Contact Mechanics

Contact Mechanics
SOLID MECHANICS AND ITS APPLICATIONS

Volume 103
Series Editor: G.M.L. GLADWELL

Department of Civil Engineering
University of Waterloo
Waterloo, Ontario, Canada N2L 3GI
Aims and Scope of the Series

The fundamental questions arising in mechanics are: Why?, How?, and How much?
The aim of this series is to provide lucid accounts written by authoritative researchers
giving vision and insight in answering these questions on the subject of mechanics as it
relates to solids.
The scope of the series covers the entire spectrum of solid mechanics. Thus it includes
the foundation of mechanics; variational formulations; computational mechanics;
statics, kinematics and dynamics of rigid and elastic bodies: vibrations of solids and
structures; dynamical systems and chaos; the theories of elasticity, plasticity and
viscoelasticity; composite materials; rods, beams, shells and membranes; structural
control and stability; soils, rocks and geomechanics; fracture; tribology; experimental
mechanics; biomechanics and machine design.
The median level of presentation is the first year graduate student. Some texts are mono-
graphs defining the current state of the field; others are accessible to final year under-
graduates; but essentially the emphasis is on readability and clarity.
For a list of related mechanics titles, see final pages.

Contact Mechanics
Proceedings of the 3rd Contact Mechanics International
Symposium, Praia da Consola~ao, Peniche, Portugal,
17-21 June 2001
Edited by
l.A.C. MARTINS
lnstituto Superior Tecnico,
Departamento de Engenharia Civil, Lisbon, Portugal
and
MANUEL D.P. MONTEIRO MARQUES

CMAF and Faculdade de Ciencias,
Universidade de Lisboa, Lisbon, Portugal
SPRINGER-SCIENCE+BUSINESS MEDIA, B.Y.

A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-90-481-6099-0 ISBN 978-94-017-1154-8 (eBook)

DOI 10.1007/978-94-017-1152-8
Printed on acid-free paper
All Rights Reserved

© 2002 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 2002
Softcover reprint of the hardcover 1st edition 2002
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming, recording
or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
CONTENTS
Preface........................................................................................... ix
Third Contact Mechanics International Symposium..... .... xi
1. Dynamics and impact

Numerical dynamics of granular materials ..................................................... .
Jean Jacques Moreau
Measurements of impacts with friction... ............. ....... .... ........ .......... ............. 17
Friedrich Pfeiffer
Formulation and well-posedness of unilateral multibody dynamics......................... 25

Patrick Ballard
Collisions in systems made of rigid bodies... .... ........... ...... .... ............ ......... ..... 33
Eric Dimnet
Impacts with global dissipation index at reentrant corners................................. ... 45

Christoph Glocker
Understanding impact through continuous medium vibrations.............................. 53

Laetitia Paoli, Michelle Schatzman
Solving rocking block problems with multiple impacts.............................. .......... 65

Cuneyt YUmaz, YUdirim Hurmuzlu
A controllability criterion for linear juggling mechanical systems........ ....... ......... .... 75
Bernard Brogliato, Mongi Mabrouk, Arturo Zavala Rio
On integrating stiffmultibody dynamics with contact and friction........................... 83

Mihai Anitescu, Florian A. Potra
Analysis of systems with multiple frictional contacts....................................... ... 91

A.P. Ivanov
vi
2. Instabilities, oscillations and waves

Initiation of friction instability on a plane fault system................................. ....... 99
loan R. lonescu, Michel Campilio, Cristian Dascalu, Pascal Favreau, Christophe
Voisin
An example of stick-slip waves. . .... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... . . . . . . . . . . . 109

Franck Moirot, Quoc-Son Nguyen
Longitudinal waves in elastic rods with discontinuous cross sections.. .. .. .. .. ...... .. .. .. . 117
Werner Schiehlen, Bin Hu, Peter Eberhard
The influence of contact properties on friction-induced brake vibrations................... 125

M. Rudolph, K. Popp
Analysis of eigenvalue problems modeIling friction: sufficient conditions of non-

uniqueness for the elastic equilibrium......... ...... .... ........ ..... ................. ....... .... 133
Riad Hassani, Patrick Hild, loan lonescu
3. Contact models, results and applications

Adhesion of viscoelastic spherical solids.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .... 141
K. L. Johnson, J. A. Greenwood
A model of adhesion added to contact with friction........................................... 161

Christian Talon, Alain Curnier
Analysis of a "walking" punch.................................................................... 169

D.A. Hills, A. Sackfield
Phenomenological model of friction accounting for subsurface plastic deformation in

metal forming....................................................................................... 179
Stanislaw Stupkiewicz, Zenon Mroz
Thermoelastic instabilities in automotive disc brakes - finite element analysis and

experimental verification........................................................................... 187
Yun-Bo Yi, J.R. Barber, D.L Hartsock
Friction coefficient evolution during wear tests related to infrared emission............... 203
Nadine Stalin, Jean Christophe Eytard
A contact formulation for electrical and mechanical resistance. ............................ 211

G. Zavarise, D. Boso, B.A. Schrefler
The real area of contact - a combination of experimental and numerical approaches.... 219
Frank Bucher, Rob S. Dwyer-Joyce
vii
Prediction of wheel wear for rail vehicles - methodology and verification.. . . . . . . . . . . . . .. 229
Tomas Jendel. Mats Berg
Multi-body modeling of paper calendering unit by contact dynamics fonnulation.... .... 237
Erno Keskinen. Sirpa Launis. Juha-Matti Kivinen
4. Mathematical analysis
Existence and uniqueness for quasi static contact problems with friction................... 245
Lars-Erik Andersson, Anders Klarbring
Augmented lagrangian methods for a class of non convex contact problems in struc-
tural mechanics...................................................................................... 261
W.R. Bielski, A. Galka, J.J. Telega
Approximation of quasi static Signorini problems with local friction by a mixed

method................................................................................................ 269
Marius Cocou, Remi Rocca
A beam in adhesive contact........................................................................ 277

W. Han, K.L. Kuttler, M. Shitlor, M. Sofonea
Coulomb fluid-solid interface law in lubrication............................................... 285

Guy Bayada, Mahdi Boukrouche
Elastodynamic friction problem with a "surface inertia" perturbation....................... 293

Jean-Claude Paumier, Yves Renard
Solvability of thenno-viscoelastic contact problems with Coulomb friction and nonlin-

ear heat conductivity.. .. .. .. .. . .. .. .. .. .. .. .. .. . .. .. .. . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ...... .. .. ... 301
Christof Eek, Jir, Jarusek
A simplified model of impact..................................................................... 309

Jeongho Ahn, David E. Stewart
Convergence for a time discretization of dynamic contact problems with friction........ 317
E. Pratt, J.-M. Ricaud
Solution methods for structural optimization in contact rod problems...................... 325

Isabel N. Figueiredo, Joaquim J. Judice, Silverio S. Rosa
5. Numerical methods
Unilateral contact, friction and adhesion: 3D cracks in composite materials............... 333
Michel Raous, Yann Monerie
The unilateral frictional contact of a piezoelectric body with a rigid support.............. 347
Paolo Bisegna, Frederic Lebon, Franco Maceri
viii
Contact between 3D beams with rectangular cross-sections.. ............................. ... 355

Przemyslaw Litewka, Peter Wriggers
Algorithms for thermoelastic wear problems................................................ .... 363

Peter lreman, Anders Klarbring, Niclas Stromberg
Convergence studies for 3D smooth frictional contact elements based on the quartic
Bezier surfaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... . . 371
Lovre Krstulovic-Opara, Peter Wriggers
Multicontact structures and parallel computing........ ................ ......... ....... ......... 379
P. Alart, M. Barboteu
A new approach to the 2D transient rolling contact problem............................. .... 387

J.A. Gonzalez. R. Abascal
Numerical approximation of the elastic-viscoplastic contact problem using noncoin-

ciding finite element meshes.................................................. .................. 395
Jose R. Fernandez-Garcia, Juan Viano, Patrick Hild
Contact in the Arlequin framework....... . . . . . . . . . . . . . . . . . .. .. . . . . . . . . . . . . . . . . . . . . . . . . . ..... . . . . . 403

Hachmi Ben Dhia, Malek Zarroug
Appendix A. Scientific program of the Third Contact

Mechanics International Symposium..... ........ 411
Appendix B. List of participants in the Third Contact

Mechanics International Symposium............. 419
PREFACE
This volume contains 44 papers presented at the Third Contact

Mechanics International Symposium (CMIS 2001) held in Praia da
Consola9ao, Peniche (portugal), June 17-21,2001.
This Symposium was the direct continuation of the first two CMIS held
in Lausanne (1992) and in Carry-Le-Rouet (1994). Other related meetings,
in what concerns scientific topics and participants, took place in the nineties
at La Grande Motte (1990), Vadstena (1996), Ferrara (1997), Munich (1998)
and Grenoble (1999).
The Symposium aimed at gathering researchers with interests in a wide
range of topics in theoretical, computational and experimental contact
mechanics. The call for papers mentioned topics in tribology, mathematical
formulations and analysis, numerical methods in non-smooth mechanics,
impact problems, instabilities and technological problems.
The total number of participants was 102, from Universities and
Research Institutes of 19 countries. The Scientific Committee reviewed 102
submitted abstracts, and the final program consisted of 6 main lectures, 43
oral communications and 36 poster presentations (see Appendix A).
The papers in this book correspond to almost all the main lectures and
oral communications, and they are assembled in 5 chapters:
• Dynamics and Impact
• Instabilities, Oscillations and Waves
• Contact Models, Results and Applications
• Mathematical Analysis
• Numerical Methods.
We thank all the authors for their valuable contributions to this volume.
We are indebted to the members of the Scientific Committee for their help in
refereeing the submitted abstracts and manuscripts. We also thank the Series
editor, Prof. Graham Gladwell, for his assistance in the revision process.
We gratefully acknowledge the support of our research centers, the
lCIST (lnstituto Superior Tecnico) and the CMAF (Universidade de Lisboa),
and the support of the institutions listed in page viii. In particular, we thank
ix
x
the Funda9ao Calouste Gulbenkian for partially funding the publication of
this volume.
We address special heartfelt thanks to our colleagues of the Organizing
Committee, Eduardo Borges Pires, Fernando Simoes, Ant6nio Pinto da
Costa and Anca-Maria Toader, for their great effort, efficiency and
enthusiasm in the organization of the Symposium.
Joao Martins and Manuel Monteiro Marques
Lisbon, February 2002

THIRD CONTACT MECHANICS
INTERNATIONAL SYMPOSIUM
CMIS 2001
June 17-21, 2001
Praia da Consola~iio, Peniche, PORTUGAL
1. Organizing Committee
J.A.C. Martins, Dep. Eng. Civil and ICIST, Instituto Superior Tecnico, Lisboa
M.D.P. Monteiro Marques, Fac. Ciencias da Universidade de Lisboa and CMAF, Lisboa
E.B. Pires, Dep. Eng. Civil and ICIST, Instituto Superior Tecnico, Lisboa
F.M.F. Simoes, Dep. Eng. Civil and ICIST, Instituto Superior Tecnico, Lisboa
A. Pinto da Costa, Dep. Eng. Civil and ICIST, Instituto Superior Tecnico, Lisboa
A.-M. Toader, Faculdade de Ciencias da Universidade de Lisboa and CMAF, Lisboa
2. Scientific Committee
J.R. Barber, Michigan, USA A. Klarbring, Linkoping, Sweden
S. Cescotto, Liege, Belgium F. Maceri, Rome, Italy
A. Curnier, Lausanne, Switzerland J.J. Moreau, Montpellier, France
G. Del Piero, Ferrara, Italy F. Pfeiffer, Miinchen, Germany
M. Fremond, Champs sur Marne, France M. Raous, Marseille, France
C. Glocker, Zurich, Switzerland J.R. Rice, Cambridge, USA
J. Haslinger, Praha, Czech Republic D. Stewart, Iowa, USA
M. Jean, Marseille, France W.J. Stronge, Cambridge, United Kingdom
K.L. Johnson, Cambridge, United Kingdom P. Wriggers, Hannover, Germany
J.J. Kalker, Delft, The Netherlands
xi
xii
3. Sponsors and Acknowledgments
The Organizing Committee of CMIS 2001 gratefully acknowledges the

following institutions for their fmancial or material support:
• Instituto de Engenharia de Estruturas, Territorio e Constru'Yao (ICIST),

• Centro de Matematica e Aplica'Yoes Fundamentais (CMAF),
• Reitoria da Universidade Tecnica de Lisboa,
• Instituto Superior Tecnico,
• Funda'Yao para a Ciencia e a Tecnologia (FCT),
• Funda'Yao Calouste Gulbenkian,
• Funda'Yao Luso-Americana para 0 Desenvolvimento (FLAD),
• Camara Municipal de Peniche,
• Regiao de Turismo do Oeste,
• Camara Municipal de Obidos,
• Banco BPI.
The Organizing Committee of CMIS 2001 thanks Prof. Joao Bento and
Prof. Eduardo Pereira of the ICIST, for having kindly provided some
database software, and Dr. Cristian Barbarosie of the CMAF, for his help
during the meeting.
The Organizing Committee of CMIS 2001 also thanks the staffs of
Camara Municipal de Peniche, Hotel Atlantico Golfe and ICIST, for their
efficient collaboration.
NUMERICAL DYNAMICS
OF GRANULAR MATERIALS
Jean Jacques MOREAU

Laboratoire de Mecanique et Genie Civil
UniversiU Montpellier II/CNRS, Montpellier, Prance
Abstract Contact Dynamics is a numerical method, suitable for computing the

dynamical motion of large collections of rigid bodies, with Coulomb fric-
tion taken into account in the event of contact. The principles of the
method are sketched, in particular the way possible collisions or other
nonsmooth features of the evolution are handled. As an example of
application to granular dynamics, the construction of dry deposits and
banks is simulated, in order to investigate their microstructure: force
chains, geometrical anisotropy, Cauchy stress and some unexpected fea-
tures of force transmission.
1. Introduction
1.1. Addressed problems
In diverse domains, computational methods are needed for the statics
or the dynamics of collections of rigid bodies subject to the constraints
of non-interpenetrability, with friction taken into account in the event of
contact. This includes the dynamics of machines, in particular robots,
the dynamics of masonry works submitted to transient actions (earth-
quakes, gusts of wind or impacts), animated computer graphics and
numerical simulation in granular mechanics. Possibly, some variables
are added in order to also account for a certain deformability of the
bodies without essentially changing the computational strategies; see
Jean (1999). Of course, for the handling of non-interpenetrability and
friction, much may be learned from the rich literature devoted to the
treatment of contact between deformable media, discretized through fi-
nite elements; see e.g. Alart and Curnier (1991), Chabrand et al. (1998),
Cristensen et al. (1998), Curnier (1984), Vola et al. (1998).
The main part of the lecture at the Symposium consisted of projected
animations, intended to show what sort of knowledge may be gained
J.A.C. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 1-16.
© 2002 Kluwer Academic Publishers.
2
from the numerical simulation of granular processes. The present paper
begins with an overview of the 'Contact Dynamics' time-stepping tech-
nique used in these simulations. More detailed accounts of the method,
originated in Moreau (1988b), may be found in Moreau (2000), Moreau
(2001). Finally, some drawings created in the course of numerical exper-
iments are shown and commented.
1.2. Analytical setting

The analytical description of a multibody system begins with the
choice of a parametrization - at least a local one - of the set of possible
configurations through an element q := (ql, ... ,qn) of Rn. In terms of
this parametrization, the geometric effect of the non-interpenetration of
the system members, or of their confinement by external obstacles with
prescribed motion, is assumed expressed by a finite set of inequalities
ga(t,q)?O, aE{1, ... ,~}, (1)
where the functions ga(t, q) are measures of the gaps between bodies,
counted as negative if the configuration q at time t involves overlap.
The event of equality in (1) for some value of a, corresponds to contact
between a member B of the system and a body B' which may be another
member of the system or an external obstacle.
Non-interpenetrability is a unilateral constraint; as always in Mechan-
ics, its description cannot reduce to merely specifying the corresponding
set of feasible states (such, in contrast, is the meaning of the word 'con-
straint' in Optimization and similar contexts). Some information must
be added about the mechanical process through which the geometrical
restriction is enforced. For instance the use of servomechanisms to se-
cure (1) should result in other motions than those investigated in what
follows.
Generically, let us conceive a contact law as a relationship involving
the contact force n a experienced by B from B' at the contact point Ma
and the local velocity Ua of B relative to B', relationship which a priori
depends on time and on the configuration attained by the system, say
(2)
For numerical and analytical purposes, the definitions of the elements
n
Ua and a of E3 are extended, at least locally, to configurations with
nonzero ga(t, q). Saying that (2) models a contact phenomenon involves
that this law yields n a = 0 whenever ga(t, q) > o. Of course, more
complex situations may be addressed, for instance with contact actions
not reducing to single forces.
3
The system motion on a time-interval I, with origin to, is described

by a function t H q(t) of I into Rn. Classical Kinematics requires
of this function to be locally absolutely continuous, so that its deriva-
tive q' makes an element of Ctoc(I,Rn ) from which q may be retrieved
through Lebesgue integration. The components of q'(t) are commonly
referred to as the velocity components of the system with regard to the
parametrization in use.
When dealing with stereodynamical problems one may also charac-
terize the system velocity by some otherwise defined set of components,
making an element u of R n whose dependence on t may generically be
called the velocity function. Commonly, one attaches to each rigid body
an orthonormal frame emanating from its mass center and principal with
regard to body inertia. The components relative to these axes of the spin
vector of the body are entered as elements of the column u(t). Retriev-
ing from the function u the evolution of q then rests on the integration
of elementary kinematical relationships. The moments, about the same
axes, of the forces experienced by the body are concomitantly entered
at the corresponding places in the row of the covariant components of
the forces acting on the system.
The classical framework of 'smooth' dynamics requires of u to be in
its turn locally absolutely continuous on the interval I. By Lagrange's
method or any stereo dynamical practice, one writes the dynamical equa-
tions as the following equality of elements of R n
A(t,q)~~ = F(t,q,u) + ~ro, (3)
to hold for almost every t. Here A denotes in general the n x n inertia

matrix, diagonal and constant in (t, q) if u has been constructed accord-
ing to the technique of mass centers and principal axes mentioned above.
Expression F comprises certain standard terms, sometimes referred to
as 'centrifugal' and 'gyroscopic', and also the covariant components of
possible applied forces, supposed given as functions of time, the position
of the system and its velocity. The elements r O , a E {l, 2, ... ,K}, are
made of the covariant components of the respective contact forces.
In order to take the contact laws (2) into account, one has to connect
the elements r O and u of R n with the vectors R O and Uo of E3. By
the kinematical analysis of the way the parametrization (q) has been
constructed, one obtains an expression of the form
(4)
where Go : R n ---+ E3 denotes a linear mapping, depending on t and q.
In case the contact labelled a involves an external obstacle, the term
4
Wa E E 3 , a known function of t and q, accounts for the given motion of

this obstacle. No attention is paid at this stage to the imagined motion
preserving contact or not.
By applying the standard virtual power machinery, one correlatively
obtains the following expression for the covariant components of the
contact forces acting on the system at Ma (a pair of opposite forces if
8' is a member of the system, a single force if it is an external obstacle)
(5)
with G~ : E3 -+ R n denoting the transpose of Ga.
As said before, the elements associated with each 0, such as G a here or
a common normal unit na, directed toward B, invoked in the sequel, are
conventionally extended in a smooth way to neighbouring configurations
without requiring ga = O.
1.3. Approximation
Relationships (1) to (5) convey all the retained information about the
mechanical system. For the numerical treatment of evolution problems,
it is easy to conceive time-stepping procedures of approximation to the
differential equation (3), under the kinematical relationship connecting
q and u. Through (4) and (5), the contact laws (2) are entered into the
management of each time-step. The difficulty comes from the necessity
of taking the non-interpenetration inequalities (1) also into account.
In the majority of the techniques described in literature and applied
in commercial software packages, this is achieved through the classical
trick of penalizing inequalities. Mechanically, this means that the strict
non-interpenetrability of a pair of bodies is approximately replaced by
elastic repulsion forces which become effective when the bodies come
close to each other. Such a procedure amounts to approximate the joint
conditions (1) and (2) by relationships connecting the same variables,
but smoother. In principle, the resulting system of differential equations
should be sufficiently regular for standard time-stepping procedures to
apply. Similar interaction laws are used in the computer simulation of
molecular motions, hence the name of Molecular Dynamics (abbr.: MD)
commonly given to this approach.
The drawback is that the need of precision requires of the artificial
repulsion laws to be very stiff. Numerical stability in integration then
calls for the use of very short time-steps and frequently also for the
introduction of some artificial damping or artificial increase of inertia.
When treating dynamical applications, the effect of such alterations of
the mechanical data may blur the picture. Significant simulations of
loose, collisional, flows of granular materials have been obtained in that
5
way, but when dense collections of bodies are concerned (masonry works
or compact granulates) the method is mainly applied to quasi-static
evolutions in which only a succession of equilibrium states is looked for.
For these reasons, other numerical strategies may be preferred in uni-
lateral multibody dynamics.
In the line of the traditional stereodynamic practice are the approaches
qualified as event-driven (abbr: ED). Starting from a state in which
certain contacts are in effect, one attempts to calculate the subsequent
motion under the provisional assumption that these contacts persist and
that their status - sticking or sliding - in regard to Coulomb law is in-
variant. Calculation is then the same as with bilateral constraints, but
requires to watch the evolution of some indicators. If, for instance, the
normal components of some contact forces take directions incompatible
with unilaterality, one concludes that the subsequent motion has to be
calculated otherwise. Finding the further status of contacts without hav-
ing to try all combinations is a nontrivial matter. It is usually reduced
to Linear Complementarity Problems; see Moreau (1966), Pfeiffer and
Glocker (1996), Vola et al. (1998), Abadie (2000).
The ED strategy becomes unpractical if the number of contacts in
presence is too large. Some time-stepping integration procedures with
preselected time-intervals have been developed instead; see Pang and
Stewart (1999), Stewart (1998). The following one belongs to this class.
2. The 'Contact Dynamics' approach

2.1. Handling non-interpenetration in terms of
velocity
For every label Ct, it has been agreed to extend, at least locally, the
definition of contact descriptors such as the normal unit n a , to configu-
rations with nonzero gao Then put
IC (t ).= { {V E E3 I V.no ~ O} if go(t, q) ~ 0
a ,q. E3 otherwise.
If go (t, q) ~ 0, this is the set of the values of U:, the contact right-
velocity, which are kinematically compatible with non-interpenetration.
The elements for the proof of the following may be found in Moreau
(1999):
Lemma. Let a motion be defined by a locally integrable velocity function
u : I --t Rn. If the corresponding expression of Uo : I --t E3 , as it
results from (4), satisfies Uo E 1C0(t, q) for almost every t and if the
non-interpenetration inequality (1) holds at the initial instant to, then it
holds throughout I.
6
The decisive move of the Contact Dynamics strategy consists in com-
plementing the contact laws (2) so as to obtain relationships in the same
variables, containing the same stipulations as before and additionally
the following ones: • in all cases UrJ/ E Ko, • if Uo E interior Ko, then
RO = O. In other words, these relationships secure the implications
go(t, q) ::; 0 :::} nO. Uo ~ 0 (6)

nO.Uo > 0 :::} RO = O. (7)
We propose to say that a package of information concerning the possi-

ble contact labelled a, if it possesses these properties, is a contact law of
prospective type. The underlying idea is that such a law does not prop-
erly govern the values ofUo and RO at the actual instant, but their limits
on the right of this instant, assumed to exist. In fact, if nO. Uo > 0, the
concerned instant is followed by a contactless time-interval. Since RO
must vanish over this interval, the same should be true for its right-limit.
Once all the contact laws (2) have been complemented in that way,
the Lemma shows that condition (1) becomes redundant, provided it is
satisfied at the initial instant. It then becomes easy to imagine time-
stepping procedures for the approximation of the evolution problem.
The very structure of the Coulomb model of friction induces to orga-
nize the step computation as an implicit scheme, at least with regard to
the velocity function u, i.e. it is the unknown value of this function at
the step-end which, as an approximation, is entered into the law. A con-
tact law of the prospective type is adequate here, since step-end values
mimick right-side limits. The possibility of contacts to get loose at every
instant is then managed automatically, without needing any analysis of
complementarity conditions.
Each time-step makes a nonlinear problem whose computational cost
grows rapidly with the number of contacts in presence. As a compen-
sation for this cost, the method accepts much larger step-lengths than
those required by the regularization techniques referred to in 1.3. Usu-
ally an iteration procedure a la Gauss-Seidel is applied. It amounts to
treat cyclically a sequence of dynamical problems, each of them involv-
ing a single contact. In dense collections of bodies with slowly evolving
contact list, computational efficiency may be greatly enhanced by tak-
ing as starting guess in iterations the contact forces calculated at the
antecedent time-step, whenever the same contact was already in effect.
The Contact Dynamics strategy may also be applied with other solvers
used at each time-step; see Johansson and Klarbring (2000).
A single computation step is needed to check whether a given position
of the system is that of a possible equilibrium. One just has to launch
7
iterations with zero initial velocity; if the step-end velocity is found zero,
the values obtained for the contact forces agree with equilibrium.
2.2. Nonsmoothness in time and collisions

The sudden occurrence at some instant tc of a new contact, i.e. a
collision, is expected to generate a velocity jump, the dynamics of which
cannot be governed by the differential equation (3). Even in the absence
of events of this sort, dry friction at a contact point has long been known,
in some cases, to stop the existence of a smooth solution at a catastrophic
instant, on the left of which the contact forces, as well as the derivative
u', may become unbounded; see Genot and Brogliato (1999). This is a
dynamic analogue to the locking situations familiarly met in the statics
of frictional systems. Around year 1900, such a lack of solution for
an apparently well set problem seemed inadmissible to P. Painleve and
induced him to question the very concept of a contact force. Hence the
misleading denomination of 'Painleve's paradox' which has been given
to the observation. Today, one is accustomed to see models reaching the
limits of their validity domain. Such is the case for smooth dynamics at
a catastrophic instant of any sort, leading to enlarge the function space
where u is looked for; see Moreau (1988b), Stewart (1998).
A readily available mathematically framework is that of Rn-valued
functions of locally bounded variation on the time-interval I; notation
Ibv(I, Rn), cf. Moreau (1988a). With every u in this space, an Rn_
valued measure on J is classically associated that we shall denote by
du: this is the differential measure (or Stieltjes measure) of u. In the
smooth case, this measure admits the derivative u' as density function
relative to dt (the Lebesgue measure on J). A natural generalization of
(3) therefore is
A{t,q)du = F{t,q,u)dt + Lds a , (8)

a
where the Rn-valued measures ds a constitute the 'components', relative
to the parametrization in use, of the respective contact impulses dS a
which are E 3-valued measures. This is a measure differential equation.
Concerning the theoretical foundation of such an extension of Classical
Dynamics, one may refer to Moreau (1989).
On a time interval of smooth motion dS a = na dt, while at the instant
tc of a collision, each measure dS a is expected to present an atom whose
value is nothing else than the percussion vector of the traditional the-
ory of shocks, say pa. The latter is a priori unknown and the question
arises whether it may be involved in some phenomenological relationship
playing at instant tc a role similar to that of (2) on intervals of smooth
8
motion. The laws of restitution of Newton, Poisson or others amount to

pragmatic tricks of this sort, but it is now clear that the empirical coef-
ficients involved in such laws cannot in general be identified a priori; cf.
Stoianovici and Hurmuzlu (1996), Ivanov (1997). Apart the simple case
of the collision of two, otherwise free, nearly spherical objects, the out-
come of a collision can be predicted with some precision only through a
detailed analysis of the contact interaction which actually involves a cer-
tain amount of deformation of the concerned bodies. When so studied,
the process takes place on a nonzero time interval, contact forces and
accelerations exist in the proper sense, leaving smooth mechanics valid.
Another classical approach consists of a multiple scaling technique, using
some microtime variable in the analysis of the process.
A critical issue, when restitution laws are used, is the energy balance
of the instant process. A theoretical framework securing the dissipativ-
ity required by the Second Law of Thermodynamics may be found in
Fremond (1995).
2.3. The CD handling of nonsmooth evolutions

Formally, at each time-step of a CD computation scheme, say [til tf]
('i' as in initial, 'f' as in final), tf = tj + h, the algorithm is ready to
face collisions. Throughout this interval, the elements A, Go, W o are
approximated by constants, namely the value they take at the 'midpoint'
(tm = tj + h/2, qm = qj + h ui/2) and the element F by the value it takes
at (tm, qm, ud· It is also at (tm, qm) that inequalities (1) are checked in
order to determine the set J m of the labels a to be treated as effective.
By integrating both members of (3) over the interval, or both members
of (8) as well, one obtains
Uf = Uj + hA- I F + A-I Lpo, (9)

°
where the quantities po are the covariant components of the contact total
impulses, i.e. the integrals of the dso. The relationship that (2), (4),
(5) establish between rO and u, does not commute in general with time
integration because contact laws are nonlinear. When the motion one
intends to approximate is smooth, the variation of rO and U over the
time-step may be small enough for the decision of connecting po with
the final velocity Uf by the said relationship to merely reflect the choice
made in 2.1 of a time-stepping strategy of the implicit type.
But if some collisions occur in [tj, tf], i.e. J m includes labels it did not
contain at the antecedent step, large variations of U are expected. The
designation of any value of U to be connected with pO, constitutes an
9
assertion about the collisional process, the phenomenological quality of
which has to be checked in regard to physical circumstances.
To Uf corresponds the final value UO!f of the local velocity. The above
choice amounts to connect with it, through the admitted contact law,
the total local impulse on [ti' tf]. If this interval contains a single col-
lisional instant t e , the initial and final values of UO! may be viewed as
approximants of the left and right limits U;;(te) and Ut(te).
Therefore, the collisions met in the course of the above computation
happen to be treated according to the following rule: for each contact
a present at instant t e , the contact percussion pO! is related, through the
admitted contact law, to the right limit ut(te).
Contact laws here are assumed of the prospective type. In view of (6)
and (7), this yields the implication pO! f:. 0 ~ nO!. UO! = 0, expressing
that the collision is soft, i.e. completely inelastic. But the possibility
nO! . UO! > 0 with pO! = 0 is left open. This model of soft collision
in multicontact systems thus improves on the mere assumption of zero
Newton restitution coefficient, because all the contacts present at instant
te are involved together in the dynamics.
This time-stepping scheme also produces plausible results if a fric-
tional catastrophe is met; see Moreau (1988b).
An extension of the preceding enables CD algorithms to produce
bouncy collisions: it consists in connecting, through the admitted con-
tact law, the total local impulse with some weighted mean of UO!i and
UO!f; see Moreau (1998b), Moreau (2000), Moreau (2001). The choice of
the weighting coefficients (possibly distinct for the normal and tangen-
tial components of velocities) directly determines some restitution rules,
improving as before on those of Newton.
Anyway, such laws of restitution do not elude the criticism made in
2.2. Their advantage is only that their implementation in a CD algo-
rithm adds nothing to the computation cost.
3. Numerical simulation in Granular Mechanics

3.1. What can be expected from computation
Granular mechanics has long been a topic in Civil Engineering as a
part of Soil Mechanics. For a certain number of years, it has also at-
tracted the attention of another public, coming from the Condensed Mat-
ter and Disordered Systems community. It is expected that the forms of
statistical reasoning used in the latter domain could provide some insight
into the many intriguing features of the granular behaviour. On this pur-
pose, the model consisting of rigid grains which interact only through
10
contact and Coulomb friction, without any adhesive effect, makes a suf-
ficiently rich concept for investigating some fundamental questions.
For Civil Engineers, dry sand , to which the above model is relevant,
is only a special instance of soil.
In the numerical treatment of most engineering problems, the soil
material s viewed as a continuous medium, for which some empirical
constitutive law should be available. Finite Element software packages
are then applied to the corresponding boundary value problem. The
difficulty lies in the determination of the constitutive law which above all
has to be tested against the experimental information available about the
concerned soil. Engineering needs may be met through this approach,
but no insight into the intimate granular behaviour is gained.
While the abbreviation FEM (for Finite Element Method) is used to
refer to the above strategy, the representation of a granular sample as
a collection of solids is called a Distinct Element Method (abbr. DEM,
sometimes also read as Discrete Element Method). Numerical simulation
may be conducted thanks to the methods referred to in the foregoing.
Clearly the boundary problems of Civil Engineering cannot directly be
handled that way since the number of grains involved in real soils exceeds
the possibility of computers. The availability of scaling rules, which
would allow one to approximate these boundary problems through DEM
with oversized grains and possibly with more elaborate interaction laws
than Coulomb contact, is still an object of investigation; cf. Nouguier
et al. (2000).
3.2. Force chains
Figure 1. A two-dimensional model of railway ballast.
It is known from experiments with assemblies of photoelastic grains

that the transmission of forces in the bulk of a granular material is
11
concentrated in chains of grains more loaded than the surrounding ones.

