to a compressive value causing
compaction. During perforation and mostly during the “tunneling” phase, shears
reach large absolute values and large rotations of the main stress tensor axes may
take place locally. Thus, the concrete shear behavior with a wide range of pressures
or mean stresses has to be known.

Figure 2.5. Triaxial compression tests on MB50 concrete.

Loading paths in the deviatoric- mean stress plane (from [BUZ 99])
 Dynamic Behavior of Concrete: Constitutive Models 61

Figure 2.6. Triaxial compression tests on MB50 concrete. Strain-longitudinal stress diagrams

Figure 2.5 shows loading paths for concrete tri-axial tests. Figure 2.6 presents the
relationship between axial stress and axial strain. We can note the brittle-ductile
transition for a confinement pressure of about 50 MPa.

2.1.5.1. Strain rate ranges

Is it possible to establish a relationship between a dynamic loading and the strain
rate it would impose on the structure material? Generally not. The strain rate is
meaningful for a model describing the structure’s movement. It is necessary to make
a difference between the scale of the structure and the local scale as they have been
defined above.

At the scale of the structure, the model describes a beam or a slab element under
bending, for instance. The model implies a specific time-scale – that of the period of
the bending mode taken into account. Thus, we implicitly admit that the propagation
time for a distance corresponding to the size of the element is much shorter. This
modeling scale is consistent for seismic calculations and for the overall effect of an
impact on a structure. Within that frame, the strain rates can reach the 1/s order of
magnitude.

On the local scale, we can observe the wave propagation in a continuum. The
strain rate can be quite high (10 to 1,000/s). The strain rate is a function of the strain
level (which depends on the loading) and the wave velocity (which depends on the
62 Dynamic Behavior of Concrete and Seismic Engineering

material). The strain rate notion is interesting if it can be associated with a volume
element where it is almost uniform. The higher the rate, the smaller the
corresponding volume. Therefore, we have to note that in continuum mechanics, a
very high strain rate can be incompatible with a given representative elementary
volume, especially for a material like concrete.

The notion of strain rate becomes meaningless if we consider a shock wave

(finite variation of the strain amplitude for an infinitesimal period of time).

2.2. Fast dynamics applied to concrete

This chapter presents the fundamental concepts that are useful to study a
structure subject to dynamic loading. Many studies and applications have been made
for metals, but here, the emphasis is placed on the specificities of applications to
concrete.

2.2.1. Impacts and waves

2.2.1.1. Uniaxial strain state

In a uniaxial strain state, which can correspond to certain impact conditions, the
stress and strain tensors have the following forms:

§ H1 0 0 · § V1 0 0 ·
¨ ¸ ¨ ¸
H ¨ 0 0 0¸ V ¨ 0 V2 0 ¸ [2.2]
¨ 0 0 0¸ ¨0 0 V ¸
© ¹ © 2¹

To study such a strain state, the most commonly used hypothesis is that the
material can be described by means of perfect elasto-plastic behavior and that the
plastic flow is incompressible. Thus, the partition between elastic and plastic strains
is made:

H H e  H p œ H ie  H ip 0 i 2,3 [2.3]

and:

traceH p 0 œ H1p  2H 2p 0 [2.4]

 Dynamic Behavior of Concrete: Constitutive Models 63

If the plasticity criterion is of the Tresca or von Mises type:

V1  V 2 Y0 [2.5]

In the elastic phase the relationships between stresses and strains are as follows:

E 1  v 
V1  O  2 P  H1
1  2v 1  v 
H1  K  43 G  H1
and [2.6]
v
V2 V1
1 v

At the plasticity threshold:

§ v · 1 v § K 2·
V1  V 2 ¨1  1  v ¸ V1 Y0 and V1
1  2v
Y0 ¨ 2G  3 ¸ Y0 [2.7]
© ¹ © ¹

This elastic limit in a uniaxial strain state is also called Hugoniot’s elastic limit
(HEL). Beyond the flow threshold:

2
V1 K H1  Y0 [2.8]
3

The observable behavior is shown in Figure 2.7, together with the hydrostatic
strain behavior and the uniaxial strain state behavior (simple compression).
64 Dynamic Behavior of Concrete and Seismic Engineering

V1 c
b

(K/2G + 2/3)Y0
K

Y0 a

2/3 Y0
H1
E

Figure 2.7. Axial stress-axial strain relationship: a) uniaxial stress state (simple
compression; b) isotropic or hydrostatic state; c) uniaxial strain state (oedometric)

Beyond the elasticity threshold, we can see that the uniaxial strain behavior
curve is parallel to the hydrostatic strain state behavior. By extrapolation, it is often
admitted that the behavior obtained in a uniaxial strain state, apart from the elasticity
threshold, is identical to the relationship between the pressure and the volume
variation in a hydrostatic case. However, we must not forget that this result relies on
such hypotheses as the von Mises criterion and plastic strain incompressibility.

Moreover, we can note the velocity of an “elastic” wave:

K  43 G
ce [2.9]
U

This conventional chain of reasoning can be extended to a Drucker Prager (or

Mohr Coulomb) criterion, which is more realistic with a material such as concrete.
The conclusions will then be different. The plasticity criterion is written the
following way:

V1  V 2 A  B V 1  V 2  [2.10]

Under oedometric loading, the plastification threshold (HEL) appears for the
stress level as:
 Dynamic Behavior of Concrete: Constitutive Models 65

A 1  v 
V1 [2.11]
1  2v  B

Note that this threshold only exists if A ! 0 and B 1 – 2v. Beyond that
threshold, the lateral stress equals:

1 B A
V2 V1  [2.12]
1 B 1 B

If we retain the hypothesis of an incompressible plastic flow, which in that case

corresponds to a non-associated flow:

3K 1  B  2 A 1  B 
V1 H1  [2.13]
3 B 1  B  3  B 

In this case, we can observe that the conclusion considering the von Mises
criterion established above is no longer valid. The volume-pressure relationship
cannot be derived through a simple translation from the oedometric path strain-stress
relationship. Nevertheless, if the deviatoric behavior is well known, the deduction is
still possible with the hypothesis of an incompressible plastic flow.

Here we have been considering perfect plasticity models. Yet, whenever there is
strain hardening, the plastic flow is either dilating or contracting, and although it is
not impossible, expressing the volume-pressure relationship from the strain-stress
relationship can be quite difficult.

2.2.1.2. Creation of a shock wave

A shock wave will be created during an impact if the impact speed is sufficient
for the stress and strain levels to be located within the “positive curvature” part of
the axial stress–axial strain curve. The remark means that a shock wave is possible
in uniaxial strain, but is impossible in uniaxial stress. The stress level is located in
the upper part of Hugoniot’s curve, as shown in Figure 2.8. In such conditions, a
shock wave is created with a velocity of:

R
v [2.14]
U
66 Dynamic Behavior of Concrete and Seismic Engineering

R is the slope of the straight line linking the material’s representative points
before and after the shock. We notice that there may be two different situations; if
the shock is not too violent (Figure 2.8(i)), the shock wave will be preceded by an
elastic precursor, as the shock front velocity is lower than that of the elastic waves.
If the shock is very violent (Figure 2.8(ii)), its velocity can be higher than that of the
elastic waves, and there is no longer a precursor.

V1
(ii)

(i)
H1

Figure 2.8. Strain-stress relationship and stress levels involving a shock wave,
with or without precursor (i and ii respectively)

2.2.2. Impact and shock polar curve

2.2.2.1. Conservation equations

A shock wave is a non-stationary speed discontinuity that is necessarily
associated with a stress discontinuity. D is the propagation speed of the discontinuity
surface. On a specific point of the discontinuity surface (of normal line n), the mass
and momentum conservation equations take on a particular shape (the brackets
represent the jump of the considered value when passing the discontinuity surface):

Mass conservation a U  v  D bn 0 [2.15]

Momentum conservation aV b n U aV b V  D  n [2.16]

It is interesting to write these equations in the case of a one-dimensional medium

that is at rest after the shock (Figure 2.9). The equations are:
 Dynamic Behavior of Concrete: Constitutive Models 67

Mass conservation U  D  V  U0 D [2.17]

Momentum conservation V U0 DV [2.18]

U V V D U0 V 0 V=0

Figure 2.9. Shock wave propagating with a velocity D within an initially at rest medium

The strain after the shock can then be expressed (the compressions are positive):

U V
H 1 H [2.19]
U0 D

Figure 2.10. Shock polar curve for an elastic material

It is interesting to express the stress level reached as a function of the material

speed of the solid (the corresponding diagram is often called shock polar curve). For
an elastic material, the result is a straight line (Figure 2.10).

2.2.3. Shock between two solids

During a shock between two solids, waves are generated within each of them.
The shock polar curve is a convenient way to determine the stress level associated
with those waves. For instance, let us consider the shock of a metal projectile on a
68 Dynamic Behavior of Concrete and Seismic Engineering

concrete element (Figure 2.11). Experience shows that it is reasonable to do a

representation in a uniaxial strain state. The shock polar curves (Figure 2.12) show
that the interface stress necessary equilibrium at the time of the shock can only be
reached for values V3 and V3.

Figure 2.11. Metal projectile hurled at speed V2 against a concrete element;

image before and after the impact

Figure 2.12. Shock polar curve diagram

(full line for concrete and dotted line for metal)

2.3. Scabbing

The scabbing failure phenomenon occurs with materials whose tensile strength is
definitely lower than their compression strength, like concrete and rocks. Dynamic
loading, such as on impact, generates a compression wave like that seen previously.
The compression duration corresponds to a propagation and reflection movement of
the waves inside the projectile. Then the stress level is back to zero and an unloading
 Dynamic Behavior of Concrete: Constitutive Models 69

stress wave propagates throughout the material. Besides, when the compression
wave reaches a stress-free side, an unloading stress wave propagating from the
opposite direction is also emitted to respect the zero-stress condition. These two
unloading waves meeting cause traction of the same intensity as the compression of
the initial wave. It is easy to understand that such an intensity, which can be borne in
compression leads to failure when it is exerted in traction (Figure 2.13). A strain
wave is then reflected on a free end by changing signs.

localization of failure in traction

Figure 2.13. Lagrange’s diagram representing the reflection

of a strain increment on a free side

2.4. Effect of a shock wave on the structure of materials

A shock wave passing through a material can induce irreversible modifications

to its characteristics. The most obvious variation involves the relative density – this
is called compaction. For a dusty material, such as a highly damaged concrete, a
model example of the relationship between pressure and the relative density can be
the following:

1
§ 1 a ·
U U0 ¨ a  P [2.20]
© 1  b ¸¹
70 Dynamic Behavior of Concrete and Seismic Engineering

where a defines the initial density and b is a flexibility coefficient.

Figure 2.14 represents the pressure-volume relationship. In high intensity

compaction, the powder becomes a solid. When a shock wave passes through, there
is a pressure jump up to the value P. Due to the dynamic and viscous effects, the
evolution follows the so-called Rayleigh line. Unloading occurs in parallel to the
solid behavior curve (there is no decompaction).

Figure 2.14. Compacting caused by a shock wave passing through a dusty material

The expression of the dissipated energy is the following:

1 § U ·
W PV0 ¨1  0 ¸ [2.21]
2 © U ¹

2.5. Modeling types

2.5.1. Behavior description theoretical frames

To describe the dynamic behavior of concrete, engineers still mostly use

continuum mechanics. Then the resolution of structure problems is achieved by
means of the finite element method. Therefore, behavior models or constitutive
models are developed within that frame.

Taking the previous remarks about loading paths into account, we are led to
consider three model types (Figure 2.15):
– damage mechanics, which aims to describe the stiffness loss due to cracking;
 Dynamic Behavior of Concrete: Constitutive Models 71

– plasticity or viscoplasticity theory, which describes sliding and shear;

– pressure-volume relationships or hydrodynamic state laws for compaction.

Figure 2.15. Diagram of the elastic domain for concrete and of the different irreversible
phenomena in the plane of the first two invariants of the stress tensor

2.5.2. Integrating sensitivity to the strain rate

2.5.2.1. Models with time dependence

If we want to integrate the observed sensitivity to strain rate into the behavior
model, the latter can intervene within the framework of conventional models.

Linear or non-linear visco-elasticity will reveal an apparent stiffness increase

with the strain rate. If failure modeling is of the perfect brittle type and if the failure
criterion is expressed as maximum strain, we will find that strength increases with
the strain rate.

Visco-plasticity will show up as an apparent strength increase, in terms of the

maximum stress reached rising with the strain rate.

Rate effects can also be introduced within the evolution of damage.

It is necessary to justify the introduction of the dependence on rate. Two reasons

justify the use such a process:
– a physical argument, because we wish to use the model to describe a real event
that has been observed experimentally. We must refer to Chapter 1 and see how
cautiously these apparent dynamic strength increases must be interpreted;
– a theoretical argument, because dependence on rate within a viscoplastic or
“viscous-damage” framework allows us to preserve a well-posed structure
calculation problem, even in the presence of softening.
72 Dynamic Behavior of Concrete and Seismic Engineering

2.5.2.2. Phenomenological law

The apparent strength of numerous materials depends on the loading rate used to
test it. The experimental results related to strength in a uniaxial stress state are
generally represented in a diagram of strength versus strain rate on a logarithmic
scale. For some rate ranges, strength varies more or less linearly with the strain rate
logarithm. The observation has led to empirical modeling:

EH
V K  H  [2.22]

By analogy, models have been designed for application to concrete, with the
intention of deriving the strain rate effect from tests carried out at different rates.
The main problem with this approach, apart from the precautions to be taken when
interpreting the results of the tests, involves converting to a tensorial formulation. A
rather simplistic approach in this field could lead to models that have dubious
thermodynamic acceptability (dissipation not necessarily demonstrated outside the
radial loading paths).

2.5.3. Elasto-plasticity and criteria

Most of the models have been built on an elastoplastic theoretical framework. If

the plasticity is perfect, the plasticity criterion corresponds to the failure criterion,
but many models differentiate the elastic limit criterion from the failure criterion.
The loading surface evolves from the first to the second with the evolution of a
strain-hardening variable.

These criteria are expressed in terms of the stress state and the strain-hardening
variables:


f V , ki  0 [2.23]

The material independence leads us to express the criterion as a function of the

three stress invariants:
– the first invariant (or pressure);

I1 trace V 3 p [2.24]
 Dynamic Behavior of Concrete: Constitutive Models 73

– the second invariant, the von Mises equivalent stress;

3
V eq J2 V  2
s:s [2.25]

– the third invariant, i.e. Lode’s angle;

1
1
1 1 3 3J 3
J3
3
trace V 3  3

3

V ijV jk V ki  3
cos  3T  3
2J22
[2.26]

The criteria are generally isotropic, but they express asymmetry between traction
and compression. Among the criteria used for concrete are Coulomb’s and Drucker
Prager’s criteria [DRU 52].

1 § 3·
 
f V
2

s : s  D trace V  Yh  ¨¨ 0  D  ¸
2 ¸¹
[2.27]
©

This criterion is now used as a reference for the shear failure of geomaterials.
More elaborate criteria, taking the third invariant into account, for example, are used
for other models. We cite those defined by Ottosen, Argyris, and Willam Warnke as
examples.

Figure 2.16. Representation of a “cap” surface

The criteria linked to shear failure such as those below, do not express high-
pressure irreversible phenomena. In the plane of the first two invariants, the
representation of the criterion gives an “open” curve on the pressure axis. The
eponymous “cap” surface closes the elastic domain (Figure 2.16); it is associated
with compaction.
74 Dynamic Behavior of Concrete and Seismic Engineering

2.5.4. Damage

The subject of damage mechanics seeks to represent the irreversible micro

cracking and stiffness-degradation processes which occur in some materials. The
threshold above which this phenomenon appears can be represented by one criterion,
as in the case of plasticity. The criterion can be expressed in terms of strains instead
of stresses, which is not the same in dynamics in the presence of viscosity. It can
also be asymmetric, i.e. a function of extensions and not compressions. For example,
Mazars’ criterion [MAZ 84], which has been used for concrete, is a function of the
principal strains if they are positive. The most widely used model is the scalar
isotropic damage model [LEM 85]. The behavior of damaged real material is
expressed by the law used for the undamaged material, except the usual stress is
replaced by the effective stress.

V
H [2.28]
E (1  D)

There are also damage models that depend on time.


D H , D,"  [2.29]

This type of formulation derives its justification from the fact that the creation
and propagation of cracks are phenomena that cannot be instantaneous [SUA 84].

2.5.5. Notion of a state law

A common hypothesis on which to build up a material behavior model involves

splitting the stress tensor and the strain tensor into their spherical and deviatoric
parts.

V  p I  s D and H T I  ec [2.30]

Then we can define separately the relationship between the spherical parts
(pressure and volume variation) called isotropic behavior or state law by abusing the
language, and the relationship between the deviatoric parts (which represents the
deviatoric behavior). This approach has proved quite interesting in the case of
metals, as both relationships are then independent, and two independent experiments
 Dynamic Behavior of Concrete: Constitutive Models 75

allow us to identify a behavior model. For a cohesive material like concrete, the
irreversible volume and deviatoric phenomena are coupled. Nevertheless, taking the
techniques developed for metal materials into account, the same partition has been
carried out in certain models for use with concrete, and the technique is then
enriched using coupling.

In the case of very high intensity dynamic loadings such as those generated by
solid explosives, it is accepted that interior loads can be represented by a pressure
scalar field, as in a perfect fluid. This means that the order of magnitude of the shear
stress is smaller than those of normal stresses: in this case, we resort to so-called
“hydrodynamic” codes.

Within this framework, concrete is treated as a porous material, and the state law
involves the relationship between pressure and porosity, with potentially other
thermodynamic variables such as temperature and phase states being used.

U
P  P , T , " P 1 [2.31]
U0

The relationship between pressure and density is often expressed as a

polynomial. In calculation codes, an example of such a law is “P-D model” [HER
69], D being the porosity index.

This model type is generally derived from plate impact tests. During these tests a
shock wave passes through a sample of the material and the shock velocity D and
the material speed u are measured [MCQ 70]. It seeks the coefficients of a law of the
type:

D A  Bu n [2.32]

The relationship between pressure and volume variation can be deduced

according to the hypotheses in section 2.2.1.

2.5.6. Location limiter and time sensitivity

Softening, which is also known as “cohesion loss” or “negative strain-

hardening”, is a characteristic shown by concrete in its post-failure behavior. In a
transient dynamic calculation, having a good description of that phase is important,
for, even if the concerned area is small, the global reaction of a structure is quite
76 Dynamic Behavior of Concrete and Seismic Engineering

strongly dependent. Location limiters (either explicitly or implicitly) induce the

dependence of behavior on the strain rate.

For instance, in visco-plasticity, owing to a study of wave dispersion, we can

establish a characteristic length (K is the coefficient of viscosity and ce the velocity
of elastic waves):

2K ce
Ac [2.33]
E

We can then wonder if this time-dependence can be mixed up with the

rheological aspect of the behavior. As a matter of fact, both aspects are separated by
the values of the viscosities in play. In fact, the value of a characteristic length is
centimetric [PIJ 87] and therefore must be related to a specific coefficient of
viscosity.

A c | 102 m l K | 104 Pa.s [2.34]

Such a coefficient of viscosity value (taken as a visco-plasticity coefficient in

Perzyna’s model) is without any noteworthy effect on the macroscopic behavior (for
example, the apparent strength as a function of stress/speed). If we are to apply a
coefficient of viscosity to express a rheological aspect, like the dependence of
compaction on the strain rate [GAR 98b], We will have to introduce a coefficient
about 108 Pa.s.

2.6. Models

An illustrative selection of models that are used in fast dynamics codes to

represent concrete behavior are presented below. This collection is obviously
restricted and does not pretend to be exhaustive. It is not a critical or comparative
study of the models. For each model, a reference is provided, affording the reader
the opportunity to find details related to each, as explained by its author. The “basic”
models, such as Prager’s model, are not included.

2.6.1. Elasticity-based model

2.6.1.1. Cedolin
– Reference: [CED 77].
 Dynamic Behavior of Concrete: Constitutive Models 77

– Principle: non-linear elasticity, spherical and deviatoric parts separated.

– Application field: low pressure level.
– Can model: sliding.
– Cannot model: negative strain-hardening, compaction, damage.
– Triaxial stress state:

V1  60 MPa; V1 ! V 2 ! V 3 [2.35]

– Strain-stress relationship:

V oct 3K s H oct and W oct Gs J oct [2.36]

1
H oct  H1  H 2  H 3  [2.37]
3

and

2ª 1/2
2 2 2º
J oct  H 1  H 2    H 2  H 3    H 3  H 1  [2.38]
3 ¬« ¼»

Ks K s  H oct  ˜ Gs Gs  J oct  [2.39]

2.6.2. Models based on the theory of plasticity

2.6.2.1. Elwi, Murray

– Reference: [ELW 79].
– Principle: elasto-plasticity.
– Application field: low pressure.
– Can model: shear.
– Cannot model: compaction and strain-hardening.
– Model data: Argyris failure surface (5-parameter surface, not closed on the
hydrostatic axis).
78 Dynamic Behavior of Concrete and Seismic Engineering

2.6.2.2. “CONC”
– Reference: model implemented into “Radioss” code by “Mecalog”.
– Principle: elasto-plasticity with non-associated flow.
– Application field: monotonous-dynamic loading.
– Can model: shear and dilatancy.
– Cannot model: negative strain hardening.
– Model data: failure surface (Ottosen type).

f V m , r , T  0,

with:

I1 3 3J3
Vm ,r 2 J 2 , cos(3T ) [2.40]
3 2 J 23/2

The loading surface contained between the initial elastic surface and the failure
surface.

relasticity 
r V m , T  k k y , V m  [2.41]

According to the value of k we can represent different behaviors of concrete,

including the traction region, transition region and compression region.

2.6.2.3. Bicanic, Zienkiewic

– Reference: [BIC 83].
– Principle: elasto-viscoplasticity.
– Application field: cyclic loading.
– Can model: damage.
– Cannot model: compaction.
– Model data: uses two surfaces (Mohr/Coulomb):
- the threshold surface, FD, limit between the elastic zone and the plastic zone;
- the stress limit surface FF.
 Dynamic Behavior of Concrete: Constitutive Models 79

– FD and FF vary with accumulated visco-plastic work;


FD V , W p , k  
0 FF V , W p  0 k W p  W p [2.42]

where Wp = viscoplastic work, Wp* = energy dissipated at time t*’, t* = time when
FF is reached, and where k controls energy dissipation and represents the
deterioration of threshold surface;

2.6.2.4. Han and Chen

– Reference: [HAN 85].
– Principle: non-associated plasticity;.
– Application field: high pressure.
– Can model: elasto-plastic phase, strain hardening.
– Cannot model: damage.
– Model data: 4-parameter failure surface (Ottosen, Hsieh-Ting-Chen), and 5-
parameter elastic threshold surface (Willam-Warnke).

Figure 2.17. Elastic threshold surface and failure surface of

the Han and Chen model (from [HAN 85])

The shape of the failure surface is different from the shape of the elasticity initial
surface. As is the case with Ottosen, the equation of the loading surface is:

f r  kr f 0 [2.43]
80 Dynamic Behavior of Concrete and Seismic Engineering

where k is a factor the form of which is k k V m , k0  .

2.6.2.5. Pietruszczak
– Reference: [PIE 88].
– Principle: non-associated plasticity.

Figure 2.18. Pietruszczak failure surface, (a) space of the main stresses;
(b) meridian plane; (c) deviatoric plane (from [PIE 88])

– Application field: low or medium confinement pressure, monotonous loading.

– Can model: the sensitivity of the material to confinement pressure, contraction-
dilatancy transition, brittle-ductile transition and strain hardening.
– Cannot model: compaction, unloading.
– Model data: failure surface.

2.6.2.6. Sandler et al. (CAP)

– Reference [SAN 76].
– Principle: associated plasticity.
– Application field: behavior under impact, high pressures.
– Can model: compaction.
– Cannot model: damage.
– Model data: shear failure criterion and “cap” surface.
 Dynamic Behavior of Concrete: Constitutive Models 81

2.6.3. Models based on damage mechanics

2.6.3.1. Mazars
– Reference: [MAZ 84].
– Principle: isotropic damage (explicit formulation).
– Application field: low pressures (structures, beams), traction.
– Can model: damage, negative strain hardening.
– Cannot model: compaction, sliding.
– Model data: coefficients of elasticity, threshold strain and parameters of
calculation of D:

- behavior law: V 1  D  ª¬O  trace H  ‚  2PH º¼ ; [2.44]

2 2 2
- equivalent strain: H ¢H 1 ²  ¢H 2 ²  ¢H 3 ² . [2.45]
  

If H D 0  H there is damage, and the threshold surface is written:

f  H , D  H  H D 0 0 [2.46]

The variable D integrates traction damage as well as compression damage:

D D t Dt  D c Dc [2.47]

2.6.3.2. La Borderie
– Reference: [LAB 91].
– Principle: scalar damaging (implicit formulation).
– Application field: cyclic loading, structures.
– Can model: stiffness loss (deterioration under traction stresses, remnant strains,
stiffness recovery).
– Cannot model: compaction.
82 Dynamic Behavior of Concrete and Seismic Engineering

– Model data: decomposition of the stress tensor, within the principal axis, in a
positive part and in a negative part, to take into account the independence of
stiffness in traction and compression. Two damage variables are defined by:

V V v E1 D1 E 2 D2
H
E0 1  D1 

E0 1  D2 

E0
   
V  traceV ‚ 
E0 1  D1 
 
fc V 
E0 1  D2 
[2.48]

where D1 and D2 are the damage variables, the other parameters being characteristics
of the material.

2.6.3.3. Dubé
– Reference: [DUB 94].
– Principle: visco-damage (implicit formulation).
– Application field: cyclic loading, dynamic problem.
– Can model: stiffness loss (= deterioration under traction stress), residual
strains, stiffness recovery and rate phenomena.
– Cannot model: compaction.
– Model data: the formulation is close to La Borderie’s model, but it introduces,
in the same way as Perzyna, time dependence in the evolution of the damage
variables (f being the damage threshold function)

n
§¢f²·
D ¨ m ¸ [2.49]
© ¹

2.6.4. Model coupling damage and plasticity

2.6.4.1. Ulm
– Reference: [ULM 93].
– Principle: associated elastoplastic model + damage (implicit formulation).
– Application field: cyclic loadings, low-rate dynamic and monotonous loadings.
– Can model: monotonous loadings, contracting plastic behavior for high
hydrostatic pressures, dilating plastic behavior for low hydrostatic pressures.
 Dynamic Behavior of Concrete: Constitutive Models 83

– Model data: criterion of plasticity (4-parameter modified Willam-Warnke’s

criterion). Coupling takes place in the following way:
- the coefficients of elasticity (K and G) evolve with the inelastic strain;
- a strain-hardening parameter depends on cracking.

Figure 2.19. Trace of Willam Warnke’s criterion in

the deviatoric plane (from [ULM 93])

2.6.5. Model coupling damage and mechanics of porous media

2.6.5.1. Burlion
– Reference: [BUR 97].
– Principle: porous media [COU 91], visco-plasticity and 1-parameter scalar
damage.
– Application field: high-pressure dynamic loadings.
– Can model: compaction, damage.
– Cannot model: sliding (explicitly).
– Model data: the threshold surface of elasticity used is adapted from Gurson’s
criterion [GUR 77] initially used for some metals.
84 Dynamic Behavior of Concrete and Seismic Engineering

3J 2 §q I ·

F V m ,V , f  V m2
 2q1 f cosh ¨ 2 1
© 2V m

¸  1   q3 f 
¹
2
 [2.50]

The coefficients q and f characterize the material.

Coupling results from the fact that the damage scalar variable is linked to
porosity. At the initial state, the damage variable is not equal to zero, and can
decrease in the event of compaction.

Figure 2.20. Representation of Gurson’s criterion in the plane

of the first two invariants (from [BUR 97])

2.6.6. Model deriving from a hydrodynamic approach

2.6.6.1. “ARMOR”
– Reference: [MAR 94].
– Principle: spherical and deviatoric separation of parts, perfect plasticity and
damage.
– Application field: concretes and rocks under high pressure.
– Can model: compaction.
– Cannot model: negative strain hardening.
– Model data:
i) the volume mechanism is represented by a relationship between pressure and
volume variation:
 Dynamic Behavior of Concrete: Constitutive Models 85

1 U
p pP  p tr V  P = 1 [2.51]
3 U0

ii) the deviatoric mechanism is defined by a threshold of plasticity in the plane

(p,q)

3 2
q2 sij 3J 2 [2.52]
2

The material’s threshold curve depends on a damage variable D:

S S s 1  D   S f  D  [2.53]

where SS = threshold for the undamaged material and Sf = threshold for the damaged
material (Figure 2.8).

Figure 2.21. “ARMOR” model, elastic field in the density-pressure plane

and flow threshold surface in the p-q plane (from [MAR 94])

2.6.6.2. Holmquist
– Reference: [HOL 93].
– Principle: spherical and deviatoric separation of parts, perfect plasticity and
damage.
– Application fields: high stresses, high strain rates and high pressure.
– Can model: compaction, damage.
– Cannot model: negative strain hardening.
86 Dynamic Behavior of Concrete and Seismic Engineering

– Model data:
i) volume-pressure relationship, state equation;

U
elastic field: P K elastic P with P 1 [2.54]
U0

transition field: P ª¬1  F  K elastic  F K1 º¼ P [2.55]

“The Hugoniot’s upper part”:

P  Plock
P K1P  K 2 P 2  K 3 P 3 with P [2.56]
1  Plock

ii) plasticity threshold f P, İ , D  given by the effective stress:

ª A 1  D   BP N 1  C1nH º V V
V ¬ ¼,
[2.57]
fc

Figure 2.22 illustrates the different terms of the formulation.

 Dynamic Behavior of Concrete: Constitutive Models 87

Figure 2.22. Homlquist’s model, threshold plasticity surface, damage-inelastic strain

relationship, density-pressure relationship (from [HOL 93])

2.6.7. Endochronic models

2.6.7.1. Bazant and Bhat

– Reference: [BAZ 76].
– Principle: similitude between a strain-stress relationship and the evolution of
strain in a viscous body model. The time variable is replaced by a variable
depending on the accumulation of plastic strain.
– Application field: cyclic loadings.
– Can model: plastic dilatancy, strain hardening (positive or negative).
– Cannot model: compaction.
88 Dynamic Behavior of Concrete and Seismic Engineering

2.6.8. Discrete element method

Modeling a medium as an assembly of particles is an alternative to

representation as a continuous medium. The concept has been applied to both fluid
and solid mechanics, and to pulverulent media where there is an identity between the
model’s particle and the grain of the material. It has recently been extended to
cohesive media. Two approaches co-exist in this frame:

– The medium is treated as undeformable particles with a mass (ED). The

stresses and strains are represented by interactions between the particles (distance
and binding force). Thus, behavior, both elastic and inelastic, is completely
expressed by the linking conditions of EDs [CUN 97].
– The medium is considered as if composed of EDs which are deformable solids,
the behavior of which is described through elasticity and plasticity in continuum
mechanics. The links between EDs only express local phenomena, such as contact
loss and friction [MOR 96].

2.6.8.1. Camborde and Mariotti

– Reference: [CAM 99].
– Type: interactions between particles.
– Each particle is representative of a mesoscopic scale consisting of a few
aggregates, cement and empty space. Initially, each link that is of the cohesive type
transmits compression and traction loads. Two types of elastic forces are present
(Figure 2.23): the normal strength of stiffness Kn and the shear force of stiffness Ks.

The inelastic behavior involves a traction failure threshold T on Fn and a Mohr-

Coulomb shear failure threshold on Fs, defined by the cohesion C and a friction
angle I. In a traction failure (Fn < 0), the link is simply suppressed. In a shear failure
(Fn > 0), we change from a cohesive type law to a friction type law (2.24).

'U n=U tn- U (t-1)n

'U s=U ts- U (t-1)s

Kn Ks

Figure 2.23. Details of the particles’ interaction 'U n normal displacement,̓ 'U s
tangential displacement (distortion) ( from [CAM 99])
 Dynamic Behavior of Concrete: Constitutive Models 89

Porosity is introduced not on the geometric level, but through an inner variable
that controls the evolution of stiffness Kn, as is the case with the Holmquist and
Johnson-type model.

Figure 2.24. Domain of the contact forces and state equation (from [CAM 99])

2.6.8.2. Donzé
– Reference: [DON 98].
– Type: interaction between particles.
– One original feature of this model is definition of the interactions between
particles. Two interacting elements are not necessarily in contact. Two elements a
and b of radii Ra and Rb will be in interaction if J (Ra + Rb) ! Da,b, with the
interaction coefficient J (! 1), Da,b being the distance between the centers of EDs.
This definition of interacting elements is different from the one typically used with
spherical EDs, where only the elements in contact interact (J= 1). This definition of
initial interactions allows us to increase the number of interactions for one element,
and therefore to model the action of the matrix between the concrete aggregates. The
interaction force between two EDs has a normal component Fn and a tangential
component Fs. Figure 2.11 shows the cyclic loading response of the bonding
between two EDs. A Mohr-Coulomb criterion is used to characterize the failure of
the bonding between two EDs.
90 Dynamic Behavior of Concrete and Seismic Engineering

Figure 2.25. Relative displacement-normal load answer of a bonding between

two EDs and strength field limit (from [DON 98])

2.6.8.3. “LMGC” type

– Reference: [MOR 96].
– Type: “dynamic contact”.

The directing idea of this model is dealing separately with the elasticity of the
medium and the adhesive contact laws where damage can appear. The laws linking
the elements are of the graph type and describe the contact physical behavior.
Between two elements, two physical principles exist: the existence of contact that is
described using a Signorini condition, and the potential friction between elements.
To this law, a new law is added to build a cohesive medium, namely adherence. Two
media stand out, the micro-medium between discrete elements respecting the
adhesive friction contact laws, and the macro-medium which describes the elastic
continuous medium governed by the first principle of dynamics and the elements’
elastic (or viscoelastic) strain laws.

2.7. Conclusion

2.7.1. Main features of the models

Most models are constructed within an elastoplastic framework, with a criterion

determined by the first stress invariant. Taking the loading field into consideration,
models generally include:

– a plastic flow due to shear;

– a plastic volume strain due to compaction;
– scalar damage due to cracking (usually).
 Dynamic Behavior of Concrete: Constitutive Models 91

Inevitably, whatever the model, the number of the parameters to be identified

will be important, and complete identification will require several experiments. In
practice, completing identification of all the parameters of a model is difficult. It
requires having a concrete typology with the average values corresponding to each
type for each model. Then the user only has to identify the type of concrete using a
few simple tests, and can infer the whole set of corresponding parameters.

2.7.1.1. Taking the dynamic aspect into account

Should sensitivity to strain rate be taken into account in a concrete behavior
model? When considering the models presented, it is easy to see that opinions are
divided. The works of the “GEO” network [BAI 99] tried to answer this question.
Two elements clearly emerge.

If dynamic compression tests (simple or with low confinement) are simulated

using dynamic structure calculation carried out with models that are not sensitive to
the strain rate, the apparent strength increase is quite well-rendered. Actually, the
non-homogenity of the stress state and the “inertial” confinement effect has been
established.

Unlike metal materials, concrete failure is accompanied by great dilatancy,

which induces radial displacements slowed down by inertia. An increase of pressure
within the test structure, and a higher failure threshold will follow [GAR 98a], [LEN
00].

Comparisons of static and dynamic compaction tests have rarely been conducted.
Such comparisons are consistent if the stress path is the same in both cases.
Experiments and simulations performed to that effect obviously show the
importance of the strain rate on the compaction behavior [GAR 98b].

Rather paradoxically then, models integrating sensitivity to the strain rate do so

in relation to the shear induced flow instead of the irreversible volume variation.

2.7.2. Contribution of distinct elements

This modeling type, initially designed for aggregate materials, is used to

represent cohesive materials. Dynamic compression test simulations confirmed the
conclusions of mechanics of continuous media studies with regard to the strain rate
sensitivity.

Simulating structure tests with these methods allows us to describe post-ultimate

phase aspects such as chip projection. This goal is difficult to achieve using the
92 Dynamic Behavior of Concrete and Seismic Engineering

mechanics of continuous media and numerical finite element codes (except in the
case of advanced codes, where finite elements meet EDs).

2.8. Bibliography

[BAI 99] BAILLY P., “Comportement des ouvrages en dynamique rapide, comportement
dynamique du béton”, final scientific report of the research group formed in the “GEO”
laboratories, 1999.
[BAZ 76] BAZANT Z.P., BHAT P.D., “Endochronic theory of inelasticity and failure of
concrete”, Journal of the Engineering Mechanics Division, Vol. 102, N° EM4, August
1976.
[BIC 83] BICANIC N., ZIENKIEWICZ O.C., “Constitutive model for concrete under
dynamic loading”, Earthquake Engineering and Structural dynamics, Vol. 11, pp. 689–
710, 1983.
[BIS 91] BISHOFF P.H., PERRY S.H., “Compressive behavior of concrete at high strain
rates”, Materials and Structures, Vol. 24, pp. 425–450, 1991.
[BUR 97] BURLION N., Compaction des bétons: éléments de modélisation et caractérisation
expérimentale, PhD Thesis, Ecole Normale Supérieure de Cachan, 1997.
[BUZ 99] BUZAUD E., DON D., CHAPELLE S., GARY G., BAILLY P., “Perforation
studies into MB50 concrete slabs”, 9th International Symposium on Interaction of the
Effect of Munitions with Structures, Berlin, Germany, 1999.
[CAM 99] CAMBORDE F., Modélisation du comportement dynamique du béton, application
aux problèmes d’impacts et d’explosions, Thesis, Ecole Centrale de Nantes, 1999.
[CED 77] CEDOLIN L., CRUTZEN R.J., POLI S.D., “Triaxial stress-strain relationship for
concrete”, Journal of the Engineering Mech. Division, Vol. 103, N° EM3, June 1977.
[CHE 75] CHEN ANDRIE C.T., CHEN WAI-FAH, “Constitutive relations for concrete”,
Proceedings of the American Society of Civil Engineers, Vol. 101, N° EM4, August 1975.
[COU 91] COUSSY O., La Mécanique des Milieux Poreux, Editions Technip, Paris, 1991.
[CUN 97] CUNDALL P.A., “Distinct element models of rock and soil structure”, Analytical
and Computational Methods in Engineering Rock, Allen & Unwin Eds, Ch. 4, pp. 129–
163, London, 1997.
[DAR 90] DARVE F., “The expression of rheological laws in incremental form and the main
classes of constitutives equations”, Geomaterials Constitutive Equations and Modeling,
Ed. Darve, Elseiver Applied Science, London, pp. 123–148, 1990.
[DON 98] DONZÉ F.V., MAGNIER S.A., “Numerical simulation of impacts using the
discrete element method”, Mech. Cohes.-Frict. Mater., Vol. 3, pp. 257–276, 1998.
[DRU 52] DRÜCKER D.C., PRAGER W., “Soil mechenics and plastic analysis or limit
design”, Quarterly of Applied Mathematics, Vol. 10, pp. 157–175, 1952.
 Dynamic Behavior of Concrete: Constitutive Models 93

[DUB 94] DUBÉ J.F., Modélisation simplifiée et comportement visco-endommageable des

structures en béton, PhD Thesis, University of Paris 6, 1994.
[ELW 79] ELWI A.A., MURRAY D.W., “A 3D hypoelastic concrete constitutive
relationship”, Journal of the Engineering Mechanics Division, Vol. 105, N° EM4, August
1979.
[GAR 98a] GARY G., BAILLY P., “Behavior of quasi-brittle material at high strain rate.
Experiment and modeling”, European Journal of Mechanics, A/Solids, 17, N° 3, pp. 403–
420, 1998.
[GAR 98b] GARY G., BAILLY P., GATUINGT F. “Testing concrete at high strains and high
rates of strain”. 3rd International Symposium on Impact Engineering’98, Singapore,
Singapore, 7–9 December 1998.
[GUR 77] GURSON A.L., “Continuum theory of ductile rupture by void nucleation and
growth: Part I – Yield criteria and flow rules for porous ductile media”, Engng. Materials
and Technology, Vol. 99, pp. 2–15, 1977.
[HAN 85] HAN D.J., CHEN W.F., “A non-uniform hardening plasticity model for concrete
materials”, Mechanics of Materials, 4, pp. 283–302, 1985.
[HER 69] HERMAN W., “Constitutive equation for the dynamic compaction of ductile
porous materials”, Journal of Applied Physics, Vol. 40, N° 6, May 1969.
[HOL 93] HOLMQUIST T.J., JOHNSON G.R., COOK W.H., “A computational constitutive
model for concrete subjected to large strains, high strain rates, and high pressures”, 14th
International Symposium on Ballistics, Quebec, Canada, 26–29 September 1993.
[LAB 91] LA BORDERIE C., Phénomènes unilatéraux dans un matériau endommageable:
modélisation et application à l’analyse de structures en béton, PhD Thesis, University of
Paris 6, 1991.
[LAS 88] LASRY D., BELYTSCHKO T.B., “Localization limiters in transient problems”,
Int. J. Sol. Struct., Vol. 24, pp. 581–597, 1988.
[LEM 85] LEMAITRE J., CHABOCHE J.L., Mécanique des matériaux solides, Dunod,
Paris, 1985.
[LEN 00] LE NARD H., BAILLY P., “Dynamic behavior of concrete: the structural effect of
compressive dynamic strength increase”, Mechanics of Cohesive-Frictional Materials,
Vol. 5, pp. 491–510, 2000.
[LEV 98] LE VU O., Etude et modélisation du comportement du béton sous sollicitation de
grande amplitude, application au problème de la pénétration dynamique, Thesis, Ecole
Polytechnique, Palaiseau, France, 1998.
[LOR 87] LORET B., “Application de la théorie des multimécanismes au comportement des
sols”, Manuel de rhéologie de géomatériaux, presses de l’ENPC, 1987.
[MAR 94] MARIOTTI C., THOMAS F., “Loi de comportement ARMOR pour géomatériaux
sous chargement dynamique”, Journal de Physique IV, Vol. 4, September 1994.
[MAZ 84] MAZARS J., Application de la mécanique de l’endommagement au comportement
non linéaire et à la rupture du béton de structure, PhD Thesis, University of Paris 6, 1984.
94 Dynamic Behavior of Concrete and Seismic Engineering

[MCQ 70] MC QUEEN R.G., MARSH S.P., TAYLOR J.W., FRITZ J.N., CARTER W.J.,
“The equation of state of solids from shock wave studies”, in High Velocity Impact
Phenomena, R. Kinslow (Ed.), Academic Press, New York, USA, 1970.
[MOR 96] MOREAU J.J., JEAN M., “Numerical treatment of contact and friction: the
dynamic contact method”, Proc. Eng. Systeme Design and Analysis Conference, ASME,
New York, USA, Vol. 4, pp. 201–208, 1996.
[PER 63] PERZYNA P., “The constitutive equations for rate sensitive plastic materials”,
Quaterly Applied Mechanics, Vol. 20, N° 4, pp. 312–332, 1963.
[PIE 88] PIETRUSZCZAKS, JLANG J., MIRZA F.A., “An elastoplastic constitutive model
for concrete”, Int. J. Solids Structures, Vol. 24, N° 7, pp. 705–722, 1988.
[PIJ 87] PIJAUDIER-CABOT G., BAŽANt Z., “Non-local damage theory”, J. Eng. Mech.,
ASCE, Vol. 113, pp. 1512–1533, 1987.
[PON 95] PONTIROLI C., Comportement au souffle des structures en béton armé – analyse
expérimentale et modélisation, PhD Thesis, ENS Cachan, 1995
[SAN 76] SANDLER I.S., DIMAGGIO F.L., BALADI G.Y., “Generalized CAP model for
geological materials”, Journal of the Geotechnical Engineering Division, Vol. 102,
N° GT7, July 1976.
[SUA 84] SUARIS W., SHAH S.P., “Rate sensitive damage theory for brittle solids”, Journal
of Engineering Mechanics, Vol. 110, N° 6, June 1984.
[ULM 93] ULM F.J., “Un modèle d’endommagement plastique: application aux bétons de
structure”, Report from a PhD Thesis, Ecole Nationale des Ponts et Chaussées, 1993.
[ZAO 93] ZAOUI A., PINEAU A., FRANÇOIS D., Comportement mécanique des matériaux,
Hermes Science Publications, Paris, 1993.
 Chapter 3

Seismic Ground Motion

3.1. Introduction

Seisms are sudden fracture failures, with aftershock that can reach several
meters, on areas that can exceed 10,000 km². The sudden sliding emits seismic
waves that propagate in the ground, causing multi-directional vibratory movements
ui (x,y,z; t) in horizontal and vertical directions that generate vibrations on the
surface of existing structures. Depending on their amplitude, frequency and duration
characteristics, these vibrations can damage buildings or even cause their complete
collapse. Seisms not only affect man-made structures, but also lead to the damage or
collapse of “natural structures” like soils, causing liquefaction, settlement or the
formation of unstable slopes.

The main aim of “seismology engineering” is either to estimate a priori the main
features of such vibrations so that surface structures can be dimensioned safely, or to
estimate the potential damage that could accrue on pre-existing natural or man-made
structures. Whilst, on the one hand, seismology attempts to analyze the space and
time repartition of vibrations (their location, depth, size and frequency of
earthquakes), in close connection with “conventional” geologists and seismologists,
it will also attempt to quantify the associated vibrating motions, and characterize
them using parameters representing their “damaging power”. The aim is to
understand the phenomena that cause vibrations and the physical values that control
them as well as possible, so that future seismological events can be predicted.

Chapter written by Pierre-Yves BARD.

96 Dynamic Behavior of Concrete and Seismic Engineering

Only those aspects linked with seismic motion itself will be dealt with here:
readers interested in seismology can read [BER 03], [LAM 96], [LAM 97] and
[MAD 91].

This chapter briefly outlines three areas:

i) techniques for the measurement of seismic motions for engineering purposes;
ii) qualitative characterizations, with a view to para-seismic dimensioning; and
iii) physical descriptions of the phenomena affecting the characteristics of such
motions, including their origin, propagation and site effects.

3.2. Measuring seismic motions

3.2.1. Differences between seismological and accelerometer networks

The first seismological instruments were built almost 2,000 years ago by the
Chinese, who developed a device for locating the epicenter of an earthquake. The
first “modern” instruments, which date back to the end of the 19th century, aimed at
detecting and recording motions that were imperceptible to man. Since then, a vast
range of seismological instrumentation has come online, with increased sensitivities
and frequency responses, designed to locate quake event maxima (whether local,
regional or global) and to provide information on the inner structure of the terrestrial
globe.

For technological reasons, these very sensitive instruments were initially

intended to monitor violent events in their immediate vicinity. This is why
Californian engineers since the 1930s have tended to design low gain instruments.
The first accelerometers (called “strong motion” instruments) were low gain devices
because these make it easier to record acceleration directly. Even now, when the
velocimeters available have wide recording dynamics, the “strong motion” ground
acceleration has endured as a testing parameter. It has also been kept alive by the
fact that most of the instrument networks set up globally to monitor violent motions
still use accelerometers. The logic behind the siting of stations remains quite
different; “strong motion” instruments are typically installed in high economic stake
areas (city areas, important works). The ways in which seismological networks use
data on strong motions also remain quite different. Besides “traditional” seismology,
a more practical “violent motion seismology” has developed, which does not
necessarily leave aside relatively sophisticated treatment or modeling: this forms the
subject matter of this chapter.
 Seismic Ground Motion 97

This chapter deals with the characterization of seismic stresses using values
suited to para-seismic design, and details the ways in which these values are
estimated and the way they are used to predict future earthquakes using either
empirical or more physical models.

3.2.2. Accelerometer networks

Several thousand accelerometers are in operation all over the world. Organized
as regional networks, they are sometimes installed as low span very dense local
networks, or installed inside a wide range of civil engineering works (buildings,
dams, nuclear power stations and geo-technical works). Generally, there are two
kinds of instrument: “open field” instruments, intended for recording the ground
motion (or seismic action); and “structure” instruments, used for recording the actual
seismic behavior of the monitored structure.

Until the last decade, the only purpose of the latter instrumentation was to
improve para-seismic knowledge and safety regulation. Telecommunications
advances have, however, allowed two other uses in quasi-real time: real time alert
generation in the case of big distant and remote earthquakes (like the Pacific
subduction area earthquakes and their effects in Mexico) and crisis management to
make a chart of the most severely hit areas in a very short time, for co-ordination
and decision-making in relief efforts (for example, gas network management in
Taiwan in 1999 – Tokyo Gas manages over 3,000 stations on its own behalf, or
high-speed trains – Shinkansen in Japan and Mediterranean TGV in France).

The most important open field networks, as far as volume is concerned, have
been installed in the Pacific area, notably Japan, Taiwan, California and Mexico.
Europe and countries around the Mediterranean with a few exceptions are somewhat
behind, but efforts have been made to create new networks (such as the RAP in
France) or to improve them (as in Italy).

Structure instrumentation is far less widespread, and the data obtained is

typically under-used, for reasons probably arising from “community cultures” that
are more or less modeling confident. Whilst glaring gaps in the networks still exist
in Europe, the USA and Japan have started their own consistent and sustained
instrumentation programmes. Their results are not always easy to “import” into
Europe because of differences in building processes, particularly when modeling
ancient buildings.
98 Dynamic Behavior of Concrete and Seismic Engineering

3.2.3. Accelerometer data banks

The rate at which new recordings are collected is constantly increasing, and a
cursory inspection of recent strong motion seismological data demonstrates they
have all been caused by “astonishing” recordings, the analysis and interpretation of
which has forced engineers to re-evaluate certain hypotheses in the field.

Whereas this data was spread confidentially until the 1980s, the advent of the
Web has forced a state of mind change thanks to the volume now available online.
Several sites are listed at the end of this chapter, but the system evolves so rapidly
that many of these may already be obsolete.

Two Internet site categories are given, some presenting recordings from specific
networks and others gathering data from quite varied sources to compile as
databases.

3.3. Quantitative characterization of seismic movements

Real accelerograms possess different forms of time, amplitude and frequency

content. In most cases, para-seismic dimensioning is not carried out on complete
accelerograms (used only in non-linear dynamic analysis cases), but is performed
using a restricted number of reducing quantities which are representative of the
damage a structure might receive.

Only those most commonly used are referred to; more elaborate descriptions are
given in [BET 03], [JEN 83] and [KRA 96].

3.3.1. Time maximum values

The most direct and simple quantity is the acceleration maximum, often referred
to as the “PGA” (peak ground acceleration). For a long time, when only analog
recordings were available, this was the only quantity available. As we will see later,
it remains in use because it links directly to the spectral characterization that para-
seismic engineers prefer, i.e. the acceleration response spectra.

For destructive earthquakes, PGA data has sensitivity better than 1 m/s². Even if
until the 1970s it was generally thought that ground acceleration could never be
more than 1 g, observations have since indicated values in excess of 20 m/s², even
along the vertical component (which may explain some reported discrepancies in
construction works with spans resting on their supports).
 Seismic Ground Motion 99

Obviously, the relationship of this quantity to the damage a seismic signal can
cause is somewhat obscure. It gives no information on the duration or spectral
content of the stress, both essential parameters for modeling non-linear oscillators
with several degrees of freedom, which lie at the root of paraseismic calculations.

Nevertheless PGA-macroseismic connections are often used quantitatively.

Everybody has to be aware that qualitative relationships are marred by huge
uncertainties and should never be used without explicitly taking into account the
associated standard deviation (a minimum factor 3). Most existing correlations show
that whenever intensity rises by 1 degree (MM or MSK), the PGA is multiplied by a
factor of 2 or 3.

To be able to take spectral content into account in an approximate way whilst

keeping the simple concept of a maximum value, we consider the ground velocity
(“PGV”) and displacement (“PGD”) maxima as well. Whereas the maximum
acceleration is mainly associated with the high-frequency spectral content (beyond 5
Hz), the maximum speed is associated with the intermediate frequencies (between 5
and a few Hz), and the maximum displacement is associated with low frequencies
(below 1 Hz).

For a destructive earthquake, values of PGV/PGD typically range from a few

cm/s to over 1 m/s, and from a few millimeters to a few meters. They are far more
sensitive to the size of the shock than the maximum acceleration, because size
influences the low frequency content far more than the high frequency content.

Nevertheless, neither of these two values shows a better correlation with macro-
seismic intensity. Furthermore, their estimation is less direct than that of the PGA,
since acceleration histories have to be integrated numerically. For analog or first
generation digital instruments, it is also less reliable because of the greater
sensitivity of the numerical process to low frequency noise.

However, it should be noted that knowledge of all three values gives a good idea
of the frequency level and content of a given accelerogram. Furthermore, the non-
dimensional ratio, PGA x PGD/(PGV)2, which relates to the width of the excited
frequency response, generally varies little from one recording to the next, with
values typically ranging from 1 to 10, and even from 2.5 to 7.5 [BET 03].

3.3.2. Spectral characterizations

Whilst characterizations that are more complex are often used in seismic
engineering, spectral characterizations convey information that is richer and more
consistent than mere maximum values. Unfortunately, two different spectral
100 Dynamic Behavior of Concrete and Seismic Engineering

characterizations are used by two distinct communities: the “producers” and “users”.
Seismologists prefer using Fourier spectra, due to mathematical practicability and
the possibility of establishing useful relationships with wave emission and
propagation physics. On the other hand, “users” like civil engineers (structure,
geotechnology) have become accustomed to reasoning in terms of response spectra,
as they are easily adapted to the simplified modeling of civil engineering structures.

3.3.2.1. Response spectra

The origin and interest in response spectra lie in the fact that they reduce (as a
rough approximation) the seismic behavior of a building to a simple oscillator with 1
degree of freedom. Representation as a response spectrum directly accesses the
motions of a structure’s center of gravity.

Let us consider a 1-dof linear viscoelastic oscillator, characterized by its

frequency f and damping ]. Due to an earthquake characterized by an acceleration
a(t), the oscillator will undergo a relative displacement x(t), and an absolute
acceleration x"(t) + a(t). Response spectra are defined as the time maximum
responses of the oscillator for a relative displacement, relative speed and absolute
acceleration:

Sd (f,[ ) = Max t {x(t)}

Sv (f,[ ) = Max t {x'(t)}

Sa (f,[ ) = Max t.{x"(t) D + D a(t)}

On varying the frequency (f) of the oscillator with a constant damping (]), we
obtain three curves, Sd, Sv, Sa defining respectively the response spectra for
displacement, speed and acceleration. These are usually calculated for damping
discrete values: 0%, 2%, 5%, 10%, and 20%, with 5% as the most frequent value.

With the following expression for the basic equation of a mass (m), stiffness (k),
and damping (c) of a simple oscillator with 1 degree of freedom:

m [ a(t) D + D x"(t) ] =  [ c x ’

Ȧ0  k/m  ;ȟ 1
2 c/  km 
 Seismic Ground Motion 101

Ȧ1 = Ȧ0 1  ȟ 2  
The expression for x(t) conventionally derives from Duhamel’s integrals:

t
X(t) =  1/ Ȧ1 ³ a  IJ  exp [  ȟȦ0  t  IJ ] sin [Ȧ1  t  IJ ] sin [Ȧ1  t  IJ ] dIJ
0

However, it can also be derived fast and simply using the Fourier transform.

According to these formulae, we can then see the following properties:

Sa (f,ȟ) = Max t.{a (t) }= a max = PGA ]

Sd (0, ȟ) = Max t.{  d (t) } = d max PGD ]

Sa (f, 0) = 4ʌ 2 f 2 . Sd (f , 0)

Sv (f,ȟ) | 2ʌ f . Sd (f,ȟ)

Sa (f,ȟ) | 4ʌ 2 f 2 . Sd (f,ȟ)

The last two formulae are only approximate with non-zero damping; two new
quantities have been introduced, called speed (S’v) and acceleration (S’a) pseudo-
spectra, which can be defined as:

S, v (f,ȟ) 2ʌ f . Sd (f,ȟ)

S,a (f , ȟ) 4ʌ 2 f 2 .Sd (f,ȟ) 2ʌ f . S, v (f,ȟ)

The latter relationships allow a quadri-logarithmic representation of the pseudo-

response spectra, with the logarithm of the frequency (or the period T) on the
abscissa, and the logarithm of S’v on the ordinate. S’a and Sd values can then be read
at once in relation to the diagonal axes, since:
102 Dynamic Behavior of Concrete and Seismic Engineering

log S,a = log S, v + log(2ʌ) + log (f) = log S, v + log(2ʌ)  log (T)

and as a correlation:

log Sd = log S, v  log(2ʌ)  log (f) = log S, v  log(2ʌ) + log (T)

Thus, it is quite easy, from a given accelerogram, to derive response spectra.