A two-dimensional model of railway track is shown on Figure 1. In view
of the elastic deformation of the rail, the force exerted by a passing wheel
is distributed between the closest supporting concrete blocks, in relation
with their distances. The blocks, as well as the ballast grains, are drawn
with levels of gray corresponding to their respective loads, i.e. the sum
of the normal components of the experienced contact forces. This makes
visible that forces are transmitted to the ground across the ballast layer
along some preferred chains of grains.
3.3. Deposition anisotropy

The response of a grain collection to some mechanical action strongly
depends on the way it has been prepared or, more generally, on its past
evolution. Some message from history is therefore written in the cur-
rent state of the granulate. As a part of this message the statistical
distribution of the directions of the normals at the grain to grain con-
tacts is commonly viewed as a descriptor of a possible microstructural
anisotropy. Here is an intriguing example.
A usual way of preparing granulates consists of producing a random
rain of grains under gravity. If this rain is received on a fixed horizontal
ground, a certain statistical anisotropy of the collected granular mass is
expected, since the vertical direction plays a special role in the deposition
process. The surprise was to discover experimentally that the most
frequent directions of contact normals are not vertical nor horizontal but
at some angle on both sides of the vertical. The physical experiment
was performed with the model of two-dimensional granulate called a
Schneebeli material, i.e. a collection of cylindrical rods of equal lengths
stacked parallel and observed laterally; it was well reproduced by CD
numerical simulations, cf. Daudon et al. {1997}.
Figure 2. Histograms of contact directions

12
Instead of circular objects, the two examples presented on Figure 2
concern convex irregular polygonal grains of random shapes, with two
different distributions of sizes. In each sample, for all contact points de-
tected in the rain-deposited layer, the normal directions - in an extended
sense, since contact may affect a vertex - are recorded. Histograms of
the distributions are constructed by dividing the 360 0 range about the
origin into 24 equal sectors. Each sector is filled in gray up to a radius
proportional to the number of contact points whose normals have the
corresponding directions. For legibility or for quantitative treatment,
one may choose to fit a continuous curve to the distribution, the graph
of a probability density expressed as a function of the angle () by a linear
combination of cosine and sine of multiples of 2(). The same preference
for two oblique directions as with circular grains is found.
Three-dimensional simulations yield similar results: in the granular
layer created by raining spherical grains of dispersed sizes, contact di-
rections of largest frequency make an angle with the vertical.
3.4. The stress tensor

In the Civil Engineering situations referred to in 3.1, the granular
material is treated in the framework of classical Continuum Mechanics,
with the Cauchy stress field as central concept. This macroscopic ap-
proach does not prevent one, at the stage of assessing the constitutive
laws that govern the medium behaviour, to refer to smaller scale fea-
tures. Elaborating from micromechanical quantities some tensor which
could be identified as the stress is not trivial, cf. Goddard (1998). Since
the very time of Cauchy, this question has mainly been addressed for
assemblies of points subject to some interaction potentials. Granular
media offer a very different sort of microstructure.
Figure 3. Quality of the stress tensor estimate.

13
On Figure 3 it is shown a part of a two-dimensional collection of irreg-

ular polygonal grains (grain to grain friction: 0.4) submitted to a 'biaxial
test', i.e. the granulate is squeezed between frictionless rectilinear ver-
tical boundaries in imposed motion while some containing pressure is
exerted by horizontal ones. The representation of contact forces as line
segments makes the force chains visible. The programme allows one
to superimpose a circular probing area over which the average stress
is computed according to the definition introduced in Moreau (1997),
practically equivalent to some other ones found in literature. If this ten-
sor is used in the standard way to express the normal and tangential
components of the tension upon a cut with direction {}, the results are
sinusoidal functions of {} whose graphs are drawn in gray. On the other
hand, 31 diameters of the probing area have been selected. By comput-
ing, for each of them, the resultant of the contact forces exerted by the
grains on one side upon the grains on the other side, one estimates the
corresponding tensions. When the plots of the normal and tangential
components of these vectors are compared with the preceding sinusoids,
one concludes that the computed tensor, in spite of the small size of the
sample and the inhomogenity due to force chains, plays with satisfactory
precision the role expected from the Cauchy stress in continuous media.
The cross drawn in the circular area represents the principal stresses.
3.5. Banks created by avalanches of dry grains

Figure 4 has been created in the course of the following two-dimensional
simulation. Random polygonal grains are deposited at the top of a rigid
incline, made of a collection of fixed grains with similar characteristics,
so as to simulate roughness. Thanks to a landing plate placed at the bot-
tom of the incline, a layer of some thickness accumulates, sporadically
perturbed by avalanches.
The angle that the free surface makes with horizontal fluctuates be-
tween two values: the angle of repose Orep (approx. 25°), observed after
an overall avalanche has remodelled the layer, and some larger value
OHm (approx. 29°), the limit angle, attained as the result of progressive
accretion. At OHm, the layer becomes unstable. The line of fixed grains
which makes the incline floor was precisely copied from the free surface
of a previously created pile, slightly before the limit state.
Similarly to what was done in 3.4, the programme allows one to choose
a line segment and to calculate the resultant contact force transmitted
across such an imagined cut.
For a line segment parallel to the free surface (the difference between
Orep and OHm is neglected here), one finds a vertical force. This feature
14
-
.;§~:::::. .-
~~~~m~·:g~~l~~,~;,. The deposited layer nearly at rest
with conjugate
directions
Figure 4- Two cuts in a bank.
looks natural since the bank is made of layers successively deposited by

avalanches. When such a layer stops, the supporting forces it experi-
ences from the material beneath should equilibrate its weight, and this
verticality of transmitted forces is expected to persist after other layers
possessing the same property have been superimposed. This confirms
the assumption commonly made in Civil Engineering that a freshly de-
posited bank is everywhere in a state of incipient failure.
For a vertical line segment, one finds a force parallel to the free sur-
face. This observation, which seems novel, is connected in an essential
way with the preceding one through the symmetry of the Cauchy stress
tensor. It provides a convincing insight into the following subject which,
in recent years, started a lot of speculation and controversy.
Let a conical pile be created by pouring grains from a source onto
a rough rigid horizontal ground. Some experimentalists have been sur-
prised to observe that the distribution of the pressure exerted upon the
ground was not proportional to the height of the material above and
even that a local minimum of ground pressure may be present at the
vertical of the apex.
Actually, the fact that the free surface is not horizontal makes clear
that the statics of the granular assembly has nothing to do with hy-
drostatics. In a three-dimensional CD numerical experiment not shown
15
here, we have reproduced the construction of a conical pile of 14.000

spherical grains with dispersed sizes; see Moreau (2001). Computed re-
sults reveal first that a conical test surface portion placed inside the
peripheral bank of the pile and parallel to the free surface transmits a
vertical force, indicative of a state of incipient failure. Secondly, a test
surface in the shape of a cylinder coaxial with the conical pile is found
to transmit average forces parallel to the free surface. As before, this
couple of observations reflects the symmetry of the stress tensor. If such
a cylinder is used to delimit some 'central core' of the pile, this core
consequently has part of its weight supported by an arching action from
the surrounding bank, which explains the experimental findings.
References
Abadie, M. (2000) Dynamic simulation ofrigid bodies: modelling of frictional contact,
in Impacts in Mechanical Systems. Analysis and Modelling, edited by B. Brogliato,
Springer-Verlag, Berlin Heidelberg, 61-144.
Alart, P., Curnier A. (1991) A mixed formulation for frictional contact problems prone
to Newton like methods, Comput. Meth. in Appl. Mech. Engng. 92, 353-375.
Brogliato, B. (1999) Nonsmooth Mechanics, 2d. edition. Springer-Verlag, London.
Brogliato, B., ten Dam, A. A., Paoli, L., Genot, F. and Abadie, M. (2001) Numerical
simulation of finite dimensional multibody nonsmooth mechanical systems, ASME
Applied Mechanics Reviews, to appear.
Chabrand, P., Dubois, F., and Raous, M. (1998) Various numerical methods for solv-
ing unilateral contact problems with friction, Mathl. Comput. Modelling 28, 97-
108.
Christensen, P. W., Klarbring, A., Pang J. S., and Stromberg N. (1998) Formulation
and comparison of algorithms for frictional contact problems, Int. J. Num. Meth.
Engng. 42,145-173.
Curnier, A. (1984) A theory of friction, Int. J. Solids Struct. 20, 637-647.
Daudon, D., Lanier, J., and Jean, M. (1997) A micromechanical comparison between
experimental results and numerical simulation of a biaxial 2D granular material,
in Powders and Grains 97, edited by R. P. Behringer and J. T. Jenkins, Balkema,
Rotterdam, 219-222.
Fremond, M. (1995) Rigid body collisions, Physics Letters A 204, 33-41.
Genot, F., and Brogliato, B. (1999) New results on Painleve paradoxes, European
Journal of Mechanics, A/Solids, 18, 653-677.
Goddard, J. D. (1998) Continuum modelling of granular assemblies, in Physics of Dry
Granular Materials, edited by H. J. Herrmann et al., Kluwer, Dordrecht Boston
London, 1-24.
Ivanov, A. P. (1997) The problem of constrained impact, J. Appl. Math. Mech. 61,
341-353.
Jean, M. (1999) The Non Smooth Contact Dynamics method, in Computational Mod-
eling of Contact and Friction, edited by J. A. C. Martins and A. Klarbring, special
issue of Computer Meth. in Appl. Mech. and Engng. 177, 235-257.
Jean, M. (2001) Simulation numerique discrete de materiaux granulaires, in Mi-
cromecanique des materiaux granulaires, edited by B. Cambou and M. Jean, Her-
mes, Paris.
16
Johansson, L., and Klarbring, A. (2000) Study of frictional impact using a nonsmooth
equation solver, ASME J. Appl. Mech. 67, 267-273.
Jourdan, F., Alart, P., and Jean, M. (1998) A Gauss-Seidel-like algorithm to solve
frictional contact problems. Computer Meth. Appl. Mech. Engng. 155, 31-47.
Kunze, M., and Monteiro Marques, M. D. P. (2000) An introduction to Moreau's
sweeping process, in Impacts in Mechanical Systems. Analysis and Modelling, edited
by B. Brogliato, Springer-Verlag, Berlin Heidelberg, 1-60.
Moreau, J. J. (1966) Quadratic programming in mechanics: dynamics of one-sided
constraints, SIAM J. Control 4, 153-158.
Moreau, J. J. (1988a) Bounded variation in time, in Topics in Nonsmooth Mechanics,
edited by J. J. Moreau, P. D. Panagiotopoulos, and G. Strang, Birkhiiuser, Basel
Boston Berlin, 1-74.
Moreau, J. J. (1988b) Unilateral contact and dry friction in finite freedom dynam-
ics, in Nonsmooth Mechanics and Applications, edited by J. J. Moreau and P. D.
Panagiotopoulos, CISM Courses and Lectures, Vol. 302. Springer-Verlag, Wien
New York, 1-82.
Moreau, J. J. (1989) An expression of classical dynamics, Ann. Inst. H. Poincare Anal.
Non Lineaire, 6 (suppl.), 1-48. Volume also available as Analyse Non Lineaire,
edited by H. Attouch, J.-P. Aubin, F. Clarke, and I. Ekeland, Gauthier-Villars,
Paris.
Moreau J. J., (1997) Numerical investigation of shear zones in granular materials,
in Proc. HLRZ- Workshop on Friction, Arching, Contact Dynamics, edited by P.
Grassberger and D. Wolf, World Scientific, Singapore, 233-247.
Moreau, J. J. (1999) Some basics of unilateral dynamics, in Unilateral Multibody Con-
tacts, edited by F. Pfeiffer and Ch. Glocker, Kluwer, Dordrecht/Boston/London,
1-14.
Moreau J. J. (2000) Contact et frottement en dynamique des systemes de corps rigides,
Rev. Europ. des Elements Finis 9, 9-28.
Moreau J. J. (2001) An introduction to unilateral dynamics, in Novel approaches
in Civil Engineering, edited by M. Fremond and F. Maceri, Springer-Verlag, to
appear.
Nouguier, C., Bohatier, C., Moreau, J. J., and Radjai, F. (2000) Force fluctuations in
a pushed granular material, Granular matter 2, 171-178.
Pang, J. S., and Stewart, D. E. (1999) A unified approach to discrete frictional contact
problems, Int. J. Engng. Sci., 37, 1747-1768.
Pfeiffer F., and Glocker Ch. (1996) Multibody Dynamics with Unilateral Contacts,
John Wiley and Sons, New York.
Stewart, D. E. (1998) Convergence of a time-stepping scheme for rigid body dynamics
and resolution of Painleve's problem, Arch. Rational Mech. Anal. 145, 215-260.
Stoianovici, S. P., and Hurmuzlu, Y. (1996) A critical study of the concepts of rigid
body collision theory, J. Appl. Mech. 63, 307-316.
Vola, D., Pratt, E., Jean, M., and Raous, M. (1998) Consistent time discretization
for a dynamical frictional contact problem and complementarity techniques, Rev.
Europ. des Elements Finis 7, 149-162.
MEASUREMENTS OF IMPACTS
WITH FRICTION
Friedrich Pfeiffer
Lehrstuhl fUr Angewandte Mechanik, Technische Universitiit Miinchen
D-85747 Garching, Germany
Abstract Existing models of impacts with friction consider compression and ex-
pansion phases, where during compression impulses are stored, and
where during expansion they are released with frictional losses, both
in normal and tangential contact directions. Measurements performed
with a specially designed ballistic machine verify these theories and give
an indication for improvements.
1. Introduction
Impacts with friction play an important role in machines and mecha-
nisms. Therefore many efforts have been made to establish good models,
where the most realistic ones are based on Moreau's ideas (1986, 1988),
continued and extended by Glocker (1995). The theory takes into ac-
count compression and expansion phases during the impact and con-
siders storing and restoring energy in normal and tangential directions.
Friction is included by the corresponding complementarities, restoring is
governed by Poisson's friction law. The theory has been applied success-
fully to many industrial problems, but in the meantime it has been also
systematically verified by laboratory tests. For this purpose a machine
was designed, where a rotating mechanism on a rotating arm releases a
disc within milliseconds, at the same time controlling nearly exactly the
translational and the rotational speeds of the disc. These speeds can be
pre-programmed. The ballistic flight of the disc, the contact with the
ground and the disc-ground behavior are tracked stroboscopically and
by a high-speed camera. Special marks assure a precise evaluation of
the experiments. More than 600 experiments have been carried through
with various material pairings. The outcome confirms the theory excel-
lently, and, moreover, it has induced a small correction of the corner law
during expansion.
17
l.A. C. Martins and M.D.P. Monteiro Marques (eds.). Contact Mechanics. 17-24.
18
First ideas for impact modeling came from Newton (1687) and Poisson
(1833). The interest in problems of that kind arose anew in the last fif-
teen years when Moreau (1986, 1988) and Panagiotopoulos (1985, 1993)
started to consider mechanical systems with unilateral properties first
more in a statical sense, then more and more also dynamically. Both
scientists have the merit to have created and established this area with
great impact to modern mechanical sciences. A very good review of
impact dynamics is presented by Brogliato (1996). Murty (1988) gives
a collection of practical rules to deal with the numerical problems of
complementarity and, more generally, of nonlinear optimization.
The author's own institute develops research in that area for sixteen
years, very much related though to applications of practical relevance. A
survey of these activities is given by Pfeiffer, Glocker (1996). In partic-
ular an impact theory including frictional effects has been developed by
Glocker (1995) and experimentally confirmed by Beitelschmidt (1998).
The paper at hand is mainly based on these findings.
2. Experimental Test Set-Up

The theory of impacts with friction has been published elsewhere
(Glocker (1995), Beitelschmidt (1998), Pfeiffer, Glocker (1996)). There-
fore we shall focus here on the experimental side. In designing a test
set-up for measuring impacts with friction a first principal decision with
respect to the experiments referred to the geometrical type of impact,
plane or spatial. Colliding bodies moving in a plane are connected with
linear complementarity problems, spatial contacts generate nonlinear
complementarities. Therefore motion in a plane was considered where
one body is a disc and the other one the ground. On this basis some
further requirements had to be defined:
• maximum translational velocity 10 mls

• maximum rotational velocity 40 rps
• throw direction 0° - 90°
• release time < 12 ms

• encoder main axis 1600 points
• encoder momentum axis 400 points
• throwing disc diameter 50 mm

thickness 20 mm
weight 300 g
19
• continuous variable velocity control

• translation and rotation decoupled
• disturbance-free support and release of disc
• mass balance, statically and dynamically
• electric drives (pulse width modulation with 250 steps)
• automatic control for the throwing process, the release of strobo-
scope and camera
momentum drive
release unit
release unit
Figure 1. Throwing machine
As a result, the machine of Fig. 1 was designed and built, which met
all requirements. A release unit containing the disc is mounted at the
end of a rotating arm with mass balance. The unit itself drives the disc
giving it a prescribed rotational velocity. Main drive and momentum
drive are decoupled allowing to control the two speeds independently.
The rotation of the arm mainly generates a translation, the rotation
of the release unit a rotation of the disc. The flight of the body is
photographed under stroboscopic exposure in a dark room before and
after hitting his target. From the evaluation of the photographs one can
20
calculate the velocities and the position of the body immediately before
and after the impact.
Figure 2 depicts the structure of the test set-up. A computer performs
all control calculations, processes sensor data, evaluates control torques,
releases stroboscope and camera and records all measured data. Within
this overall structure we find for each drive an individual control con-
cept, which has thoroughly been optimized with regard to the above
requirements (Beitelschmidt (1998)).
flash sensor
camera
AT-BUS card
Figure 2. Structure of the complete test set-up
A typical sequence of events for the test procedure is the following:
Data Input
Throw Planning
Initialization Hardware
Initialization Process Control
Trajectory Planning
6 Sensor Acquisition
'"'" Drive Control
8
e
p.., Release Disc
Release Flash, Camera
De-Initialization Process Control
Storage of Results
Post-processing
All computer codes have been realized in C++, which was feasible
due to the fact that the PC-Mode activities are not critical with respect
to time.
21
3. Results
The evaluation of the measurements as recorded by the camera and
the processor was straightforward. Figure 3 illustrates the method and
shows additionally two photographs of experiments. Especially the rub-
ber disc experiment shows nicely a reversal of the trajectory due to the
disc's rotation. The experimental process provided thus a very precise
and well reproducible basis for determining the properties of impacts
with friction.
a)
b)
c)
Figure 9. Disc trajectory during an experiment

a) method of evaluation b) photograph steel c) photograph rubber
22
In the following we shall give only a few examples out of more than
600 experiments performed with axisymmetric and with eccentric discs.
In all cases the comparisons with theory are good to excellent (Beit-
elschmidt (1998)). In the following diagrams we shall use dimensionless
velocities and impulses defined by
gTA gNC
'Y , 'YNC=-.-,
iJTc
'YTC= - . - ,
-9NA -9NA -9NA
gNE iJTE gTEO
'YNE = , 'YTE = - . - , 'YTEO = - . - , (1)
-gNA -9NA -9NA
where the indices N, T refer to normal and tangential directions. The
indices A, C, E are the beginning and the end of the compression phase,
and the end of the expansion phase, respectively. The kinematical mag-
nitude g is a relative velocity in the contact zone. Experiments usually
generated a negative normal velocity (-gN A) at the beginning. The first
figure shows results of experiments with the PVC test body.
I§I .. ~/~"
0.5
.
J"]ZI'-"
#,.,r
~'
Ai...."
o • .- -"
-0.5 ,,.'N.
-"til'
,,~
...., ....
~~
-1
/~ pvc
4.
I--+--i
Theory ---------.
-1.5 -1 -0.5 o 0.5
y
Figure 4. Dimensionless tangential relative velocity
after vs. before the impact, PVC-body
The experiments are marked by crosses, the dotted line shows the the-
oretical result. For small tangential relative velocities before the impact,
sticking occurs, and the rolling constraint between disc and ground is
fulfilled after the impact. If the relative velocity is big enough, the body
slides throughout the impact and has a redcuced tangential relative ve-
locity at the end of the impact. At this impact no tangential reversion
occurs.
23
A similar diagram for a rubber-body is shown in Fig. 5. For most

of the impacts the tangential relative velocity has changed during the
impact: the bodies collide with a negative relative velocity and separate
with a positive velocity. The inclination of the line through the origin is
-ENET. If EN is known from another simple experiment one can evaluate
the coefficient of tangential reversibility from this plot. For this series
of experiments the parameters EN = 0.75 and ET = 0.9 were identified.
2.5
1.5
w
;t::
•
0.5
0
/o-f--o
-0.5
-8 -6 -4 -2 o 2
y
Figure 5. Dimensionless tangential relative velocity
after vs. before the impact, rubber-body
If the tangential relative velocity increases further, sliding occurs in

the contact point during the impact. Then it is not possible to restore
the elastic potential energy during the phase of expansion. For very high
velocities the rubber body slides during the whole impact and the effect
of tangential reversibility is not further visible. In Fig. 5 two lines are
plotted for comparing theory with experiment. The new theory includes
a correction with respect to Glocker's theory (1995). If we consider
the contact point of two bodies, where Coulomb's friction applies, and
that point of the contact zone, where the spring force resulting from the
storage of impulse applies, we come out with two force laws in series.
This gives a modification of the complementarities with respect to the
friction cone, and thus a modification of the final results (Beitelschmidt
(1998)).
24
4. Summary
Impacts with friction play an essential role in machines and also in
biology. Therefore we need good models being verified by sound ex-
periments. This paper gives a survey of such experiments which have
been performed by an especially designed throwing machine. The exper-
imental results compare excellently with calculated values from existing
theories. A slight improvement of the theory could be deduced from
the experimental findings. It concerns the frictional complementarities
during expansion, which have been slightly modified.
An additional verification of the theory has also been performed by
applying it to a large variety of industrial problems, where measurements
were carried through within an industrial environment. Some typical
examples are given by Pfeiffer, Glocker (1996), Pfeiffer (1996), Pfeiffer,
Fritz, Srnik (1997) and Pfeiffer, Stiegelmeyr (1997).
References
Beitelschmidt, M. (1998) Reibstope in Mehrkorpersystemen, Dissertation TU-Miinchen.
Brogliato, B. (1996) Nonsmooth Impact Dynamics, Springer, London.
Glocker, Ch. (1995) Dynamik von Starrkorpersystemen mit Reibung und StofJen.
Fortschr.-Ber. VDI. Reihe 18, Nr. 182, VDI-Verlag, Diisseldorf,
Moreau, J.J. (1986) Une formulation du contact a frottement sec; application au calcul
numerique, Technical Report 13, C.R. Acad. Sci. Paris, Serie II.
Moreau, J.J. (1988) Unilateral Contact and Dry Jilriction in Finite Freedom Dynam-
ics, Non-Smooth Mechanics and Applications, CISM Courses and Lectures, 302,
Springer Verlag, Wien.
Murty, K.G. (1988) Linear Complementarity, Linear and Nonlinear Programming,
Sigma Series in Applied Mechanics (ed. White, D.J.), Heldermann Verlag, Berlin.
Newton, I. (1687) Principia, Corol.
Panagiotopoulos, P.D. (1985) Inequality Problems in Mechanics and Applications,
Birkhiiuser, Boston, Basel, Stuttgart.
Panagiotopoulos, P.D. (1993) Hemivariational Inequalities, Springer Verlag, Berlin,
Heidelberg.
Pfeiffer, F., and Glocker, Ch. (1996) Multibody Dynamics with Unilateral Contacts,
Wiley & Sons, New York.
Pfeiffer, F. (1996) Assembly processes with robotic systems, Robotics and Autonomous
Systems 19, 151-166.
Pfeiffer, F., Fritz, P. and Srnik, J. (1997) Nonlinear Vibrations of Chains, J. of Vi-
bration and Control 3, 397-410.
Pfeiffer, F., and Stiegelmeyr, A. (1997) Damping Towerlike Structures by Dry Jilriction,
Proc. of DETC '97, ASME Design Eng. Techn. Conf.
Poisson, S.D. (1833) 1hliU de mechanique, Bachelier, Paris.
Srnik, J., and Pfeiffer, F. (1997) Dynamics of CVT Chain Drives: Mechanical Model
and Verification, Proc. of DETC '97, AS ME Design Eng. Techn. Conf.
FORMULATION AND WELL-POSEDNESS
OF UNILATERAL MULTIBODY
DYNAMICS
Patrick Ballard
Labomtoire de Mecanique des Solides - Ecole Polytechnique
91128 Palaiseau Cedex - Jihlnce
Abstract The classical theory of rigid bodies systems undergoing perfect bilateral
constraints has received firm mathematical foundations for a long time.
We extend it to the case where there are perfect unilateral constraints
also. The formulation follows the line initiated by Schatzman (1978) and
Moreau (1983). We give a parametrization-free formulation and try to
identify the most general form of impact constitutive equation which is
compatible with weU-posedness. Then, weU-posedness is proved under
the assumption of analyticity of the data, since it is known, thanks to
Bressan (1960) and Schatzman (1978), that uniqueness does not hold
in the Coo setting.
1. Background: multi body systems with perfect

bilateral constraints
Consider a finite collection of rigid bodies undergoing perfect bilateral
constraints. It is classically associated with a manifold Q (called the
configuration manifold of the system) of finite dimension d (called the
number of degrees-of-freedom of the system).
A motion of the system is a curve on Q, that is a mapping q(t) from
a real interval I into Q. When a motion is smooth enough, there exists
a time-derivative q(t) in tangent space Tq(t)Q also called a (generalized)
velocity of the system. We shall also denote it by (q( t), q( t) ). An arbi-
trary element v (also denoted by (q, v), q being the basepoint of v) of
the tangent bundle TQ is called a (virtual) state of the system.
Given the mass distribution in each of the solids, one classically ob-
tains the kinetic energy K which appears to define a positive definite
quadratic form on each tangent space TqQ of the configuration man-
ifold, endowing it with a Riemannian structure. The induced scalar
25
26
product and norm will be denoted by (., .)q and 1I·lI q , whereas the du-
ality product between tangent space TqQ and cotangent space T;Q will
be denoted by (., .)q. With this notation, we have:
11 vllq·2
K(q,v) = 2"1
The modelling of forces makes use of duality. Internal and external
forces acting on the system in the configuration q are represented by
a linear form f E T;Q. Given an arbitrary virtual velocity (q,v) of
the system, the real number (f, v}q is the virtual power of internal and
external forces for the virtual velocity (q, v). The cotangent vector f is
usually allowed to depend on the current state (q,v) and also on time t.
Hence, we are given a mapping f: TQ X lR+ -+ T*Q satisfying:
V(q,V)ETQ, VtElR+, IIQ(f(q,v;t))=q, (1)
(II Qis the natural projection on T*Q), called the force mapping.
Since such a structure appears systematically in the modelling of the
dynamics of rigid bodies systems with perfect bilateral constraints, we
are led to make the following definition.
Definition 1 A simple discrete mechanical system is a pair (Q, f) whe-
re:
• Q is a finite-dimensional Riemannian manifold called the configu-
ration manifold.
• f: TQ X lR+ -+ T*Q is a mapping satisfying (1), called the force
mapping.
Now, consider an arbitrary smooth motion q(t) of the system. The
power of inertial forces is by definition the time derivative of the kinetic
energy:
:tK(q,q) = :t~llq(t)II:(t) = (~q(t),q(t)) q(t) = /~~q(t),q(t))

\ q(t)
,
where D/dt denotes the covariant derivative along the curve q(t) associ-
ated with the Levi-Civita connection, and ~ the isomorphism from TQ
onto T*Q canonically associated with the Riemannian metric. For any
virtual velocity (q, v) ofthe system, the real number (~Dq/dt, v}q is nat-
urally called the virtual power of inertial forces in the virtual velocity
(q, v). Given any local parametrization of the system (local chart), it
admits the well-known representation:
~~q = (:t~iK(q,q) - ~iK(q,q)) dqi.

27
The fundamental principle of classical dynamics asserts that the vir-
tual power of inertial forces should equal the virtual power of external
and internal forces, for any virtual velocity. We obtain immediately the
equation of motion:
\:It, b~q(t) = f(q(t),q(t);t).
Next, we are given an initial state (qO, vol E TQ. Then, the evolution
problem associated with the dynamics of rigid bodies system with perfect
bilateral constraints is the Cauchy problem:
Problem I. Find T > 0 and q E 0 2 {[O, T[; Q) such that:
• (q(O), q(O)) = (qO, vol,
• \:ItE[O,T[, b~q(t)=f(q(t),q(t),t).
Well-posedness is ensured by the following.
Theorem 2 (Cauchy) The Riemannian configuration manifold is as-
sumed to be of class 0 2 and the mapping f : TQ x R+ -+ T*Q is of class
0 1 . Then, there exists a unique maximal solution for problem 1.
More precisely, theorem 2 states that there exists Tm > 0 (Tm E
R+ U {+oo}) and qm E 0 2 ([0, Tm [, Q) being a solution of problem I such
that any other solution of problem I is a restriction of qm. Of course, we
expect that Tm = +00, in which case the dynamics is said to be eternal.
This situation cannot be taken for granted, in general. In the usual cases
encountered in mechanics, eternal dynamics is ensured by the following
general sufficient condition.
Theorem 3 The configuration manifold Q is assumed to be a complete
Riemannian manifold. The effort mapping f is supposed to admit the
following estimate:
IIf(q,v;t)lI; ~ l(t) (1 +d{q,qo) + IIvll q) ,
for all (q, v) E TQ and almost all t E [to, +00[, where d{·,·) is the
Riemannian distance and l{t), a (necessarily nonnegative) function of
L}oc{lR; R).
Then, the dynamics is eternal: Tm = +00.
2. Formulation of unilateral multibody dynamics

The consideration of elementary examples shows that the dynamics
of rigid bodies systems can lead to predictions of the motion in which
28
some bodies of the system overlap in the real world. Of course, this
cannot be allowed. Hence, very often, one has to add the statement of
non-penetration conditions to a simple discrete mechanical system. This
is a simple occurrence of a unilateral constraint. Here, we shall recall
briefly the general formulation of the resulting evolution problem. This
is due essentially to Schatzman (1978) and Moreau (1983).
Consider a simple discrete mechanical system according to defini-
tion 1. A unilateml constmint is a restriction on the admissible motions
of the system which is expressed by means of a finite number n of smooth
real-valued functions /.Pi on the configuration manifold Q, so that the set
of all admissible configurations A is given by:
A={qEQ;'ViE{1,2, ... ,n}, /.pi(q)~O}. (2)
The set of all active constraints in the admissible configuration q E A is

defined by:
J(q)={iE{1,2, ... ,n}; /.Pi(q)=O}.
The following hypothesis is usual in this framework. The functions /.Pi
are assumed to be functionally independent in the sense that, for all
q E A, the d/.pi(q) (i E J(q)) are linearly independent in T*Q.
The cone of admissible right velocities in the configuration q is defined
by:
V(q) = {v E TqQ; 'Vi E J(q), (d/.pi(q),v}q ~ O},
and we denote by N*(q) its polar cone for the duality (TqQ, T;Q):
N"(q) = {t, A; d\Oi(q) ; Vi E J(q), .I; ~ 0, Vi ric J(q), A; = o},

the polar cone of V(q) for the euclidean structure of TqQ being N(q) =
~(N*(q)) (~= b- 1 ).
The equation of motion has to be corrected with some reaction efforts
R:
n q·+ - f( .+.) R
"T
L.
- q,q, t + . (3)
We require the unilateral constraint to be perfect, that is, the reaction
effort R has to take values in the normal cone N* (q). It is well known
that the dynamics of such a system involves some time-discontinuities of
velocity called impacts. Therefore, we cannot expect that the equation
of motion (3) should be satisfied in a classical sense, but rather in a
distributional sense. Actually, it is usual to require that R should be
a vector-valued mesasure rather than a general distribution. Hence, we
29
define the class of motions M M A(Ij Q) (motions with measure acceler-
ation) to be the set of all absolutely continuous motions q(t) from a real
interval I to Q admitting a right velocity q+(t) at every instant t of I
and such that the function q+(t) has locally bounded variation over I.
Bounded variation is classically defined only for functions taking values
in a normed vector space. However, for any absolutely continuous curve
q(t) on a lliemannian manifold, parallel translation along q(t) classically
provides intrinsic identification of the tangent spaces at different points
of the curve and so, the definitions can easily be carried over to this
case (for a precise mathematical setting, see Ballard (2000)). Any mo-
tion q E MMA(Ij Q) admits a left and right velocity, q- and q+, in the
classical sense at any instant. Moreover, any motion q E MMA(Ij Q) is
intrinsically associated with the covariant Stieltjes measure Dq+ of its
right velocity q+. The equation of motion takes the form:
bDq+ = f(q, q+j t) dt + R,
where dt denotes the Lebesgue measure. It remains to express the con-
dition that R should take values in N*(q) when R is a measure.
Convention. We shall write:
R E -N*(q(t))
to mean: there exist n nonpositive real measures Ai such that:
n
R= L Ai dc,oi(q(t)) and
i=1
'v'iE{1,2, ... ,n}, SUppAiC{tj c,oi(q(t»=O}. (4)
With this convention, the final form of the equation of motion is:
R = bDq+ - f(q(t), q+(t)j t) dt E -N*(q(t)) (5)
It is classical that this equation of motion has to be completed with
some constitutive impact equation describing the outcome of any impact
occurrence. We shall assume an impact constitutive equation of general
form:
'v't, q+ (t) = F (q(t), q- (t» . (6)
To ensure compatibility with the equation of motion (5), the mapping
F should satisfy:
F(q,v-) E V(q),
F (q, v-) - v- E -N(q). (7)
30
Moreover, we add the assumption that the kinetic energy of the system
cannot increase during an impact:
'v' q E A, 'v' v - E -V (q), (8)
Now, we formulate the evolution problem associated with the dy-

namics of rigid bodies systems with perfect bilateral and unilateral con-
straints. The initial condition is assumed to be compatible with the
realization of the constraint: Vo E V(qo).
Problem II. Find T > 0 and q E MMA([O, T[; Q) such that:
• (q(O),q+(O)) = (qO,vo),
• 'v't E [0, T[, q(t) E A,
• R ~ ~Dq+ - f(q(t), q+(t); t) dt E -N*(q(t)),
• 'v'tE]O,T[, q+(t)=F(q(t),q-(t)).
The equation of motion is understood in the sense of convention (4), and

the impact constitutive equation is supposed to fulfill requirements (7)
and (8).
3. Well-posedness of unilateral multibody

dynamics
To study existence and uniqueness of solution for problem II, we need
to state some regularity hypotheses on the data. It was pointed out by
Bressan (1960) and Schatzman (1978), through a striking example, that
under the strong hypotheses that the data are all of class Coo there may
exist non-unique solutions for problem II. It was Percivale (1985) and
Schatzman (1998) who were the first to notice that this trouble could be
overcome by assuming that the data are analytic. But their analysis was
limited to one-degree-of-freedom systems. The general case is treated in
Ballard (2000), Ballard (2001) and is now briefly recalled.
From now on, we assume the following
Regularity hypothesis. The Riemannian configuration manifold Q,
the functions 'Pi and the mapping f : TQ X jR+ -7 T* Q are analytic.
Proposition 4 Let qo E A and vo E V(qo). Then, there exist Ta > 0,

an analytic curve qa : [0, Ta[-7 Q and n analytic functions Aai : [to, Ta[-7
jR such that:
31
• (qa(O), q:(O)) = (qo, VO),
• \I I E [0, Ta[, ~ ~ qa(l) = f (qa(I), qa(I); I) + ~ >.,..( I) dl",(qa( I)),

. ~~
vZ
== [10'2T~[.,.
" ,n, Aai(t) ~ 0, rpi(qa(t)) ~ 0, Aai(t) rpi(qa(t)) = O.
Moreover, the solution of this evolution problem is unique in the sense

that any other analytic solution (T, q, A1, ... , An) is either a restriction
or an analytic extension of (Ta, qa, Aa1' ... , Aan)·
An early proof of proposition 4 can be found in Lotstedt (1982).
Corollary 5 There exists an analytic solution (Ta, qa) for problem II.
Naturally, the analytic solution furnished by corollary 5 will cease to

exist at the first instant of impact. This is the reason why we have
considered the wider class MMA which contains motions which are not
differentiable in the classical sense. Considering motions in MMA will
allow US to extend the solution beyond the first instant of impact. But,
we must ensure that admitting the wider class of solutions MMA will not
introduce parasitic solutions. This is the aim of the following theorem.
Theorem 6 Let (Ta, qa) be the solution for problem II furnished by

corollary 5, and (T, q) be an arbitmry solution for problem II. Then,
there exists a real number To (0 < To ~ min{Ta, T}) such that:
ql[O,To[ = qal[O,To[·
In other terms, there is local uniqueness for problem II.
Local uniqueness is the difficult part in the study of well-posed ness of

problem II.
Corollary 7 There exists a unique maximal solution for problem II.
It was noticed above that the analytical solution for problem II fur-
nished by corollary 5 fails to exist at the first instant of impact. To
overcome this fact, we have proved that local uniqueness still holds in
the wider class of motion MMA which allows impacts. However, the
maximal solution for problem II may stop to exist at finite time for un-
physical reasons. In other terms, we still do not know if the class MMA
is wide enough. Actually, it is wide enough as shown by the following
theorem which should be brought aside theorem 3.
32
Theorem 8 The configumtion manifold Q is assumed to be a complete

Riemannian manifold and the mapping f is supposed to admit the fol-
lowing estimate:
IIf(q,v;t)lI; ~ l(t) (1 +d(q,qo) + IIvll q) ,
for all (q, v) E TQ and almost all t E [0, +00[, where d(·,·) is the
Riemannian distance and l(t), a (necessarily nonnegative) function of
L}oc{1R; JR).
Then, the dynamics is eternal, that is, the maximal solution for prob-
lem II is defined on [0, +00[.
References
Ballard, P. (2000) The dynamics of discrete mechanical systems with perfect unilateral
constraints, Archive for Rational Mechanics and Analysis 154, 199 - 274.
Ballard, P. (2001) Formulation and well-posedness of the dynamics of rigid bodies
systems with perfect unilateral constraints, Philosophical 7ransactions of the Royal
Society Serie A (to apppear).
Bressan, A. (1960), Incompatibilita. dei teoremi di esistenza e di unicita del moto per
un tipo molto comune e regolare di sistemi meccanici, Annali della Scuola Normale
Superiore di Pisa Serie III, Vol. XIV, 333 - 348.
Lotstedt, P. (1982), Mechanical systems of rigid bodies subject to unilateral con-
straints SIAM J. Appl. Math. 42, no 2, pp 281 - 296.
Moreau, J.J. (1983) Standard inelastic shocks and the dynamics of unilateral con-
straints, in Unilateral problems in structural analysis (G. Del Piero and F. Macari
Eds), Springer-Verlag, Wien, New-York, 173 - 221.
Percivale, D. (1985) Uniqueness in the elastic bounce problem, I, Journal of Differ-
ential Equations 56, 206 - 215.
Schatzman, M. (1978) A class of nonlinear differential equations of second order in
time, Nonlinear Analysis, Theory, Methods (1 Applications 2, No 2, 355-373.
Schatzman, M. (1998) Uniqueness and continuous dependence on data for one dimen-
sional impact problems, Mathematical and Oomputational Modelling 28, No. 4-8,
1-18.
COLLISIONS IN SYSTEMS MADE OF
RIGID BODIES
Eric Dimnet
Laboratoire Lagrange, Laboratoire Central des Ponts et Chaussees
58, boulevard Lefebvre, 75732 Paris cedex 15.
Abstract We describe the collision of a point with a fixed plane and the simul-
taneous collisions of a collection of rigid bodies following the theory of
collisions by M. Fremond. We then give some existence and uniqueness
results for simultaneous collisions of a collection of rigid bodies, and
we show some numerical simulations of the evolutions of collections of
solids during which such collisions happen.
1. Introduction.
In this paper, we discuss instantaneous collisions involving rigid and
deformable solids by describing the interior percussions of the system
made of the colliding bodies. The system made of all the colliding el-
ements is a deformable system: its form changes, even if it is made of
rigid elements! If the duration of a collision is small compared to the
duration of the evolution, we assume that the collision is instantaneous;
thus the velocities are discontinuous.