However, a given response spectrum (for a given damping value) corresponds to an
infinite number of time histories with the same acceleration and displacement
maxima but different durations and phases. The repartition of the energy arrival in
time is not constrained by the response spectrum data. The absence of equivalence
constitutes a great handicap when non-linear time studies have to be led, and using
even adapted accelerograms to a given response spectrum is a difficult task which
can only be performed by seismologists if they are to appear realistic.

Elastic response spectra are often used in paraseismic engineering because, as a

first estimate, simple structures can be compared to a 1-dof oscillator with a known
period and damping. Caution must be urged in this regard, as the conventional
formula T = N/10 (with N being the number of floors) has been imported from the
USA where most buildings are framework constructions and are not applicable in
France, where the prevailing structures have stiffer walls for which the formula T =
N/25 is more appropriate. The motions of the center of gravity can be reasonably
well estimated, as long as the structure is assumed to have a linear elastic behavior.

This last hypothesis is obviously not true when the structure has been seriously
damaged: that is why other non-linear spectra have been developed, among which
the simplest correspond to perfect elasto-plastic behavior. They involve the
introduction of an additional parameter, the ductility demand, (P), which defines the
relationship between the maximum displacement of the elasto-plastic structure and
that of the associated elastic structure (with the same low acceleration stiffness and
damping).

As estimating the dynamic response of non-linear systems is difficult, it has

recently been proposed that dynamic non-linear analysis is replaced by a static non-
linear analysis, called “push-over analysis”. The gross result of this is a curve that
links the applied horizontal force to the displacement. Procedures have also been
proposed for converting elastic response spectra into “stress-strain” curves, or to
transform the latter into “Sa-Sd” curves, linking force to acceleration by means of the
modal mass.
 Seismic Ground Motion 103

3.3.2.2. Fourier spectra

For seismology, the most commonly used spectral representation resorts to
Fourier transform, an algorithm well-assimilated and quick to use from a numerical
point of view. The Fourier acceleration spectrum is defined by:

f D
Fa (f ) A(f) . exp(iij (f)) = ³ a(t)exp(  i2ʌft)dt = ³ a(t) exp(  i2ʌft) dt
f 0

where A refers to the modulus, M is the phase and D is the total duration of the
recording.

Similar expressions exist for speed Fv(f) and displacement Fd(f) spectra, and the
conventional relations of Fourier transform are quite convenient and useful:

f fN
a(t) = ³ Fa (f) . exp(+i2ʌft) df = 2 ³ Fa (f) . exp (+i2ʌft) df
f 0

where fN is the maximum frequency which can be calculated taking the sampling
time step 't: fN = 0.5/'t.

Fa (f ) i. 2ʌf. Fv (f) =  4 ʌ 2 f 2
fN D
Fd (f) = ³ Fa (f ) df = 1/2 ³ a 2 (t) dt
0 0

This last relationship (Parseval’s theorem) allows us to link time descriptions

with spectral data: it is particularly used in many “spectral” models that allow the
quantitative parameters of seismic movement to be linked not only to the physical
properties that characterize the seismic source (emitting waves), but also to their
crust (i.e. very deep) propagation and site effects. However, it should be noted that
reconstructing the time signal from its Fourier spectra is only possible if both the
modulus A and the phase M are known. In general terms, only relationships between
the Fourier modulus and various parameters describing both origin and propagation
are well-established.

Therefore, methods and practices for describing the non-stationary time state of a
seismic signal and its consequences for the strength of building works are still
evolving. Some quite interesting tracks have been opened using “group delay time”
or the phase M derivative in relation to frequency f.
104 Dynamic Behavior of Concrete and Seismic Engineering

3.3.2.3. Relationships between response and Fourier spectra

A qualitative correspondence exists between these two spectral representations,
which both express the frequency repartition of energy of the seismic signal.
However, no strict mathematical relation between the elastic response spectra and
the Fourier spectra exists. The only relationship that can be demonstrated is an
inequality, given by:

Sv (f , 0) t Fa (f )

3.3.2.4. Generic spectral shapes

These spectra are characterized by simple shapes (Figure 3.1): once smoothed, a
Fourier spectrum A(f) is characterized by a plateau shape between two frequencies fc
and fmax. The first (the corner frequency) is inversely proportional to the size of the
failure area, but it is also linked to the stress drop caused by the earthquake and to
the directivity process, that is, to the position of the receptor at the front or at the
back of the failure propagation direction. For destructive earthquakes, it is generally
less than 1 Hz, and can drop well below 0.1 Hz for very high magnitude
earthquakes. The second frequency is linked to inelastic attenuation phenomena
which affect the waves during their in-depth and surface propagation, and to
processes that govern failure dynamics on the crack. This frequency is generally
higher than 4-5 Hz, and can sometimes exceed 20 Hz. Below fc, the acceleration
Fourier spectrum modulus varies as f², which corresponds to a plateau for the
displacement Fourier spectrum, usually written :0, and is linked to the properties of
the source by the following relationship:

:0 M 0 .R șij / (4ʌȡ R c3 )

where M0 is the seismic moment P.D0. S, Pisthe shear stiffness at focal depth, D0 is
the average reactivation on the fault, S is the total failure area one the fault plane, R
is the focal depth, and c is the wave propagation speed (and thus generally that of
the S waves if we are considering horizontal component forces).
 Seismic Ground Motion 105

Figure 3.1. Generic shapes of acceleration Fourier spectrum, for increasing

magnitudes ranging from 3 to 7 (these curves were established on the
basis of scale laws specified in section 3.4.1)

Beyond fmax, the acceleration Fourier spectrum modulus decreases in fD, but
with an exponent D that can vary according to the site and the event.

As shown in Figure 3.2, the acceleration response spectrum, represented as a

function of the period, has a characteristic shape, with an ordinate equal to the PGA
at the origin. It has a plateau between two periods depending on the site and the size
of the event and a level typically between 2 and 3 PGA, followed by a branch
decreasing first as 1/T, and then as 1/T².

Figure 3.2. Example of a real response spectrum for a 5.5 magnitude earthquake
(San Salvador 1986), and generic shape of the pseudo-acceleration response spectrum,
as defined in most para-seismic regulations (EC8 in this case)
106 Dynamic Behavior of Concrete and Seismic Engineering

3.3.3. Features of hybrid characterizations

Other representative values are used in the field: from a physical point of view
they are more significant than peak values (PGA, PGD and PGV) and simpler than
spectral representations. Many proposals have been made; here we will only refer to
those that have been accepted by the majority of the para-seismic community.

3.3.3.1. Spectral intensity

This value was introduced initially by Housner to characterize the stress average
level of the most “common” structures: he considered the response spectrum integral
on periods ranging from 0.1 to 2.4 seconds:

T = 25
SI (ȟ) | ³ Sv(T,ȟ) dT.
T = 0.1

As a rule, SI is calculated for high damping values (]: typically 20%), in order to
integrate the already highly smoothed spectrum.

Other values can be defined by changing the integration bounds, altering the
mute variable (period or frequency), or by carrying out the integration on a
logarithmic axis. None really stands out as superior to any other in practice.

However, the notion of spectral intensity has come back in force in Japan over
the last two decades, namely IS SI(0.2)/2.4 s (then associated with the speed
response spectrum average level on the 0.1 – 2.5 s range). The reason for this is the
excellent correlation observed with the intensity of damage, and with a significant
damage apparition threshold for IS t 30 cm/s. The correlation has proved good
enough for Tokyo-Gas to use it in Tokyo and its suburbs, where over 3,000
instruments calculate it directly and transmit it in real time to a monitoring center,
for the purpose of mapping the damage estimated layout to decide which safety
floodgates should be activated.

3.3.3.2. Duration, average quadratic acceleration and the Arias intensity

Many characterizations are based on the increasing monotonous function W(t),
defined by:

t
W(t) ³ a 2 (IJ) dIJ
0
 Seismic Ground Motion 107

Its final value Wf, is directly linked to the energy contained in the signal a(t). It
can also be connected to the spectral content using Parseval’s relation (see above).
For this reason, Arias used it to define the “Arias intensity”, viz: IA = S/2g. Wf.

This is also used to define a “high phase duration”, D (H1, H2), given by:

D (İ1 , İ 2 ) = T (İ 2 )  T (İ1 )

where T(Hi) is defined by: W(T(Hi)) = Hi. Wf.

If H1 is nearly always chosen as equal to 5%, the values of H2 are either 75%
(which involves quite short a duration) or 95% (which, on the contrary, leads to a
rather long duration).

Correlatively, as soon as duration has been determined, we can define an

“average quadratic acceleration”, arms (H1, H2) for this duration:

a rms (İ1 , İ 2 ) = ª¬ İ 2  İ1  Wf / D  İ1 , İ 2  º¼

Thus, once a duration has been defined, we can estimate time parameters owing
to the Fourier spectrum.

3.3.4. Caveats regarding differential motions

In addition to considering pure translation motions, the rapid space variations

shown by seismic motions should not be neglected either. These variations have
multiple origins, including the differences due to wave propagation, decorrelations
related to heterogenity or the complex wave field and local site effects.

Thus, the origin of spatial variability of seismic motions is multiple:

– whenever the angle of incidence is oblique (T z 0°), their horizontal
propagation speed is finite and equal to c/sin T. From then on, at two sites separated
by a distance d (projected along the propagation direction), the signal (despite its
shape and amplitude being identical for both sites) will be delayed by W = d sin T/c.
The delay has to be compared with the predominant period of the signal T = 1/f and
can induce significant effects when the ratio W = d sin T/c exceeds a threshold of 0.1.
The “horizontal propagation” effect will become increasingly significant when the
108 Dynamic Behavior of Concrete and Seismic Engineering

propagation speed c is low (therefore for weak stiffness grounds), the angle of
incidence is great (therefore for surface waves), and the frequency is high;
– in reality, as the propagation medium is not homogenous (especially near the
surface, and the incident wave field is not limited to one single wave), an infinity of
them is generated on each point of the failure area, and they therefore arrive with
different azimuths, different angles of incidence, different polarizations and different
frequency contents. This effect is particularly marked in a near field; they give rise
to multiples whenever they meet a heterogenity (e.g. reflection phenomena,
refraction, geometric diffraction) whilst propagating. As the propagation delays
associated with each individual wave vary from one wave to the other, the resulting
signal is greatly modified from one site to the other, even if its overall spectral
characteristics remain unaltered. These signal modifications are usually measured by
the coherence;
– for structures with a large basemat into the ground (e.g. bridges and dams), the
geological and geotechnical conditions can vary greatly along the foundations and
involve significant variations of frequency content, as well as the phase
modifications mentioned above.

Besides transient “dynamic” differential motions, residual static motions

associated with the sudden or continuous creep of active faults or sliding activated
or re-activated by earthquakes can exist. For example, for the new Rion-Antirion
Bridge, it was necessary to take into account the “continuous” opening of the Gulf
of Corinth, at a rhythm of about 2 cm a year, and the potentiality of a major
earthquake that might bring about sudden spreading over 10 cm. Such imposed
differential displacements should be taken into account in town planning and design,
but it should be remembered that the damage associated with surface failure
accounts for only about 1% of the total damage earthquakes cause.

3.4. Factors affecting seismic motions

The physics of seismic wave emission on fault propagation within the Earth’s
crust involves diverse phenomena associated with an extensive literature, and in
some cases, are not well understood themselves. For this reason, the indications
presented below are deliberately simplified: any reader interested in looking deeper
into the subjects could profitably consult the works of [AKI 80], [BER 03], [BET
03], [KRA 96], [MAD 91] and [PEC 85].
 Seismic Ground Motion 109

3.4.1. Spectral signature of the seismic source

The term “seismic source” refers to wave emission caused by the creep of a
finite sized fault area; this creep is neither instantaneous nor homogenous on the
failure zone, and the emission process is so complex that it forms a research field in
its own right.

When simplifying to extremes, the source of an earthquake (Figure 3.3) can be

viewed as a failure starting at a focus F (given by the ([K) coordinate hypocenter)
which then propagates along the fault plan ([K) at a failure speed VR, causing at
each point a sliding 'u ([Kt). At the end of each earthquake, sliding at these points
reaches a final value 'Uf([K).The source is completely determined by the space-
time sliding function 'u and the waves emitted at point ([K) have a displacement
signal proportional to w/wt ['u ([Kt)]. At present, we do not know how to
anticipate future earthquakes; nevertheless, with good instrument conditions, we can
discover its characteristics for high wavelengths (several kilometers) and low
frequencies (generally lower than 1 Hz) a posteriori. We are also beginning to
reconstruct a stochastic description of the short wavelength and high frequency part.

However, without going into all the relevant details, it is now possible to explain
certain general aspects of the spectral content of emitted waves, thanks to a few
overall features of the seismic source. These overall features together with their
effects are described below (Figure 3.3):
– the 'Uf ([K) quantity is reduced to the surface F of the failure zone (LW for a
rectangular fault, Sa² for a circular fault), and to the average sliding D0. This enables
the definition of a fundamental quantity indicating the importance of the earthquake,
namely the “seismic moment”, M0 = P S D0, where P is the shear stiffness of the
Earth’s crust at the level of the fault. This quantity controls the low frequency level
of the emitted waves, expressed in N.m;
– the stress drop 'Vexpresses the shear stress relaxation between the state
immediately prior to failure, V, and the state immediately following failure end, V.
Elastic dimensional analysis allows the following rough estimate to be obtained: 'V
= C P D0/Lc where Lc is the dimension characteristic of the fault (W for a rectangular
fault, a for a circular fault), and C is a shape coefficient close to 1;
– the energy released during an earthquake corresponds to the work of shear
stresses. It can be roughly approximated by the work of the average stress E = S.
[(V + V)/2]. D0. Thus, we can write E = M0. [(V + V)/2P]. Because, deep into the
ground, the shear stiffness P and the average stress (V + V)/2 do not differ
significantly, we can see that the seismic moment is a good indicator of the total
energy released;
110 Dynamic Behavior of Concrete and Seismic Engineering

– the failure speed typically ranges from 2 to 3.5 km/s, but it can sometimes
exceed the speed of the waves S. In such cases, the failure is termed “supersonic” or
“super shear”. Together with the dimensions of the failure area, this speed controls
the wave emission time and the length of near-field strong motions. Moreover, it
plays a part in directivity phenomena which are similar to the well known Doppler
effect: a receiver located at the front of the failure propagation direction will receive
waves emitted for a shorter time the nearer the failure speed is to wave propagation
speed (S), which involves more intense motions. The opposite is true for a site
located at the rear of the failure front;
– the final “global” parameter used is the “rising time” (W). This is the time
necessary for sliding at a specific point to pass from 0 (pre-failure) to its final value
'Uf.. In reality, W varies from one point to another on the failure area. Nevertheless,
for simple models, we assume the sliding function is identical on all points, i.e. that
'u([Kt) = Uf([K S(t – d/VR), S(t) being a function the derivative which has a
bounded support [0, W], and d being the distance between the current point ([K and
the hypocenter ([K The simplest function is a linear ramp function (see Figure
3.3).

Figure 3.3. Schematic representation of the failure process. The diagram on the left
represents the real history of the sliding field on the fault D (x,y;t) (isochrones of the failure
front at successive instants), whereas the diagram on the right illustrates its simplified
representation as a (L,W) dimensioned rectangular fault, in which the dislocation function
D(t) follows the same time history on each point (ramp function characterized by its rising
time W and its uniform final dislocation D0)

The analysis of these parameters for numerous earthquakes demonstrates they do

not vary from one another, and therefore that the following three “scaling laws”
exist:
– a “geometric” law; W/L | c1, with c1 | 0.5 (except for important strike-slip
earthquakes, where width W is limited by the thickness of the brittle part of the
Earth’s crust, i.e. 15-20 km);
 Seismic Ground Motion 111

– a law associated with a relative constancy of the stress drop: D0/Lc | c2, with c2
| 10-5 and 10-4 (which gives 'V = C P D0/Lc | P c2 | 3. 105 to 106, i.e. 0.3 at 3 MPa);
– a “dynamic” law: WVR/L | c3, with c2 | 0.2 (the rising time is proportional to
the total length of failure propagation).

By integrating these empirical scaling laws into simple models, we can show
that:
– the magnitude of the moment MW = 2/3 log M0 – 6 varies like the logarithm of
the fault’s surface;
– the seismic moment, and the very low frequency content vary as the cubed
fault’s dimension and as 10 1.5 MW;
– the corner frequency (which is linked to the emission length of the waves on
the failure area), varies as the inverse of the fault’s dimension, and then as 10 -0.5 MW;
– from then on, the level of the plateau of the acceleration Fourier spectrum is
roughly proportional to the fault’s dimension, and then it varies as 10 0.5 MW.

3.4.2. Effects of propagation in the Earth’s crust

Once emitted, seismic waves propagate inside the Earth’s crust: these deep
propagation effects are characterized by three main phenomena, which are discussed
in the following sections.

3.4.2.1. Geometric expansion

This is simply the decrease in wave amplitude due to distance, in close contact
with energy preservation.

For “volume” waves, the wave front is spherical, and the conservation of energy
(which is proportional to the squared amplitude on this surface) involves using a
geometric expansion term varying in 1/R.

For surface waves in which the energy concentrates in the direct vicinity of the
surface, the wave front is cylindrical and the geometric expansion term varies in
1/R0.5.

These theoretical dependencies are of course, only valid for far fields, and for a
perfectly homogenous space (or semi-space). In practice, the existence of near-field
terms, the non-homogenity of the propagation medium and the multiplicity of the
waves bring about geometric expansions with terms which are markedly different
from the exponents – 1 and – 0.5.
112 Dynamic Behavior of Concrete and Seismic Engineering

3.4.2.2. Multiple pathways

As it meets heterogenities in the crust during propagation, the wave emitted from
the source will be partially transmitted, partially reflected, and possibly partially
diffracted. There are also possible conversions between radial P and S waves.
Overall, this results in the existence of many additional waves with different paths
that will prolong the signal, which decrease faster with the distance of direct waves.

For certain strong earthquakes, energy phases corresponding to reflections of

direct waves at the base of the crust appear at distances ranging from 50 to 100 km
This was evident in San Francisco in 1989 during the Loma Prieta earthquake
(located 80–100 km away).

Effects like these are almost impossible to predict with simple models. However,
their numerical modeling is relatively simple, providing the structure to the Earth’s
crust is known (location and characteristics of the main discontinuities).

3.4.2.3. Inelastic attenuation

As is the case in any real material, these waves undergo some energy loss
(mainly via aggregate friction), which is characterized in seismology by a quantity
factor Q that is related to the damping coefficient ] by the relationship Q = 0.5/].
The definition of Q is linked to the relative energy loss on a wavelength O, chosen to
be equal to 'E/E = – 2S/Q. For a frequency f, plane waves propagating in the R
direction with a speed c, have an associated spatial dependency given by:

A(R) = A 0 (R). exp (  i 2ʌf t) ˜ exp (  ʌf r/Q c)

for which A0 (R) corresponds to the geometric expansion decrease:

ª¬ A  R+n O  A(R) ˜ exp   ʌf n O /Q c  A  R  ˜ exp   ʌ n/Q  º¼

This expression shows that for a given distance R, higher frequencies will be
attenuated more than lower frequencies, assuming Q has the same value at high and
low frequencies. This effect is often used to account for the high frequency decrease
of the Fourier spectrum.

Typically, inside the Earth’s crust, Q values are a few hundred (generally higher
for P waves than for S waves), and this effect only becomes evident when the
distance corresponds to a number n of value of Q wavelengths. However, in the
most superficial materials, Q can reach values equal to only a few dozen: if we take
 Seismic Ground Motion 113

a Q of 20 on a 200 m deep layer, with 200 m/s speed values, we calculate loss
factors of exp (– S/20) at 1 Hz (negligible), and of exp (– S) at 20 Hz (significant!).
This indicates that the high frequency decrease effect is due either to propagation
over long distances within poorly attenuating media or to a very superficial effect
within highly attenuating materials. We note that the values of Q found in the
literature generally reflect two phenomena: on the one hand, the intrinsic inelastic
attenuation, and on the other, energy losses through diffusion on short wavelength
heterogenities – this often results in a frequency dependency of Q that is thought to
be linked to the size of the heterogenities. This dependency is expressed generically
as:

Q = Q0 f Į with 0 d Į d 1

3.4.3. Site effects

Seismic motions can also be noticeably modified by close surface geological

conditions. This is evident by comparing their wavelengths (O = c/f, for c varying
from 100 m/s to 2 km/s, and f varying from 0.5 to 10 Hz, i.e. for O varying from 10
m to 4 km), with the dimensions of surface heterogenities that are in the same range.
Because such heterogenities can be quite important, both mechanically and
geometrically, the interferences between the incident wave fields can be very intense
and cause modifications of the spectral characteristics of seismic motions.

Seismic effects are thus wave propagation effects leading to amplifications (or
de-amplifications) located in space which can reach very high factors (greater than
10 in some extreme cases). They mainly affect topographic reliefs and sedimentary
fillings. In the next section, we briefly explain the basic physics for these typical
configurations. The curious reader will find a more comprehensive description in
[BAR 99], which discusses the remaining unsolved questions, and gives an account
of the diverse estimation methods used.

We also feel it is important to mention an additional site effect category – those

characterized by ground located failures caused by intense vibrations. These include
liquefaction affecting saturated sandy soils (causing subsidence) and slope
instability (landslips, sliding, both superficial and deep). If the latter arise in static
conditions, the earthquake is only a starting element, but the former are specific to
earthquakes. Taking both of these phenomena into account depends more on
particular dispositions (especially for foundations or ground improvement) than
structure calculation, so they are not covered in detail here.
114 Dynamic Behavior of Concrete and Seismic Engineering

3.4.3.1. Effects of topographic relief

After destructive earthquakes, the damage reported is significantly worse at the
hilltop than the base. The most outstanding French examples took place in Rognes
and Vernegues (1909 earthquake) and in Castillon (1887 earthquake). This
“qualitative” observation has been confirmed by numerous instrument
measurements, which revealed amplitude relationships between the summit of a
relief and its foot which sometimes reach a 3 to 4 factor difference in their time
maximum (PGA, PGV) and a factor higher than 10 in the spectral field (Figure 3.4).
The number of experimental studies on the subject is extremely low compared with
the studies dealing with the underground amplification, so it is not yet possible to
derive statistically significant empirical laws. Theoretical and numerical models also
predict a seismic motion systematic amplification perpendicular to the convex parts
of a topographical relief (cliff edge, hilltop), in addition to sensitivity to the field
characteristics of the incident wave (wave types, azimuth, incidence). The sensitivity
of seismic motions to the topographic relief seems to be linked to three physical
phenomena:
– sensitivity of surface motion to the incidence angle, especially around the limit
angle for SV waves;
– focusing of seismic waves in a convex relief (summit) and defocusing in a
concave relief (foot);
– volume and surface wave diffraction over all the surface irregularities, which
generates interferences (constructive and/or destructive depending on the position
and the frequency).

The known facts on this topic can be summarized as follows:

– as far as quality is concerned, agreement between theory and observations is
quite satisfactory;
– a convex relief generates amplification whereas a concave relief brings about
de-amplification;
– the amplification is generally higher for horizontal component forces than for
vertical component force. In the case of 2D reliefs (ridges), the horizontal
component perpendicular to the crest’s axis is often more amplified, and the
transverse stiffness is weaker than the longitudinal stiffness;
– amplification level is linked to the topographic slenderness (height/width): the
stiffer the average slope, the higher the summit amplification;
– summit amplification (as well as base de-amplification) shows a strong
frequency dependency. The maximum effects correspond to wavelengths similar to
the relief’s horizontal dimensions, favoring an interpretation of the topographic
 Seismic Ground Motion 115

effect by means of lateral interferences of the diffracted waves on all the surface
irregularities;
– qualitatively, the situation is rather confused. Besides a few well documented
cases for which the in situ instrument observations reveal moderate amplifications
(crest/base spectral amplification between 2 and 3), in good accordance with the
numerical modeling, in many cases the observed amplifications are far more
important than the theoretical previsions obtained from sometimes sophisticated
models, which are either two- or three-dimensional. Thus, an important proportion
of the instrumented sites showed spectral amplifications around 10, whereas this
value was reached by only two simulations, and these resorted to other effects rather
than geometry alone (e.g. mechanical contrast).

Figure 3.4. Example of topographic amplification in the village of Castillon (maritime Alps),
severely damaged during the 1887 Ligure earthquake. The instrument transfer functions
(on top) show the amplification on the “summit” site compared with the
“Mercier” site (from [NEC 95])
116 Dynamic Behavior of Concrete and Seismic Engineering

To conclude this section, whilst theoretical convex relief focusing certainly plays
a significant part, it does not seem to be the only physical phenomenon involved.
The fact that only this phenomenon has been incorporated into the French national
para-seismic regulations is quite justified. We should, however, be aware that more
important effects that are not well delineated at present do take place, but can only
be grasped by measurement.

3.4.3.2. Effects of the sedimentary or alluvial cover

3.4.3.2.1. Observations
Earthquake damage is generally more significant in sediment-filled areas than on
rocky outcrops. The intensity local increments observed commonly reach 1 or 2
degrees in intensity (MM or MSK scale) and have sometimes exceeded three
degrees (Mexico and San Francisco for instance), indicating in certain cases the
complete control of damage distribution on the nature of the ground. These
observations have given rise to large numbers of instrumental studies that have
confirmed the existence of amplification phenomena, and to as many theoretical and
numerical works, the results of which are summarized below. The curious reader is
referred to the different synthesis articles in the bibliography.

3.4.3.2.2. Physical explanations for amplification

The basic phenomenon responsible for amplification is the trapping of seismic
waves – particularly the S waves – into low mechanical (shear) stiffness superficial
formations. For horizontally-stratified formations (“1D” structures), this only affects
the incident volume waves that travel vertically back and forth between the surface
and the substratum-sediment interface (Figure 3.5 left). When these are not very
stiff, superficial formations present thickness side variations whether they are 2D
(valleys) or 3D (basins). We can also discern generation of local surface waves
(through diffraction on lateral heterogenities) that can reflect between the valley or
basin sides.

The interference between trapped waves causes resonance phenomena, the

features of which (frequencies, natural modes, amplification) depend on the
geometric and mechanical characteristics of the structure. Diffraction and trapping
phenomena also have other consequences apart from amplifying certain frequencies.
 Seismic Ground Motion 117

Figure 3.5. Principle of site effects associated to the alluvial cover: the incident waves are
trapped therein and their vertical reflections (figure on the left) cause a resonance
phenomenon characterized by amplification peaks at certain frequencies (fundamental and
harmonic). The curves on the right illustrate the variability of the amplification according to
damping (full line: ] = 2.5%, T = 0°; dotted line: ] = 0.5%, T = 0°) and the angle of
incidence (dashes: z = 2.5%, T = 60°)

3.4.3.2.3. Spectral signature

Within the frequency field, trapping effects are expressed through the strong
frequency dependency of surface amplification (or “transfer function” H(f), and the
relationship of the surface Fourier spectrum As(f) with that of the depth incident
motion A(f). These transfer functions can be calculated with exact methods for a
limited number of configurations (1D stratified media, “canonical” valleys or basins
perfectly semicircular, elliptical or spherical geometrically speaking). Here we only
give results for a simple model with only one sediment layer (medium 1) resting on
a rocky substratum (medium 2) by means of a horizontal interface.

In this case, the complex transfer function for vertically incident S waves is
given by the relationship:

H(f) = 2 C/ ª¬ C ˜ cos  2ʌ f h/ȕ1   i ˜ sin  2ʌ f h/ȕ1  º¼

where h is the depth of the layer, f the frequency, C the mechanical impedance
contrast U 2 E 2 / U1E1 , Ui  the density of medium i and ELthe S wave speed of
medium i.

This general formula gives access to resonance frequencies fn and to the

corresponding amplifications (Figure 3.5 right):
118 Dynamic Behavior of Concrete and Seismic Engineering

– resonance frequencies;

f0 ȕ1 / 4 h (fundamental mode)

fn (2n+1) f 0 (harmonics)

where n is a positive integer;

– amplification: in the absence of any inelastic damping, the resonance
amplification – apart from free surface effect – is equal to the mechanical impedance
contrast:

A0 H  f0  / 2 C ȡ 2 ȕ 2 / ȡ1 ȕ1 (all modes)

– in the presence of damping, the amplifications are reduced, in a more and more
marked way for higher harmonics. As for the fundamental mode, the formula is still
simple:

A0 C/ 1 + 0.5 ʌ[l C 

where ]is the damping inside the surface layer.

Whilst the value of the fundamental frequency is rarely modified, the same is not
true of amplitudes (except when we are interested in the vertical component force
and when the S wave speed E1 has to be replaced by the P wave speed D1). Certain
numerical calculations for deep valleys have shown over-amplification factors up to
4, with very deep modifications of transfer functions beyond the fundamental
frequency f0. In addition, lateral reflections generate other harmonics that can
combine to create wide band amplification areas.

3.4.3.2.4. Time signature

In the time field, these effects alter maximum amplitudes, wave shapes and
motion durations, especially for 2D or 3D structures.
 Seismic Ground Motion 119

Maximum values

Figure 3.6. Example of the appearance of site effects in the time field. RAP recordings of
25 February 2001 (M =4.7) earthquake in the city of Nice. The “Mont-Boron” station is
on rock, whereas the four others are on the Paillon’s alluvial deposits; the “Alsace-
Lorraine” station is the thickest

It is almost impossible to give general indications, as the effects depend too

much on the position of the frequency band amplified in comparison with the
incident spectral content. For example, a site that is not very thick (high f0) may give
rise to strong PGA amplification during a moderate earthquake nearby, but no
particular effect during a wide amplitude earthquake far away – even if the absolute
incident PGA levels are identical! Conversely, very thick sites may not affect, or
even reduce, the PGA during moderate earthquakes nearby, yet amplify it
considerably for those far away, while the transfer functions remain unchanged.

3.4.3.2.5. Duration
Because motion duration is rarely taken into account in paraseismic design, few
attempt at quantifying the consequences of resonance and trappings on duration
have been published. From a qualititative point of view, however, valley and basin
configurations seem to prolong motions in a significant way ([BEA 03], [COR 03],
[PAR 03]). The study of prolongation may even constitute a good method for
detecting the presence of 2D or 3D effects (Figure 3.6).

3.4.3.2.6. Modifications due to non-linearity

As with most other materials, grounds deteriorate when subjected to high
mechanical stresses: as shown by [SEE 69]’s pioneering works, the secant stiffness
120 Dynamic Behavior of Concrete and Seismic Engineering

modulus decreases and the damping increases when the imposed strain increases:
G = Gmax 1./(1+J/Jr). Such behavior might be expected to lead to a decrease of both
frequencies and resonance associated amplifications, as well as reductions in the
high-frequency content (maximum acceleration, especially).

The question is knowing where the strains imposed by an earthquake are located
(they can quite often exceed 10-3, and sometimes reach 10-2) with regard to the
critical strain values (Jr). Seismologists and geotechnicians disagree on this point,
the former believing that accelerometer observations were best explained by
reference to linear visco-elastic behaviors, whilst the latter measured (Jr) values
under 10-3 under laboratory conditions. In the last two decades, observations have
tended to reconcile both points of view, owing to the simulation of far less non-
linear behaviors under lab conditions, notably for very plastic grounds [VUC 91],
and the observation of non-linear in situ effects, especially in sand layers. Some
quantitative disagreements still exist, since even the best accelerometer data seems
to reveal a slightly to appreciably less non-linear behavior than that predicted by
numerical models derived from laboratory measurements [BON 03].

It is important to grasp the reason for inconsistencies between in situ and

laboratory observations, particularly for seismic zones such as France. Currently,
French regulation (PS92) authorizes a 20% decrease of the high frequency content in
soft ground, implicitly assuming highly non-linear grounds, and has proved to be
rather conservative. The EC8 recommendations allow high frequency amplifications
of up to 80%, whilst the latest American propositions reach 200%.

3.5. Conclusions

Strong motion seismology is a relatively new subject which has evolved a great
deal over the last decades, advances often being due to the questions raised by
“abnormal” damage and intensity observations made during destructive earthquakes,
generally via accelerometer recordings. Whilst estimating seismic motions was
mainly empirical in the last century, many models that were satisfactory from a
physical point of view emerged in the following decades, and have been
progressively adjusted and calibrated to real time instrument measurements.

The percolation time between “normal observations” and their use in seismic
engineering practice remains quite long (about a decade and even more for
conventional regulations) for the following reasons: if we simplify in the extreme,
the seismic force F which a building will have to withstand is proportional to three
terms: F v az. S’a(T,])/Qs, where:
 Seismic Ground Motion 121

– az is the rated acceleration mainly depending on the seismic area (regional

zoning);
– S’a(T,]) is a spectral shape standardized with regard to the rock, taking the
magnitude of the most frequent earthquakes into account, but which mostly depends
on the surface geological conditions;
– Qs (behavior coefficient) is a reduction factor which takes into account the
resistance reserves of the structure, linked to the more or less ductile behavior of its
constitutive materials and their assembling mode: this coefficient can vary from 1 to
1.5 for very brittle structures (non-reinforced masonry), and up to 8 for very ductile
structures (steel).

Logically, evaluation of all these terms should be performed in separate and

independent steps. This is rarely the case, as so many uncertainties exist when
evaluating the unknown factors. This is coupled to the inertia of building practices in
different countries, some of whom do not think it is possible to clearly distinguish
unknown parameters from design parameters. Final regulations for a given country
are usually a compromise reached when it is difficult to separate the safety
coefficients inherent to each step, and there is a reluctance to modify any unknown
factor values (az for example), without modifying the whole regulations.
Nevertheless, such weaknesses will be corrected eventually, which will enhance the
weak links, and assist in fields where uncertainties are the highest and greater efforts
have to be made.

As far as unknown factors and seismic motions are concerned, despite

improvements made in the last decades, many uncertain areas (motions very close to
the source, importance of non-linearities, possible maximum motions) remain.
Experience shows that only instrument observations can lead to further progress in
practice. It is therefore absolutely essential to support the installation of more
instrumentation on open field sites, as well as inside structures, particularly since the
cost of an accelerometer (between €3,000 and €10,000) is derisory if compared with
the cost of rebuilding a bridge, a dam, a nuclear power station or a 10-floor building;
moreover, because of recent advances in telecommunications techniques,
accelerometer maintenance costs are now comparatively modest.

3.6. Bibliography

[AKI 80] AKI K., RICHARDS P.G., Quantitative Seismology. Theory and Methods, Freeman
and Co, San Francisco, USA, 2 Volumes, 932 pages (re-edited into a single volume in
2002), 1980.
122 Dynamic Behavior of Concrete and Seismic Engineering

[BAR 99] BARD P.-Y., RIEPL-THOMAS J., “Wave propagation in complex geological
structures and local effects on strong ground motion”, in Wave Motion in Earthquake
Engineering (Chapter 2), E. Kausel and G.D. Manolis (Eds.), WIT Press (“Advances in
Earthquake Engineering” Series), ISBN 1-85312-744-2, pp. 38–95, 1999.
[BEA 03] BEAUVAL C., BARD P.-Y., MOCZO P., KRISTEK J., “Quantification of
frequency-dependent lengthening of seismic ground motion duration due to local geology:
applications to the Volvi area (Greece)”, Bull. Seism. Soc. Am., 93-1, pp. 371–385, 2003.
[BER 03] BERNARD P., Qu’est-ce qui fait trembler la terre ? A l’origine des catastrophes
sismiques, EDP Sciences, collection “Bulles de Sciences”, 287 pages, 2003.
[BET 03] BETBEDER-MATIBET J., Génie parasismique, 3 volumes, Hermes, Paris, 2003.
[BON 03] BONILLA L.F., COTTON F., ARCHULETA R.J., “Quelques renseignements sur
les effets de site non-linéaires en utilisant des données de forage: la base de mouvements
forts Kik-net au Japon”, VIe Colloque National de l’AFPS, Ecole Polytechnique, 1–3 July
2003, Vol. I, pp. 181–192, 2003.
[COR 03] CORNOU C., BARD P.-Y., DIETRICH M., “Contribution of dense array analysis
to basin-edge-induced waves identification and quantification. Application to Grenoble
basin, French Alps (II)”, Bull. Seism. Soc. Am., 93-6, pp. 2624–2648. 2003.
[JEN 83] JENNINGS P.C., “Engineering seismology”, in Earthquakes: Observation, Theory
and Interpretation, H. Kanamori and E. Boschi (ed.), North-Holland, pp. 138–173, 1983.
[KRA 96] KRAMER L.K., Geotechnical Earthquake Engineering, Prentice Hall International
Series in Civil Engineering Mechanics, Upper Saddle River, 1996.
[LAM 96] LAMBERT J., LEVRET-ALBARET A. (Ed.), Mille ans de séismes en France,
Catalogue d’épicentres paramètres et références, Ouest Editions, Presses académiques,
80 pages, 1996.
[LAM 97] LAMBERT J. (Ed.). Les tremblements de terre en France, Editions BRGM,
196 pages, 1997.
[MAD 91] MADARIAGA R., PERRIER G., Les tremblements de terre, Presses du CNRS,
211 pages, 1991.
[NEC 95] NECHTSCHEIN S., BARD P.-Y., GARIEL J.-C., MÈNEROUD J.-P., DERVIN
P., CUSHING M., GAUBERT C., VIDAL S., DUVAL A.-M., “A topographic effect
study in the Nice region”, Proceedings of the Fifth International Conference on Seismic
Zonation, Nice, France, 17–19 October, II, pp. 1067–1074, 1995.
[PAR 03] PAROLAI S., BARD P.-Y., “Evaluation of amplification and frequency-dependent
lengthening of duration of seismic ground motion due to local effects by means of joint
analysis of sonogram and standard spectral ratio”, Journal of Seismology, 7, pp. 479–492,
2003.
[PEC 85] PECKER A., La dynamique des sols appliquée aux séismes, Presses des Ponts et
Chaussées, 1985.
[SEE 69] SEED H.B., IDRISS I.M., “Influence of soil conditions on ground motions during
earthquakes”, J. Soil Mech. and Found. Div., ASCE, 95 (SM1), 1969.
 Seismic Ground Motion 123

[VUC 91] VUCETIC M., DOBRY R., “Effect of soil plasticity on cyclic response”, ASC
Journal of Geotechnical Engineering, 117(1), pp. 89–109, 1991.

List of accelerometry websites

European accelerometry data

European accelerometer data: http://www.isesd.cv.ic.ac.uk/ESD/frameset.htm

France: http://www-rap.obs.ujf-grenoble.fr
Greece: http://www.itsak.gr
Italy: http://www.serviziosismico.it, http://www.dstn.pcm.it
Switzerland: http://seismo.ethz.ch
Turkey: http://angora.deprem.gov.tr

Japanese accelerometry data

K-NET network (1,000 stations with geotechnical information): http://www.k-net.bosai.go.jp/

KiK-NET network (500 coupled surface-depth stations): http://www.kik.bosai.go.jp/

Californian accelerometry data

http://quake.crustal.ucsb.edu/scec/smdb
http://www.consrv.ca.gov/cgs/smip

World accelerometry data

http://agram.wr.usgs.gov/docs
http://mceer.buffalo.edu
http://perun.wdcb.rssi.ru/SMDB/
http://www.ednes.org/CGDS/
http://www-socal.wr.usgs.gov/smdata.html
National Geophysical Data Center (NGDC):
htpp://www.ngdc.noaa.gov/seg/hazard/strong.html
This page intentionally left blank
 Chapter 4

Soil Behavior: Dynamic

Soil-Structure Interactions

Introduction

As suggested by the name, soil-structure interaction aims at assessing the

response of a structure resting on the ground and subjected to any stimulation, whilst
taking into account coupling with the support medium and the soil, which has its
own deformability and even resistance characteristics. This interaction is expressed
through modifications of the motion of the soil near the structure as compared with
the open field – without any structure configuration; in the same way, the motion of
the structure is different from that which would result from the hypothesis of
an infinitely stiff soil: a structure “embedded” at its base corresponding to the
hypothesis taken into account in paraseismic building codes. This interaction is of
variable importance, depending on the nature of the soil, the characteristics of the
structure and its type of foundations. For light structures with superficial
foundations, it can be almost negligible. However, whenever it is necessary to study
the seismic response of a structure, and to consider it as an integral part of a whole
system consisting of the ground, foundations and structure, soil-structure interaction
analysis becomes increasingly important when building structures like dams, nuclear
facilities, liquefied natural gas tanks and very high buildings.

Chapter written by Alain PECKER.

126 Dynamic Behavior of Concrete and Seismic Engineering

Many factors need to be considered to completely solve a soil-structure

interaction problem. They include: the definition of the seismic unknown factor and
the resulting motion, soil behavior under cyclic stress, an estimate of soil response in
the open field and the response of structures under dynamic loading. In this chapter,
only those aspects dealing with the behavior of soils under cyclic loading will be
tackled, together with soil-structure interaction modeling.

The extensive literature dealing with the soil-structure interaction reflects both
the complexity of the phenomenon as well as the interest in the subject shown by the
scientific community. Two general syntheses have proposed a classification of the
methods used to take soil-structure into account ([LYS 78] and [IDR 80]). Both
publications stressed at the time that the study of interaction phenomena was
essentially limited to cases involving linear problems, approaches for non-linear
problems still being rare. However, in the last few years, important advances have
been made as far as our understanding of non-linear phenomena is concerned, which
makes broaching these problems possible.

4.1. Behavior of soils under seismic loading

4.1.1. Influence of the nature of soils on seismic movements

Observations made on sites during real earthquakes clearly show the influence of
the nature of surface layers on the seismic motion recorded. This fact has
been acknowledged for about 20 years and has led to response spectra being
interpreted according to the nature of the soil.

The recent major earthquakes that have affected the world (Mexico, 1985; Loma
Prieta, 1989; Northridge, 1994; Kobe, 1995) confirmed the following facts: alluvial
soils tend to amplify incident motion, especially at low frequency. As an illustration,
recordings of the Loma Prieta earthquake obtained on rocky sites around San
Francisco showed a maximum acceleration of about 0.10 g; recordings of the same
earthquake made on alluvial sites showed maximum accelerations two to three times
as high (Table 4.1 [COL 90]), with spectra presenting important low frequency
peaks. As the epicenter was far from the recording sites (about 85 km), this could
not be a local effect due to the source (directivity). By the same token, as the
recording sites were near to one another (within a radius of a few kilometers), it
could not be an effect due to propagation between the source and the site inside the
Earth’s crust. The only parameter that could have affected the nature of the recorded
motion was the geological nature of the sites, i.e. the mechanical characteristics of
the soil near the surface.
 Soil Behavior: Dynamic Soil-Structure Interactions 127

However, we should not deduce from the previous observations (or other similar
ones made in Mexico) that alluvial soils systematically amplify seismic
accelerations. If we again consider the San Francisco sites, it seems that at the time
of the 1957 earthquake, which also originated on San Andreas’s fault but nearer the
city (between 15 and 20 km), the accelerations registered on rocky sites were also
about 0.10 g, as the earthquake had a lower magnitude (5.3 instead of 7.1). Yet at
the surface of the alluvial sites, the recorded accelerations for the earthquake were
1.5 to 2 lower than those on the rock (from 0.05 g to 0.07 g).

To be able to estimate such differences, it is necessary to have a thorough

knowledge of soil behavior under cyclic loading so that it can be integrated into
elaborate calculation models. At the present time, even if many aspects remain to be
clarified, our knowledge of soil behavior has progressed and the calculations models
have developed sufficiently to allow an evaluation of the phenomena that will satisfy
engineers.

Ground maximum
Station Stratigraphy acceleration
1957 1989
Golden Gate Park Rock 0.13
Market/Guerrero St Rock 0.12
State Building Sand 0.10
+ clayed sand (60 m)
Mason/Pine St Rock 0.10
Alexander Building Clayed silt + sand (45 m) 0.07 0.17
Southern Pacific B. Soft clay 0.05 0.20
Rincon Hill Rock 0.10 0.09
Oakland City Hall Clay, sand (30 m) 0.04 0.26
+ stiff clay (270 m)

Table 4.1. Maximum acceleration in San Francisco (from [COL 90])

On the basis of these observations, the various paraseismic building codes

acknowledge the necessity of taking into account the geological nature of the soil in
the definition of seismic stresses. This is expressed in the way response spectra are
expressed differently according to the nature of the ground, which is characterized
by the average propagation speed of shear waves in the 30 uppermost meters of the
ground layer [COL 02].
128 Dynamic Behavior of Concrete and Seismic Engineering

4.1.2. Experimental description of soil behavior

Both the dynamic response calculations of a soil profile and soil-structure

interaction problems usually consider seismic motion are caused by a shear wave
propagating vertically from the substratum. Under such conditions, a soil element
taken from the soil layer is subjected to the stress cycles represented in Figure 4.1.

%u
T'0 T'0 T'0
%h
K0 T'0

K0 T'0 K0 T'0

Figure 4.1. Idealized loading sequence

Initially, in the case of a horizontal soil profile, the element is balanced under the
real vertical stress V’v and the real horizontal stress K0 V’v, where K0 is the thrust
coefficient of the ground at rest. The passing shear wave is discerned by the
application of a shear stress W(t) on the horizontal sides, and then on the vertical
sides of the soil element (to preserve equilibrium conditions). Under the effect of the
stress, the sample is subjected to a simple shear strain, which, for an elastic behavior
material, involves a volume variation equal to zero. The shear strain, also called
distortion, is described in Figure 4.2.

'u
Ȗ [4.1]
'h

If in a laboratory we reproduce symmetric and constant amplitude strain cycles

similar to those in Figure 4.1, we obtain the curves shown in Figure 4.2. The latter
shows that in the plane (W, J), soil behavior is characterized by a hysteresis loop, the
area and inclination of which is determined by the amplitude of the strain. The wider
the latter, the more important loop area is, and the less horizontally inclined it
becomes. In addition, it shows that experimentally, the shape of the hysteresis loop
is unaffected by the loading rate.
 Soil Behavior: Dynamic Soil-Structure Interactions 129

Figure 4.2. Cyclic strain-stress curve

When cyclic loading is not closed, the behavior becomes more complicated to
describe. An example of this is given in Figure 4.3. Up to b, the path followed is
identical to the one in Figure 4.2 (first loading curve then discharge curve). At point
b, where Vb < Va, the loading sign gets inverted again; the path followed is given by
curve bc, then possibly by cHc if the loading changes its sign again in c. Yet if the
loading is continued beyond c, the path is represented by ca, then beyond c it
follows the first loading curve anew.
130 Dynamic Behavior of Concrete and Seismic Engineering

Figure 4.3. Any cyclic loading

Associated with these shear strains, volumic strains also appear (Figure 4.4).
These irreversible volumic strains express the fact that soil behavior is non-linear.
These irreversible volumic strains cause hardening in a dry material. Thus, even for
closed cycles that are symmetric and have identical strain amplitudes, the hysteresis
loop obtained during the fourth loading cycle is different from that obtained during
the first cycle. The latter loop is less inclined on the horizontal and has a smaller
area. For a saturated, low-permeability soil, the strains occur at constant volume,
because the interstitial water does not have time to drain off the skeleton. However,
the tendency for volume variation exists, and involves an increase of interstitial
pressure, and therefore a decrease in effective strain.

The few simplified examples above show the complexity of the soil behavior,
which is highly non-linear and inelastic.

The modeling adopted in practice and described below only refers to the
deviatoric behavior of the soil (Figure 4.2). The volume variation aspects are often
neglected, except in cases where real elasto-plastic behavior laws are used.
 Soil Behavior: Dynamic Soil-Structure Interactions 131

Figure 4.4. Volume variations under cyclic loading

4.2. Modeling soil behavior

An exhaustive description of soil behavior can be obtained from a state of

balance, characterized by a stress field V and a strain state (characterized by tensor

H ), if it is possible to determine the stress field (or the new strain state)

corresponding to the new state of balance after applying strain increment d H (or

respectively of a stress increment d V ). In most cases, time also plays a part in the

expression of behavior law, but as a rule, for most soils, this parameter can be
neglected. Establishing the behavior law is the ultimate goal of the description of the
soil behavior. Nevertheless, because of the complexity of the experimental behavior
of the soil, establishing a true behavior law is both delicate and costly. At present, no
universal behavior law exists: each law available in literature has advantages as well
as drawbacks and limitations.

When confronted with these difficulties, geotechnicians often prefer to use

procedures commonly employed in soil mechanics, which involve anticipating the
stress mode to which an element isolated inside the soil layer will be subjected and
reproducing the corresponding loading path, either inside a sample or in situ. The
parameters measured during the test can then be used directly for the calculations.
For example, a similar approach is used in soil mechanics to study the subsidence of
a compressible clay layer due to the weight of a backfill, the extension of which is
important relative to the layer thickness: the test used would be an oedometric test
with no radial strain.

We have to realize that such an approach is not the same as determining a

behavior law, even if the stress-strain curves obtained are represented by
mathematical equations. This modeling of soil behavior is only valid for loading
132 Dynamic Behavior of Concrete and Seismic Engineering

paths close to those used to establish the model; extrapolation to different paths
would be erroneous and invalid. Furthermore, quite often, the modeling obtained in
this way only partly represents the physical phenomena; thus, the equivalent linear
visco-elastic modeling (which will be described later) does not describe or
incorporate volume variations (subsidence) under purely deviatoric loadings.
Moreover, the stress paths represented in tests are only rudimentary idealizations of
real stresses. This type of approach represents a compromise between the
phenomenon to be modeled and ease of implementation. If it is used with good
judgment, however, it is quite a powerful tool.

Before attempting an experimental description of the phenomena to be modeled

by mathematical representation, it is important to realize that on the time scales
involved in seismic stresses, most soils exhibit non-drained behavior during the
stress. Soil permeability is not sufficient (compared to the loading rate) to permit
drainage. As a consequence, in the approach described above, we reason in terms of
total stresses. Once again, this is a simplification, as real soil behavior is controlled
by the actual stresses.

The rest of this text will be limited to examining soil behavior before failure. The
study of soil behavior on failure gives rise to different approaches made necessary
by the adopted schematization. If we had a real behavior law at our disposal, such a
distinction would not be necessary; in fact, the behavior law would allow us to
reproduce soil behavior from the smallest strains (quasi-elastic strains) to the very
high strains associated with failure.

For further descriptions of soil behavior, see [HAR 78], [PEC 8] and [PRE 78].

What emerges from the experimental statements in section 4.1.2 is that the soil
cannot be represented by an elastic model, at least as soon as the strains become
significant.

The non-linearity appearance thresholds generally correspond to low strains (in

the range 10-4 to 10-6). However, we must distinguish between reversible (or quasi-
reversible) and irreversible non-linearities, as the appearance thresholds of the latter
are higher (10-4 to 10-3).
 Soil Behavior: Dynamic Soil-Structure Interactions 133

Cyclic
Behavior Elasticity and Analysis
degradation for
linearity plasticity method
saturated soils
Very low 0 d J d Js Practically Practically Non-degradable Linear
linear elastic
Low Js d J d Jv Non-linear Weakly Practically non- Linear
Elasto-plastic degradable equivalent
Average Jv d J Non-linear Elasto-plastic Degradable Non-linear
to large

Table 4.2. Behavior field for cyclic stresses

The appearance thresholds for these non-linearities, respectively named Js and Jv,
depend on the nature of the material, which is roughly characterized by its plasticity
index IP.

Table 4.2 and Figure 4.5 allow us to characterize each of the behavior fields as
well as the analysis methods used to express their behavior in digital studies.

Figure 4.5. Shear cyclic strain

The stresses generated by major earthquakes expected in the French context

could induce strains causing significant linearity losses, or even irreversible strains
in the highest cases (where J t Js or Jv).
134 Dynamic Behavior of Concrete and Seismic Engineering

As indicated in the previous table, each domain corresponds to a behavior type,

the characterization of which involves measuring specific parameters.

4.2.1. An experimental description of elastic soil behavior (J d Js)

For strains lower than 10-6 to 10-5, soil behavior is typically linear elastic. Some
saturated materials can present a slight damping of viscous origin (a few percent), in
which case soil behavior can be characterized by an elastic or possibly visco-elastic,
linear type of behavior law. In the case of an isotropic material, the modulus of shear
elasticity G (similar to Lame’s modulus P in continuous medium mechanics) and a
volumic bulk modulus B allow the complete characterization of the behavior.
Alternately, it is possible to use the propagation rate of the elastic waves Vs (shear
waves) and Vp (compression waves) linked to the previous values by:

§ 4 ·
G = ȡ Vs2 B = ȡ ¨ Vp2  Vs2 ¸ [4.2]
© 3 ¹

where U is the density of the material.

Measuring these values can be either done in situ (Vs, Vp) or under laboratory
conditions on intact samples (G, B).

The measurements correspond to very specific techniques. Among the in situ

techniques used, we include the logging suspension test, spectral analysis of SASW
surface waves, or any other drilling measure technique (crosshole or downhole test)
as examples; in the laboratory, the most appropriate test is the resonant column test.

The study of soil behavior within its elastic field is important, and a whole range
of problems exist for which the models are valid: cases of vibrations of well
conditioned machine blocks, low amplitude seismic stresses like those caused by
geophysical tests are examples.

4.2.2. Linear visco-elastic models for medium strain domains where Js d J d Jv

In this strain field, more or less marked non-linearities appear in the stress-strain
curve.
 Soil Behavior: Dynamic Soil-Structure Interactions 135

Figure 4.6. Linear visco-elastic model

As the visco-elastic models reveal hysteresis loops under harmonic stresses, it is

tempting to represent soil behavior using such models (Figure 4.6) for those stresses.
However, the linear visco-elastic model leads to an energy dissipation cycle
depending on the stress frequency, which contradicts the experimental observations
(section 4.1).

We must establish equivalence between the actual material and the model: the
equivalence is based on the equal dissipative qualities of both the material and the
model.

For one-dimensional stresses, the stress-strain relationship in the Kelvin-Voigt

model is written as follows:

IJ = G Ȗ + C Ȗ [4.3]

where G and C are the spring and dashpot constants, and J and J are respectively
the strain and strain rate.

Under harmonic stress:

Ȗ = Ȗ m eiȦt [4.4]

the previous relationship is written:

ª CZ º
Wm = G «1 + i » J m = G* J m [4.5]
¬ G ¼
136 Dynamic Behavior of Concrete and Seismic Engineering

where G is a complex modulus.

The dissipated energy during a loading cycle is equal to:

'W = ʌ C Ȧ Ȗ 2m ʌ Im (G ) Ȗ 2m [4.6]

where Im (G*) refers to the imaginary part of G*.

As C is constant, we note that the dissipated energy depends on the stress

frequency f = Z/2S.

If we refer to Figure 4.7, without going into too much detail, it is possible to
define a damping term for a material that is frequency-independent, a standard
expression of the energy 'W dissipated during a cycle. The standardization is made
owing to the elastic energy W accumulated during a cycle:

1
W = G J2 [4.7]
2

Therefore, we can define either a damping ratio E or a loss coefficient K,

independent of the stress frequency:

1 'W
K = 2E = [4.8]
2S W
 Soil Behavior: Dynamic Soil-Structure Interactions 137

Figure 4.7. Definition of standard damping

The energy dissipated during a cycle then takes the form:

'W = ʌ G Ș Ȗ 2m [4.9]

The purpose of the equivalent linear visco-elastic models is to establish a

relationship of the form:

IJm G Ȗ m [4.10]

*
where G is a complex modulus that has to be chosen so as to involve the same
stiffness and energy dissipation properties as the real material. Several models have
been proposed for this purpose. Their characteristics are summarized in Table 4.3.

The first two models are due to Berleley’s team [SEE 70]; the third was
developed by [DOR 90]. We note from Table 4.3 that the first model respects the
dissipated energy but overestimates stiffness; the second model respects stiffness but
underestimates dissipated energy. Only the third model satisfies both parameters.
138 Dynamic Behavior of Concrete and Seismic Engineering

Energy dissipated during *

Complex modulus ~G ~
a AW cycle
G* = W/J modulus

Material
S G K J2m G

Model 1
G [1 + i K] S G K J2m G 1 + K2

GeiT

Model 2 T K2
K = 2 sin S G K J 2m 1 -
2 4 G

Model 3
G 1 - K2 + i K S G K J2m G

Table 4.3. Characterization of equivalent linear visco-elastic models

An alternate representation of the data in Figure 4.2 is obtained by plotting both

the norm of the secant modulus of the hysteresis loop (G/Gmax) and the equivalent
critical damping ratio E, as a function of strain (Figure 4.8). This figure clearly
shows the notion of strain threshold (Js), for which neither strain nor its dependence
on a plastic index are constant.
 Soil Behavior: Dynamic Soil-Structure Interactions 139

Figure 4.8. Characteristic results for clay

Correlatively, Figure 4.9 shows that for cyclic shear strains J lower than Jv | 10-4,
volume strain can be neglected. For saturated soils the result is an insignificant
variation of the interstitial pressure, and therefore deterioration of properties is
absent.

To conclude, within this strain range, soil behavior is definitely non-linear,

though it stays markedly elastic because the permanent changes in its microstructure
are negligible.
140 Dynamic Behavior of Concrete and Seismic Engineering

In the linear visco-elastic modeling described below, the properties of the

material (secant modulus, damping) are adjusted to a “medium” strain level to
approximately account for the non-linearity of the behavior.

The results developed above in the case of unidirectional stresses immediately

become generalized in the case of harmonic stress with all the limitations for the
experimental validity of such an extension that we have mentioned previously.

The behavior law is written in a form similar to a generalization of Hooke’s law:

*
H = / : V [4.11]
~ ~ ~

*
where tensor / is formed thanks to the complex volume and shear moduli. The
~
previous behavior law leads to the same shaped solutions as the linear elastic law,
hence its unquestionable appeal.

In the case of an isotropic material, the behavior is entirely described using two
moduli, G and B, and theoretically, using two damping ratios associated with the
shear and volume strains, as well as their dependence on the amplitude of the shear
strain. In practice, both critical damping ratios are chosen to be identical.

The variation of these characteristics with regard to strain takes the shape given
in Figure 4.8. It is the conventional form for the results of the tests:

G = G(Ȗ) , Ș = Ș(Ȗ) [4.12]

As practice stands at the moment, measuring these values can only be performed
under laboratory conditions with intact samples. The trial best suited to the range of
strains required is the resonant column test.

Such models are extensively employed in common practice. When they are used
together with an iterative process allowing values of modulus G and the loss
coefficient K to be selected as those compatible with the average level of resultant
strain, they yield strain and acceleration values that compare favorably with those
obtained using more sophisticated models or observed during real earthquakes. Such
models have the virtue of simplicity: they only require measurement of three
parameters (one more than the elastic model): shear modulus, volume modulus and
loss coefficient. As the soil has a non-linear behavior, these parameters depend on
the state of stresses and strains (Figure 4.8). The main limitation of these models is
the fact that they are unable to give irreversible strain values: those calculated by the
 Soil Behavior: Dynamic Soil-Structure Interactions 141

model are necessarily equal to zero. Whenever a more accurate representation of the
soil is necessary, use of non-linear models is inevitable.

4.2.3. High strain domain non-linear models where J t Jv

In this strain domain, important changes in micro-structure occur (rearrangement

of grains), which generates significant volume and shear irreversible strains (Figures
4.3 and 4.9). These microstructure changes involve volume variations inside
unsaturated materials, and an increase of interstitial pressure within saturated
materials. This interstitial pressure increase can lead to canceling the actual stresses
and therefore the soil strength: it is a liquefaction phenomenon. Apart from this loss
of strength, the increase in interstitial pressure brings about a decrease in the initial
stiffness, which is a function of the state of the real stresses withstood by the soil
(Figure 4.10). Both factors combined – the stiffness loss and the loss of strength –
make saturated soils subjected to amplitude strains in which J is higher than Jv
highly degradable.

As this type of behavior generates significant irreversible strains, it can only be

studied using non-linear models. Accumulated experience shows that strain
hardening elasto-plastic models are the most appropriate for describing this behavior
([PRE 78], [PRE 87]); the models have to be written using real stresses as these
govern soil behavior. In saturated soils, coupling of the fluid phase with the solid
phase can be described within the frame of porous medium mechanics due to Biot’s
relationship.

Measuring the values that are used to formulate the behavior laws must be
carried out in laboratory conditions, on intact samples, under strictly controlled
environment conditions and over a range of stress paths that are all the more
numerous because the behavior law includes a greater number of parameters. In any
case, these tests are difficult to implement, because taking geo-material samples is
always delicate and can induce sample modifications, especially deep sampling. The
nature of the soil to be sampled (clay, marl) plays a part in the extent of the
modification. The properties most susceptible to modifications are soil stiffness and
the capacity of saturated sands to withstand liquefaction. At present, the most
versatile tool for obtaining this data is the triaxial device, which makes it possible to
reach important strain amplitudes with different stress paths.
142 Dynamic Behavior of Concrete and Seismic Engineering

s
4 different preperation techniques

Figure 4.9. Irreversible interstitial pressure and volumic strain

 Soil Behavior: Dynamic Soil-Structure Interactions 143

Figure 4.10. Variation of hysteresis loops with the number of cycles

4.3. Linear soil-structure interactions

For the study of the soil-structure interaction in linear systems, the models
adopted to represent the soil are either the elastic model (section 4.2.1) or the
equivalent linear visco-elastic model (section 4.2.2). The non-linearities in the
behavior of the soil can be taken into account by selecting G shear modulus and E
damping modulus values compatible with the average strain induced in open field by
the seismic stresses. This approach involves neglecting additional non-linearities
linked to the soil-structure interaction, such as soil plastification at the ends of
superficial foundations, or along the pile shafts. Comparison of the results obtained
with non-linear calculations shows this approach gives acceptable results when
foundations show sufficient safety regarding their ultimate load.

4.3.1. Illustration of the soil-structure interaction effect

The influence of soil-structure interactions on the response of a building can be

illustrated using the analog model shown in Figure 4.11 [WOL 85]. The structure is
represented by a mass and a spring placed at a height h above the foundations. The
bond between the structure and its foundations is represented by a stiff bar. The
latter is lying on the ground, and its interaction is modeled via an impedance
function for the foundations. In this model, what stands out is the simplified
representation of impedance functions as springs and dashpots that are independent
144 Dynamic Behavior of Concrete and Seismic Engineering

of frequencies; the dashpot theoretically explains both radiation damping and

material damping as defined in section 4.2.2. With a view to simplifying the
presentation, we assume that material damping can be neglected relative to the
radiation damping (elastic behavior of the soil), which is a valid assumption for a
homogenous medium and low to medium amplitude seismic stresses.

Figure 4.11. Soil-structure interaction simplified model

The system in Figure 4.11 has 3 degrees of freedom:

– the horizontal displacement u1 of mass m;
– the displacement u0 of the foundations;
– the rotation T of the foundations,

and is subjected to a horizontal displacement of the supporting soil, with a pulsation

harmonic Z and an amplitude ug.

The dynamic balance equations of the system can be derived from Lagrange’s
equations [CLO 75] if we take qi as generalized variables:
– q1 = u, relative displacement of the mass in relation to A;
– q2 = u0, displacement of the foundations;
– q3 = T, rotation of the foundations.

The relationship between the absolute displacement and ut of the mass m and the
previous variables is obvious:

ut u g + u 0 + u + hș [4.13]
 Soil Behavior: Dynamic Soil-Structure Interactions 145

with T as the total kinetic energy:

1  2
T= m (u g  u + hș) [4.14]
2

and V is the potential energy:

1
V
2
 2
ku 2  k h u 0  k T T2  [4.15]

and GW is the work of non-conservative (damping) stresses:

įW =  Cuįu  
 + Ch u 0 įu 0  Cș șįș  [4.16]

Lagrange’s equations are written as follows:

d § wT · § wT · § wV · GW
¨ ¸¨ ¸¨ ¸ [4.17]
dt © wq i ¹ © wq i ¹ © wq i ¹ Gq i

Using the previous notations and taking the acceleration, speed and displacement
relationships into account, this means:

x =  Ȧ2 x
x = iȦx,  [4.18]

When we introduce the critical damping ratios, we obtain:

iȦC iȦCh iȦCș

ȟ= , ȟh ȟș [4.19]
2k 2k h 2k ș

­mZ2 (u 0  u  hT)  k(1  2i[)u mZ2 u g

°° 2 2
®mZ (u 0  u  hT)  k h (1  2i[h )u 0 mZ u g [4.20]
° 2 2
°̄mhZ (u 0  u  hT)  k T (1  2i[T )T mhZ u g
146 Dynamic Behavior of Concrete and Seismic Engineering

When we introduce the following notations:

mȦs2 K , mȦh2 k h , mh 2 Ȧș2 kș [4.21]

and then eliminate u0 and T between the three previous equations, we obtain:

ª Z2 Z2 1  2i[ Z2 1  2i[ º Z2
«1  2i[  2  2  »u ug [4.22]
¬ Zs Zh 1  2i[ h ZT2 1  2i[T ¼ Zs2

Taking into account the fact that [, [h, [T  1, the previous equation becomes:

ª Z2 Z2 Z2 º Z2
«1  2i[  2  2 (1  2i[  2i[ h )  2 (1  2i[  2i[T ) » u ug [4.23]
¬ Zs Zh ZT ¼ Zs2

Let us now consider a simple oscillator with 1 degree of freedom and the same
~ and damping ~
mass m, with its characteristic pulsation Z [ submitted to the
harmonic displacement ~
u g with a pulsation Z at its base (case of a structure
embedded at its base). The harmonic response of the oscillator is given by:

§ Z2 · Z2
¨¨1  2i[  2 ¸¸ u u g [4.24]
© Z ¹ Z 2

The equivalent oscillator will have the same response as the structure in Figure
4.11 if the following equations are verified:

1 1 1 1
 2  2 [4.25]
Z2 2
Zs Zh ZT

2
Z 2
Z 2
Z
[ [  [ h  [- [4.26]
Zs2 Zh2 ZT2
 Soil Behavior: Dynamic Soil-Structure Interactions 147

2
Z
u g ug [4.27]
Zs2

Figure 4.12. Influence of the soil-structure interaction

148 Dynamic Behavior of Concrete and Seismic Engineering

The previous equations are obtained by equating the real part with the imaginary
part of equations [4.23] and [4.24] and for [4.27] by assuming a resonance situation
(Z = Z~ ).

Equations [4.25] and [4.26] assume the soil-structure interaction results in:
– decreasing the characteristic pulsation Zs of the embedded basis structure (Z 
Zs);
~
– increasing the damping of system ( [ ! [) with regard to the embedded basis
structure;
– decreasing the effective incident stress at the basis of the structure ( u g  ug).

The conclusions are shown in Figure 4.12, which represents a circular

foundations lying on an homogenous elastic semi-space, and the relative variations
Z /Zs, [ , u g /ug are a function of the non-dimensional parameters:

h Ȧs h m
h 1,s= , m [4.28]
r Vs ȡr 3

in which r is the radius of the foundations, and U and Vs are respectively the volumic
density and rate of the waves S within the soil (equation [4.2]).

Figure 4.12 clearly shows that the influence of the soil-structure interaction is all
the more marked if the foundation soil is soft (increasing s) or if the structure is
massive (increasing m).

4.3.2. Expression of a soil-structure problem

Before examining the different methods employed to take soil-structure

interactions into account, the problem is worth formulating from a general point of
view. This formulation aims at dealing with phenomena due to finite elements. In
fact, the problem is so complex that resorting to digital methods cannot be avoided,
but in the rest of this chapter, we will try to point out steps that can be dealt with
analytically and those that are amenable to existing solutions.

Motion equations are obtained by referring to Figure 4.13, which schematizes a

soil-structure unit.
 Soil Behavior: Dynamic Soil-Structure Interactions 149

Figure 4.13. Superposition theorem for the soil-structure interaction

If we refer to the mass, damping and stiffness matrices as [M], [C] and [K], the
equation of motion is written:

> M @^u`  >C@^u `  >C@^u ` ^Qf ` [4.29]

As the source of the motion (seismic focus) is generally not included in the
model, the load vector ^Q f ` has values not equal to zero only on the outer border of
the model.

When there is no structure, the equation of the open field motion is similar in
shape to that described by [4.29]. With the f indexes designating the mass, damping
and stiffness matrixes related to the open field alone, this equation will be:

> Mf @^u f `  >Cf @^u f `  > K f @^u f ` ^Qf ` [4.30]

Equation [4.30] can only be solved by making certain assumptions about the
nature and the direction of the incident field propagation.

Assuming that:

^u` ^u i `  ^u f ` [4.31]
150 Dynamic Behavior of Concrete and Seismic Engineering

Equation [4.31] defines the interaction displacement ^u i ` which satisfies the

equation:

> M @^u i `  >C@^u i `  > K @^u i `  ^Qi ` [4.32]

with:

^Qi ` ª¬> M @  > M f @º¼ ^

u f `  ª¬> C@  > Cf @º¼ ^u f `  ª¬> K @  > K f @º¼ ^u f ` [4.33]

The load vector ^Qi ` is determined from the open field displacements. For linear
systems, the superposition theorem is illustrated in Figure 4.13 [LYS 78]: the
interaction problem is divided into the sum of an open field soil response problem
[4.30] with a source problem [4.32] where the applied forces ^Qi ` have components
not equal to zero only at nodes common to the soil-structure. This last problem is
analogous to a machine vibration problem. The total displacement for the integration
problem is then given by equation [4.31].

Equation [4.33] clearly shows that there is interaction as soon as a mass or

stiffness difference develops between the soil and the structure. To simplify matters,
let us suppress the damping term in the equation and restrict the problem to one of a
structure lying on the ground, being subjected to the vertical propagation of volume
waves (shear or compression). Under such conditions, in the open field, all points of
the ground surface will be driven by the same motion. If the foundations of the
structure are infinitely stiff, the last term of the equation becomes equal to zero, and
the load vector ^Q i ` is restricted to:

^Qi ` ª¬> M @  > M f @º¼ ^

uf ` [4.34]

The stresses ^Qi ` applied at the basis of the structure generate a movement of
the support that is equivalent to a force of inertia field within the structure.
Consequently, interaction only results from the inertia developed inside that
structure. This is referred to as the inertial interaction, and its effect is illustrated by
the example in section 4.3.1.

On the other hand, let us consider a structure laid in the ground, the mass of
which is equal to zero outside the ground, and equal (in value and repartition) to the
 Soil Behavior: Dynamic Soil-Structure Interactions 151

mass of the soil for the part that is in the ground. The expression for loads ^Qi `
becomes:

^Qi ` ª¬> K @  > K f @º¼ ^u f ` [4.35]

These only result from the soil and structure stiffness difference for the part
embedded in the ground. Even without any mass difference, there is interaction; it is
referred to as kinematic interaction. It results from the stiffness of the foundations
that prevents them from following motions imposed by the ground. We saw before
that it was strictly equal to zero in some cases; it can be low in others (foundations
on flexible piles) or highly significant (stiff structure strongly cross-braced and sunk
deep in the ground).

In the most general case, interaction comprises both an inertial interaction and a
kinematic interaction.

Figure 4.13 and the previous arguments illustrate the two best methods for
solving the soil-structure interaction. Figure 4.13a corresponds to the overall
methods, the solutions to which are obtained via direct resolution of equation [4.29].
They do not resort to any notion of superposition, and thus they are theoretically
suited to non-linear problems. Alternatively, sub-structure methods rely either on the
division of Figure 4.13b-4.13c, or on similar divisions to solve the problem step by
step. These methods can indeed only be applied to linear problems suitable for
superposition.

4.3.3. Superposition theorem

Dividing the soil-structure interaction into inertial and kinematic interactions (as
discussed in section 4.3.2) not only has the advantage of allowing us to illustrate the
essential phenomena, it also generates a resolution method based on the sub-
structuring principle, the validity of which relies on superposition theorem [KAU
78], [ROE 73]. This theorem establishes that the response of the model in Figure
4.14 (on the left) subjected to a üg acceleration on its basis can be obtained the
following ways:
– either in one step by solving the equation:

> M @^u`  > K @^u`  > M @^I` u g [4.36]

152 Dynamic Behavior of Concrete and Seismic Engineering

where:

– ^u` represents the vector of the relative displacements of the system with
regard to the foundation;
– ^I` represents a unit vector giving the direction of the stress ug ;

– > M @ and > K @ represent the mass and stiffness matrices of the system.

To simplify matters, the damping terms are omitted;

– or in two steps by expressing the vector of relative displacements ^u` as the

sum of the “kinematic” displacements ^u cin ` and of the inertial
displacements ^u iner ` :

^u` ^u cin `  ^u iner ` [4.37]

and simultaneously solving the two equation systems [4.38] and [4.39] below:

u cin `  > K @^u cin `

ª¬M sol º¼^  ª¬M sol º¼^I` 
ug [4.38]

> M @^u inert `  > K @^u inert ` u cin `  ^I` 

 ¬ªM st ¼º ¬ª^ u g ¼º [4.39]

in which [Msol] and [Mst] represent the mass matrices of the soil and structure parts:

>M@ ª¬M sol º¼  ª¬M st º¼ [4.40]

The equivalence of [4.36], [4.38] and [4.39] is obtained by a mere addition of the
two last values whilst taking [4.37] and [4.40] relationships into account.
 Soil Behavior: Dynamic Soil-Structure Interactions 153

Figure 4.14. Superposition theorem (from [KAU 78])

Equation [4.38] gives the response of a structure without a mass to seismic stress
ug . The solution gives kinematic interaction displacements that are used as loads in
[4.39], when imaginary inertial forces are applied to the structure.

When resolving equation [4.39], the modeling of the soil is indifferent: it can be
represented either with finite elements or using a stiffness matrix representing
foundations and the soil defined at the soil-structure interface. This stiffness matrix
results from the condensation of all degrees of freedom of the soil at the interface
[PEC 84]; the condensation is only possible for a resolution in the frequency field.
Within such a framework, the stiffness matrix is formed from the complex moduli
(section 4.2.2), which take damping into account. The stiffness matrix consists of a
real part (representing the stiffness of the foundations) and a fictitious part that
integrates all the damping phenomena (material and radiating). The terms of the
matrix depend on the frequency.

In the case of a structure with rigid foundations, it is legitimate to replace the

stiffness matrix (NxN), N being the number of interface nodes, with a (6x6) matrix
that gives the rigid body movements of the foundations; this matrix is called an
impedance matrix, and it can be conceptually represented by springs and dashpots
depending on the frequency. The result is that the solution of the kinematic
interaction problem is completely defined by rigid body movements of the
weightless structure; the latter can then be replaced by weightless foundations
subjected to the same seismic stress.

Examining the structure of equation [4.39] reveals that the solution ^u inert ` can
be interpreted as the displacement vector that is related to a fictitious support
154 Dynamic Behavior of Concrete and Seismic Engineering

submitted to the rigid body movements (translations and rotations) of the

foundations.

Thus, with rigid foundations, the overall problem can be divided into three sub-
problems:
– determining the motion of weightless rigid foundations subjected to the seismic
stress, this step representing the solution to equation [4.38];
– determining the impedance matrix of the foundations, this matrix including a
real part and a fictitious part which both depend on the frequency;
– calculating the dynamic response of the structure linked to the impedance
matrix and its supports subjected to the kinematic interaction movement calculated
in the first problem).

Insofar as the foundations are perfectly stiff, the procedure is strictly identical to
that leading to the one-step resolution of the overall system (equation [4.36]). The
interest of such division clearly appears whenever it is possible to simplify one of
the three steps of the calculation.

The diffraction problem (step a) still exists except for structures with shallow
foundations subjected to the vertical propagation of volume waves; in such cases,
resolution of step a) is identical to that of the response of an open field soil profile,
since the kinetic interaction is equal to zero. The second step solution can be
avoided for certain configurations by using the results from impedance functions
published in literature. The third step is essential; however, it is simpler and more
familiar to engineers, as it originates in conventional dynamic analysis of the
structures

4.3.4. Practical modeling of the soil-structure interaction

One of the main arguments proposed for using sub-structure methods rather than
global methods is that they are both easy to use and less expensive. It is probably
true for structures with shallow foundations submitted to vertical volume wave
propagation; in this case, we have seen the kinematic interaction is equal to zero,
which suppresses the diffraction problem resolution step, and that analytic or already
published solutions to the impedance problem exist. Besides, if the reference
movement is defined at the ground surface, the only possible step is dynamic
analysis of the structure. Sub-structure methods have the advantage of allowing
some modifications without making it necessary to start the whole analysis again: a
modification of the characteristics of the structure does not require a new dynamic
analysis of the structure; modifying the characteristics of the project earthquake
enables the use of the impedance problem solution.
 Soil Behavior: Dynamic Soil-Structure Interactions 155

As soon as the study structure is either partially or completely buried, resolving

the diffraction problem becomes complex; the rigorous solution becomes as delicate
to obtain as the global solution to the problem. Nevertheless, simplified methods to
take this kinematic interaction into account have been proposed in the case of stiff
foundations ([HAL 75], [KAU 78]). It is nonetheless true that, in the most general
case, sub-structure methods become less competitive than global methods. Besides,
the latter can solve non-linear problems as well, yet they remain limited to two-
dimension geometries.

In the rest of this chapter, we mainly endeavor to enhance particular aspects

related to soil modeling. For structure specific problems, the reader can refer to
[CLO 75]. However, we restrict ourselves to finite element calculations, which are
the only ones possible when implementing either global or sub-structure methods
requiring resolution of the diffraction problem.

4.3.4.1. Model of soil behavior

In practice, we use the equivalent linear visco-elastic model. This model has the
advantage of allowing the non-linearities of the behavior to be taken into account
owing to an iterative pattern. It is especially well suited to global methods; with each
iteration, the properties of the soil (moduli and damping) are adjusted, within each
element, to the average strain while the element is stressed.

When using sub-structure methods for linear systems, the non-linearities of the
soil are taken into account by using those characteristics that are compatible with the
average strain level of each open field element.

4.3.4.2. Nature and propagation direction of the incident wave

The commonly used model is the plane wave type, which assumes that the
horizontal movement of the incident field is created by the vertical propagation of
SH shear waves, and vertical movement is created by the vertical propagation of P
compression waves. These assumptions give no indication of the resulting
movement obtained after the waves reverberate on the structure. Besides, we can see
that waves generated after interaction with the structure mainly consist of Rayleigh
waves.

Other hypotheses related to the nature of the incident field are possible and can
be taken into account. Breaking down the displacements into open field and
interaction displacements shows that once the open field displacements have been
calculated, the nature of the incident field no longer explicitly influences the
formulation of the interaction problem ([4.32] and [4.33]). Sub-structure methods
are quite well-adapted to taking diverse incident fields into account, providing the
open field movement can be determined.
156 Dynamic Behavior of Concrete and Seismic Engineering

4.3.4.3. Geometrical modeling of the medium

The finite element formulation is standard; nevertheless, some conditions have to
be respected. Transmitting high frequencies imposes a maximum dimension on the
elements, at most equal to a fraction of the corresponding wavelength. Typically, we
use a value between 1/8 and 1/5 of the wavelength:

1 1 Vs
h max d to [4.41]
5 8 f max

where fmax represents the highest frequency to be transmitted, and Vs the propagation
rate of the shear waves. This criterion is generally applied to the vertical dimension
of the mesh because, considering the generally admitted assumption of wave vertical
propagation, the displacement field varies faster vertically than horizontally,
especially some distance away from the structure.

The extension of the finite element mesh constitutes one of the most critical
problems in the resolution of a dynamic problem involving propagation phenomena
using the finite element method. As a matter of fact, without any special conditions,
the side and lower limits of the model are open surfaces that completely reverberate
the wave fronts that hit them. The energy carried by these waves is reverberated
back to the structure instead of being carried ad infinitum inside the soil. As the only
energy dissipation takes place through material damping, the model has to be
extended so that the waves reverberated at the limits do not reach the structure while
its response is being estimated. The procedure soon makes the calculation cost
prohibitive.

To free us from such reverberations, some special devices called absorbing

boundaries have been developed. Located at the ends of the model, these boundaries
are supposed to represent the exact stress conditions existing at that limit, due to the
presence of soil outside the model.

Generally speaking, the side boundaries of the model can be divided into local
boundaries or consistent boundaries.

The local boundaries generally consist of localized dashpots, the characteristics

of which depend on the mechanical properties of the medium around them. These
boundaries do not couple the different degrees of freedom of the nodes along the
boundary and perfectly absorb only the waves with a normal incidence [LYS 69].
They can advantageously be implanted into time or frequency calculations.
 Soil Behavior: Dynamic Soil-Structure Interactions 157

Unlike local boundaries, consistent boundaries couple all the degrees of freedom
of the boundary nodes and perfectly absorb all kinds of waves. Formulating these
boundaries ([KAU 74], [LYS 72] and [WAA 72]) involves frequency dependent
terms; therefore, they can only be used for resolutions in the frequency field. There
are no consistent boundaries to represent the effect of the semi-space underlying the
model. This is why the lower boundary of the model is supposed to be rigid. If it is
chosen to be deep enough (about a structure wide), the reverberation phenomena on
this boundary become negligible. Actually, the field of the waves reverberated by
the structure mainly consists of surface waves which die down fast with depth.
Besides, we can take advantage of this property by having a moving lower
boundary, with a variable meshing, which is deeper the lower the studied
frequencies. In fact, low frequencies die down more slowly but they require coarser
meshing to ensure a correct transmission of the wave [LYS 81].

4.3.4.4. Digital integration pattern

It is theoretically possible to choose a digital integration pattern that is either a
time, a frequency or a mode pattern. These aspects will not be explained in detail
below; for more information, see [PEC 84]. We will only mention that for linear
problems, the frequency integration pattern is best suited to solving soil-structure
interaction problems because of the formulation of the soil behavior law (section
4.2), the dependence on frequency of the impedance matrices used in sub-structure
methods, and the absorbing boundaries used to estimate the kinematic interaction
displacements or for global method resolutions.

However, if a time or mode integration pattern is preferred, the variation of

impedances with frequencies has to be taken into account. This can be implemented
either by successive iterations allowing us to adjust the impedance to the frequency
of the soil-structure interaction mode, or by resorting to analog models using
springs, dashpots and additional mass (Figure 4.15).

Such models, which originate in cone models [WOL 95], lead to very satisfying
approximations of the impedance functions [PEC 94]. Nevertheless, due to the
presence of an additional mass, they require the use of precautions related to the
definition of the actual movement of the foundations. They can be equally used for
either mode analysis or time integration.
158 Dynamic Behavior of Concrete and Seismic Engineering

(a) (b)

Figure 4.15. Analog model for the impedance matrix

4.4. Non-linear soil-structure interactions

In most cases, linear modeling of the soil-structure interaction represents a

reasonable approximation of the phenomena. However, extreme situations do exist,
in which taking into account the geometric or behavior non-linearities is the only
possible approach representative of the phenomenon. Taking these phenomena into
account currently represents a topic that is developing fast, and for which important
progress has been achieved in recent years. Hereafter, we restrict ourselves to
identifying the phenomena and superficially point to the way to deal with them.

4.4.1. Geometric non-linearities and uplift of the foundations

It has often been observed after major earthquakes that slender structures with
shallow foundations show good behavior and seem to have been protected from
seismic aggression by their propensity to move independently of the foundation soil;
these additional “degrees of freedom” are provided by the possible uplift at the soil-
foundation interface. This uplift is expressed by transformation of the kinetic energy
into potential energy, due to the raising of the center of gravity, and through
lengthening of the vibration periods. According to the frequency content of the
stress, the longer vibration period can shift the vibration mode towards a lower
amplification spectrum region.

4.4.2. Non-linearities of behavior

Such non-linearities come from the local plastification of soil near the foundation
element. In the case of pile foundations, plastification takes place along the upper
part of the pile shaft, at a height typically equal to 4 or 6 diameters; for superficial
 Soil Behavior: Dynamic Soil-Structure Interactions 159

foundations, the plastification is likely to take place at the angles of the base, owing
to the oscillation moment of the structure.

The extreme soil plastification case corresponds to mobilizing the ultimate

capability of the foundations. If loading is applied permanently, such a situation
would bring about failure through loss of the supporting capability of the
foundations. Because under seismic stress, the applied loading varies with time, this
temporary mobilizing of the strength capability does not lead to failure, but the
appearance of irreversible displacements. These are not necessarily harmful to the
good behavior of the structure. Their evaluation allows us to design foundations by
relying on performance criteria (performance-based design) rather than on a safety
criterion with regard to “failure” [PEC 00].

4.4.3. Modeling the non-linear soil-structure interaction

The most direct method for taking geometric or behavior non-linearities into
account is still the finite element method. Thanks to a behavior law adapted to the
materials and the soil-structure interface elements, this method is the most versatile.
However, we should not lose sight of the fact that implementing this method is quite
complex and requires modeling, digital analysis and soil and structural dynamics
competences. Furthermore, in spite of considerable computer advances, the
calculation times are still quite long. The result is that the method is best suited to
verifications, but not to pre-dimensioning, as this requires many varying studies.

To get round such difficulties, sub-structuration methods similar in spirit to the

methods developed for the study of linear phenomena have been developed recently.
We merely present the outlines that give birth to the dynamic macro-element
concept. For a more comprehensive presentation of the subject as well as an
exhaustive bibliographic study of those methods, see [CRE 01a], [CRE 01b] and
[CRE 02].

The general philosophy of the method involves defining two sub-structures: the
soil and the soil-foundation interface on the one hand, and the structure on the other.
The division is made at the level of the foundations. The soil-interface sub-structure
is conceptually sub-divided into the neighboring field and far field (Figure 4.16); the
exact boundary between both sub-fields is unknown, but it does not explicitly
intervene in the macro-element concept.

The far field corresponds to the region where the soil-structure interaction can be
neglected; the behavior in this area is governed by the propagation of seismic waves
and the energy dissipation there is mainly viscous radiative damping. Quite
160 Dynamic Behavior of Concrete and Seismic Engineering

naturally, and by analogy with the impedance concept, the distant field is modeled
using a spring and a dashpot independent of the frequency for simplification.

The neighboring field represents the medium part interacting with the
foundations: all the behavior non-linearities (geometric, material) are concentrated
with their potential couplings. This element is formulated in terms of global
variables as the stress wrench acts on the foundations and its stiff body movements.

The whole displacement of the foundations equals the sum of an elastic

displacement uel, a plastic displacement upl and a displacement linked to the uplift
uup.

u tot u el  u pl  u up [4.42]

Figure 4.16. Dynamic macro-element concept

The plastic displacement upl is calculated on the basis of a conventional plastic

model defined by the data of a failure criterion, a loading surface and a flow rule for
strain hardening. The failure criterion is provided by the carrying capability of the
foundations [PEC 97]; the strain hardening law combines both isotropic and
kinematic strain hardenings and the flow rule is non-associated.

The displacement caused by the uplift is estimated from an uplift model for
foundations lying on elastic soil [CRE 98].

The macro-element thus constituted is placed in series with the one representing
the far field (Figure 4.16, on the left). Conceptually it can be represented by an
assembly of springs and “brush springs” that couple all the degrees of freedom of
the foundations.
 Soil Behavior: Dynamic Soil-Structure Interactions 161

Though for the moment it is limited to two-dimension geometries, this macro-

element provides us with a simple and effective tool for the resolution of soil-
structure interaction problems involving non-linear phenomena. Confronted with
finite element digital simulations or with experiment results (CAMUS 4 model
tested on the CEA’s vibrating table), this model gives results that are quite
satisfactory from a practical point of view, easy to use and less cumbersome than the
finite element method.

4.5. Bibliography

[CLO 75] CLOUGH R.W., PENZIEN J., Dynamic of Structures, McGraw-Hill, 1975.
[COL 90] COLLECTIF, AFPS, Rapport de mission, Séisme de Loma Prieta, 1990.
[COL 02] COLLECTIF, EUROCODE 8, Design of Structures for Earthquake Resistance,
Part 1: General Rules, Seismic Actions and Rules for Buildings, CEN/TC250/SC8/N317,
May 2002.
[CRE 98] CREMER C., Elaboration of a Soil Structure Macro Element for Foundation Uplift,
mémoire de stage DEA, ENS Cachan, 1998.
[CRE 01a] CREMER C., Modélisation du Comportement non Linéaire des Fondations
Superficielles sous Séisme, Thesis, LMT Cachan, 2001.
[CRE 01b] CREMER C., PECKER., DAVENNE L., “Cyclic Macro Element of Soil Structure
Interaction: Material and Geometrical Non-linearities”, International Journal for
Numerical and Analytical Methods in Geomechanics, 25, pp. 1257–1284, September
2001.
[CRE 02] CREMER C., PECKER A., DAVENNE L., “Modelling of Non Linear Dynamic
Behavior of a Shallow Strip Foundation with Macro Element”, Journal of Earthquake
Engineering, Vol. 6, N°2, pp. 175–211, 2002.
[DOR 90] DORMIEUX L., CANOU J., “Determination of Dynamic Characteristics of a Soil
based on Cyclic Pressumeter Test”, 3rd Int. Symposium on Pressumeters, Oxford, UK,
pp. 159–168, 1990.
[HAL 75] HALL J.R., KISSENPFENNIG J.R., “Special Topics on Soil-Structure
Interaction”, International Seminar on Extreme Load Conditions and Limit Analysis
Procedures for Structural Safeguards and Containment Structures, Paper U2.2, Berlin,
1975.
[HAR 78] HARDIN B.O., “The Nature of Stress-strain Behavior for Soils”, State of the Art,
International Conference on Earthquake Engineering and Soil Dynamics, Pasadena,
USA, 1978.
[IDR 80] IDRISS J.M., KENNEDY R.P., “Analysis for Soil-Structure Interaction Effects for
Nuclear Power Plants”, Report by the Ad Hoc Group on Soil Structure Interaction of the
Committee on Nuclear Structures and Materials of the Structural Division of ASCE, 1980.
162 Dynamic Behavior of Concrete and Seismic Engineering

[KAU 74] KAUSEL E., “Forced Vibrations of Circular Foundations on Layered Media”,
Soils Publication N° 336, Department of Civil Engineering, Massachusetts Institute of
Technology, Cambridge, USA, 1974.
[KAU 75] KAUSEL E., ROESSET J.M., “Dynamic Stiffness of Circular Foundations”.
Journal of Engineering Mechanics Division, ASCE Vol. 101 N° EM6, 1975.
[KAU 78] KAUSEL E., WHITMAN A., MURRAY J., ELSABEE F., “The Spring Method
for Embedded Foundations”. Nuclear Engineering and Design. N° 48, 1978.
[LYS 69] LYSMER J., KUHLEMEYER R.L., “Finite Dynamic Model for Infinite Media”,
Journal of Engineering Mechanics Division, ASCE Vol. 95 N° EM4, 1969.
[LYS 72] LYSMER J., WAAS G., “Shear Waves in Plane Infinite Structures”, Journal of
Engineering Mechanics Division, ASCE Vol. 98 N° EM1, 1972.
[LYS 78] LYSMER J., “Analytical Procedures in Soil Dynamics”, State of the Art, ASCE
Conference on Soil Dynamics and Earthquake Engineering, Pasadena, California, USA,
1978.
[LYS 81] LYSMER J., RAISSI M., TAJIRIAN F., VAHDANI S., OSTADAN F., “Sassi – A
System for Analysis of Soil-structure Interaction”, Geotechnical report N° 81–02,
University of California, Berkeley, USA, 1981.
[PEC 84] PECKER A., Dynamique des sols, Presses de l’Ecole Nationale des Ponts et
Chaussées, 1984.
[PEC 94] PECKER A., “Seismic Design of Shallow Foundations”, State of the Art, Proceed.
10th European Conference on Earthq. Engin, Ed. Balkema, Vienna, Austria. pp. 2905–
2910, 1994.
[PEC 97] Pecker A. “Analytical formulae for the seismic bearing capacity of shallow strip
foundations”, Seismic Behavior of Ground and Geotechnical Structures, Seco and Pinto
(Eds..), Balkema, 1997.
[PEC 00] PECKER A., PENDER M.J., “Earthquake Resistant Design of Foundations: New
Construction”, Geo Eng 2000, Vol. 1, pp. 313–332, Melbourne, Australia, 2000.
[PRE 78] PREVOST J.H. “Plasticity theory for soil stress-strain behavior”, Journal of
Engineering Mechanics Division. Vol. 104, N° EM5, 1978.
[PRE 87] PREVOST J.H., Modelling the behavior of geomaterials, Notes de cours”, Non-
linear soil mechanics and dynamic soil-structure interaction, Lausanne, Switzerland,
1987.
[ROE 73] ROESSET J.M., WHITMAN R.V., DOBRY R., “Model Analysis for Structures
with Foundation Interaction”, Journal of Structural Engineering Division, ASCE, Vol. 99
N° ST3, p. 89–416, 1973.
 Soil Behavior: Dynamic Soil-Structure Interactions 163

[SEE 70] SEED H.B., IDRISS I.M., “Soil Moduli and Damping Factors for Dynamic
Response Analysis”, Report EERC 70-10, Earthquake Engineering Research Center,
1970.
[WAA 72] WAAS G., “Earth Vibration Effects and Abatement for Military Facilities”,
Report 3, US Army Engineer WES, Vicksburg, Miss., USA, 1972.
[WOL 85] WOLF J.P., Dynamic Soil Structure Interaction, Prentice Hall, Englewood Cliffs,
New Jersey, 1985.
This page intentionally left blank
 Chapter 5

Experimental Methods
in Earthquake Engineering

Introduction

Considering the vast size of most civil engineering structures, direct structure
experimentation cannot generally be considered. Apart from a few particular cases
of structure instrumentation on high-risk sites in which earthquakes are anticipated,
model experimentation in the laboratory has been preferred. The most natural
approach uses the shaking table, which involves reproducing the motion of the soil
on which the model has been laid. By imposing acceleration on the table
identical, for example, to the one measured on the ground surface during an
earthquake owing to jacks, we can reproduce the behavior of actual structures. That
behavior results from the properties of stiffness (evaluative in the case of concrete
cracking and steel plasticity), of damping and of the distributed mass of the
structure.

Nevertheless, directly transposing the results of model laboratory tests to actual

structures encounters several difficulties, which include the need to use laws of
similarity which are often ill-adapted to concrete material, the piloting of the table,
and the accuracy of the accelerogram representation at the model level (considering
the filtering carried out by the table itself).

Chapter written by Alain MILLARD, Pierre PEGON and Jean-Claude QUEVAL.

166 Dynamic Behavior of Concrete and Seismic Engineering

With the development of numerical methods, a new method has evolved: the
pseudo-dynamic method, which is complementary to shaking table tests.

In principle, it involves numerically integrating, in the course of time, the

equations of dynamics, written for the structure into a relative motion reference
frame:

M a r  t   C v r  t   Fint ª¬d r  t  º¼ M a e  t 

where M and C represent the viscous damping and mass operators, Fint the inner
forces (apart from viscous effects), ar(t), vr(t) and dr(t) the accelerations, speeds and
displacements related to the structure with regard to its base considered as fixed, and
ae(t) the driving acceleration of the same base.

At a given time, the equation is numerically integrated over a short period of

time, giving an approximation of the displacement d(t). The displacement is then
applied to the model, fixed at its base, through a few wisely located jacks that allow
an estimation of the internal forces (Fint) which will be used for the numerical
integration on the next interval.

The main features of the method are the fact that it is possible to work on models
that are much larger than shaking tables, and that we can proceed slowly.
Nevertheless, rate effects are only taken into account through the viscous damping
operator, and the displacements are only imposed on the structure at a small number
of points, which is not well adapted to real structures with a well-distributed mass.

As we may note, the theoretical concepts on which both experimental methods

are based are rather simple. The actual difficulty lies in the interpretation of the
results and their transposition to real structures, as has been demonstrated in
numerous model tests; it also lies in the implementation aspects, which
fundamentally condition the quality of the results. Thus, in the case of shaking tables
for example, it is essential to control the electro-mechanical system that drives the
table. In the same way, for the pseudo-dynamic method, the splitting into time
periods as well as the integration algorithm of the dynamics equations have a
considerable influence on the final quality of the results.

This is the reason why, in this chapter, the stress is deliberately laid on these
aspects that may at first seem to be more “technological” (numerical as well as
experimental), yet which remain linked to the dynamic behavior of the structures.
The test examples given, obtained from the ELSA and TAMARIS European
facilities, illustrate our remarks explicitly and concretely.
 Experimental Methods in Earthquake Engineering 167

5.1. The pseudo-dynamic method

5.1.1. Introduction

Even if performed in a quasi-static way, the pseudo-dynamic test (PSD) on the

one hand combines on-line calculations and control and on the other hand the
continuous measurement of the properties of a given structure, allowing researchers
to simulate the dynamic response of a structure in a realistic way.

The motion equations of a discrete system modeling the structure during the test
are integrated by a step by step numerical algorithm. The viscous damping and
inertia forces are modeled analytically, which is simple when compared to the task
of modeling the non-linear forces that develop in the structure, and are measured
experimentally. In most cases, it is impossible to model them accurately. This
process allows both the hysteresis damping linked to irreversible strains and the
damage of structural materials (typically the main dissipation source during an
earthquake) to be taken into account.

Figure 5.1. Pseudo-dynamic method

Let us consider the building in Figure 5.1. To simulate its response to an

earthquake, a recording of the acceleration of an earthquake (either natural or
artificial) is supplied as input to the computer in charge of the PSD method. The
horizontal displacement of the building floors (where the mass is supposed to be
168 Dynamic Behavior of Concrete and Seismic Engineering

concentrated) is calculated for a short time step using an appropriate time integration
algorithm. These displacements are then imposed on the structure via servo-
controlled hydraulic jacks leaning on a reaction wall. Load cells measure the
load level needed to impose the required displacement on each floor, and these
values are fed back to the computer to be used at the next time step.

Because the inertia is modeled, the experiments do not have to be carried out in
real time, which allows us to test large size structures with relatively low hydraulic
power. In this regard, PSD tests are complementary to those carried out with shaking
tables which are conducted in real time on extremely reduced structure components
or structure models.

5.1.2. History of the PSD method

The PSD method was created in Japan [TAK 75] over 30 years ago, and the
basic implementations, as well as numerous improvements, were made. Initially, the
integration methods used were the central finite difference method (explicit
Newmark’s method), then an implementation with operator splitting of the
trapezoidal method (implicit Newmark’s method) [NAK 90]. The latter allows the
use of an unconditionally stable scheme for cases including stiff structures and a
certain number of degrees of freedom.

The PSD method was further developed in the USA [SHI 84]. There, a host of
improvements were introduced, including the use of iterative implicit schemes [SHI
91].

The experience accumulated in both Japan and the USA has shown that success
is determined by the quality of the implementation of the PSD method (see [ELB
89], [NAK 90] and [SHI 87]). Many elements in this implementation have a strong
tendency to bring about errors. The measurement or control errors tend to
accumulate [SHI 87], and if they get too high, they can eventually overcome the
response [NAK 87].

All of this takes place as if a differential equation system was integrated with a
computer that would contain only a small number of significant digits.

5.1.3. The ELSA laboratory

The European Laboratory of Structural Assessment (ELSA) [DON 92] is a

laboratory of the European Community Research Common Center (see Figure 5.2)
and was inaugurated in 1992. It consists of a 16 m high and 21 m wide reaction wall
 Experimental Methods in Earthquake Engineering 169

and of 2 test stands that can withstand loads about a few thousand kN, i.e. the loads
necessary to strain and seriously damage real size structures. Although the wall is by
far the longest in Europe, there is one more or less similar in the USA, in San Diego,
but both are surpassed in size by the Japanese Tsukuba 21 m high wall.

Figure 5.2. Reaction wall at the ELSA

The implementation carried out at the ELSA was the first to use fully numerical
control algorithms for the motions of the jacks [MAG 91], [MAG 93]. These allow
very accurate control and implementation of the different algorithms available for
the numerical integration of motion equations.

It is not usually possible to test structures as big as bridges or oilrigs.

Nevertheless, the seismic loading will often only damage some parts of them. For a
lot of bridges, the damage will mostly concentrate at the base of the piers. In these
cases, we can estimate that the remaining structure can be numerically modeled
using the finite element method. It is then easy to combine the PSD test for one part
of the tested structural portion with the time integration of the motion equations of
the model of the remaining structure or numerical structure. For a bridge, the piers
constitute the tested structure, whereas the bridge deck is the numerical structure.
This kind of sub-structuration test forms the essence of the PSD method [DER 85],
but had only been used in practice for small components in the early 1990s. At the
ELSA large-scale tests were carried out for the first time [PIN 95a, PIN 96].
170 Dynamic Behavior of Concrete and Seismic Engineering

5.1.4. Comparison with shaking tables

At this stage, we can already compare the advantages and drawbacks of reaction
wall PSD tests with shaking table dynamic tests.

Advantages: the PSD tests are carried out on real scale or only slightly reduced
models, which are easy and cheap to build without any scale effect of the basic
materials. The tests are carried out slowly, so they can be carefully observed. As
they can be stopped at any time, the models can be loaded to failure and it is
possible to potentially add instrumentation where it is necessary. It is also possible
to use the sub-structuration method with them.

Disadvantages: PSD tests are carried out more slowly than they would be in real
time, which does not suit all material types. We only drive few nodes in the
structure, which is not an accurate situation when the mass is too widely distributed.
Usually we can only control the model in the horizontal plane, which prevents the
vertical components of earthquakes being taken into account.

5.2. The conventional pseudo-dynamic method

5.2.1. Algorithms

The basic hypothesis of the PSD method is that the dynamic behavior of a
structure can be represented using a discrete model with a small number of degrees
of freedom (dof). The motion equations of this idealized structure are expressed as a
system of second order differential equations:

Ma (t )  Cv (t )  r (d (t )) f (t ) [5.1]

where M and C are respectively the mass and damping matrices, a(t), v(t) and d(t)
the acceleration, speed and displacement vectors, r(d(t)) the originally non-viscous
internal forces and f(t) the external forces applied to the structure.

An approximation of the solution to equation [5.1] can be obtained using a step

by step direct integration method. The solution at each time step is obtained
according to the values obtained at the previous time step, in order to have the
displacement of the structure submitted to any external loading. From this point
onwards, the time interval ª¬0.T º¼ on which the integration has to be effected is
divided into equal time steps of size 't .
 Experimental Methods in Earthquake Engineering 171

The integration schemes used at the ELSA belong to the so-called “E-Newmark”
family. For these schemes, when the displacement, speed and acceleration vectors at
time t are known, the values of the same vectors at time t + 't are given by
Newmark’s formulae:

't 2
d n 1 d n 1  't 2 E a n 1 d n 1 d n  'tv n  (1  2 E )a n
2 [5.2]
n 1 n 1 n 1 n 1 n n
v v  'tJ a v v  't (1  J )a

and verify the equilibrium system expressed in equation [5.1], shifted in time thanks
to parameter D:

Ma n 1  (1  D )Cv n 1  D Cv n  (1  D )r n 1  D r n (1  D ) f n 1  D f n [5.3]

If we choose D = 0, E = 0 and J = ½, we obtain the well-known central finite

difference method. The pattern is explicit, as it is possible to calculate dn+1 from
known quantities then impose this displacement, measure the reaction load r n 1 ,
and finally calculate a n 1 and v n 1 .

This scheme is perfectly fitted to the PSD method. Unfortunately, it is

conditionally stable. The solution may become unstable (i.e. increase indefinitely)
when the value of the reduced frequency :0 2S f 0 't is higher than 2. Here f 0 is
the highest natural frequency of the dynamic system considered. If f 0 is high, the
time step becomes very small. Then we have to impose many displacement
increments on the structure, some of which are likely to be smaller than the accuracy
of the control system, which may bring about very large errors.

The frequency f 0 may be high for stiff structures including a small number of
dofs, or for most structures when the number of dofs becomes important. The
excellent accuracy of the displacement checking system used at the ELSA (see
section 5.2.2) allows test stiff structures owing to this explicit scheme without any
real difficulty. The scheme is also used to the exclusion of all others by the
continuous PSD method (see section 5.2). The number of dofs can become very high
in sub-structuring tests (see section 5.2.3), and in these cases, other integration
schemes have to be introduced.

If we choose E (1  D )2 4 and J (1  2D ) 2 , with D  > 1 3, 0@ , we again

find the D method [HIL 77]. This scheme is implicit, because the displacement d n 1
172 Dynamic Behavior of Concrete and Seismic Engineering

depends on the acceleration a n 1 linked to the load r n 1 , which is a function of the

displacement d n 1 . Generally it implies an iterative resolution process. Such an
approach is used by the PSD method [SHI 91], even if it is not easy to implement in
so far as the structure itself is involved in the iterations. The implementation can be
considered as a modified Newton-Raphson method converging in a linear way at
best. Actually a sub-relaxation coefficient is introduced into the successive iterations
to force convergence in a monotonous way and without any parasitic discharge. The
stability of this scheme will be discussed later.

Nevertheless, it is possible to introduce an implicit D-Newmark scheme without

resorting to iterations. To do this we use an operator splitting (OS) method ([COM
97], [NAK 90]). This maintains the stability of the scheme, as it is implicit for the
elastic part of the response, but requires no iteration whatsoever, as it remains
explicit as far as the non-linear part is concerned. The OS method is based on the
following approximation of the reaction force r n 1 :

r n 1 (d n 1 )  r n 1 (d n 1 )  K I (d n 1  d n 1 ) [5.4]

where K I is a stiffness matrix which has to be higher than (or equal to) the tangent
K T stiffness matrix of the system. That condition ensures the unconditional
stability of the scheme. In this case, the time step 't can be any time step, and thus
it is chosen according to the experiment carried out and not to verify any condition
of stability. The D-OS scheme appears as the natural extension to the implicit case of
the explicit scheme of the central finite differences.

As structures tend to soften, K I can be chosen as the elastic stiffness matrix

K E or the initial stiffness matrix of the structure. The experimental installation
obviously allows us to obtain a very good approximation of this matrix.

Eventually, we should note that the Newton-Raphson method of the iterative

scheme also uses a K I matrix with the same properties in terms of stability as those
of the D-OS scheme. If the matrix is chosen properly, then the scheme will be
unconditionally stable as well.

5.2.2. Implementation at ELSA

The checking system implemented at ELSA was quite original compared with
previous analog implementations. As this system is entirely digital, it can easily be
 Experimental Methods in Earthquake Engineering 173

modified, improved and can take new types of sensors into account. Therefore, it is
easy to adjust the system by introducing hybrid solutions to implement the PSD
method, especially as far as the sub-structuration technique is concerned (see section
5.2.3).

The control system operates in the following way: each degree of freedom,
corresponding to the displacement of a jack, is driven by a PC equipped with the
interface needed to receive data from the experiment and send orders to the jack
([MAG 91], [MAG 93]). In the system, a digital displacement sensor measures the
output signal. The real state of the system is then described by means of a variable
y k , which is updated at each step of the sampling. This variable is compared with
the displacement orders d k . In fact, the control algorithm operates from the
ek d k  y k error (and other information) to produce correction orders u k .
Finally, these numerical orders are transformed into a u(t) step-wise continuous
signal, which is sent to the jack’s servo-valve. From the displacement d n at the
beginning of the step to d n 1 at the end of the step, the control unit generates a
function in the shape of a ramp followed by a plateau. This function represents the
displacement path the jack will have to follow to gradually pass from d n to d n 1 .
The ramp is generated in the shape of a discrete sequence d k that is used as orders.
It is presented on the left in Figure 5.5.

In such cases, the control algorithm itself is a numerical implementation of a

PID (Proportional Integral Derivative) analog method.

Note that the displacement rate for making the ramp is generally specified. The
length of a loading step is therefore proportional to the displacement increment, thus
it is unknown beforehand. On the other hand, the force sent back to the central
processing unit is an average of values supplied by the acquisition of the last points
of the displacements’ plateau. The control units are geographically distributed
according to the position of the jacks. Each control unit is connected to the central
processing unit via a high bitrate optical fiber. Each unit works independently, in
real time, for each of the checked dofs in the structure, and it is controlled by a
central processing unit that integrates the motion equations.

The structural displacement is measured by an optical data line counter,

engraved with a 2 Pm resolution [MAG 91]. Units allowing a 0.5 or 1 m
displacement are currently used. These digital sensors are directly connected to the
control units, without resorting to any digital/analog conversion and without any
analog pre-conditioning. The sensors therefore require no calibration and deliver a
signal that gives a correct measure of the speed, by counting on a given time period
174 Dynamic Behavior of Concrete and Seismic Engineering

(the sampling period, for example). The sharp measure of the sensors allows us to
reach an accuracy of the displacement control around 50 Pm.

It should be noted that the displacement sensors are mounted on a supporting

structure independent of the movement of the jacks or those of the reaction wall on
which they rest (see Figure 5.1). In fact, as the stiffness of the wall is not infinite, it
bends with the motion of the tested structure. This means the lengthening of the jack
is not characteristic of the motion of the tested structure alone, especially if we
consider the accuracy of the displacement. Therefore, an independent reference
system has to be introduced, and we should ensure its neutrality (for example, its
mass should be such that no resonance system can be induced by the vibrations
either of the hydraulic system or of the control system).

The hydraulic system which powers the jacks operates under a 210 bar pressure
and delivers a 25-liter per second maximum flow. The jacks are two different types:
maximum force: 0.5 and 1.0 MN; displacement: r 250 mm and r 500 mm.

5.2.3. The sub-structuration method

The PSD method is a hybrid method that combines the digital integration of the
motion equations of a complex structure (condensed on a reduced number of dofs)
with the measure of the reaction forces that results from the imposed motion.
Despite its potential, direct experimentation on very large civil engineering
structures such as bridges would be difficult: apart from the size problem, the
simultaneous control of a large number of dofs would be quite laborious. Yet in
order to deal with such cases, it is possible to expand the application field of the
PSD method, at least whenever part of the structure can be modeled; this is called
the sub-structuration procedure [DER 85].

This procedure takes advantage of the hybrid nature of the PSD method by
combining the modeling of part of the structure (numerical structure) with the real
testing of the remaining structure (tested structure). The procedure naturally applies
to a bridge, as its biggest part, the deck can often be considered as elastic linear, and
can therefore be modeled with finite elements. Only the piers subjected to damage
will be tested in the laboratory. Another advantage, which is particularly welcome in
the case of bridges, lies in the fact that we can impose asynchronous seismic
loadings or loadings that have different amplitudes along the foundations.

The numerical part of the structure has to be modeled suitably, and then the
resulting motion equation discrete system must be integrated in time. The numerical
model can reveal a number of dofs that are far more important than those checked in
the lab. Therefore, it is not straightforward to modify the software developed by the
 Experimental Methods in Earthquake Engineering 175

central processing unit implementing the PSD method and to expand it to the sub-
structuration case.

An alternative solution, far more consistent with the decentralized character of

the implementation at ELSA, has been chosen. It involves implementing the sub-
structuration due to two processes using two different computers. The first process is
carried out by the central processing unit responsible for the dofs tested in the
laboratory, whereas the second process – the one in charge of the modeled part –
operates with a working station implementing the finite element software. Both
processes exchange minimum information via the network. The advantage of such
an approach is obvious. The modifications in laboratory software are minor and the
process dedicated to the numerical part does not specifically need to deal with the
tested part. The communication between both processes has to be safe and has to
potentially allow connections between two different operating systems, while being
standard on the Internet. This is why Berkeley’s communication system (Berkeley
sockets) has been chosen. Practical details about this implementation can be found in
[BUC 94].