33
JAC. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 33-43.
@ 2002 Kluwer Academic Publishers.
34
2. Instantaneous collisions of rigid solids.

2.1 Collision of a rigid body with a plane.
piD~x,t)
x(t)
Figure 1. Collision of a point with a rigid fixed plan and sliding after the collision.
The solid is reduced to a point. Let us consider a point, moving

above a rigid fixed plane. Its position at time t is x( t). The system
made of the point and the plane is deformable because the distance of
the point to the plane changes. The deformation velocity is the velocity
ofthe point with respect to the plane, U(t) = dx(t)/dt. In many circum-
stances the duration of the collisions of the point and the plane is small
compared to the duration of the considered evolution: thus we consider
that the collisions are instantaneous. There is the velocity U- (t) before
a collision at time t, and the velocity U+(t) after the collision. Let a
virtual velocity V (T) be a bounded variation function.
We specify the interior forces of the system point-plane by defining
their virtual work. The virtual work of the interior forces is a linear
function of the velocity of deformation which has to be equal to 0 for
any rigid system motion. In our situation, because one element of the
system (the plane) is fixed, the rigid system motions, i.e., the motions
which do not change the form of the system, are reduced to the motion
with zero velocity.
Consider the actual motion shown on figure 1 and a virtual velocity
which is discontinuous at time t. The virtual work of the interior forces
we choose is
J/
wint(tl,t2' V) = - 12 Rint(T). V(T)dT
(1)
_pint-(t) . V-(t) - pint+(t) . r(t)
35
where Rint (r) is the contact force between the point and the plane. Both
quantities, pint-(t) and pint+(t) are percussions.
The virtual work of the acceleration forces is
wacc(tl, t2, V) = J/ md~JT) . V(r)dr

12
+m(U+(t) -
~ -+
U-(t)). V (t)!V (t).
(2)
Let us note that the virtual work of the acceleration forces is such that
the actual work is the variation of the kinetic energy between the times
tl and t2. We suppose that there are no exterior forces applied to the
system. The equations of motion result from the principle of virtual
work
\::Itl, \::ft2, \::IV, wacc(tl, t2, V) = wint(tl, t2, V) (3)
They are
m dU
dr
=_Rint ' (4)
almost everywhere and
0= pint-(t) - pint+(t), m(U+ (t) - U- (t)) = _pint(t), (5)
with pint(t) = 2pint-(t) = 2pint+(t) at any time t. The power of the

interior forces becomes
A more sophisticated theory involving the two percussions font-(t)

and pint+(t) is investigated in Fremond (2000).
From now on we consider only the collisions and do not investigate
the smooth motion.
The rigid body is not reduced to a point. A solid with mass m,

center of mass G, mass moment of inertia I, velocity of center of mass
(j and rotation velocity n,collides with the plane at a unique point A
(figure 2). The equations of motion at the time of collision result from
the principle of virtual work
W,\::Iw,
m(U+ - U-). y++y- + I(n+ - n-). w++w-
_ 2_ _ _ 2 (7)
= _pint. D(V+,w+,A)+D(V-,w-,A»),
.... .... .... 2 r;-t
D(V,w,A) = V +w X GA.
36
where DCV,w,A) is the velocity of deformation of the system at point

A, i.e. the relative velocity of the point with respect to the plane. They
are
m(tJ+ - 71-) = _pint,
(8)
I(rt+ (to) - rt- (to)) = -GA X pint.
The situation where the contact occurs at points, lines or surfaces is in-
vestigated in Dimnet and Fremond (2000), Dimnet (2001), and Fremond
(2001) .
A consequence of the chosen work of the interior forces in (7) is that
the interior percussion pint is applied at the collision point A.
Figure 2. Collision of a rigid body with a rigid fixed plan. The contact point is A.
Constitutive laws. The interior percussion pint is split between

a dissipative percussion pd which modelizes all the dissipative interac-
tions between the colliding solids and a reaction to the impenetrability
condition preac. The expression of the work of the interior force in (6)
and (7) leads to assume that pd depends on the velocity of deformation.
Thus
pint = pd( D(71+, rt+, A) + D(71-, rt- ,A) ) + preac. (9)

2
We assume that the dissipative percussion results from a pseudo-potential
of dissipation </>d
ftd E &</>d( D(71+, rt+, A) ; D(71-,rt- ,A)). (10)

37
Let us recall that a pseudo-potential of dissipation, as introduced by

Moreau, is a convex function which is positive with value zero at the
origin. The impenetrability reaction is active only when the risk of
interpenetration is present, i.e. when the contact is persistent after the
collision. Thus
(11)
where the normal vector N is directed upward (figure 1). Let us define
the set
K -_ {Ix x>
i5(U-, It-,A). N}
. (12)
- 2
Then relation (11) is equivalent to
preac E 8IK( i5(U+, It+, A) ; i5(tJ-, It-, A) . N), (13)
where IK is the indicator function of the set K. Thus by defining the

functions
(14)
and
¢=¢d +IK,
-
(15)
the constitutive law (9) becomes
pint E 8¢(i5([J+, It+, A) ; i5(U-, It- ,A)). (16)
Let us note that the pseudo-potential ¢ depends on the velocity of de-

formation before the collision via the function i K .
Examples of constitutive laws depending on a pseudo-potential.

The following examples concern only the normal percussion: the tan-
gential percussions are zero. The first example shows the equivalence
between the use of a linear constitutive law and the usual use of a resti-
tution coefficient for the collision of a point (example of figure 1), whereas
the second example shows how we modelize collisions with adhesion.
A very simple pseudo-potential is a quadratic function ¢d(i5) = (1/2)
kN(i5.N)2, which corresponds to a linear dissipative percussion
(17)
for a point. In the case of the collision of a point with a plane (figure 1),
one can easily show, cf Dimnet (2001), that the linear percussion leads
38
to a normal restitution coefficient e = (kN - m)/(kN + m) if kN >m

and e = 0 if k N ::; m.
-c
Figure 3. Constitutive law describing a collision with adhesion.
..
.u X10-a
.0
:+
:u ..
,.
I .~ : .:~ \- •. :
::.,t+.: ' •~
.~--------~--~--~---~
3 ....
0.1
":1
.:
.. ..
. !t.:
...
0.1
o. . :
:
0.5
00 •
.
.. ";41 ...
:
0.1
,
0.2 0.3 0.4 0.5 D.. 0.7 0.' G.I 1

00
.• ----.,----.J
O·~~---oC_U'---_~'------c....'=-.--;,----"""'U-~--,,7- •.•
Figure 4. Experimental measurements: The normal part of the percussion, PN ,

versus the normal part of D, DN' on the left side and the ratio PN / PT versus the
tangential part of D, Dr' on the right side.
Collisions involving adhesion are collisions such that the percussion

has to be large enough for separation to occur after. The graph shown
in figure 3 has this property (kN is the slope of the linear parts of the
law assumed greater than m (otherwise contact of the solids is always
maintained in collisions)). The two parameters C and II characterize the
properties of adhesion:
• if the relative velocity before the collision is small k:'!::mC) (InNI : ;
the adhesion forces are strong enough for the solids to remain in
contact after a collision.
39
• if two solids are in contact, an exterior percussion has to be larger
than II to separate them, see Dimnet (2001).
A constitutive law which does not result from a pseudopoten-

tial: the Coulomb law. Experiments of collisions in two dimensions
of dry rigid solids with a plane (dry angular steel particles colliding a
dry marble plane, (see figure 4) show that the normal behaviour can be
modelized by a pseudo-potential: PN(DN) E a~N(DN)' even quite well
by a quadratic one. But as far as the tangential behaviour is concerned,
PT = P .T (T is the tangent vector in the plane of experiments) can not
be considered as a function of the only variable DT = D . T: it depends
also on PN = p. N (see figure 4). This behaviour is described by
PT :::; /-LPN, /-L > 0 and

if PT < /-LPN then DT = IT,
if PT = /-LPN then :3A > 0 such as
(18)
. . . . . . + ---=++
D(U ,n ,A)T +D(U ,n ,A)T = -APT,
-# ..... - -:::+- .....
This constitutive law, called the Coulomb law, does not derive from
a pseudo-potential.
2.2 Simultaneous collision of a collection of rigid

bodies.
Figure 5. Numerical simulation of the crumbling down ofa collection of rocks.
When more than two solids collide together at the same time, the
relative velocities of all the couples of colliding solids are discontinuous, it
is a simultaneous collision. This situation happens very often in nature or
40
in numerical simulations of the evolution of collections of solids. Figures
5 and 6 show examples of such simulations.
Figure 6. Numerical simulation of the filling of a hopper with polyhedrons.
Model of simultaneous collisions. Consider N solids with mass

mi, center of mass is Gi mass moment of inertia tensor are Ii colliding
together at time t. We assume that all the contacts are reduced to a
point. But multiple collisions can occur between solids i and j at points
A,j,k' The set Si,j contains the points of contact of the two solids. If
the two solids do not collide, this set is empty. The percussion ~intk
is applied at the collision point Ai,j,k' Let the virtual velocities of iiie
centers of mass Gi be ~ and the virtual rotation velocity be Wi . Let us
define the vector V = (~, Wi) and the function
Di,j(V, Ai,j,k) = ~ +Wi X GiAi,j,k - (~+Wj x GjAi,j,k), (19)

which gives the velocity of deformation at point Ai,j,k (the relative ve-
locity of the point). The principle of virtual work may be written
41
By using the constitutive laws (16)
P!11:t E Dcp .. (Di,j(U+,Ai,j,k)) + Di,j(U-,Ai,j,k)) (21)

~,J,k ~,J,k 2 '
which take into account the impenetrability condition, the principle (20)
becomes
In order to use law (16), we assume that there exist a normal vector
directed from solid j toward solid i with i < j.
This implies that one of the solids has to be regular at the contact
point but. not both. The more general situation where irregular solids
collide afangular vertices is studied in Dimnet and Fremond (1999a),
Fremond(2000) and Fremond (2001).
The function
(23)
is convex, positive, with value zero at the origin (it is a pseudo-potential).

Thus the function
(24)
is also a pseudo-potential. Let us define the scalar product
(U, V) = L {mJTi· ~ +IJ2i .,:q.

N
(25)
i=l
Thus relation (22) is
( fj+ - (r, V - fj+ ; fj-) 2 <1'>( fj+ ; fj- ) - <I'>(V), (26)

42
or by having R 6N equipped with the scalar product (.,.)

~+ ~-
-("0+ - 0-) E 8~(U ; u ), (27)
or by letting
0++0-
X= 2 ' (28)
20- E 2X + 8~(X). (29)

The equation (29), assuming 0- is known, has one and only one solution
because the operator X --t 2X + 8~(X) is maximal, surjective and
strictly monotone. It is classical that the equation (29) is equivalent to
the minimization problem
(30)
The interior percussions are described by the Coulomb law.

If we set in Coulomb law (18) the normal percussion PN at its ac-
tual value, X, the constitutive law results from a pseudo-potential of
dissipation. Thus finding X is equivalent to solving the equation
where the vector PN(X) contains the actual (but unknown!) values of
the quantities /-LPN which intervenes in the tangential constitutive law
(18). One can try to solve this problem by an iterative method
+ ~(PN(Xn), Y) -
X n +1 = Arglnf {
y2
Iy E R 6N
(20-, Y) }
.
(32)
It can be shown, cf Dimnet (2001) that the method converges if the

coefficients /-Li,j,k of law (18) are not too large.
References
C. Cholet, (1998a), Chocs de solides rigides, these de I'Universite Pierre et Marie
Curie, Paris.
E. Dimnet, M. Fremond, (1999a), Chocs de solides rigides, 4e Colloque national de
calcul des structures, Giens.
E. Dimnet, M. Fremond, (1999b),Choc instantane d'un solide deformable sur un sup-
port rigide, 14e congres franc;ais de mecanique, Toulouse.
43
E. Dimnet, M. Fremond, (2000), Instantaneous collisions of solids, European Congress

on Computational Methods in Applied Sciences and Engineering, Barcelona.
E. Dimnet, M. Fremond, R. Gormaz, J. San Martin, (2001), Collisions involving solids
and fluids, in Novel approaches in civil ingeneering, Springer Verlag Heidelberg
2ool.
E. Dimnet, (2001), Chocs de solides et integration des equations d'evolution des
systemes multi-solides, these de l'Ecole Nationnale des Ponts et Chaussees, in
preparation.
M. Fremond, (1995), Rigid bodies collisions, Physics letters A, 204, 33-4l.
M. Fremond, (2000) , Collision of a wedge with a plane, Computational and Applied
Mathematics, Vol 19, N°2, pp1-1O.
M. Fremond, (2001), Non-smooth Thermomechanics, Springer-Verlag, Berlin.
P. Germain, (1973), Mecanique des milieux continus, Masson, Paris.
J. J. Moreau, (1966), Fonctionnelles convexes, Seminaire sur les equations aux derivees
partielles, College de France, Paris.
IMPACTS WITH GLOBAL DISSIPATION INDEX
AT REENTRANT CORNERS
Christoph Glocker
IMES - Center of Mechanics, ETH Zentrum
CH-8092 Zurich, Switzerland
Abstract A geometric interpretation of Moreau's frictionless multi-contact impact law is

presented and extended to the case of reentrant corners. The methods used are
based on the geometry of cones. The main construction of the impact law for the
convex case is done by the unique orthogonal decomposition of the pre-impact
velocity wi th respect to two orthogonal closed convex cones, one of them approxi-
mating the non-smooth boundary of the admissible domain in some neighborhood
of the point of impact.
1. Introduction
Newton's classical kinematic impact law provides a way to calculate the post-
impact velocities of two colliding particles. It reverses the sign of the relative
velocity 'Y at the impact and takes into account dissipation by a coefficient of
restitution € such that 'Y+ = -€ 'Y-. When this impact law is applied to multi
contact problems, a formulation in terms of inequalities is required, such as done
by Pfeiffer and Glocker (1996). Here we assume in addition that"the coefficients
of restitution of the different contacts are all equal to each other. Such an event
is then called a Newtonian impact with global dissipation index, and it agrees
with Moreau's non-smooth dynamical equations, see e.g. Moreau (1988), when
the latter are evaluated at a single instant of time. We present a geometric
interpretation of this class of impacts which is based on the decomposition of
the pre-impact velocity with respect to a pair of orthogonal convex cones. Most
of these results are already found in Moreau (1988). We have put them together
to find a natural generalization to the non-convex case of re-entrant comers
which is addressed at the end of the paper. The resulting formulation goes far
beyond simple unilaterally constrained motion and includes, for example, the
situation of two rectangular blorks hitting each other at their comers.
45
J.A.c. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 45-52.
46
2. Cones and Convexification
In this section some basic definitions, notations and properties of cones used
in non-smooth analysis are put together. These cones characterize sets in the
neighborhood of a chosen point and indicate whether convexification, as the
next step available after linearization, is possible or not. As a standard tool for
inequality systems, these cones will be used throughout the paper. The material
is mainly taken from the books by Aubin and Ekeland (1984) and Rockafellar
(1972), but applied to finite-dimensional manifolds.
Let M be a proper n-dimensional Riemannian manifold. We denote the
tangent space to M at q E M by TqM, the cotangent space by r: M, and the
duality pairing between r:M and TqMby (', '}q. The inner product and the
norm on TqM induced by the metric on M is denoted by (', ')q and 1I·lI q , and
the corresponding expressions on r: M by (', .); and 11·11;, respectively. We
will also need the natural vector bundle isomorphisms ~ : T M -+ T* M and
U= ~-1.
Let V denote the set of admissible points on M which is assumed to be a
closed subset of M with piecewise smooth boundary avo
In order to charac-
terize the behavior of av in some neighbourhood U of a point q E V we adapt
the definitions of the contingent and the tangent cone in Aubin and Ekeland
(1984) to manifolds. We denote the contingent cone to the set Vat the point
q E V by KqV, and the tangent cone by CqV, respectively. Then
whereqn ~ qdenotes the convergence ofqn toqinV, andexpp : TpM -+ M,

v -+ 'Yv(l) is the exponential map of M at p which determines the point
p = 'Yv (t = 1) on M that is passed at t = 1 by a geodesic 'Yv (t) emanating
from point p = 'Yv(t = 0) with velocity v = 'Yv(t = 0). Note that 'Yhnu n (1)
constitutes a sequence of points Pn converging on 'Yv to P = 'Yv(t = 0) when
Un = V is fixed and hn ..j.. O. This ensures the existence of precisely one
geodesic connecting 'Yv (0) and 'Yhn Un (1) for sufficiently large n.
Both cones in (1) are closed. The tangent cone CqV is convex, whereas
convexity must not be expected for KqV. One always has CqV C KqV C
TqM which is obvious when setting ql = ... = qn in (1). V is said to
be tangentially regular at a point q if KqV = CqV; otherwise we call q a
reentrant corner point of V. Tangential regularity applies in particular for the
interior points of V because then CqV = KqV = TqM, but also for smooth
portions of av for which we have TqaV = -CqV n CqV (and CqV = KqV),
and finally for smooth submanifolds V with CqV = KqV = TqV.
47
Figure 1. The cones Kq V, CqV and ci-v at a point q EVe M.
Finally let us set up a pair of closed convex cones (CqV, ctV) orthogonal
to each other via the variational inequality
CfV := {v E TqM I (v, u}q :::; 0 \lu E CqV}. (2)
For convenience, we denote the elements of ctv by vl. when v are the corre-
sponding elements of CqV. One observes that CqV = TqM {:} C;V = {O}
and vice versa. If CqV is a half-space then CiV degenerates to a ray orthogo-
nal to the boundary of Vat q. If CqV is a subspace of TqM, then ctv is the
classical othogonal vector space complement in TqM. According to (2), the
elements of (CqV and ctV) are characterized by the inequality
(3)
Figure I shows the cones Kq V, CqV and ctv for the situations of a tangentially
regular set and a re-entrant comer. More examples on how these cones look
like may be found in (Glocker, 2001)
3. Orthogonal Decompositions
In this section we review briefly the decomposition of a vector with respect to
a pair of orthogonal convex cones and discuss some other related formulations.
The following statement is contained as a special case in Moreau's Theorem,
see e.g. Rockafellar (1972) for the full version and the proof, and generalizes
the classical orthogonal vector decomposition.
Theorem 3.1. Let U be a finite-dimensional real inner product space, and
(R, Rl. ) an orthogonal pair of closed convex cones in U. Any u E U can then
be decomposed uniquely into a sum u = v + vl. such that
(4)
48
Figure 2. Orthogonal cones R, RJ. and the orthogonal tangent cone of Rat (.), Cf)R.
This decomposition is depicted in the left part of Figure 2. For u ¢ R URi. one
obtains v =f. 0, vi. =f. 0, whereas vi. = 0 or v = 0 as soon as u E R or u E Ri..
Apparently is v the nearest point to u in the set R, which is usually denoted by
v = ProxR(u). Thecorrespondingmapu -t proxR(u) is called aproximation
which is, in fact, a projection because proxMu) = proxR(u). In addition, this
projection is orthogonal since (proxR(u) , (u - ProxR(u))) = O. Of course,
the same properties apply for the second term vi. in the decomposition due to
symmetry, i.e. vi. = proxRJ.(u) .
Proposition 3.2. Let U, Rand Ri. be defined as in Theorem 3.1. Then (4) is
equivalent to any of the two conditions
(5a)
(5b)
where C~ Rand C;J.Ri. denote the cones orthogonal to the tangent cones to
R at v and Ri. at vi., respectively.
The proof may be found in (Glocker, 2(01). The cone C~ R is depicted in the
right part of Figure 2 for some elements v E R.
4. The Impact Equations

In this section the impact equations of the system are stated, following the
approach of Ballard (2000). We start with a motion in the interior of V which
is classically governed by the second order differential equation
~q(t) ~u(t) = f(q(t), u(t)) . (6)
Here, ~u(t) E Tq(t)M denotes the covariant derivative ofu along the curve
q(t) with velocity q(t) = u(t) E Tq(t)M , and f E T;(t)M are the applied
49
forces. Note that Tq(t)M = Cq(t) V holds in this particular situation, because
q{t) E int V by assumption.
Suppose now that q{t) from (6) reaches the boundary of V. In order to
ensure that q{t) does not leave V, a discontinuity in the velocity u{t) has to be
permitted. To include such events in the formulation of the dynamics, equation
(6) is rewritten as an equality of measures (Moreau, 1988)
~q(t) Du = f{q{t), u{t)) dt + dR, (7)
where Du is the covariant differential measure of the velocities u{t) which

are now assumed to be of bounded variation, and dR is the percussion force
measure which takes into account the (probably impulsive) reactions.
In order to obtain the impact equations, (7) is integrated over a singleton {t}
which yields
u+ (t) - u- (t) = ~q(t) R(t) (8)
with u+(t) and u-{t) the right and the left limit of u{t), and R{t) the impact
percussion. In accordance with perfect constraints, the most natural cone of
possible reactions and percussions is ~q(t) C~t)V, hence we claim
(9)
which incorporates already one part of the impact law: For q{ t) E int V one has
C~t)V = {O},henceR(t) = Oandu+(t) = u-(t) by (8), i.e. no impact. Note,
however, that (9) together with (8) is still insufficient to determine uniquely
a post-impact velocity u+{t) or, in other words, to single out one particular
element -~q(t) R(t) from C~t)V. The missing information will be added in
the next section by means of constitutive equations, called the impact law.
5. A Geometric Impact Law

We give now a geometric interpretation of Moreau's impact law (Moreau,
1988) which is a unilateral version of Newton's classical kinematic impact law,
see Glocker (2001) for a detailed discussion. For brevity, explicit dependence
on time t is suppressed.
With u- being the pre-impact velocity of the system, we propose the fol-
lowing geometric construction of the impact law: According to Theorem 3.1,
perform an orthogonal decomposition of u- into
(10)
such that the vectors v and v.l satisfy
(11)
50
Figure 3. The geometry of impacts with global dissipation index.
This decomposition is unique. The tenu v plays the role of the tangential com-
ponent of u- which remains unchanged by the impact. The normal component
vol is "inverted" by the impact rule V X := -EVol, where E is the global coeffi-
cient of restitution, a magnitude expressing the global dissipation behaviour of
the impact when chosen between 0 and 1. The post-impact velocity is then set
to be u+ := v + V X which yields
u+ =V - EVol (0 ~ E ~ 1), (12)
according to the construction shown in Figure 3, see e.g. also Brogliato (1999).
By taking the difference of (12) and (10) the impact percussion R in (8) is
identified as
(13)
From Figure 4 one recognizes that the impact law is always energetically con-
sistent: The kinetic energy T satisfies 2 T+ = lIu+ II~ ~ lIu-ll~ = 2 T-,
where equality holds for E = 1 and maximal dissipation is achieved for E = O.
In tenus of a minimization problem v and vol are the nearest points to u-
in the sets CqD and CiD, respectively. The corresponding maps are called
proximations and are denoted by
(14)
For example, the impact law (10)-(12) might equivalently be stated in tenus of
proximations as
(15)
51
when the first equation in (14) is used. Further, we recognize that the proxima-
tion in (15) becomes the identity whenever u- E Cq'D. In this case u+ == u-,
thus no impact occurs.
6. Extension to Re-entrant Corners

Finally we extend the impact law (10)-(12) to re-entrant corner points. Note
that such a situation can not be generated by the intersection of smooth simple
unilateral constraints, but occurs in practice when, for example, two rectangular
blocks hit each other at their corners. The generalization of the impact law
is done in two succeeding steps: By virtue of Proposition 3.2 and equations
(4), (5a) we first leave the symmetric formulation (11) and replace it by the
equivalent conditions
v E Cq'D, vi. E C;-Cq'D c civ. (16)
In a second step, the convex tangent cone Cq'D determining the set of admissible
post-impact velocities in the regular case has to be replaced by the (larger non-
convex) contingent cone KqV in order to access the whole space of physically
admissible velocities for re-entrant corner configurations. This yields
v E KqV, vi. E C;-KqV C CfV (17)
and is together with (10), (12) already the desired generalization. The geometri-
cal meaning of this impact law is sketched in Figure 4. Note that the orthogonal
decomposition as performed in (11) no longer applies, because convexity of the
two participating cones is required in Theorem 3.1. A representation in terms
of proximal points also fails: Due to the lack of convexity of KqV one is no
longer able to express (17) as a minimization problem. The extended statement
is: Find the stationary points of the function
(18)
where IN(-) is the indicator of the set N. The solution set of this problem
contains, among others, the proximal points to u- in KqV and is set up by the
zeros of the generalized gradient of (18). These are the values of v in KqV
which satisfy
- - i.
DE 8<pq{v) = bq (v - u-) + 8IKq V{V) = bq (v - u-) + bq Cv KqV, (19)
i.e. the impact law (17).
Due to the non-convexity of Kv{qo) one can not expect the solution set of
(19) to consist of one element only. Non-unique post-impact velocities must be
allowed to occur for certain configurations, such as those discussed in (Glocker,
2001). An impact law for re-entrant corners was also introduced in Kane et
al. (1999) in the sense that (18) is taken as a minimization problem leading,
however, to a reduced set of solutions.
52
Figure 4. Impact at a re-entrant corner.
7. Conclusion
In this paper a geometric interpretation of two extensions of Newton's kine-
matic impact law were presented: Moreau's multi-contact impact law for fric-
tionless unilateral constraints, and a version which is capable of handling even
the reentrant corner problem. Although clearly originating in the collision prob-
lem of rigid bodies under simple unilateral constraints, these methods are useful
for many impact problems in general finite-degree-of-freedom dynamics.
References
Aubin, J., and Ekeland, I. (1984) Applied Nonlinear Analysis, John Wiley & Sons, New York.
Ballard, P. (2000) The dynamics of discrete mechanical systems with perfect unilateral con-
straints, Arch. Rational Mech. Anal. 154, 199-274.
Brogliato, B. (1999) Nonsmooth Mechanics, 2nd ed., Communications and Control Engineering
Series, Springer, London.
FIimond, M. (1995) Rigid bodies collisions, Physics Letters A 204, 33-41.
Glocker, Ch. (2001) A geometric interpretation of Newtonian impacts with global dissipation
index, Cahiers Stephanois de Mathematiques Appliquees Publication de l'Equipe d'Analyse
Numerique 2, UPRES EA 3058, Universite Jean Monnet, Saint-Etienne, online publication
at http://wwwean.univ-st-etienne.fr. 1-14.
Kane, c., Repetto, E.A., Ortiz, M., and Marsden, J .E. (1999) Finite element analysis of nonsmooth
contact, Compo Methods Appl. Meck Eng. 180, 1-26.
Moreau, J. J. (1988) Unilateral contact and dry friction in finite freedom dynamics, in Non-
Smooth Mechanics and Applications, edited by J. J. Moreau and P. D. Panagiotopoulos,
CISM Courses and Lectures 302, Springer, Wien, 1-82.
Pfeiffer, F., and Glocker, Ch. (1996) Multibody Dynamics with Unilateral Contacts, John Wiley
& Sons, New York.
Rockafellar, R. T. (1972) Convex Analysis, Princeton Univ. Press, Princeton, New Jersey.
UNDERSTANDING IMPACT THROUGH
CONTINUOUS MEDIUM VIBRATIONS
Lretitia Paoli
UMR CNRS MAPLY and Equipe d'Analyse Numerique (EA9058) Faculte des Sciences,
Universite Jean Monnet, 42029 St-Etienne Cedex 2 France
Michelle Schatzman
UMR CNRS MAPLY
Universite Claude Bernard - Lyon 1, 69622 Villeuroonne Cedex France
Abstract We write a model for the impact of a slender bar making an angle
(J with a rigid foundation, starting from first principles. We obtain
some explicit solutions, with the help of asymptotics and simplifying
assumptions. We compare the corresponding results to the experiments
of Hurmuzlu and Stoianovici, and we obtain good qualitative agreement
in all the cases and good quantitative agreement for 5 out of 8 of the
bars in their experiments. We rule out the role of inharmonicity due
to heterogeneity as a cause for loss of macroscopic energy when the bar
makes the angle 7l" /2 with the rigid foundation, concluding thus that
this loss is mainly due to plastic deformation.
1. Introduction
In a seminal series of experiments Stoianovici and Hurmuzlu, 1996
dropped a slender bar onto a massive rigid foundation while the initial
position of the bar made an angle (J with the horizontal. They found
that the apparent coefficient of restitution, defined aB the ratio of the
post-collision to pre-collision vertical velocities, measured with the help
of a high-speed camera system, depended strongly on (J.
In numerical simulations presented in Paoli and Schatzman, 1999, we
have obtained results with an excellent qualitative agreement and a good
quantitative agreement with those of Stoianovici and Hurmuzlu, 1996;
in our discrete model we considered a frictionless rigid contact on the
foundation, while Stoianovici and Hurmuzlu, 1996 included a nonlinear
damped model of the contact with the foundation.
53
l.A.C. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 53-64.
54
The analysis of Hurmuzlu, 1998 offers a way of understanding the
relation between continuous medium modes and restitution: it is based
on energy arguments, giving an ansatz for the form of the elastic en-
ergy stored into the continuous media modes, and then identifying the
relevant parameters for one set of geometrical data; these values are
subsequently validated by comparing experiment to theoretical results
in other experiments.
This explanation is quite satisfactory, since it accounts well for the
experimental observations; however, it contains an argument which we
do not understand completely, namely the choice of the ansatz. In this
article, we give a different and more theoretical argument, relying on
the analysis of the dynamics of the system in infinite dimension, i.e.
continuous medium modelling. This analysis does not answer all the
questions, but we hope that it sheds a bit more light on a problem
which would deserve to become classical.
The article is organized as follows: in section 2, we establish a La-
grangian formulation for the impact of the bar; this formulation in-
cludes the continuous medium modes; there are two different kinds of
constraints: one type of constraints is the non penetration constraint;
we have to impose three equality constraints to remove any indetermi-
nacy in the coordinates. All these constraints will be treated through
the use of Lagrange multipliers. We obtain thus a complicated system
of ordinary and partial differential equations with unilateral constraints.
In order to get to practical results, we make approximations and sim-
plifying assumptions in section 3; we estimate the physical quantities in
terms of the small parameter which is the ratio c of the vertical velocity
of the bar before impact to the velocity c of compressive waves in the
beam. We also assume that the reaction force due to the non penetration
constraint is constant throughout the first microcontact time interval;
in order to prove that this is close to truth, much mathematics might be
needed. Nevertheless, this is a simple enough assumption, which leads
to a number of definite conclusions: first, we obtain a kinetic coefficient
of restitution when there is only one contact interval; this coefficient is
qualitatively correct for the eight steel bars which have been tested by
Stoianovici and Hurmuzlu, and quantitatively correct for the five longer
ones.
In section 4, we go beyond the first microcontact time interval and
we produce an approximation of the solution, which enables us to find
the smallest initial angle for which there is a second micro contact time
interval soon after the first. We also obtain qualitatively correct results
for all bars, and they are quantitatively correct for the five longer ones.
55
2L/e
L/e
x -L/2 o L/2 x
Figure 1. Choice of axes and coordi- Figure 2. The regions involved in
nates for the problem. the definition of the function it in (17)
A crucial element of the analysis developed here is that no energy re-

mains stored in the compressive modes after the end of the first contact;
the argument underlying this conclusion is based on an explicit solution
of the wave equation with constant coefficients; should there exist any
inhomogeneity of material or geometrical origin, this conclusion is no
more warranted.
In the short section 5, we claim that for small heterogeneities, this
statement is still approximately true. The details are not given here for
lack of space.
2. The Lagrangian formulation for impact

We assume that all motions of the slender bar are planar. Its cross
section is circular, with radius R, and its length is L. Young's modulus is
called E and the mass density of the material p. The area of the section
is A = 11'R2 and the moment of inertia for the cross-section of the beam
is I = 11'R4 /4. The wave velocity is e = J E / p; finally we let ", = J I / A.
Let x denote the longitudinal coordinate along the bar, with origin at
its middle; let u(x, t) and v(x, t) be respectively the longitudinal and
the transversal deformations of the bar; in the fixed reference frame, the
coordinates of the center of mass are denoted by X and Y. This implies
that the x averages of u and v vanish:
L/2 jL/2
j u(x, t) dx = v(x, t) dx = o. (1)
-L/2 -L/2
56
The bar makes an angle 8 with the horizontal axis and the vanishing
moment condition
L/2
/ xv (X, t) dx = 0, (2)
-L/2
removes the indeterminacy on the angle 8. Figure 1 pictures the choice

of coordinates.
After adding Lagrange multipliers related to the equality constraints
(1) and (2), the Lagrangian is given by
C = pAL
x2 +2 y2 + "211L/2 pA[(u - . ..
vO)2 + (u + x)202 + 2uvO
-L/2
+ v 2] dx - "211L/2 [EAu~ + Elv~x] dx + I-Ll 1L/2 udx (3)

-L/2 -L/2
L/2 lL/2
+ 1-L2 l V dx + I-La xv dx.
-L/2 -L/2
Define cp(Y,u,v,8) = Y + (u - L/2) sinO + v cos 8; then the non-

penetration constraint is
q>(Y, u( -L/2, ·),v(-L/2, .), 0) ~ o. (4)
Then, the motion of the bar can be, at least theoretically, described
as an extremum of the action A = f C+ dt, under the constraint (4);
in fact, when the constraint is satisfied, this extremum of the action
is a maximum; in order to understand why this must be so, it suffices
to consider a linear case, where the sense of the variational inequality
corresponds to a maximum of the action.
When the constraint (4) is not active, the coefficients of each varia-
tion must vanish; when the constraint is active, the variations of Y, 8,
u( -L/2, t) and v( -L/2, t) must satisfy the following inequality
8Y + 8u( -L/2,.) sinO + (u( -L/2,·) - L/2) 80 cos 0

+ 8v(-L/2,·) cos 8 - v( -L/2, ·)80 sinO ~ o.
Therefore, when the constraint is active, there exists a non negative

Lagrange multiplier, A, having the dimension of a force, such that the
57
following relations are satisfied:
pALl' = A, (5)
d
pA-
dt
l L/ 2
-L/2
•
[v(vO - u)
•
+ O(u + x)2 + uti] dx (6)
= A((u(-L/2,.) - L/2) cosO - v(-L/2,·) sinO),

AEux (-L/2,·) = -AsinO, (7)
Elvxxx(-L/2,·) =A0080. (8)
When the constraint is not active, A vanishes in (5)-(8). The remaining
equations are:
pA(utt - 2iJvt - 8v - iJ2(u + x») - EAu xx = 1-'1, (9)

u x (L/2,.) = 0, (10)
pA(vtt + 2iJut + (u + x)8 - viJ2) + Elv(4) = 1-'2 + 1-'3X, (11)
vxxx (L/2,·) = vxx (-L/2,·) = vxx (L/2,·) = O. (12)
The values of 1-'1, 1-'2 and 1-'3 can be estimated by integrating over
[-L/2, L/2] respectively (9), (11) and (11) multiplied by x. We find
thus
AsinO ACOSO
l
1-'1 = --L-' 1-'2 = --L-'
(13)
L3 L L/2 •..
1-'3 12 = 2"ACOSO + pA x(20ut + uO) dx.
-L/2
3. Approximation and resolution of the equation

during the first contact interval
At this point, we make approximations which we do not justify math-
ematically; however, they enable us to find some results: assume that
initially all the velocities vanish, except y(O), which is strictly nega-
tive; the most important scale is £ = I"V(O)I/c. In the experiments of
Stoianovici and Hurmuzlu, 1996, £ was less than 10-3 . Since we will
be interested in what happens during a contact interval, we will assume
that the constraint q> is active during an interval [0, T], and that A is
constant throughout this time interval. If the impact were vertical, we
would expect A = EA£; we include one more unknown parameter by
letting
A=EA£(, (14)
where ( will be determined as a result of the foregoing calculations. We
will also neglect the variation of 0 throughout the time interval [0, T].
58
The time r, itself, can be found on physical grounds: it is the time needed
by compressive waves to go from one end of the bar to the other and
back, i.e. r = 2L / c. Therefore, we expect that all the velocities X, Y, u,
iJ and LO will be of order EC, that the corresponding accelerations will be
of order EC2 / L and that the difference between the spatial coordinates
at time t E [0, r] and 0 will be of order EL. These approximations can