We have already said that system described by equation [5.1] could be integrated
in time by a step by step-type numerical scheme. The speed and displacement
vectors at step (n + 1) are expressed as a function of the acceleration vector and
vectors obtained at the previous time step (see equations [5.2]). If these expressions
[5.2] are introduced into the D-off-center shape of the equilibrium described in [5.3],
we obtain the following linear system for acceleration a n 1 :

ˆ n 1
Ma fˆ n 1 [5.5]

where M̂ is a pseudo-mass matrix and fˆ n 1 a pseudo-force vector. If we take into

account approximation [5.4] of the reaction forces of the OS scheme, the expression
of both quantities is given by:

Mˆ M  J (1  D )C  E't 2 (1  D ) K I
fˆ n 1 (1  D ) f n 1  D f n  (1  D )r n 1  [5.6]
D r n  (1  D )v n 1  D v n  D (J'tC  E't 2 K I )a n

Now if we distinguish the data coming from the numerical sub-structure (S),
from the data coming from the tested part (T), the [5.5] system can be rewritten as
follows:
176 Dynamic Behavior of Concrete and Seismic Engineering

ª S Mˆ ij S
Mˆ iT 0 º ª a nj 1 º ª S ˆ n 1
fi º
« » « » « »
« S Mˆ
Gj
S
Mˆ GT  T Mˆ GT T ˆ
M G J » x « aTn 1 » « S fˆGn 1  T fˆGn 1 » [5.7]
« » « » « »
« 0 n 1 T ˆ n 1
T
Mˆ IT M IJ »» «« aJ »»
T ˆ «
¬
fI »
¼
¬« ¼ ¬ ¼

where the indexes i, j refer to the dofs inside the numerical structure, G , T the dofs
common to both parts and I , J the dofs inside the tested structure. Now if we
condense the accelerations a j , which are inside the numerical structure, we obtain
the following control equation, for the whole dofs present in the lab:

ª( S Mˆ GT  S Mˆ G j S Mˆ ji1 S Mˆ iT )  T Mˆ GT T Mˆ G J º ª a n 1 º
« »x« T »
« T ˆ T ˆ n 1
¬ M IT M IJ »¼ «¬ a J »¼
[5.8]
ª ( S fˆGn 1  S Mˆ G j S Mˆ ji1 S fˆin 1 )  T fˆGn 1 º
« »
« T ˆ n 1 »
¬ fI ¼

and the numerical structure is ruled by:

S
Mˆ ij a nj 1 S
fˆin 1  S Mˆ iT aTn 1 [5.9]

Equation [5.8] allows us to stress the differences between a simple PSD test and
a test with sub-structuration. As far as sub-structuration is concerned, an additional
mass matrix related to the connection nodes (term between brackets in the first
member of [5.8]) should be added to the experimental pseudo-mass. In the same
way, at each time step, the pseudo-force vector should include a contribution (term
between brackets in the second member of [5.8]) that will become the numerical
sub-structure. It is interesting to note that during the time-step loop, the data flow
between the digital process and the experimental process is balanced. If the
experimental part needs the contribution of forces on the connection nodes, the
numerical process will require acceleration on the same connection nodes to be able
to develop according to equation [5.9]. Let us also note that such an implementation
is mainly asynchronous: each process has to wait for the other to be able to go
ahead. That approach is natural within the frame of the conventional PSD method,
where the implementation length of a loading step in displacement is not set a
priori. As we will see, this is no longer the case with the continuous PSD method
(see section 5.2).
 Experimental Methods in Earthquake Engineering 177

The measurements supplied by networks of accelerometers installed in

earthquake areas (see [OLI 91], for instance) clearly show that the ground motion is
not uniform near its surface. Due to the reflections and refractions of seismic waves
on soil layers that have different behaviors, measuring points not far away from each
other can be subjected to important relative displacements.

Recent numerical studies (see [PIN 95b] for example) have shown that for
reinforced concrete bridges, whether insulated or not, synchronous loadings lead to
an upper bound of the response. Therefore, the traditional method based on that
loading type would be particularly reliable. However, it should be noted that an
asynchronous stress could substantially modify the distribution of damage inside a
structure. It is thus important to be able to carry out tests for these asynchronous
loadings. In the case of a synchronous-type seismic load, the equations of motion are
written within a relative coordinates system linked to the moving base. The
earthquake is taken into account through the system inertia forces that are introduced
as external forces into the equation of equilibrium system [5.1]:

f  MIb [5.10]

for which b is the intensity of the base acceleration and I the vector which takes into
account the direction of the seismic loading at the level of each dof.

This approach is particularly well suited to PSD tests, as the base of the tested
structure does not move. How can we process for an asynchronous test with the sub-
structuration technique? The different parts of the tested structure, loaded differently
at the base, will in no way be physically connected to one another. In the case of a
bridge, only the piers of which are tested, the condition is verified: each pier will be
tested within a different relative coordinates system, whereas the numerical
structure, the motion of which will be described within an absolute coordinates
system, will ensure the synchronization between the whole relative coordinates
systems. In practice, as far as the tested structure is concerned, an asynchronous test
will be carried out in a way similar to a synchronous test. On the other hand, the
calculations related to the numerical structure are different [PEG 96a, PEG 00].
Furthermore, if a damping matrix is associated with the numerical structure, we will
have to be careful about the introduction of a consistent formulation [PEG 96b], i.e.
a formulation independent from the coordinates systems.

5.2.4. Illustration

An example of the experimental installation carried out for an irregular bridge

test on a scale of 1:2.5 [PIN 95a, PIN 96] can be seen on the left in Figure 5.3.
178 Dynamic Behavior of Concrete and Seismic Engineering

Figure 5.3. Typical structures tested at ELSA

As the deck has been modeled, the piers can be put side to side, which allows a
reduced space. In addition to horizontal jacks, vertical jacks can be seen at the front
of the piers, where they apply the vertical loading due to the weight of the deck. On
the right in Figure 5.3 is a real size 4-storey frame in a configuration including a
filled in wall which was tested in the same period.

5.3. Continuous pseudo-dynamic method

5.3.1. Continuous method principle

The conventional PSD method requires considerable implementation time. The

structure is stopped when it reaches the specified displacement, so the reaction force
can be measured. The break, which can be about several seconds long, harms the
regularity of the loading and causes interfering relaxation phenomena inside the
structure. With the continuous PSD method [MAG 98], [MAG 00], the movement of
the jacks is not deliberately stopped; therefore, it can follow the specified
displacement quite accurately, because it no longer includes any discontinuity. The
load can now be measured for each sampling period of the control system and the
motion equations are immediately integrated, at the same frequency, to continuously
provide a new displacement target. To achieve this goal, it has been necessary to
 Experimental Methods in Earthquake Engineering 179

insert the finite element method into the control program instead of the ramp
generator.

Figure 5.4. Conventional and continuous pseudo-dynamic method

We have seen that both PSD conventional and continuous methods use the same
integration schemes. However, for each bi acceleration value at the base (read in the
recorded acceleration file with a sampling period 'T ), the continuous method will
introduce a series of nss values with bi  k / nss ( 0  k  nss ) accelerations obtained by
interpolation between bi and bi 1 . For each 'T large step, the step by step
integration process will be carried out nss times, at the 'tst operating period of the
control system (typically 1 or 2 ms), as if O sub-steps were introduced into the
conventional method (see Figure 5.4).

The O time dilatation factor between a PSD test and a real time test is given by:
O nss 'tst / 'T . For a test on a several-storey building, we will typically obtain:
nss 500 , 'tst 2ms , 'T 5ms , i.e. O = 200. One second of the real earthquake
will develop in 200 seconds in the test.

The results of a series of tests carried out on a 3-storey, 10.4 m high steel
structure have shown that the PSD method could be implemented 10 to 20 times as
fast as the conventional method, depending on the importance of the masses present
in the experimental device. Let us actually note that besides the static stresses linked
to the displacements, the force cells of the jacks measure inertia forces about O 2
times lower than the real inertia forces, and therefore they can often be neglected.
Nevertheless, when the mass of the tested structure is large and O is smaller than
| 10 , the inertial effects have to be taken into account when we try to resolve the
equations of motion.
180 Dynamic Behavior of Concrete and Seismic Engineering

For a given test, 'tst and 'T are constant. Then the dilatation factor can be
modified interactively just by modifying nss during the experiment. Actually, we
always try to decrease O as much as possible while keeping the control error within a
satisfactory interval (typically < 50 Pm), which indeed depends on the nature and the
complexity of the test. In any case, as with the conventional PSD method, it is
important to keep the errors as low as possible, and the motion regular. The absence
of relaxation imposed on the structure by the PSD method definitely has a big part to
play in it. Let us also note that as the reaction force is now taken into account with a
high frequency (500 or 1,000 Hz), an excellent noise filter of this analog
measurement is thus introduced, which is far better than in the case of the
conventional PSD method where the number of measuring points of the force taken
into account is much lower.

In practice, the values of O can be close to the unit for a simple test (one dof in
the lab with potential sub-structuration) with negligible forces of inertia in the
experimental device. They can be much higher, up to 400 for a complex tested
structure.

5.3.2. Implementation at ELSA

When tests with several jacks have to be carried out, it is necessary to achieve
perfect synchronization of each jack. This requires some refinement in the
architecture of the control system. The difficulty involves coordinating and carrying
out for a specific time period of the control system (1 or 2 ms) all the tasks
associated with a cycle of the PSD method: measuring, calculation of the movement
and control of the displacement. Therefore, new control units, more powerful and
more modular have been introduced, communicating with the computer in charge of
the integration of the equations of motion, owing to a new broadband system.

As shown in Figure 4.5, each control unit consists of 3 modules integrated on the
same card: signal processing and storing, acquisition and control, communication
and exchange of data. These cards can then be plugged on the same single
motherboard and work together owing to the data bus of the host computer, as well
as with its processor that is in charge of the integration algorithm. All these
processors (the slaves checking while the master is integrating) operate in real time
with an appropriate operating system (TNT on 32 bits), synchronized by a clock.
Moreover, as a major task via the network, the master processor also provides the
interface with other computers working with more standard operative systems
(Windows NT, 2000 or XP).

There lies another key point of the new architecture: easy data access. In using
the possibilities of active object distribution via the network (DCOM technology by
 Experimental Methods in Earthquake Engineering 181

Microsoft), it is possible to transfer the data acquisition, processing, storing and

visualizing tasks via the master processor, and potentially the sub-structuration
calculations to other computers in an integrated way. The data related to the
configuration of the test, including the parameters useful for each control device,
could be modified any time as well. Finally, it is possible to open an access to the
test on the net, beyond the local network. This will give external users the
opportunity to directly interact with the test. We can even contemplate sub-
structuration tests where the different parts of a same structure are tested on different
test-platforms.

Figure 5.5. Architecture of the new network tested at ELSA

182 Dynamic Behavior of Concrete and Seismic Engineering

As we have already seen, in a PSD test, it is essential for the displacement

control to be carried out in an accurate way. As a matter of fact, during the step by
step process, errors can accumulate until they distort the response of the structure.
Moreover, if the test procedure is applied to a structure, the response of which
strongly depends on the stress rate, a sharp measure of the speed will also be
important in order to correctly assess the reaction forces linked to the phenomenon.

We should also note that the parameters characteristic of the tested structures are
generally not well known. As a consequence, the control algorithms have to work
within an extremely variable environment including sudden and generally
completely unforeseen disruptions. In such situations, whenever we try to impose a
position tracking control, the PID fixed gain algorithm cannot operate efficiently
enough on the whole possible behavior range of the tested structure. It then becomes
necessary to use alternative control algorithms.

Thus, to ensure sufficient accuracy of the displacement and speed of the jacks, a
recent adaptive algorithm developed at the University of Bristol [STO 90a, STO
90b, STO 92] has been inserted into the control systems. The principles of the MCS
(Minimal Control System) algorithm which ensures a minimum synthesis of the
control are the following:
– no assessment of the parameters of the system to control is required (not even
on-line). The control algorithm requires neither any adaptive observer nor any
control by recovery;
– the parameters (unknown) of the system to control can vary with time, even
with a relatively wide bandwidth;
– the unknown quantities (such as the speed), which are highly contaminated by
noise, can anyway retroact on the system to control;
– the implementation of the algorithm is very simple.

5.3.3. Sub-structuration for the continuous method

With the conventional PSD method, running the test was mainly asynchronous.
A test step in the laboratory could last between 1 and 4 seconds depending on the
size of the imposed displacement. This context was extremely advantageous for sub-
structuration, since the loading process of the numerical sub-structure could also
operate in a completely asynchronous way. At last, as both processes wait for each
other, the calculation time of the digital part could even become longer than the
implementing time of a loading step in the lab. It was then quite easy to carry out
iterative non-linear calculations at the level of the numerical process.
 Experimental Methods in Earthquake Engineering 183

The context completely changes when we consider the continuous PSD method.
The experimental process is synchronous and the physical time between two
integration time steps is extremely short: 1 or 2 ms. As it is difficult to carry out a
non-linear calculation step in so short a period of time, a delayed approach like the
one used previously (see section 5.2.3) is no longer possible, as the experimental
process cannot wait for the numerical process any longer. Therefore, the coupling
method should be synchronous and the data exchange should take place without any
interruption.

The integration algorithm for the whole structure (both numerical and
experimental) should have the following properties: be second order, be stable
within the limits of stability of the explicit scheme implemented for the experimental
part, introduce minimum dissipation and operate with non-stop data exchange.

To do so, we implement a partitioned approach that features the following

characteristics ([MAG 98], [MAG 00]):
– an implicit scheme for the numerical part, coupled with the explicit scheme of
the central finite differences used for the experimental part;
– two different time steps for each part of the structure ('t for the experimental
part and 'T for the numerical part);
– sub-cycling of the explicit part with a number of sub-steps which can be very
large ('T / 't | 1, 000);
– assessment in advance of the state of the numerical structure, and interpolation
of the forces to be imposed on the experimental structure by the numerical part to
limit the digital dissipation of the scheme.

The master computer carries out the integration of the equations of motion of the
numerical structure as its priority task. On each 1 or 2 ms cycle of the control
system, the remaining time is used for the basic acquisition and data transfer tasks.
These basic tasks can be given priorities and the main one can be dedicated to the
calculation of the motion of the numerical structure. This strategy has been used to
obtain preliminary results for a simple structure [MAG 98]. An alternative consists
of having the calculations of the numerical structure carried out by another
computer: this strategy has successfully been used in a recent test campaign on
bridges.

5.4. Final comments

As we have seen, modeling plays a crucial part for the PSD method: choice of
the system of the equations to integrate and model (whether linear or not) of the
184 Dynamic Behavior of Concrete and Seismic Engineering

numerical part in the sub-structuration method. This part does not stop here [PEG
98]. Modeling can be used to carry out the design of the experiment itself. As it
allows seismologists to carry out a dynamic simulation (by keeping all the dofs of
the model) as well as a PSD simulation (by condensing the dofs of the mass matrix),
it is used to optimize the number and positions of the jacks, in order to choose the
best-suited integration scheme and the best integration step. It allows us to ensure
that the jacks will not saturate as far as either forces or displacements are concerned.
At last, experiment is often used to identify models which will later be used either
for parametric studies or to select a reinforcement technique.

Implementing the PSD method has often led to the propagation of excessive
errors. The extreme care taken for its implementation at the ELSA, both in its
conventional and its continuous version, showed that it could be used with
confidence for quite varied situations: concrete, metallic, masonry or mixed
constructions, bridge piers, variedly shaped walls, structures including various
passive insulation systems, models of historical monuments, reinforcement and
repair technical studies.

The PSD method itself remains a research topic, as is shown by the recent
publication of new integration schemes [CHA 02], [WAN 01] or by the involvement
of ELSA in the continuous method where aspects of the regularity of loading and
sub-structuration keep on being improved.

5.5. Shaking table tests

5.5.1. Introduction

Several techniques can be used to test reinforced concrete structures in

earthquakes and especially:
– centrifuge tests;
– pseudo-dynamic tests;
– shaking table tests.

The shaking table tests consist of placing the structure on a table operated in
order to reproduce a given seismic load.

A shaking table generally consists of:

– a table;
– a guiding system along 1, 2 or 3 directions;
 Experimental Methods in Earthquake Engineering 185

– a loading system, with 1 or several degrees of freedom;

– a system generating pressurized hydraulic oil;
– a servo-control system;
– a driving system;
– a reaction basemat;
– acquisition and processing means.

The table plate is generally square. It is made of welded assemblies with

reinforcements in order to have free-free frequencies as high as possible (over 50
Hz).

The table can be loaded in a horizontal mono-axial mode, in a vertical-horizontal

bi-axial mode or in a three-axial mode. It is guided by sliding rail systems or directly
by transverse jacks. In the case of a sliding rail system, the maximum loading
frequencies are lower than 100 Hz because of the operating gaps.

As for multi-axial tables, these are driven directly by jacks in all directions.
There are geometrical compensations between the various axes in order to have
perfectly rectilinear displacements in the three directions. All the jacks are equipped
with hinges. The hinges limit the stiffness of the table as well as its rotations.

The hydraulic power system allows us to supply the jacks with pressurized oil
(210 bars) with outflows varying according to the number of pumps in service.

In order to control the table, electronics bays are needed to drive the jacks. The
jacks are controlled with closed loops. The basic closed-loop control system consists
of:
– an amplifier for the control of the servo-valve and the servo-controller;
– feedback of a LVDT displacement sensor to measure the position of the piston.

A servo-controller compares the feedback signal of the LVDT displacement

transducer with a reference signal (control or program). For any difference between
the feedback signal and the reference, the servo-controller injects an error signal into
the servo-valve in order to bring the difference back to zero. The performance of the
basic control is limited because of the frequency response of the different
components and the maximum outflows of the servo-valves.

Whenever the tested specimens are very big and heavy and the oil outflows are
quite important, it is necessary to use multi-stage servo-valves using two closed-loop
control circuits: an outer loop and an inner loop.
186 Dynamic Behavior of Concrete and Seismic Engineering

In big servo-valves, there is usually a pilot stage (outflow from 1 to 60

liters/min), which is used only to drive an enslaved stage allowing an important
outflow (up to 700 liters/min capacity). A displacement sensor on the enslaved stage
provides a signal proportional to the position of the enslaved stage and then roughly
proportional to the output flow of the servo-valve. There also exists an inner loop
identical to the previous one but which uses the LVDT feedback of the pilot stage.

In spite of the settings of the various loop gains, this kind of control does not
allow us to obtain a flat transfer function within a wide frequency range. This is the
reason why three variable controls are used on some systems.

The system combines the information from each of the following signals:
– displacement of the low frequency;
– speed in the middle frequency range;
– acceleration for high frequencies.

The servo-controller subtracts the feedback for each of the three variables of the
corresponding reference signal, which thus form three error signals. The three errors
are then amplified and added in order to make up a global error signal making it
possible to drive the system. In order to stabilize the control system, pressure-
measuring feedback is injected at the level of the error signal. The technique is
generally used on the hydraulic circuit to improve the stability of the system. The
function stabilizes the high frequencies produced during the seismic tests to provide
more stable acceleration and above all it allows us to reduce the influence of the
resonance frequency of the oil column of the jacks.

For multi-axial tables where several jacks enable the table to be driven in a given
direction, the control systems are split into degrees of freedom, that is, that the
reference signals do not correspond to the signals sent to each jack but to the control
in one degree of freedom. From the reference signal and the geometrical positions of
the jack, the control system generates a control signal per jack.

In spite of the control loops, it is difficult to obtain a flat transfer function for the
whole frequency range. Therefore, a computer is used. From a pre-test (impulse or
random-loading type) it measures the transfer functions for each degree of freedom.
Then, from these transfer functions, the reference signal is corrected and a control
signal is determined, which makes it possible to obtain the desired accelerogram on
the table. After each test, the actual response is measured on the table and compared
with the reference signal. An iterative calculation system allows the control signal to
be corrected again, in order to converge to the desired response.
 Experimental Methods in Earthquake Engineering 187

As the jacks have to move important masses (a table and model), they are fixed
on a reaction basemat, the mass of which is generally at least 20 times bigger than
the mass of the models to be tested. Furthermore, to avoid generating any vibrations
inside the test hall, the reaction basemat is laid on suspension set at a very low
frequency ( 1 Hz). This suspension consists of springs or anti-vibration masses and
dampers.

Finally, digital acquisition methods enable us to digitize and store all the
responses of the various sensors implanted on the models.

5.5.2. Characteristics and performance of shaking tables

Generally, due to the large dimensions and masses of tested structures, the
loading system consists of one or several electro-hydraulic jacks. In fact, they are
the only equipment that allows the application of high-amplitude motions to
important-mass models at low cost and in a small space. In general, the jack strokes
are about 250 mm, compared with the electro-dynamic strokes of vibrating pots,
which are smaller at 50 mm. The other advantage of jacks is that they allow very
low frequency motions and even continuous strokes. One of their negative aspects is
seen if we consider the frequency response of the servo-valves: the maximum
responses are lower than 100 Hz (500 Hz at the most for some jacks with limited
maximum strokes).

The performance of shaking tables is determined by three limiting factors:

– displacement limit corresponding to the maximum stroke of the jack or the
jacks;
– a speed limit corresponding to the maximum outflow supplied to the jacks. The
outflow is restricted by the maximum flow provided by the hydraulic pumps or by
the maximum flow possible through the servo-valves;
– an acceleration limit that is a function of the mass loaded on the table of the
mass of the table itself and of the maximum force of the jacks.

The maximum force of the jacks depends on the section of the piston, the oil
pressure supplied by the pumps and the head losses in the servo-valves. The pressure
commonly used in hydraulic groups is 210 bars, and at high frequency, head loss in
the servo-valves is about 40 bars.

Thus, the maximum acceleration on the table is given by:

Ȗ = F/  M t  M s  [5.11]
188 Dynamic Behavior of Concrete and Seismic Engineering

with:
– Mt = mass of the table;
– Ms = mass of the tested structure;
– F = maximum force of the jacks.

The maximum speeds reached by hydraulic systems are about 1 m/s. For large
installations, the maximum outflows cannot be supplied exclusively by the hydraulic
pumps, which deliver a constant outflow. During tests with sinusoidal loads, the
average necessary outflow Qm is given by Qm = 2 Qk/ʌ. (Qk being the peak outflow).
For time history tests, i.e. when we try to reproduce a seismic motion versus time
(acceleration or displacement) for a limited period, it is possible to increase the
speed by using auxiliary accumulators, which are often placed as reserve powers on
the hydraulic network. Thus, once pressurized, they allow us to supply the system
for short moments with the amount of oil necessary to pass the speed peaks.

Shaking tables have another limit, namely a frequency limit. The frequency
range can be limited by:
– the first frequency specific to the table;
– the dynamic response of the servo-valves;
– the resonance frequency of the oil column.

Figure 5.6. General view of the 4 shaking tables of the TAMARIS facility in Saclay
 Experimental Methods in Earthquake Engineering 189

As far as the eigenfrequencies of the table is concerned, it is a direct function of

the dimensions of the table and its design features. The most common tables are
between 1 m x 1 m to 3 m x 3 m in size. The largest tables are 6 m x 6 m large
(CEA/Saclay in France, Berkley in the USA). A few exceptional tables can reach 15
m x 15 m (Tadotsu in Japan). Plans are underway to allow coupling between several
large-size tables. The tables themselves are generally made of steel, though some are
made of pre-stressed concrete. As they are quite heavy, they are problematic as far
as the mass or the maximum acceleration are concerned. In order to improve
performance, tables can be made of aluminum or composite materials.

With small shaking tables that use small servo-valves, the maximum frequencies
due to the responses of servo-valves are about 100 to 150 Hz. With other
installations requiring high-outflow servo-valves, the maximum frequencies are
about 50 to 100 Hz. To improve the frequency efficiency with some big
installations, it is possible to use several smaller servo-valves mounted in parallel,
themselves used as drivers to operate the racks allowing high outflows.

The resonance frequency of the oil column is due to the compressibility of the
compressed oil inside the jack chambers. Although oil has a very high bulk
modulus, the stiffness of the oil column is low enough for the frequency of the oil
column to be within the relevant frequency range of the table.

The stiffness of the oil column is given by the formula K = 4 S2 B/V where:

 S is the section of the jack piston;

 B is the oil bulk modulus (= 1.4 x 109 N/m²); and
 V is the whole volume of oil enclosed in the jack.

1/2
The oil column frequency is then given by F = 1/2ʌ.  K/  M t  M s   .

For large shaking tables, the frequency ranges from 8 to 20 Hz depending on the
load (for example, the frequency of the oil column of the AZALEE table in the
vertical direction with a 50 ton model is about 20 Hz).

Another limitation is due to the stiffness of the table. Though it is reinforced and
therefore quite stiff, the table is supported either by vertical hinged jacks or by truss
rods or sliding rails. Such equipment has a finite stiffness which influences the
behavior of the model and which should be taken into account for the calculation of
the eigenfrequencies of the model. In many cases, a 6 Hz frequency model will have,
once laid on the table, a bending frequency of 5 to 5.5 Hz.
190 Dynamic Behavior of Concrete and Seismic Engineering

Figure 5.7. Test with a 1/3-scale load-bearing wall model on the AZALEE table

The advantage of using a shaking table lies in the fact that it is the only test
means that allows simulation of structure inertia with a distributed mass. In fact, in
the case of pseudo-dynamic tests, the jacks only apply forces on nodes where all
masses are supposed to be lumped. On the other hand, because the jacks have to set
both the table and the model in motion, it is quite easy to see that the limit is linked
to the maximum load supported by the table. In Europe, several laboratories carry
out shaking table tests: Table 5.1 gives the characteristics of the tables they use for
testing civil engineering structures.

The largest European table is located at the CEA in Saclay. It offers 6 degrees of
freedom, is 6 m x 6 m in dimension, and can support a 100-ton structure. The
biggest table is Japanese and is located in Tadotsu. It is 15 m x 15 m and can support
1,000 tons. A few projects with large tables are being studied in the USA (a multi-
axial, 4 m x 4 m, long stroke (1 m) table at the University of Buffalo) and in Japan
(a multi-axial long stroke table that can support 1,200 tons).
 Experimental Methods in Earthquake Engineering 191

LEE/NTUA MASTER AZALEE

BRISTOL LNEC
Characteristics Athens ISMES CEA/Saclay
England Portugal
Greece Italy France
Dimensions (m) 3x3 4x4 4x4 5.6 x 4.6 6x6
Mass of the table
3 10 11 40 26
(tons)
Mass of the
15 10 30 40 100
specimen (tons)
Maximum height
4 11 12
(m)
Number of degrees
5 6 6 3 6
of freedom
Translations X,Y,Z X,Y,Z X,Y,Z X,Y,Z X,Y,Z
Rotations șx, șy, șz șx, șy, șz șx, șy, șz șx, șy, șz
Horizontal
displacement ± 150 ± 100 ± 100 ± 175 ± 125
(mm)
Horizontal
± 700 ± 1,000 ± 550 ± 200 ± 1,000
speed(mm/s)
Horizontal ± 1.8 and
± 4.5 ±2 ±3
acceleration (g) 1.1
Vertical
displacement ± 150 ± 100 ± 100 ± 175 ±125
(mm)
Vertical speed
± 700 ± 1,000 ± 440 ± 200 ±1,000
(mm/s)
Vertical
±7 ±4 ±2 ± 0.6
acceleration (g)

Table 5.1. Characteristics of the main European tables

Other smaller tables exist in several private European laboratories (including

SOPEMEA in France, ANSALDO and ENEL in Italy and others in Spain, Germany,
etc.).
192 Dynamic Behavior of Concrete and Seismic Engineering

(Acceleration) (g)

(Speed)
(m/s)
(Bi ax ial horizontal
excitation)
(Displacement) (m)

(Frequency) (Hz)

Figure 5.8. Maximum performance of the AZALEE table

under horizontal-horizontal biaxial load

(Acceleration) (g)

(Speed)
(m/s)
(Mono axial horizontal
excitation)
(Displacement) (m)

(Frequency) (Hz)

Figure 5.9. Maximum performance of the AZALEE table

under horizontal mono-axial load
 Experimental Methods in Earthquake Engineering 193

(Acceleration) (g)

(Velocity)
(m/s)
(Mono axial vertical excitation)
(Displacement) (m)

(Frequency) (Hz)

Figure 5.10. Maximum performance of the AZALEE table

under vertical mono-axial load

5.6. Laws of similarity

As discussed above, the main limitation of shaking tables is the mass they can
support. In many civil engineering cases, because of the masses and dimensions, it is
impossible to carry out tests on real-size models.

For reinforced concrete, the minimum 6-cm thickness seems to be a limit value
that allows the preservation of the mechanical characteristics of concrete. For
conventional civil engineering models, this gives a reduced 1/3 scale. With a
reduced scale, it is also necessary to represent the metallic frames, which raises the
question of the accuracy with which it represents the mechanical characteristics of
the steels. As small-diameter steels are not manufactured in the same way as large-
diameter steels, they have different characteristics, particularly for the elongation
limit as well as for the steel-concrete bond.

The tables tests carried out on reduced-scale models involve using laws of
similarity to retrieve the stresses and the strains.

In most cases, the materials are kept identical, as it is impossible to find other
materials with the same multiple characteristics of concrete.
194 Dynamic Behavior of Concrete and Seismic Engineering

Depending on the tests and models, two laws of similarity are used for shaking
table seismic tests.

The first law of similarity is a speed law. It can be used when gravity does not
influence the model’s behavior or when permanent stresses do not prevail. For the
same speed, the same stresses will also be found in the material. For a geometrical
reduction scale n, this involves applying a horizontal acceleration that is n times as
high as for scale-1 and is contracted in time by an n coefficient. As gravity remains
constant, the vertical acceleration (or gravity) is then n times as low as it should be.

For civil engineering tests, generally, the gravity effect is important. Therefore,
to get back to the normal loads and the stresses, it is necessary to apply a vertical
acceleration that will have the right horizontal acceleration ratio. As gravity cannot
be modified, another law of similarity called the acceleration law is used. In this
case, both horizontal and vertical acceleration levels applied are similar to scale-1
acceleration, and the signals are contracted in time by a factor n . In order to find
the right stresses, because the masses have to be in 1/n² (instead of 1/n3, if we keep
the same material), it is necessary to load the floors or the reduced-scale model with
additional masses.

5.7. Instrumentation

During seismic tests on civil engineering models, the following conventional

instrumentation is used:
– accelerometers (mainly controlled or piezo-resistive types to have the lowest
possible pass-band or a pass-band starting from zero);
– wire displacement transducers to allow measurement, even when the model has
transverse displacements, and to allow measurement of important displacements;
– LVDT-type displacement transducers to measure crack openings;
– strain gages stuck on the metallic frames during manufacturing or
stuck directly onto the concrete.

This instrumentation enables us to determine:

– the overall behavior of the model, by measuring the eigenfrequencies, the
eigenmodes, the overall displacements and the accelerations at different points of the
model; and
– the local behavior, by punctually measuring the strains and crack openings;
– the displacement transducers allow us to measure either the relative
displacements (if they are fitted to a supporting frame loaded on the table or directly
 Experimental Methods in Earthquake Engineering 195

fitted on the model) or absolute displacements (when they are fitted on a supporting
frame outside the table).

Accelerometers stuck at different points allow acceleration to be measured at a

given point for determining torsion accelerations, or calculation of the average
acceleration at floor level by summation, differentiation or averaging. Thanks to
minute software that takes the dimensions of the model into account, it also allows
us to obtain normal and shear force values at each floor and each level.

5.8. Loading

The models placed on a shaking table can be submitted to three different types of
loading:
– sinusoidal scanning loading for measuring the eigenfrequencies of the model
with the associated damping;
– random noise loading for measuring the eigenfrequencies of the model together
with the associated damping;
– time history loading where time accelerations are applied in a mono-axial or
multi-axial way.

With sinusoidal loading, scanning is continuous. The scanning rate is adjusted

according to the damping of the structure to avoid being too fast to measure the
resonance peak correctly (rates of about 1 or 2 octaves/min are commonly used for
damping ranging from 2 to 5%). During the tests, a control system allows the table
to be driven with constant acceleration over the whole frequency range. The
acceleration control can take place on the peak value of the basic signal, i.e. on the
signal measured on the table and band-pass filtered with a filter focused on the
loading frequency, or on the value of the return signal (in which case, the return is
disrupted by all the distortions measured on the table).

Damping is calculated either:

– from the width of the resonance peak at – 3 dB using the formula:

ȕ = 'F/2F0 , where 'F is the peak width and F0 is the resonance frequency; or

– from the resonance over-voltage using the formula:

ȕ = 1/2Q , where Q is the resonance over-voltage ratio.

196 Dynamic Behavior of Concrete and Seismic Engineering

For random noise loading, we start from a flat spectrum in g²/Hz. The software
generates a random signal whose frequency band energy corresponds to that of the
spectrum defined for the tests. During the test, driving software checks that the
energy level remains constant within each frequency range throughout. Acquisition
is carried out using a certain number of samples (a minimum of 64 to obtain a 5%
accuracy). The duration of the samples depends on the analysis frequency band: it is
longer when we want to analyze low frequencies. For each sample, a PSD is
calculated and an average is obtained for all samples. Comparing the PSDs of the
sensors with the PSD of the table sensor enables the identification of the resonance
frequencies. This kind of test is generally less disadvantageous for reinforced
concrete structures: unlike a sinus test, it stays at the resonance frequency for a
shorter time and the maximum amplitudes reached are less important.

The time history tests involve applying a given seismic load to the table and in
reproducing it as well as possible. The load can correspond to any of the following:
– a real earthquake recorded on site (real accelerogram);
– acceleration derived from pre-dimensioning calculation;
– acceleration generated from the spectrum of oscillators (synthetic
accelerogram) for a given damping.

5.9. Conclusion

In conclusion, the advantages of shaking table tests are as follows:

– they allow simulation of inertia loads on a structure with a distributed mass;
– they allow tests to be carried out with vertical loads;
– they reveal some dynamic behaviors (vertical normal loads due to the opening
and closing up of cracks);
– it is not necessary to use calculations to take the structure damping into
account since they give the real response of the structure.

However, their drawbacks include the following:

– limitation of the loaded mass, which dictates carrying out tests on reduced
scale models, which may not be truly representative;
– the stiffness of the table influences the eigenfrequencies of the model;
– the soil-structure interaction is difficult to take into account (generally, the
table is considered as hard soil (rock-type));
 Experimental Methods in Earthquake Engineering 197

– driving and reproducing the accelerogram on the table when the model is very
heavy is difficult, deteriorates noticeably during the test for the current driving
systems, and cannot take into account sudden variations in the mechanical
characteristics of the models.

5.10. Bibliography

[BUC 94] BUCHET P., PEGON P., “PSD Testing with Substructuring – Implementation and
Use”, JRC Special Publication, No. I.94.25, JRC, Ispra, Italy, 1994.
[CHA 02] CHANG S.-Y., “An improved on-line dynamic testing method”, Engineering
Structures, Vol. 24, pp. 587–596. 2002.
[COM 97] COMBESCURE D., PEGON P., “D-operator splitting time integration technique
for pseudodynamic testing: error propagation analysis”, Soil Dynamics and Earthquake
Engineering, Vol. 16, pp. 427–443, 1997.
[DER 85] DERMITZAKIS S.N., MAHIN S.A., “Development of Substructuring Techniques
for On-Line Computer Controlled Seismic Performance Testing”, Report N° UCB/EERC-
85/04, Berkeley, USA, 1985.
[DON 92] DONEA J., JONES P.M., MAGONETTE G., VERZELETTI G., “The pseudo-
dynamic test method”, in Recent Advances in Earthquake Engineering and Structural
Dynamics, V.E. Davidovici (Ed.), Ouest Editions, Nantes, France, pp. 769–780, 1992.
[ELB 89] EBERG E., The Pseudodynamic Test Method – Numerical Algorithms and
Implementation, Doctoral Dissertation, Division of Structural Engineering, The
Norwegian Institute of Technology, Trondheim, Norway, 1989.
[HIL 77] HILBER H.M., HUGHES T.J.R., TAYLOR R.L., “Improved numerical dissipation
for time integration algorithms in structural dynamics”, Earthquake Eng. Struct. Dyn.,
Vol. 5, pp. 283–292, 1977.
[MAG 91] MAGONETTE G., “Digital control of pseudo-dynamic tests”, in Experimental
and Numerical Methods in Earthquake Engineering, J. Donea and P.M. Jones (Eds),
Kluwer Academic Publishers, Dordrecht, The Netherlands, pp. 63–99, 1991.
[MAG 93] MAGONETTE G., “The digital control of pseudodynamic tests”, Experimental
Methods in Earthquake Engineering and Structural Dynamics, J. Mazars and E. Booth
(Eds.), AFPS-SECED, Saint-Rémy-lès-Chevreuse, France, pp. 50–62, 1993.
[MAG 98] MAGONETTE G., PEGON P., MOLINA F., BUCHET P., “Development of fast
continuous pseudo-dynamic sub-structuring tests”, Proc. of the Second World Conference
on Structural Control, 28 June and July 1998, Kyoto, Japan, 1998.
[MAG 00] MAGONETTE G., PEGON P., BUCHET P., BONELLI A., “Continuous pseudo-
dynamic testing with non-linear sub-structuring”, Proc. of the Second European
Conference on Structural Control, 2–6 June 2000, Paris, France, 2000.
198 Dynamic Behavior of Concrete and Seismic Engineering

[NAK 87] NAKASHIMA M., KATO, H., “Experimental error growth behavior and error
growth control in on-line computer test control method”, BRI-Report No. 123, Building
Research Institute, Ministry of Construction, Tsukuba, Japan, 1987.
[NAK 90] NAKASHIMA M., KAMINOSONO T., ISHIDA M., ANDO K., “Integration
Technique for substructure pseudodynamic test”, Proc. 4th U.S. National Conference on
Earthquake Engineering, Vol. 2, Palm Springs, California, USA, pp. 515–524, 1990.
[NAK 93] NAKASHIMA M., AKAZAWA T., SAKAGUCHI O., “Integration method
capable of controlling experimental error growth in substructure pseudo dynamic test” (in
Japanese), J. of Struct. Constr. Engng. AIJ, Vol. 454, pp. 61–71, 1993.
[OLI 91] OLIVEIRA C.S., HAO H., PENZIEN J., “Ground motion modelling for multiple-
input structural analysis”, Structural Safety, Vol. 10, pp. 79–93, 1991.
[PEG 96a] PEGON P., “PSD testing with sub-structuring: the case of asynchronous motion”,
JRC-Special Publication, No. I.96.42, Ispra, Italy, 17 pages, 1996.
[PEG 96b] PEGON P., “Derivation of consistent proportional viscous damping matrices”,
JRC-Special Publication, No. I.96.49, Ispra, Italy, 26 pages, 1996.
[PEG 98] PEGON P., “The numerical simulation of buildings subjected to seismic loading:
application of an integrated modelling strategy”, invited lecture, Proc. 11th European
Conf. on Earthquake Engng., Paris, France, 1998.
[PEG 00] PEGON P., PINTO A.V., “Pseudo-dynamic testing with sub-structuring at the
ELSA laboratory”, Earthquake Engineering & Structural Dynamics, Vol. 29, pp. 905–
925, 2000.
[PIN 95a] PINTO A.V., VERZELETTI G., PEGON P., MAGONETTE G., NEGRO P.,
GUEDES J., “Pseudo-dynamic testing of large-scale R/C bridges”, EUR Report n° 16378
EN, JRC, Ispra, Italy, 1995.
[PIN 95b] PINTO P.E., “Research and practice on seismic design of bridges in Europe”,
Proc. of the U.S. National Seismic Conference on Bridges and Highways, San Diego,
California, USA, 1995.
[PIN 96] PINTO A.V. (Ed.), “Pseudo-dynamic and shaking table tests on RC bridges”,
ECOEST/PREC8 Concerted Action Report No. 5 (EC-HCM), LNEC, Lisbon, Portugal,
1996.
[PIN 02] PINTO A.V., PEGON P., MAGONETTE G., MOLINA J., BUCHET P., TSIONIS
G., “Pseudodynamic tests on a large-scale model of an existing RC Bridge using non-
linear substructuring and asynchronous motion”, EUR report No. 20525 EN, Ispra, Italy,
2002.
[SHI 84] SHING B., MAHIN S.A., “Pseudo-dynamic test method for seismic performance
evaluation: theory and implementation”, Report No.UCB/EERC-84/01, Berkeley,
California, USA, 1984.
[SHI 87] SHING B., MAHIN S.A., “Cumulative experimental errors in pseudo-dynamic
tests”, Earthquake Eng. Struct. Dyn., Vol. 15, pp. 409–424, 1987.
 Experimental Methods in Earthquake Engineering 199

[SHI 91] SHING B., VANNAN M.T., CATER E., “Implicit time integration for pseudo-
dynamic tests”, Earthquake Eng. Struct. Dyn., Vol. 20, pp. 551–576, 1991.
[STO 90a] STOTEN D.P., BENCHOUBANE H., “Empirical studies of an MRCA algorithm
with minimal controller synthesis”, Int. J. Control, Vol. 52, No. 4, pp. 823–849, 1990.
[STO 90b] STOTEN D.P., BENCHOUBANE H, “Robustness of a minimal controller
synthesis algorithm”, Int. J. Control, Vol. 51, No 4, p. 967–983, 1990.
[STO 92] STOTEN D.P., “Implementation of MCS on a servo-hydraulic testing machine”,
Proc. I Mech E – Part 1, Journal Sys. Cont. Eng, Vol. 206, No. 13, pp. 189–194, 1992.
[TAK 75] TAKANASHI K., UDAGAWA K., SEKI M., OKADA T., TANAKA H.,
“Nonlinear earthquake response analysis of structures by a computer-actuator on-line
system”, Bull. of Earthquake Resistant Structure Research Centre No. 8, Institute of
Industrial Science, University of Tokyo, Japan, 1975.
[WAN 01] WANG Y.-P., LEE C.-L., YO T.-H., “Modified state-space procedures for
pseudo-dynamic testing”, Earthquake Engng Struct. Dyn., Vol. 30, pp. 59–80, 2001.
This page intentionally left blank
 Chapter 6

Experiments on Large Structures

Introduction

Realistic modeling of the behavior of civil engineering structures is the first and
most difficult step in dynamic or seismic analysis. The parameters such analysis can
define include mass distribution, damping characteristics, the stiffness of the
strength system as regards inertia, the influence of secondary elements and various
interaction phenomena. Forced or ambient vibration testing of large civil
engineering structures are generally used to validate these parameters at the design
stage.

The ability of particular numerical methods or mathematical models to represent

the different interaction phenomena that determine the dynamic behavior of civil
engineering structures can only be judged by comparing the forecasts they give with
well-documented experimental results. Thus, dynamic tests allow the creation of
experimental databases that can be used to develop and validate specific digital
models, taking particular conditions or special interactions into account.

From a practical perspective, the dynamic testing of real structures is very

important when assessing the safety of large works such as dams and bridges,
because they characterize the fundamental properties of a structure, like the mass
and stiffness at certain vibration frequencies, and also provide information regarding
damping. Variations in these properties give a good indication of the changes the

Chapter written by Patrick PAULTRE and Jean PROULX.

202 Dynamic Behavior of Concrete and Seismic Engineering

physical properties of the works will undergo. It is thought that soon, dynamic
testing will be used to not only to detect, locate and quantify damage, but also to
assess the likely productive life of all civil engineering works. This chapter briefly
describes the instrumentation used to perform dynamic testing. Both forced
vibration and ambient vibration tests, widely used for assessing civil engineering
structures, are presented. Some examples of real-life applications on large structures
– a cable-stayed suspension bridge, a pedestrian footbridge, a large gravity dam and
a large arch dam – are used to illustrate different test techniques. Tests used to detect
damage in a civil engineering structure are also explained, and the most recent
research in this area is described.

6.1. Instrumentation

The main consideration in any experimental investigation is the accuracy and

reliability of the instrumentation being used. In a dynamic test, all the instruments
can modify the measured signals. Such modifications must be taken into account
during data processing. To do this, it is necessary to consider all the components of
the instrument, from the measuring device to the analog-to-digital converter of the
acquisition device. Appropriate corrections can then be made at the data processing
stage. A review of the different instruments used for tests on large civil engineering
structures is presented and includes accelerometers, hydrophones, strain and
displacement gauges, and connection wires and data acquisition devices.

Force-balanced or piezoelectric accelerometers are commonly used for structure

testing. At the low frequencies typically associated with civil engineering structures,
the most commonly used devices are of the former type, generally seismic-
mass accelerometers whose frequencies range from 0 to 300 Hz, the 0 to 50 Hz
range being the most common. The advantage of using such accelerometers is that
load amplification is not required.

An accelerometer behaves as a low-pass filter that reduces the amplitude and

modifies the high-frequency phase of the measured signal. Figure 6.1 shows the
frequency response curve of an accelerometer with a cut-off frequency of 50 Hz (3
dB reduction, or about 30% of the amplitude of the measured signal).
 Experiments on Large Structures 203

Figure 6.1. Frequency response of some measuring devices

Several types of dynamic acquisition devices are available. We recommend use

of devices that can carry out several acquisitions simultaneously (track and hold),
thereby eliminating some signal aligning corrections. Acquisition devices must be
fitted with anti-aliasing “hardware” filters before the analog-to-digital conversion,
to prevent frequencies near the acquisition devices getting into the range of interest,
because of aliasing during the conversion into digital format. As an example, Figure
6.2 illustrates a 0.4 second long 95 Hz signal. Suppose that the sampling frequency
of the data acquisition device were 100 Hz, the black spots representing the digitized
values. The system would then record a 0.2 s-period signal instead of the real one,
and the latter would have aliased to 5 Hz. In this case, use of a 50 Hz “hardware”
filter would ensure the original 95 Hz signal almost completely died down before
being digitized.

Controling the cut-off frequencies of these filters is essential, as they may change
from test to test. The frequency response curve of the hardware filter should also be
204 Dynamic Behavior of Concrete and Seismic Engineering

determined in order to make the necessary corrections when processing the signal. A
frequency response curve for a filter with a 20 Hz cut-off frequency is shown in
Figure 6.1.

0.8

0.6

0.4
Amplitude

0.2

-0.2

-0.4

-0.6

-0.8

-1
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40

Time, s

Figure 6.2. Frequency aliasing

Hydrophones are high-accuracy pressure sensors used to measure hydrodynamic

pressures in dam reservoirs. These sensors behave as high-pass filters, as Figure 6.1
shows. The strain and displacement sensors do not require any measuring
corrections, but the signal from such instruments has to be amplified before
acquisition.

The different instruments are connected to the data acquisition device via
several-conductor electric wires, which can reach several hundred meters in length
due to the large dimensions of typical structures. These wires often cause noise with
or without any load amplifier. Noise originates mainly from three sources:
electromagnetic noise, ground loops and tribo-electric effects.

Electrical wires conveying large amounts of alternating current cause

electromagnetic effects. Such cables are surrounded by powerful electromagnetic
fields that vary over time, and these induce static voltages in parallel cables. We
recommend the use of sheathed wires with a conducting surface, which protects the
inner part from external electrostatic and electromagnetic fields. The wires carrying
signals should be placed as far as possible from electrical wires, and the use of
wound cables liable to turn a transformer should be avoided.

A ground loop arises when a common connection inside a system is grounded to

more than one point. This happens, for example, when the shielding of the wire
 Experiments on Large Structures 205

carrying the measured signal is grounded at both extremities. Both groundings are
unlikely to have the same potential, and the potential difference causes currents to
flow within the loop. The signal will therefore be modulated by the potential
difference, which will induce noise in the measuring system. We can avoid forming
such a loop by grounding only one end of the signal carrying wires.

Tribo-electricity effects are due to the static electricity produced by friction

between two different surfaces. Thus, the bending and folding of a cable, a shock or
pressure can create an electrical charge between conducting wires and the outer
shield, or between the conducting wires themselves, which generates a voltage
between both ends of the wire. For this reason, we recommend use of low-noise
wires. These are generally tested along their whole length by manufacturers, to
guarantee low noise levels. When adding connectors at the ends of these wires, we
should take care not to create leaking spots. Using carbon tetrachloride as a solvent
and xylene as a cleaning agent is also helpful.

6.2. Dynamic loads

The development of reliable testing methods and high-frequency analog-to-

digital data recording devices has facilitated the dynamic testing of large civil
engineering structures. Two main dynamic loading methods are used.

The first test involves applying a harmonic load, via an unbalanced mass, as an
exciter (the principle of which is illustrated in Figure 6.3). The harmonic forced
response is measured and the process is repeated over a given frequency range. The
amplitude of the harmonic force applied to the structure is proportional to the
squared excitation frequency. The main drawback of this method is that it is difficult
to generate an appreciable harmonic force at low frequencies, and impossible to
obtain the static response of the tested structure. The testing method proceeds with
rather low increments of the excitation frequency to obtain a sufficient frequency
resolution, which involves protracted experiment times.

Despite this, the advantages of the method are appreciable, which explains why
it is still used so often. To begin with, the whole power available for excitation
concentrates at only one frequency. It also enables structure linearity to be tested by
the successive introduction of different amplitude harmonic forces for the same
excitation frequency: this is easily achieved by varying the mass or its eccentricity.
206 Dynamic Behavior of Concrete and Seismic Engineering

Figure 6.3. Schema of an unbalanced rotating mass exciter connected to a structure and
installation of an exciter on the crown of a concrete dam

The second method involves applying a stationary random excitation and

measuring the stationary response. This modal test technique is used quite widely in
mechanics in aeronautics. It is seldom used on civil engineering structures because
applying the random load and measuring it accurately is far from easy.

An alternative to this method is being used increasingly in civil engineering and

involves using ambient environmental loads as exciters, choosing an accelerometer
as the input reference signal and treating all the others as outputs. These approaches
are called output signal-only methods or under ambient load test methods.

6.3. Data processing

Dynamic tests aim to determine the main dynamic properties of a structure,

including its resonance frequencies, modal damping and vibration modes, by
processing signals from it in both time and frequency fields. How is the data
obtained actually processed?

Forced vibrations: in the case of forced vibration tests, data processing is

relatively easy; it involves establishing the frequency response curves. These are
generally better defined with ambient vibration tests and allow a better assessment
of damping owing to the half-power bandwidth method. Frequency response curves
are obtained from the raw data gathered under forced vibrations (see Figure 6.19
section 6.7.2). The main steps involved in data processing under forced vibrations
are given below:
 Experiments on Large Structures 207

– For each excitation frequency, we must:

- determine the exact excitation frequency and the force generated by the exciter,
- calculate of the amplitude and phase delay for this frequency for each
measuring unit. This can be done using the least squares method. The amplitude is
standardized with regard to the force, for the latter is not constant as it varies with
the frequency square number. From this step, the signal from a measuring unit at a
given frequency is replaced by only two values, namely the amplitude and the phase,
- correct the amplitude and the phase delay according to the frequency
responses of the measuring and recording instruments (Figure 6.1).
– Likewise, for each measuring unit, we must:
- establish the frequency response curves (amplitude and phase),
- identify the resonance frequencies (peaks) and of the corresponding phases,
- determine of the modal damping associated with each frequency identified
owing to the half-power bandwidth method,
- determine the forms of resonance from the amplitudes and phases of each
measuring unit. Continuous duty accelerations are transformed into displacements
and drawn, taking into account the phase delays in relation to a reference point
(generally the excitation point).

Examples of frequency responses obtained under forced vibrations for tests on

buildings and large dams are presented in the following sections.

Ambient vibrations: tests under ambient vibrations involve far more complex
data processing, as calculating frequency responses is conducted in the frequency
domain. The method involves calculating the frequency content owing to Fourier
transform and obtaining power spectral density (PSD) curves. A maximum
recording period should be used to optimize resolution within the frequency domain,
and to identify neighboring modes. Rectangular, tapering extended cosine or
exponential windows can be applied to the recorded signals. Exponential windows
favor the first part of the recorded signal, whereas cosine windows favor the central
portion.

To soften the effects of local vibrations, a time and frequency average is used.
We therefore obtain average spectral density curves for several positions on the
structure. In the case of real test modal techniques, once the excitation source is
measured, it is possible to calculate the cross-spectrum (the ratio of Fourier
transforms of the input and output signals). However, measuring the excitation
208 Dynamic Behavior of Concrete and Seismic Engineering

source is often impossible and it is necessary to use a technique similar to modal

techniques. This involves selecting a reference measuring position that will stay
fixed during all the tests, and using this to calculate the cross-spectra of the obtained
signals with regard to the reference signal. From this, we can estimate a coherence
function from the cross-spectra [BEN 00]. This indicator varies from 0 to 1. A high
value, associated with a peak on the power spectral density curves, reveals a mode
specific to the structure, whereas local vibration modes lead to low values. Some
examples of power spectral densities and coherence factors are shown in Figure
6.14.

Resonance can be directly identified from PSD curves using the coherence
function as an additional indicator. It is possible to draw the eigenmodes for the
amplitudes and phases associated with each resonance, as is the case with the forced
vibration tests. Damping is often difficult to estimate because of the amount of noise
in PSD curves, as the stress level of the structure under ambient vibrations is
generally very low. To overcome this, a new method for identifying dynamic
properties in the time domain has been developed, and it has been shown to provide
an improved assessment of damping and the vibration modes from ambient vibration
signals [AND 99].

6.4. Application to buildings

The finite element method is used to model very complex large civil-engineering
structures all over the world. The ability of any mathematical model to represent the
real behaviors of the structure can only be validated by comparing analytical results
with experiment results. Obviously, such an approach cannot be achieved at the
project definition stage or during designing, but it does make it possible to verify
designing hypotheses. This method was used during the rebuilding of Montreal’s
Olympic Stadium roof, as the dynamic characteristics of the Stadium’s tower had to
be determined to extremely high precision. Besides determining the dynamic
characteristics, repeated vibration tests can also be used as an instrument to measure
damage in a structure. This technique has been successfully applied to a two-storey
building subjected to higher and higher seismic stress levels causing appreciable
damage in the structure.
 Experiments on Large Structures 209

6.4.1. The slanting tower at the Montreal Olympic Stadium

The main objective of the forced vibration tests carried out on the tower of the
Montreal Olympic Stadium was to obtain vibration frequencies and modes to update
a three-dimensional finite element model of the structure.

Figure 6.4. Instrumentation of Montreal’s Olympic Stadium slanting tower

Researchers were especially interested in assessing the modal damping, since the
lower part of the tower had been built in concrete (up to 133 m high), whereas the
higher part was later completed in steel. Figure 6.4 shows a front view of the tower,
and plane views of the instrumented levels.

Frequency sweeping from 0.5 to 9 Hz was carried out with an unbalanced

gyrating mass exciter located (i) at the top of the tower (level 577 or 176 m high)
and (ii) at the top of the concrete section (level 372 or 133 m high). With a 4 Hz
frequency, the eccentricity of the masses was varied from 100 to 40% to reduce the
excitation level. Accelerometers were placed on those levels and three others to
measure the longitudinal, transverse and vertical responses.

Figure 6.5 shows the frequency responses obtained with standard amplitudes and
phases at station 3 (longitudinal and vertical). The arrangement of the instruments
was crucial for mode assessment and to allow distinction between torsion modes and
bending modes.
210 Dynamic Behavior of Concrete and Seismic Engineering

Figure 6.5. Frequency response of the higher floor in the slanting tower

Four torsion modes including the fundamental mode, and four bending modes
were identified over the frequency range used. Three other coupled modes could
also be observed. Ambient responses were also measured under the effect of wind,
and the results were correlated to the dynamic properties obtained under harmonic
vibrations. Vibrations due to the motion of the outer panoramic lift were recorded,
both to estimate the amplitude of the resulting accelerations and to identify the
specific modes excited when it is operating.

In total, 11 experimental modes were identified, among them the first torsion
mode at 1.02 Hz and the second bending mode at 1.35 Hz. This enabled a three-
dimensional finite element model, including the interaction with the flexible roof, to
be constructed. It predicted a fundamental bending mode and a second torsion mode.
Modal damping was identified for all modes, and this enabled calibration of
Rayleigh’s damping on the first mode (1.29%) and the eleventh mode (2.55%) to be
calculated.

6.4.2. Reinforced concrete building

A real-size reinforced concrete building was constructed at the Sherbrooke

University structure laboratory, as part of a project to assess the performance of very
high efficiency concrete structures under seismic loads. The building was subjected
to increasing seismic load levels using the pseudo-dynamic test method presented in
this book. Here we only discuss the second objective of that study, which was to
measure the damage caused by dynamic testing. The signature of the undamaged
building was obtained and used as a reference to determine the damage.
 Experiments on Large Structures 211

Figure 6.6. Real size very high efficiency concrete building after pseudo-dynamic tests

The building was 5 m long in the E-W (East-West) direction, and 4 m long in the
N-S (North-South) direction. Each floor was 3 m high. The building was a stiff
nodal element structure supporting bidirectional slabs made from 70 MPa high
strength concrete (see Figure 6.6).

The building was equipped with strain and displacement gages, and
accelerometers set on both floors to measure acceleration in the three directions. The
harmonic forced vibration tests were carried out with an unbalanced gyrating mass
exciter placed on the roof, slightly off-axis, in order to load the torsion modes.
Figure 6.7 illustrates the instruments on both floors of the building, the location of
the exciter and those of the loads added to represent the weight of the service loads.
Vertical accelerations allowed the identification of the vibration modes associated
with the floors.
212 Dynamic Behavior of Concrete and Seismic Engineering

Figure 6.7. Instrumentation of the building

Figure 6.8 presents the frequency response curves measured on the first floor and
the roof before damage, and after applying the standardized El Centro earthquake
signal with a 0.129 g maximum acceleration. The effects of the damage caused by
the earthquake are obvious in the figure and mostly involve an amplitude reduction,
a notable increase of the modal damping (by a factor of 2) and a downward shift of
the vibration frequencies. It is possible to quantify the damage and to locate it from
this data, using appropriate system identification techniques [DOE 96]. Ultimately,
we obtain damage “quantities” according to the stiffness reduction of the structural
components (frames either parallel or perpendicular to loading directions, for
example).