be justified by qualitative arguments; it would require a complete and
heavily mathematical paper to give a fully rigorous justification.
Under these assumptions, we keep the first order terms in E, and
equations (9) and (11) become, with the help of (13):
Asin8
pAutt - EAu zz = - -L-' (15)
(4) Acos 8 6x
pAvtt + Elv = --L- + L2ACOS 8 . (16)
We complete these equations with the respective boundary conditions

(7), (10) and (8), (12) and with zero initial conditions for the position
and velocity. Both equations can be solved fairly explicitly; for (15), on
the time interval [0, r] we get a solution in closed form:
u{x, t) = (E(U{X, t) - c2t 2/2L) sin 8,

where U is given on the domain of interest by
o if (x, t) E R 1 ,
u{x, t) = { ct - x - L/2 if (x, t) E R2, (17)
2{ct - L) if (x, t) E R31
with the regions Rl, R2 and R3 of the plane being depicted in Fig. 2.
It is important to observe that
u{x, r) = Ut{x, r) = o. (18)

The value of the solution v is more complicated; we need to introduce
the eigenfunctions of the beam operator v t-+ v(4) with two free ends on
[0,1]; for this purpose, we define the numbers ek as the strictly positive
solutions of cosh ek cos ek = 1; it is a well-known and classical fact that
ek (2k + 1)1r/2 for all k = 1,2, .... It is possible to define normalized
fV
eigenfunctions satisfying the following identities:
¢i
4) = et¢k, ¢Z{O) = ¢Z{I) = 0, ¢Z'{O) = ¢Z'{I) = 0,
(19)
111¢k12dx = 1, ¢k{O) = 2.
59
The space L2(0, 1) is spanned as a Hilbert space by the functions ¢k,

k = 1,2, ... , and the functions 1 and x. The normalized eigenfunctions
for the beam operator on (-L/2, L/2) are then
1
tPk(X) = -¢k
..fL
(x-L + -21) (20)
We call ¢ the unique polynomial ¢ which is orthogonal to 1 and x

and satisfies ¢",( -L/2) = 1, ¢"( -L/2) = ¢"(L/2) = ¢/"(Lj2) = 0; its
second derivative is ¢"(X) = x+(L/2) - (2(x+L/2)2 / L)+(x+L/3)3 j L2.
The decomposition of ¢ on the eigenfunction basis can be calculated very
easily; its components are (¢, tPk) = 2L 7/2 k4. e
With the help of (14), equation (16) can be rewritten as
Vtt + /'1,2C2v(4) = c(c2 cos 8¢(4) (21)
with boundary conditions (12) and
v'" (_ L/2, t) = c( c~s 8 . (22)

/'1,
We seek a solution of (21), (12) and (22) under the form v = w +

c( cos 8¢j /'1,2 and we find the following result, where Wk is the eigenfre-
quency
(23)
We use now relations (5) and (6) to find Y and 0, where we neglect
the quadratic terms in c; we find
.. c2
Y = c( L' (24)
.. ~
8 = -6 L2 c( cos 8, (25)
from where we infer the relations

Y(T) - Y(O) = 2Lc(( -1), 8(T) - 8(0) = -12c(cos8(0). (26)
Of course, we have used the initial values Y(O) = -cc and 0(0) = O. In
order to determine (, we use the constraint (4); in the formula for the
difference of q, at times T and 0, we keep only the terms of order 1 in c,
which leads to
L
Y(T) - Y(O) - "2 cos 8(0)(8(T) - 8(0)) + v( -L/2, T) cos 8(0) = 0, (27)
60
since u( -L/2, r), u( -L/2, 0) and v( -L/2, 0) vanish. We substitute into
(27) the values given by (26); according to (23) and (19), (20), the value
ofv(-Lj2,r) -v(-Lj2,0) is
-Lj2 ) = 4L 3 t::( cosO '" 1- cos(2~~",jL)

v( ,r ",2 L- t4 .
k>1 "'k
Define
_ 2L 2 " , 1- coswkr.
J-t - ",2 L- t4 '
k~1 "'k
after straightforward simplifications we obtain the equality ( = (1 + (3 +
J-t) cos2 0(0))-1. From this relation, we may now deduce the kinetic or
apparent coefficient of restitution; arguing as for the impact of a rigid
bar, it is the ratio between the vertical velocities of the impacting end
before and after impact, i.e. the ratio
e-
Y(r) - O(r)L(cosO(O))j2 .
- IY(O)I '
we obtain finally
1 + (3 - J-t) cos2 0(0)
e= -~-..:......:..--:::--~ (28)
1 + (3 + J-t) cos2 0(0)'
4. The second microcontact time interval

Relation (28) is valid if there is only one micro contact time interval.
Let us find when and for which initial 0 there might be a second
microcontact time interval, i.e. whether the constraint (4) can be active
at a later time, very soon after the end of the first micro contact time
interval. After the time r, and as long as the constraint is not active,
the Lagrange multiplier >. vanishes; denoting by (r, a) the maximum
interval on which the constraint is not active, conditions (18) imply that
u vanishes throughout the interval (r, a). Thanks to (5), the acceleration
of Y vanishes and so does the acceleration of 0, up to second order terms,
thanks to (6). Therefore
Y(t) = Y(r)+cc:(2( -l)(t-r), O(t) = O(r) -12cc((t-r)(cosO(0))/L.

The function v solves the partial differential equation
(29)
61
Table 1. The critical angles and times for the second microcontact time interval
bar number radius length ()c (degrees) t' U - T (10- 3 8)

1 6.35mm 63.5mm 57.04 1.29 0.03
2 6.35mm 100mm 64.64 2.29 0.08
3 6.35mm 200mm 70.36 5.44 0.42
4 6.35mm 300mm 72.80 8.24 0.96
5 6.35mm 400mm 74.30 11.12 1.72
6 6.35mm 600mm 76.09 17.60 4.09
7 3.2mm 400mm 77.17 22.62 3.50
8 3.2mm 600mm 78.57 33.68 7.82
with free end boundary conditions, and initial data at time T deduced
from (23). Therefore, v( -L/2, t) - v( -L/2, T) is given explicitly by
v{ -L/2, t) - v( -L/2, T)
_ r 4L3 O~ COS Wk(t - 1 + COSWkT
- C.. ",2 cos L....J 4 e
T) - COS(Wkt) -
•
k~l k
Up to second order terms, the time u = T(t' + 1) is defined by the

condition
(2( - l)t' + 6(t' cos2 00
r
+ .. cos
211 2L2
170 ",2
~ COSWkTt' - COSWkT(t'
L....J
k~l
4
k
e + 1) - 1 + COSWkT _
-
0
•
(30)
We sought numerically the value Oc of 0 and t' for which (30) vanished
for the first time; the results are given in table 1 for Stoianovici and
Hurmuzlu bars, which are made of steel with E = 2.11011Pa and p =
7.86kg x m- 3 .
In order to compare our theoretical results with the experimental
measurements of Stoianovici and Hurmuzlu, 1996 and Hurmuzlu, 1998,
we have represented at figure 3 the plots of the coefficient of restitution
e given by (28) on the angle interval [Oc, 71'/2]; on the plot, the angle
is measured in degrees. This figure is to be compared to figure 12 of
Hurmuzlu, 1998, and we see that for bars 4 to 8, our results seem closer
to the experimental results than the numerical results of Stoianovici and
62
. e /'
e
/
r- ~
'r- bo" 8.2
/
/
(J (J
e e
/
= 80"
u-
IL7 _ Bill: 4
(J
e e
r-
I- a..,S
r- ... 6
I-
e e
ur-
u r-
r-
Bar 7
r- B.B
(J
Figure 3. The restitution coefficient for bars 1 to 8 (left to right and then top to
bottom).
Hurmuzlu, 1996. However, our results are much worse for the first three
bars. It should be emphasized that our analysis ignores friction, and
this may account for the discrepancies.
Another experimental comparison is the time duration between the
first and the second impact; in the case of bar 3, we have performed ex-
tensive numerical experiments, reported in Paoli and Schatzman, 1999;
the observed time which can be read on Fig. 5 of that paper is 0.410- 3 S
which is completely compatible with the theoretical value found and
reported in table 1.
63
5. The possible role of inharmonicity

The important fact (18) holds only if the compressive modes of the bar
are harmonic, i.e. their frequencies are integer multiples of the funda-
mental; then for (hr /2, relation (28) shows a restitution coefficient equal
to 1.
However, the coefficient of restitution 1 is never observed, not even in
our numerical experiments.
Therefore, we wondered whether inharmonicity, whether of geomet-
rical, material or numerical origin, could account for the observations.
We considered therefore a slightly heterogeneous wave equation
p(x, (1)Utt - (E(x, (1)u x )x = 0,

with initial conditions
U(x, -L/2) = 0, ut(x,L/2) =-1 (31)

and boundary conditions
x(-L/2,t) =0, u x(L/2,t) =0. (32)

We assumed that
p(x, (1) = Po + (1pt{x/(1), E(x,{1) = Eo + (1El (x/{1), (33)

with Pl and El periodic with respect to their argument. The litera-
ture on the homogenization of the wave equation, for instance Brahim-
Otsmane et al., 1992, Francfort and Murat, 1992 or Gerard, 1991, is not
very helpful here, since it considers strong oscillations and shows often
that the derivatives do not converge strongly.
Nevertheless, we were able to show that the first time r({1) after which
ux(t, -L/2) becomes positive again tends to 2L/co as {1 tends to o. After
going back to physical coordinates, which introduces a discrepancy in the
geometry, since the characteristics for the {1-problem lie at a distance at
most (1 from the characteristics for the limit problem, we claim that the
difference between Ut(x, r({1)) and its limit for (1 = 0 is at most O({1)
except on a set of total measure O({1) indicating thus that the global
kinetic energy at time r({1) is O({1)-c1ose to the initial kinetic energy.
The same analysis could be performed in the case of strong and fast
oscillations, i.e. c( x, (1) = Cl (x / (1) and it would probably lead to a net
loss of macroscopic kinetic energy in the bar after impact. This is a
feature which might be worthy to implement in impact dampers. As
it stands, we believe that the macroscopic loss of energy observed for
(J = 1r/2 is really the loss due to plastic deformation.
64
References
Brahim-Otsmane, S., Francfort, G. A., and Murat, F. (1992). Correctors for the ho-
mogenization of the wave and heat equations. J. Math. Pures Appl. (9),71(3}:197-
23l.
Francfort, G. A. and Murat, F. (1992). Oscillations and energy densities in the wave
equation. Comm. Partial Differential Equations, 17(11-12}:1785-1865.
Gerard, P. (1991). Mesures semi-classiques et ondes de Bloch. In Seminaire sur les
Equations aux Derivees Partie lies, 1990-1991, pages Exp. No. XVI, 19. Ecole Poly-
tech., Palaiseau.
Hurmuzlu, Y. {1998}. An energy based coefficient of restitution for planar impacts of
slender bars with massive external surfaces. ASME J. Appl. Meek., 65:952-962.
Paoli, L. and Schatzman, M. (1999). Dynamics of an impacting bar. In Wunderlich,
W., editor, European Congress on Computational Mechanics, Munich. CD-ROM.
Stoianovici, D. and Hurmuzlu, Y. {1996}. A critical study of the applicability of rigid
body collision theory. ASME J. Appl. Meek., 63:307-316.
SOLVING ROCKING BLOCK PROBLEMS
WITH MULTIPLE IMPACTS
Cuneyt Yilmaz
Mechanical Engineering Department, Southern Methodist University
Dallas, TX-75275-0337 USA
Yildirim Hurmuzlu
Mechanical Engineering Department, Southern Methodist University
Dallas, TX-75275-0337 USA
Abstract In this paper, we present a solution to the multiple impact problem

that may arise in the rocking blocks. We use an approach based on
the impulse-momentum methods, the energetic coefficient of restitution,
and the impulse transmission ratio.
1. Introduction
Multiple impact problems still have many difficulties and unanswered
questions (see Marghitu and Hurmuzlu 1995, and Brogliato, 1996). The
rocking block problem, where multiple impacts are likely to occur, is one
of the simplest rigid body impact problems that may involve multiple
collisions.
Housner (1956) introduced the first study to derive the mathematical
equations of a free-standing rigid block under base excitation. Aslam
et al (1980) investigated the earthquake rocking response of rigid bod-
ies, and defined angular coefficient of restitution by using experimental
data. Lipscombe and Pellegrino (1993) presented a solution for free rock-
ing prismatic blocks, and discussed the angular coefficient of restitution
for it. Moreau (1994) used a complementarity dynamical formulation
to solve the multi-body impact problems. This research was followed
by Cholet (1998) and Johansson (1999). Johansson (1999) investigated
the rigid body impact dynamics by using a linear complementarity al-
gorithm considering Coulomb's friction effect. Cholet (1998) used an
adaptation of Moreau's sweeping process (see also Fremond 1995). This
65
J.A.C. Martins and M.D.P. Monteiro Marques (eds.). Contact Mechanics, 65-73.
66
work produces unique and energetically consistent solutions. The only
drawback of the approach is that the problems are formulated in terms
of parameters that do not have obvious physical meanings.
Ceanga and Hurmuzlu (2000) introduced a method that produces
unique and energetically consistent solutions in multiple impact prob-
lems. They developed a new approach that used the energetic coefficient
of restitution, and proposed a new constant, that they called the "Im-
pulse Transmission Ratio". This parameter is physically meaningful and
more effective in dealing with multiple impact problems. They applied
the method to the multiple impacts that took place in an linear N-ball
chain.
2. Problem Description
In this article we consider the system given in Fig. (1). The impact
problem takes place as a result of the block striking the left surface (at
Od while resting on the right surface (at O 2 ). The block is symmetrical
with a width of 2w, a height of 2h, mass m, and a centroidal moment
of inertia of fern. To simplify the calculations, we choose a coordinate
system whose origin is attached to the center of the block and its x-axis
parallel to the horizontal edge of the block at the impact instance. At
the instant immediately before impact, the block is undergoing a non-
centroidal rotation about 02 with an angular velocity of w-. In addition,
we consider frictionless contacts only and choose surface inclinations of
fh and (}2 (see Fig. (1)) at the the respective contact points 0 1 and O2 •
In this paper we apply the Impulse Transmission Ratio (ITR) concept
that was developed in Ceanga and Hurmuzlu (2000) to resolve the diffi-
culty that is encountered in the present problem. In the next section we
use compliance contacts to derive the ITR for the rocking block problem.
3. Impulse Transmission Ratio

We use the compliant model that is shown in Fig. 2 to derive the ITR
for the block problem. Our objective here is to find the relationship
between the normal impulses 71 and 72 at the contact points. In addition,
for simplicity, we use m = h = w = 1 and assume that the block strikes
the surface with a velocity of 1 m/s.
We may write the equations of motion from the block shown in Fig. 2.
Solving these equations yields the following expressions for the x and y
components of the displacements at contact points:
67
Y
m,Icm
h
x
r1 r2
h
Vern
W
.1. w
Figure 1. Rigid Block Diagram while the Impact Occurs
Figure 2. Compliant Model of Rigid Block
2(,81 -a1P1) sin (y'iiit) 2(,81-a1P2) sin (ffit)

3/2 - 3/2
x _ qX _
q1 ____P-"-1 _ _-;-_ _--:-_..;",P=-2_ __
- 2- (1)
f(P2 - pt}
68
(2)
where P1, (12, ).1, ).2, a1, a2, a3, (31, (32, (33, and f are functions of m, h,
w, I cm , (h, (h, k, and 'Y. The impulses acting at the two contact points
can be computed as follows:
~71 = -k lot (sin(h qr + cos(h qf) dt (3)
(4)
Using'Y» 1, we obtain ~72 + a~71 such that:

+ w 2 ) cos((h - (h)
(h 2
a =
21;:- + (h 2 + w 2 ) + (w 2 - h 2 ) cos 2(h - 2hw sin 262
([ -21;:- + (w 2 - h 2 )] cos(61 + ( 2 ) - 2hw sin(61 + (2 )
+ . (5)
2 ~ + (h 2 + w 2 ) + (w 2 - h 2 ) cos 262 - 2hw sm262
On the other hand, when'Y« 1 we obtain ~72 = o. Accordingly, we
can specify the bounds for ITR as follows:
(6)
4. Velocity-Impulse Relationships
Using the laws of conservation for the impulse and momentmn, we
obtain the following equations:
(7)
(8)
where, m is the mass, Icm is the centroidal moment of inertia of the
block, l'cm and w are the linear and angular velocities of the mass center
of the block respectively, 71 and 72 are the impulses, and r1 and r2 are
the vectors from the mass center to the contact points. The velocities
at the contact points can be written as follows:
(9)
(10)
69
where V1 and V2 are the velocities of the block at 0 1 and 02 respectively
(see Fig. (1)).
We use Eqs. (7), (8), (9), and (10) to derive the velocity expressions
in terms of the collision impulses.
5. Bouncing patterns
We consider the case where the block strikes the external surface at
0 1 while resting at 02 (vr- i= 0 and v~- = 0, or more specifically
W- = Wo i= 0). There are two possible bouncing patterns that result
from the collision at 01, they can be enumerated as follows:
5.1. Single impact

This case arises when the non-impacting end bounces at the onset
of the collision (i.e. its normal velocity becomes positive immediately).
This means that the slope of the normal velocity v~ is positive at the
onset of impact. Thus, we may write the condition for the occurrence of
this case as follows:
-dv~ = --sm(h
dv~. dv~
+ -COS(}2 =
dT1 dT1 dT1
(h 2 + w 2) COS((}l - (}2) + [-2 ~ + (w 2 - h 2)] COS((}l + (}2)
2~ + (h 2 + w 2) + (w 2 -
h 2) cos 2{}2 - 2hw sin2(}2
2hw sin((}l + (}2)
> 0 (11)
2 ~ + (h + w ) + (w 2 - h2) cos 2{}2 - 2hwsin2(}2
2 2
The impulse at the end of the collision at 0 1, T{, can be found by

using the energetic definition of the coefficient of restitution to obtain:
ei r10 f v? dT1 + iT!

Tf
v? dn = 0
where e1 is the coefficient of restitution at 0 1 •

By solving this equation, we obtain,
f _ 2(1 + et}r1 WO cos (}1
T1 - r2 cos 2 (}1 + sin (}1 (-2r3 cos (}1 + r 4 sin()t}
The post-impact velocities can be found by substituting the final im-
pulse into the respective velocity expressions.
5.2. Simultaneous Collision At Both Ends

If the condition in Eq. (10) is violated, simultaneous impacts at 01
and O2 take place. We use the energetic definition of the coefficient of
70
restitution to compute the final impulse for the first collision at 01 as:
f 2(1 + edr1WOCos81
71 =
rg + r10
and the final impulse for the first O2 collision as:
72f = r1 r 1S
r17(r 2 cos 82 + r4 sin2 82 - r3 sin(282))
2
[V
r 14+ e2 -r16a+r 214]
where
r7 :1 [r3(asin82 - sin8d + r2 cos 81 + rsacos82]
ra :1 [r3(acos82 - cos8d + r4 (sin 81 - a sin 82)]
rg - r2 cos2 81 + sin81 [r3a cos 82 + r 4(sin 81 - a sin 82)]

r10 cos 81 [rsa cos 82 + r3( -2 sin 81 + a sin 82)]
r 14 = (-r4-rS) cos(81+82) +m[(h2 +w 2) cos (81-82)-2hwsin(81 +82)]
r 1S = (1 + edwo cos 81
r 16 = [(rs - r 4) cos(81 - 82) + (r 4 + rs) cos(81 + 82)

+2r3 sin(81 + 82)][-r2 - r4 - (r2 - r4) cos(282) + 2r3 sin(282)]
r17 = r 2 cos2 81 + sin8I[ar3 cos 82 + r 4(sin81 - a sin82)]

+ cos 8dars cos 82 + r3( -2 sin 81 + a sin 82)]
Now, once again, we check the normal velocity at Ot. If it is positive,
there would be no more impacts. Otherwise, additional impacts may
71
emerge. In this case, the computations will continue in a similar manner.

We just switch the notations for the two ends of the block, and follow
the procedure that we presented above. The process continues until all
normal velocities become positive.
Finally, we may obtain the upper limit of the impulse transmission
ratio by considering the inside of the square root in the impulse equation
as follows:
(12)
Simplifying the expression given in Eq. (12) yields the identical upper
bound for the compliant case, which was presented in Eq. (5).
6. The limits of ITR

One of the important questions that we may ask in the present study is
why the two conditions that are given in Eqs. (5) and (10) are identical.
This means that the upper limit of the ITR a is positive if and only if
the condition in Eq. (10) is violated. On the other hand, when we have
Eq. (10) satisfied, we have a negative upper limit for Q. This means
that there is no valid ITR for these types of collisions. This is expected,
because when the block undergoes single impact, the ITR is meaningless
(there is no impulse at O2 ), and hence no valid range for it can be found.
Now, we analyze the bouncing pattern for a rectangular block with
a height to width ration of r = wjh. Substituting r into Eq. (10) and
simplifying yields:
3(1 + r2) cos((h + ( 2) - (5 - r2) cos(81 - ( 2) - 2rsin{81 + ( 2) > 0

5 + 5r2 + 3(r2 - 1) cos 282 - 6r cos 282
(13)
Using this inequality, we obtain the regions of single and multiple

impacts depicted in Fig. (3). Setting the numerator of Eq. (13) equal
to zero we can draw the boundary curves and obtain the limiting angle
values shown in the figure. A choice of an angle pair in the unshaded
regions will result in a single impact. Whereas, a choice of an angle pair
in the shaded regions will result in multiple impacts. The partitioning
of the 81-82 plane according to the bounce patterns at the contact points
depends on the specific r values. When r < ...ti, we have a disjoint mul-
tiple impact region and connected single impact region (see Fig. (3.a)).
On the other hand, when r > ...ti the situation is reversed as shown
in Fig. (3.b). In addition, as expected, we can observe that for wide
72
I.SO
1.25
1.00
'0'
g 0.7S
",- Single Impact
D 01 - Tan-'[(2r'.l)l(3rl] aJ-
9 1 -Tan" [(3r)l(r'+4l]
0.25
r< .Jf
0.00 .....,...",...-:-:---:-:--'-:-:-:......,...,,.,......:.;~~
l .s<l 1.25 1.00 0.7S 050
92 (Rad)
(a)
Figure 9. Bouncing Patterns
blocks (large r values) multiple impacts are more likely occur, while for
a narrow block (r is small) we are more likely to have a single impact.
1. Conc1 us ion
In this paper we developed a new approach to solve the multiple im-
pact problem of a rocking block. The methodology is based on the use of
impulse momentum methods. The approach uses the Impulse Transmis-
sion Ratio that was developed previously to solve the multiple impact
problems in a linear chain of balls (Ceanga and Hurmuzlu, 20(0). The
method also utilizes the energetic coefficient of restitution and yields
energetically consistent solutions (although this issue is not valid in the
present paper since friction is neglected).
The problem considered here is simplified by neglecting friction at the
contact points. This was a necessary simplification in this initial stage
of the development. Yet, including friction will be a good natural step
for future research efforts.
References
Aslam, M., Godden, W.G 1956, Scalise, D. T., 1980, "Earthquake Rocking Response
of Rigid Bodies" , Journal of Structural Division, Feb. 1980, pp. 377-392.
Brogliato B., 1996 Nonsmooth Impact Mechanics: Models, Dynamics and Control,
Springer Verlag, LNCIS 220.
Ceanga V., Hurmuzlu Y. 2000, "A New Look At an Old Problem: Newton's Cradle" ,
to appear in the ASME Journal of Applied Mechanics.
Cholet C. 1998 Chocs de Solides Rigides, Ph. D. Thesis, University of Paris 6 and
LCPC-CNRS, March, 1998)
Fremond M. 1995, "Rigid bodies collisions", Physics Letters A, 204, pp 33-41.
73
Housner, G. W., 1956, "Limit Design of Structures to resist Earthquakes", Proc. of

the World Conference on Earthquakes Engineering, pp. 51 - 5t1.
Johansson, L., 1999, "A Linear Complementary Algorithm for Rigid Body Impact
with Friction" , European Journal of Mechanics, , vol. 18, pp. 703-717.
Lipscombe, P. R., Pellegrino, S., 1993, "Free Rocking of Prismatic Blocks", Journal
of Engineering Mechanics, Vol. 119, No.7, pp. 1387-1410.
Marghitu D. B., Hurmuzlu Y. 1995, "Three Dimensional Rigid Body Collisions with
Multiple Contact Points", ASME Journal of Applied Mechanics, 62:725-732.
J.J. Moreau, 1994 "Some numerical methods in multibody dynamics: application to
granular materials", European J. of Mechanics A/Solids, vol. 13, no 4, pp. 93-114.
Stronge, W. J., 1990, "Rigid Body Collisions with Friction", In Proceedings of Royal
Society, London, A 431, pp. 169-181.
A CONTROLLABILITY CRITERION FOR
LINEAR JUGGLING MECHANICAL
SYSTEMS
Bernard Brogliato
INRIA RhOne-Alpes, ZIRST Montbonnot, 655 avenue de l'Europe
38334 Saint lsmier cedex, France
Mongi Mabrouk
Laboratoire de Mecanique Appliquee Rene Chaleat
24 chemin de l'Epitaphe, 25030 Besan(,!on cedex, France
Arturo Zavala Rio

Universidad Autonoma de San Luis Potosi, Faculdad de lngenieria
A'll. Dr Manuel Nava 8, San Luis Potosi, SLP 78290, Mexico
A bstract This paper deals with the controllability of a class of nonsmooth

complementarity mechanical systems. Due to their particular
structure they can be decomposed into an "object" and a "robot",
consequently they are named juggling systems. It is shown that
the controllability properties of the "object" can be characterized
by nonlinear constrained equations, or generalized equations. Ex-
amples are presented, including a simple model of backlash, whose
accessibility is shown. The main focus of the work is about linear
jugglers.
1 Introd uction
Manipulating objects by pushing and hitting (also called nonprehen-
sile manipulation) is an important robotic task, see Akella et al (2000)
and references therein. It is easily recast in the setting of so-called jug-
gling systems (a class that encompasses systems with dynamic backlash,
75
I.A.C. Martins and M.D.P. Monteiro Marques (eels.), Contact Mechanics, 75-82.
76
manipulators with dynamic passive environments, controlled structures,

hopping machines etc). Studies on the controllability of such nonlin-
ear nonsmooth dynamical systems require the development of specific
analysis tools, due to their very particular features Brogliato and Zavala
(2000). This paper is dedicated to investigate a way to characterize
the controllability properties of a subclass of juggling systems, which we
choose to name linear jugglers. It appears that despite the fact that this
class of jugglers may represent the simplest juggling systems, their con-
trollability is not easy to establish in general since anyway they remain
highly nonlinear dynamical systems.
2 System's dynamics
Let us consider the following class of complementarity dynamical sys-
tems that we may call linear jugglers
Mlih = AT>.
(1)
O~Aq+B..l>'~O
The "proxM" denotes the proximation in the kinetic metric and the
times tk generically denote impact times. The tangent cone to the do-
main Aq + B ~ 0 is defined as V(q(t» = {x E lR 2 X lR 2 I Ax =
~ ~
AIXI + A2X2 ~ O}. In (1) qi E lRnl/2xl, q2 E lRn2/2xI, qT = (q[, qf),

A = (A I ,A2) E lRmx (nI/2+n2 /2), E E lRn2/2xnu, MI E lR¥-x¥-, M2 E
lR 2 X 2 , and B E lRm are all constant, >. E lRm , and we may choose
~ ~
Moreau's collision rule with restitution e Mabrouk (1998).
3 A controllability criterion
In view of this let us investigate another path to characterize the con-
trollability properties of linear juggling systems as in (l).We make the
following
Assumption 1. Let Al be made of the first ~ columns of A. The
matrix Al satisfies rank(At) ~ 2.
77
The control problem is decoupled into two parts: the control of the
zl-dynamics using A as the input (i.e. using impacts), then the control
of the overall system with u. Here we focus on the first part only.
Let us formulate the following controllability problem which we name
Controllability Through the Impacts (CTI)
Problem 1 (CTI). Given (q},vt) and (qf,v~) with vi =F 0, find
{ql(k)h$k$n, {lh(k)}2$k$n, {lJ2(k)h$k$n (2)
such that (qt{tr), ql(ti» = (q}, vt) and (ql(t n), ql(tt» = (qf, v~).
Notice that n ~ 2 by construction. Controllability is here to be
understood as a general assertion and obviously can be replaced by ac-
cessibility. The constraint on vi will be made clear later. The reachable
subspaces are defined as 'R,n [(qt,vt)] = {(qf,v~) I problem 1 possesses
at least one solution}, and denote tin [(qt,vi)] = U3$k$n'R,k [(q},vi)].
Following the standard terminology Nijmeijer and van der Schaft (1990),
the dynamics of the object may be called accessible when tin [(q}, vt)]
contains an open set for any n ~ 3, and accessible in N impacts if
tin [(q},vt)] contains an open set for all n ~ N. Let us denote xf =
xr
(qf(2), ... ,qf(n-1», = (qf(l), ... ,qf(n», xf = (qf(2), ... ,qf(n-
1», xT = (qf(2), ... ,qf(n».
The main goal of this note is to prove the following
Lemma 1. Let assumption 1 hold. Problem 1 has a solution if and
only if the constrained equation H(Xl' X2, X4) = 0, G(XI' X2, X4) ~ 0 has
a solution, where H(·) and G(·) are some nonlinear functions. Further-
more H(xI, X2, X4) = HX4 (Xl, X2) and G(XI, X2, X4) = GX4 (Xl, X2), where
H X4 (·) and GX4 (-) are linear.
The functions H (.), G(·), HX4 (.), GX4 (.) can be constructed as follows
AIj(vi,A,MI,M2,X4)XI + BIj(vi,A,q},qf,MI ,M2,X4) = 0
J. E {1 , ... ,m(n-l)}
(3)
78
that is a set of nl + n equalities and 2(n - 1) inequalities for each j .•

A first step may be to study the linear equations under linear in-
equality constraints in Xl parameterized in X4 (Le. the first and third
lines in (3)), then to incorporate the constraints on X4. In view ofthis,
let us reformulate problem 1 as follows
Problem 2. Let the initial and final states (qi,vi) and (q~,vt) with
vi =1= 0 be given. Does there exist n ~ 2 such that there exists j E
{1, ... , m(n-l)} such that rank(Azj) = nI/2? If yes, then choose Xl =
Xim(X4)+xt such that AZj (X4)xim(X4) = -BZj (X4), where xt lies in the
kernel of AZj (X4). Then (3) is transformed into
:FZj (e, A, MI, M2)X4 + (hj (vi, e, M I , M2, A, vt) = 0
llZj (vi, x4)[xim(X4) + xt] + .7Zj(ql, q~, vi, X4) ~0 (4)
K:zj (vi,A,e,Ml,M2) + Lzj(e,MI,M2,A)X4 ~ 0

which is a linear equation in X4 with a linear inequality and a nonlinear
one. The second step is: does there exist n ~ 2 such that there exists j E
{1, ... ,m(n-I)} such that rank(:Fzj ) = T?
If yes choose X4 = x~m +xi
such that :FZj x~m = -gZj' where xi lies in the kernel of :Fzj . Then
choose xi such that llZj(x~m + xi)[xim(X4) + xt] + .7Zj(x~m + xi) ~ 0
and K:Xj + LXj [x~m + xi] ~ o. The last step consists of solving the
equation for X2 (it suffices that rank(Vxj) = n)
CZj (A).[xim(x~m + xi) + xt-] + VZj (A)X2 + eZj (B) = O. (5)
Under the stated conditions of problem 2, the whole space lR,n 1 3

(q~, v;t) is accessible from any (qt, vi): consequently controllability over
Fl is true. It is clear that the first two steps can be reversed.
Remark 1. Using the Karush-Kuhn-Tucker optimality conditions, it is
then not difficult to derive a necessary condition for problem 1 to possess
a solution, under the form of a generalized equation
(6)
where !(x) = 1/J(JR+)2n-dLx+K) is a convex function, and 1/J(JR+)2n-d·)

is the indicator function of the set (lR,+)2n-l.
79
Notice that the proposed criterion (problem 1) is not local in nature

as in the classical smooth nonlinear case, as the definition of the sets
'R,n [(qL vt)] shows.
4 Example: dynamic backlash model

As an illustration let us consider the impacting pair in figure 1 that
models dynamic backlash. Let us consider the case n = 4, and let us
choose an arbitrary sequence Ij = {I, 2,1}, i.e. i(2) = 1, i(3) = 2,
z·(4) -- .1 Thus A i1(2) -- A i1(4) -- 1, A i1(3) -- -1 , ~
,di(2) - ,di(4) -- - 1,
- ~
A2 i(3) - 1 Bi(2) - Bi(4) - 0 Bi(3) - L N i (2) - N i (4) - 11m1 N i (3)_
-' - -, -'1-1- '1-
-11m2. In other words we choose three impacts and to collide with each
constraint alternatively. Then the following simplified form of (3) can
be computed, where we drop all indices for convenience
AX1 + 8X2 + C = 0
(7)
1l(X4).X1 + I(x4) ~0
which is a set of 6 equalities and 6 inequalities. Let us choose the second

option for (7). From the first equation in (7) it follows that
X2 = _(8T 8)-1 [8T Ax1 + 8 T C] (8)
The second and third lines of (7) are a linear equation with linear in-
equality in X4, and the fourth line is a linear equation in Xl parameterized
by X4. The following is true
Lemma 2. The dynamics of the object part of the impacting pair (Le.
the dynamics of the mass md is accessible in 4 impacts.
Proof: The backlash model does not satisfy assumption 1, so that
the constrained equation in (3) is slightly modified. One has in (7)
-1 -1 0 1 -1 1 L-qi
0 0 1 0 0 0 -q}+L
A= 1 0 8= 0 -1 0 0 C= 0
0 -1 0 0 1 0 L
0 0 0 0 0 -1 qi
(9)
80
fo--------.
. . ._ . . . ._ _ _. . ._ ..... M2
.. ,
,,
'
.. ,
, L+L =L
1 2
Figure 1: The impacting pair
v=( o) (10)
(He)m2(eml-m2) (He)ml
(ml+ma)2 - ml+m2
1 -vi
m1+ema v +
:F= (He)ma
ml+ma -1 0 g= ml+m2 1
+ (He)am1m2 +
1 0 0 eV1 - (ml +ma)2 v1
(11)
-1
0
-ev+ +
I
(1-te)2ml m2 v+ +K:x4
(ml +m2)2 I
-qf1l12
1l=
-1 1
I= 0
~+
ml+m2 VI + c. X4 ml+m2 VI + C. X4
~+
1
vt
0 ::#
VI
(12)
(13)
To prove the accessibility, it is sufficient to show that the constrained

equation in (7) possesses a solution for a set of data qf and that vt
81
contains an open set. To this end let us first examine the third and
second lines of (7), i.e. the X4 subproblem. From the third line and (11)
one deduces
(14)
where X4,i is the ith component of X4. The inequalities in (14) define
a domain Dl C IR3 for X4 E F that clearly has a nonempty interior:
int{Dl) -=J. 0. Actually Dl is the intersection of three half-spaces whose
boundaries are three planes. From the second line in (7) and from (10)
it follows that X4 has to belong to a plane PI that is orthogonal to
the plane (O, x, y) where 0 is the origin and x, yare the first two
coordinates. The intersection of PI with (O,x,y) is a line L 1, which in
turn intersects (O, y) at the point {O, -£ (fi~)~~' 0). Clearly the signum
of £ can be varied arbitrarily by modifying vt, which corresponds to
translating PI along its normal 'D in (10). Since int{Dd -=J. 0, it follows
that there exists a set 81 for vt such that PI n Dl =1= 0 for all vt E 81,
and int{8d -=J. 0.
Let us now examine the denominators of 11. and I. The values of X4
which yield singularities belong to two planes P2 and P3, which are both
orthogonal to (0, X, y) in the 3-dimensional X4 space. Therefore even in
the worst case where P2 n PI n DI -=J. 0 and P 3 n H n DI =1= 0, the set
of admissible X4 remains of nonempty interior and so does 8 1 3 vt. It
remains now to examine the last inequality in (7), which provides the
existence of Xl E IR2. The first and third inequalities yield XI,2 ~ qt
or XI,2 ~ qt and XI,1 ~ qi or XI,I ~qi. The second inequality defines
a half plane whose boundary is a line containing the origin O. There
exists a set 8 2 with nonempty interior such that for every qt E 82 the
intersection of these three half planes has a nonempty interior. Finally X2
can be calculated from (8). Since 81 and 82 have non-empty interiors,
the reachable set 'R,4[(qi, vi)] contains an open set. Since the above
reasoning can be redone with new initial data chosen as the quantities
at time t2, one concludes that 'kn [(qi,vi)] contains an open set for all
n ~ 4.
82
References
S. Akella, W.H. Huang, K.M. Lynch, M.T. Mason (2000) Parts feeding
on a conveyor with a one joint robot, Algorithmica, vo1.26, pp.313-344.
B. Brogliato (1999) Nonsmooth Mechanics, Second Edition, Springer
London CCES.
B. Brogliato and A. Zavala-Rio (2000) On the control of complementary
slackness juggling mechanical systems, IEEE Transactions on Auto-
matic Control, vo1.45, no 2, pp.235-246, February.
M. Mabrouk (1998) A unified variational model for the dynamics of per-
fect unilateral constraints, European Journal of Mechanics A/Solids,
vo1.l7, no 5, pp.819-842.
H. Nijmeijer and A. J. van der Schaft (1990) Nonlinear Dynamical Con-
trol Systems, Springer Verlag.
ON INTEGRATING
STIFF MULTIBODY DYNAMICS
WITH CONTACT AND FRICTION
Mihai Anitescu
Department of Mathematics University of Pittsburyh,
Pittsburyh, PA 15213, U.S.A.
Florian A. Potra
Department of Mathematics, University of Maryland, Baltimore Oounty,
Baltimore, MD 21250, U.S.A.
Abstract We present a complementarity-based time-stepping method for rigid

multibody systems with contact and friction that accomodates stiffness
through a linear implicit technique. The method is consistent whenever
the stiff forces originate in springs and dampers. When the stiffness
parameters increase towards infinity the subproblem to be solved in
one step approaches the one where the stiff elements are replaced by
joints, under the assumption that the friction cone of the limit system
is pointed.
1. Introduction
In Anitescu and Potra (2001) we proposed a time-stepping method
for stiff multibody dynamics, subject to joint, contact and frictional
constraints. The method is set up in a velocity-impulse framework, and
is shown to be consistent for a wider class of configurations than the
ones covered by classical mechanics, similar to previous such algorithms
cf. Anitescu and Potra (1997), Anitescu and al. (1999), Stewart and
Trinkle (1995) and Stewart (2000).
1.1 Constraints and requirements

Our approach covers several types of constraints. In the following q
and v will constitute the position and, respectively, velocity vector of a
system of several bodies.
83
I.A.C. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 83-90.
84
Joint Constraints. Such constraints are described by equations
e(i)(q) =0, i=1,2, ... ,m. (1)
Here, e(i)(q) are sufficiently smooth fWlctions. We denote by y(i)(q) the
gradient of the corresponding function, or
y(i)(q) = vqe(i) (q), i = 1,2, ... ,m.
The impulse exerted by a joint on the system is c~)y(i)(q) where c~) is a
scalar related to the Lagrange muliplier of classical constrained dynam-
ics, see Haug (1989).
Noninterpenetration Constraints. These constraints are defined in
terms of a continuous signed distance function between the two bodies
q,(q), cf. Anitescu and al., (1996). The noninterpenetration constraints
become
q,(j)(q) ~ 0, j = 1,2, ... ,po (2)
The fWlction q,(q) is generally not differentiable, especially when the
bodies have flat surfaces. This situation is generally remediable by con-
sidering different geometric primitives, cf. Cremer and Vanecek (1995),
which result in noninterpenetration constraints being expressed in terms
of several inequalities involving differentiable functions q,(q). We denote
by
(3)
which we call the normals. In sequel, q,(j) will be called contact (j),
though the contact is truly active only when q,(j)(q) = O. When the
contact is active, it can exert a compressive normal impulse, cW)n(i)(q)
on the system, which is quantified by requiring cW) ~ o. The fact that
the contact must be active before a nonzero compression impulse can
act is expressed by the complementarity constraint
Frictional Constraints are expressed via a discretization of the fric-