Figure 6.8. Frequency responses of the undamaged and damaged building

 Experiments on Large Structures 213

6.5. Bridge application

Bridges are probably the best characterized civil engineering structures as far as
behavior is concerned, and specific dynamic test techniques have been developed for
this type of structure. The tests are used for long-term damage and behavior
monitoring as well as damage assessment following an earthquake. The following
section examines two dynamic test applications on bridges (a pedestrian footbridge
and a mixed cable-stayed/suspension bridge).

6.5.1. Pedestrian footbridge

As a rule, pedestrian footbridges are lightweight, slender structures that

sometimes have vibration problems. In order to select the right vibration reduction
techniques (passive or active damping devices), it is essential to know the dynamic
properties of the structure and to estimate the vibration level with service loads as
well.

The 60 m long pedestrian footbridge in Sherbrooke (Figure 6.9) is one of the

earliest structure applications built using reactive powder concrete. It consists of a
post-tensioned truss, the diagonal RPC beams of which are confined in thin stainless
steel tubes, and a (30 mm) ultra thin slab. The innovative features of the structure
have given birth to a novel behavior monitoring and instrumentation program.

Figure 6.9. RPC pedestrian footbridge (Sherbrooke, Quebec, Canada)

Figure 6.10 illustrates the instrumentation of the footbridge, and the locations of
the vertical acceleration gages under a load due to pedestrians or pedestrian groups
crossing.
214 Dynamic Behavior of Concrete and Seismic Engineering

Figure 6.10. Instrumentation of the footbridge

Two different kinds of loading were used for the tests. In the first series of tests,
several pedestrians walked randomly across the footbridge, walking in an
unsynchronized way, in order that no vibration mode was favored. This was done in
order to extract the dynamic properties (frequencies and damping and vibration
modes) of the bridge under ambient vibrations. 120 s samples were recorded at 100
Hz for this part of the test.

Figure 6.11. Signals measured on the footbridge with different loading cases

The purpose of the second series of tests was to estimate the vibration level
under service loads. Pedestrian groups performed several “controlled” crossings, at
different speeds and with different numbers of pedestrians on the footbridge. Figure
6.11 illustrates the signals measured while pedestrians were crossing, and Table 6.1
gives the vertical acceleration maximum values measured on the bridge deck.
 Experiments on Large Structures 215

Values respectively ranging from 0.022 g (0.22 m/s2) to 0.14 g (1.40 m/s2) could be
observed with walking and running pedestrian groups. Design codes often specify
the comfort criteria from critical accelerations under service loads, and frequently
depend on the resonance frequencies.

Dynamic tests allow these criteria and their relationships with the dynamic
properties of footbridges to be checked. For this footbridge, the critical acceleration
specified by British standards (BS 5400) and by the Eurocode (EC2.2) is 0.76 m/s2,
whereas the critical acceleration specified by the Ontario code (ONT 83) is 0.48
m/s2. These values can be exceeded in certain loading cases specified in Table 6.1.

Maximum accelerations
Loading cases
(g) (m/s2)
1 pedestrian, walking 0.0135 0.132
2 pedestrians, walking 0.0133 0.130
3 pedestrians, walking 0.0159 0.156
4 pedestrians, walking 0.0219 0.215
Pedestrian group, walking 0.0222 0.218
1 pedestrian, running 0.0384 0.377
2 pedestrians, running 0.0604 0.593
3 pedestrians, running 0.1053 1.033
4 pedestrians, running 0.1352 1.326
Pedestrian group, running 0.1425 1.398

Table 6.1. Maximum accelerations on the footbridge

6.5.2. A mixed cable-stayed/suspension bridge

As far as road bridges are concerned, dynamic tests can take place with
controlled traffic, using test vehicles (lorries and trailers) with a specified axle load,
or under ambient vibrations generated by the wind or steady traffic. Bridge
responses are measured by accelerometers, strain gauges and displacement
transducers. Generally, the strains and displacements are used to obtain the dynamic
amplification factor, whilst accelerations are used to obtain the dynamic properties,
even if the accelerations can also be integrated to provide displacement responses.
216 Dynamic Behavior of Concrete and Seismic Engineering

Built in 1947, the suspended Beauharnois bridge is 177m long, and crosses one
of the supply canals of an hydroelectric power station south-east of Montreal. The
bridge underwent important structure modifications after a steel orthotropic deck
slab and a mixed cable-stayed suspension system were installed. Dynamic tests were
conducted on the new structure to calibrate a finite element model capable of taking
the non-linear behavior of the cables and suspension into account.

The eastern part of the symmetric bridge deck was instrumented (Figure 6.12).
Low-frequency accelerometers were placed at each suspender cable to measure the
vertical, transverse and longitudinal responses of the bridge deck and of the east
tower, under the influence of ambient and controlled traffic loads. Displacement
transducers were placed at position 14, close to the east tower, and strain gauges
placed on the main trusses.

West Tower East Tower

[mm]

54483 176784 54483

(a)

19 Accelerometers
15 16
167 kN 134 kN 183 kN 41 kN Accelerometers' reference

17 18
20
223 kN 57 kN 1 2 3 4 5 6 7 0 8 9 10 11 12 13 14
Test vehicles with axle loads West tower Measuring locations (east span)

(b) (c)

Figure 6.12. Instrumentation of the bridge and test vehicles

Two lorries and a trailer were used to cross the bridge at 75 different speeds,
according to various positions on the bridge. The data recorded during the tests as
well as the data obtained with usual traffic were used to calculate the dynamic
amplification factor, a parameter which had rarely been calculated for stay-cable and
suspension bridges.

Figure 6.13 shows the displacement versus the time, obtained thanks to LVDT
n°2 (Position 14) with one lorry crossing at a speed of 76 km/h (48 mph). The
maximum dynamic response was then derived from the measured signal, as shown
in the figure. The maximum static response was obtained by filtering the dynamic
response with a low-pass filter. The cut-off frequency and the bandwidth of the filter
 Experiments on Large Structures 217

were calibrated from responses measured during low-speed crossings (quasi-static

tests). These tests allow the measurement of the static deformation of the bridge and
an equivalent to the line of influence to be measured at a selected point.

The dynamic amplification factor can be determined, as indicated in Figure 6.13,

and an average calculated for each series of vehicles crossing, by working out the
relationship between both response parameters. The same procedure is applied for
strain measures. As is often the case with bridges, the positions of the two
displacement transducers on the bridge were limited by the depth of the river and its
currents. To overcome this problem, accelerations were integrated to give a rough
estimate of the displacements to assess values of the dynamic amplification factor in
different locations.
Displacement, mm
Volts

Time, s

Figure 6.13. Assessment of the dynamic amplification factor (DAF)

Strain Displacement* Displacement**

Vehicle
Min. Max. Min. Max Min. Max.
10-tire lorry 1.19 1.20 1.10 1.11 1.06 1.07
Lorry and trailer 1.08 1.09 1.03 1.04 1.01 1.02
2 lorries in line 1.09 1.10 1.05 1.06 1.03 1.03
2 lorries side by side 1.10 1.10 1.06 1.07 1.04 1.05
All tests 1.12 1.12 1.07 1.07 1.03 1.04
*Measured displacements (LVDT) **Average of the integrated accelerations

Table 6.2. Average values of the DAF

218 Dynamic Behavior of Concrete and Seismic Engineering

Table 6.2 presents the average dynamic amplitude factor values corresponding to
strains measured on the main trusses, for the displacements measured (LVDT at
position 14), and the integration of the average accelerations recorded for all
positions. The highest values were obtained for a single lorry, whereas strain
dynamic amplification factor values were higher than the values obtained for
displacements. This can be explained by the fact that strains are affected more by the
contribution of higher modes than displacements, and also because the lowest
vibration frequencies are below the 2 to 5 Hz range generally associated with
commercial vehicles.

Figure 6.14. Separation between torsion and bending modes

Ambient vibration tests (steady traffic) were also carried out. As many as 70
ambient vibration response measurements were recorded for 120 seconds with a 100
Hz sampling frequency. The frequency content and coherence functions for vertical
acceleration are shown in Figure 6.14. The responses of both sides of the bridge
deck were added and subtracted to separate bending and torsion modes, which are
often quite close to each other. Graphs on the right illustrate the frequency contents
of the added data leading to the identification of bending modes, whereas those on
the left illustrate the results obtained with subtracted answers where torsion modes
are isolated. A total of 23 bridge deck modes (10 in bending, 8 in torsion, 4 lateral, 1
longitudinal) and 8 tower modes could be identified.
 Experiments on Large Structures 219

1st bending mode 1st torsion mode

fmef =0.631 Hz fmef =1.271 Hz

fexp =0.647 Hz fexp =1.221 Hz

2nd bending mode 2nd torsion mode

fmef =0.816 Hz fmef =2.097 Hz

fexp =0.867 Hz fexp =2.020 Hz

3rd bending mode 3rd torsion mode

fmef =1.291 Hz fmef =2.912 Hz

fexp =1.477 Hz fexp =3.015 Hz

4th bending mode 4th torsion mode

fmef =2.041 Hz fmef =4.159 Hz

fexp =2.386 Hz fexp =4.419 Hz

5th bending mode 5th torsion mode

fmef =3.112 Hz fmef =5.328 Hz

fexp =3.503 Hz fexp =5.505 Hz

6th bending mode 6th torsion mode

fmef =4.266 Hz fmef =6.675 Hz

fexp =4.675 Hz fexp =6.769 Hz

Figure 6.15. Comparisons between the bridge vibration

modes measured and calculated

To calibrate the numerical model, initial tensions within the main suspenders and
stays were calculated using a program aimed at the non-linear analysis of suspension
bridges, which was used to rehabilitate the structure. A three-dimensional model
was used to carry out an analysis of eigenvalues; finally, the results from the
numerical model and its accuracy were checked against the experimental results
(modes and frequencies).
220 Dynamic Behavior of Concrete and Seismic Engineering

Thanks to this method, the experimental modes of the bridge deck and the tower
were predicted with an average accuracy of 5%. The first four torsion and bending
frequencies, the experimental and numerical modes, together with the corresponding
frequencies are illustrated in Figure 6.15. The determining factors for the calculation
of eigenvalues and vectors in this case were found to be the cable tensions and the
density of the highway

6.6. Application to large dams

The owners of large dams are increasingly concerned about the seismic safety of
their works, and the best methods to assess them formed one of the main themes of
the ICOLD Congresses (Large Dam International Commission).

In situ dynamic tests are acknowledged as the most reliable ways to assess the
dynamic properties of dams [HAL 88, COL 90]. With recent advances in finite
element modeling techniques that take complex interaction phenomena between
different sub-structures into account (dam, foundation, reservoir, ice covering),
compiling a reliable database on the dynamic behavior of dams is now greatly
simplified. Ambient and forced vibration tests are also quite useful, but are carried
out at significantly lower loading levels.

Recording accelerometer networks are already in place on some large dams

([DAR 95], [DAR 01]) and allow the behavior of these works to be studied in real
time, especially during reservoir filling and emptying cycles. It is widely known that
hydrodynamic pressures and wave absorption at the bottom of the reservoirs play an
important part in the dynamic response of concrete dams, using hydrodynamic
pressure measurements made during forced vibration tests on gravity and arch dams
([DUR 88], [PRO 97]).

The following sections describe three forced vibration test campaigns undertaken
to assess the seismic behavior of large concrete dams, and highlight the use and
efficiency of the non-destructive dynamic techniques used.

6.6.1. Assessment of a response spectrum on the crown

The Beauharnois gravity dam is located near Montreal (Quebec, Canada) and
runs across the Saint-Laurent river. This 20m high concrete structure includes 36
alternators, and has a crown that is almost 1 km long. It was constructed in three
phases in 1930, 1950 and 1960. The gravity dam is a channel-head with the power
 Experiments on Large Structures 221

station juxtaposed downstream. The latter consists of concrete foundation, including

the turbines and alternators, and a metallic superstructure.

A dynamic test campaign was carried out in collaboration with the owners,
Hydro-Quebec, in order to determine the response spectrum on the dam’s crown.
Important works had been planned on the site and a building was to be erected on
the crown to shelter the numerous transformers and other electrical equipment. It
was therefore expedient to estimate the response spectrum on the crown, and this
was assessed using data regarding the dynamic properties of the works.

Figure 6.16. Sectional drawing of the dam with test instrumentation

As the Beauharnois dam produces a large amount of the total energy supply to
the City of Montreal, stopping a group during the test period was out of the question.
However, as repair work was underway on one of the alternators, the necessary
measurements were made on an inoperative system, without contamination from
noise due to the operation of the alternator and the water flowing through turbines.
Nonetheless, some noise from the neighboring groups was measured.

An unbalanced mass exciter used to generate a horizontal harmonic force on the

crown was placed inside the water intake valve shelters. Figure 6.16 is a sectional
drawing showing the location of the accelerometers. As the test results were used to
calibrate a two-dimensional finite element model, it was necessary to place
instruments on the water intakes (upstream), the concrete power station and its
metallic superstructure.
222 Dynamic Behavior of Concrete and Seismic Engineering

Speed (Sv) cm/s

Period (T), s

Figure 6.17. Response spectrum on the crown of Beauharnois dam

The frequency responses from each station were used to extract the resonance
and damping of the system (the process will be illustrated for another dam in the
next section). As the numerical model developed for the structure was two-
dimensional, the identification was carried out from the first mode of the dam
observed at a frequency of 8 Hz. The model was then used to estimate the influence
of the different sub-structures in the dynamic response, by calculating the frequency
response for a unit harmonic load on the crown, and comparing it to the
accelerations measured during the tests.

Once the model had been calibrated with a reasonable confidence coefficient, it
was used to calculate the response spectrum for a structure located on the crown.
Several earthquakes, based on the dimensioning spectrum for the Montreal region,
were generated and applied to the model. The crown’s responses were then used to
assess several spectra for different earthquakes: the average values for ten artificial
earthquakes are shown in Figure 6.17. The resultant response spectrum was then
used to design buildings located on the crown of the dam.

6.6.2. Study of foundation-ice-reservoir-dam interactions

Forced vibration dynamic tests provide well-defined frequency response curves;

thus, the identification of dynamic properties is easier than it is with ambient
vibration tests. Indeed, such tests require significant logistics, namely using an
 Experiments on Large Structures 223

exciter and a generator set capable of supplying the necessary power (about 8 kW).
Nevertheless, the accuracy of the results enables the validation of the numerical
models, taking interactions between the different sub-structures into account, and
thus facilitates validation of the other modeling techniques used to estimate the
seismic response of hydraulic works.

From this perspective an ambitious project was undertaken, in collaboration with

Hydro-Quebec, aimed at studying the dynamic interaction phenomena taking place
between the dam, the foundation and the reservoir ([PRO 97], [PAU 02]). These
sub-structures interact during an earthquake and play quite an important role in the
dissipation of energy. Moreover, in Nordic countries, the ice cover represents
another sub-structure liable to influence the dynamic response of the system.

The 84 m high Outardes 3 density-dam was chosen for the test campaign. It is
located in the northern part of Quebec and its reservoir is covered with ice almost 6
months a year. Very little experimental data on the dynamic behavior of structures
with such ice coverings exists, so the test campaign was also geared towards
assessing the effect of ice on the dam’s response. Forced vibration measurements
were carried out in July and February, with daily temperatures ranging from –10 to
–15°C in the latter case. The ice cover was measured and found to reach a thickness
of up to 1.5 m along the upper bank.

The unbalanced mass exciter used for the tests can generate a force up to 90 KN
within a 0 to 20 Hz range. Figure 6.18 illustrates the instruments used for the forced
vibration tests. As the aim is to identify as many resonance values as possible within
a given frequency range, it is sometimes necessary to change the exciter’s location.
When placed at the center of the structure, the latter will tend to favor symmetric
modes. Although modifying the orientation of the stress in the horizontal plane is
possible, it is sometimes necessary to move the excitement point to excite certain
anti-symmetric modes as well.

The exciter positions used for the Outardes-3 tests are indicated by double
arrows in Figure 6.18, and the accelerometers locations are identified by cubes. The
latter were placed on the crown, in various locations in the inspection galleries and
on the lower bank in order to model a three-dimensional representation of the
resonance shapes.

To assess the reservoir-dam interaction, the hydrodynamic pressures generated

by the motion of the dam and wave propagation inside the reservoir were measured.
These pressures were collected by hydrophones mounted at regular intervals on a
cable. The upper part of Figure 6.18 (summer tests) shows the cable network from
224 Dynamic Behavior of Concrete and Seismic Engineering

which the hydrophones were suspended. The main cable was used to connect both
banks of the reservoir, and secondary cables were placed perpendicular to the crown,
linking the dam with the main cable. This floating assembly allowed a series of
hydrophones to be suspended in several places inside the reservoir, and the
hydrodynamic pressure response at 0, 30, 60 and 90 m from the upper bank was
obtained. The distance between each transducer was 15 m, and ranged from 15 m to
75 m underwater. For the winter tests, with the ice cover as shown at the bottom of
Figure 6.18, the cable network was unnecessary and the ice was merely drilled in
several places to obtain measurements similar to those conducted in summer.

Figure 6.18. Instrumentation of the Outardes 3 dam both in summer and in winter
 Experiments on Large Structures 225

Figure 6.19 illustrates the typical forced vibration signals obtained. The
hydrodynamic pressure is shown in the first two graphs ((a) and (b)) for two
different depths along the upper bank, whereas the acceleration response obtained on
the lower bank is illustrated in (c). A 5 V pulse is emitted for each mass rotation
(graph (d)), which allows the exact rotation frequency of the exciter to be assessed (a
little over 8 Hz in this case).

Figure 6.19. Examples of signals measured under forced vibrations

The signals were used to calculate the frequency responses for each measuring
position and to extract the dynamic properties of the dam. Both graphs at the top of
Figure 6.20 show the responses obtained on the crown for two exciter positions. By
comparing the amplitudes and phases for the summer (dotted lines) and winter tests
(solid lines), it is clear that amplitudes are far smaller when there is ice, and that
resonance frequencies are generally decreased. Therefore, the ice cover imparts an
additional stiffness to the reservoir-dam set and also increases modal damping
indicated by smoothing of the peaks.

Figure 6.20 also illustrates the displacement effect of the excitation point. For
results obtained in summer, the first resonance (which corresponds to a symmetric
mode) is more loaded in a central position (block H) than in the quarter point
position (block M). Conversely, the second resonance (associated with an anti-
symmetric mode) is almost invisible at the central stimulation position and highly
stressed at the other.
226 Dynamic Behavior of Concrete and Seismic Engineering

Figure 6.20. Frequency responses on the crown

Comparing both series of tests reveals a new mode near the first resonance. This
is mostly visible with responses obtained inside inspection galleries, as illustrated by
the diagrams in the middle of Figure 6.20. Amplitudes are drawn at the same scale
to highlight the response reduction in passing from one gallery to another one. The
double peak is clearly identified in the case of the first gallery, 20 m below the
crown.

Ice cover also causes modifications similar to hydrodynamic pressures. The

diagrams at the bottom of Figure 6.20 show responses along the upper bank of the
reservoir (actually, 3 m away from the bank in the winter tests) at respective depths
of 15 m and 65 m. The damping increase and resonance shift are obvious, as in the
case of acceleration responses: the new mode can be observed as well. It should be
remembered that the water level difference between both series of tests can be
neglected and cannot account for the resonance frequency variation.
 Experiments on Large Structures 227

Resonance modes are obtained from amplitudes and phase delays for the
different measuring positions on the crown. They make it possible to detect the
appearance of a new mode due to interaction between the dam and the ice covering.
This has a symmetric shape, as is the case with the first mode observed in summer.

Figure 6.21. Finite element model for the dam, the reservoir and the foundation

These results have been used to calibrate a finite element model developed for
the dam-reservoir-foundation set. This model is illustrated in Figure 6.21, which
shows the mesh for the three structures. In this study, EACD-3D software [FOK 86],
designed for three-dimensional seismic analysis of concrete dams, was used to carry
out a parametric analysis. Effects due to dam and foundation stiffness, water
compressibility (energy dissipation through upstream wave propagation) and dam-
reservoir interactions were studied. The frequency responses on the crown, inside
galleries and in the reservoir were compared and proved each sub-structure made its
own contribution to the dynamic response of the construction works [PRO 97].

The results of the study were also used to develop a two-dimensional model of
the ice covering [BOU 02], which was added to EAGD software used for the
seismic analysis of gravity dams [FEN 84]. A parametric study confirmed the
appearance of a new mode, together with the stiffness and damping modifications
previously observed during the tests.
228 Dynamic Behavior of Concrete and Seismic Engineering

6.6.3. Study of the effect of the water level inside the reservoirs

Dynamic test techniques can be used to study the effect water level inside
reservoirs has on the dynamic response of concrete dams. A multiple test campaign
was completed on a large dam in Switzerland, as part of a collaboration between the
Swiss Federal Water and Geology Office, Electricité de France, Hydro-Quebec and
Emosson S.A.

Forced vibration measurements were carried out on the Emosson arch-dam,

located on the French-Swiss border near Martigny. This 180 m high arch
(represented by a finite element model in Figure 6.22) was submitted to forced
vibrations by an unbalanced mass exciter, and the hydrodynamic pressure and
acceleration responses for different water levels in the reservoir were derived [PRO
01].

The yearly variations of water levels inside the reservoir are shown in Figure
6.23. The test periods – there were four of them  were chosen with “empty” and
“full” lake conditions, plus two intermediate levels. Access to the dam, which is
located in an Alpine region, is limited by a snow covering which generally still
prevails in spring during the empty reservoir conditions. The reservoir fills up in
summer, and usually reaches its maximum operating level (roughly 225 x 103 m3) in
September.

Dynamic tests were conducted in a way similar to those described in the previous
section. The frequency responses enabled several resonance values within
frequencies ranging from 0 to 10 Hz to be identified, and derivation of the
corresponding modal damping.

180 m

Figure 6.22. Model of the Emosson arch-dam

 Experiments on Large Structures 229

Figure 6.23. Selection of the test dates during the filling-up cycle (source: Emosson S.A.)

Figure 6.24 shows the variation of frequencies with water levels for the first
seven modes. As the water level rises, resonance frequencies should decrease with
the added mass from the reservoir. However, in the lower part of the curves in
Figure 6.24, it is evident that the frequencies rise with the water level, and the trend
gets more marked for higher modes. The same phenomenon was also observed
during the ambient vibration tests at the Mauvoisin dam ([DAR 01], [DAR 00]).

This phenomenon has been attributed to an increase in the stiffness of the dam
due to construction joints being tightened by the increased hydrostatic pressure.
When water reaches a certain level, the phenomenon is counterbalanced by the
added mass of the reservoir and the resonance values decrease. Figure 6.24 shows
the level was reached between July and August.

Figure 6.24. Effect of the water level on the reservoir

230 Dynamic Behavior of Concrete and Seismic Engineering

As for the project previously described, the experimental results have been used
to develop and calibrate a finite element model of the same type as that shown in
Figure 6.21. The predictions of this model, in terms of the frequency response of the
acceleration on the crown, have been compared with test results for each measuring
position. The main objective of this project was to model the influence of water
level variation inside the reservoir on the dynamic response. For this structure, a
10% reduction of the dam stiffness simulated the opening of the joints in the dam
due to a decreasing hydrostatic pressure.

6.7. Conclusion

A numerical model is only valid if it can accurately represent the behavior of a

structure. Representing the dynamic behavior of a civil engineering structure is far
more difficult than representing its static behavior, and an engineer’s experience is
required to determine the important parameters governing the response of the
structure. As far as dimensioning or verifying the safety of a large structure is
concerned, all the techniques presented in this chapter have been shown to constitute
reliable and economical methods to confirm the hypotheses used for the building up
of physical models. Furthermore, these techniques are at the root of new damage
prediction methods in which multiple tests are used to follow the behavior of a
structure over time.

6.8. Acknowledgements

The authors would like to acknowledge the support of the Civil Engineering and
Natural Science Research Committee in Canada, the Quebec Technology and Nature
Research Fund, as well as that of industrial partners who have contributed to the
different research projects on the dynamic behavior of large civil engineering
structures, notably the Ministry of Transport in Quebec, Hydro-Quebec, the Swiss
Water and Geology Federal Office, Electricité de France and Emosson S.A.

6.9. Bibliography

[AND 99] ANDERSEN P., BRINCKER R., PEETERS B., ROECK G. (DE), HERMANS L.,
KRÄMER C., “Comparison of System Identification Methods Using Ambient Bridge
Test Data”, 17th International Modal Analysis Conference (IMAC), Kissimee, Florida,
USA, pp. 1035–1041, 1999.
 Experiments on Large Structures 231

[BEN 00] BENDAT J.S., PIERSOL A.G., Random Data: Analysis and Measurement
Procedures, third edition, Wiley Interscience, New York, USA, 2000.
[BOU 02] BOUAANANI N., PAULTRE P., PROULX, J., “Two-Dimensional Modeling of
Ice Cover Effects for the Dynamic Analysis of Concrete Gravity Dams”, Engineering and
Structural Dynamics, 31, pp. 2083–2102, 2002.
[COL 90] COLLECTIF, US National Research Council Panel on Earthquake Engineering for
Concrete Dams, Earthquake Engineering for Concrete Dams: Design, Performance, and
Research Needs, National Academy Press: Washington, DC, USA, 1990.
[DAL 93] DALY J.W., RILEY K.G., MCCONNELL K.G., Instrumentation for Engineering
Measurements, second edition, John Wiley & Sons, New York, USA, 1993.
[DAR 95] DARBRE G.R., “Strong-motion Instrumentation of Dams”, Earthquake
Engineering and Structural Dynamics, 24, 1101-1111, 1995.
[DAR 00] DARBRE G.R., SMET C.A.M. (ED), KRAEMER C., “Natural Frequencies
Measured from Ambient Vibration Response of the Arch Dam of Mauvoisin”,
Earthquake Engineering and Structural Dynamics, 29, pp. 577–586, 2000.
[DAR 01] DARBRE, G.R., PROULX J., “Continuous Ambient-vibration Monitoring of the
Arch Dam of Mauvoisin”, Earthquake Engineering and Structural Dynamics. 31,
pp. 475–480, 2001.
[DEI 82] DEINUM P.J., DUNGAR R., ELLIS B.R., JEARY A.P., REED G.A.L., SEVERN
R.T., “Vibration Tests on Emosson Arch Dam, Switzerland”, Earthquake Engineering
and Structural Dynamics, 10, pp 47–470, 1982.
[DOE 96] DOEBLING S.W., FARRAR C.R., PRIME M.B., SHEVITZ D.W., Damage
Identification and Health Monitoring of Structural and Mechanical Systems from
Changes in their Vibration Characteristics, Los Alamos National Laboratory, USA, LA-
13070-MS, 127 pages, 1996.
[DUR 88] DURON Z.H., HALL J.F., “Experimental and Finite Element Studies of the Forced
Vibration Response of Morrow Point Dam”, Earthquake Engineering and Structural
Dynamics, 16, pp. 1021–1039, 1988.
[FEN 84] FENVES G., CHOPRA A.K., EAGD-84, “A computer program for earthquake
analysis of concrete gravity dams”, Report No. UCB/EERC-84/11, University of
California, Berkeley, USA, 1984.
[FOK 86] FOK K.L., HALL J.F., CHOPRA A.K., EACD-3D, “A computer program for
three-dimensional earthquake analysis of concrete dams” Report No. UCB/EERC-86/09,
University of California, Berkeley, USA, 1986.
[HAL 88] HALL J.F., “The Dynamic and Earthquake Behavior of Concrete Dams: Review of
Experimental Behavior and Observational Evidence”, Soil Dynamics and Earthquake
Engineering, 7(2), pp. 58–121, 1988.
[HAR 88] HARRIS C.M. (Ed.), Shock and Vibration Handbook, third edition, McGraw-Hill,
New York, USA, 1988.
232 Dynamic Behavior of Concrete and Seismic Engineering

[MCC 95] MCCONNELL K.G., Vibration Testing: Theory and Practice, John Wiley & Sons,
New York, USA, 606 pages, 1995.
[PAU 95] PAULTRE P., PROULX J., TALBOT M., “Dynamic Testing Procedures for
Highway Bridges Using Traffic Loads”, ASCE J. Struc. Eng., 121(2), pp. 362–376, 1995.
[PAU 97] PAULTRE P., PROULX J., “Dynamic Testing of Large Scale Structures”,
International Association for Bridges and Structural Engineering (IABSE), 7(1), p. 29–
34, 1997.
[PAU 02] PAULTRE P., PROULX J., CARBONNEAU C., “An experimental evaluation of
ice cover effects on the dynamic behaviour of a concrete gravity dam”, Earthquake
Engineering and Structural Dynamics, 31, pp. 2067–2082, 2002.
[PRO 97] PROULX J., PAULTRE P., “Experimental and Numerical Investigation of Dam-
Reservoir-Foundation Interaction for a Large Gravity Dam”, Canadian Journal of Civil
Engineering, 24(1), pp. 90–105, 1997.
[PRO 01] PROULX J., PAULTRE P., RHEAULT J., ROBERT Y., “An Experimental
Investigation of Water Level Effects on the Dynamic Behavior of a Large Arch Dam”,
Earthquake Engineering and Structural Dynamics, 30(8), pp. 1147–1166, 2001.
 Chapter 7

Models for Simulating the Seismic

Response of Concrete Structures

7.1. Introduction

This chapter will present the principles of the mode most frequently used for the
simulation of the collapse behavior of reinforced concrete structures subjected to
seismic loading. Non-linear calculation methods have proved their utility for the
preparation and interpretation of dynamic, pseudo-dynamic and static tests on
shaking tables and reaction walls, but now they are starting to be used to validate
certain design rules, as well as for the seismic reassessment of existing structures
which are more vulnerable to earthquakes.

The quality of this type of modeling, which is not currently regulated, depends
on numerous factors. Thus, these models should take into account:
– the material effects, which may give rise to local non-linearities;
– the structural effects (mass distribution and behavior of the bonds);
– the environmental effects (support-structure interaction).

Within this context, two aspects affect the modeling, i.e. the structure
discretization and the local behaviors of the materials and “interfaces”, either

Chapter written by Didier COMBESCURE, Nicolas ILE, Jacky MAZARS and Jean-Marie
REYNOUARD.
234 Dynamic Behavior of Concrete and Seismic Engineering

between the materials themselves or between the structure and its support. We
should endeavor to deal with both aspects consistently to achieve relevant results.

We present below a few examples of the opportunities afforded by the literature

to deal with analysis of the seismic operation of concrete structures. Generally, non-
linear models are divided into two large families, depending on the scale
represented: on the one hand, global and semi-local simplified models (fiber and
multilayer models), based on bar or beam elements, and on the other hand, more
comprehensive models that rely on volume or surface finite element meshing.

Each has advantages and drawbacks shown revealed by applications, implying

different types of structures tested in the laboratory on a shaking table or with a
reaction wall owing to a pseudo-dynamic method.

7.2. Different discretization families

7.2.1. Global modeling

Such models reproduce the behavior of a whole structure element, such as a

beam, column, concrete wall or masonry wall, a whole storey or even a whole
building. The laws of behavior are generally uniaxial and link a global strain-
curvature, shear, strain, extension, relative displacement with its associated global
stress: moment, shear load or normal load.

Multilinear curves are often used to represent the behavior of reinforced concrete
elements: concrete cracking follows the elastic behavior; it brings about lower
stiffness then steel plastification and sometimes concrete crushing or steel failure.
Simplified laws allow us to reproduce the main phenomena observed with cyclic or
seismic loading (decrease of the overall stiffness, strength degradation under cyclic
loading, recovered stiffness when cracks close).

Figure 7.1 shows a shear law for the study of not very slender reinforced
concrete walls, as well as the modeling of a masonry in-filled wall owing to two
diagonal connecting rods. These global laws generally apply to beams or diagonal
beams.
 Models for Simulating the Seismic Response of Concrete Structures 235

Figure 7.1. Overall models of shear and masonry in filled walls

7.2.2. Semi-global modeling

Multi-fiber and multi-layer models are examples of semi-global models: they

rely on beam elements and cover the conventional laws of beam theory – the
hypothesis of plane sections that remain plane for instance. Here, the inner stresses
are calculated from uniaxial laws that reproduce the behavior of each material
(concrete, steel, etc.) instead of empirical laws. Such an approach by of this
“simplified method” frames results in generating problems of moderate size and
with a quality standard that is respectable when compared to those obtained using
conventional 2D or 3D finite element calculations, particularly if the discretization,
the behavior models and the boundary conditions are well-defined.

Multi-layer discretization is well suited to dealing with plane problems involving

slender structures. Later, among the applications, we will present an example
illustrating “multi-layer” elements and including a damage model that will be
described further in section 7.4.2. These elements include two nodes and consist of
superimposed layers with non-linear behavior, specialized according to whether they
include reinforcements or not (Figure 7.2). The Navier-Bernouilli hypothesis on
cross-sections attributes a uniaxial behavior to the layers [MAZ 98]. As far as layers
236 Dynamic Behavior of Concrete and Seismic Engineering

are concerned, introducing reinforcements relies on a partial pressure law that

implies non-sliding between steel and concrete [LAB 91]. Thus, inside each layer:

H H steel H concrete Eb concrete modulus

Ea steel modulus
E 1  C  E  C
a
b b a
b
Ea

E homogenized (a + b) modulus
H inel . 1  C  H a
b inel . concrete  Ca b H inel . steel

steel area H inel . concrete concrete inelastic strain

Ca b
concrete area
H inel . steel steel plastic strain

beam element
composite layers
(concrete + steel)

Figure 7.2. Principle of discretization by multi-layer beam elements

Following the same principle, multi-fiber discretization is more efficient,

because it makes it possible to study the behavior of structures, for slender elements
at least. We will present the principle behind writing this type of element more
thoroughly, based on a Timoshenko beam element with shear strains ([GUE 97],
[PEG 93]).

The shear and axial strains inside fiber i are written respectively as functions of
the general strains of the beam element and of the section geometry:

In 2D (multi-layer beams):

H x i H x  yi ˜ M z

and
 Models for Simulating the Seismic Response of Concrete Structures 237

 J y i Jy

In 3D (multi-fiber beams):

 H x i H x  yi M z  zi M y

 y y i J y  zi Ix

 J z i J z  yi Ix

where Hx is the average axial strain, yi is the distance between fiber i and the average
fiber, Iy and Iz are the bending curvatures (Iy = dTy/dx and Iz = dTz/dx), Ix = dTx/dx
is the unit angle for torsion, and Jy and Jz are average shear strains.

The bending moments My, Mz and the normal force Nx are obtained by
integrating the axial strains V x i corresponding to the strains H x i :

Nx ³ V dS , M ³ z V dS and M
x y x z  y V x dS
³ [7.1]
s s s

In the case of a Timoshenko beam element, shear loads Ty, Tz and the torsion
moment Mx are calculated by integrating the shear stresses W y i and W z i derived
from J y i and J z i :

Ty ³W dS , T ³W dS
y z z [7.2]
S S

Mx ³ y W
s
z  z W y dS  [7.3]

For a shear elastic material, the behavior law is written as follows:

W y i D y G  J y  and W z i
i
D z G  J z i [7.4]
238 Dynamic Behavior of Concrete and Seismic Engineering

where G is the elastic shear modulus: G = E/2(1+Q), and parameters Dy and Dz are
correction factors that take non-uniform shear stress distribution into account.

We observe that all kinematic hypotheses presuppose the absence of any

warping. Torsion behavior does not respect Saint-Venant’s theory, and the shear
strain distribution is only valid for circular sections.

Figure 7.3. Principle of multi-fiber models

When beam elements without any shear strain are used (Bernouilli’s
assumption), the shear forces are generally chosen to be equal to the bending
moment derivative.

Figure 7.4. Example of a curvature-moment relationship

 Models for Simulating the Seismic Response of Concrete Structures 239

Multi-fiber or multi-layer models have several advantages in comparison to

global models. First, the parameters of the model are the characteristics of the
material and the section geometry instead of the curvature-moment or displacement-
stress global curve. Thus, a section analysis with a fiber model allows us to identify
the first-load curve of a curvature-moment global model. Figure 7.4 shows the
results of the section analysis of a column and a beam.

Figure 7.5. Influence of the axial stress on the resistance and ductility of a column

Secondly, modeling takes into account the effect of the normal stress on both the
bending moment and the ductility of the section. The curvature-moment
relationships in Figure 7.5 derive from the analysis of a column section submitted to
different axial stresses, and clearly show not only the section modulus increase, but
also the decrease of ductility caused by the normal stress rise.

7.2.3. 2D and 3D fine models

The subject here involves the most exhaustive finite element description
possible. Among others, mass elements will be used in 3D, as well as shell elements,
interface or joint elements or even plane or asymmetric elements for associated 2D
problems (see Figures 7.19 and 7.30). For further details, refer to the wide literature
on the subject.

At this level of modeling, each material – masonry, concrete, reinforcement,

mortar joints or dry joints – can be represented. Associating these laws can then
allow us to predict the global behavior of a structure as well as its local response, but
heavy calculation often restricts their use to the analysis of a structure element,
geometrically simple specific structures or structures with simplified loading (static
240 Dynamic Behavior of Concrete and Seismic Engineering

loading equivalent to seismic forces for instance). Hence, the use of such models
should be reserved for the detailed analysis of the operation of a studied object. As
we shall see later, this can be the case for better understanding the behavior of a
model during an experiment, or for deepening the study of an existing structure
(when re-assessing nuclear or industrial facilities or a dam for example). This
necessarily implies both accurate discretization work and choice of constitutive
models and boundary conditions. Such modeling can also prove useful in identifying
the parameters of the global or semi-global models described above.

7.3. Behavior laws for concrete

The quality of modeling mainly relies on the ability of constitutive laws to

reproduce the physical phenomena characterizing complex composite materials like
reinforced concrete. For 2D and 3D modeling, these constitutive laws lean on such
theories as the theory of damage and plasticity. Uniaxial laws are usually enough for
global and semi-global models. Hereafter we describe and explain the principles of a
few laws used for concrete and steel.

7.3.1. Semi-empirical mixed models

Semi-empirical laws are generally uniaxial and are directly based on

experimental observations. They allow reproduction of post-cracking softening, the
unilateral behavior of concrete (renewal of stiffness when the cracks close up), as
well as softening after compression strength has been reached, taking this softening
confinement and the compression strength into account. Below, we present a law
inspired by Hognestad’s model [HOG 51].

During compression, or with monotonous loading, the law is the parabolic type,
becoming linear with softening (Figure 7.6a). A plateau with residual stress not
equal to zero can be defined after the softening. When using civil engineering
conventions (compression positive strain), the following values are obtained:
 Models for Simulating the Seismic Response of Concrete Structures 241

For 0<H<Hc0, V H § H · [7.5]

¨ 2.0  ¸
V c0 H c0 © H c0 ¹
For Hc0<H, V [7.6]
1  Z H  H c0 
V c0
On the plateau, V V pt [7.7]

where Vc0 is the peak stress, Hc0 is the peak strain, Z is the slope after the peak and
Vpt represents the residual stress.

Under cyclic loading, the compression non-linearity generally comes with

stiffness degradation and the appearance of plastic strains.

In traction, concrete has strength, yet after cracking, its behavior is brittle. The
law is linear for each traction part (Figure 7.6) and:
– up to the peak, the behavior is linear;
– softening is linear up to the ultimate strain.

Vc0
Z

Vpt

Hc0
a. Compression b. Traction

Figure 7.6. Example of compression and traction constitutive laws for concrete

7.3.2. Damage model

Based on damage mechanics [LEM 90], [MAZ 86], the model [LAB 91] fits the
description of the damageable behavior generated by the creation of micro-cracks
(decrease of stiffness). During the cycles, the operation is linked to micro-crack re-
closing. Two scalar damage variables are used, one in traction D1 and the other in
compression D2, acting on the stiffness of the material and generating permanent
strains. The 3D constitutive laws are given below:
242 Dynamic Behavior of Concrete and Seismic Engineering

İ İ e  İ p : partition of the plastic and permanent strain tensors [7.8]

V V Q
He    V  Tr V1  [7.9]
E0 1  D1  E0 1  D2  E0

E1 D1 wf  V  E2 D2
Hp   1 [7.10]
E0 1  D1  wV E0 1  D2 

where E0 is the Young’s modulus for the healthy material and Q is the Poisson’s
coefficient.

V V   V  , with V  and V  are respectively the “traction stress” and the

“compression stress”; D1 and D2 are respectively the traction and compression
damage variables. Their evolution beyond a specific threshold, between  (healthy
material) and 1 (broken material), is linked to the evolution of the local elastic
energy.E1 and E2 are constants whose values calibrate the evolution of the
permanent strains, and f V  is the function of the crack opening-closing up
adjustment.

Figure 7.7. Uniaxial response of the model with unilateral damage (from [LAB 91])

Figure 7.7 gives the uniaxial response of the model with a compression-traction
alternate loading. Beyond the peak in traction, we can see that damage (D1) evolves,
unloading reveals a Young’s modulus affected by crack opening, E0 (1-D1), and a
 Models for Simulating the Seismic Response of Concrete Structures 243

residual strain gradually gets cancelled out (in equation [7.10] 0<df V  / V  1 
until the cracks in compression close up ( df V  / V 0 for the closing-up stress Vf
). Beyond a compression threshold, the damage (D2) evolves, which creates non-
linearity, reducing the modulus, E0 (1-D2), and definitive permanent strain. When Vf
is reached, the cracks open again, which reactivates the effects of D1
( 0  df V  / V  1 . For further details refer to [LAB 91].

7.3.3. Plasticity model for concrete

Plasticity models are more common in the world of digital calculation of

structures, mainly because they were developed for metals. Adaptations to the
specific behaviors of materials like concrete have been recent developments, but
give effective models that agree with the damage model principles and can describe
the appearance of cracks and the effects of their opening and closing. Here, we
present the concrete model in detail, which were developed at the INSA in Lyons
[MAR 99], within the framework of plasticity theory. The failure surface is
described by two criteria: Nadai in compression and bi-compression, and Rankine
for traction cracking. The flow behavior law is associated and the strain-hardening
rule (whether positive or negative) is isotropic. The behavior of cracked concrete is
described using the distributed cracking concept, which considers the cracked
material as a continuous medium. When the failure surface in the traction field has
been reached, the biaxial plasticity is given up, and an orthotropic law is activated.
Thereafter, the process of describing cracking involves three independent
relationships between stress and strain, as defined in a system of coordinates parallel
and perpendicular to cracking direction. The unilateral feature of concrete cracked
during cyclic loading is dealt with by the restoration of the corresponding stiffness.

7.3.3.1. Uncracked concrete

Two failure surfaces are defined for the compression and traction domains (see
Figure 7.8). Nadai’s 2-parameter failure surface, of the Drucker-Prager type, is
defined using a linear relationship between the octahedral stresses W oct and V oct .
The mathematical expression of the surface for the compression domain is defined
by:

W oct  aV oct
f comp V oct ,W oct   f cc 0, V 1  0 and V 2  0 [7.11]
b
244 Dynamic Behavior of Concrete and Seismic Engineering

Maximum principal stress criterion

Nadal compression criterion

Figure 7.8. Failure and loading domains in the main stress plane

For the traction domain (pure traction or compression-traction), the failure area
is expressed as follows:

ftrac  ft c 0, V 1  0 and/or V 2 ! 0 [7.12]

where:

2J2 2
V oct I1 / 3 and W oct V eq [7.13]
3 9

V eq is the Von Mises equivalent stress:

E 1 2 E
a 2 ,b and E f cc / f ccc [7.14]
2E  1 3 2E  1

where:
'
– f t = concrete strength in uniaxial traction;
– f c' = concrete strength in uniaxial compression;
 Models for Simulating the Seismic Response of Concrete Structures 245

– f cc' = concrete strength in biaxial compression for V xx / V yy 1 and W xy 0 ;

– I 1 = first invariant of the stress tensor;

– J 2 = second invariant of the stress deviator.

The parameters of criterion a and b , functions of D and E , are derived from

three tests: uniaxial compression f’c, uniaxial traction gt’ and biaxial compression f’cc,
( V xx / V yy 1 and W xy 0 ).

7.3.3.2. Cracked concrete

When the failure area in the traction domain has been reached, the behavior of
the material is considered as uniaxial in each orthotropic direction, given by the
directions parallel and perpendicular to the crack. The direction of the crack is
perpendicular to the direction of the traction maximum main stress (Figure 7.9); it is
traditionally determined owing to axis rotating matrices.

Y
y x
)
V
V
) X

Figure 7.9. Axes linked to the crack

During the loading increment that causes cracking in a given material point,
passing from a biaxial elastoplastic model to an uncoupled orthotropic model
requires specific treatment, and several connecting rules are used to ensure
continuity between both models. The behavior of cracked concrete is then described
by a cyclic uniaxial law in all directions. The stress tensor within the local reference
axis is complemented by the shear stress and elastically calculated with a restricted
modulus of transverse elasticity to account for the meshing effect of the crack that
corresponds to friction at both surfaces of the crack lips. This effect is implicitly
taken into account in the law by arbitrarily decreasing the initial modulus of
elasticity in shear using a shear transfer factor K . Best values depend on the kind of
problem, but an initial value of factor K ranging from 0.2 to 0.4 seems capable of
246 Dynamic Behavior of Concrete and Seismic Engineering

giving correct results [ILE 00]. This parameter can be adjusted. Whenever there is a
pronounced rotation of the main directions, a second series of cracks is created
perpendicular to the existing one.

7.3.3.3. Cyclic uniaxial law

The uniaxial law implemented allows us to account for the main phenomena
observed when loading consists of a few alternating cycles.

Stress

fc PF

E0

ft 2
1
8 Ets
3
Htm
7 Strain
E1 9
E1 4
-ft

E2

6
5

-fc

Figure 7.10. Response of the model to a traction-compression cycle

Let us examine (Figure 7.10) the behavior of a point initially in traction which
cracks completely before being subjected to opposite compression reloading. The
concrete is elastic until the strength in simple traction is reached: ft (path 1), then it
cracks according to negative stiffness (slope Ets, path 2), up to a Htm H rupture
trac
strain.
Beyond this, crack opening occurs without any stress (path 3). When the stress
changes its direction, an increasing compression stress is necessary for the crack to
gradually close up (slope E1 z E0, path 4). The latter is considered to be completely
closed for a stress lower than –ft, the value at which stiffness is completely restored
(path 5). The description of path 4 is based on tests that show the opposite lips of a
crack do not coincide, that they become strained under the effect of a stress which
tends to close the crack up, and that the stiffness of the uncracked concrete is only
restored with complete closing. Nevertheless, the crack closes up with a stress equal
 Models for Simulating the Seismic Response of Concrete Structures 247

to zero, provided the strain is higher than a specific threshold: 3*Htm. Path 5 follows
the non-linear law of concrete in simple compression to a new inversion of loading,
which involves a discharge along the E2 slope straight line (path 6) passing a focal
point (fc; H0), with H0 fc /E0 , as suggested by Mander et al. [MAN 88] and Park
[PAR 90].

A concept similar to the damage concept affecting the modulus can also be found
with stress intensity. When the stress exceeds -ft (path 7), the E1 modulus
corresponding to the crack closing up is still found. Paths 8 and 9 obey the same
rules as paths 3 and 4. The behavior of an initially compressed point (or a spot that
has not completely cracked due to a reversed loading) is described by similar laws,
presented in detail elsewhere [ILE 00].

7.3.4. Cyclic models for steel

Because of its geometry, the reinforcement can only be modeled with uniaxial
laws, even for calculations involving 2D or 3D problems.

As a rule, the laws used are elasto-plastic, with or without hardening. The most
exhaustive laws [MEN 73] successfully reproduce phenomena including non-linear
strain hardening (Figure 7.11a), Bauschinger’s effect with cyclic loading (Figure
7.11b) and compression bar buckling when frames are not close enough (Figure
7.11c).

a) Behavior with monotonous loading

248 Dynamic Behavior of Concrete and Seismic Engineering

b) Behavior with cyclic loading

c) Taking buckling into account

Figure 7.11. UNI-STEEL behavior laws (from [MEN 73])

7.3.5. Taking construction layouts and second-order phenomena into account

Beyond material behavior, the quality of the model involves taking into account
some phenomena that develop at the interfaces between materials (between steel and
concrete, between soil and structure) or phenomena that are related to the severity of
calculation assumptions (such as the effects associated with the transverse load or
the effects of the geometric second order for slender elements). A few indications to
help take them into account are given below.
 Models for Simulating the Seismic Response of Concrete Structures 249

7.3.5.1. Influence of the transverse reinforcement (stirrups)

Due to the formalism of multi-layer and multi-fiber models, some building
arrangements can directly influence material laws. In this way, stirrups increase the
ductility of an element and act on its strength. They are taken into account by
increasing, in the behavior law of concrete, stress at the peak and/or by raising the
post-peak phase to improve the failure strain when the volume ratio of the stirrups
increases. As far as column-type or beam-type elements are concerned, stirrups
delay the buckling of steel under alternating loading.

7.3.5.2. Anchoring and overlapping of the reinforcements

The phenomena linked to the steel/concrete bond (lack of overlapping or lack of
anchoring) that tend to soften the structure and increase displacements can be
modeled by specific non-linear and non-dissipative laws. Taking these phenomena
into account can become really important in old reinforced concrete structures that
do not respect the latest construction layouts (generally having inadequate
overlapping or anchoring length). A modeling example ([CON 01], [MON 00] and
[XIA 97]) involves drawing a parallel between a steel model and a steel-concrete
interface model using an adapted law that splits the total strain into two parts: one
related to the behavior of steel and the other describing sliding between steel and
concrete [ELI 93].

7.3.5.3. Taking transverse load non-linearity into account

In Timoshenko beam elements, cross-sections do not stay orthogonal to the
longitudinal axis. This can be taken into account by additional stresses linked to the
shear diagonal cracking and enables verification of the transverse load collapse
modes. The law describing layer behavior is no longer uniaxial, but should take the
shear component forces into account [GHA 98]. As far as spatial problems (multi-
fiber element) are concerned, the problem is somewhat more complex, but
introducing a warping function is a good way to proceed [CAZ 03].

7.3.5.4. P-Delta effects

For very slender elements supporting an important axial load – a bridge pier for
instance – a second-order bending effect due to the displacement at the top is added
to the front-mass inertia effect. The digital implementation of this phenomenon for
seismic conditions can be found in [GHA 98].

7.3.5.5. Structure-support interaction

This effect can be decisive. The support – typically the soil – has its own
behavior and, at the interface, generates rotation and/or uplift effects. The main
phenomena can be taken into account either using a 2D or 3D representation of the
250 Dynamic Behavior of Concrete and Seismic Engineering

soil foundation or by using an integrating interface element, the softening of the

support and the non-linear effects associated with the behavior of the support and
the uplifts [CRE 01].

7.4. A few examples with their validation through experiments

Because of the complex phenomena they reproduce, and therefore the number of
law parameters they use, using non-linear models involves permanent confrontation
with results from shaking table, pseudo-dynamic and static tests. The experimental
results used here come from tests carried out with the CEA AZALEE shaking table
(Saclay-France) and the JRC ELSA reaction wall (Ispra-Italy) within the scope of
two research programs dealing with reinforced concrete building structures, namely:
– the French CAMUS (Conception et Analyse des Murs sous Seisme)
program; and
– the ECOEST-ICONS (Innovative Seismic Design Concepts for New and
Existing Structures) European programs.

7.4.1. Application of the semi-global method to a four-storey structure

Within this context, validating non-linear calculation models first requires using
the results from elementary tests on structure elements like columns, then applying
these models to several-storey structures in which each element is submitted to
series of complex and realistic loadings. The second stage notably allows the quality
of the hypotheses linking the different elements to be checked. In the next section,
we present a few results derived from the second phase of the study on an RC frame
in a “multi-fiber beam” modeling context.

7.4.1.1. Experimentation
Within the framework of the ECOEST-ICONS European projects, two scale-1,
4-storey reinforced concrete frames were tested on the Ispra (Italy) JRC reaction
wall using the pseudo-dynamic method. Both frames had the same dimensions and
reinforcements, and differed only in the presence of masonry in filled walls (Figure
7.12 and [COM 96]). The reinforcement in these frame structures use building
arrangements similar to those used from 1940 to 1970 in Mediterranean countries:
smooth steel-bars, insufficient numbers of widely spaced stirrups with regard to
current standards, and low reinforcements at nodes. The unfilled walls were made of
perforated clay bricks.
 Models for Simulating the Seismic Response of Concrete Structures 251

Frame without Frame with

filling filling
Seismic level 475 years – 2.18 m/s2
Displacement 60.8 mm 10.2 mm
Shear load at the basis 209 kN 754 kN
Seismic level 975 years – 2.88 m/s2
Displacement 116.7 mm(1) 22.3 mm
Shear load at the basis 217 kN 846 kN
Seismic level 2000 years – 3.73 m/s2
Displacement / 40.6 mm
Shear load at the basis / 529.2 kN

Table 7.1. Main experimental results of pseudo-dynamic tests on 4-storey frames carried out
at the ELSA (Ispra, Italy). Tests stopping before the end of the signal

0.20 0.20 0.50 0.20

2.30 4.65 4.65

0.50

2.0 x 1.1 1.2 x 1.1 2.20

0.50

2.20
1.30 m
2.0 x 1.1 1.35 m 2.65 m
1.2 x 1.1

0.80 m
0.50
0.30 m
1.30 m 2.20
2.0 x 1.1 1.2 x 1.1

0.80 m 1.30 m
0.50

0.30 m 0.30
2.60 m 2.20
1.2 x 1.1
1.30 m 2.0 x 1.9
0.80 m 0.80 m

0.20 0.20 0.60 0.20

2.30 4.60 4.60

Figure 7.12. 4-storey frame tested at ELSA (CCR, Ispra, Italy)

The artificial signals used for these high-level pseudo-dynamic tests

corresponded to earthquakes with 475, 975 and 2,000 year return periods. Front
displacements and shear loads at the basis of the structure measured during the tests
are given in Figure 7.1 for the seismic levels applied.
252 Dynamic Behavior of Concrete and Seismic Engineering

Second floor of the frame without any filling

Figure 7.13. Examples of the relative displacement-transverse load relationship between

storeys for an earthquake with a 975 year return period

For both structures, little damage could be observed for the 475-sequence period
earthquake, whereas important displacement values (tests stopping before the end of
the signal to allow the structure to be repaired) were reached for 975 (Figure 7.14)
and 2,000 sequence period earthquakes respectively, for both the infilled frame and
the frame without any infill.

The collapse mode is also different from one structure to the next: the damage is
concentrated within the second-floor columns for the frame without any infill
whereas the ground floor that has important openings (door and windows)
constitutes the critical storey of in-filled frame.

7.4.1.2. Modeling and comparison with experiments

For time dynamic calculations, the reinforced concrete frame has been modeled
with beam elements and the fiber model described previously. Each column and
beam has been modeled in three parts: the two plastic hinges have been chosen to be
non-linear, whereas the central part is considered as elastic with cracked stiffness.
The masonry in filled walls has been modeled as diagonal truss rods (stiffness and
strength) and has been determined with plane stress 2D refined calculations. These
calculations allow us to estimate the drop in stiffness and strength caused by the
appearance of openings.

Figure 7.14 shows a test-calculation comparison for the frame without infill for a
high earthquake level (475-sequence period). Thus, non-linear models can assess the
displacement and global strength of a structure. Furthermore, these calculations
highlight the concentration of strains and damage on the second floor, as was the
 Models for Simulating the Seismic Response of Concrete Structures 253

case during the tests. We should note that the damage mode with conventional linear
calculations is hardly ever predictable.

Figure 7.14. Test-calculation comparison for the unfilled structure

and the 475 sequence period earthquake

Within the frame of para-seismic regulations, the justification of that structure

with elastic calculation would involve using behavior coefficients respectively of 2.9
and 4.3 for earthquakes with 475 sequence periods (peak ground acceleration equal
to 2.18 m/s2) and 975 year sequence periods (peak ground acceleration equal to 2.88
m/s2). For the first earthquake level, the steel must have plasticized but no damage
could be observed. The 975 year sequence period test was stopped as initial crush
had started on the second floor, and the structure had to be reinforced before being
submitted to a further series of tests.

7.4.2. Semi-global and local models applied to concrete walls

Reinforced structure walls are quite widely used in France. Recent research has
allowed us to work out their para-seismic efficiency better. We will deal below with
work related to two specific studies:
254 Dynamic Behavior of Concrete and Seismic Engineering

– the case of weakly reinforced walls loaded in their plane, for which two
modeling types are used: the “multi-layer” beam simplified finite element study or a
more exhaustive 2D finite element study;
– the case of U-shaped walls loaded in two directions, the behavior of which is
described using a 3D multilayer thin shell approach integrating the local plastic
model.

7.4.2.1. CAMUS experimental program

Within the framework of the CAMUS program [COI 98], two 1/3-scale models
(CAMUS I and II) consisting of two weakly reinforced slender walls have been
tested on the AZALEE-CEA shaking table. The test layout has also been used for an
additional model (CAMUS III) by the ICONS program that was dimensioned
according to Eurocode 8 recommendations. The CAMUS model consists of two load
bearing walls without any opening, linked together by six floors, and by a lower
longitudinal girder anchored to the shaking table (Figure 7.15). Each wall is 5.10 m
high, 1.70 m long and 6 cm thick. The walls are cast in two steps (first the
longitudinal beam and levels 2 and 4, then levels 1, 3 and 5) in order to simulate the
concrete work cast stages on each floor. Additional masses have been added to the
higher and lower parts of each floor (except the ground floor) to impose a maximum
load close to the vertical stress normally present on the foundation of conventional
buildings (1.6 MPa).

The first two models have been designed to obtain a multi-storey operating mode
with concrete cracking and steel plastification on several floors (“French-style”
design). The third model – CAMUS III – had the same geometry as the first two
models and was designed according to Eurocode 8 regulations so as to have an
ultimate moment at the base of the wall next to the one in CAMUS I. The amount of
reinforcement is given in Table 7.2. It is worth noting that the bending vertical
reinforcement for the CAMUS II model is minimal (1I4.5 between the second and
the fifth floor and 1I4.5 + 1I5 on the first floor). The amount of reinforcement in
the CAMUS III model is definitely higher than the one used for the first model, and
reinforcement discrepancies are obvious on the higher levels.
 Models for Simulating the Seismic Response of Concrete Structures 255

1.70 1.70

0.90
0.90
0.90
21

0.90
6

0.90
10

0.60
2.10 2.10

Front view Side view

Figure 7.15. CAMUS mock-up

Most tests were conducted with an artificial acceleration meter deduced from the
Nice S1 spectrum of PS92 regulations, with the timescale contracted by factor
1/ 3 to respect the similarity rules. CAMUS I and II models have been the
subjects of additional tests to compare a proximal earthquake nearby (San Francisco
or Melendy Ranch-type natural signal) with a distal earthquake (Nice-type artificial
signal). The levels carried out during the test campaign were as follows:
– CAMUS I: Nice 0.24 g, San Francisco 1.11 g, Nice 0.24 g, Nice 0.40 g, Nice
0.71 g;
– CAMUS II: Nice 0.10 g, Nice 0.23 g, Nice 0.52 g, Nice 0.51 g;
– CAMUS III: Nice 0.42 g, Melendy Ranch 1.35 g, Nice 0.64 g, Nice 1 g.
256 Dynamic Behavior of Concrete and Seismic Engineering

CAMUS I
Levels At both ends of the walls Central reinforcement
th 2
5 floor 1I4.5 = 15.9 mm 4I5 = 78.4 mm2
4th floor 1I6 = 28.2 mm2 4I5 = 78.4 mm2
3rd floor 1I6 + 1I8 + 1I4.5 = 94.4 mm2 4I5 + 2I5 = 110 mm2
nd 2
2 floor 2I6 + 2I8 + 2I4.5 = 189 mm 4I5 + 2I5 + 2I = 138
mm2
1st floor 4I8 + 2I6 + 2I4.5 = 289 mm2 4I5 + 2I5 + 2I = 138
mm2
CAMUS II
Levels At both ends of the walls Central reinforcement
th 2
5 floor 1I4.5 = 15.9 mm none
th 2
4 floor 1I4.5 = 15.9 mm none
rd 2
3 floor 1I4.5 = 15.9 mm none
nd 2
2 floor 1I4.5 = 15.9 mm none
1st floor 1I4.5 + 1I5 = 35.5 mm2 none
CAMUS III
Levels At both ends of the walls Central reinforcement
th 2
5 floor 2I8 + 2I4.5 = 132 mm 2X5I4.5/200 = 159 mm2
4th floor 4I8 + 2I4.5 = 233 mm2 2X5I4.5/200 = 159 mm2
3rd floor 4I8 + 2I4.5 = 233 mm2 2X5I4.5/200 = 159 mm2
nd 2
2 floor 4I8 + 2I6 + 2I4.5 = 289 mm 2X5I4.5/200 = 159 mm2
1st floor 4I8 + 2I6 + 2I4.5 = 289 mm2 2X5I4.5/200 = 159 mm2

Table 7.2. Reinforcements of CAMUS models

 Models for Simulating the Seismic Response of Concrete Structures 257

2D
2D flexible
3D 1D flexible 1D
Model support
embedde embedde support flexible Test
type without
d d with support
interface
interface
1st
bending 10 Hz 10.3 Hz 8.5 Hz 7.3 Hz 7.4 Hz 7.3 Hz
mode
1st
pumping 45 Hz 40 Hz 28.3 Hz 22.3 Hz 19 Hz 20 Hz
mode

Table 7.3. Main natural modes, test-calculation comparisons

7.4.2.1.1. Interest of simplified modeling for para-seismic studies

Importance of linking conditions at the support-structure interface

This analysis was conducted mainly on the CAMUS I model. For this analysis,
the determination of natural modes was the subject of comparative analyses of
calculations between the multi-layer model and a refined 3D model based on
embedding conditions on a rigid foundation. The results show an excellent
correlation between both calculations led to elasticity. Preliminary tests revealed that
the two main modes were the first bending mode (in the plane of the walls) and the
first vertical pumping mode, which was particularly visible, due to the flexibility of
the table, a versatility not taken into account for the first calculations (Table 7.3). An
accurate description, including the model, table, anchoring bars and the model-table
interface mortar as an option was performed within a 2D EF model. For the two
main modes, the results show the importance of the interface being softer than the
remaining foundation), the taking into account of which allows the experimental
results. As far as the “multi-layer” calculations are concerned, the flexibility of the
table was adjusted to take the effects of the interface mortar into account. As we can
see, the results compared with the experiment prove quite convincing.
258 Dynamic Behavior of Concrete and Seismic Engineering

Figure 7.16. CAMUS model: mechanical diagrams of the simplified model for stiff support
and flexible support options; the dots represent the masses of the floors

Behavior along the loading sequence (CAMUS I)

The entire loading sequence has been modeled on a multi-layer discretization
basis, with a flexible support and the use of the damage model presented in section
7.4.2. As shown previously, two types of accelerograms have been used, one
representative of an earthquake far away (Nice), the other one of an earthquake
nearby (San Francisco), with a far shorter signal and higher acceleration peaks. The
earthquakes applied have been successively: Nice 0.25 g, San Francisco 1.13 g, Nice
0.4 g, Nice 0.71 g (the quoted intensity value is the peak ground acceleration).
Material parameters used for calculation were concrete: Young’s modulus E =
30,000 MPa, maximum compressive strength, fc = 35 MPa, traction strength, ft = 3
MPa; steel: Young’s modulus E = 200,000 MPa, elastic limit fe = 414 MPa, and
strength at failure, fr = 480 MPa).

Figure 7.17. Displacement at the top for the complete seismic sequence,
test on the left, calculation on the right
 Models for Simulating the Seismic Response of Concrete Structures 259

NOTE.– Some structural damping has been taken into account using Rayleigh’s
conventional description, both coefficients of which have been adjusted at 1% on the
first mode and 2% on the second – advances have recently been made on the subject
[RAG 99]. Finally, precautions have been taken as regards the finite element mesh
to ensure objective results on a global scale, at least. Figure 7.17 compares the top
displacement results obtained experimentally as well as by calculation for the whole
loading sequence. Other results concerning load distribution throughout the
damaged or plasticized areas have been obtained [MAZ 98]. All results are of good
quality, which tends to lend credibility to such simplified methods.

Emphasizing the interaction between local behavior and overall function

The dialog between experiment and modeling groups has revealed the
importance of interaction on local and global developments. During the test at the
bottom of the wall we have been able to measure a dynamic variation of the vertical
load with a frequency twice that of the bending moment and raising intensity.
Analysis led to an appreciation of the link between the two phenomena: progressive
cracking of the structure with cracks “breathing” (opening-closing) during loading,
and the activation of the vertical pumping mode due to the flexibility of the table
and the evolution of damage. Within this context, the damage model presented in
section 7.4.2, which takes both effects into account, has revealed a direct connection
between the closing-up stress (Vf) and the intensity variation of the vertical load. A
parametric study was conducted and Figure 7.18 shows this mutual influence for
two particular values of Vf, 3 MPa and 1.35 MPa, with the latter value leading to a
connection with the vertical load variations measured.
260 Dynamic Behavior of Concrete and Seismic Engineering

Vertical Initial model Recalibrated Tests (kN)

force model
Nice 0.25 g 115 119 138
S-F 1.13 g 150 160 198
Nice 0.40 g 119 132 146
Nice 0.71 g 140 190 248

Figure 7.18. Connection between closing up stress and dynamic variation of the load at the
bottom of the model. In the upper left hand corner, uniaxial response of the model for two
values of Vf (3.5 and 1.35 MPa). In the upper right hand corner, consequences on the
variation of the vertical load at the bottom. Below, table of the maximum values of the
compressive vertical load at the model-table interface

7.4.2.1.2. Two-dimensional discretization and plasticity models

As the stress is horizontal and parallel to the planes of the walls, half of the
structure is modeled using a biaxial local approach. An example of the complete FE
mesh of the model is presented in Figure 7.19. The mesh of the wall was chosen so
that the connection with the steel elements could be made in a position quite close to
the exact position, in accordance with the reinforcement layout. The mesh allows the
bar stoppers to be reproduced, as well as the concrete construction joints. Two-node
bar elements have been used to represent the reinforcement.
 Models for Simulating the Seismic Response of Concrete Structures 261

Mock-up
Maquette

Shaking
TSa table

Figure 7.19. 2D meshwork of the CAMUS model

5.0

4.0

3.0
Displacement (cm)

2.0

1.0

-1.0

-2.0

-3.0

-4.0
6.9 7.4 7.9 8.4 8.9 9.4 9.9 10.4 10.9 11.4 11.9
Time (s)
EXPERIENCE
CALCULATION

Figure 7.20. Test-calculation comparison: CAMUS I model, Nice 0.71 g signal

Figures 7.20, 7.21 and 7.22 show the calculation-test comparison of the front
displacement for an earthquake level that caused important damage to each model.
In each case, the non-linear plastic model is capable of estimating the overall
behavior of the structure quite well. At the local level, the digital results highlight
operating modes similar to the experimental modes – there lies the interest in such
modeling (Figure 7.23).

In the case of the CAMUS I model, the behavior is clearly influenced by a strong
localization of the strains at the level of the bar stoppers on the second floor, and by
the appearance of inclined cracks due to the shear force. Such behavior is quite
262 Dynamic Behavior of Concrete and Seismic Engineering

compatible with the operation observed during the last test: appearance of “fan-
shaped” cracks resulting from a truss operation functioning followed by bending-
predominant failure mechanism with bar failure at the level of the second-floor
stoppers.

In the case of the CAMUS II model, the behavior is of the multi-block type,
with horizontal cracks appearing at each concrete construction joint. Steel is
plasticized at the bottom of the lower floors. This kind of behavior is well expressed
in both the computed and experimental moment-curvature curves (Figures 7.24 and
7.25). The influence of the variation in vertical load dynamic is obvious in the
plastic plateau area (when the cracks are wide open and the steel plasticized)
because of the disruption brought about: the position of the plastic plateau depends
on the vertical load value.

With the CAMUS III model, designed according to the Eurocode 8 regulations,
the operating mode is quite different from the two modes previously described: a
plastic hinge at the foot of the wall and a quasi-elastic behavior above the critical
area. The large plastic rotations and dissipation of energy appeared mainly on the
first floor (Figure 7.23).

4.00
EXPERIENCE
EXPERIENCE
CALCULATION
CALCUL
3.00
Displacement (cm)

2.00

1.00

0.0

-1.00

-2.00

-3.00

-4.00
7.0 8.0 9.0 10.0 11.0 12.0 13.0 14.0

Time (s)

Figure 7.21. Calculation-test comparison: CAMUS II model, Nice 0.51 g signal

 Models for Simulating the Seismic Response of Concrete Structures 263

3.0
EXPERIENCE
EXPERIENCE
CALCULATION
CALCUL
2.0
Displacement (cm)

1.0

0.0

-1.0

-2.0

-3.0

-4.0
2.52 3.02 3.52 4.02 4.52 5.02 5.52 6.02 6.52

Time (s)

Figure 7.22. Calculation-test comparison: CAMUS III model, Melandy Ranch 1.35 g signal

Note that the collapse mode observed using the CAMUS I model could not be
predicted by elastic calculation. Moreover, due to the disruption caused by the shear
load, modeling based on a fiber model could not accurately reproduce this mode of
behavior. The CAMUS tests, together with the digital analysis carried out, have
shown the strong interaction that exists between the vertical and horizontal
directions of the vibration.

CAMUS I CAMUS II CAMUS III

Figure 7.23. Isovalues of the vertical strains in concrete

264 Dynamic Behavior of Concrete and Seismic Engineering

Moment (KNxm) Moment (KNxm)

200 150
150 1st floor base 2nd floor base
100
100
50 50

0 0
-50
-50
-100
-150 -100

-200 -150
-25 -10.0 -5.0 0.0 5.0 10.0 -20 -15.0 -10.0 -5.0 0.0 5.0 10.0 15.0 20.0
Curvature (mrad/m) Curvature (mrad/m)

Figure 7.24. Experimental moment-curvature relationships – CAMUS II – 0.51 g

Moment (KNxm) Moment (KNxm)

200 150
150 1st floor base 2nd floor base
100
100
50
50
0 0
-50 -50
-100
-100
-150
-200 -150
-250 - 200
-15.0 -10.0 - 5.0 0.0 5.0 10.0 -15.0 -10.0 -5.0 0.0 5.0 10.0 15.0 20.0 25.0
Curvature (mrad/m) Curvature (mrad/m)

Figure 7.25. Digital moment-curvature relationships – CAMUS II – 0.51 g

The direct consequence of mass uplift is the appearance of a compression-

traction dynamic vertical load, the magnitude of which can be considerable,
depending on the values of the mass-weights, as seen before. It is important to note
that the dynamic non-linear calculations are the only ones capable of estimating
variation ranges of the dynamic vertical loads required in the designing stage.

7.4.2.2. ICONS experimental program on u-shaped walls

The structure was a 3.60 m high load-bearing wall, with a u-shaped constant
section developed in accordance with Eurocode 8 regulations. The girder of the u-
section was 1.50 m long, with flanges 1.25 m long and 25 cm thick. The wall was
built on a foundation: a reinforced concrete square block 3.50 m long on each side
 Models for Simulating the Seismic Response of Concrete Structures 265

and 1 m high. Another reinforced concrete block, 2.50 m long on each side and 60
cm high was placed at the head of the wall. The top and side views of the model are
presented in Figure 7.26. The three specimens tested on the ELSA reaction wall are
all identical.

The vertical reinforcement is ensured by steel bars 10 and 12 mm in diameter

with a percentage U equal to 0.0056. The shear load reinforcement consisted of I8
steel bars with 125 mm-wide spacing in the wings and 75 mm-wide spacing in the
girder. The spacing of the diameter-8 stirrups was 90 mm wide. The wall section is
shown in Figure 7.27.
a) 16/100

b) 20

60

A
c) 12/150
85

75 125

i
360
250
150

B B
65

15

f) 16/100
250 100 d) 20/200
A
e) 16/100
100

Figure 7.26. Top and front views of the wall

The seismic action was simulated by horizontal forces applied on the center
plane of the upper longitudinal girder applied using imposed displacement. Equal
displacements were thus imposed on each end of the upper longitudinal girder,
therefore rotation within the horizontal plane of the girder was blocked. The
experimental assembly plan and the axis positioning are shown in Figure 7.28.
266 Dynamic Behavior of Concrete and Seismic Engineering

375 750 375

250 250 250

250
375  ) 12  ) 10  ) 12
Y

) 8/75 ) 8/90
250 X
500
 ) 10  ) 10
100
250 ) 8/125 0

187.
5
) 8/90
375
187. Position of axes
 ) 12 5  ) 12

250 1000 250

A - A SECTION

Figure 7.27. Wall section

MAIN REACTION WALL

X Reaction
X+ flange wall

Z
Z
Y

X Y X- flange

Ame

Elevation Top view

Figure 7.28. Experimental assembly plan

To simulate a vertical permanent load, before each cyclic test a 2,000 kN vertical
load was applied via 6 cables arranged so that the result was close to the center of
inertia of the area. Three different cyclic tests were carried out on the u-structure
with the reaction wall: firstly, a USW1 test with horizontal loading along the Y-axis,
then a USW2 test with horizontal loading along the X-axis, and finally, a USW3 test
with horizontal loading along the X- and Y-axes.
 Models for Simulating the Seismic Response of Concrete Structures 267

10
8

Y-average displacement (cm)

6
4
2 vol 3 vol 1
0
-2 vol 2 vol 4

-4
-6
-8
-10
-10 -8 -6 -4 -2 0 2 4 6 8 10
X-average displacement (cm)

Figure 7.29. Y-average displacement as a function of the

X-average displacement in the USV3 test

The displacements applied during the USW3 test are presented in Figure 7.29.
Two program levels (± 4 cm and ± 8 cm) were involved. They were applied in a
“butterfly-wing” manner: cycle n° 1 (vol 1 in the figure) carried out in the direct
direction (clockwise), then cycle n° 2 (vol 2) in the indirect direction (anti-
clockwise), then cycle n° 3 (vol 3) in the indirect direction and finally cycle n° 4
(vol 4) in a direct direction as well. Afterwards the same procedure was used for the
upper level, which led to failure during cycle n° 4 as shown by Figure 7.29.

Modeling and comparison with experimentation

Upper
Sole

Wall

Figure 7.30. Finite element mesh of the U-shaped non-rectangular wall

268 Dynamic Behavior of Concrete and Seismic Engineering

A 3D model based on multi-layer thin shell modeling has been adopted to

simulate the behavior of this structure. It should be noted that with such modeling,
transverse shear is not taken into account, and the actual 3D behavior (transverse
dilation) is not correctly described. Nevertheless, the approach allows the bi-bending
of the different parts of the wall and the shear in the plane of the flanges and girder
web to be represented quite well. The wall is modeled using integrated DKT-type
elements, whereas four-node thin shell-type elements with linear elastic behavior are
used to represent the upper sole. Vertical and horizontal steel reinforcements are
modeled using off-center bar-type elements. The concrete-steel adherence is
assumed to be perfect and the wall to be perfectly embedded at its bottom. The
meshwork used for the wall is presented in Figure 7.30.

The behavior of the wall is ruled by shear in the flanges as well as in the girder
web of the wall. As a matter of fact, the flanges change roles during the cyclic test,
as most of the shear imposed in the Y-direction is taken up again by the compressed
flange. During the 4 cm cycles, the steel reinforcements located at the flange-web
junction tends to buckle and the concrete starts to exhibit spalling. During the 8 cm
cycles, most of the steel at the ends of the flanges and the girder webs buckle, the
concrete gets crushed and collapse takes place because of the shear in the
compressed flange. The calculation-experiment comparison of the displacement-
load cyclic curves recorded during the USW3 test is presented in Figure 7.31. Figure
7.32 shows the drawing of the loads in both directions: when the displacement in
one direction changes, while keeping the displacement on the orthogonal direction
practically constant, the load required in the latter direction to keep the
corresponding displacement constant drops.

800 800
600 600
400
400
200
X Load (KN)

Y Load (KN)

0 200
-200 0
-400
-200
-600
-800 CALCULATION -400 CALCULATION
EXPERIENCE EXPERIENCE
-1000 -600
-100 -80 -60 -40 -20 0 20 40 60 80 100 -100 -80 -60 -40 -20 0 20 40 60 80 100
Displacement at the top X (mm) Displacement at the top Y (mm)

Figure 7.31. USW3 test (XY dir.): comparison with cyclic calculation

The effect of such coupling on the load-displacement diagram for both directions
then consists of quasi-vertical unloading, which can be observed in Figure 7.31.
 Models for Simulating the Seismic Response of Concrete Structures 269

Thus, the hysteresis loops are broader and the dissipated energy higher. An
important strength cyclic deterioration occurs during the cyclic loading. This can be
seen in Figure 7.32, as collapse takes place during the last loop of loading for 8 cm
displacement in the X direction and –8 cm displacement in the Y direction.

Comparing Figures 7.31 and 7.32 shows that the important phenomena observed
during the tests (stiffness deterioration, strength cyclic deterioration) are correctly
reproduced by the 3D model shell. Recent works have shown that for the same
problem it was also possible to achieve good results with enriched Timoshenko
multi-fiber elements [KOT 00].

800
CALCULATION
EXPERIENCE
600

400

200
Y Load (KN)

-200

-400

-600
-1000 -800 -600 -400 -200 0.0 200 400 600 800

X Load (KN)

Figure 7.32. USW3 test: graph of the load evolutions in X and Y directions during the test

7.5. Conclusions

In this chapter, we showed some peculiarities of the functioning of RC structures

under seismic loading and demonstrated the need to have efficient modeling tools to
analyze their responses. This results in two major aspects to be considered:
– representation of the structure with its associated masses;
– relevance of the behavior models to use for both materials and interfaces
(between the materials and between structure and support).

The choice should be made according to the expected accuracy of the results and
the opportunity of expressing simplifying hypotheses.

With slender structures, frames and structural walls, the effectiveness of semi-
global models (multi-fiber or multi-layer beams) has been proved, if the constitutive
270 Dynamic Behavior of Concrete and Seismic Engineering

models are suited to the phenomena to emphasize (material models with damage and
models for the interfaces that give rise to strong local non-linearity). Such models
are especially recommended to access global and collapse-mode behaviors.

A good calculation experiment correlation should not leave aside the difficulty in
finding local results (like crack opening or steel deformation) for important damage
levels. Such difficulties are all the more awkward as the criteria that herald damage
of the structure (and therefore its potential collapse) are generally local in para-
seismic regulations.

To obtain more realistic apprehension of local values, a solution lies in more

complete 2D and 3D discretization. We could state that the plastic model coupled
with a correct description of the limit conditions and of loading allows us to justify
the essential mechanisms of the behavior of a wall under cyclic loading. The main
drawback of this local approach lies in the amount of calculation generated by the
great number of degrees of freedom used, far more than with simplified models.
Nevertheless, the plane stressed walls are quite suitable for 2D modeling, and from
this point of view, current computer means are quite well adapted. The behavior of
walls that have a U-shaped section and are loaded in bi-bending is modeled using a
multi-layer thin shell 3D approach, based on the 2D local model. In the last case,
although the amount of calculation generated is higher than the 2D approach, it is
still quite reasonable and much lower than massive 3D modeling.

Whilst highlighting these model’s possibilities, we should not conceal their

limitations and it is important to note that the local phenomena that take place at the
concrete-steel interface can only be represented approximately according to the
concrete model used. The shear transfer factor K allows some of the transmission of
the shear stresses in the cracking plane to be taken into account, but it cannot
correctly represent the mode II cyclic energy dissipation.

Furthermore, for all the presented models, the modeling quality is affected by the
parameter identification procedures used. Some experimental procedure regulations
for the commonly admitted characterization of materials could prove quite useful.

7.6. Bibliography

[CAZ 03] CAZAUX G., Modélisation tridimensionnelle du comportement sismique

d’ouvrages en béton armé – Développement de méthodes simplifiées, PhD Thesis, LMT-
ENS de Cachan, 2003.
[COI 98] COIN A. et al., “Recherche CAMUS, Rapport final, Présentation générale,
Conclusions et perspectives”, Ministère de l’Equipement, Plan Génie Civil, Convention
9670011, 10 pages, 1998.
 Models for Simulating the Seismic Response of Concrete Structures 271

[COM 96] COMBESCURE D., Comportement sismique des portiques en béton armé
comportant des murs de remplissage en maçonnerie, PhD Thesis, Ecole Centrale Paris,
1996.
[COM 01] COMBESCURE D., SOLLOGOUB P., “Validation of non linear constitutive laws
used for the seismic evaluation of existing industrial facilities”, 4th International
Conference on Fracture Mechanics of Concrete Structures, Cachan, France, 2001.
[CRE 01] CREMER C., Modélisation du comportement non linéaire des fondations
superficielles sous séisme. Macro-élément d’interaction sol-structure, PhD Thesis, LMT –
ENS Cachan, 2001.
[ELI 93] ELIGEHAUSEN R., BALAZS G.L. “Bond and detailing”, in Bulletin d’Information
CEB, N° 217, Comité Euro-International du Béton, 1993.
[GHA 98] GHAVAMIAN S., Méthode simplifiée pour la simulation du comportement
sismique des structures en béton armé. Traitement des effets de l’élancement et estimateur
d’erreurs, PhD Thesis, LMT-ENS Cachan, 1998.
[GUE 97] GUEDES J., Seismic behavior of reinforced concrete bridges. Modelling,
numerical analysis and experimental assessment, PhD Thesis Dissertation, University of
Porto, Portugal, 1997.
[HOG 51] HOGNESTAD E., “A study of combined bending and axial load in reinforced
concrete”, Bulletin Series 339, University of Illinois, USA, 1951.
[ILE 00] ILE N., Contribution à la compréhension du fonctionnement des voiles en béton
armé sous sollicitation sismique: apport de l’expérimentation et de la modélisation à la
conception, PhD Thesis, INSA-Lyon, 262 pages, 2000.
[KOT 00] KOTRONIS, P., Cisaillement dynamique de murs en béton armé. Modèles
simplifiés 2D et 3D, PhD Thesis, LMT-ENS Cachan, 2000.
[LAB 91] LABORDERIE C., Phénomènes unilatéraux dans un matériau endommageable:
modélisation et application à l’analyse des structures en béton, PhD Thesis, University of
Paris 6, 1991.
[LEM 90] LEMAITRE J., CHABOCHE J.L., Mechanics of Materials, Cambridge University
Press, Cambridge, UK, 556 pages, 1990.
[MAN 88] MANDER, J.B., PRIESTLEY M.J.N, PARK R., “Theoretical stress-strain model
for confined concrete”, Journal of Structural Engineering, Vol. 114, N° 8, pp. 1804–
1825, 1988.
[MAZ 86] MAZARS J., “A description of micro- and macroscale damage of concrete
structures”, J. of Engineering Fracture Mechanics, Vol. 25, N° 5/6, pp. 729–737, 1986.
[MAZ 98] MAZARS J., “French advanced research on structural walls: an overview on recent
seismic programs”, Proceedings of the 11th European Conference on Earthquake
Engineering, Invited lecture, Paris CD-Rom, 1998.
272 Dynamic Behavior of Concrete and Seismic Engineering

[MEN 73] MENEGOTTO M., PINTO P., “Method of analysis for cyclically loaded
reinforced concrete plane frames including changes in geometry and non-elastic behavior
of elements under combined normal force and bending”, IABSE Symposium on Resistance
and Ultimate Deformability of Structures Acted on by Well-defined Repeated Loads,
Lisbon, Portugal, 1973.
[MER 99] MERABET O., REYNOUARD J.M., “Formulation d’un modèle élasto-plastique
fissurable pour le béton sous chargements cycliques”, Contrat EDF/DER, Rapport final,
84 pages, N° INSA-1/943/002, 1999.
[MON 00] MONTI, G., “Consistent insertion of bond-slip into beam fiber elements for
biaxial bending”, 12th World Conference on Earthquake Engineering, Auckland, New
Zealand, 2000.
[PAR 90] PARK Y.S., “Fatigue of concrete under random loadings”, Journal of Structural
Engineering, Vol. 116, N° 11, pp. 3228–3235, 1990.
[PEG 93] PEGON P., “A Timoshenko simple beam element in Castem 2000”, JRC Technical
Note, N° I.93.05, Ispra, Italy, European Commission, 1993.
[RAG 99] RAGUENEAU F., Fonctionnement dynamique des structures en béton. Influence
des comportement hystérétiques locaux, PhD Thesis, LMT-ENS Cachan, 1999.
[XIA 97] XIAO Y., MA R., “Seismic retrofit of RC circular columns using prefabricated
composite jacketing”, Journal of Structural Engineering, Vol. 123, N° 10, ASCE, 1997.
 Chapter 8

Seismic Analysis of Structures: Improvements

Due to Probabilistic Concepts

8.1. Introduction

Structures, including buildings, construction works and industrial facilities, have

to be designed to resist earthquakes. To ensure this, engineers obtain data about the
open field motions where the structure is to be built, and employ modeling and
calculation methods to enable them to determine its likely response, in order to
verify the design is acceptable.

The degrees of sophistication of the methods used will depend quite heavily on
the importance of the structure and on the costs incurred if it collapses.

As a consequence, when designing large construction works or nuclear facilities,

the most elaborate methods available are implemented. For example, we will care
about the correlation of open field seismic motions within the considered soil area,
and of the influence of structural motions on those of the soil (soil-foundation
interaction). We will take non-linear behaviors into account. We will also
characterize the motions of the components’ anchoring points in order to carry out
separate detailed calculation, etc.

The common denominator in all conventionally used methods is what is usually

called the modal method. The method, based on the linear behavior of the soil and

Chapter written by Jean-René GIBERT.

274 Dynamic Behavior of Concrete and Seismic Engineering

the structures, is presented in a quite assertive deterministic light, through the

seismic source data, through the whole approach for obtaining the response and the
conclusion as far as strength or collapse are concerned.

The current chapter aims to show that this deterministic approach can sometimes
lead to quite misleading results and that, in order to control things and to improve
the predictions, it is necessary to become aware of the probabilistic feature of the
problem raised and of the more or less justified hypotheses made. Thus, we are led
to ask ourselves questions such as:
– What model should be chosen to represent the motion of the soil?
– Is the conventionally-used “oscillator response spectrum” data still relevant?
– How can we preserve the simple feature of the new methods proposed while
keeping them consistent with regulations, especially in the case of non-linear
behavior?

8.2. The modal method

For further details see [DAV 85], [DAV 88] and [GIB 88].

8.2.1. Data about the seismic source

Open field motion is a rather complex acceleration time signal (Figure 8.1)
characterized by a level variable, the Jsmax maximum acceleration and frequency
content.

In the deterministic approach one or several earthquake levels (operation and

safety levels, for example) corresponding to different structure strength requirements
are associated with a site.

The frequency content results from various compilations and averages of

measured seismic signals. As we do not usually have any recordings of the
concerned site, we often just use general standardized data such as the “Regulatory
Guide 1-60” spectra (Figure 8.2), possibly corrected with the “site effects”.
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 275

Figure 8.1. Soil acceleration as a function of time during an earthquake:

open field measurements

Figure 8.2. Standardized spectra (at 0.15 g) of the “Regulatory Guide 1-60”
276 Dynamic Behavior of Concrete and Seismic Engineering

Thus, this results in data that are unlikely to represent a real motion but,
considering the way to characterize the frequency content and the modal method
using the data, which constitutes a consistent set.

Let us examine the situation more accurately. The frequency content of the
seismic signal is characterized by the oscillator response spectra (ORS), the
definition of which is determined by the following:
– given a high degree of freedom oscillator (mass, spring, damper system) with
imposed movement at anchoring points (Figure 8.3): xsoil(t). The oscillator is
characterized by its resonance frequency f and damping ratio H;
– given x(t) the relative motion of the oscillator mass (x(t) = xstruct(t) – xsoil(t)),
the maximum value reached by the absolute value of x(t) will be called S (f,H).

Figure 8.3. Diagram of the seismically excited harmonic oscillator

By varying f and H, we obtain a curve network that is the ORS characterizing the
xsoil(t) signal.

The ORS are typically represented in the form of:

– pseudo-velocities Spv (f,H) = 2Sf S (f,H);
– pseudo-accelerations Spa (f,H) = (2Sf) 2 S (f,H).
The ORS include 3 frequency bands (Figure 8.4):
– the very low frequency band ( 0.1 Hz) for which we can observe an
asymptotic behavior in S (f,H) = (xsoil) max;
– the frequency band where the seismic energy is located and where we can
observe an amplification effect all the wider the smaller H is;
– the high frequency band (! 20 to 30 Hz) where we can observe a behavior in S
(f,H) = Jsmax /(2Sf) 2.
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 277

Figure 8.4. The three frequency ranges of an oscillator response spectrum

8.2.2. Calculation of structural responses using the modal method

Let us consider a linear behavior structure, with the xsoil(t) motion imposed on a
specified number of its points (for instance, structures with stiff soil foundation for
which we can suppose that the anchoring points within the soil follow the open field
motion).

The x(r, t) relative motions with regard to the soil confirm the system (r = space
variable):

M G 2 x/G t 2  A G x/G t + Kx =  M U Ȗsoil (t) [8.1]

with M, A, K being the operators of inertia, viscous damping and stiffness for the
structure blocked at its anchoring points. U represents the space function “unit
translation in the earthquake direction”, and Jsoil(t) is the soil acceleration (seismic
signal).

)n (r), Zn and mn being the modal elastic lines, resonance pulsations and
generalized masses of the non-dampened structure with displacement limit
conditions equals zero at the anchoring points. The an(t) modal contributions
confirm the following uncoupled system (when supposing that the damping operator
becomes diagonal):

d 2 a n / dt 2  2 H n Zn da n / dt + Ȧn 2 a n   q n / m n  Ȗsoil  t  [8.2]
278 Dynamic Behavior of Concrete and Seismic Engineering

where Hn and qn are respectively the modal damping and the seismic participation
factor (the projection of the MU force of inertia on mode q n = ()n, M U)).

The ORS data allows the maxima reached by the absolute values of modal
contributions to be obtained without calculation from:

a n max
¬ n n n
ª q / m Ȧ 2 º S f , İ 
¼ pa n n  [8.3]

where fn = Zn/2S = resonance frequency.

Obtaining the variables useful for strength diagnostics (relative displacements

and stresses in certain points of the structure) requires an assumption.

Actually, if we are interested in the relative displacement at a specific point r, for

instance, the latter will be a linear combination of the modal contributions:

x  r, t  ¦a
n
n  t  )n  r  [8.4]

To obtain the absolute value of x, the so-called simple quadratic combination

rule is applied:

X max 2 (r) ¦a n max 2

)n2  r  [8.5]
n

This rules implies, as will be shown later, the statistical independence of the
different terms of the sum. In particular, the previous formula cannot be applied with
modes whose frequencies are beyond the seismic range.

The summation is then restricted to the first N modes (the higher limit of the
seismic range) and the contribution of the higher modes is represented by a
“pseudo-mode”, determined by virtue of an associated static solution:

2
N ­ N ½
x max 2  r  ¦a n max 2
)n2  r   Ȗs max ®° X s  r   ¦ ªq n / m n Ȧ 2 º ) n  r  ¾°
¬  n ¼  [8.6]
n 1 °¯ n 1 n 1 ¿°

with the Xs (r) static solution given by K Xs =  M U.

 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 279

REMARK 8.1.– The coherence of the method is quite obvious. An earthquake is

defined due to the response maxima of oscillators, with the maxima being precisely
the values that enable us to rule about the strength of the structure. As a matter of
fact, this implies that collapse is determined by a threshold that is exceeded once
(collapse from excessive strain and not from oligo-cyclic fatigue).

8.3. Criticism of the modal method

The drawbacks of the modal method are numerous and include:

– the simple quadratic combination hypothesis cannot always be verified;
– knowing the response maxima is not always sufficient. For example,
calculating a complex installation must be performed in several steps. The
calculation for buildings is effected, with the main elements represented through
their mass. To calculate the equipment, the motions of their anchoring points have to
be characterized. This is called the floor spectrum problem;
– collapse can be caused by combination of several cycles, which therefore
involve several relative maxima (oligo-cyclic fatigue, cumulative plastic strain);
– if we want to take a highly non-linear behavior into account, the modal method
is ineffective.

Therefore, other approaches have to be used. For instance, we might think of

resorting to time calculations. Carrying out shaking table tests implies a time
approach as well. The obvious question that must then be asked deals with
associating a temporal seismic signal with the ORS data.

In order to bring solutions to this set of questions, it is necessary to revert to the

definition of a seismic signal.

Observation of the seismic motions on a same site shows noticeable time rate
differences from one signal to the next. Such differences result from the complex
phenomena involved, and it would be mistaken to believe that it is possible to give a
detailed explanation of them.

The only solution involves representing the seismic motion of a site through a
random process that will be characterized by means of average values. The
previously defined ORS can be considered as such values. Are they relevant ones,
that is to say, is the ORS data enough to determine the average values associated
280 Dynamic Behavior of Concrete and Seismic Engineering

with the response of the structure and which are necessary to predict the collapse
probability?

In order to answer that question, we will first recap a few ideas about random
processes.

8.4. A few reminders about random processes

For more information, see [GIB 88], [PRE 90] and [KRE 83], though these are
more theoretical.

8.4.1. Definition and characterization of a time random process

Consider a set of time instants arranged in ascending order. Each is associated

with an Xn random variable. The set of these variables constitutes a random process.

The process will thus be completely characterized by the joint repartition

functions and probability densities of the Xn set. The discrete process defined above
can be more generally applied to a continuous process, denoted X(t), by having the
between tn intervals tend to zero. In that case, complete characterization obviously
requires an infinite amount of information.

From a practical point of view, we will only be able to consider a limited number
of joint probabilities, and therefore the characterization will be imperfect.

In most cases, considering only the p2 (x1,t1; x2,t2) joint probability densities or
two Xn variables will be enough. If the probability density completely characterizes
the process, the latter is said to be Markovian.

8.4.2. Second order characterization

Rather than using the p2 function, we will consider the associated moment
function, limiting ourselves to order 2 (which, in the case of a Gaussian process, is
enough to characterize it). Hence if (D) is the set of values X can assume, we can
define:
(i) the process average;
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 281

m (t) = x p1  x, t  dx
³ [8.7]
D

(ii) the correlation function of the process;

ȡ  t1 , t 2  ³³x 1 x2 p2  x1 , t1 ; x 2 , t 2  dx1 dx 2 [8.8]

( D) ( D)
DD

The process is said to be stationary if these characteristics stay invariant through

time translation. Thus, we will obtain:

m = x p1  x  dx and ȡ W 
³ ³³x x p
1 2 2  x1 , x 2 , IJ  dx1 dx 2 [8.9]
D DD

with W = t2  t1.

It is interesting to mention that for stationary processes, the “ergodicity” quality

allows us to determine m and U (W) owing to time averages carried out on an x(t)
realization:

T /2 T /2
m = lim 1/ T
T of ³x  t  dt and ȡ  IJ  lim1/ T
T of ³x  t  x  t +W  dt [8.10]
T /2 T /2

In the case of stationary processes, the power spectral density (PSD), which is
the Fourier transform of U (W), is quite commonly used. We can note:

f
S (f) = 2
f
³ ȡ  IJ  exp  2iʌfIJ  dIJ [8.11]

Quite often the signal will have an average equal to zero (the seismic signal for
example) and we will obtain:
282 Dynamic Behavior of Concrete and Seismic Engineering

f
Variance of X (t) = ı 2 ȡ  0 ³ S (f) df = m 0 0 order spectral moment
0

Physically, for the signal, the PSD represents the energy distribution according to
the frequency. The notion of dX/dt(t) derivative process will also be used; it is
defined as a quadratic mean by:

^
Expected value: E ª¬ X  t+İ   X  t   / İ  dX/dt (t) º¼
2
` o 0

İ o 0

For a stationary process, the derivative process exists if d2U/dW2 exists for W = 0.

The derivative process is “orthogonal” to the initial process for:

E > X (t) dX/dt (t)@ dȡ/dIJ (0) = 0, and its PSD is equal to (2ʌf) 2S (f).

Thus, the variance of the derivative process for a zero- average signal is:

ı 2der  2ʌ 2 ³ f 2S  f  df =  2ʌ 2 m 2 [8.12]
0

where m2 = second order spectral moment.

8.4.3. Response of a linear system to random stress

Generally we know how to express the second order characteristics of the

response of a linear system characterized by its transfer functions to stress in the
shape of a space-time random field that is also second order characterized.

In order not to make this chapter too heavy, we will develop the formulation for
the case described previously of a structure loaded with an imposed acceleration of
its anchoring.
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 283

The imposed acceleration will be considered as a zero-average stationary time

random process *(t), characterized by its PSD S(f). The structure is characterized by
its by-pulse response G (r, W) (response at point r and instant W to an anchoring
acceleration corresponding to a unit pulse exerted at instant W = 0).

Let us consider the resting structure at W = 0 and let us exert the *(t) process from
that moment. The relative displacement of the structure at point r and instant W can
be considered as a stationary random process of time t: X (r, W, t), the PSD of which
is expressed by:

2
Sx  r, IJ, f  H  r,IJ,f  S  f  [8.13]

H (r, W, f) has the meaning of a transfer function and is given by:

H  r,IJ,f  ³ G  r, u  exp  2iʌfu  du [8.14]

0

If W o f, the transfer function tends to the conventional expression of the

Fourier transform of the by-pulse response, written h(r, f).

Formula [8.13] then gives:

2
Sx  r, f  H  r, f  S (f) [8.15]

which represents the PSD of the structure response set in r.

In [8.15] we can express the H function owing to the natural mode basis of the
structure:

H (r, f) = ¦H
n
n f  )n  r  [8.16]

with H n  f   
  q n / m n  / Ȧ2  2iİ n Ȧn Ȧ + Ȧn 2  Ȧ = 2ʌf  .

Hence:
284 Dynamic Behavior of Concrete and Seismic Engineering

Sx  r, f  ¦¦ Hn m
n (f) H m  f  ) n  r  ) m  r  S (f) [8.17]

For a wide frequency band source signal (quasi-random noise), it is possible to

integrate [8.17] into the frequency domain. For example, if we take S (f) = 1, we
obtain the response variance:

ıX2  r  ¦¦ ı nm 2
) n  r  ) m (r) [8.18]
n m

with Vnm2 = 1/[2ZnZm (HnZn+ HmZm) (1+Onm2)] and Onm = «Zn  Zm «/ (HnZn+ HmZm).

In the case of quite distinct modes (Onm !! 1 for n different from m), the non-
diagonal terms in the double summation can be neglected:

ı X2  r  ¦ı
n
n2
) n2  r  [8.19]

with Vn 2 = 1 / (4 HnZn3).

Equations [8.18] and [8.19] can be interpreted in the following way.

Inside the natural mode basis of the structure, modal contributions are the motion
variables. Their evolution over time results from the random stimulation (here the
unit random noise). Thus, they are An(t) random processes and we obtain:

X  r, t  ¦A
n
n  t  )n  r  [8.20]

The Vnm2 values represent the correlation coefficients of the An(t) values with one
another (for the same value of t), whilst Vn2 values represent their variances.
Equation [8.19] shows that if the modes are very distinct, the An(t) will be
statistically “decorrelated”.
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 285

8.4.4. Using stochastic equations

After these basic reminders, it seems timely to say a few words about stochastic
equations, as they represent an important investigative tool as far as the calculation
of randomly stimulated structures is concerned. Although engineers seldom use
them, for simple cases, they can provide us with quite useful solutions for
approximate methods and the interpretation of digital simulations.

Consider a not necessarily linear N dof (degree of freedom) structure, stimulated

by a stationary Gaussian random noise (“no memory” random process). Its response
is a “one memory step” (Markov) 2N dimension vectorial process (the dofs and
their first derivatives). Such a process is completely characterized, as we have
already said, by its second order joint probability or, which is equivalent, by its
transition probability p2 (y, t / y0, t0), with t > t0, and y represents a 2N-component
vector.

We can show that such probability verifies a so-called Fokker-Planck equation,

with partial derivatives according to y and t, the terms of which can be explained
from those of the motion equations of the structure.

Let us take as a simple example a 1-dof dampened oscillator, with a stiffness

derived from a potential:

d 2 x/dt 2  2İȦ dx/dt + d  U  x   / dx = f  t  [8.21]

Initially the oscillator is at rest. We will take y0 = 0 and t0 = 0. Furthermore, we

will only be interested in its set response. Then the transition probability does not
depend on t time and it will be written as q (x, v), with v = dx/dt. The source process
F(t) is a S0 level random noise (its correlation function will be S0 G(W), with G(W) =
delta functional).

The Fokker-Planck equation confirmed by q (x, v) is:

v wq/wx  w /w ª¬ 2İȦv + dU/dx  q º¼   S0 /2  w 2 q/w 2 v = 0 [8.22]

the solution to which is:

286 Dynamic Behavior of Concrete and Seismic Engineering

¬  
q (x, v) = C exp ª   4İȦ/S0  v 2 /2  U  x  º
¼
[8.23]

being a constant such that ³³ q dx dv = 1.

8.4.5. Extrema statistics in a stationary process

Second order representation leads to simple formulation for linear systems at

least, but it is not enough as far as seismic analysis is concerned. As a matter of fact,
the values handled are the maxima reached instead of the average quadratic values.

This section will present the main formulations at our disposal, while keeping in
mind the fact that a definitely more complex problem is involved.

8.4.5.1. Wide-band processes and narrow-band processes

Extrema statistics depend greatly on the bandwidth of the process being
investigated.

Typically we distinguish between the following:

– wide-band processes, the PSD of which covers a wide frequency range. The
soil’s seismic motions without any site effect belong to this category;
– narrow-band processes, the PSD of which corresponds to a narrow peak (the
whole energy is located around a middle frequency f0). The response of a weakly
dampened harmonic oscillator with wide-band noise belongs to this category.
– intermediate processes consisting of a low number of well-separated narrow
peaks. The response of a structure on its first natural modes belongs to this category.

Each category corresponds to a quite characteristic rate of the signal. In

particular, wide-band noise presents isolated extremas, whereas narrow-band noise
consists of “extrema packages” (Figure 8.5).

The bandwidth is characterized by the parameter:

1/ 2

į = 1  m12 /m 0 m 2  [8.24]

where m0, m1 and m2 are order 0, 1 and 2 spectral moments.

 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 287

For example, the bandwidth of the response of a one-random noise harmonic

oscillator is:

1/ 2
į =  4İ/ʌ  [8.25]

Figure 8.5. Wide-band (a and b) and narrow-band

(c and d) processes PSD and time rate

8.4.5.2. Envelope of a narrow-band process

A narrow-band process consists of both amplitude and frequency-modulated
frequency f0 oscillations. The amplitude modulation gives conditions for the analysis
of extremas. It can be represented by a new process, whose envelope can be defined
by:

2 2
E2  t  X 2  t   > dX/dt (t) @ /  2ʌf 0  [8.26]
288 Dynamic Behavior of Concrete and Seismic Engineering

The envelope process is wider band. If X(t) is Gaussian, E(t) obeys a Rayleigh
law.

8.4.5.3. Statistical study of the absolute maximum reached by a zero-average

process in a given time period
This section will be directly useful for the seismic analyses described above. We
will present below a model that can be applied to wide-band Gaussian processes
before trying to extend it to narrow-band processes.

8.4.5.3.1. Average number of threshold crossings per time unit

We consider a positive threshold xm. We will deal with the crossings of that
threshold owing to the process:

X  t    x m or X  t  ! x m

The crossing average frequency is given by:

f
f  xm  ³ v p  x m , v  dv [8.27]
f

where p (x,v) are the joint probability density of the process and its derivative.

For a Gaussian process we have:

p  x, v  1 / 2ʌııder  exp ^ 1 2  x 2 / V 2  v2 / ıder 2 ` [8.28]

Once the calculations are over, by using [8.12] and introducing the a-
dimensional threshold r = xm /V we obtain:

1/2
f r 2  m 2 / m0  
exp  r 2 / 2  [8.29]

REMARK.– f  0  / 2  m2 / m0 1/2 represents the average frequency of the process

according to the Rice formula.
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 289

8.4.5.3.2. Determination of the reliability function and the peak factor

Reliability is defined as the probability for the process not to cross the threshold
in a given T time period, which is usually chosen to be non-dimensional.

N = f (0) T represents the average number of semi-cycles performed by the

process during the T time period. The reliability function is noted: W (r, N).

For wide-band processes and for relatively high thresholds (r t 2.5 to 3), the
“threshold-crossing” occurrence is both rare and isolated.

Therefore, we can represent its occurrence owing to a Poisson distribution:

Probability (n occurrences in the T time period) =

^ª¬f  r  Tº¼ / n!` exp ª¬f  r  Tº¼

n

Reliability involves a zero occurrence probability. Hence, by replacing f(r) with

its [8.29] expression:

W  r, N 
¬ 
exp ª  N exp r 2 / 2 º
¼ [8.30]

we can recognize Gumbel’s asymptotic law.

From reliability, we can define the absolute maximum reached by the process
during the time period T. This is a stochastic variable the probability density of
which is:

p max  r, N  wW/w r [8.31]

The most frequently used average value is the average maximum, which is also
called peak factor P (N) when it is referred to the mean square deviation.

f
ȝ  N ³ rwW/wr dr [8.32]
0
290 Dynamic Behavior of Concrete and Seismic Engineering

Figure 8.6. Rate of the reliability function – comparison between the different
estimations for a narrow-band process

The function W varies from 0 to 1 when r varies from 0 to infinity. The crossing
occurs quite suddenly around such r-values as W(r) = ½, which is an increasing
function of N (Figure 8.6).

Thus, we can explain integral [8.30] owing to expression [8.32]. If N is not too
large (N t 100), we have:

1/2
ȝ  N ª¬ 2 Ln  N  º¼ [8.33]

It is also possible to explain the maximum standard deviation (related to the

process standard deviation):

ı max  N  ʌ  6
1/2
> 2 Ln (N)@1/2 [8.34]
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 291

8.4.5.3.3. Extension to a narrow-band process

As the maxima of a narrow-band process come in the shape of packages, the
Poisson model for the calculation of W is no longer valid. Instead of applying the
model to the process itself, we could try to apply it to its envelope.

For a Gaussian process, the average number of maxima per package is given by:

1
E ª¬ N maxi  r  º¼
«¬  
ª1  exp  ʌ/2 1/2 įr º
»¼
[8.35]

where the bandwidth parameter G appears.

Thus, we can derive from it a threshold crossing average frequency by the

envelope:

f  r  / E ª¬ N maxi  r  º¼ [8.36]

and use this frequency as previously.

It will also be possible to adjust semi-empirical formulas inspired by what comes

before, the most sophisticated of which is Vanmarcke’s formula:

W  r, N     ^
ª1  exp  r 2 / 2 Ln2 º exp  N exp  r 2 / 2
¬« »¼  

¬«  į »¼ ¬ 
ª1  exp   ʌ/2 1/2 1.2 r º / ª1  exp r 2 / 2 º
 ¼  ` [8.37]

Figure 8.6 shows a comparison between different models and simulation results
carried out with a reduced 0.01 (G = 0.11) damping harmonic oscillator.

Figure 8.7 shows the results of digital simulations concerning the evolution of
the peak factor and the maximum standard deviation as a function of N and G.
292 Dynamic Behavior of Concrete and Seismic Engineering

Figure 8.7. Abacuses to determine the peak factor of a narrow-band process (numerical
calculation results of a harmonic oscillator stimulated by random noise)

8.5. Improvements to the modal method

Let us return to the seismic analysis problem.

What we said before will be used to rephrase the definition of seismic data,
namely the oscillator response spectra (ORS).

The seismic signal should be considered as a random process characterized by

average values suited to the response we are looking for, typically predicting the
collapse of the structure. According to the regulations and practice, the ORS
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 293

characterization is used, although in some cases it is not the best suited, as we will
show later.

Therefore, the problem involves specifying the meaning of ORS in probabilistic

terms. As we recalled at the beginning, the ORS represents the maximum reached
during the seismic stimulation by the response of a harmonic oscillator. The notion
was analyzed in probabilistic terms in the previous section.

One ORS point represents a random variable. Calculating the response to a given
seismic signal represents a carrying out of that random variable. Figure 8.8 shows
such calculation results. We can observe quite fluctuating curves that reflect the
straggling. Actually, the regulation curves used by engineers are much smoother.
Though they result from various averages not necessarily consistent with a well-
defined random process, they should be considered as a statistic average associated
with a random process.

Figure 8.8. OSR associated with a seismic signal (the higher the damping,
the lower the statistical fluctuation)

The first goal of probabilistic seismic analysis is therefore to adjust a random

process to the regulation ORS data considered as the statistic average of the
“maximum reached” random variable. In fact, the problem has several solutions that
may involve quite different results in the case of very non-linear structure behaviors.

Let us return to the criticisms of the modal method listed above. The first
concerned the simple quadratic combination hypothesis. As a matter of fact, there
are actually two distinct causes of non-verification of this hypothesis.
294 Dynamic Behavior of Concrete and Seismic Engineering

8.5.1. Complete quadratic combination

The first cause is that modal contributions are statistically dependent processes.
This is most obvious when two natural modes have neighboring frequencies.

The effect is obvious in the expression of the response variance of a structure

represented by its natural modes (equation [8.18]). The parameter characterizing the
modal contribution-coupling is written:

Ȝ nm Ȧn  Ȧm /  İ n Ȧn  İ m Ȧm  [8.38]

We can give an expression on the equation [8.18] model for the maximum. Thus
we obtain the complete quadratic combination formula:

N
x max 2  r  ¦a nmax 2
)n2  r  2 ¦ 
a nmax a mmax ) n  r  ) m  r  / 1  Ȝ nm2 
n=1 couples n,m  n z m 

2
­° N ½°
 Ȗsmax 2 ®Xs  r   ¦  
q n / m n Ȧn 2 ) n  r  ¾ [8.39]
°¯ n 1 °¿

Furthermore, we can improve the formula by taking the transient feature of the
seismic stimulation into account in introducing, instead of the modal damping,
equivalent modal damping given by:

2
 İ n equ İ n ª¬1  exp  2İ n Ȧn T  º¼ / ª¬1  exp  İ n Ȧn T  º¼ [8.40]

where T is the earthquake duration.

Formula [8.40] takes into account the fact that when the earthquake takes place,
the structure at rest includes in its response a transient phase all the more important,
compared with the T duration, the smaller the damping H.
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 295

8.5.2. Peak factor effect

The second error cause lies in the “peak factor effect”: the transposition of
formulae that can be applied to a variance calculation to maxima is generally not
valid. To understand such an effect, we will consider the special case of a structure
that responds according two almost similar contribution modes, the resonance
frequencies of which are quite separate (O12 !! 1).

In this case, the simple quadratic combination formula well applies to the
variance calculation:

2
ıx2 ¦Į ı
n
n n [8.41]

Transposition to maxima would give:

x max 2 ¦Į a
n
n n max 2 [8.42]

The latter formula is only correct if the Pn peak factors associated with each An(t)
process are equal to each other and the X(t)(PX) as well. If both modes have
neighboring damping ratios, the Gn bandwidths will also be close, and providing that
both resonance frequencies are of similar orders, both peak factors will not be very
different either.
296 Dynamic Behavior of Concrete and Seismic Engineering

Figure 8.9. Illustration of the peak factor effect

On the other hand, the GX bandwidth of the X(t) double peak process will be
definitely wider. It can be estimated owing to the formula:

1/2
į x | ªĮ 2  8 / ʌ  İ º with: Į = Ȧn  Ȧm /  Ȧn  Ȧm  [8.43]
¬ ¼

valid in the case: H1 = H2 = H o G1 = G2 = G = (4H / S)1/2  1.

For example: f1 = 4 Hz, f2 = 6 Hz, H = 0.01 o GX = 0.26 and G = 0.11.

In Figure 8.9, we represented the evolution of the peak factor according to the
bandwidth, for the response process of 1-dof and 2-dof oscillators. The curves allow
us to determine the P1, P2 and PX peak factors. In the previous example, we have, for
100 half-cycles carried out with an average 5 Hz frequency, P1 | P2 | 2.5 and PX |
3.0, that is, a 1.20 ratio. This ratio represents the error made when carelessly
applying the maximum quadratic combination formula. This error always tends to be
underestimated. In our example, it is equal to 20%.

There exist correction processes of the complete quadratic combination formula

that consist of introducing a multiplying function determined from abacuses into the
cross-terms of the sum (see for instance [GIB 88]).
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 297

REMARK.– Peak factor corrections are based on stationary process considerations.

As a first approximation, it is possible to take the transient feature into account by
replacing the damping H with the equivalent damping given by [8.40].

8.6. Direct calculation of the floor spectra

The seismic calculation of a complex structure is performed in several steps:

– the first step involves the calculation of buildings, including the soil-
foundation interaction effects and simplified modeling of large equipment;

– the second step involves detailed calculation of large equipment;

– in some cases a third step is necessary to study the secondary equipment, i.e.
the pipework connecting the main equipment.