tion cone, cf. Stewart and Trinkle (1995) and Anitescu and Potra
(1997, 2001). For a contact j, we take a collection of coplanar vec-
tors di (q), i = 1, 2, ... , me which span the plane tangent at the contact
(though the plane may cease to be tangent to the the contact normal
when mapped in generalized coordinates, see Anitescu and al. (1996)).
The cover of the vectors di(q) should approximate the transversal shape
of the friction cone. In two-dimensional mechanics, the tangent plane
is one dimensional, its transversal shape is a segment, and only two
85
such vectors d1 (q) and d2 (q) are needed in this formulation. We de-
note by D(q) = [d 1(q),d2 (q), ... ,dmc(q)]. A tangential impulse will be
:E~1 (1ic4(q), where {1i ~ 0, i = 1,2, ... ,me. The friction model, which
ensures maximum dissipation for given normal impulse en and velocity
v, as well as that the total contact force is inside the discretized cone, is
D(q)Tv + .Ae ~ 0 ..L {3 ~ 0, Jl.Cn - eT{3 ~ 0 ..L .A ~ o.
Here e is a vector of ones of dimension me, e = (1,1, ... , l)T, Jl. is
the friction parameter and {1 is the vector of tangential forces {3 =
({31, f32, ... , (3mc ). The additional variable .A is approximately equal to
the norm of the tangential velocity at the contact, if there is relative
motion at the contact, or IID(q)TvII =I- 0 cf. Stewart and Trinkle (1995)
and Anitescu and Potra (1997).
Dynamical data of the simulation. We denote by M(q) the sym-
metric, positive definite, mass matrix of the system and by k(t, q, v) the
external force.
2. A scheme for stiff multi body dynamics with

contact and friction
To include these results in a time-stepping scheme, we formulate all
geometrical constraints at the velocity level. For joint constraints, we
thus require V q9(i)T (q)v = v(i)T (q)v = O. For noninterpenetration con-
straints, we require that, whenever such a constraint is active, we must
have ft<p(j)(q) = n(j)(q)T v ~ o. The set of indices j of active con-
straints is denoted by A. The active set is updated at every integration
step based on normal velocities at contact at the end of one step and on
level of interpenetration. If a new contact appears, a collision resolution
needs to be applied, see Anitescu and Potra (1997). We do not discuss
in detail here strategies for updating the active set, we will assume that
it is available at every step of the integration procedure.
Let h be the time step. If, at some time to), the system is at position
q(l) and velocity v(l), then we choose the new position to be q(l+l) =
q(l) + hv(I+1), where v(l+1) is the solution of the following mixed linear
complementarity problem that collects all the constraints introduced
above
---
M(l) -v -n -D 0 v(l+1) - M v(l) - hiJi) 0
ifI' 0 0 0 0 Cv 0 0
ii.T 0 0 0 0 Cn + 0 P
jjT 0 0 0 E 13 0 (J
0 0 t-t -ET 0 .A 0 (
(5)
[l n~ 1
86
= 0,
c- =
Here 1/- -- [1/(1) , 1/(2) , ... , I/(m)] ,II
[ II ~ 0, [ ~ 1~ O. (6)
11,11,···,11,
[C(1) C(2) C(m)]T n- = [n(i!) "n(jI)
••• , n (js)] , r;~~

'-"ft
=
n , c(h)
[c(jI) "ys)]T ,~R = [R(i!)T
n ,···,v-It ~ ,~
R(h)T , ... ,~
R(is)T] ,
15 = [D(jt) , D(h), ... ,D(js)], >: =
[A (j I) , A(j2), ... ,A(js)],
ji = diag(t-t(jt) , t-t(h), ... ,t-t(js)f, and E = diag(e(jt) , e(h), ... ,e(js)) are
the lumped LCP data, and A = {ii, h, ... ,js} collects the active contact
constraints. The vector inequalities in (6) are to be understood comp~
nentwise. To simplify the presentation we have eluded the dependence
of the geometrical ~ameter~n the data of the simulation.
The quantities M(l) and k(l) are the key of the successfull stiffness
resolution. They correspond to a linearly implicit modification of the
classical Euler scheme:
ij{i) [M (q{l)) - h2 "V qk (t(l) , q(l), v(l)) - h"Vvk (t(l), q(l), v(l)) ](7)
iJii k (t(l) , q(l), v{l)) - "Vvk (t(l), q(l), v(l)) v{l). (8)
If Ki(l) is positive definite (but not necessaily symmetric), then (5-6) is

guaranteed to have a solution, that can be found by Lemke's algorithm,
cf. Anitescu and Potra (2001).
3. The case of stiffness due to springs and

dampers
In multi body dynamics, the most widely encountered cases of stiff
forces are generated by springs and dampers attached between two points
of the system. In this case the total external force will have the following
expression:
k(t, q, v) k1 (t, q, v)
(9)
E~~-Y1 ('Ykt/J(k) (q) + 8k ("V qt/J(k) (q)T v) ) "Vqt/J(k) (q)
Here kt{t,q,v) represents the nonstiffforces, t/J(k) (q), k = 1,2 ... ,nc'Y
express the distance functions between two points of the system, even-
tually offset by a parameter to account for the initial energy stored in
the spring. The quantities 'Yk and 8k , k = 1,2, ... , nc'Y represent the
spring and, respectively, the damper parameters.
To apply the scheme (5-6) we use the following approximations
no-y
"Vqk(t,q,v)) ~ - L'Yk"Vqt/J(k)(q)"Vqt/J(k)T(q), (10)
k=1
87
n6-y
V vk (t, q, v)) ~ - L ~k V q¢(k) (q)V q¢(k)T (q). (11)

k=1
Using these approximations when computing ij{i) (7) and k(i) (8), it
was shown, cf. Anitescu and Potra (2001), that
1 The matrix M(l) is positive definite for any value of h and the
scheme (5-6) is thus unconditionally consistent.
2 An energy inequality holds from one step to the next, which ensures
that a system with constant mass and linear mappings defining the
stiff forces is stable.
3 If damping is the dominant stiffness effect and the friction cone
is pointed, then we obtain that, for high stiffness, the solution set
of the time-stepping scheme (5-6) approaches the solution set of
the problem with the springs and dampers replaced by rigid joint
constraints.
4. Further stability results

In this section we extend the results from Anitescu and Potra (2001),
in that we consider the case where we can have either elastic or damping
dominating effects for large values of the stiffness parameters.
In the force (9) we assume that Tk ::> dk for k = 1,2, ... , ny and that
~k ::> Tk for k = n-y+l, n-y+2, ... ,nch. Here n-y is a fixed natural number
n-y ~ n-yo. For large values of the stiffness parameters, the elastic effects
will dominate in the first n-y components of the sum defining (9), whereas
the damping effects will dominate in the last n-yo - n-y components.
It is intuitively clear that for large stiffness the springs and dampers
will behave like rigid links. Therefore, as we will show later in the limit,
stiff forces will be replaced by
V't//k) (q(I)T v(l+1) = - ¢(k)tl »), k = 1,2, ... , n-y (12)
V't/J(k) (q(l)T v(l+1) = 0, k = n-y + 1, ... , n-yo.
We thus construct the following limit linear complementarity LCP:
M '"
-v -n -D 0 v(l+1) - M v(l) - hk~') 0
-:;;T
v 0 0 0 0
'"
ell r 0
en
ll
,
ri.T 0 0 0 0 + 0 15
jjT 0 0 0 E {3 0 u
0 0 Ji. -EfT 0 A 0
(13)
nn~
88
[~ 0, [ ~ 1~ 0, [ ~ 1~ 0. (14)
We use the same notations as in (5-6) in addition to the mass matrix M

replacing with ki = kd t(l) , q(l) , v(l)), the non-stiff force, replacing
Mi,
kW, and ~ replacing V, where
~= [v, V'l/P) (q(l»), V.,p(2) (q(l»), ... , v.,p(n-y.s) (q(l») ] .
and
We now make the assumption that the friction cone of the limit system
(13-14) is pointed. This is a key assumption for proving that, in the
nonstiff case, as h -7 0, the sequence created by this algorithm converges
in a measure differential inclusion sense, cf. Stewart (2000). In our case,
this is equivalent to assuming that any solution of (13-14) for a zero free
term (v(l) = On, rv = Om+n.s-y and k~') = On) must satisfy v(l+l) = 0,
en
~/I = 0, = 0 and ,B = O.
We now focus on the LCP (5-6), where M (7) and k (8) are com-
puted using the approximations (10) and (11). We assume that we have
a sequence of stiffness parameters, -r:,6~, k = 1,2, ... ,ny6, P = 1,2, ...
such that all elements of the sequence increase to infinity, but the dom-
inant stiffness or damping characteristic of the individual component is
6P 7f
preserved, or :t -7 0, k = 1,2, ... , n'Y and ~ -7 0, k = n'Y + 1, n'Y +
'Yk k
2, ... , n'Y6 . Denote by v(i+l),p, ~, ~, ~P, .xp a solution of (5-6) thus ob-
tained at any p. Although in some perfectly behaved cases there may
be an unbounded choice of .xp, we can always choose .xp to be the rel-
ative velocity at the contact. Therefore we can assume without loss of
generality that we can choose .xp such that II.xPII ~ CcIlV(l+l),PII, where
Cc is a constant that depends only on the friction cone, but not on p.
For k = 1, 2, ... , n'Y6, define dt = h( 6: !h;Y:) , ~ = 6t J;Y: and
~,m+k = -h'rk (.,p(k) (q(l») + hV.,p(k) (q(l»)T v(l+l),p)

-h6~V.,p(k)(q(l»)Tv(l+l),p.
With this notation, we obtain, for k = 1,2, ... , n-y6, that

V.,p(k) (q(l»)T (v(l+l),p) + dt~,m+k + ~.,p(k) (q(l») = O.
89
Denote Dr = diag ( 1m, lit, ~, ... ,~-yo ), Vi = diag ( 1m, 1;,~, ... ,~-yo)
and Tv = [Om,t/J(1)(q(l)),t/J(2)(q(l)), ... ,'ljin-yo)(q(l))]. With these nota-
tions and using ~ = [~, ~,m+1' ~,m+2' ... ,~,m+no-y] in the framework
of the limit problem (13-14) we obtain that v(l+1),p, ~, ct:., fjP, .>:p must
satisfy
M -n -D 0 v(l+1),p
-/I - M v(l) - hkl') 0
-;::p
-:::;T
/I DPI 0 0 0 Cv 77-
lTv 0
fiT 0 0 0 0 cP
~n + 0 P
jjT 0 0 0 E fJP 0 U
0 0 J.t -E
~T
0 ),.P 0 (
(15)
l ~;~]Tlli]{= 0, lct:.]
~; ~ 0, lli]
{~ O. (16)
-;::p
Let AP00 = Ilv(l+l),p , cv' cPn' JJ,
RP ),.PII 00· Assume that a subsequence of
~-
the solution is unbounded, or, without loss of generality, that A~ ~ 00.

By dividing with A~ through (15-16), scaling the solution by A~, and
taking p ~ 00, we obtain that (13-14) has a solution for zero free term
where at least one of the entries not belonging to'>: is not zero (recall that
Ccllv(l+l),PII ~ II).pl!' 'tip). This is a contradiction with the pointedness of
the friction cone. Therefore the sequence of solutions must be uniformly
bounded and taking the limit in (15-16) we immediately see that (13-14)
must be satisfied for any accumulation point of the sequence.
This shows that, as the stiffness increases, the problem to be solved
behaves increasingly like one where the stiff elements are replaced with
joints. The equation for elasticity-dominated elements in (12) is different
from the equation of a joint, but it enforces an even faster return to
equilibrium (the feasible position of a corresponding joint constraints)
as it points toward the constrained limit manifold.
These findings, together with the consistency of the scheme are vali-
dated by several simulations available in VRML format at ''http://www.
mcs.anI.gov/ anitescu/SIMULATIONS/simulations.html", where we used
ranges of le2-1e16 for the stiffness parameters. The LCP was indeed
solvable for the entire range of parameters, and the large stiffness pa-
rameters behaved essentially as joint constrained simulations.
90
Acknowledgments
Thanks to Todd Munson and Mike Ferris for providing and support-
ing PATH, see Dirkse and Ferris (1995) and Munson (2000). This work
was supported by the National Science Foundation, through the awards
DMS-9973071{MA) and DMS-9996154 (FAP), as well as the Mathemat-
ical, Information, and Computational Sciences Division subprogram of
the Office of Advanced Scientific Computing, U.S. Department of En-
ergy, under Contract W-31-109-Eng-38 (MA).
References
Anitescu, M., and Potra, F .A., " A time-stepping method for stiff multibody dynamics
with contact and friction", Reports on Computational Mathematics MCS-P884-
0501, Mathematics and Computer Science Division, Argonne National Laboratory,
200l.
Anitescu, M., and Potra, F.A., "Formulating rigid multi-body-dynamics with contact
and friction as solvable linear complementarity problems", Nonlinear Dynamics
14, 231-247, 1997.
Anitescu, M., Stewart, D. and Potra,F. A., "Time-stepping for three-dimensional rigid
body dynamics", Compo Methods Appl. Mech. Eng. 177(3-4), 183-197, 1999.
Anitescu, M., Cremer, J., and Potra, F. A., "Formulating 3D contact dynamics prob-
lems", Mechanics of Structures and Machines 24(4), 405-437, 1996.
Cremer, J., and Vanecek G., "Building simulations for virtual Environments", Pro-
ceedings of the IFIP International Workshop on Virtual Environments, October
1994, Coimbra, Portugal.
Dirkse, S. P., and Ferris, M. C., "The PATH solver: a non-monotone stabilization
scheme for mixed complementarity problems", Optimization Methods and Software
5, 123-156, 1995.
Haug, E. J., Computer Aided Kinematics and Dynamics of Mechanical Systems, Allyn
and Bacon, Boston, 1989.
Munson, T. S., "Algorithms and Environments for Complementarity", Ph.D Thesis,
Department of Computer Science, University of Wisconsin-Madison, 2000.
Stewart, D. E., and Trinkle, J. C., "An implicit time-stepping scheme for rigid-body
dynamics with inelastic collisions and Coulomb friction", International Journal for
Numerical Methods in Engineering 39, 2673-2691, 1996.
Stewart, D., "Rigid-body dynamics with friction and impact", SIAM Review 42 (1),
3-29,2000.
ANALYSIS OF SYSTEMS WITH MULTIPLE
FRICTIONAL CONTACTS
A. P. Ivanov
Moscow State Textile University named after A.N.Kosygin
M.Kaluzhskaya 1, Moscow 119991, Russia
Abstract Different types of paradoxical situations in systems with two or more frictional
constraints are discussed. Along with well-known Painleve paradoxes of non-
uniqueness and non-existence, a new kind of singularity is detected: co-
existence of several stable solutions. Besides, it is shown that in a regular case
where a unique solution exists, this solution might be unstable.
1. Introduction
Given a constrained mechanical system, one should first formulate it on

the level of differential equations and then investigate these equations. Here,
only the first part of this program is discussed, i.e. the determination of
accelerations and constraint forces for given values of co-ordinates and
velocities. Such a problem is known to be well-posed in the case of ideal
(frictionless) constraints where the use of complementarity conditions leads
to unique solution, cf. Cottle et al. (1992). Thus, the accelerations can be
expressed by certain functions of co-ordinates and velocities, leading to a set
of ordinary differential equations.
On the contrary, in the presence of dry friction one cannot be sure to
obtain unique solutions for the accelerations. Such a paradoxical situation
was detected by Painleve (1905) who gave examples of non-existence and
non-uniqueness. Methods of overcoming the paradoxes were discussed by
many authors, including Lecomu (1905), Klein (1909), Prandtl (1910), and
others. It was established that in the case of non-existence, the system
experiences a sudden ''jump'' to the non-paradoxical region. In the case of
non-uniqueness, selection of a "proper" solution is connected with the
91
92
adoption of some additional assumption. For instance, Painleve suggested
that, if the two bodies would separate in the case of smooth surfaces, they
must separate in the presence of dry friction, too. Klein proposed to apply
stability arguments.
In more recent papers by Lotstedt (1981, 1982), Pang and Trinkle (1996),
and Trinkle et al. (1997) some sufficient conditions for the absence of
frictional paradoxes were obtained.
In this paper, the problem in question is considered from a standpoint of
bifurcation theory. First of all, it is shown how to interpret Painleve
paradoxes in terms of fold bifurcations. Next, the case of two frictional
constraints is studied. As proved by Whitney (1955), typical singularities of
differentiable plane-to-plane mappings are either folds or cusps. However, in
the present context the mappings are not differentiable, but piecewise linear.
It is shown that in this case the double-fold bifurcation should be added to
the list of typical bifurcations. Then a mechanical example is discussed.
2. Systems with one frictional constraint
Consider a multibody system with co-ordinates q E R n subject to a single

unilateral constraint ql ~ 0 with ql denoting the distance between two bodies
with rough surfaces. The Lagrange equations have the form
~[OL)_ oL =Q+R(1) (1)

dt oq oq
with Q being the generalized force and R(I) being the constraint force. For
a fixed instant to and given values of q and q, system (1) is to be solved
with respect to q and R(l). We suppose that at t =to the normal velocity at
the point of contact vanish (i.e. ql = 0) while the tangential velocity does
not. In this case, the tangential components of reaction force ~I) , • •• ,R~I) can
be expressed as linear functions of the normal component R?) by means of
Coulomb's law. Substituting these expressions in equation (l) and then
excluding q2'" .qn' we arrive at the single equation for ql and R?)
.. -AR(l)
ql - 1 +
B (2)
where A and B are certain functions of q, q and t. Equation (2) is to be

solved taking account of the following complementarity conditions:
.. > 0, R(l)
ql- I
>
-
0,ql
.. R(I)
I
=0 . (3)
93
Obviously, in the case A> 0 equation (2) has a unique solution which
agrees with (3): if B > 0 (B < 0), then ij} = B, R?) = 0 (ij} = 0, R?) = -B / A).
Paradoxes arise in the case A < 0: if B > 0, both the above solutions are
suitable, while for B < 0 no solution exists. All these possibilities can be
represented in the (B,x) - plane (see Fig.1) with
if ij} > 0
(4)
if Ril) ~ 0
x x
B B
.....
....
A>O
A<O
Figure 1. Equilibria curves
We discuss the stability of these solutions (in a sense). For such purpose,
we apply a quasi-static approach, usually adopted in bifurcation theory. The
variables qj' ifj (j * 1) are "frozen", and A and Bin (2) are fixed. Further,
small violations of the constraint q} ~ 0 are possible, so that q} = -E, E < O,Fl > O. It is easy to show that an equilibrium
with x < 0 is stable ¢:;> A > o. Concerning a solution with x> 0 , it is stable
since x = 0 for B fixed. Thus, in the left diagram in Fig. I, both branches are
stable while in the right diagram the upper branch is stable and the lower one
is unstable.
It is worth noting that in the regular case we have a unique "proper"
solution for small initial perturbations. In paradoxical case, however, in the
absence of stable solution (B < 0) the system experiences a velocity jump
after which the quantity A becomes positive.
94
3. Systems with two frictional constraints
Now suppose that the system under analysis is subject to frictional

constraints ql ~ 0 and q2 ~ 0 .
3.1. General formulation
Equation (1) should include two constraint forces R(I) and R(2) . Excluding
the unknowns Q3" .. ,qn' we get the following equation similar to (2):
.. - A R(I) A 1)(2) B .. - J R(l) A R(2) B (5)

ql - II I + 12L'2 + P q2 - f~1 I + 22 2 + 2'
where ~2) is the normal component of the second constraint force. We

formulate complementarity conditions for the second constraint similar to (3)
and introduce variables XI and x 2 like (4). Then, equation (5) can be treated
as a piecewise linear mapping from the X - plane to the B - plane.
According to Whitney's theorem (1955), in typical case, a differentiable
mapping may have singularities of two types: folds and cusps (see Fig.2). In
the first case (Fig.2,a), different points in the B - plane have two or none pre-
images, while in the second case (Fig.2,b) the number of pre-images varies
from one to three.
c c: a
Figure 2. Fold and cusp
To determine the type of singularity of (non-differentiable!) mapping

associated with system (5), one should look for a solution in each quadrant
on (XI ,X2 ) - plane. The results depend on the coefficients Ay in (5) and are
summarized as follows.
95
l. All > 0, ~ > 0, ~ = All ~ - A12~1 > °- regular case where unique
2.
3.
All > 0, ~ > 0, ~ < 0, AI2 > °-
solution exists for any B» B2 (cf. Cottle et al. (1992»;
cusp;
~I > 0, ~ >0, ~ <0, AI2 <0 - fold;
4. AII~ <0, ~ <0 - fold;
5. AIIA22 <O,~>O,AII~I >O-fold;
6.
7. All < 0, ~ < 0, ~ > °-
~I~ <O,~>O'~I~I <O-cusp;
double fold. To imagine this singularity, one
should bend a sheet of paper in halves and then bend it once more
along a line which is non-orthogonal to the line of the first bending;
8. ~I <0, ~ <O'~<O'~I <O-fold;
9. All <O,~<O,~<O'~I>O-CUSp.
3.2. Discussion of stability
Now we argue the stability of solutions of equation (5) in the same sense
as previously in Section 2. Thus, we "freeze" variables qj' (t (j > 2) and
present reaction forces as some functions of small deformations:
(6)
Generally speaking, there exist three types of solutions: (i) those where
°
Xi and X 2 are positive (i.e. both deformations vanish); (ii) those where
Xix2 < (i.e. one of the deformations is non-zero); (iii) solutions those where
Xi and X 2 are negative (both deformations are non-zero). Solutions of the
first type are stable since for zero reactions and BI and B2 being constant,
we obtain from (5) ~ = x2 = 0. Furthermore, a solution of the second type is
stable ~ the corresponding element of matrix A is positive (i.e. An > 0 for
a solution where R?) =0, RJ2) > 0 etc.).
The third case is most complicated: in general, it is impossible to make
conclusion on stability without specifying the strain-stress functions (6). A
special subcase where matrix A =IIAy II is symmetric can be studied with the
help of Kelvin theorem: the solution is stable ~ A is positive definite. Due
to the presence of dry friction, A is non-symmetric in general. An example
was considered by Popp and Rudolph (2001) who considered a system with
linear functions (6) and detected regions of stability and of instability as
well. Some other conditions of stability and instability which are invariant to
the choice of functions (6) were obtained by the author (2002).
We apply these arguments to the list of singularities.
96
1. The unique solution may be stable or unstable depending on strain-
stress functions (5) (we denote such kind of stability (D»;
2. One stable solution, or two stable (S) + one unstable (U);
3-4. No solution or S+U;
5. No solution, S+U, or D+U;
6. S, D, or S+U+D;
7. No solution, U+U, S+U, or S+U+U+U;
8. No solution, or S+N;
9. U, or S+U+U.
These results allow to understand the dynamics of system (1). In such
cases where either no solution, or no stable solution exists, the system
experiences an instantaneous velocity jump - impact by friction. If there
exist one stable and one or more unstable solutions, the stable one is realised
for most initial conditions. Finally, if we have two stable solutions at once,
each of them can be realised for certain initial conditions. This situation is
similar to the famous example of bifurcation - the Euler pinned column.
3.3. Example
Consider a unbalanced wheel, rotating in (unilateral) contact with two

orthogonal walls (see Fig.3) (Painleve (1905) considered a similar example
in the case of one wall).
Figure 3. Wheel with two frictional contacts
The equations of motion are (we assume that the wheel has unit mass):
97
x=X +Rz -/1I RI' y=y +~ +f12Rz

p2ip =M -a~ +bRz -/-4(R +b)RI - /12(R +a)Rz
ql =y-R-b?O, q2 =x-R-a?O, a=rcosrp,b=rsinrp (7)
where x, yare co-ordinates of mass centre G, rp is angle between OG

and x - axis, r =1 OG I, R - wheel radius, p - gyration radius, X, Y,M -
external forces and angular momentum, RI ,Rz - normal reactions of the
constraints, /11,/12 - coefficients of dry friction. Equations in form (5)
(omitting superscripts) can be obtained by double differentiation of ql and
q2 with respect to (7). We get finally the following expressions for the
elements of matrix A:
p2 ~I = p2 + a 2 +a(R +b)flt, p2 ~2 = /12p2 +a(R +a)/12 - ab

p2 Azi = _fltp 2 -b(R +b)flt -ab, p2Az2 = p2 +b 2 -b(R +a)/12
p2!1 = p2 + r2 + R(a/11 -b/12) + flt/12 (p2 + r2 + Ra + Rb).
(8)
In the simplest case where the wheel is homogeneous and flt =/12 =/1 ,
we have a = b = 0 and AlJ > 0, A22 > 0,!1 > o. Therefore, a unique solution
of system (5) exists for any external forces. However, the stability problem
is still open. To solve it, let reaction forces be similar linear functions of
deformations: RI = gEl + / &pRz = gE2 + /&2· Then equation (5) is
81+ /&1 + gEl +/1(/&2 +gE2)=-BI

(9)
82 + f&2 + gE2 - /1(/&1 + gEJ = -B2
If BI and B2 are negative, system (9) has an equilibrium where EI and

E2 are positive. Linear analysis leads to the following condition of stability:
(10)
In particular, if the damping coefficient / is small, the solution to (9) is

unstable. This corresponds to squeal of breaks (cf. Popp, Rudolph (2001».
In general case of non-symmetric wheel (r *- 0) the quantity ~ I might
be negative if a < 0 and flt is sufficiently large. Similarly, Az2 might be
negative if b > 0 and /12 is sufficiently large. Then, !1 might be negative if p
and r are small with respect to R. Therefore, we can expect to encounter
here all the cases 1-9. For example, if a=O,b=-r,r=p=O.IR, then
AlJ > 0, A22 > 0, AI2 > 0,!1 = p2(2 + 10/12 - 8/-4/12) Thus, for /11 sufficiently
98
large, we obtain the cusp (case 2). Here for certain values of external forces
we have three possibilities: RI > 0, IS> 0 (unstable), iiI > 0, IS> 0 (stable),
and ii2 > 0, ~ > 0 (stable). Which of these stable solutions is "proper",
depends on small perturbations of initial position and velocities. This
example is considered in Ivanov (2002) in more detail.
Acknowledgments
This work is supported by the Russian Foundation for Basic Research

(project 99-01-00281).
References
Cottle, R.W., Pang, lS., and Stone, R.E. (1992) The linear complementarity problem.
Academic Press, Chichester.
Ivanov AP. (2002) On dynamical paradoxes in systems with dry friction. To appear
Klein, F. (1910) Zu Painleves Kritik des Coulombschen Reibungsgesetze. Zeitsch.Math.Phys.
58, 186-191.
Lecomu, L. (1905) Sur la loi de Coulomb. C.R.Acad.Sci.Paris. 140,847-848.
Lotstedt, P. (1981) Coulomb friction in two-dimensional rigid body systems. ZAlvIM. 61,605-
615.
Lotstedt, P. (1982) Time-dependent contact problems in rigid body mechanics. Math.Progr.
Study. 17, 103-110.
Painleve, P. (1905) Sur 1es lois du frottement de glissement. C.R.Acad.Sci.Paris. 121, 112-
115; 141,401-405,546-552.
Pang, l-S., Trinkle, lC. (1996) Complementarity formulations and existence of solutions of
dynamic multi-rigid-body contact problems with Coulomb friction. Math.Progr.73, 199-
226.
Popp, K., Rudolph M. (2001) Brake squeal. Detection, utilization and avoidance ofnonlinear
dynamical effects in engineering applications. Shaker Verlag. Aachen. 197-225.
Prandtl, L. (1910) Bemerkungen zu den Aufsatzen der Herren F.Klein, R.v.Mises und
G.Hamel. Zeitsch.Math.Phys. 58, 196-197.
Trinkle, lC., Pang, l-S.,Sudarsky, S., Lo, G. (1997) On dynamic multi-rigid-body contact
problems with Coulomb friction. ZAlvIM. 77,267-279.
Whitney, H. (1955) On Singularities of Mappings of Euclidean Spaces I. Mappings of the
Plane into the plane, Ann.Math. 62, 374-410.
INITIATION OF FRICTION INSTABILITY
ON A PLANE FAULT SYSTEM
loan R. IONESCU
Labomtoire de Mathematiques, Universite de Savoie, 73376 Le Bourget du Lac, France
Michel CAMPILLO
Labomtoire de Geophysique Interne et Tectonophysique, BP53, 38041 Grenoble, France
Cristian DASCALU
Institute of Applied Mathematics, P.O. Box 1-22, 70700 Bucharest, Romania
Pascal FAVREAU
Institute of Crustal Studies, University of California, Santa Barbam, CA 93106, USA
Christophe VOISIN
Dept. of Geological Sciences, San Diego State University, San Diego, CA 92182, USA
Abstract We study an unstable elastodynamic process during the initiation phase (i.e. the
period between a perturbation of a unstable state and the onset of rupture propa-
gation associated with the seismic wave radiation). We consider the elastic anti-
plane problem for a system of finite faults under a slip-weakening friction law.
A spectral analysis is used to determine the existence, or not, of a catastrophic
evolution ofthe slip. We find that long initiation durations are expected. We also
investigate the possibility of defining an effective friction law for a finite fault
with a small scale heterogeneity. The "spectral equivalence" between an het-
erogeneous fault system and an homogeneous fault is pointed out. Surprisingly
good agreements are found between the heterogeneous fault model and the ho-
mogeneous fault with an effective friction law. Finally we analyze the initiation
pattern as a possible signature of instability and we show how the weakening
rate is transmitted in the elastic medium through a "domain of confidence".
99
J.A. C. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 99-108.
100
1. Introduction
Laboratory experiments on friction (Dieterich (1979), Ohnaka et al. (1987),
Ohnaka and Kuwahara (1990» have pointed out the existence of a phase of
slow but accelerating motion, called the initiation or nucleation phase. This
initiation stage precedes the propagation of the dynamic instability along the
fault surface. The pre seismic slip associated with the nucleation process should
be recognized as a manifestation of some phenomena precursor to the rupture.
Following these experiments, we shall consider here a slip-weakening friction
law.
The elastic quasi-static problem with slip dependent friction was studied
in Ionescu and Paumier (1996) where results concerning non homogeneous
bifurcation of the static equilibrium positions were obtained. Having in mind
the multiplicity of the equilibrium positions they concluded that it is difficult
to predict the new equilibrium position with a quasi-static analysis and that a
dynamic analysis is required.
We shall concentrate here on the elastodynamic analysis of the friction in the
anti-plane case. More precisely, we focus on the initiation of the shear process
during the weakening stage to point out simple mathematical properties of its
unstable evolution as a result of a slip dependent friction law. Our aim is to
describe the growth of the instability in a form very simple to evaluate and to
interpret. The present paper does not discuss the long term evolution of the
system as done in Cochard and Madariaga (1994), Geubelle and Rice (1995).
2. The evolution problem

We consider the 2D anti-plane shearing on a bounded fault region r f in-
cluded in the plane y = 0 in an homogeneous linear elastic space. The fault
region can be composed of a set of simple faults on which the contact is de-
scribed by a slip dependent friction law. We assume that the displacement
field is 0 in directions Ox and Oy and that U z does not depend on z. The
displacement is therefore denoted simply by w(t, x, y). The elastic medium
has the shear rigidity a, the density p and the shear velocity c = Va /
p.
The non-vanishing shear stress components are (Jzx = T(: + a8x w(t, x, y)
and (Jzy = T':t + a8y w(t,x,y), and the normal stress on the fault plane is
(Jyy = -8 ( 8 > 0). Let us assume in the following that the slip and the
slip rate are nonnegative. Having in mind that we deal with a fault plane and
with the evolution of one initial pulse, we may put (for symmetry reasons)
w(t, x, y) = -w(t, x, -y), hence we consider only one half-space y > O. The
equation of motion is
(1)
101
°
for t > and y > 0. The boundary conditions on y = °are:
W(t,x,O) ° if x ¢ r/, (2)
aoyw(t,x,O) +r~ = ~(x,w(t,x,O)))S, if Otw(t,x,O) > 0, (3)
aoyw(t,xO) +r~ < ~(x,w(t,x,O))S if Otw(t,x,O) = 0, (4)
where ~ (x, s) is the coefficient offriction which is a function of the slip sand
may be non-homogeneous. The initial conditions are denoted by Wo and WI,
that is,
w(O,x,y) = wo(x,y), Otw(O,x,y) = WI(X,y). (5)
Since our intention is to study the evolution of the elastic system near an un-
°
stable equilibrium position, we shall suppose that r;:O = S~(x, 0). We remark
that taking was a constant satisfies (1)-(4); hence w == is an (unstable) equi-
librium position, and wo, WI may be considered as small perturbations of the
equilibrium.
In order to give a non dimensional formulation we introduce a, the charac-
teristic length, and we put Xl = Xf a, X2 = Y f a.
°
Suppose that the initial perturbation is small and the nonlinear function ~
may be approached in a neighborhood of s = by its linear approximation i.e.
(6)
and we introduce the non-dimensional function:
where (7)
We can state now the following linearized evolution problem:
oltw(t, Xl, X2) (cfa)2\1 2w(t, XI, X2) (8)
w(t, Xl, 0) 0, for Xl ¢ r I, (9)
OX2 W(t,XI,0) -,B(XI)W(t,XI, 0), for Xl E r/,(lO)
W(0,XI,X2) = WO(XI,X2), Otw(O, XI, X2) = WI (Xl, X2). (11)
3. The spectral problem

Let us consider the following eigenvalue problem connected to (8)-( 11): find
cP : R x R+ -7 Rand..\2 such that:
\1 2CP(XI' X2) ..\2cp(XI' X2) for X2 > 0, (12)
CP(XI,O) = 0, for Xl ¢ r I, (13)
OX2CP(XI,0) = -,B(XI)CP(XI,O) for Xl E r I' (14)
Since we deal with a symmetric operator we have real-valued eigenvalues ..\2,

i.e . ..\ is real or purely imaginary.
102
Two techniques are used to solve the above eigenvalue problem. The first
one is based on the equivalence with the following hyper-singular integral
equation for <P(XI, 0):
,B(XI)<P(XI,O) = -~FP
7r
r <p(s,O)KI~As I-s - ~l I) ds,
ir, Xl
(15)
where A 2:: 0, K I is the modified Bessel function of the second kind and the
integral is taken in the finite-part sense. In the case of homogeneous single
fault system (i.e. r / = [-a, a], and ,B(XI) = canst) this integral equation
has been solved by Dascalu et al. (2000) for small values of A using a semi-
analytical method. Recently, this method have been improved by Dascalu and
Ionescu (2001) to work in the case of a system of multiple and homogeneous
faults (i.e. ,B(XI) = ,Bk on the fault k) and for arbitrary A.
The second technique, developed in Voisin et al (2001) uses a finite element
approach. To do this the finite fault zone is embedded in a bounded elastic
domain n =] - L, L[ x ]0, L[. The infinite elastic half space is limited by a
fictitious boundary all over which the displacement is negligible, i.e. a null
displacement all along r d, the part of the boundary of n which is not on the
fault r /' is imposed. The variational formulation of (12)-(14) is
r V<P . VVdXldx2 - ir,

in
r ,B<PVdXI = _A 2 r <PVdXldx2,
in
for all functions v E Vh (Vh is a finite element space of dimension N, com-
posed of continuous and affine functions over each triangle) such that v = 0
on rd.
4. Spectral expansion and the dominant part

Let us denote by (A~, <pn) the associated eigenvalues and the eigenfunctions
of (12)-(14) and let N be such that A5 > AI > ... > Ah-l > a > Ah > ....
The solution of (8)-(11) can be generically written (in its spectral expansion)
as:
w = wd+ww,
where w d is the "dominant part" and WW is the "wave part", given by:
~l sinh(cIAnlt/a)
f;:o [cosh(cIAnlt/a)Wn + a
0 I
CiAni Wn]<Pn(Xl, x2),(16)
~ 0 sin(cIAnlt/a) I
n~[COS(CIAnlt/a)Wn + a CiAni Wn]<Pn(XI, X2). (17)
where wg, W~ are the projections of the initial data on the eigenfunctions.
We remark that the part of the solution associated with positive eigenvalues
103
Full solution Dominant part
-+_.... n. I
dstO'lCe (m)
OQO """ """
r-;--+--+---J. .. .,..~ "'--4--+--1--1-•. ,,· ~

,,-rt-:l:--+--ll.. ~. g -r-r--::±:-+-I-...,.. ..§.
-r----I· .. .,..f co---+···~f
-i -i
Figure 1. Comparison between the "Full Solution" Otw(t, x, 0) (computed with a FD

method) and the "Dominant Part" OtWd(t , x ,0) (analytical expression) during initiation t E
[0, Tel. Note that the two solutions are indistinguishable (upper plots). The lower plots are
enlarged views of the upper ones in a time window just after the application of the initial condi-
tions (t=O). Note that in the lower plots the slip rate is of order oflO- 4 msec- 1 and in the upper
ones is of order of Imsec- 1 . The violation of causality is clear, but the difference between
complete and dominant solution has an amplitude less than the initial perturbation.
A2 will have an exponential growth with time. Hence, after a while this part
will completely dominate the other part which has a wave-type evolution. This
behavior is the expression of the instability caused by the slip weakening fric-
tion law. The use of the expression of the dominant part leads to a solution in
which the perturbation has been severely smoothed by the finite wavenumber
integration. The propagative terms are rapidly negligible and the shape of the
slip distribution is perfectly described by the dominant part.
104
The spectral approach has the advantage to be extended relatively easily to

other cases. Indeed, for an infinite fault (which implies a continuous spectrum)
the analytical expression of the dominant part was first obtained by Campillo
and Ionescu (1997) in the anti-plane case, and thereafter by Favreau et al.
(2000), (2001) in the in-plane and in the 3D cases.
The accuracy of the approximation of the dominant part is illustrated by
the numerical comparison. The dominant part was compared in Campillo and
Ionescu (1997), Favreau et al. (2000),(2001), Dascalu et al. (2000) and Voisin
et al. (2001) with the full solution computed by a finite differences method
described in Ionescu and Campillo (1999). In all these cases the difference
was found to be of the order of the initial perturbation, which is negligible
with respect to the final amplitude of the solution (see Figure 1 top) .
The dominant part is not a complete solution but a part of the solution as-
sociated with the real positive eigenvalues. Since all propagating terms are
omitted, the dominant part is not causal, and is not expected to be so. In the
case of the initiation, the amplitude of the mismatch between complete and
dominant solution at the causality limit scales with the initial perturbation (see
Figure 1 bottom).
Recently, Knopoff et al. (2000) presented an analytical study of the initia-
tion of shear instability under slip weakening friction for an infinite homoge-