If we want to avoid time calculations and therefore continue to apply the modal
method concept, then at the start of the second and third steps we will need the ORS
associated with the motions of the floors on which the large equipment rests or to
which the secondary equipment is anchored. However, the modal method such as it
has been described so far does not allow us to obtain such information when it is
applied to the first step.

The objective of the so-called “floor spectrum direct calculation” method is to

answer that question. As in the previous section, it implies resorting to an accurate
probabilistic definition of seismic loading. On this occasion we will be driven to
study the transient feature of the loading and of the response of the structure, which
has only been touched on so far.

8.6.1. Representation of non-stationary processes

If staying at a second order characterization, a general (non-stationary) process

fulfills a correlation function depending on two time variables:

ȡ  t1 , t 2  or ȡ  t, IJ  with: t =  t1  t 2  / 2 and W t 2  t1

A PSD can be defined either by carrying a double Fourier transform of the

correlation function or by performing a simple Fourier transform (with regard to W).
298 Dynamic Behavior of Concrete and Seismic Engineering

Then we obtain either a PSD that is a function of two frequencies f1 and f2, or a PSD
that is a function of frequency f and time t.

Obviously, adjusting a PSD to two variables during a series of observations or of

experiments is far more problematic and often illusive. A way to simplify involves
supposing that the U (t, W) function evolves more slowly according to the t variable
than according to the W variable (which often proves to be true). It is then easier to
carry out adjustments.

We can also assume that the t evolution occurs in a deterministic way, and thus
adjust a frequency content varying in time according to some given law. We can be
still more restrictive by considering an average, and therefore constant with time
frequency content, and by making only the overall amplitude time-dependent. The
latter assumption leads us to the divisible model of seismic stimulation:

*t a t Ft [8.44]

where a(t) is a deterministic envelope and F(t) is a S(f) PSD stationary random
process.

As we will see, the latter representation is fairly easy to adjust to conventional

seismic data. Generally a simple envelope (half-sine) is taken. Even a constant level
slot gives satisfactory results because the most important effect to take into account
is the start of the stimulation for t = 0 time and its stopping for time t = T.

8.6.2. Adjusting a separable process from the ORS data

We have seen that the seismic stimulation is given to an engineer in the shape of
a set of ORS drawn up for several damping values. Thus, the problem involves
adjusting an a(t) envelope and a S(f) PSD from Spv(f,H) or Spa(f,H) functions.

As we remember that the ORS have the meaning of maximum statistic averages,
we can show that the relationship:

2 2
ªSpv  f, 0  º E * f  [8.45]
¬ ¼

is quite well confirmed.

 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 299

The second term of [8.45] represents the mathematical expectancy of the f

parameter random variable:

TT
2
³³ *  t  *  t  exp ª¬2iʌf  t  t º¼ dt dt
, , ,
* f  ˜ [8.46]
0 0

By using [8.46] we obtain, if we suppose a slow a(t) variation:

T
2
ªSpv  f, 0  º | 1 S  f  a 2  t  dt [8.47]
¬ ¼ 2 ³
0

We take a standardized envelope with a given (half-sine) shape. The integral

given by [8.47] only depends on the unknown T. In order to obtain T from the ORS,
for example, for an f frequency centered within the seismic range and for two given
damping values, we can consider the ratio:

R  f,İ,İ‘  Spv  f, İ‘  / Spv  f, İ  [8.48]

This relationship characterizes the duration of the stimulation. Actually, if we

consider the case when the duration of the transient phase of the response of the f
frequency and the H damping frequency oscillator where H‘ is small with regard to T,
for instance, R is given by:

R  f,İ,İ‘   İ/İ‘ 1/2 ȝ  N, į‘ o / ȝ  N,į  [8.49]

From Figure 8.7 abacuses, the latter relationship allows us to obtain N and
therefore T. Then relation [8.47] gives S(f).

8.6.3. Determination of the floor spectra

As we now know the characteristics of the source separable process, it is

possible to determine those of the structure response. To do so, all we have to do is
apply equation [8.13] which gives the PSD of the response at point r and time W
300 Dynamic Behavior of Concrete and Seismic Engineering

after the beginning of the stimulation, in which the expression of the H transfer
function is slightly modified to take the envelope of the source process into account:

2
SX  r,IJ,f  H  r,IJ,f  S  f  [8.50]

with:

W
H  r,IJ,f  ³ G  r,u  a  IJ  u  exp  2iʌfu  du [8.51]
0

Figure 8.10. Floor spectra: comparison between the proposed method (synthetic spectra) and
direct time integration calculation (10 realizations)

The last step involves passing from SX (r, W, f) (r being the position of the floor
or the anchoring considered) to the associated ORS that will be the “floor spectra”
sought. To obtain this, we also go through a separable process representation.

On the one hand, we consider the average according to W of SX (r, W, f) that we

take as PSD: Sfloor (f), and on the other hand the integral according to f of Sx (r, W, f),
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 301

which represents the square of the standard deviation of the response VX (r, W)),
which is taken, after being standardized, as an envelope: afloor(W).

Finally, we can reconstruct the oscillator spectra [Spv] floor (f, H)] using equations
[8.47] and [8.49]. An illustration of the floor spectrum calculation for which the
carrying structure behaves as a 1-dof oscillator is presented in Figure 8.10.

8.7. Creation of synthetic signals and direct numerical integration

So far we have examined seismic analysis methods that do not require any direct
time integration of the movement equations. The main interest of such methods lies
in them being more “physical”, far lighter and far more flexible to use (depending
on the problem to deal with we adjust the number of modes used), and in them being
perfectly consistent with the regulation data of the seismic source.

Nevertheless, they cannot be applied in two cases:

– when we want to carry out an experimental analysis, in the case of the shaking
table assessment of equipment, the good behavior of which has to be guaranteed
under seismic loading defined by a given ORS, for instance. The jacks of the
shaking table then need a load time law;
– when the structure has to be represented by a highly non-linear model. Then
the modal theory cannot be applied.

We could also prefer the time method for some linear but complex structures, the
modal core truncation criteria of which are not really clear or involve a very high
number of modes. As we said at the beginning of this chapter, the seismic data
associated with a site is seldom of the time kind because of the absence of
measurement data, of the “envelope” feature that we mean to give it, etc.

Therefore, the problem is as follows: from an ORS set, how can we define a
random process, the ORS of which will represent the average maxima like that
defined above?

Such a problem is not well put. Is there one solution to it? Are there several? Do
these solutions have any relationship with the physical phenomena brought into
play?

As far as we know, these questions have not been satisfactorily answered yet.
302 Dynamic Behavior of Concrete and Seismic Engineering

In practice, various methods allow us to make random processes, the average

characteristics of which are adjusted on an ORS family within an eigenfrequency
and damping range, and thus to produce time creations that can be used for both
calculation and experimentation.

These methods even include the correction of the most obvious non-physical
characteristics (for instance, time averages not equal to zero for acceleration
signals).

The nature of the generated signals can be quite different from one method to the
next. It would not be that important for a linear behavior structure because in this
case the ORS represent relevant values as far as the response maximum prediction is
concerned. Besides, that is the reason why the adjustment of a separable process
involves satisfying results for the direct calculation of the linear floor spectra.

In the case of a marked non-linear behavior (modeled or real structure), such

relevance is not at all obvious. Thus, from response average maxima, we can obtain
quite different results according to the method used (with adjustment on the same
ORS).

As an illustration we will hereafter give the principles of the quite commonly

used “random phase harmonic sinusoid” method.

Generally, the adjustment is carried out on a given S (f, H) ORS (H being given).
The probabilistic model used is the separable type;

*t a t Ft [8.52]

The stationary process F(t) results from a sum of sinusoids with deterministic Ai
amplitudes and Mi random phases that are independent and equiprobable within the
[S, S] interval:

N
Ft ¦ A sin  2ʌ f t+ij 
i 1
i i i [8.53]

The fi values are a set of discrete frequencies that describe the studied range. We
often take:
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 303

fi i'f with 'f 1/ T [8.54]

T being the duration of the signal. According to the Shannon theorem, the studied
[0, fmax] frequency range is conditioned by the choice of the time discretization slot
1/2fmax = T/2N with fmax = N 'f.

The problem involves determining the Ai values and T from the ORS. For this
we can show that the PSD discretized (with the 'f slot) corresponding to the F(t)
process is given by:

S  i'f  Ai 2 T / 2 [8.55]

Thus, we are brought back to the problem in the previous section: the
relationships expressed in [8.47] and [8.49] allow us to obtain S (f) and T from the
ORS; equation [8.55] allows us to obtain the Ai values.

REMARK.– Regarding the time integration methods: the seismic analysis must
result in obtaining statistical characteristics concerning the absolute maxima
(averages and possibly standard deviation) of different physical values
(displacements, accelerations, stresses). Thus, time calculation represents the
realization of the corresponding random variables. In order to carry out their
statistics with reasonable accuracy, many calculations have to be done. Therefore the
time method is cumbersome and expensive, especially in the case of non-linear
behaviors. We lay great stress on the fact that the result of one only calculation,
however complex it may be, brings only little information. The potentially chaotic
behaviors of the non-linear systems we will deal with in the next section only
reinforce that assertion.

Let us notice however that if the calculation includes several quite isolated
maxima, it will be possible to take advantage of all these maxima, not only of the
most important of them, which will allow us to decrease the number of calculations
to carry out for any given accuracy.
304 Dynamic Behavior of Concrete and Seismic Engineering

8.8. Seismic analysis of non-linear behavior structures

8.8.1. Introduction

Generally speaking, the dynamic study of non-linear systems is a mathematics

topic that is quite extensive and complex with many remaining gray areas. Unlike
linear cases with which we have been occupied so far, there is no theoretical frame
strong enough to allow us to draw predictive methods engineers could use in an
acceptable way. It is not surprising, as the non-linear feature does not correspond to
the general case, whereas the linear feature corresponds to quite specific properties.

As a consequence, we will be able to propose valid predictive methods only if

the problem becomes particular, i.e. if we can precisely know the nature of the non-
linearities encountered, and those of the stimulation sources, as losing the linearity
property forbids characterization of the structure as a “responsive system”
independently of the source.

Let us note that the type of result we are interested in, regarding the response
(here essentially the maxima), also plays an important part.

In this section we are not dealing with developing theories about non-linear
systems but, after clarifying the main types of non-linearities commonly used in
practice, with illustrating the associated dynamic (seismic) behaviors and describing
the simplified methods used by engineers for direct prediction or interpretation and
thus for a reasonable use of time digital simulations.

8.8.2. Main non-linearities of seismically-loaded structures

Non-linear behaviors can be generated by:

– geometry due to the important displacements of the structure (bridge stay-
cables or high-voltage cables, for instance) or due to limit conditions (the impact of
two neighboring structures like pipework with its supports, for example, or sliding at
the level of anti-seismic bearing devices); or
– behavior laws of materials (for example, the plastic behavior of steel or of
reinforced or pre-stressed concrete extensively illustrated in this book).

In dynamics, it is important to distinguish non-linearities leading to great energy

dissipation (like plastic or visco-plastic behaviors, or friction) from those implying
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 305

only a little dissipation, like great movement effects or moderate shocks that can be
deemed elastic.

It is also necessary to distinguish the weak non-linearities which often appear at

the level of imperfect linking and which do not question the modal structure of the
system from the strong non-linearities that radically modify the mode structure and
the behavior.

The former are taken into account through a slight modification of the resonance
frequencies and mostly through the introduction of a relatively small damping term
compared with the stiffness and inertia terms. In most such cases we give up an
actual physical modeling to merely globally adjust a viscous dampening coefficient
Hn for each mode, which corresponds to linear modeling.

The latter can only be dealt with using specific non-linear models. In current
practice, it does not prevent engineers from deriving equivalent linear models from
these models, within restricted parametric ranges.

This last aspect will be the focus of our attention in the rest of the chapter.

We will not systematically study the different non-linear models used in seismic
analysis, but we will rely on two characteristic examples:
– the elasto-plastic behavior model, which is the most often used, or the
“according to Coulomb’s law adherence-sliding-type” behavior, which has the same
nature;
– the “elastic shock-type” behavior model.

8.8.3. Notion of “inelastic spectra”

This notion is the most used method in seismic analysis practice for taking the
elasto-plastic behavior of the structure into account.

Consider a perfect elasto-plastic m-mass 1-dof oscillator loaded by an J(t)

imposed acceleration of its support. It is characterized by its resonance pulsation Z
in its elastic phase, its limit elastic displacement xe, its limit force mZ2xe and its
ductility coefficient P (Figure 8.11, where D = 0).

Suppose that J(t) consists of two opposed I intensity impulses in the shape of 'W
time slots and separated by a time period W, with 'W << W (Figure 8.12).
306 Dynamic Behavior of Concrete and Seismic Engineering

Figure 8.11. Diagram of a bi-linear-strain hardening elasto-plastic oscillator

The idea involves comparing the limit intensities of the impulses bringing about
collapse that can be calculated thanks to:
– elastic analysis: we will write that the displacement remains lower than xe;
– elasto-plastic analysis: we will write that the displacement remains lower than
Pxe.

We can distinguish three frequency ranges:

(i) If ZW << 1, it is the low frequency range of the ORS. The maximum relative
displacement of the mass depends little on the stiffness law and it is given
approximately by xmax = IW. A purely elastic analysis gives a limit intensity Ile = xe/W.
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 307

An elasto-plastic analysis leads to Il = P xe/W. Therefore, taking plasticity into

account involves admitting a higher limit intensity of a factor P.
(ii) If ZW | 1, it is the medium frequency range of the ORS. We can suppose that
the maximum displacement of the mass occurs before the second impulse takes
place. The elastic analysis will give Ile = xe Z.

J(t)

1/'t

'W

'W

–1/'t W

Figure 8.12. 2-impulse stimulation model

When writing the preservation of energy, the elasto-plastic analysis will give:

I12  Ȧ x e 2  2Ȧ2 x e  ȝ x e  x e  [8.56]

As a consequence, taking plasticity into account involves admitting a higher

limit intensity of a factor (2P  1)1/2.
(iii) If Z'W d 1, it is the high frequency range of the ORS. The maximum
displacement of the mass is conditioned by the application of the first acceleration
step. The elastic and elasto-plastic analyses will give the same results as far as the
limit intensity is concerned.
308 Dynamic Behavior of Concrete and Seismic Engineering

These results are extrapolated to the case of an average seismic stimulation:

 Constituting the “inelastic spectrum” involves applying the reduction factors
defined above within the three frequency ranges to the original ORS. Then we will
use the new spectrum to apply the modal method the way we would do with a linear
behavior structure.
 In the para-seismic Eurocode, we use a more continuous reduction factor
depending on the frequency but based on the same principles (Figure 8.13).

Figure 8.13. Inelastic reduction coefficient for a 4-ductility factor

dispersion effect between successive signals

Applying the technique raises some problems:

– the most serious one lies in the application of a system that cannot be reduced
to 1 dof. Actually it is difficult to define “basic solutions” (the equivalent of linear
natural modes); moreover, the result we are trying to achieve generally cannot be
represented in the shape of a sum of the contributions of these basic solutions.
Applying the modal method starting from those hypotheses can bring about very
serious errors;
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 309

– on the other hand, the behavior law is not generally the perfect elasto-plastic
type (for example the strain-damping model in Figure 8.11);
– the nature of the source plays a part. As a matter of fact the model is based on
the assumption of a seismic signal consisting of rare and isolated impulses that have
the oscillator enter the plastic field. This is more or less achieved with wideband
signals, yet it is completely wrong for narrow-band signals (floor motions).

This latter point can be illustrated by taking different seismic motions and by
numerically calculating the reduction factors. Figure 8.13 shows their evolution
according to Z impulses (compared with the average impulse of the seismic signal)
for a given P value.

REMARKS:
1) Besides the factors discussed above, the inelastic spectrum technique assumes
that the highest impulse will cause collapse (notion of absolute maximum contained
in the ORS). This is not always the case. In fact, the Pxe limit value can be reached
by combining several intrusions into the plastic domain. Vanmarcke has proposed a
calculation method (see [VAN 73]) that applies whenever such intrusions are rare
and isolated. It is based on the following ideas:
– the plastic threshold overstepping statistics are little modified by the non-linear
behavior. The considerations developed in the summaries about maximum statistics
can therefore apply (average overstepping number, Poisson model etc.);
– the plastic displacement during an excursion is calculated according to the
same hypotheses as for the inelastic spectra (ZW | 1 case).
310 Dynamic Behavior of Concrete and Seismic Engineering

Figure 8.14. Diagram of a sliding oscillator

Thus, we can obtain the average value of the residual plastic displacement, the
probability for the cumulative plastic displacement to exceed a specific value, etc.

2) Structures presenting a series of adherence and sliding phenomena that can be

represented owing to a Coulomb’s law belong to the above category. Thus, the 1-dof
sliding oscillator represented in Figure 8.14 obeys the same behavior law as the
perfect elasto-plastic oscillator when the adherence force equals the friction force.
The sliding displacement plays the same part as the previous plastic displacement.
We can mention studies on the sliding problem carried out with a technique similar
to the reference inelastic spectrum technique (see [NED 99], [NOE 93] and [SAR
98]).

8.8.4. Conventional method of stochastic linearization

The flaws of the inelastic spectrum method result from the fact that the
probabilistic nature of the seismic source is unknown. Thus, a solution may consist
of coming back to the separable probabilistic model defined in the previous sections
and trying to statistically analyze the response of the non-linear system.

The stochastic linearization method tries to solve the problem in a quite

comprehensive way. We are going to present its principles while shedding light on
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 311

its limitations. In order to simplify, we will use a 1-dof system for the presentation,
as extending the method to N-dof systems does not raise any specific problem.

Given a non-linear oscillator with a motion equation:

d 2 x / dt 2  h  x, dx/dt  Ȗ  t  [8.57]

J(t) being stationary, we are interested in the proven solution for [8.57], for example,
we wish to replace this oscillator with an equivalent linear oscillator of an equation:

d 2 y/dt 2  2Ȧeq İ eq dy/dt  Ȧeq 2 y Ȗ  t  [8.58]

The y(t) displacement, the solution to [8.58], does not verify [8.57], but we want
to adjust Zeq and Heq so that it is as close as possible to its solution x(t). The
adjustment is done by minimizing, in the mean squares sense, the difference
between the non-linear term and the equivalent linear term both applied to the exact
solution of the problem (equation [8.57]). The Zeq and Heq thus obtained are such that
the variance of the response calculated with [8.58] corresponds quite well to the
solution value of [8.57].

Let us call *(t), X(t) and Y(t) the random processes associated with J(t), x(t) and
y(t).

Unfortunately, the statistic characteristics of the X(t) “non-linear response”

process are not known a priori. In practice, we will therefore try to minimize the e(t)
gap between the non-linear term and its linear equivalent applied to the Y(t)
“equivalent linear response” process. This approximation explains the low efficiency
of the method in some cases:

et 
h  Y,dY/dt   2Ȧeq İ eq dY/dt+Ȧeq 2 Y  [8.59]

Zeq and Heq are constants given by the equations:

2 2
¬ ¼ 
w E ª«e  t  º» / w Ȧeq 2  ¬ ¼ 
w E ª«e  t  º» / w İ eq Ȧeq  0 [8.60]
312 Dynamic Behavior of Concrete and Seismic Engineering

For a Gaussian stimulation (the Gaussian linearization method), the E[e(t)2]

variance can be expressed according to the co-variance coefficients of Y(t) and
Y/dt(t) which themselves are functions of Zeq and Heq. The [8.60] system consists of
2 equations in 2 unknowns. Its solution will give the characteristics of the equivalent
linear oscillator.

We can also show that:

Ȧeq 2 E ª¬wh  Y, dY/dt  / wY º¼ [8.61]

2Ȧeq İ eq E ª¬wh  Y, dY/dt  / w  dY/dt  º¼ [8.62]

The equivalent linear stiffness and damping represent the average statistics of the
“tangent” stiffness and damping values associated with the different values of the
response of the equivalent linear oscillator.

More generally, the method can apply in the non-stationary case of stimulation
owing to a separable process. We are going to discuss the equivalent linearization
method in two simple examples.

8.8.4.1. Application to an elasto-plastic oscillator

The formalism described above does not apply directly to elasto-plastic or
adherence-sliding behavior types. As a matter of fact, conventional models (perfect
elasto-plastic models with bilinear strain hardening) imply dependence on the
history of movement that cannot be represented by a mere h (x, dx/dt). Thus, we
have to resort to either simplified models representing the plastic cycles according to
the movement maximum amplitude or to more sophisticated models that introduce
an additional subsidiary variable confirming a differential equation (Wen-Bouc’s
model [BOU 94], for example). Figure 8.15 shows the aspect of cycles obtained
with Wen-Bouc’s model for different values of its adjustable parameters.
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 313

Figure 8.15. Different cycles shapes obtained owing to Wen-Bouc’s model

Figure 8.16. Adjustment attempt of a linear oscillator equivalent to a

bilinear strain-hardening elasto-plastic oscillator

Such models give quite good results as far as the prediction of the variances and
PSDs of the response, for linearities that are either weak or with strain-hardening
slope/elastic slope ratios higher than 0.5 or so.
314 Dynamic Behavior of Concrete and Seismic Engineering

For strong non-linearities with slope ratios lower than 0.5, the non-linear
calculation reveals the existence of two phenomena:
– a phenomenon at the elasto-plastic cycle frequency;
– a low-frequency phenomenon that corresponds to the cumulative effects of
plastic displacements regarding the inelastic spectrum method referred to above.

A 1-dof linear oscillator cannot represent this “two-frequency” behavior, thus, in

that case we can observe discrepancies as far as the variance estimate is concerned,
as well as poor PSD results (Figure 8.16).

8.8.4.2. Application to a 1-dof elastic shock oscillator

Figure 8.17 shows the diagram of a 1-dof-shock oscillator and the aspect of the
behavior law. Kc is the stop stiffness, which is typically high relative to that of the
oscillator. This law is the f(x) type and it fits quite well with the stochastic
linearization technique.

Figure 8.17. Diagram and law of a 1-dof-shock oscillator

However, applying this technique here gives very disappointing results:

– The graphs in Figure 8.18 extracted from [GUI 90] derive from non-linear
digital simulations. They show the physical behavior of the shock oscillator. They
represent the mass displacement PSDs for different stationary wide-band noise type
stimulation levels (the K no-dimension associated parameter being the ratio between
the play and the standard deviation of the response of the without shock oscillator).
For high K values (very important play and therefore no shocks or few shocks) or
small K values (play nearly equal to zero, hence an almost perfect connection with
the stop), we have the behavior of a weakly dampened linear oscillator (we took H =
0.01). On the other hand, when K is in the order of 1, the PSDs are quite flat. This
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 315

expresses a frequency dispersion effect but which does not correspond to a damping
effect.
– Figure 8.19 illustrates the stochastic linearization method. Note that it is
incapable of representing the frequency element.

Figure 8.18. Shock oscillator: evolution of the shape of the PSD

of the answer as a function of reduced play

8.8.4.2.1. Conclusion
The principle of conventional stochastic linearization expressed for systems with
several degrees of freedom can seem appealing owing to its general feature.
Nevertheless, as we have just illustrated, its application is often problematic. Things
go off smoothly when the effect of non-linearities is weak, whereas very non-linear
behaviors are poorly modeled, especially those that correspond to energy dispersion
phenomena within one or several frequency bands. We are then led to particularize
the method to the non-linearity type studied (which, from our point of view, is
inevitable whatever the method used) and to use such restrictive hypotheses that,
when compared with such appropriate simplified methods as the inelastic spectrum,
their utility is often quite questionable.
316 Dynamic Behavior of Concrete and Seismic Engineering

Figure 8.19. Failure of the adjustment of an equivalent linear oscillator with the
conventional method (comparison of the response PSDs)

8.8.5. Random parameter stochastic linearization

To overcome the drawbacks we have just mentioned, several strategies come to

mind:
– since it is necessary to enrich the frequency representation, we could associate
a several-dof system with an initially 1-dof system, for instance. Thus, two
equivalent linear oscillators could be associated with the oscillator in the previous
sections, one of which with a very low resonance frequency. Obviously, such an
approach does not simplify matters. Besides, it is utterly ineffective with a shock
oscillator (see Figure 8.20 and [GUI 90]);
– another idea involves keeping the initial number of dofs whilst giving the
parameters of the equivalent linear oscillators a random feature. Such an idea is the
subject of rather general developments ([FOG 96] and [BEL 99]) that we will not
present here. We will merely illustrate the method using the 1-dof-shock oscillator
as an example and underline some difficulties that arise when the number of degrees
of freedom is increased.
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 317

Figure 8.20. Adjustment attempt of a 15-equivalent oscillator set

to represent the frequency stretching due to shocks

8.8.5.1. Modeling of a 1-dof-shock oscillator owing to a random parameter linear

oscillator
Let us take the case of an oscillator stimulated by a *(t) wide-band noise (quasi-
random noise) and study its stationary response.

The general principle results from the observation of the response signal
obtained by numerical simulation. The latter consists of amplitude and frequency
oscillation sequences randomly distributed (Figure 8.21).

If the maxima (most of the time corresponding to contact with the stops) are
transferred into an amplitude/frequency diagram, the points are organized along a
curve that characterizes a deterministic relationship between the frequency and the
amplitude (Figure 8.22).
318 Dynamic Behavior of Concrete and Seismic Engineering

Figure 8.21. Example of a response signal of a random noise shock oscillator

Figure 8.22. Amplitude-frequency relationship of oscillation sequences (the full line has
been obtained analytically, the dot cloud results from digital simulations)
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 319

In the present case of a shock oscillator, the relationship can be found

analytically by considering the periodic regime set for the free oscillator (without
any stimulation) instead of the dampened oscillator, which is illustrated in Figure
8.23.

Figure 8.23. Periodic regime/ratio for the 1-dof, free and non-dampened shock oscillator

The idea involves writing that these sequences correspond to the response of a
Z0-resonance pulsation and H-damping coefficient linear oscillator, to random noise,
the intensity of which is different from the initial random noise by an S(Z0) factor:

d 2 x/dt 2  2İ Ȧ0 dx/dt  Ȧ0 2 x S  Ȧ0  *  t  [8.63]

the pulsation Z0 being random (H can be considered as constant and thus as

deterministic).

As far as the random process is concerned, we will be able to represent the X(t)
response process using two slow time variation processes: the Xmax(t) amplitude and
the \(t) phase shift which characterize the sequences, and a fast time variation
process that represents the 4(t) phase of the oscillation within a sequence. The :(t)
320 Dynamic Behavior of Concrete and Seismic Engineering

pulsation process is linked to the Xmax(t) process in a deterministic way using the Z0
function:

:t Ȧ0  X max  t   [8.64]

Xt X max  t  cos  4  t   <  t   [8.65]

dX/dt  t   X max  t  Ȧ0  X max  sin  4  t   <  t   [8.66]

d4 / dt  t  Ȧ0  X max  [8.67]

As a consequence, after having specified the deterministic relationships Z0(xmax)

and S(Z0) (and therefore S(xmax)), the problem will consist of characterizing the
random processes Xmax(t) and \(t) in terms of probability density. Then it is easy to
obtain the PSD of X(t).

It is possible to show that from a theoretical point of view (see [FOG 96] and
[BEL 99]), each of the probability densities confirms a Fokker-Planck differential
stochastic equation that can be solved.

If, in our example, the equation gives an analytical solution, it is not the same
with more complex cases. Thus, for several-dof systems, the differential problem is
large, and its solution is far beyond the scope of engineering calculations. Therefore,
from a practical point of view, in order to evaluate the probability density rate we
will merely use simple physical approximations that will reproduce the frequency
dispersion phenomenon.

In short, the different steps of the method are:

– determination of the amplitude/pulsation relationship. In our example, we will
take the relationship corresponding to the full-line curve in Figure 8.22;
– determination of the amplitude/source intensity relationship. In our example,
we adjust S(xmax), to preserve the average intensity of the oscillator (actually we can
simply explain the average energy of the non-linear oscillator as well as that of the
equivalent linear oscillator);
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 321

– estimation of the probability density of the amplitude and the phase shift. In
our example, we can explain the joint probability density q (x, v) of the W(t) and
dX/dt(t) processes by applying formula [8.23] with the expression of the potential
corresponding to the behavior law in Figure 8.17:

Ux
2 1 2 1 2 ¬ 
1 Ȧ 2 x 2  1 Ȧ 2  Ȧ 2 ª x  e  H  x  e    x  e  H   x  e  º
¼

with Z12 = k/m and Z22 = (k + K)/m.

From q (x,v) we can derive the p(xmax) probability density of the Xmax(t)
envelope process;
– obtaining the PSD of the response.

We can eventually express for the general case the PSD of the response X(t) of
the oscillator by;

2
Sx Y  ³ H  Ȧ, Ȧ0  x max  , İ  S  x max  p  x max  dx max [8.68]
0

which corresponds to the statistic average according to xmax of relation [8.15].

In our example H represents the transfer function of the Z0(xmax) resonance

pulsation and H the damping 1-dof linear oscillator:

H  Ȧ, Ȧ0 , İ  
1 / Ȧ2  Ȧ0 2  2iİȦȦ0  [8.69]

Figure 8.24 shows the good harmony between the PSDs obtained by numerical
simulation and those obtained using the equivalent linearization method.
322 Dynamic Behavior of Concrete and Seismic Engineering

Figure 8.24. Comparison between the random coefficient linearization method

and numerical simulations (PSD of the response)

8.8.5.2. Comments on several-dof systems

The previous formalism can be extended to a several-dof system [BEL 99].
However, concerning the obtaining of Zn(xmax) relationships, a delicate problem is
raised, as n ranges from 1 to N which is the number of dofs. These relationships can
be obtained from periodic regimes associated with the free and non-dampened
system (see Figure 8.23 for 1 dof).

As a matter of fact, if for 1 dof, there is often one only such stable regime, things
become far more complex as soon as we reach 2 dofs. Then we can observe
numerous stable or unstable periodic regimes, chaotic regimes, etc. These different
regimes correspond to paths concerning different areas of the phase space that refers
to the space of the y vectors defined above [CAU 63]. The intensity and the space
distribution of the source will favor a particular regime.

Actually the problem is quite complex and it is not certain that the methods
currently proposed within the scope of the formalism of the random coefficient
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 323

equivalent linearization take these aspects into account. From a practical point of
view we should be cautious when applying these methods.

Nevertheless, we can be quite optimistic in the case of seismic analysis that

concerns us here, as the transient feature of the stimulation inevitably restricts the
excursions of the system into the phase space, which makes the dynamic behavior
much simpler.

8.9. Conclusion

Probabilistic seismic analysis lies within the wider scope of random dynamics,
which aims at developing tools to predict responses to complex loading sources that
can only be characterized by statistical averages.

The specificity of the seismic source derives from its complex and transient
characteristics. Moreover, it is difficult to model because of the complexity of the
earthquake generating mechanisms, the media through which waves propagate and
the fact we lack “experimental” data.

Another aspect of seismic analysis is that it falls within the scope of quite
heavily regulated designing methods.

In this chapter, we tried to explain where probabilistic seismic analysis stands

with regard to conventional design methods. It specifies the scope of their validity
by supplying designers with a means to assess their hypotheses, it proposes
improvements and thus allows the extension of their application field; it can even
stand in for them in some of the more delicate cases that require mastery of safety
margins.

As we have shown, probabilistic approaches are constantly evolving, especially

as far as non-linear behavior is concerned.

8.10. Bibliography

[BEL 99] BELIZZI S., BOUC R., “Analysis of Multi-degree of Freedom Strongly Non-
Linear Mechanical Systems with Random Input”, Parts I and II, Probabilistic Engineering
Mechanics, vol. 12, n° 3, p. 229-244 and 245-256, 1999.
[BOU 94] BOUC R., “The Power Spectral Density of Response for a Strongly Non-linear
Random Oscillator”, Journal of Sound and Vibration, vol. 175, n° 3, p. 317-331, 1994.
324 Dynamic Behavior of Concrete and Seismic Engineering

[DAV 85] DAVIDOVICI V., Génie parasismique, Presses de l’ENPC, 1985.

[DAV 88] DAVIDOVICI V., Recent Advances in Earthquake Engineering and Structural
Dynamics, Ouest Editions-Presses Académiques, 1992.
[FOG 96] FOGLI M., BRESSOLETTE P., BERNARD P., “Dynamic of the Stochastic
Oscillator with Impacts”, European Journal of Mechanics, A/Solids, vol. 15, n° 2, p. 213-
241, 1994.
[GIB 88] GIBERT R.J., Vibrations des Structures, Interactions avec les fluides, Sources
d’excitation aléatoires, Editions Eyrolles, 1988.
[GUI 90] GUILHOT P., Analyse de la Réponse de Structures Non Linéaires Sollicitées par
des Sources d’Excitation Aléatoires – Application au Comportement des Lignes de
Tuyauterie sous l’Effet d’un Séisme, PhD Thesis, Pierre et Marie Curie University,
December 1990.
[KRE 83] KREE P., SOIZE C., Mécanique Aléatoire, Dunod, Paris, 1983.
[NED 99] NEDJAI H., GIBERT J.R., “Seismic Analysis of a Structure with Sliding Bearing”,
PVP Conf., Boston, USA, 1999.
[NOE 93] NOE H., BETBEDER-MATIBET J., GOUTELARD H., “Les Modes Propres en
Base Glissante. Théorie et Applications”, 3e Colloque National de Génie Parasismique,
Saint-Rémy-lès-Chevreuse, 1993.
[PRE 90] PREUMONT A., Vibrations Aléatoires et Analyse Spectrale, Presses
Polytechniques et Universitaires Romandes, Switzerland, 1990.
[SAR 98] SARH K., Réponse d’une Structure Multimodale Glissante sous Excitation
Aléatoire, PhD Thesis, Pierre and Marie Curie University, 1998.
[SCH 01] SCHEIMAN P., Analyse de la Réponse de Structures Non Linéaires à deux Degrés
de Liberté Sollicitées par des Sources d’Excitation Aléatoires, PhD Thesis, University of
Evry-Val-d’Essonne, 2001.
[VAN 73] VANMARCKE E.H., VENEZIANO D., “Probabilistic Seismic Response of
Simple Inelastic Systems”, 5e Conf. Mondiale d’Ingénierie Sismique, Rome, Italy, 1973.

Additional references

[BER 84] BERNARD P., FOGLI M., BRESSOLETTE P., LEMAIRE M., “Un Algorithme de
Simulation Stochastique par Markovianisation Approchée. Application à la Mécanique
Aléatoire”, Journal de Mécanique Théorique et Appliquée, vol. 3, n° 6, p. 905-950, 1984.
[CAU 63] CAUGHEY T.K., “Equivalent Linearization Technics”, Journal of the Acoustical
Society of America, p. 1706-1711, vol. 35, n° 11, November 1963.
[CAU 71] CAUGHEY T.K., “Non-linear Theory of Random Vibrations”, Adv. Appl. Mech.,
vol. 11, p. 209-253, 1971.
 Seismic Analysis of Structures: Improvements Due to Probabilistic Concepts 325

[HAD 82] HADJIAN A.H., “A Re-evaluation of Equivalent Linear Models for Simple
Yielding Systems”, Earthquake Engineering and Structural Dynamics, vol. 10, p. 759-
767, 1982.
[IWA 73] IWAN W.D., “A Generalization of the Concept of Equivalent Linearization”,
International Journal of Non-Linear Mechanics, vol. 8, p. 279-287, 1973.
[KAN 57] KANAI K., “Semi-empirical Formula for the Seismic Characteristics of the
Ground”, Univ. Tokyo, Bull. Earthquake Res. Inst., vol. 35, p. 309-325, 1957.
[LAB 86] LABBE P., “Evolution des Méthodes Spectrales en Analyse Sismique. Un Apport
de la démarche Probabiliste”, 1er Colloque National de Génie Parasismique, Saint-Rémy-
lès-Chevreuse, January 1986.
[ROB 80] ROBERTS J.B., “The Yielding Behaviour of a Randomly Excited Elasto-plastic
Structure”, Journal of Sound and Vibration, vol. 72, n° 1, p. 71-85, 1980.
[ROB 93] ROBERTS J.B., SPANOS P-T.D., Random Vibration and Statistical Linearization,
John Wiley & Sons, 1993.
[SOI 88] SOIZE C., “Steady-state Solution of Fokker-Planck Equation in Higher Dimension”,
Probabilistic Engineering Mechanics, vol. 3, n° 4, p. 196-206, 1988.
[SPA 80] SPANOS P-T.D., “Formulation of Stochastic Linearization for Symmetric or
Asymmetric MDOF Non-Linear Systems”, Journal of Applied Mechanics, vol. 47,
p. 209-211, 1980.
[SPA 81] SPANOS P-T.D., “A Method for Analysis of Non Linear Vibrations Caused by
Modulated Random Excitation”, International Journal of Non-Linear Mechanics, vol. 16
p. 1-11, 1981.
[WEN 89] WEN Y.K., “Methods of Random Vibration for Inelastic Structures”, Journal of
Applied Mechanics, vol. 42, n° 2, p. 39-52, 1989.
This page intentionally left blank
 Chapter 9

Engineering Know-How:
Lessons from Earthquakes and
Rules for Seismic Design

9.1. Introduction

This chapter aims to show what can be learnt from the study of earthquakes,
describes the main failure modes of reinforced concrete and discusses how current
designs rely on these studies, as well as theoretical models, to improve structural
strength and ability to withstand earthquakes..

Some items in the current PS92 French regulations [COL 95a] will be
approached, but all of the topics discussed are framed within the standards set by the
Eurocodes that are destined to replace them.

9.2. Lessons from earthquakes

The sudden application of seismic action to buildings amplifies their design and
building faults in a typical manner (Figure 9.1): poor architectural layouts, under-
dimensioning and brittle failure modes all become evident during structural damage
or collapse.

Chapter written by Philippe BISCH.

328 Dynamic Behavior of Concrete and Seismic Engineering

Photo Philippe Bisch

Figure 9.1. Poor-quality concrete in a column

9.2.1. Pathologies linked to overall behavior

The seismic loads that cause the most damage are horizontal inertia loads the
structure has not been designed to withstand: permanent loads do less damage
because they are directed downwards. Horizontal loads have to be distributed
between transverse bracings, vertical structures designed to transfer inertial loads
down to the foundations. As far as the overall behavior of any structure is
concerned, two main problems are typically identified:
– the layout of the transverse bracing structures is such that the center of stiffness
is offset from the inertial load resultant, hence the structure is subject to a vertical
axis torsion that tends to overload bracing located on the edges of the building.
Figure 9.2 shows a building block after torsion around a stiff portion, namely the
stairwell. The right hand part, supported by a reinforced concrete column, has
collapsed due to a general rotation imposed on the vertical axis;
– bracing structures do not have any continuous stiffness and geometry along the
height of the building; consequently, at levels where discontinuities occur, seismic
loads have to be transferred to a “diaphragm” horizontal plane. Moreover, since
stiffness can differ significantly between two successive levels, strains are
concentrated in the less rigid storeys. This is evident in Figure 9.3, which shows a
rather stiff villa built on piles that had its lower level destroyed by displacements
imposed by the overall motion.
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 329

Photo Philippe Bisch

Figure 9.2. Collapse of a building showing overall torsion

Photo Philippe Bisch

Figure 9.3. Rupture at the level of a lightened storey

A building’s overall dynamic behavior is determined by the strength and

ductility of its structural elements. In fact, except for specific constructions for
which preserving a quasi-elastic behavior is desirable, large incursions into the post-
elastic field are usually included in the design (see section 9.5.2). In linear element
structures such as beams and columns, this involves formation of “plastic hinges”
that will suffer alternating rotation cycles during a seismic disturbance. These induce
steel and concrete strains far higher than those at the ultimate limit state. Roughly
speaking, with such flexible structures, we can assume that the displacements
reached during incursions into the plastic field are equal to those calculated using
the elastic hypothesis. To a first approximation, the structures are subjected to
impose displacements, which allow engineers to carry out sound design, by
evaluating the rotation likely to be caused in the plastic hinges.

To avoid strains causing compression and crushing, areas that are joined by the
plastic hinges should be confined by transverse reinforcements (stirrups). The final
strength of sections is determined by the extent to which the energy dissipation
330 Dynamic Behavior of Concrete and Seismic Engineering

inside the hinges counterbalances resistant moments, and whether sufficient ductile
rotation can take place. Figure 9.3 shows a case in which the piles were able to
absorb the imposed displacements without rupture.

Photo Philippe Bisch

Figure 9.4. Rupture of a column due to the shock between two neighboring structures

Two dynamically independent blocks can shock each other if the hinge
separating them is not open enough. Figure 9.4 shows local bending due to a shock
inside a column. The phenomenon can lead to total building collapse.

9.2.2. Problems linked to local under-design

Figure 9.5 shows a section at the head of a column that has been broken because
of the horizontal load transmitted to it. The section has several features that
contribute to this poor performance: smooth construction joints with little strength in
the concrete to withstand the shear load, no concrete confining reinforcements and
insufficient longitudinal reinforcements. No rotation ductility could be induced
because of these deficiencies, and the reinforcements broke in traction in the joints,
which opened because the rotation imposed was localized.

Figure 9.5. Rupture of a section due to insufficient longitudinal reinforcements

 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 331

Photo Philippe Bisch

Figure 9.6. Rupture of a column due to buckling

In Figure 9.6, failure has occurred because of the instability of the column:
combined bending and extra normal forces due to the vertical component of an
earthquake have crushed the concrete.

Figure 9.7 shows a silo after being subjected to a high horizontal acceleration.
Plastic hinges appeared at both ends of the columns, where the bending moments
were highest. In this case, the behavior of both hinges is quite different, due to
reinforcement differences. In the lower area, regular and tightly spaced frames have
confined the concrete within a “cage”, allowing strong plastic rotation (obvious in
the photograph) and holding the longitudinal reinforcements together, preserving the
limit moment during cycles. It should also be noted that the frames were well-
anchored, and consequently did not get torn down by the increased confinement
pressure.

In the upper part of this plate, the existence of smooth construction joints,
together with insufficient longitudinal reinforcements joining the section, generated
failure even before the plastic hinge had been able to form.
332 Dynamic Behavior of Concrete and Seismic Engineering

Photo Philippe Bisch

Figure 9.7. Rupture of a silo column with formation of a plastic hinge

Photo Philippe Bisch

Figure 9.8. Rupture of a column showing the influence of the shear load

As shown in Figure 9.8, bending failure is influenced by shear loads. As it takes

place, the column suffers the influence of insufficient transverse and longitudinal
reinforcements, and failure combines both effects.

In the case depicted in Figure 9.9, longitudinal reinforcements have allowed the
embedding section to withstand the moment quite well, but failure due to the shear
load started before a plastic hinge could form, because of insufficient transverse
reinforcements. This case highlights the importance of the sequence in which
plastification and failures occur – if the shear load-generated brittle failure had been
delayed beyond the onset of bending plastification of the embedding section, a
plastic hinge could have formed, which would have increased the chances the
column could have withstood the earthquake (see section 9.3.4).
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 333

Photo Philippe Bisch

Figure 9.9. Failure of a column due to shear load

Photo Philippe Bisch

Figure 9.10. Failure of a short column

The influence of shear load can be especially prominent in short columns, as

shown in Figure 9.10. In such a situation (for example, in ventilation spaces), an
imposed horizontal strain results in a prominent shear load, because the bending
length is limited and the column is stiff.

A similar phenomenon may also take place in short piers cut into a wall by
openings (Figure 9.11).

Figure 9.12 shows the deficiency of a column section with regard to

compression. The longitudinal reinforcements have suffered instability caused by
compression. They are not held firm by the concrete cover, which is ejected because
the stirrups are too widely spaced. The buckling force within the reinforcements is
334 Dynamic Behavior of Concrete and Seismic Engineering

proportional to the inverse of its buckling length squared, therefore stirrup spacing is
the main parameter controlling the phenomenon (see section 9.6.5.2).

Photo Philippe Bisch

Figure 9.11. Rupture of a column due to shear load

Photo Philippe Bisch

Ph °12
Figure 9.12. Rupture of a column due to buckling of longitudinal reinforcement

9.2.3. Problems linked to construction layout

Figure 9.13 shows a failure diagram for one of the reinforced concrete walls used
in the CAMUS 1 experiment. In this non-ductile wall, the reinforcements are laid
out so that cracking and deformation are distributed throughout the wall height,
unlike the situation in ductile walls, where the strain is mainly due to the rotation of
a plastic hinge at the base (see section 9.6.4.1).

In this case, longitudinal reinforcement failure occurs within an inclined crack

located on the second level and is initiated within the bar stopper section. Actually,
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 335

because of the influence of the shear load, this level tends to behave as trusswork
(vertical braces and bracing struts, floors being the stays). The bar stopper causes a
sudden variation of the strong vertical stress, thereby initiating failure.

Figure 9.14 shows the cover concrete bursting in a column section with a bar
stopper and a tie bar placed in a plane parallel to the outer side. Such a layout is
forbidden in seismic areas, where tie bars must be laid out inwards inside the
concrete mass to avoid such bursting. A similar cover concrete bursting
phenomenon can be observed in cases where coatings have insufficient seams.

Figure 9.13. Failure scheme for a reinforced concrete wall

Photo Philippe Bisch

Figure 9.14. Rupture of the cover in a column due to a reinforcement tie bar
336 Dynamic Behavior of Concrete and Seismic Engineering

9.3. The aims of anti-seismic protection standards

9.3.1. Standardization of anti-seismic design

As far as construction is concerned, standardization codifies design and

calculation methods. When they are correctly applied, they help designers produce
constructions that ensure the safety of people and goods for their entire planned
lifespan. In the building field, design standards serve to supervise the calculation
methods used, whilst other regulations govern the strength and strain limits of
building materials and construction layouts. In Europe, prevailing standards rely on
a general philosophy of safety that is based on a semi-probabilistic approach using
safety partial coefficients.

In France, anti-seismic protection rules were only developed after the

Orleansville earthquake in Algeria (1954). The first standardizing text (AS 55) was,
however, used only in Algeria. The earthquake in Agadir, Morocco (1960) a few
years later gave birth to the PS 69 rules that remained as guiding principles for over
20 years. After the El Asnam earthquake in Algeria (1980), it became clear that the
PS 69 rules did not ensure sufficient safety in some situations, and a review was
instigated. As such, a 1982 addendum contained some missing construction layouts
to the PS 69 rules. New regulations were requested to compensate for deficiencies in
PS 69, and these incorporated more recent research discoveries. This led to the PS
92 standard [COL 95a], which is currently in use.

The PS 92 standard is devoted to the anti-seismic design of commonly used

buildings, according to the BAEL standard. The materials addressed are reinforced
or pre-stressed concrete, steel structure frames and timbers. The text includes all
general measures concerning the targets to achieve, the seismic actions to take into
account, the calculation methods to be used, and strict definitions of safe
foundations. It also contains a chapter about concrete pre-cast elements. Its structure
and the concepts used are similar to those in the very latest codes.

In 1971, the European Commission decided to launch a program aimed at

harmonizing technical specifications in the construction field in across the Common
Market. The Commission introduced standardized technical rules for construction
work design (“Structural Eurocodes”). The first version of these standards was
introduced for experimental work. The codes use experimental standards from ENV
1991 to ENV 1999, the first parts of which were published in the early 1990s. After
a three-year experimentation period, the ENV experimental standards (see [COL
00a], [COL 00b] and [COL 01, which give those parts of Eurocode 8 ENV related to
buildings) were transformed into EN European standards. In their final format, all of
the design standards therein (approximately 60) were gathered into ten subsets that
make up the Eurocodes included in EN 1990 to EN 1999. Anti-seismic design is the
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 337

direct concern of Eurocode 8 (EN 1998) which deals with different types of building
in earthquake areas and comprises six parts. Parts 1 [COL 03b] and 5 are needed to
design buildings; they cover about the same scope as the PS 92 rules. Several
Eurocodes are generally necessary for the complete definition of a reinforced
concrete construction work: EN 1990 gives the general philosophy and operation
combinations, EN 1991 defines the operations to take into account in addition to
earthquakes, EN 1992 [COL 03a] gives common design rules for reinforced
concrete, EN 1998 introduces the action of earthquakes and the additional
arrangements needed in a seismic situation, whilst EN 1997 deals exclusively with
foundations.

The PS 92 standard will still apply in parallel with Eurocode 8 (co-existence

period), probably up to about 2010; the national standard will then disappear, and
will be replaced by the European standard. The general structure and the main
layouts of both standards are quite close, which should make passing from one to the
other easier, despite the greater complexity of Eurocode 8.

As far as anti-seismic construction is concerned, standardization has a strong

status in France [BET 97]. In fact, lawmakers made anti-seismic protection
compulsory, by introducing the 22 July 1987 law (Article 41) and the 2 February
1995 law (the so-called Barnier law). Whereas only new constructions were affected
by the first law, the second expanded the protection to existing facilities, which
raised technical and economic problems, because many existing housing types were
non-compliant. The different orders and decrees specifying how these laws should
be enforced defined seismic zoning in France, classified different constructions
according to their importance with regard to public safety, and fixed standards and
other technical rules.

9.3.2. Main objectives of anti-seismic protection

Because of the violent dynamic features of seismic action and the inaccurate
knowledge we have about its likely effects, building in seismic areas involves
additional efforts in terms of design and construction quality. The anti-seismic
standards give recommendations that complement other design standards; in fact,
constructions must at least respect the objectives of strength, practicality and
durability set for constructions in non-seismic situations and that are subjected to
these other standards (especially Eurocode 2 in the case of reinforced concrete
structures). However, some Eurocode 8 verifying rules are more restrictive than
those in Eurocode 2, and a well-informed designer can limit the number of
verifications required. The goals of Eurocode 8 should be considered besides the
goals of Eurocodes, and both lie within the framework of EN 1990, which describes
the basics of Eurocode design.
338 Dynamic Behavior of Concrete and Seismic Engineering

The main objective of an anti-seismic design standards is to ensure three things:

the protection of human life during an earthquake, the operational continuity of
constructions important for emergency services, and the limitation of structural
damage. A major obstacle to this aim lies in the randomness of seismic action, the
statistical characteristics of which are difficult to determine in countries with low or
medium seismicity. Furthermore, earthquake protection has a cost, and some
countries have low disposable resources.

In the case of Eurocode 8 (Part 1), these general objectives are expressed by the
following mandatory prescriptions:
– Non-collapse prescription: structures must be designed and built to withstand –
without any local or global collapse – an earthquake, the aggressive level of which
corresponds to an overlapping probability over 50 years determined by the National
Authorities (10% in principle, which corresponds to a 475-year-return period). This
is the reference seismic action. After the event, the structure must preserve a notable
residual strength capacity (especially with regard to permanent loads), and should be
able to withstand a replica without collapsing.
– Damage limitation prescription: the structure should be designed and built so
that, under the effect of an earthquake more important than the previous one, the
structure will not suffer damage the repairing cost of which would be higher than the
cost of the structure itself, or which would limit its normal use. The earthquake level
sought corresponds to an overlapping probability over 10 years set by the national
authorities (usually 10%, i.e. a 95 year-return period). In order to make the
designer’s task easier, such seismic action can be derived from the reference seismic
action by affinity.
– The specific protection assigned to certain types of constructions that are
important for public safety is ensured by means of a JI multiplying coefficient
(“coefficient of importance”) which is applied to the seismic action. This amounts to
increasing the return period (within the limits of the values considered for the
coefficient) or to decreasing the overlapping probability over a given period of time.
In order to make such differentiation possible, constructions are classified into four
importance classes, and a JI value is assigned to each class. Class IV corresponds to
construction works vital for public safety, whereas class II corresponds to standard
buildings.

9.3.3. Verification method

9.3.3.1. General principles

The Eurocode system is based on a series of common principles that apply to the
use of structural materials, either in standard operation situations or in particular
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 339

circumstances (an earthquake, for instance). Such principles aim at ensuring

structure strength, usability and structure durability for an expected lifespan. They
are laid out in EN 90 regulations [COL 02], which also introduce the general
Eurocode philosophy.

The reliability of the approach used is demonstrated by calculating the

achievement of the goals targeted and is linked to the introduction of partial safety
coefficients for each of the materials, the actions and the calculation methods.
However, achieving these objectives is partly connected to the design methods used:
specific protection measures, the quality of the studies and of their execution, as
well as the follow-up of potential failures have contributed in an essential way to the
success of the objectives.

It is an acknowledged fact that the goals targeted have been reached by

verification of the boundary conditions of structures in project situations. These
situations are basic and are represented by combinations of actions in which the
seismic action is considered to have a higher occurrence probability higher than any
other accidental action.

The seismic situation is a specific one that gives rise to particular action
combinations that are expressed the following way:

¦G
j dl
k, j  P  AEd  ¦\
i tl
2,i Qk ,i [9.1]

In the above expression, “+” means “combined with” instead of a mere algebraic
operation. Gk,j refers to the characteristic values (in the probabilistic sense of the
word) of the permanent actions applied to the structure, P is the action representing
pre-stressing, AEd the calculation seismic action equal to the JI AEk product of the
coefficient of importance (see section 9.3.2) by the value characteristic of the
seismic action, Qk,i the values characteristic of the accompanying variable actions
(i.e. those that may be present during the seismic situation) and \2,i accompanying
coefficients expressing the intensity of the variable actions present when the
earthquake occurs. These coefficients are chosen as equal to their corresponding
values of the action quasi-permanent parts. Such combinations are used for
verifications in the considered limit states; they should not be used in a stand alone
manner to calculate the masses to be introduced into dynamic calculations (section
9.4.3.1).

Two structure-limited states classes are considered: ultimate limit states (ULS)
which address the safety of people, structures and possessions, and service limit
340 Dynamic Behavior of Concrete and Seismic Engineering

states (SLS) that are concerned with the normal operation of the structure and its
elements, people’s comfort and the appearance of the construction.

Calculation is far from the only design element in seismic situations, where
general design layouts are essential elements too.

Material fatigue – which will not be dealt with in this chapter – must also be
taken into account when designing construction works subject to important and
repeated dynamic loads, such as railway bridges and seaside works.

9.3.3.2. Application of Eurocode 8 (Part 1)

To be compliant with the Eurocode objectives defined in section 9.3.2, the
following limit states must be checked:
– ultimate limit states: verifying these involves non-collapse and other kinds of
structural failure that might endanger human lives;
– damage limitation states: these are associated with the appearance of damage
and correspond to conditions under the structure is not able to fulfill its duties any
more, while remaining stable.

A feature common to anti-seismic design standards is that verifications required

for specified levels are enough to demonstrate the strength of the construction work
with regard to all seismic events on intermediate levels. The methods do not
necessarily directly apply to intermediate levels. Forgetting this can lead to
erroneous interpretations.

9.3.3.3. Verifications
We need to verify that the structure has enough strength and ductility at the
ultimate limit state, and do this by taking second order effects into account. The
verification can be achieved by introducing acceptable incursion limits into the post-
elastic state. A technique for predicting post-elastic behavior involves applying a
dividing coefficient, called the behavior coefficient (sections 9.4.3.3 and 9.5),
determined by the overall behavior characteristics of the structure and constitutive
material, to the seismic actions calculated from a linear elastic hypothesis. The more
ductile the structure is and the better controlled its behavior, the higher the value of
this coefficient. Application involves dimensioning structures for strength values
lower than elastic.

We also need to verify the joint openings between two adjacent buildings or
between two blocks of a same building to ensure they are sufficient to avoid shocks.
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 341

Under seismic actions that are reduced by the behavior coefficient, the code
demands that the engineers must demonstrate the overall stability of the structure,
the strength of the soil (without any intensive permanent strain) and the strength of
the foundations.

In addition, the non-structural elements should neither be dangerous for people

nor affect the dynamic behavior of the structural elements.

At damage limitation condition, the code recommends that strains be limited

(relative displacements between floors) to avoid partition walling and facade
damage according to their brittleness.

9.3.4. Capacity-design method

To design a structure to withstand seismic action economically, incursions into

the post-elastic field are tolerated. Because such incursions take place in cycles, it is
necessary to avoid reaching the limit strain of the constitutive material of the
structure, as beyond these points, integrity cannot be ensured, as the material
deteriorates, lowering strength. Good design not only seeks to ensure the strength of
constitutive materials, but to preserve their ductility and stable behavior during
cycles as well. The ductility objective is achieved by adopting some design and
verification rules that are more restrictive than those used in non-seismic situations,
by taking incursions into the post-elastic field into account, and setting up specific
construction layouts.