neous fault in the anti-plane case. They have considered the perturbation on
the fault (i.e. of the friction constitutive equation) and obtained an elegant and
complete solution of w. More recently, Ampuero et al. (2001), which used a
different technique, confirmed the qualitative behavior of the solution given by
Campillo and Ionescu (1997) via the dominant part w d •
5. Stability analysis
One can easily remark that w == 0 is a stable position if >'5 < 0 (i.e. N = 0).
In this case the dominant part w d vanishes and the system has a stable behavior.
Hence it is important to obtain a simple condition on the distribution /3(X1)
which determines the positiveness of the eigenvalue >'5.
Let us suppose in the following that we deal with a homogeneous fault sys-
tem (i.e. /3(xd = canst). In this case the spectrum (>';(/3))n~O is a function
of the non-dimensional parameter /3. Let 0 < /30 ::; /31 ::; ... be the intersection
points of the curves /3 ----7 >'~(/3) with the axis >.2 = 0, i.e. >'~(/3k) = O. The
constant /30 depends only on the geometry of the anti-plane problem (i.e. the
distribution of the faults) and it is independent on all physical entities involved
in our problem. This non dimensional parameter gives quantitatively the limit
between the stable (/3 < (30) and unstable (/3 > (30) behaviors of the fault.
In the case of a homogeneous single fault system (i.e. r f = [-a, aD the
value of /30 was computed by Dascalu et al. (2000) to be /30 = 1.15777388 .....
105
For each representative physical quantity which is included in the non di-
mensional parameter f3 (the friction weakening slope SJ.L'(O) = (J.Ls -J.Ld)S/ L e,
the interface stiffness G/ a, the fault half length a, the elastic bulk modulus G,
etc... ) we can define a "critical" value.
6. Renormalisation of a heterogeneous fault

Friction is a phenomenon that concerns both microscopic and macroscopic
scales. The phenomenon is observed in seismology at the scale of the seismic
waves, that is kilometric. The smallest scales of heterogeneity cannot be ob-
tained directly. Even the laboratory measurements Dieterich (1979), Ohnaka
and Kuwahara (1990) do not represent the local boundary condition at the mi-
croscopic scale but the macroscopic frictional behavior of the elastic bodies in
contact at the scale of the samples. We aim to check the assumption that there
exists an equivalent macroscopic friction law for the problem of a fault with
small scale strength heterogeneity. By equivalent, we mean that this "macro-
scopic" effective law is sufficient to describe the global behavior of the fault.
Our analysis concerns primarily the initiation phase which is an unstable and
highly dynamic stage of rupture.
COMPLETE SOLUTION VB EFFECTIVE SOLUTION
.,go.s
5
(e) 0
0.6
iil 0.4
>
~ 0.2
en O~~~~~~~~~
-4000 -3000 -2000 -1000 0 1000 2000 3000 4000
POSITION (ml
Figure 2. Comparison between two slip rate distributions Otw(t, x, 0) on the fault computed
with a FD method at the same times during the initiation phase. The slip rate corresponding to
the heterogeneous model of the local friction (continuous line) vs the slip rate corresponding to
the homogeneous model with the effective friction law (dashed lines).
Let us suppose in the following that we deal with linear slip-weakening

friction, i.e. the constitutive friction is well defined by the parameter f3 dur-
ing the initiation phase. We consider a fault rt;e
composed of N I segments
[bk' dkj, k = 1,2, .. , NI, all identical and of weakening rate f3loe, separated
by rigid barriers. We aim to show that the dynamic behavior of this complex
fault system is equivalent to the dynamic behavior of a simple homogeneous
fault r?UiV = [b1 , dnl with a weakening rate f3equiv.
106
The initiation develops according to a finite set of eigenfunctions associated

with positive eigenvalues that govern the exponential evolution of the instabil-
ity. The process evolution is dominated by the greatest positive eigenvalue >'5.
Indeed, after a period of time the term which involves exp( ct>.o) completely
dominates all other terms in the series, hence we can write:
w(t, x, y) ~ [ch(ct>'o)W~ + sh(ct>.o)]Wp]<Po(x, y). (18)
We define the effective or equivalent friction as the slip dependent function

which generates the same first positive eigenvalue as the one associated with
the heterogeneous problem. This means that we look for f3equiv such that
(19)
In Figure 2 we have plotted the spatial distribution of the slip rate at different
times during the initiation phase for a heterogeneous fault (solid lines) with
a weakening parameter f31oc. With dashed lines we have plotted the spatial
distribution of the slip rate at the same times on a homogeneous fault with a
weakening parameter f3equiv computed from (19). Note that the homogeous
fault gives a good description of the heterogenous fault at a macroscopic scale.
7. The initiation pattern

To characterize the unstable behavior of a fault will be a step in the earth-
quake prediction. Let us suppose that we deal with a slow initiation (i.e. (18)
holds) and let us introduce the ratio, given by:
( ) __ 8x2 <PO(XI, X2) (20)

, Xl, X2 - "" ( )'
'%'0 Xl, X2
This ratio represents the information about the weakening rate of the fault,
when it is defined. Let us analyze now the parameter , in three different
cases. If we use the expression of the dominant part (see Campillo and Ionescu
(1997» in the case of the homogeneous and infinite fault we deduce that the
parameter, is constant in all the elastic space, i.e. we have
__ 8yW d(t,x,y) - f3
, (X, y ) - a d( ) - , everywhere in R x R+, (21)
w t,x,y
In the case of a single finite fault the function 'Y has its support in a narrow
band of the size of the fault length. The general shape of'Y defines an initiation
pattern that qualitatively characterizes the unstable behavior of the fault. The
most interesting point is the existence of a domain, including r f' over which
,(x, y) = f3 (see Figure 3). We now define the "domain of confidence" as
107
X103
4. 2
4.25
4. 3
4.35
4. 4
4 .45
Figure 3. Spatial distribution of the derivative of strain with respect to displacement, param-
eter ,(x, y) computed with a FD method, during the initiation on a single finite homogeneous
fault. Note the constant value of, in abroad region around the fault Dc(fJ) called the "domain
of confidence".
As it follows from Campillo et al. (2001), Voisin et al. (2001) in the case of
a heterogeneous fault, described in previous section, close to each individual
fault segment, an initiation pattern develops in the elastic medium, associated
with a local domain of confidence over which ,(XI,X2) = {3loc' But now,
the striking feature is the existence of a wide domain over which ,(Xl, X2)
is nearly constant, independent of the individual fault segments but closely
related to the whole fault system (see Figure 4). All over this domain, we have
,(Xl, X2) = (3equiv. That is, over this wide domain, it is possible to measure
the collective behavior of all the fault segments, similar to the behavior of a
homogeneous fault.
References
Ampuero J.-P., l-P' Villote and FJ. Sanchez-Sesma (2001) Nucleation of rupture under slip
dependent friction law: simple models of fault zone, submitted to J. Geophys. Res..
CampilIo, M. and Ionescu, I. (1997) Initiation of anti-plane shear instability under slip-dependent
friction, J. Geophys. Res.• 102, 20,363-20,371.
Campillo, M., Favreau, P., Ionescu I.R. and Voisin, C. (2001) On the effective friction law of an
heterogeneous fault, J. Geophys. Res .• 106. 88, 16,307-16322.
Cochard, A and R. Madariaga (1994) Dynamic faulting under rate dependent friction, Pure
Appl. Geophys. 142,419-445.
Dascalu, C., Ionescu, R.I., and CampilIo, M. (2000) Fault finiteness and initiation of dynamic
shear instability, Earth Planet. Sci. Lett.. 177, 163-176.
Dascalu C. and I.R. Ionescu (2001) Spectral analysis of instabilities for the wave equation with
application to eathquake initiation, submitted.
Dieterich, J.H. (1979) Modeling of rock friction 1. Experimental results and constitutive equa-
tions, J. Geophys. Res .. 84, 2161-2175.
108
x 10'
1
1. 1
1. 2
1. 3
1 •
1. 5
1. 6
1. 7
1. 8
1. 9
Figure 4. Spatial distribution of the derivative of strain with respect to displacement, param-
eter ,(x, y) computed with a FD method, during the initiation on a heterogeneous fault. Note
the constant value of, in a "collective domain of confidence" Dc ({3equiv) .
Favreau, P., M. Campillo and I.R. Ionescu (2000) The initiation of inplane shear instability
under slip dependent friction. Bull. Seism. Soc. Am. 89, 1280-1295.
Favreau, P., M. Campillo and I.R. Ionescu (2001) The initiation of shear instability in three
dimensions: theoretical and numerical studies, to appear in 1. Geophys. Res..
Geubelle, Ph. and J.R Rice (1995) A spectral method for 3D elastodynamic fracture problem, 1.
Mech. and Phys. Solids, 43, 1791-1824.
Ionescu I.R. and Campillo M. (1999) The influence of the shape of the friction law and fault
finiteness on the duration of initiation, 1. Geophys. Res. , 104, 3013-3024.
Ionescu, I.R., and J.-C. Paumier (1996) On the contact problem with slip dependent friction in
elastostatics, Int. 1. Eng. Sci. 34(4),471-491.
Knopoff L., J.A. Landoni and M.S. Abinant (2000) The causality constraint for fractures with
linear slip-weakening, 1. Geophys. Res., 105, 28035-28044.
Ohnaka, M., Kuwahara, Y. and Yamamoto, K. (1987) Constitutive relations between dynamic
physical parameters near a tip of the propagating slip zone during stick slip shear failure,
Tectonophysics, 144, 109-125.
Ohnaka, M. and Kuwahara Y. (1990) Characteristic features of local breakdown near a crack-
tip in the transition zone from nucleation to unstable rupture during stick-slip shear failure,
Tectonophysics, 175, 197-220.
Voisin C, I. Ionescu, Michel Campillo, Riad Hassani, and Q.L. Nguyen (2001) Initiation process
on a finite segmented fault: a spectral approach, to appear in Geophys. 1. Int ..
AN EXAMPLE OF STICK-SLIP WAVES
Franck Moirot
PSA, Centre Technique La Garenne-Colombe
Paris, France
Quoc-Son Nguyen
Laboratoire de Mecanique des Solides, CNRS-umr7649
Ecole Polytechnique, Paris, France
Abstract An analytical solution representing a family of stick-slip waves is obtai-

ned in a simple example modelling the dynamic behaviour of an elastic
cylindrical tube in contact with Coulomb's friction with a rigid and rota-
ting cylinder. This family of waves, representing the non-trivial periodic
responses of a continuous system of one space variable, is not classical
in the literature.
1. Introduction
It is well known that in frictional contact of solids, a contact point
may have a slip or stick or separation regime according to Coulomb's law
of dry friction. For dynamical problems, the study of the propagation
of these zones on the contact surface may be interesting in different ap-
plications : ultrasonic motors (cf. Zharii, 1996), seismic predictions and
simulations (cf. Cochard and Madariaga, 1995), mechanical vibrations
and instabilities induced by friction (cf. Martins and Simoes, 1995, Vola
et al., 1999, Oancea and Laursen, 1997, Oestreich et al., 1996). In the
particular case of a solid in contact with a moving obstacle, some recent
results of the literature on the flutter instability of the steady sliding
solution (cf. for example Adams, 1995 or Martins and Simoes, 1995 or
Vola et al., 1999) have suggested that we investigate dynamic periodic
responses. A possible transition of this response to a cyclic response in
the spirit of Poincare-Andronov-Hopf bifurcation, cf. (Nguyen, 2000),
is the long term goal. As an example, the study of the phenomenon of
brake squeal has been discussed in this context, Moirot, 1998. A simple
modelling of a drum brake Ie? ds to the study of an academic example of
109
1.A. C. Martins and M.D.P. Monteiro Marques (eds.). Contact Mechanics. 109-116.
110
encased cylinders, and has enabled us to obtain an interesting example
of stick-slip waves. Such an example is new in the literature.
2. An encased-cylinder problem and the

reduced equations
The mechanical response in plane strain of an elastic cylinder, of in-
ternal radius R and external radius R*, in frictional contact on its inner
surface with a rotating rigid cylinder of radius R + d and of angular
rotation n as shown in Figure 1, is considered when the displacement
is assumed to be homogeneous on its outer surface. Coulomb's law of
dry friction is assumed with a constant friction coefficient j. An in-
teresting approximation reducing this two-dimensional problem to an
one-dimensional problem of contact with friction is discussed here. Our
objective is to demonstrate the existence of periodic dynamic responses
in the form of stick-slip waves. Although stick-slip solutions have often
been discueeds in statics, we emphasize that such a non-trivial example
of stick-slip vibrations is not classically known in the literature for conti-
nuous systems. The construction of periodic dynamic responses has prin-
cipally been discussed numerically, (cf. for example Oancea and Laursen,
1997, Oestreich et al., 1996, Vola et al., 1999), the work of Zharii, 1996
on surface waves generated by piezo-electrical sources on a half-space
boundary in contact with an obstacle remains an exception, and his re-
sults deal only with slip waves. Non-dimensional variables are introduced
_ u _ it _ r _ pR2 n2 ~ _ R* J< _ d _ 0- • _ du
U - R' a - E' r - R' , - E ' ." - If' U - R' t - ~d, u - dt·
Governing equations are the kinetic relations, the fundamental law, the
linear elastic constitutive equations, the boundary conditions, the uni-
lateral contact conditions and Coulomb's laws:
€= (\7u)s,
Diva = ,iL,
a = (1+ V )(1-2v)Tr(€) 1+ l~v€'
u(~, (), t) = v(~, (), t) = 0, (1)
a rr (1 , (), t) = -p((),t), a r o(l,(),t) = -q((),t),
u ~ 8, p ~ 0, p( u - 8) = 0,
Iql ::; jp, q(l - v) - jpl1 - vi = 0.
The steady sliding solution is

111
An interesting modelling of the problem is obtained when the displace-
ment is approximated in the form
u = U{O, t}X{r}, v = V{O, t}X{r}, X{r}

1 e
= ~2 _ 1 {-;- - r}. {3}
In this approximation, the following local equations are obtained from
virtual work equation
(; - bU" - DV' + gU = P,
{ V - aV" + DU' + hV = Q, (4)
P ~ 0, U - 8 ~ 0, P(U - 8) = 0,
IQI ~ I P, Q(1 - V) - I P11- VI = o.
bA _ 2ii+2(e2-1)b h - 2eb - - 1-11

°
where ' denotes the derivative with respect to and a = ~~, b =
b -- 2(1+11)'
1 A
"(B' g - "(B , - "(B' a - (1+11)(1-211)' =
- ~ + 1+e > 0 B = ~ + 1-3e > 0 D = aCl-bC2 0 1 =
~ -1 2 , ~ -1 4' A'
- -
2l:~~{ 1 > 0, O2 = - 2l:~~{ 1 + 2~2 > O. The coupling coefficient
D between the normal and tangential displacements can be positive or
negative according to the values of v and ~. The steady sliding solution
is given by Ue = 8, Ve = 8Ig/h, P = Pe and Qe = I Pe.
The steady sliding response is unstable, cf. Martins and Simoes, 1995
or Adams, 1995. The proof of this result is straightforward under the
assumption of sliding motions. In this case a small perturbation motion
is described by U = Ue, V = Ve + V*, P = Pe + P* and Q = Qe + Q*. It
follows that
V* - aV:' + IDV: + hV* = o. (5)
A general solution is sought in the form V* = esteikO. It follows that
-s2 = ak 2 + h + iklD. If I = 0, then S = ±iWk with w~ = ak 2 + h.
Thus two harmonic waves propagating in opposite senses of the form
cos (kO ± wkt + IP) are obtained as in classical elasticity. If I > 0 and
D > 0, then S = ±(Srk +iSik), Srk > 0, Sik < 0, thus a general solution
ofthe form V* = e±Srkt cos (kO ± Sikt + IP) is obtained and represents two
waves propagating in opposite senses : an exploding wave in the sense of
the implied rotation, and a damping wave propagating in the opposite
sense. If I > 0 and D < 0, the exploding wave propagates in the opposite
sense since the previous expression of S is still valid with Srk > 0 and
Sik > o.
3. Existence of stick-slip waves

It is expected that in some particular situations, there is a dynamic bi-
furcation of Poincare-Andronov-Hopf's type. This means that since the
112
steady sliding response is unstable and there is a flutter instability, the
perturbed motion may eventually become a periodic response. This tran-
sition has been obtained numerically in many examples, cf. for example
Oestreich et al., 1996 or Vola et al., 1999, but a mathematical proof is
still lacking since the classical proof in Hopf's theorem cannot be applied
to the dynamic equations with unilateral contact and dry friction. The
existence of possible periodic dynamic solutions is our centre of inter-
est. A periodic solution is sought in the form of a wave propagating at
constant velocity :
U = U(4)), V = V(4)), 4> = () - ct (6)
where c is the non-dimensional wave velocity, U and V are periodic

functions of period T = 2:.
The physical velocity of the wave is thus
C = IclRO and the associated dynamic response is periodic of frequency
IclkO. The propagation occurs in the sense of the rotation when c > o.
According to the regime of contact, a slip wave, a stick-slip wave, a slip-
separation wave or a stick-slip-separation wave can be discussed. Only
the two first cases will be discussed here. The governing equations of
such a wave follow from (1) :
(~- b)U" - DV' + gU = P,

{ 2 - a)V" + DU' + hV = Q,
(c
(7)
P ~ 0, U ~ 0, P(U - 0) = 0,
IQI ~ f P, Q(1 - V) - f PI1 - VI = o.
For a slip wave, there is no separation and Q = f P everywhere when
there is positive slip (1 - V ~ 0). Thus the governing equations of a
positive slip wave are :
U = 0, -DV' + go = P, (c2 - a)V" + hV = fP (8)
It follows that V = Ve + ~ with

(c2 - a)~" + fD~' + h~ = 0 (9)
Thus ~ = ercp with r2 - 2ar + {3 = 0, a = 2(i!!a) ' {3 = c2~a. It

follows that r = a ± Ja 2 - {3. This expression shows that no periodic
solution can be obtained and thus no periodic response under the form
of a positive slip wave can be found.
A solution of (7) in the form of a stick-slip wave is now sought. It is
assumed that there is a positive slip regime in the interval (0, 'liT) and
a stick regime in the interval ('liT, T) :
113
- in the stick region (WT, T), governing equations are

, 1 D
U = 8, V = --, P = - + g8 > 0, Q = hV, (10)
e e
- in the slip region (0, WT), governing equations are
U = 8, P = g8 - DV' ~ 0, Q = f P, 1 + eV' ~ 0, (11)
(e 2 - a)V" + fDV' + hV - fg8 = 0. (12)

The last equation shows that V = Ve + Ll. in the interval
(0, WT) with
(9). The case a < (3 is considered thus e > a+ t:f2. Let w = ";(3 - a 2 •
2 2
Since r = a ± iw, the general expression of Ll. is Ll. = eCtCP(M cos w¢> +
Nsinw¢». There are 4 unknowns (e, W,M,N). The continuity of the
displacement gives V(O) = V(T). The stick region leads to two supple-
mentary equations V'(O) = V'(WT) = -lie. Thus, the following system
of equations must be satisfied :
M = ealJ!T (M cos wWT + N sin wWT) _ T(l; IJ!) ,

{ ~l = aM +wN, (13)
~l = aealJ!T (M cos wWT + N sinwWT)+
wealJ!T( -M sinw'ItT + N cosw'ItT)
e
2
> a+ ~ 4h '
Ifg8 + hLl.1 < f( ~ + g8)
{
(14)
g8 - DLl.' ~ 0, 1 + ell.' ~°V¢> E (WT, T),
V¢> E (0, WT).
A family of stick-slip waves is considered for which the continuity of Q
at ¢> = T is ensured :
(15)
This condition implies that hV(T_) = fg8 + 1..f!- and leads to the follo-
wing system of governing equations and inequalities after the elimination
of M and N:
T(l - w) + ~ealJ!T sinw'ItT = 0,
{ 1 + ~ealJ!T sinwWT - ealJ!T cos wwT = 0,
(3- (16)
a -- ~,
-fD
v - a2,
- c2h-a' W -_.1(3
Ifg8 + hLl.1J!1 ::; 1..f!- + fg8, 1 + ell.' ~ 0.
The equations can be solved numerically by Mathematica for different
values of k and f. It is found that e must have the sign of D; these
waves propagate in the sense of the previous exploding perturbed mo-
tions. The value of 8 must be strong enough to avoid possible separations.
114
The results (phase diagram, displacement and reactions) are given res-
pectively in Figure 2 and Figure 3 for the particular case of a common
metal with f = 1, k = 8 and D < O. In this case, the physical velocity
is c = 1255m/s and the wave propagates in the opposite sense to the
implied rotation.
4. Transition of the steady sliding to a stick-slip

wave after perturbations
Since the steady sliding solution is unstable, a small or finite pertur-
bation of this solution will lead to a different response. Some numerical
calculations of the system (4) have been performed in order to study
the possibility of transition to a periodic response. It has been found
that the rate of convergence and the limit response depend strongly on
the initial conditions. Some stick-slip solutions are stable and present a
specific domain of attraction. Figures 4,5 and 6 present the numerical
results obtained after a direct integration of the system following an ex-
plicit scheme of time integration with a common regularization of the
friction force-slip velocity relation.
References
Adams, G. (1995). Self-excited oscillations of two elastic half-spaces sliding with a
constant coefficient of friction. J. Appl. Mech., 62 :867-872.
Co chard, A. and Madariaga, R. (1995). Dynamic faulting under rate-dependent fric-
tion. Pure Appl. Geophys., 142 :419--445.
Martins, J. and Simoes, F. (1995). On some sources of instable/illposedness in elasti-
city with Coulomb's friction. In Contact mechanics, New York. Plenum Press.
Moirot, F. (1998). Etude de la stabilite d'un equilibre en presence du frottement de
Coulomb. Application au crissement des freins a disque. These, Ecole Polytech-
nique, Paris.
Nguyen, Q. (2000). Stability and Nonlinear Solid Mechanics. John Wiley and Sons,
Chichester.
Oancea, V. and Laursen, T. (1997). Stability analysis of state-dependent dynamic
frictional sliding. Int. J. Nonlinear Mech., 32 :837-853.
Oestreich, M., Hinrichs, N., and Popp, K. (1996). Bifurcation and stability analysis
for a non-smooth friction oscillator. Arch. Appl. Mech., 66 :301-314.
Renard, Y. (1998). Modelisation des instabilites liees au frottement sec des solides
elastiques, aspects tMoriques et numeriques. These, Univ. Grenoble.
Vola, D., Raous, M., and Martins, J. (1999). Friction and instability of steady sliding
squeal of a rubber/glass contact. Int.J. Num. Meth. Engng., 46 :1699-1720.
Zharii, O. (1996). Frictional contact between the surface wave and a rigid strip. J.
Appl. Mech., 63 :15-20.
115
Figure 1 - An example of encased

cylinders with a stick-slip wave in
the case k = 2.
0.02
0.01
I
I
0.00
0.0
I ~~~_~.J
0.2 0.4 0.6 0.8
~ (radian)
36
p
..
Figure 2 - The phase diagram 32
__ _ _
of the periodic solution associated
with a stick-slip wave is given for
k = 8, f = 1 and 8 = 0.005. :l~ ~_. ~_~~
M U
~
U
(radian)
U U
l-~-/-I
1.5 ---.--.-.-~. ----.- ---~ -- ------,
-I
0.5
t ) .•
Q I I
--0.5
-1.5 ______ L _ _ _ _ _ _ -----' _ _ _ _ ~ _ _ _ _ ~_ --.--~----~.

M U U U U
-0.01 0.01 0.01 ~ (radian)
V
Figure 3 - An example of stick-slip

wave with k = 8, f = 1 and 8 =
0.005. In this case, W = 0.839, c =
12.55. The variations of V, P and Q
are given as a function of 0 ::; if> ::;
27r /k.
116
Figure 4 - A numerical illustration

of the transition to a periodic li-
mit response. A mode-3 stick-slip
wave is obtained here from a par-
0.5
ticular initial condition of displace-
ment and velocity
r -.
- 0.5
-1
'__ ',',', '... .',' ,' ,' ,Yo' """""'I

, ., '..,:: ...... " .• , ' 1
O.
II "'<.
• dl
l
...
""" ,'
.'I~,
I" I'"~
I'
~.
)6
)5
'~.,
• \1 .. ! . I" H
))
)2
II
10 20 )0 40
22
20
Figure 5 - The phase diagram of 10 20 30

the limit response (mode-3 stick-
slip wave)
Figure 6 - Limit response : rela-

tive velocity, normal pressure and
0.5
tangent force obtained by numeri-
cal calculations
- 0 .5
-I
LONGITUDINAL WAVES IN ELASTIC
RODS WITH DISCONTINUOUS CROSS
SECTIONS
Werner Schiehlen, Bin Hu

Institute B of Mechanics, University of Stuttgart
Pfaffenwaldring g, D-70550 Stuttgart, Germany
Peter Eberhard
Institute of Applied Mechanics, University of Erlangen
Egerlandstr. 5, D-9I058 Erlangen, Germany
Abstract In this paper longitudinal impact waves in an elastic rod with discontin-
uous cross section struck by a rigid sphere are studied numerically and
experimentally. In a rod with finite length, waves move back and forth
along the rod; therefore, it may happen that two incident waves cross
a discontinuity interface simultaneously. Thus, the reflected and trans-
mitted waves induced at the interface will influence the incident waves
due to the wave reflection at both ends of the rod. For these general
cases, to the authors' knowledge, there are no computational methods
for the transmitted and reflected waves. This problem will be addressed
in this paper in some detail. In the numerical simulations, the classical
Hertzian contact law and the wave equation are combined to compute
impact responses. In the experimental setup, advanced instruments like
Laser-Doppler-Vibrometers for measuring the high frequency impact re-
sponses are used. It is shown that numerical results computed by this
approach agree very well with experimental results.
1. Introduction
During the last decades, scientists and engineers have developed an
increasing interest in the solution of problems involving the impact of
solid bodies. Among solid bodies, rigid bodies and linearly elastic rods
are the simplest engineering structures. However, even wave propagation
phenomena in elastic rods resulting from longitudinal impacts by rigid
bodies may be complex due to the dynamical boundary conditions and
117
118
wave reflections. Considerable efforts have been put in understanding
the mechanism of wave reflection and transmission in rods. Many results
for longitudinal waves in rods produced by impacts were reviewed by
Goldsmith (1960), Johnson (1985) and Stronge (2000). Experimental
results were reviewed by Al-Mousawi (1986).
When an incident wave crosses an interface between two different cross
sections or materials, a part of the wave is reflected and another part
transmitted. Among the few investigations dealing with longitudinal
impact waves in discontinuous rods, the important references by Rip-
perger and Abramson (1957), Plunkett and Doolen (1963), Habberstad
and Hoge (1971), Barez, Goldsmith and Sackman (1980) are mentioned
here. However, in a rod with finite length, two waves move along the rod
in opposite directions. Therefore, it may happen that two incident waves
cross an interface simultaneously. Further, the reflected wave and the
transmitted wave induced at the interface will influence the two incident
waves crossing the interface later. For these general cases, to the authors'
knowledge, there are no computational methods with one-dimensional
wave theories for the wave propagation in the rod. This problem will
be addressed in this contribution in some detail. Longitudinal impact
waves in an elastic rod with a discontinuous cross section struck by a
rigid sphere are investigated analytically and experimentally.
2. Mathematical model for impact waves

The inhomogeneous rod under consideration can be considered as be-
ing composed of two linearly elastic homogeneous rods. The i-th rod
has Young's modulus Ei, density Pi, cross sectional area Ai and length
Li for i = 1,2, see Figure 1. Their axial displacements are denoted by
Ul(XI, t) and U2(X2, t), respectively. The striking sphere with mass m is
assumed to be rigid and its displacement is denoted by q(t).
~q
Vo..
C3 L..---L----''---'--''''''-'-~ m2, E2, v2' A2,P2
~I
Figure 1. Longitudinal impact of a sphere striking an inhomogeneous rod

119
The governing partial differential equations for longitudinal waves in

both homogeneous rods read
(1)
where CI and C2 are the wave propagation velocities in the rods to be

calculated from material constants,
(2)
At the interface of the two homogeneous rods, i.e, at Xl = L1, X2 = 0,
there are constraints on the displacements
(3)
and on the interface forces
(4)
The equation of motion of the rigid sphere reads
mij(t) = -F(t) (5)

where F{t) is the contact force and the dots denote the differentiation
with respect to time. Without loss of generality, it is assumed that the
initial conditions of the striking sphere are
q(O) = 0 and q(O) = Vo (6)

and the inhomogeneous rod is initially at rest
The boundary condition for the inhomogeneous rod at the free end X2 =
L2 reads
oU2{L2, t) = O. (8)
OX2
For the contact end Xl = 0, different impact theories give different
boundary conditions. A comparison of different theories applied to lon-
gitudinal impacts of a rigid sphere striking an elastic rod can be found
in Hu and Eberhard (1999). According to St. Venant's impact theory,
the boundary condition for the contact end reads
Ul(O, t) = q{t). (9)

120
However, this theory is not suitable here since the surfaces of the col-
liding bodies are not planar. Instead, a boundary approach combining
both, St. Venant's theory and the Hertzian contact law, is used in the
following. Details about the boundary approach are given in Hu and
Eberhard (1999). According to this approach, the boundary condition
for the contact turns into
-E A 8Ul(O, t) - F( ) (10)
1 1 8 - t.
Xl
The contact force F(t) is given by the static Hertzian contact law,
2
F(t) = ko~(t) (11)
where o+(t) is the indentation between the striking sphere and the struck
end, i.e.,
(12)
The coefficient k is a constant depending on the geometry and elastic
properties of the sphere and the struck end,
4VR (13)
where R, E and 1/ are the radius, Young's modulus and Poisson's ratio
of the striking sphere, respectively, see e.g. Johnson (1985). Now, the
impact problem of the sphere striking the inhomogeneous rod is mathe-
matically described by equations (1-8) and (10-13). In the following this
mathematical model will be further simplified.
The general solutions of the partial differential equations (1) due to
d' Alembert, is
Ul(XI, t)= II (CIt - xI) + 9l(Clt + xI), (14)

U2(X2, t) = h(C2t - X2) + 92 (C2t + X2).
In some references, for example in Stronge (2000), only a forward travel-
ing incident wave crossing an interface is considered. After the incident
wave crosses the interface, a part of the wave is reflected and another
part transmitted. Thus, it is assumed that the function II is a given
function and that before the wave crossing the functions 91 = 0, h = 0
and 92 = 0, and after the wave crossing the function 92 = O. This as-
sumption is in general not true since for finitely long rods there are two
longitudinal waves traveling in opposite directions due to the wave re-
flection at the ends. Further, the transmitted wave h and the reflected
121
wave 91 will influence the forward incident wave h and the backward
incident wave 92. To our knowledge, there are no computational meth-
ods for these general cases where two waves travel with the opposite
directions in every rod simultaneously.
Using the initial conditions (7) for the two homogeneous rods and
equation (14), one can obtain that
h(-e) = -C1 for 0 ~ e~ L1,

h(-e) = -C2 for 0 ~ e~ L2. (15)
The integral constants C1 and C2 can be set to any values since they have
no influence on the displacements U1(Xl, t) and U2(X2, t). For simplicity,
these constants are set to zero resulting in
91 (e) = 0 for 0 ~ e~ L1 ,
92 (e) = 0 for 0 ~ e~ L2 . (16)
Moreover, according to equations (3), (4) and (14), the constraints at

the interface can be written as
h(C1t - Ld + 91(C1t + L1) = h(C2t ) + 92(C2t) , (17)

E1Ad- fHC1t - L1) + 9i(C1t + L1)] = E2A2[- fHc2t) + 9~(C2t)] (18)
where the prime denotes differentiation with respect to the argument.
Further, using equations (8), (10), (11) and (14) yield the boundary
conditions for the free end
(19)
and for the contact end
(20)
Combining the constraints (17) and (18), the boundary conditions (19)
and (20) with the equation of motion of the striking sphere (5), one
obtains a set of nonlinear ordinary differential equations with time delay.
They have to be integrated with very small time steps since the impact
response during contact includes very high frequency components. The
unknown functions are h, 91, h, 92 and q. The initial conditions are
given by equations (6) and (7).
3. Measurement and Comparison

The model behind the experiment is presented in Figure 1, and an
overview of the experimental setup is shown in Figure 2. The steel sphere
122
section. The striking sphere was suspended by a thin kevlar wire forming
a 'V'. In the experiments it is released at a vertical height h above
the axis of the rod leading to the initial impact velocity of the sphere
Vo = J2gh. The rod has cylindrical sections with radii r1 = 10 mm,
r2 = 20 mm and lengths L1 = 1 m, L2 = 0.5 m, respectively. The rod is
suspended by two sets of thin kevlar wires at two locations and carefully
leveled horizontally.
Figure 2. Overview of the experimental setup
The strains were measured with strain gauges and high quality DC-
amplifiers. The strain gauges are of type 3j120XY13, made by Hot-
tinger Baldwin Messtechnik GmbH, and the glue is Z70. The two DC
signal conditioners are of type DMS805B made by Rohrer GmbH with
a frequency range up to 1 MHz. For the measurement of displacements
and velocities, Laser-Doppler-Vibrometers of type OFV-3000jOFV-302
made by Polytee GmbH were used. The measured locations for the
strain signals are marked with letters A, B, C and D in Figure 3 with
the distances 0.2m, 0.8m, 1.2m and l.4m from the struck end of the
rod, respectively. The measured locations for the velocity and displace-
ment signals are marked with x in Figure 3 and the directions of the
-30em '1' 30em '1' aOem .1. 30em .1. 30e~
Figure 3. Locations for the measured signals

123
laser beams radiated from the Laser-Doppler-Vibrometers are denoted

by long arrows. The angles between the laser beams and the axis of the
rod are set to 45 0 except at the free end.
Numerical results for the stress wave and the velocity wave in the rod
for the striking velocity Vo = 0.3 m/ s are shown in Figure 4, where both
the wave reflection and the wave transmission at the interface as well as
the wave reflection at the two ends of the rod can be seen.
velocity mJ
2 2
0.2
2
1.5 1.5 0.1

~
~
V>
S
Vl
S 0 0
-
~ q)
E .5
-I 0 .5 -0. 1
0.5
-2 -0.2
0.5 1 1.5 0.5 1 1.5
position [m] position [mJ
Figure 4. Stress and velocity wave propagation in the inhomogeneous rod
The measured experimental results agree with the numerical results

very well. As an example, Figure 5 shows the comparison between the
strain at position A strain at position C
~
~ 0.2 ... ... .. .. ... : ..............•.. .. ....... ....•. ........ ... ..
····J\WJiV\\l{
f.1
~ 0 ..... .• .... ... .•.. .. .. .. . .....
.
~-0.2~
., ~~'
.
:
o 0.5 1 1.5 2 o 0.5 1 1.5 2

Time [ms] Time [ms]
velocity at the free end
[E~fl~ o 0.5 1
Time [ms]
1.5 2
I:~m
is 0 0.5
Time [ms]
1 1.5 2
Figure 5. Comparison between measured and simulated results

124
measured and the simulated results for the strains at the locations A
and C and the velocity and the displacement at the free end of the rod,
where the simulated results are denoted by black solid lines and the
measured results by gray dashed lines. This also emphasizes that the
computational method using a set of ordinary differential equations with
time delay can be used to predict the longitudinal impact waves with
high accuracy in inhomogeneous rods.
4. Conclusions
In this paper, longitudinal wave propagation in a finite elastic rod
with discontinuous cross sections struck by a rigid sphere is studied nu-
merically and experimentally. A set of ordinary differential equations
with time delay is derived describing longitudinal wave propagation in
the finite rod. Not only is the wave reflection and transmission at the
interface with discontinuous cross sections considered, but also the wave
reflection at both ends of the rod. Experimental results show that the
boundary approach which combines the Hertzian contact law and the
wave equation with boundary conditions satisfactorily predicts the im-
pact wave in the rod with discontinuous cross sections. Moreover, the
computational method developed in this paper may also be used for
rods with different materials, or with both cross sectional and material
discontinuity.
References
AI-Mousawi, M.M. (1986) On experimental studies of longitudinal and flexural wave
propagations: An annotated bibliography, Appl. Mech. Rev. 39, 853 - 864.
Barez, F., Goldsmith, W. and Sackman, J.L. (1980) Longitudinal wave propagation
in axisymmetric structures with material and/or areal discontinuity, Exp. Mech.
20, 325 - 333.
Goldsmith, W. (1960) Impact: The Theory and Physical Behaviour of Colliding Solids,
Edward Arnold Ltd, London.
Habberstad, J.L. and Hoge, K.G. (1971) Effects of discontinuity in a cross section on
an elastic pulse, J. Appl. Mech. 38, 280 - 282.
Hu, B. and Eberhard, P. (1999) Experimental and theoretical investigation of a rigid
body striking an elastic rod, Report IB-32, Institute B of Mechanics, University of
Stuttgart, Stuttgart.
Johnson, K.L. (1985) Contact Mechanics, Cambridge University Press, New York.
Plunkett, R. and Doolen, R.B. (1963) Reflection and transmission coefficients for
stress waves in bars, J. Acoust. Soc. Am. 35, 1457 - 1458.
Ripperger, E.A. and Abramson, H.N. (1957) Reflection and transmission of elastic
pulses in a bar at discontinuity, Proc. 3rd Midwestern Conference on Solid Me-
chanics, 135 - 145.
Stronge, W.J. (2000) Impact Mechanics, Cambridge University Press, Cambridge,
U.K.
THE INFLUENCE OF CONTACT
PROPERTIES ON FRICTION-INDUCED
BRAKE VIBRATIONS
M. Rudolph and K. Popp