Ductile behavior is quite well illustrated in the case of frame structures, whatever
their constitutive material. Passing into the post-elastic field is evidenced by the
appearance of one or several plastic hinges, typically near the connections
(beam/column intersections). When the stress increases in a monotonic way, hinge
rotation increases and other plastic hinges may appear. This cyclic hinge formation
step does not correspond to any alternating characteristic of seismic action but
absorbs a lot of energy, thus turning the structure into a dissipative structure, which
can limit the damage accrued in an earthquake. For such behavior to be possible,
plastic hinges have to be capable of withstanding important rotations without
significant damage, so that the strength capacity of the structure and its capacity to
dissipate energy do not decrease. Transverse reinforcements and construction
layouts ensure such ductility. It also appears that hinge position inside the
mechanism is relevant for construction work safety, as localized effects in the
columns have to be avoided (Figure 9.15a). Finally, the degree of hyper-staticity of
the structure determines the number of plastic hinges that develop within the
framework to reach the mechanism condition, and is therefore an important
determinant of the total energy dissipation capacity of a structure.
342 Dynamic Behavior of Concrete and Seismic Engineering

(a) (b)

Figure 9.15. Undesirable plastifying modes

The maximum acceleration a structure can counterbalance is limited by the

resistant moment of the hinges, but their energy dissipation capacity allows the
elastic energy to counterbalance the energy injected by the earthquake.

Eurocode 8 recommends a design and dimensioning method called capacity

design. In this, the plastic hinges should arise in pre-determined areas called critical
areas, to allow the structure to reach a mechanism condition whenever energy is
dissipated by ductile rotation of the plastic hinges. The plastic moments inside the
hinges are determined by the resistant moments of individual sections. It is good
practice to arrange for resistant moments to be higher than design moments (by
about 20 to 30%), as the accelerations to which the structure is subject are
determined by resistant moments and not design moments.

This frame example allows the method to be illustrated: to ensure that the plastic
hinges will form where designers want them to, so that energy is dissipated
according to the planned mechanism, the areas outside the hinges must be designed
to stay elastic when the zones in which the hinges should appear (critical areas)
become plasticized. Inside the hinges, the maximum moment possible is the resistant
moment, derived from the design moment by multiplying it with an over-capacity
coefficient. The coefficient takes into account strain-hardening effects, such as the
reinforcements of bending strength in a reinforced concrete section. Because these
maximum moments are known in the critical areas, the other areas are deliberately
over-dimensioned with regard to them. This is supposed to prevent plastic hinges
forming in areas where they are not desired, for example, within the columns, for if
the latter are too numerous here, the risk of instability becomes greater (Figure
9.15b). The goal is to ensure hinging will mainly take place inside the beams during
an earthquake (Figure 9.16).

To ensure engineers grade the appearance of plastic hinges correctly and to

prevent them forming inside the columns, they have to dimension the post-potential
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 343

critical areas using the capacity design method. This leads to the inequality
expressed in the following equation:

¦M c t J Rd ¦M b [9.2]

in which the Mc moments (respectively Mb) are the design-resistant moments of the
columns (respectively beams), and JRD is an over-capacity coefficient (Figure 9.17).

Figure 9.16. Preferable mechanism

Mc1

Mb2
Mb1
Mc2

Figure 9.17. Equilibrium of a node

Capacity design also involves classifying a structure’s failure modes to ensure

that brittle failure modes cannot appear before ductile modes. This ensures a planned
dissipative operation, and avoids the existence of a “cliff” effect, i.e. a sudden drop
in structure strength if the aggression level is exceeded. For reinforced concrete,
shear load behavior is acknowledged to be brittle whereas bending moment behavior
is ductile, providing minimum construction layouts are respected. To prevent a beam
344 Dynamic Behavior of Concrete and Seismic Engineering

segment between two plastic hinges breaking under the influence of shear load, it
has to be designed for the maximum shear load that is obtained by expressing the
equilibrium using the following formula (Figure 9.18):

M DRd  M CRd
VSd ,CD J Rd ˜ [9.3]
lcl

VSd,CD

MCRd Figure 5
MDRd

Figure 9.18. Transverse load determined by capacity design

In this expression, MDRd and MCRd refer to the design resistant moment at both
ends, lcd is the segment length and JRd is the over-capacity coefficient.

The VSd,CD shear load thus obtained is used to design the beam segment and does
not have any direct connection to the shear load obtained by structure calculations
for the same element.

9.4. General design

9.4.1. Design principles

Eurocode 8 encourages designers to adopt layouts that ensure the good behavior
of the structure. These include using bracing that is continuous down to soil level,
and considering the symmetry, hyper-staticity, strength and stiffness in both
horizontal directions, strength and stiffness with regard to the vertical axis torsion,
the existence of horizontal diaphragms on different levels to distribute the seismic
loads on the bracing elements.

Regular behavior should be the goal. It is best achieved by implementing simple

and compact shapes, horizontally and vertically, and to achieve this, the structure
can be designed as a collection of dynamically-independent blocks, each of which
have regular properties. Regular structures simplify the process of performing both
action and behavior coefficient calculations.
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 345

Foundation structural elements have to be stiff enough to transmit the seismic

actions of the structure towards the soil or deeper foundations. Typically there
should be only one foundation type below a structure. If this is not possible, the
structure must be divided into dynamically-independent blocks.

During the general design stage for a structure, we can distinguish the main
elements that constitute the bracing from the secondary elements, which merely play
a supporting role. The secondary elements should not be considered as strong
resisting elements when calculating structural response to seismic action, and their
stiffness can be neglected when evaluating the dynamic behavior.

Except when the structure has been designed as non-dissipative, ductility should
also be considered by designers. Brittle failures, or the premature formation of
unstable mechanisms like hinges, must be avoided to ensure the overall ductility of
the structure. This is achieved by using capacity design (see section 9.3.4). Local
ductility can also be ensured by using the construction layouts described in later
chapters on material structure.

Non-structural elements must be verified as being able to withstand the

acceleration transmitted by the structure; this acceleration is calculated using a floor
spectrum. The calculation can be simplified by using acceleration results from a
standard estimate that uses a behavior coefficient specific to the element being
considered.

9.4.2. Regularity conditions

The regularity of a building must be considered in two parts: plane regularity and
vertical regularity. They are both desirable though not crucial.

Horizontal regularity is ensured by designing so as to restrict vertical axis

torsion phenomena. The criteria are twofold: the first relate to the symmetry and
compactness of the plane shape, and are the subjects of simple geometrical controls.
The existence of two main orthogonal planes over the whole height of the structure
reflects most of these criteria. The second planes are mechanical: we must ensure
that the floor diaphragms are stiff enough with regard to bracing elements that
diaphragm displacements on each level are stiff behaviors (overall translation and
rotation). A second set of conditions deals with the radii of torsion, defined as the
square root of the ratio of the torsion inertia around the torsion center with the
bending inertia in the considered direction. For each principal direction, the radial
deviation of the structure (i.e. the distance between the bending center of the bracing
system and the center of gravity, projected in the direction under consideration)
346 Dynamic Behavior of Concrete and Seismic Engineering

must be less than 30% of the corresponding radius of torsion, and the latter must
stay lower than the mass-gyration on each level.

Figure 9.19 presents two wall structures. In both cases, the radius of gyration is
determined by the mass of the floors, which is distributed over the whole surface.
The case shown in Figure 9.19(a) includes a central core and peripheral columns:
here, the stiffness and strength with regards to torsion focus at the center of the
building, hence the radius of torsion is low, and the building is not considered to be
regular. In the second case shown in Figure 9.19(b), bracing elements are distributed
on the outskirts, and the radius of torsion is similar to the width of the building. The
horizontal regularity allows us to carry out a plane analysis in both of the principal
planes.

(a) (b)

Figure 9.19. Plane configurations different from the point of view of torsion

Respecting vertical regularity allows designers to ensure the progressive

characteristic of first mode displacements, and to prevent modes higher than 1
exerting a non-negligible influence on the displacement of the structure under
seismic action. These conditions allow the use of the simplified calculation method.
They also make it possible to avoid delicate situations linked to sudden variations in
stiffness or mass (“transparencies” on the first level for example, or reversed
pendulums). The conditions are a complex function of the continuity of bracing
elements and progressive variation in geometrical shapes with structural height.

9.4.3. Calculation of seismic action effects

9.4.3.1. Structure modeling

Calculation of seismic actions should be based on modeling of the structure that
takes into account the influence of soil deformation, connections, non-structural
elements and the presence of contiguous structures. As a rule, models consists of
vertical elements that represent bracing elements, connected by other elements that
represent diaphragms. If the diaphragms can be considered as stiff compared to
bracing elements, floor masses and mass inertia can be concentrated on their centers
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 347

of gravity; in such a case, the model is termed a “lumped-model”, consisting of

beam vertical elements connecting the floors.

The masses are calculated by estimating those present in the structure associated
with the seismic action combinations:

¦ G  ¦\
kj Ei ˜ Qki

In practice, the nominal masses associated with the quasi-permanent variable

actions are affected by M coefficients that take into account their presence
probability during the earthquake: \ Ei M .\ 2i . These M coefficients are a
complex function of the level considered in the building and the type of variable
action (i.e. the use of the premises). They typically vary between 0.5 and 1.

For masonry or reinforced concrete elements, cracking must also be taken into
account when evaluating stiffness.

The soil/structure dynamic interaction effects must also be taken into account
when displacements are involved, for example, when modeling structures sensitive
to second order effects, very slender structures and structures built on very soft soils.
The effects of this interaction on piles have to be taken into account whatever the
situation.

9.4.3.2. Elastic calculations

Seismic actions inside a structure result from movements imposed at the
foundation level. When the behavior of the structure can be considered as visco-
elastic from a linear point of view, the structure dynamics methods allow calculation
of the actions from a pseudo-acceleration spectrum or by direct time integration
(time history calculation).

The multi-modal calculation method that uses response spectra is detailed in

Eurocode 8.

Pending concordance with the regularity conditions outlined in section 9.4.2, a

simplified method can be used. This involves applying – in each principal plane – a
horizontal static stress system on each different level. This stress system is
determined proportionally to a single mode the shape of which is given a priori. The
associated basic period can be determined by an approximate formula that includes
the height (H) of the building and on a coefficient characteristic of the bracing type.
The whole mass is assigned to that single mode for flexible buildings (T>2TC), and
85% of the mass is used for stiffer buildings.
348 Dynamic Behavior of Concrete and Seismic Engineering

9.4.3.3. Taking dissipative behavior into account

When considering the ductility of the structure by plasticizing critical areas of
elements, the non-linear behavior of the structure can be derived from equivalent
linear calculation (see [BET 91], section 3.17). In such cases, actions are calculated
from an elastic linear model using a design spectrum that incorporates a dividing
behavior coefficient. The structure is therefore designed using loads lower than
those derived from linear calculations. This allows an estimation of non-linear
behavior from linear calculation, which is more realistic from an engineer’s
perspective, and allows them to avoid the problems associated with non-linear
modeling, which is fraught with difficulties. Behavior coefficient determination is
discussed in section 9.5.

9.4.3.4. Non-linear calculations

Two calculation methods that take the post-elastic behavior of a structure into
account are considered: the so-called “push-over”, pseudo-static method, and the
“time history analysis” method, the latter being reserved for exceptional situations.
Both aim either at evaluating either the Du/De ratio defined in section 9.5.3, the
strength of existing buildings, or the location of potential plastic hinges and
damage. Such modeling can be bi-or tri-linear, but must be based on modeling that
is truly representative of the post-elastic behavior of structure elements and must
take potential damage into account.

The “push-over” method is internationally approved. It involves applying a

given distribution load system (proportionally to the loads derived from the elastic
analysis or those resulting from applying uniform acceleration to the height) and
increasing in intensity. A “capacity curve” is then drawn, which gives the load
characteristic of the seismic action (typically the shear load at the base) according to
a characteristic displacement (at the top of the building, for instance).

After this, we must verify that the structure can be deformed so that a “target”
displacement (characteristic of a one-degree of freedom system resulting from a
spectrum reading for the equivalent basic period of the system) can be achieved
without exceeding the ultimate strain. The one-degree of freedom system considered
is equivalent to the basic mode of the structure.

9.4.3.5. Taking vertical axis torsion into account

Structural torsion can prove dangerous if it is not well-controlled. It has three
main causes. The first is differential motion at the base of supporting points due to
wave propagation within the soil. If the supporting points are linked by structure
elements that are stiff enough, the seismic motion that is stressing a structure can be
compared to an overall displacement and rotation, the vertical component of which
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 349

corresponds to a torsional stress. The second cause of torsion is the natural

eccentricity of the center of gravity of each floor with regard to the torsion center of
the bracing system. This eccentricity can be calculated from structure plans, but its
value is associated with some uncertainty due to assumptions made about mass
distributions, the evolution of the torsion center during the motion, cracking, and the
appearance of plasticized or damage areas. The third cause of torsion is deformation.
Actually, when a structure is flexible with regard to torsion but eccentric, its basic
seismic response mode can be a combination of torsion and overall bending. In these
cases, implementing that mode simultaneously generates a bending motion and an
associated torsion that is amplified.

The random aspects of torsion are covered by taking into account what is termed
“fortuitous eccentricity”, which is equal to 5% of the dimension of the building in
each principal direction. Assessing the extent to which building motion amplifies
torsion can only be achieved through the use of adequate space models, but a
simplified approach involves sequentially moving the application point of the
seismic force away from the center of gravity on each level.

9.5. Behavior coefficients

9.5.1. Using behavior coefficients

The behavior coefficient expresses ductility, over-strength and overall behavior,

and is used as follows: seismic loads are calculated making a linear elastic
assumption, and the structure is then designed, using the elastic loads divided by the
behavior coefficient.

Without going into the detail of the behavior coefficient method, the assumptions
on which it is based are outlined in Figure 9.20. It represents a relationship between
a stress (which represents the seismic action: shear load at the base, for example)
and a characteristic displacement (horizontal displacement at the level of the center
of gravity or at the top of the structure). By likening the behavior of an actual
structure to one with perfect elastic-plastic behavior, an ideal linear elastic structure
would withstand a seismic action system q times as high as the plastic plateau.
However, it is accepted that the displacement obtained by taking the elasto-plastic
behavior of the structure into account is the same as it would be for the fictitious
linear structure. The rule of displacement equality can only be confirmed for flexible
structures. With stiff structures, the calculations show that there is an equivalence of
strain energies (areas encompassed in the F(d) curve), consequently, structural
displacement is more important than elastic displacement. Eurocode 8 uses the
elastic displacement calculated with the elastic spectrum as the value of the structure
350 Dynamic Behavior of Concrete and Seismic Engineering

displacement, without making any increase for weaker periods (except for the target
displacement in the “push-over” method).

F
Fe, u

Fe

Fdim

de d du d

Figure 9.20. Perfect elastic-plastic behavior

Figure 9.20 shows that value of the behavior coefficient has a limit, determined
by the ultimate displacement the structure can withstand before collapsing.
Eventually Fe ,u q max Fdim . This global criterion is expressed on reaching ultimate
strains the building cannot exceed. A law expressing the global behavior of the
structure integrates the local behaviors linked to the ductility of the materials used. It
also depends on the degree of hyperstaticity and the type of the elements used
(beams, slabs, columns, walls, piers), as well as on their distribution in space.

In practice, it is meant to represent the maximum value of the behavior

coefficient, assuming it is reached at the ULS. This means that, for a lower load
level, a smaller behavior coefficient will be used. If the load is low, the structure
remains elastic and the behavior coefficient is equal to 1. As mentioned in section
9.3.3.2, it is not acceptable practice to use the maximum behavior coefficient for a
seismic level weaker than the one for which the structure was designed.

The behavior coefficient also allows other less easy to master phenomena to be
incorporated: for instance, structure irregularity implies a reduction of the behavior
coefficient, i.e. an increase of the design strength of the structure, because
irregularity-linked phenomena are more difficult to control.

It should be noted that the behavior coefficient is not used when calculating
displacements or for determining the opening of joints.
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 351

The equivalent linear analysis is mostly based on considerations of ductility and,

as a rule, does not apply in the case of marked geometrical non-linearities, like cases
of uplift between foundation and the soil.

9.5.2. Structure behavior and behavior coefficients

A structure is dissipative if it dissipates energy during hysteresis cycles via the

ductility of its constituent materials.

A structure can be designed as non-dissipative if its behavior remains quite close

to linear elasticity: in such cases, a behavior coefficient equal to 1 would be
assigned to it, and the elastic spectrum could be used. However, when the structures
are designed according to the 2 to 6 structure Eurocode regulations, they retain a
little ductility (limited so-called “L” ductility) and over-strength linked to
hyperstaticity. Therefore, for all structural materials, a behavior coefficient ranging
from 1 to 1.5 (2 for metallic structures and mixed structures) is used, with a few
additional considerations as far as materials are concerned, but without any specific
construction layouts.

When the structure has been designed as dissipative, a behavior coefficient

higher than 1.5 (or 2) can be used, as long as construction layouts allowing the
structure to reach the expected ductility are used as well.

To determine the behavior coefficient of a given structure, the constituting

material of the structure, the bracing type, the ductility level considered and its
regularity must be taken into account. In Eurocode 8, typical data for a building that
is irregular in height is given in Table 9.1, from which it can be seen that the basic
value of the behavior coefficient is equal to 0.8.

As far as reinforced concrete frameworks are concerned, energy dissipation is

directly linked to the number of plastic hinges that have to be formed. Therefore,
hyperstaticity is taken into account by multiplying the reference behavior coefficient
by the Du/De ratio. Using static calculations, we assume the structure is subject to a
seismic action system obtained by the analysis described in section 9.4.3. The
resulting actions are proportionally increased by an D coefficient. The first plastic
hinge will appear at the De value (elastic limit), and the structure will become a
mechanism for the Du. Thus, for a reference value of 4.5 and a reinforced concrete
frame, the behavior coefficient can reach a maximum close to 7, because the Du/De
ratio will be restricted to a value of 1.5. It is possible to avoid calculating the Du/De
ratio by accepting a fixed value equal to 1.2, in the case of one-bay frames for
example. A similar approach is acceptable for high-ductility walls, for which the
reference value will be equal to 4.
352 Dynamic Behavior of Concrete and Seismic Engineering

9.5.3. Local ductility and behavior coefficients

For frame-type structures, if the elements have a homogenous strength, plastic

analysis reveals that plastic hinges form at the ends of each column or beam
element, providing brittle failure modes have not appeared before their formation.
Capacity dimensioning suggests the measures that have to be taken to obtain
behavior in accordance with the theoretical formation diagram of plastic hinges. The
dissipating capacity of the structure is linked to the hinge’s ability to withstand
rotational strain in the plastic field without a reduction in strength. This ability is
achieved using assiduous construction layouts and/or section selection. For higher
behavior coefficients, layouts are restricted. To correctly control plastic strains, it is
better if connections between elements do not plasticize. For this reason, they are
generally over-dimensioned by applying capacity design regulations, so that the first
plastic hinges form near the connections instead of forming within the connections
themselves.

The dissipative capacity of a structure is thus mainly determined by capacity

design, careful choice of sections and construction layouts. Structures are then
classified into ductility classes, each class being characterized by the more or less
strict application of these rules. The value of a structure’s behavior coefficient is a
function of its ductility class.

For ductile structures, a relationship exists between the ductility available in the
plastic hinges (expressed by a conventional rotation ductility coefficient PI,), and the
behavior coefficient. Eurocode 8 proposes the following relationship for reinforced
concrete structures:

­ PI 2q  1 if T1 t TC
°
® § TC · [9.4]
° PI 1  2 ¨ q  1) T ¸ if T1  TC
¯ © 1 ¹

which expresses the fact that stiff structures are less ductile. That relationship is
operational, as PI is connected to the layouts taken to confine concrete within the
plastic hinges.

9.5.4. Ductility classes and behavior coefficients

Whereas the PS 92 regulations proposed only one level of ductility to designers

(corresponding to a “medium” ductility), in Eurocode 8, three classes are suggested.
These are:
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 353

– the limited ductility “L” class, which corresponds to the application of

Eurocode 8, without any additional condition except for the ductility of materials.
Actually, Eurocode 8 is deemed to confer some limited ductility that can be taken
advantage of, as the behavior coefficient is limited to 1.5 in that case. This can be
quite useful in low seismic areas when economically justified. Stresses due to wind
can be higher than those due to an earthquake, at least in the short-dimension
horizontal direction (or “gable plane”). This class is not specific for reinforced
concrete, and similar layouts have been adopted for steel frames with Eurocode 3
and for mixed structures with Eurocode 4;
– a medium ductility “M” class for which layouts specific to a seismic situation
are used to ensure ductility and dissipation without any brittle failure occurring;
– the high ductility “H” class that enables energy dissipation higher than in the
previous class.

A q behavior coefficient is associated with each ductility class, dependent on the

structure type.

9.6. Designing and dimensioning reinforced concrete structure elements

9.6.1. Regulations specific to reinforced concrete in seismic areas

Specific regulations govern the design of main and secondary structural

elements, diaphragms, construction arrangements and reinforced concrete pre-cast
elements.

As far as the calculations of the bending moment (potentially composed) and

shear loads are concerned, Eurocode 2 verification rules apply.

Reinforced concrete structures must comply with the capacity design rules given
in section 9.3.4. Local ductility should be organized to allow rotation within the
plastic hinges, when some can form.

To achieve this, several conditions specific to the structure elements involved

have to be met: concrete and steel must have improved features, the reinforcements
must have sufficient ductility (length of the plastic plateau without any reduction of
strength) and strength must be exceed their elastic limit. For concrete, this equates to
compression strength and ductility. In addition, rotational ductility should be
ensured by the use of appropriate layouts, particularly where confined concrete is
concerned, and the possibility of premature failure due to shear loading must be
minimized or completely eliminated.
354 Dynamic Behavior of Concrete and Seismic Engineering

Whilst not included to date, some layouts involving novel anchoring and
reinforcement coverings will be included in future versions of Eurocode 2.

9.6.2. Main types of reinforced concrete bracing

From the point of view of stiffness, ductility and strength, the seismic behavior
of a construction is a function of the bracing type chosen. Besides the choice of a
material, bracing depends on the geometry of each constituting element, the
horizontal and the vertical layouts of its constituents, and their interactions.

For vertical elements, two types of reinforced concrete units can be used:
columns or walls. Characterization of these elements is not definite: passing from
one to the other is continuous. Geometrical non-linearity (see section 9.6.4.1) can
play an important part in the dynamic behavior of walls because sections have an
elongated shape. It is customary to treat an element as a wall if its section is equal to
four times its thickness.

Some construction types have isolated walls as bracings, and these can
potentially behave as vertical beams with one hinge at the base. Such walls can also
be connected to different floors in a construction because lintels dissipate energy by
creating plastic hinges at their ends, subjecting walls to strains. Coupled walls can
be designed to dissipate a lot of energy into lintels, and these would have a high
behavior coefficient.

Quite frequently, walls are not separated: coffers, for example, have partitions in
both directions, very high strengths, low ductility and very good behavior with
regard to torsion if distributed on the periphery of the construction. Opening rows in
the walls can delimit the ability of lintels and piers to dissipate energy.

Another type of commonly found construction combines cores (walls that can
contain stairwells or lift shafts) and load-bearing columns considered to act as
secondary structures. If the construction has only one center core, torsion can be
high.

In the case of columns used for bracing, good overall behavior can be achieved
by ensuring continuity of moments in horizontal planes, which involves placing
beams that will form frames with the columns. The continuous beam-column sets
form frames that constitute the primary structure. Frames take advantage of a
behavior coefficient that is high, as is the degree of hyperstaticity. As section 9.3.4
explained, plastic hinges have to be placed inside the beams without ignoring the
capacity design criteria.
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 355

In the case of masonry partitions placed within the frames, several annoying
phenomena can occur: structure stiffening, suppression of its regular feature,
creation of short column effect if the partition is lower than an adjoining column,
and bracing struts becoming stops in the walls. Specific layouts are taken in such
situations, especially when the whole height of the columns is designed as a critical
area.

If the beams are not placed in continuity with the columns, the latter behave as
vertical consoles and have no hyper-static reserve or high ductility. Any structure
built along these lines cannot therefore take advantage of a high behavior
coefficient.

Finally, bracing system would not be complete without horizontal elements

connecting the vertical elements to transfer horizontal inertial actions towards
vertical bracing elements. In buildings, floors constitute these horizontal structures
(“diaphragms”). Floors should behave in a monolithic way, and slab floors poured
on location do, even if the supporting system consists of pre-cast elements (either
beams or slabs). If such slabs cannot be installed, or if they are weakened by a lot of
openings, the stress transfer should be ensured by chaining and continuous bracing
struts, maintaining transfer continuity of diaphragm stresses towards vertical
elements.

Table 9.1 gives values for the basic behavior coefficients of different bracing
types according to their ductility class.

Bracing type DC “H” DC “ M”

Du Du
Frames, coupled walls, mixed systems 4.5 3.0
De De
Du
Wall bracing 4.0 3.0
De
Cores 3.0 2.0
Reversed pendulums 2.0 1.5

Table 9.1. Basic values of the behavior coefficient

356 Dynamic Behavior of Concrete and Seismic Engineering

9.6.3. Main frames

9.6.3.1. General behavior

Main frames are series of columns and beams that form frames on several floors
and several rows. The column-beam intersections, called nodes, allow loads to be
transmitted between different linear elements. Under permanent and vertical variable
loads, bending moments are typically positive through beam rows and negative near
the supports, whereas the columns are mostly subject to compression normal loads.
Under the effect of a horizontal seismic action, moments vary in a linear way
between nodes within the different elements. Thus, moments are maximum near the
nodes within the columns and beams. Moments change signs during motion because
of the alternating feature of the action. This means columns are subject to bending
moments and shear loads, which are generally more important during an earthquake
than in a normal situation, which explains why they are brittle in the event of an
earthquake.

Beam rows are typically not very sensitive to vertical components of seismic
actions, as they are designed for variable actions that are more intense than the
vertical acceleration on the masses present during an earthquake. This is not true
when the masses have their maximum value (archives room) or when the
acceleration is such that it can reverse the moment sign. On the other hand, the beam
ends near nodes sustain sign inversions of their bending moments, which can
generate traction on the lower sides for which longitudinal reinforcements are
necessary.

Plastic hinges form in areas where bending moments are maximum, near the
nodes. The latter sustain alternating loads transmitted by the linear elements and
expressed by compression and traction inner stresses that can cause damage or node
failure. This should be avoided to preserve the integrity of the structures, so nodes
tend to be over-designed with regard to the adjacent linear elements, according to
the capacity design principles. Most designs try to ensure that plastifications occur
inside the linear element beginnings near the nodes, instead of inside the nodes.

As shown in section 9.3.4, to maximize stability, engineers should avoid having

plastic hinges that form inside the columns during a plastic cycle. However, it
should be noted that it is almost impossible to avoid plastic hinges at the base of
columns to reach the ultimate isostatic condition.

Once the position of plastic columns has been ensured by the application of
capacity design rules, the ductility of the hinges in response to rotation has to be
ensured as well. This is governed by separate local ductility verification standards.
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 357

Construction layouts are required to cover the uncertainties that remain at each
stage of seismic action design, especially those concerning ductility. Layouts deal
with beam and column design, minimum reinforcements on beam upper sides, the
minimum percentage of traction reinforcements inside beams, the minimum
reinforcement percentage and the limitation of the normal compression stress inside
columns.
Finally, as a general rule, because frames can be very flexible, if no other
additional layout is taken, the relative floor-to-floor displacements should be limited
to minimize P-'effects.

9.6.3.2. Verification of composed bending sections and rotation ductility

Verification of sections subjected to composed bending is carried out according
to methods laid out in Eurocode 2. The reliability of a section thus designed is quite
well controlled, as they are based on many empirical measurements. In addition to
partial safety coefficients being applied to materials, part of the safety requirements
are based on strain tolerated limits, others on the fact that the stress tolerated by
reinforcements is limited to their steel elastic limit rather than their resistant
capacity.

In the ULS calculation conditions for a simple bending section, the concrete and
steel strains are respectively limited to 3.5% and 10%. This corresponds to a
maximum 0.0135/d bend (where d refers to the usable height of the beam). With a
length equal to the usable height, the rotation of the beam is equal to 0.0135 rads,
which is generally too small to allow justification as a plastic hinge.

Actually, and subject to proper reservations, section rotation may be far more
important, which means that the usual bending model is insufficient to represent the
behavior of a concrete plastic hinge. Plastic rotation depends on both the integration
length of the bend in the critical area and the strains reached.

As far as tensile reinforcements are concerned, the 1% conventional limit strain

is far lower than the failure strain, which is generally closer to 10%. However, the
strain limit of reinforcements is given by the bound limit, beyond which the
assumptions that make it possible to calculate bends are no longer valid.

For compressed concrete, the composed bending calculation method takes the
uni-axial behavior of the material into account. In a multi-axial stress condition, this
behavior is highly variable. When helically reinforced, concrete strength and
ductility are both improved.

Transverse reinforcements ensure confinement, and their effectiveness is

therefore directly related to the ability of critical areas to undergo plastic rotations.
358 Dynamic Behavior of Concrete and Seismic Engineering

9.6.3.3. Ductility classes and local ductility

In addition to the “L” ductility class (Eurocode 2), two others, “M” and “H”, can
be used (one class per structure). They correspond to two different ductility demand
levels involving plastic hinge rotation capacities, with “H” class materials being
higher.

To ensure such rotation capacities, a ductility coefficient should have a minimum

value in all the sections concerned, determined as shown in section 9.5.3. In
practice, the minimum value is obtained by incorporating sufficient transverse
reinforcements to confine the concrete within the core they encompass. The concrete
inside is able to stand shrinkages higher than those tolerated by non-confined
concrete without any damage.

For critical areas in columns, the following relationship between both quantities
is proposed in Eurocode 8:

bc
DZwd t 30PI vd H sy ,d  0.035 [9.5]
b0
Here, Zwd refers to the mechanical percentage of confinement reinforcements, D
expresses confinement effectiveness, Qd is the reduced normal load, Hsy,d is the strain
elastic limit of the reinforcements, bc the total width of the element and b0 is the
width of the confined concrete.

This expression shows that it is not necessary to confine the concrete if the
required strain is lower than or equal to 0.35%.

9.6.3.4. Beam/column nodes

Nodes are special strain localization areas consisting of limited dimension
volumes that demand a specific analysis of stress transfer. Beam-column nodes, for
example, transmit the loads that are concentrated within compressed areas of
concrete sections and within traction reinforcements. Such loads are often redirected
inside nodes to allow moment inversion. The nodes have to be designed to withstand
shear loads, which they are subjected to in three directions. Here again, good design
involves placing transverse reinforcements in the right position in sufficient
quantities. Compression within diagonal bracing struts should be limited and bar
anchoring within nodes plays a prominent part.

Edge nodes and frame intermediate nodes have quite different behaviors, as do
opening and closing nodes.
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 359

In order to analyze such nodes, two models have been proposed:

– a model of concrete confined by transverse reinforcements;
– a diagonal bracing strut model.

Eurocode 2 verifications adopt the same propositions as those put forward by the
International Federation of Concrete ([COL 91] and [COL 99]). These combine
Eurocode 1 and 2 approaches according to the application field.

The shear loads used are those that derive from the capacity design method,
which takes the actual beam strengths into account.

9.6.4. Reinforced concrete bracing walls

9.6.4.1. Two opposing approaches

Two concepts are used to cover the design of reinforced concrete walls. They are
based on two different methods for counterbalancing the energy injected into the
structure by an earthquake. When a structure moves, the energies playing a part
include: the kinetic energy (at its minimum when the strain of the structure is
maximum), the strain energy (elastic energy and energy dissipated as heat) and,
whenever possible, the potential energy of a structure’s dead weight.

With the first design concept, the walls are organized so as to dissipate some
energy. They are considered as isolated vertical beams liable to develop ductility
similar to that of a beam or a column inside a frame. As the maximum moment is
generally at the base of the wall, a plastic hinge is placed in the area to facilitate
energy dissipation. The confinement conditions of concrete needed to ensure
rotation of the plastic hinge are thus set up, similar to a hinge at the base of a
column. Above the critical area, as the bending moment decreases very quickly, the
rest of the wall will remain in a quasi-elastic condition. Obviously, this design
method can only apply if a plastic hinge can be formed at the base of the wall, which
is not the case if the support becomes detached from its foundation, for instance.

Other factors apart from ductility influence the dynamic behavior of a reinforced
concrete wall. These include:
– geometrical non-linearities (such as the sole of the wall lifting above the soil
due to the overall moment of tilt, or a crack opening within reinforced concrete
sections that causes mass lifting) bring into play the potential energy of the weight
involved. As this energy is maximum when the strain of the wall is maximum, part
of the energy injected into the structure by the earthquake is stored. This potential
360 Dynamic Behavior of Concrete and Seismic Engineering

energy can be restored, but in a motion phase where the reversible strain energy
decreases, since both kinds of energy are transformed into kinetic energy;

– the energy locally dissipated in the soil by irreversible dissipation, as the

strains in the ground may become very high at the ends of a wall;
– redundant walls, often found in French constructions. These are often linked
together and can form rigid sets in which the shear strains may prevail. Such
construction types have a very important strength reserves, but no ductility. In these
cases, the first concept cannot apply.

A second design concept for bracing walls has been adopted. This suggests
cracking distributed over the whole height of the wall is the best way to take
advantage of the potential energy from the dead weight. This implies that, on the one
hand, the layout of the bending reinforcements should be optimized to distribute the
cracking, which is better than cracking concentrated on the foot which stresses the
materials far more and on the other hand, that a high strain of concrete will not be
required; hence, as a rule, it is not necessary to make particular confining layouts.

Actually, each concept has its own merits and its application fields are different:
they merely overlap from a theoretical point of view. This is why Eurocode 8 has
chosen to introduce both methods, and specifies the cases each one is best suited to.
The compression normal force in a wall is an essential parameter to determine
rotation capacities for stressed areas; in fact, the maximum bend inside a section is
reached when the strains in compression concrete and the tensile reinforcements are
simultaneously at their limit values, which occurs in simple bending. When
compression is present, the neutral axis gets re-centered, and the maximum strains
cannot be obtained simultaneously. Thus, choosing between both concepts is based
on the influence of the compression normal force, then on wall density (the walls are
considered as second-type if the basic period is lower than 0.5 s).

9.6.4.2. Designing ductile walls

Ductile walls form the whole or part of bracing systems. They do not carry an
important part of vertical loads, which are mainly supported by columns. This is also
true if the number of walls is not important (two or three walls in each direction for
example). These walls are either isolated or coupled with each other in one plane by
lintels. A wall pierced with large openings consists of piers coupled by lintels, which
ensures strain compatibility that is controlled by rotation at the base.

By extension, the concept also applies to the center cores of high constructions,
as the cores play the same structural role as vessel vertical beams.
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 361

Plastic hinges must be organized at the base of ductile walls or cores, which
implies that they are connected to the foundations without the soil being able to lift,
otherwise, the hinges would not be able to form. As a consequence, the foundations,
whether deep or shallow, should be designed to oppose any lift below the critical
area planned at the foot of the wall, and to transmit the moments of tilt, considering
a possible transfer of vertical loads through longitudinal girders. In the case of pile
foundations, this implies traction within piers.

To ensure correct operation with regard to the assumptions chosen, it is better to

use walls that are regular. Sudden section reductions (thickness as well as length) on
the upper floors can cause the formation of additional plastic hinges, and an overall
less controlled behavior.

As walls or ductile cores mostly work in bending, critical areas are subject to
rotation ductility pulls related to the required behavior coefficient, in a predominant
bending situation and with restricted normal force. Thus, the behavior coefficient
can reach quite high values. Should the normal force influence the behavior of the
plastic hinge more markedly, the available ductility would be more restricted and the
behavior coefficient should logically be limited. Besides, in that kind of wall, the
reduced normal force is restricted to 0.4. Furthermore, the failure mode depends on
the influence of the transverse force and, in practice, of the wall slenderness. This is
the reason why the behavior coefficient is bearing a kw conversion factor ranging
from 0.5 to 1 depending on the slenderness.

For this type of walls, Eurocode 8 has chosen two ductile levels, as for the
frames. In order to fulfill the ductile demand, construction layouts allowing the
confinement of the critical area concrete are recommended, and capacity design
applies. Consequently, the main column construction layouts referred to above are
found in the walls. Moreover, flanges can be placed at both ends to improve the
behavior of the compressed area.

Capacity design is needed to assess safety issues relating to failure because of

shear and the potential additional acceleration caused by the hinges being designed
with over-resistance. As a consequence, it is recommended that the bending moment
diagram of a building should be displaced upwards to take into account the effect of
the shear force on the failure mode at the base. This shifting also allows an
acceleration rise above the plastic hinge to be taken into account. This compensates
for uncertainties in the distribution of bending moments and shear forces, and
ensures that areas located above the critical zone stay within the elastic region,
which should guarantee restoration after insult. It has been proved that shear load in
a wall is not reduced (with regard to its elastic behavior); therefore, designing so that
the shear load is increased by 30% within “M” ductile walls and by a higher
multiplying coefficient in “H” ductile walls assures compliance with Eurocode 2
362 Dynamic Behavior of Concrete and Seismic Engineering

safety requirements. To verify the composed bending in critical areas, the bending
moment and the normal force derived from both calculation and Eurocode 2
verification rules should be used. In the other areas, the bending moment should be
increased as explained above. Additional verifications recommended relate to local
ductility, the minimum percentage of confining reinforcements inside both ends of
the wall being determined by the required PI coefficient.

With high (H) ductile walls, verification of the shear load is recommended: it
must consider the different failure modes observed in a truss, namely: diagonal
bracing strut failure, traction diagonal failure and horizontal sliding failure. The
appearance of such failure modes is influenced by the reduced normal force and by
the slenderness ratio: D s M Ed / VEd / lw  . For low values of Ds (Ds2), the shear
load prevails and vertical and horizontal reinforcements should be placed inside the
regular part of the wall, to sew in the diagonal tension load. Distribution between
vertical and horizontal reinforcements depends on the value of Ds. For low values,
vertical reinforcements are the most efficient, but the number required depends on
the normal force. At higher values, horizontal reinforcements will assure strength, as
shown by conventional verification rules. For intermediate Ds values, both
reinforcement types should be placed inside the wall. To withstand potential
horizontal sliding in the critical area, the friction strength along a potential crack is
taken into account, as the friction coefficient retained in a cyclic stress condition can
be either 0.6 or 0.7, the actual value being a function of the surface condition, of the
dowel effect strength of vertical reinforcements that cross the sliding surface, and of
the traction of inclined bars placed across the surface.

9.6.4.3. Designing little ductile walls

Little ductile walls carry most of the vertical load (typically over 50%). Most of
the time such walls are long and not very slender. They are often redundant from a
mechanical point of view, but may be used for architectural or acoustic reasons. In
some cases, their number and position rule out hinge formation at their bases.

Providing vertical reinforcements are correctly designed, cracking can be

distributed over the height of these walls, and this can significantly influence the
deformability of the entire structure. These walls can lift off its foundations due to
tilt moments. Their behavior is geometrically non-linear, and the energy they receive
during an earthquake is partly counterbalanced by mass lifting. As a consequence,
the normal force plays a beneficial role (on condition that stresses in compression
are limited and that the out of the plane stability is ensured), since it is directly
linked to the mass liable to be lifted.

Because of the opening and closing up of cracks and mass lifting during part of
the motion, vertical impulses are applied to the wall so that a vertical motion
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 363

develops, mainly by response of the first vertical mode. Such motion gives rise to a
dynamic normal force, and this should be taken into account in a building’s design,
as it is relevant to overall strength. This dynamic normal force should not be
confused with the normal force due to an earthquake’s vertical component. Its
amplitude is still not well understood, but foundation conditions are thought to
determine the first vertical mode.

As for strength verification, the bending moment derived from dynamic

calculation can be used without significant overestimation. With the horizontal
component of the earthquake, the normal force increases or decreases with the
dynamic normal force: Eurocode 8 gives this an absolute value equal to 50% of the
permanent normal force. The shear force increases with the (1+q)/2 coefficient and
ensures that a premature brittle failure due to the shear force will not take place. As a
rule, verifications with regard to the composed bending and the shear force are in
keeping with Eurocode 2 verifications. Nevertheless, as the dynamic normal force is
taken into account, the limit strain of non-confined concrete is raised to 0.5%, which
will restrict some practical applications.

With this type of wall, the density of the reinforcements is restricted: capacity
design is not suitable (except to verify the increased shear load) and it is not
generally necessary to confine the concrete, as the strain does not concentrate inside
a plastic hinge, except if the dynamic normal force brings about excessive concrete
strains. As concrete sections are usually over-abundant, shear stresses are often
weak anyway, and web reinforcements are useless if the concrete shear strength is
sufficient.

Three main failure modes can play a part, depending on the influence of the
shear load:
– failure due to a bending moment, with formation of horizontal cracks and/or
reaching the ultimate strain of the concrete or of the vertical reinforcements at the
ends: this arises when the shear stresses are weak. Verification corresponds to the
composed bending type within horizontal sections;
– diagonal cracking failure resulting from a composed-bending shear load
combination. Analysis of such a mode is carried out by a truss method, as bracing
struts are formed by the floors and several truss rods may potentially form in a level
between two floors according to the level height/length ratio;
– failure due to sliding on a horizontal plane, verification of which is carried out
according to Eurocode 2, as the main strength terms are the same as for ductile
walls.

The method developed in Eurocode 8 originated in France, and emerged from a

series of experiments on little reinforced walls: the CASSBA test, then the CAMUS
364 Dynamic Behavior of Concrete and Seismic Engineering

test series ([BIS 98] and [BIS 02b]). It was first codified in the PS 92 rules ([COL
95a], section 11.4, paragraph 11.8.2). Although Eurocode 8 verification rules differ
from those in the PS 92 rules, the basic principles of behavior and strength modes
are the same, except the diagonal cracking failure mode is not considered in the PS
92 rules.

Another difference concerns the behavior coefficient: in Eurocode 8 this type of

wall is only considered in M ductility, therefore its value is limited to 3, yet it does
not require any specific justification. Moreover, this maximum value decreases for
walls that are not very thin. In these cases, the shear load does influence the dynamic
behavior, and the behavior coefficient is therefore lower.

Data obtained in the last series of CAMUS tests disclosed important factors.
These are outlined below, but have not yet been incorporated into the standards:
– because of the highly non-linear behavior of concrete due to cracking, linear
calculations do not allow all the phenomena that appear during the motion to be
taken into account. For instance, a stress in one direction can generate displacements
in the perpendicular direction, owing to the different extensions of the medium lines
of the walls caused by different cracking directions (whatever the cause);
– floor torsion stiffness plays an important part in the redistribution of stresses
and the ultimate strength of a building, but this stiffness varies with time because of
variable cracking conditions;
– the dynamic normal force does not influence bending, yet it modifies the strain
of concrete and the reinforcements, though uniformly.

The calculation method developed in the PS 92 rules gives satisfactory results

for the ultimate strength. However, it does not allow accounting for the
displacements that it widely underestimates. The very accurate models developed
during the experiments have given satisfactory results, but they are not yet within
designers’ reach, mainly because they do not allow the representation of a whole
construction work. In this respect, advances are expected as far as modeling and
calculation practical methods are concerned.

9.6.5. Detail designing

9.6.5.1. Anchoring and overlapping

The anchoring of reinforcements within the concrete matrix implies the stress to
which the reinforcement is subject will be transferred towards the concrete, whether
it is compression or traction. This transfer is carried out by several different
mechanisms:
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 365

– mere chemical “sticking”, which only gives low strength and does not play any
effective part in energy transfer;
– conventional friction, which can make important contribution if compression
stress is applied perpendicularly to the reinforcement axis. Such compression stress
can be either active (i.e. due to pressure applied to the anchoring area) or passive
(linked to the confinement of the anchoring area), because its shear strains (caused
by Poisson’s effect) are prevented by transverse reinforcements called confining
reinforcements. Friction strength is linked to the reinforcement surface condition. It
determines the pull out strength of plain reinforcements, though cannot account for
the pull out strength of high adherence reinforcements by itself. Conversely, a
normal traction condition on the reinforcement axis (generated by Poisson’s effect
when the bar elongates, for example) decreases the pull out strength;
– by concrete buttressing on the reinforcements acting as high adherence
reinforcements. This phenomenon involves formation of conic cracks initiating in
the excrescences. The transverse reinforcements sew the cracks. Once the maximum
strength has been reached, two failure mechanisms can develop: either crack
propagation through the coating that ends up in its separation, or cylindrical crack
formation around the bar. In both cases, strength decreases very rapidly.

The phenomena described above are influenced by different parameters,

including the geometry of the bar, concrete strength, confinement, the design of the
transverse reinforcements, coating and bar spacing. The Eurocode 2 formulae that
allow determination of the anchoring length take these different influences into
account. In a seismic situation, the strong extension or shrinking cycles of the bar
that is to be anchored can cause variations in both the confining condition and the
behavior of concrete. Sliding strength is modified in the same way as the
compression strength or the ultimate tensile strength of concrete by the application
of high amplitude cycles. The deterioration of that strength becomes increasingly
significant, by increasing either the amplitude or the number of cycles. With cycle
amplitudes set at about 80% of the static sliding strength, the deterioration is
important and must be taken into account. Consequently, as anchoring in the critical
areas is stressed according to the real capacity of the reinforcement because of the
potential rotation of the plastic hinge, it is better to design in such a way as to avoid
it. It will be the same for overlapping lengths as it is for anchoring lengths. Eurocode
8 takes overcapacity of reinforcements in critical areas into account and introduces
specific regulations for bar anchoring into the nodes.

9.6.5.2. Compressed reinforcement buckling

As was discussed in section 9.2.3 (see also Figure 9.12), the reinforcements may
buckle when they are strongly compressed, especially in plastic hinge regions. It
also arises in conventional column layouts where concrete shrinking is higher than
366 Dynamic Behavior of Concrete and Seismic Engineering

normal. A concrete cover that is not confined by transverse reinforcements cannot

stop this buckling, as it is ejected as soon as the shrinkage exceeds 0.35%. As a
consequence, buckling can only be averted by leaning longitudinal reinforcements
on transverse reinforcements that are correctly anchored into the concrete. The
spacing of the reinforcements determines the buckling length of the longitudinal
reinforcements, whilst their diameter determines their ability to counterbalance
reinforcement strain due to compression stress work.

That is the reason why both the French standards and Eurocode 8 impose both a
maximum spacing and a minimum diameter for transverse reinforcements. It is
worth bearing in mind that this spacing is equal to at least eight times the diameter
of the longitudinal reinforcement, and therefore imposes a maximum slenderness on
the latter reinforcement.

9.7. Conclusions

Design methods for reinforced concrete structures are based on experimental

results and analysis of structures subject to strains in the post-elastic field. Within
these contexts, the rotational ductility of plasticized critical areas appears to play a
particularly important role. Capacity design allows control over the location of
plasticized areas, as well as over the failure modes, which ensures good reliability in
predicted structure behavior despite the random nature of seismic stresses. The
design of both horizontal and vertical bracing structures emerges as the most
important point for ensuring acceptable building behaviors during an earthquake. It
is also determined by construction layouts and the choice of materials used.
Calculation plays a minor quantitative role in designing well-behaved buildings,
because at present, different stages in some calculations are based on unrealistic
approximations, but qualitatively, it remains an indispensable tool for identifying
relevant parameters, and understanding their collective behaviors in greater depth.

9.8. Bibliography

[BET 91] BETBEDER-MATIBET J., “Constructions parasismiques”, Traité Construction,

Techniques de l’Ingénieur, C 3290 and C 3291, 1991.
[BET 97] BETBEDER-MATIBET J., DOURY J.L., “Constructions parasismiques”, Traité
Construction, Techniques de l’Ingénieur, 3290, Paper C, 1997.
[BIS 02a] BISCH P., “Constructions parasismiques. Eurocode 8”, Traité Construction,
Techniques de l’Ingénieur, C 3292, 2002.
 Engineering Know-How: Lessons from Earthquakes and Rules for Seismic Design 367

[BIS 02b] BISCH P., COIN A., “The ‘CAMUS 2000’ research”, Proceedings of the XIIth
European Conference on Earthquake Engineering, London, UK, Paper N° 796, Oxford,
Elsevier, 2002.
[BIS 98] BISCH P., COIN A., “The CAMUS research”, Proceedings of the XIth European
Conference on Earthquake Engineering, Paris, France, T2 session, Rotterdam, The
Netherlands, Balkema, 1998.
[COL 00a] COLLECTIF, NF P 06 031-1: XP ENV 1998-1-1 “Eurocode 8: Conception et
dimensionnement des structures pour leur résistance aux séismes et document
d’application nationale”, Partie 1-1: Règles générales – Actions sismiques et
prescriptions générales pour les structures, AFNOR, 2000.
[COL 00b] COLLECTIF, NF P 06 031-2: XP ENV 1998-1-2 “Eurocode 8: Conception et
dimensionnement des structures pour leur résistance aux séismes et document
d’application nationale” , Partie 1-2: Règles générales – Règles générales pour les
bâtiments, AFNOR, 2000.
[COL 01] COLLECTIF, NF P 06 031-3: XP ENV 1998-1-3 “Eurocode 8: Conception et
dimensionnement des structures pour leur résistance aux séismes et document
d’application nationale”, Partie 1-3: Règles générales – Règles particulières aux
différents matériaux et éléments, AFNOR, 2001.
[COL 02] COLLECTIF, NF EN 1990–2002 “Eurocode 0: Bases de calcul pour les Eurocodes
structuraux”, French version, CEN, 2002.
[COL 03a] COLLECTIF, prEN 1992-1-1, “Eurocode 2: Design of concrete structures; part 1-
1: general rules and rules for buildings”, CEN, April 2003.
[COL 03b] COLLECTIF, prEN 1998–1 “Eurocode 8: Design of structures for earthquake
resistance; part 1: general rules, seismic actions and rules for buildings”, CEN, 30 October
2003.
[COL 91] COLLECTIF, CEB-FIP, MODEL CODE 1990 – Comité Euro-international du
Béton, Vol. 1 and 2, CEB, 1991.
[COL 95a] COLLECTIF, NF P 06 013 “Règles de construction parasismique”, Règles PS
applicables aux bâtiments, dites règles PS 92, AFNOR, 1995.
[COL 95b] COLLECTIF, NF P 06 014 “Règles de construction parasismique” . Construction
parasismique des maisons individuelles et des bâtiments assimilés, called PS-MI 89 rules
revised 92, AFNOR, 1995.
[COL 99] COLLECTIF, CEB-FIP, “Structural Concrete. Textbook on Behavior, Design and
Performance”, Updated Knowledge of the CEB/FIP Model Code 1990, Fédération
Internationale du Béton, Manual textbook, Vol. 1, 2, 3, FIB, 1999.
This page intentionally left blank
 List of Authors

Patrice BAILLY
ENSIB
Bourges
France

Pierre-Yves BARD
LCPC – LGIT
Grenoble
France

Philippe BISCH
IOSIS Group
Montreuil
France

Didier COMBESCURE
Formerly at CEA Saclay
Gif-sur-Yvette
France

Gérard GARY
Laboratoire de Mécanique des Solides
Ecole Polytechnique
Palaiseau
France

Jean-René GIBERT
Formerly at CEA Saclay
Gif-sur-Yvette
France
370 Dynamic Behavior of Concrete and Seismic Engineering

Nicolas ILE
Formerly at LGCIE
INSA
Villeurbanne
France

Jacky MAZARS
Laboratoire 3S-R
Institut Polytechnique
Grenoble
France

Alain MILLARD
DMT/SEMT
CEA Saclay
Gif-sur-Yvette
France

Patrick PAULTRE
Civil Engineering Department
University of Sherbrooke
Canada

Alain PECKER
Geodynamique et Structure
Bagneux
France

Pierre PEGON
ELSA
Joint Research Center
Ispra
Italy

Jean PROULX
Civil Engineering Department
University of Sherbrooke
Canada

Jean-Claude QUEVAL
DMT/SEMT
CEA Saclay
Gif-sur-Yvette
France
 List of Authors 371

Jean-Marie REYNOUARD
LGCIE
INSA
Villeurbanne
France

François TOUTLEMONDE
Laboratoire Central de Ponts et Chaussées
Paris
France
This page intentionally left blank
 Index

A, B E, F
Accelerogram, 98, 99, 102, 165, 186, Effective stress, 74, 86
196, 197, 258 Eigenmode, 194, 208
Accelerometer, 32, 96 - 98, 120, 121, ELSA laboratory, 168
123, 177, 194, 194, 202, 206, 209, Eurocode, 215, 254, 262, 264, 308,
211, 215, 216, 220, 221, 223 327, 336-340, 342, 347, 349, 351-
Alluvial soil, 126, 127 354, 357-361, 363-366
Arias intensity, 106, 107 Frequency
Attenuation, 104, 112, 113 excitation, 205, 207
Beam loading, 42, 195
multi-fiber, 237, 250 resonance, 186, 188, 189, 195,
multi-layer, 236, 254, 269 196, 226, 276, 278, 316
Brazilian test, 8, 23
G, H, I
C, D Global model/global modeling, 234,
Cap surface, 73, 80 239
Characteristic length, 76 Gurson’s criterion, 83, 84
Compaction, 7, 13, 18, 32, 56-60, 69- Hugoniot curve, 17
71, 73, 76-78, 80-85, 87, 90, 91 Hugoniot’s elastic limit (HEL), 63,
Confinement, 3, 6-8, 10, 13, 17, 18, 64,
20, 25, 27, 28, 29, 38, 41-43, 47, Hydrophone, 202, 204, 223, 224
61, 80, 91, 240, 331, 357-359, Hysteresis loop, 128, 130, 135, 138,
361, 365 143, 269
Consistent boundaries, 156, 157 Impedance, 7, 8, 20, 117, 118, 153,
Crater formation, 58 154, 157, 158, 160
Diffraction, 108, 114, 116, 154, 155 function, 143, 154, 157
Distinct element, 91 Interaction
Ductility class, 352, 353, 355, 358 inertial, 150, 151
Duhamel’s integral, 101 kinematic, 151, 153-155, 157
374 Dynamic Behavior of Concrete and Seismic Engineering

L, M, N Seismology, 95, 96, 103, 112, 120

Lagrange diagram, 69, 144, 145 Semi-local simplified models, 234
Limit state, 39, 329, 339, 340 Servo-controller, 185, 186
Liquefaction, 95, 113, 141 Shaking table, 165, 166, 168, 170,
Local model, 253, 270 184, 187, 188, 189, 190, 193-196,
Lode’s angle, 73 233, 234, 250, 254, 261, 279, 301
Modal damping, 206, 207, 209, 210, Shock
212, 225, 228, 278, 294 polar curve, 17, 66-68
Modal method, 273, 274, 276, 277, tube, 14-16, 57
279, 292, 293, 297, 308 Similarity rules, 32, 255
Non-drained behavior, 132 Site effect, 96, 103, 107, 113, 117,
119, 274, 286
P,Q, R Soil structure interaction, 125, 126,
p-delta effect, 249 128, 143, 144, 147, 148, 149, 151,
Peak ground acceleration, 98, 253, 154, 157-159, 196
258 Spectral intensity, 106
Perforation, 57, 60 Spectral signature, 109, 117
Perzyna’s model, 76 Spectrum
Plate-plate test, 18, 57 floor, 279, 297, 301, 345
Power spectral density, 207, 208, 281 Fourier, 104, 105, 107, 111, 112,
Process 117
random, 279, 280, 283-285, 292, response, 100, 102, 105, 106,
293, 298, 301, 302, 311, 319, 320 220-222, 274, 277
stationary, 12, 281, 282, 286, 297, Stefan effect, 34
302 Strain rate, 4, 12, 16, 18, 24, 27, 28,
Pseudo-dynamic method, 166, 167, 30, 39, 40, 55, 56, 61, 62, 71, 72,
170, 178, 179, 234, 250 76, 85, 91, 135
Push-over, 102, 348, 350 Sub-structuration, 159, 169, 170, 173,
P-Į model, 75 174, 176, 180-182, 184
Quadratic combination, 278, 279,
293-296, T, V, W
Rate effect, 12, 13, 34-38, 40-43, 45, Thrust coefficient, 128
46, 71, 72, 166 Timoshenko beam, 236, 237, 249
Rayleigh line, 70 Tunneling, 58, 60
Resonant column test, 134, 140 Wave
Rayleigh, 155
S surface, 108, 111, 114, 116, 134,
Scabbing, 55-58, 68 157
test, 3, 23, 24 volume, 111, 116, 150, 154