University of Hanover, Institute for Mechanics
Appelstr. 11, 30167 Hanover, Germany
Abstract Friction induced brake vibrations are investigated by theoretical and

experimental means. A multibody system is used to describe the dy-
namics of a floating caliper brake. In this model two different excitation
mechanisms for friction induced vibrations have been incorporated. The
mechanisms are based on two different properties of the frictional con-
tact between pad and disc. In the first step a constant coefficient of
friction (OOF) is included in the model, in the second step a decreasing
OOF with increasing relative velocity is assumed. Both approaches are
investigated by eigenvalue and sensitivity analysis. Results of this anal-
ysis are in good agreement with practical experience and experimental
data.
1. Introduction
In recent years comfort features of cars have become more impor-
tant to car customers. Therefore, brake suppliers have to avoid friction-
induced vibrations during the braking action. Vibrations showing fre-
quencies above 1 kHz and a narrow bandwidth are very perceptible by
the human ear and are therefore most annoying. These are commonly
called 'brake squeal'.
As there is no general concept of designing a silent brake from the very
beginning, expensive and time consuming tests have to be conducted to
investigate the noise generated by a particular brake design. Detected
noise is then fought by modifications of the existing design and repeated
tests. This procedure is supported by FE-calculations, which are only
valid for the brake under investigation.
125
126
Figure 1. 14-DOF multibody model of a floating caliper disc brake
This work is intended to foster the understanding of the phenomenon

of brake squeal in order to step closer to the main aim of developing a
silent brake from an early design stage.
As an example of a common car front wheel brake, a floating caliper
brake was chosen to be investigated in detail.
2. Brake model
For the brake a 14-DOF multibody model has been developed. This
approach cannot mirror each detail of the brake design but allows an easy
variation of design features to check their influence on squeal generation.
Furthermore, the results from a rather rough model are more general
than those from a detailed FE model for a specific brake design.
As Figure 1 shows, the model consists of 6 rigid bodies connected
by coupling elements. As rather high frequencies are involved, elastic
modes of certain components have to be considered. This is done at
the carrier by splitting it into two rigid bodies connected by a coupling
element. The stiffness and inertia properties of the disc in the model
are chosen so that the representation in the model has eigenfrequencies
which are close to frequencies of elastic modes of the disc, known from
FEM analysis. The coupling elements connect the bodies by moments
or forces which are dependent on relative displacements and velocities.
These dependencies can be linear or nonlinear; here, only linear charac-
teristics are considered. Frictional contacts are introduced between disc
and pads.
Parameter values have been determined by measurements conducted
at brake components, or by calculations using a validated FE-model of
the brake under investigation.
127
3. Brake noise test rig

For the investigation of a real brake system and for verifying simu-
lation results, a brake noise test rig has been built. This experimental
setup is able to conduct noise tests automatically, and can be adapted
for other investigations.
The knuckle and the brake, including the disc are mounted to a frame.
This frame is surrounded by a sound-absorbing housing, and the wheel
shaft is driven directly by a DC-motor. Brake pressure is generated using
an automatically actuated main cylinder from a car. Measured values
are brake line pressure, brake torque, sound pressure, disc temperature
and rotational speed. Cooling of the brake is done by airflow.
4. Investigations based on a constant coefficient

of friction during braking
4.1. Basic Considerations
Assuming a constant COF in the brake model leads to the excitation
mechanism of nonconservative restoring forces. This basic mechanism of
squeal generation can be found in many publications, starting with the
one of North (1972) until recent times as Allgaier et al. (1999) shows.
In spite of this, the basic effect is usually not clearly stated, therefore a
basic model is given here in Figure 2 a}.
Looking at the equations of motion for this system,
+[
--------
M
+ C3
--------
D
,
Cl
(J.Lh - S}C3 (s - J.L~)~:s + C2 ] [ : ]
,
- 0, (1)
...
c
we see that friction creates an unsymmetric stiffness matrix indicating

nonconservative forces. These forces can cause instability.
If the coefficient of friction is varied, 0 $ J.L $ 1, and the correspond-
ing eigenvalues A are plotted in the complex plane, the effect of mode
coupling can be observed, see Figure 2 b). Increasing the coefficient of
friction makes the frequencies of the complex modes grow closer until
the real parts move in opposite directions and instability occurs.
128
• • • ::• • iI. • • ~l
Im(>')
rtr~:
o -.~ ..... -:-,,"~ ....
: ,~IJ :
•••••• t····'lt •••••• t•••••••
o Re(>')
a) b)
Figure 2. 2-DOF model of nonconservative restoring forces (a) and related root
locus plot (b).
4.2. Analysis of brake model

As the multibody model comprises 6 bodies and 18 coupling elements,
the parameters have been divided into subsets of values. Each subset
represents a design property of the brake and in a sensitivity analysis its
values are changed simultaneously by the same amount. The subsets are
called parameter groups and are given names of practical meaning. In
these groups parameters were varied by 10% and the COF between pad
and disc was kept at a constant value of J-t = 0.4. The resulting changes
of the real parts 6Re(Ai) of eigenvalues due to parameter variation were
plotted as a bar diagram over eigenvalue index i = 1 ... 28.
As an example, Figure 3 presents results from modifications of the
parameter group 'piston' (coupling elements 11 and 12 in Figure 1).
These modifications consisted first in a symmetric increase of coupling
stiffness values, secondly in an asymmetric, and thirdly in a symmetric
displacement of connecting points. As can be seen, parameter changes
can have different effects on the eigenvalue pattern. The symmetric shift
of coupling elements, increasing the distance between them, leads to a
general stabilisation of the system with a very strong influence in the
upper frequency range of the model. This stabilizing effect has been
observed in other parameter groups as well. Thus, real brakes have
been modified in two ways so that in the first case the distance between
two connecting points has been increased and in the second case this
distance has been decreased. Both modifications have been compared
with respect to their squeal behaviour. In these two tests the first design
proved to be less noisy than the second, in the frequency range up to 5
129
....
:10 -------J-II-+-------tr-----+------+--------
-------- - --:--------l\------r--------:---------
~ -1~ n ----lt~~~-l---~~~=t===::
! : : : _ _::Jf::::::::t:::::::- :~--l~i--~--~-:::::
i i 0-+-+
: ::::: : ::::::F::::::F:::::-l~---=--t---:::::
~L-~~ __ ~ __ ~ __ ~ __ ~~
o 5 10 15 20 25 30
Eigenvalue index i
Figure 3. Change of eigenvalue real parts due to variations of the parameter group
'piston' (c=stiffness).
kHz, see Treyde (2001). Bearing in mind that the model has a rather
simple structure, and, therefore is valid mainly for the lower frequencies
of squeal, these results are a good confirmation for the approach chosen
in this work.
The main results from the investigations of the parameter influence can
be summarized as follows:
Parameters close to the friction contact play an important role,
increase in damping does not always stabilize the system,
increase of inertia parameters lowers the stability,
increase of the distance between two connecting points has a
stabilizing effect.
5. Investigations based on a falling friction

characteristic
As shown in the preceding section, the mechanism of nonconservative
restoring forces leads to results which are in good agreement with prac-
tical experience. On the other hand, it was not possible to destabilize
the chosen brake model by a parameter variation in a realistic range.
So, a further refinement of the model is needed. As mentioned above,
the vicinity of the friction contact is a very sensitive area and, thus, it
seems to be reasonable to model the friction properties more exactly.
To find out more about the friction between pad and disc, the COF
was recorded during stop brakings from 200 rpm to 0 rpm, see Popp
and Rudolph (2001). The results showed a nearly linear relationship
between the COF and the rotational speed of the disc, where the COF
130
increases with decreasing speed. This relationship depends on brake

pressure and on disc temperature, but the falling characteristic can be
realized in nearly any case, even if an accelerated motion is regarded and
temperature effects are compensated.
5.1. Basic Considerations

A falling friction characteristic is a mechanism to generate squeal. It
can be studied in detail using a 1-DOF oscillator sliding on a moving
belt and was first considered by Mills (1938,1939). For the following
derivation also see Popp and Rudolph (2001) and Magnus and Popp
(1997). A velocity dependent coefficient of friction can be given by
(2)
where the relative velocity Vrel = Vo - x is the difference between belt-

velocity Vo and the velocity of the mass x. Here, /-Lo stands for the co-
efficient of static friction and k for the slope of the friction characteristic.
Ignoring viscous damping, the equation of motion for the sliding fric-
tion oscillator reads
(3)
with the normal force Fn.

Equation (3) can be transformed using dimensionless time 'T = Jc/m t
to
x" + 2Dx' + x = Fn [sgn(vrel)/-LO - kvoJ, (4)

c
where (... )' = d~(''') and
D = - Fn k < o. (5)
2y'ciii
It is clear that, due to the friction characteristic, negative damping
occurs. This damping is dependent on the dynamic parameters m, c, as
well as on the slope k and on the normal force Fn.
5.2. Analysis of brake model

Applying a falling friction characteristic to the brake model yields a
negative damping which varies with brake pressure. For the slope an
131
16r-~~~~'--'-'~~1
14
12
.... --------t-------- --------1--+--+-1"'--+-
,....... I I I
.~.
I
I., 10
t,
C. 8
........
- 4 :::::::l::::::: _::::::::~::~:::::I::::::::
.:( 6 ,--------t-------- --------;---------1-------
]' ..... --------t--+--+- .--+--~--+-----j--------
4 -----rnn-tn-l"n-n:m--1--nn: : I.T.t
--------T------- --------T--------r-------
-m-r--- ni\2140HZ t-~--
2
2 +------.:--------+-----+-+-- -\--
o . .. .. 1 + + .. ..
-7 -6 -5 -4 -3 -2 -1 5 10 15 20 25 30
ReP'i) [10 3 s- 1] Brake pressure [bar]
a) b)
Figure 4. Destabilisation due to a falling friction characteristic (a) and related

experimental results (b).
average value of k = 0.0576 sm- 1 has been taken from the measure-
ments. A stability analysis with changing brake pressure from 0 bar to
25 bar was conducted and it was found that instability occurs at a brake
pressure of about 15 bar and a mode frequency of 2140 Hz.
Figure 4 a) shows the calculation results in a section of the complex
plane.
5.3. Experimental Verification

To get an impression of the squeal behaviour of the brake under in-
vestigation, a test sequence of 225 brakings has been programmed and
performed. This sequence comprised a pressure range from 2 bar to 30
bar and a temperature range from 30°C to 310°C, while brakings were
performed at a constant speed of 40 rpm and lasted for 12 sec.
For each braking, the noise was recorded and analyzed by FFT. Noise
of a sound pressure level higher than 60 dB was treated as a noise event.
Those events have been plotted versus frequency and brake pressure in
Figure 4 b).
We see that a squeal noise of 2350 Hz has been found at pressures
starting at about 12 bar, which is remarkably close to the predicted
squeal range starting at 15 bar with a noise frequency of 2140 Hz.
From this investigation it can be concluded that by using a falling
friction characteristic, even a rather simple brake model can display
important details of squeal noise occurrence.
132
6. Conclusions
The work described shows that contact properties play an important
role in modeling friction-induced vibrations. Even with the very simple
assumption of a constant coefficient of friction, system stability can be
affected by friction. In the related model, nonconservative restoring
forces depending on the structure of the brake lead to growing system
states. By a sensitivity analysis of the brake model based on this mecha-
nism, parameter influences on system stability have been found. They
are in good agreement with practical experience and could be confirmed
by experimental investigations of design modifications.
After observing the coefficient of friction during braking, a falling
friction characteristic has been realized. This property also is a possible
reason for friction-induced vibrations. The slope of the measured charac-
teristic included in the brake model gave a pressure dependent stability
of the model. A stability analysis showed the onset of instability very
close to observed results on the test rig concerning frequency and brake
pressure.
Summarizing, it can be said that improving the contact model be-
tween brake pad and disc leads to a significant increase of model pre-
cision; it is possible to describe the onset of squeal by a rather simple
dynamical brake model.
Acknowledgments
This study has been funded by the 'Bundesministerium fur BiIdung
und Forschung'. The authors are grateful for this support. Furthermore
the authors would like to thank Dr. T. Treyde, J. Korte and A. Stache
of TRW Automotive company for their fruitful discussions.
References
Allgaier, R., L. Gaul, W.Keiper and K Willner (1999) Mode lock-in and friction
modelling, In GAUL, L. and C. A. BREBBIA (editors): Computational Methods in
Contact Mechanics IV, 35-47, WIT Press, Southampton, Boston.
Magnus, K, Popp, K (1997) Schwingungen, B. G. Teubner, Stuttgart.
Mills, H. R. (1938,1939) Brake Squeak, Technical Report 9000 B, 9162 B, Inst. of
Automobile Engineers.
North, M. R. (1972) Disc Brake Squeal - A Theoretical Model, Technical Report,
MIRA. No. 1972/5.
Popp, K, Rudolph, M. (2001) Brake Squeal, In Popp, K. (editor): Detection, Utiliza-
tion and Avoidance of Nonlinear Dynamical Effects in Engineering Applications,
197-225, Shaker Verlag, Aachen.
Treyde, T. (2001) Bremsenquietschen, In Popp, K. (editor): Detection, Utilization
and Avoidance of Nonlinear Dynamical Effects in Engineering Applications, 227-
244, Shaker Verlag, Aachen.
ANALYSIS OF EIGENVALUE PROBLEMS
MODELLING FRICTION: SUFFICIENT
CONDITIONS OF NON-UNIQUENESS FOR
THE ELASTIC EQUILIBRIUM
Riad Hassani
Laboratoire de Geophysique Interne et Tectonophysique, Universite de Savoie
73376 Le Bourget du Lac, France.
Patrick Hild and loan lonescu

Laboratoire de Mathematiques, Universite de Savoie / CNRS UMR 5127,
73376 Le Bourget du Lac, France.
Abstract This study is concerned with the Coulomb frictional contact problem in
elastostatics. Introducing a convenient eigenvalue problem, it becomes
possible to establish sufficient conditions of non-uniqueness for the con-
tinuous model. It can be also proven that these sufficient conditions are
fulfilled under appropriate hypotheses.
1. Introduction
In structural mechanics, the Coulomb friction model is currently used
when studying the contact problems. The variational formulation of the
continuous problem in the case of Coulomb's simplified law in elasto-
statics was given by Duvaut and Lions (1972). In this case, the first
existence results obtained by Necas, Jarusek and Haslinger (1980) were
followed by Jarusek (1983), Kato (1987) and more recently by Eck and
Jarusek (1998). In these references, existence is proved if the friction
coefficient is lower than a critical value.
A regularizing procedure of the normal stresses, the so-called "nonlo-
cal Coulomb friction model" introduced by Duvaut (1980) and studied
by Demkovicz and Oden (1982), Cocu (1984), Kikuchi and Oden (1988)
permits to prove existence for all friction coefficients and uniqueness
when the friction coefficient is sufficiently small.
133
134
The "normal compliance model" introduced by Oden and Martins
(1985) (see also Martins and Oden (1987), Klarbring, Mikelic and Shillor
(1988, 1989)) leads to similar existence and uniqueness results.
The finite element problem, corresponding to the continuous static
Coulomb friction problem, admits always a solution according to the
works of Haslinger (1983, 1984) (see also Haslinger, Hlavacek and Necas
(1996)). Furthermore, the discrete solution is unique if the friction co-
efficient is small enough (the denomination "small" depends also on the
discretization parameter). Besides, some convergence results of the fi-
nite element solutions to a solution satisfying the continuous model are
established in Haslinger (1983).
Again in the finite dimensional context, several studies using truss
elements have led to examples of non-uniqueness in the static and quasi-
static context (Janovsky (1982), Klarbring (1990), Alart (1993) and Bal-
lard (1999)).
In this study, we propose sufficient conditions of non-uniqueness for
the continuous model by using an appropriate eigenvalue problem. If
the friction coefficient is an eigenvalue of this problem, then the couple
(geometry, material) is candidate to non-uniqueness. In such a case,
we give sufficient conditions for non-uniqueness to be satisfied by the
solution of the continuous model. Since there are to our knowledge no
available analytical solutions, the fulfillment of these sufficient conditions
can be illustrated only numerically by using finite elements.
2. The Coulomb friction problem

We consider an elastic body occupying a bounded domain 0 in ]R2.
The boundary r of 0 is Lipschitz and consists of three nonoverlapping
parts rD, rN and re. We suppose that the displacement field is known
on rD. On the boundary part rN, a density of forces denoted F E
(L2(rN))2 is applied. The third part is re, in frictional contact with
a rigid foundation (see Figure 1). The body 0 is acted on by a given
density of volume forces f E (L2(0))2. The notation n = (nl,n2) stands
for the unit outward normal vector on r and let us set the unit tangent
vector t = (-n2' nt}. We denote by J-L > 0 the friction coefficient on r e.
The Coulomb friction problem in elastostatics consists of finding the
displacement field u : 0 - t ]R2 satisfying (1)-(4):
u(u) = C c(u) and div u(u) +f = 0 in 0, (1)

u(u)n = F on rN, u =U on rD, (2)
where u stands for the stress tensor field in 0, c(u) = (Vu + VT u)/2
denotes the linearized strain tensor field, C is a fourth order symmetric
135
and elliptic tensor of linear elasticity and div represents the divergence
operator of tensor valued functions. On r c, we adopt the following
notation: U = unn + Utt and cr(u)n = crn(u)n + crt(u)t. The equations
modelling unilateral contact with Coulomb friction can be written:
Un ~ 0, (3)
Ut = 0 ===}
{ (4)
Ut =1= 0 ===}
Remark: Let us mention that the true Coulomb friction law involves
the tangential contact velocities and not the tangential displacements.
However, a physically meaningful problem analogous to the discussed
here can be obtained by a time discretization of the quasi-static frictional
contact evolution problem.
The variational formulation of problem (1)-(4) consists of finding u
such that
uE K, a(u, v - u) + j(u, v) - j(u, u) ~ L(v - u), \;Iv E K,

where
a(u, v) = In (CE(u)) : E(v) dO.,

L(v) = { J.v dO. + { F.v dr.
in irN
Figure 1. The elastic body n in frictional contact with the rigid foundation.
136
3. Sufficient conditions for non-uniqueness: a

spectral approach
The useful tool for obtaining non-uniqueness conditions is the fol-
lowing eigenvalue problem. It consists of finding A E C and 0 -I <I? E
(HI (0))2 such that
a(<I?) = C c(<I?) and div a(<I?) = 0 in 0, (5)

a(<I?)n = 0 on rN, <I? = 0 on rn, <I?n = 0 onre, (6)
at(<I?) = Aan(<I?) on re. (7)
Let us now consider an equilibrium position uo of Coulomb's frictional

contact problem (i.e., a solution of (1)-(4)) supposed to be "regular"
enough. The forthcoming proposition yields sufficient non-uniqueness
conditions concerning the solution uo. We suppose that uo is chosen
such that all the points of re are in a "slipping" contact, i.e.,
Proposition. Let uo be a smooth solution of Coulomb's frictional con-

tact problem (1)-(4) with Ii > 0 as friction coefficient. Assume that
there exist 0:, f3 > 0 satisfying
(8)
Moreover, suppose that one of the following two conditions (i) or (ii)
holds:
(i) Ii is an eigenvalue of (5)-(7) with the corresponding smooth eigen-
function <I? and
u~(x) ~ 0: for all x Ere. (9)
(ii) -Ii is an eigenvalue of (5)-(7) with the corresponding smooth eigen-

function <I? and
u~(x) ~ -0: for all x Ere. (10)
Then Coulomb's frictional contact problem (1)-(4) admits an infinity

of solutions. In particular, there exists to > 0 such that u l = uo + t<I? is
solution for any t satisfying It I ~ to.
Sketch of the proof. Let us notice that u l = uo + t<I? satisfies the
equations (1)-(2) for any t E lit The proof consists then to show that
the frictional contact conditions (3)-(4) are fulfilled for It I small enough.
137
The non-uniqueness conditions of the proposition require that the fric-
tion coefficient I-' (or -I-') is an eigenvalue of (5)-(7). In fact, the eigen-
value problem (5)-(7) depends only on the geometry of the problem and
on the material characteristics (more precisely the Poisson coefficient).
The variational formulation of problem (5)-(7) consists then of finding
A E C and 0 =f <P E VO such that:
where
and VO is the subspace of V O in which the functions v satisfy div O"{ v) E

{L2{O))2.
The existence of a countable set of eigenvalues and sufficient con-
ditions for the existence of real eigenvalues is proved in (Hassani, Hild,
Ionescu (2001)). In this existence study, we need to assume that the nor-
mal constraint O"n{u) is regularized as in Duvaut (1980) or as in Ionescu
and Sofonea (1993) to recover compactness properties.
If a real eigenvalue exists, then one has to consider the friction problem
having precisely this eigenvalue as friction coefficient. It suffices then to
exhibit a distribution of loads F, f and a displacement field U such that
a solution u O of (1)-(4) satisfies (8), (9) (or (8), (10)) to find a continuous
branch of solutions.
4. A numerical example to illustrate the theory

Let us consider as in (Hassani, Hild, Ionescu (2001)) a finite element
approach for the above-mentioned eigenvalue problem. The finite dimen-
sional problem we consider consists of finding Ah E C and 0 =f <Ph E V~
such that:
(11)
where V~ is a convenient finite element space. The convergence of the

discrete eigenvalue problem can be proved by using results of Kolata
(1978) (see also BabuSka and Osborn (1991)).
Unfortunately, no analytical solutions of (1)-(4) are available. That
is why we cannot directly verify the sufficient conditions given in the
Proposition. Next we try to illustrate them from a numerical point of
view. Since the convenient convergence results for the Coulomb friction
138
model do not exist, the numerical computations cannot stand for a rigor-
ous mathematical proof of the sufficient conditions for non-uniqueness.
Our aim here is only to illustrate the methodology given in the contin-
uous context.
As geometry, we choose an unit square n (in Figure 2) with Pois-
son ratio and Young modulus 1/ = 0.4 and E = 10 GPa respectively.
The body lies on a rigid inclined plane and it is loaded by a density of
gravity forces 1 and by an imposed displacement field U. The following
values are chosen: 0 = Arctan 2, <p = Arctan 2.5, 111 = 223.6 N.m- 3 ,
lUI = 0.005385. The discrete eigenvalue problem (11) is solved with an
10 20 30 40 &0
Figure 2. An elastic square lying on a rigid inclined plane, the deformed configura-
tion u~ and the Von-Mises stress field
uniform quadrilateral mesh of N x N elements for different N . Since real

eigenvalues are obtained, we represent in Figure 3 the quite satisfactory
convergence of the first real eigenvalue ).~. The notation h represents
the discretization parameter and h = 1/N.
Then we consider again the 100 x 100 mesh and the corresponding
eigenvalue equal to J.lo = 0.84232. A finite element solution, say u~,
is computed using the finite element code CASTEM with J.l = J.lo (see
Figure 2). It can be checked that all the points of r c are slipping, that
the normal stress and the tangential displacement on r c satisfy (8)-(9).
Denoting by <Ph an eigenfunction associated with J.lo, we can choose
a scalar t such that uh = u~ + t<Ph may stand for another solution of
(1)-(4).
Figure 4 depicts the Von-Mises stress fields corresponding to the maxi-
mal difference to<Ph = u~ -u~. In this example we can observe a maximal
139
0.87
0.86
0.85
:'.t
0.84
0.83
0.82
20 40 60 80 100 120 140 160 180 200
llh
Figure 3. The convergence of the first eigenvalue
value of (Uk - u~) /u~ of approximatively 20% concentrated on the lower

right corner of the body.
o 10
Figure 4. The Von-Mises stress corresponding to the difference u~ - u~ = to'h .
References
Alart, P. (1993) Criteres d'injectivite et de surjectivite pour certaines applications de
Rn dans lui meme : application ala mecanique du contact, Math. Model. Numer.
Anal. 27, 203 - 222.
Babuska, I., and Osborn, J . (1991) Eigenvalue problems, in Handbook of Numerical
Analysis, Volume II, Part 1, P.G. Ciarlet and J.L. Lions, eds., North Holland,
Amsterdam, 641 - 787.
Ballard, P. (1999) A counter-example to uniqueness in quasi-static elastic contact
problems with small friction, Int. J. Engng. Sci. 37, 163 - 178.
140
Cocu, M. (1984) Existence of solutions of Signorini problems with friction, Int. J.

Engng. Sci. 22, 567 - 575.
Demkovicz, L., and Oden, J. T. (1982) On some existence and uniqueness results
in contact problems with nonlocal friction, Nonlinear Analysis, T.M.A. 6, 1075 -
1093.
Duvaut, G. (1980) Equilibre d'un solide elastique avec contact unilateral et frottement
de Coulomb, C. R. Acad. Sci. Paris Ser. I Math. 290, 263 - 265.
Duvaut, G., and Lions, J.L. (1972) Les inequations en mecanique et en physique,
Dunod, Paris.
Eck, C., and Jarusek, J. (1998) Existence results for the static contact problem with
Coulomb friction, Math. Models Methods Appl. Sci. 8, 445 - 468.
Haslinger, J. (1983) Approximation of the Signorini problem with friction, obeying
the Coulomb law, Math. Meth. Appl. Sci. 5, 422 - 437.
Haslinger, J. (1984) Least square method for solving contact problems with friction
obeying Coulomb's law, Apl. Mat. 29, 212 - 224.
Haslinger, J., Hlavacek, I., and Netas, J. (1996) Numerical methods for unilateral
problems in solid mechanics, in Handbook of Numerical Analysis, Volume IV, Part
2, P.G. Ciarlet and J.L. Lions, eds., North Holland, Amsterdam, 313 - 485.
Hassani, R., Hild, P., and Ionescu, 1. (2001) On non-uniqueness of the elastic equilib-
rium with Coulomb friction: a spectral approach, Internal report 01-04c, Depart-
ment of Mathematics (LAMA), Universite de Savoie. Submitted.
Ionescu, 1., and Sofonea, M. (1993) Functional and numerical problems in viscoplas-
ticity, Oxford University Press, Oxford.
Janovsky, V. (1982) Catastrophic features of Coulomb friction model, in the math-
ematics of finite elements and applications IV, MAFELAP 1981, Proc. Conf.,
Uxbridge/Middlesex 1981, 259 - 264.
Jarusek, J. (1983) Contact problems with bounded friction. Coercive case, Czechoslo-
vak. Math. J. 33, 237 - 261.
Kato, Y. (1987) Signorini's problem with friction in linear elasticity, Japan J. Appl.
Math. 4, 237 - 268.
Kikuchi, N., and Oden, J. T. (1988) Contact problems in elasticity: a study of varia-
tional inequalities and finite element methods, SIAM, Philadelphia.
Klarbring, A. (1990) Examples of non-uniqueness and non-existence of solutions to
quasistatic contact problems with friction, Ing. Archiv 60, 529 - 541.
Klarbring, A., Mikelic, A., and Shillor, M. (1988) Frictional contact problems with
normal compliance, Int. J. Engng. Sci. 26, 811 - 832.
Klarbring, A., Mikelic, A., and Shillor, M. (1989) On friction problems with normal
compliance, Nonlinear Analysis T.M.A. 13, 935 - 955.
Kolata, W. (1978) Approximation in variationally posed eigenvalue problems, Numer.
Math. 29, 159 - 171.
Martins, J. A. C., and Oden, J. T. (1987) Existence and uniqueness results for dy-
namic contact problems with nonlinear normal and friction interface laws, Nonlin-
ear Analysis T.M.A. 11, 407 - 428.
Netas, J., Jarusek J., and Haslinger, J. (1980) On the solution of the variational
inequality to the Signorini problem with small friction, Boll. Unione Mat. Ital.
17-B(5), 796 - 811.
Oden, J. T., and Martins, J. A. C. (1985) Models and computational methods for
dynamic friction phenomena, Comput. Methods. Appl. Mech. Engrg. 52, 527 -
634.
ADHESION OF VISCOELASTIC SPHERICAL
SOLIDS
K.L. Johnson
Department ofEngineering, Cambridge University
Trumington Street, Cambridge CB2 1PZ, U.K.
J.A.Greenwood
Department ofEngineering, Cambridge University
Trumington Street, Cambridge CB2 1PZ, U.K.
Abstract The development of micro-probe instruments, such as the Atomic Force

Microscope, for the measurement of surface energy and other material
properties on the micro/nano scale has led to the need for continuum
mechanics models of the contact of a spherically tipped probe with the surface
of a specimen in the presence of adhesive forces. In the case where the tip and
specimen are perfectly elastic solids, such models are well developed through
the elastic fracture mechanics principle of equating the strain energy release
rate to the work of adhesion. The process of increasing and decreasing the
contact load is then fully reversible. When the specimen material is
viscoelastic, e.g. polymer or rubber, the situation is complicated by
viscoelastic dissipation. The apparent work of adhesion becomes rate
dependent and irreversible: more energy is required to separate adhering
surfaces than is returned when they come together, an effect known as
'adhesion hysteresis'. The paper will outline progress in extending the elastic
analysis to the viscoelastic situation by modelling the material as a linear
viscoelastic solid.
1. Introduction
The incentive for this research arose from the development of microprobe
instruments to measure the surface energy and other surface properties of
polymeric materials. Such probes commonly comprise an effectively
141
142
spherical tip pressed against a flat surface or, as in the case of the Surface
Force Apparatus, two crossed cylinders of equal radii, which are
geometrically and mechanically equivalent to a sphere in contact with a
plane. A typical experiment consists of loading and subsequently unloading
the contact with a controlled force, while measuring the displacement and/or
contact area, until the surfaces separate. If the surfaces adhere during the
compressive part of this cycle, a tensile force is required to pull them apart,
which is referred to as the 'pull-off' force. Microprobe instruments are
particularly advantageous for adhesion measurements for two reasons: (i)
adhesion forces become increasingly significant with a decrease in the size
of the contact; (ii) adhesion is strongly affected by surface roughness whose
effect is minimised by using a very small area of contact. This paper up-
dates an earlier 'progress report' [Johnson (2000)].
For perfectly elastic solids and reversible adhesive forces, the contact
mechanics is well developed through the elastic fracture mechanics principle
of matching the release of strain energy to the work done to separate unit
area of the surface (surface energy). See Johnson, Kendall and Roberts
(JKR) (1971), Derjaguin, Muller and Toporov (DMT) (1975), Maugis
(1992). But the adhesion of viscoelastic solids, notably rubber, is well
known not to be reversible. More work is required to separate two surfaces
than is returned when they come together, an effect known as 'adhesion
hysteresis'. This irreversibility arises from viscoelastic dissipation within the
contacting solids. The mechanism can easily appreciated in a qualitative
way. During separation the work done by the external force has to overcome
the dissipation in addition to doing work against the surface forces (surface
energy). When the surfaces are coming together the surface energy is
providing the driving force, which has to provide for the dissipation as well
as overcoming the external load. Hence the external force on separation
exceeds that on coming together by virtue of the internal viscoelastic
dissipation. In common with any non-conservative system, the response is
dependent upon the loading path.
The results of the JKR and Maugis elastic theories will be briefly
reviewed, since they form the basis of the viscoelastic analysis.
2. Elastic Solids: the JKR and Maugis-Dugdale Theories
Continuum mechanics models for the adhesion of perfectly elastic

spheres under the action of reversible surface forces are well developed. The
essential features of the JKR model are shown in Figure lao The surface
traction acting on a contact area of radius a comprises two terms: (i) a Hertz
pressure PI (a), caused by the compressive force PI, which flattens the
143
spherical surfaces and (ii) an adhesive tension Pa(a), which gives rise to the
adhesive force Pa. The net contact force P can be expressed by
and the compression <5 by
8 = (a 2 / R) - .J27r Ay a / E * , (2)
where E* is the combined elastic modulus of the two solids given by
_I=_ v12 + __
1- _ 1- v 22
E* E1 E
2 '
8 - - + -_ _ 8
(a)
(b) c c
·1
Figure 1. Adhesive contact of elastic spheres. Surface traction comprises two terms: Hertz
pressure PI and adhesive tension Pa. (a) JKR model; (b) Maugis-Dugdale model.
144
the relative radius of the two spheres lIR = lIRJ + lIR2 and the combined
surface energy of the two surfaces 11 r = rJ + r 2 - rl2· The surfaces snap
apart at a 'pull-off force Pc given by
(3)
We now introduce dimensionless variables:
so that equations (1), (2) and (3) become
(4)
(5)
and
Pc =-112. (6)
The infinite tension at the edge of the contact in the JKR model [Figure
l(a)] leads to difficulties in a viscoelastic analysis since, with a moving
contact boundary, it corresponds to an infinite strain rate. The singularity
has been removed in the contact problem by Maugis (1992), by analogy with
a Dugdale crack in an elastic-plastic material. As shown in Figure l(b), the
tensile adhesive traction is taken to have a maximum finite value 0"0 which is
assumed to be constant in a 'process zone' a ::s; r ::s; c in which the separation
of the surfaces h( r) ::s; ho' where ho is related to the surface energy by
I1r = (joh o· The width d of the process or Dugdale zone is given by
E* h 2
7r
d=c-a::;;.- Q (7)
4 Ar '
provided that d/a « 1, in which case the JKR equations (1) to (6) become
effectively asymptotic solutions to the Maugis analysis.
3. Viscoelastic contact cycle
We assume that an adhesion test is carried out under a controlled load

and that, for simplicity, both loading and unloading take place at a uniform
145
rate U =Po / to' where Po is the maximum compressive load. Such a ramp
cycle is shown in Figure 2.
To be mathematically tractable, analytical continuum mechanics
modelling of rate-<lependent materials is effectively confined to linear
viscoelasticity. We shall use the simplest model which encapsulates the
essential features of small strain polymer deformation: the 3-element solid. It
is represented in Figure 3 by two springs and a dashpot. At very low rates of
strain the material deforms elastically with a modulus EOC) represented by the
two springs in series. At high rates of strain the deformation is again elastic,
with a modulus Eo = EOC) Ik represented by the single spring. At intermediate
rates, the strain at time t after a step change in stress is given by the creep
compliance function (t) = (1 / EOC» fjJ(t) , where
Figure 2. Ramp cycle in which the load is increased and then reduced at a unifonn rate U.
Separation occurs at the pull-off force Pc.
-
1
r;;
.i
~
~
e-
o(,)
~
~
1 I--~~=-- ------- --------
To
o
10-2
Time tIT
Figure 3. Creep compliance function of a three-element model of a linear viscoelastic
material: equation (8).
146
¢(t) ={1- (1- k)exp( -t IT)}, (8)
and T = (1- k) 1] I Eoo is the relaxation time of the material. <D(t) is plotted
in Figure 3. It can perhaps be thought of as the reciprocal of a time
dependent modulus E(t).
If we refer to the surface traction distributions in Figures la or Ib, it is
apparent that there are two different locations of high elastic strain having
very different scales of size. There is the bulk deformation of the contact
arising principally from the Hertz contact pressure, whose size is
characterised by the radius a of the contact. There is also an intense
concentration of strain at the periphery of the contact produced by the
adhesive forces, whose size is characterised by the length d of the Dugdale
zone. During the cycle shown in Figure 2 the representative time for bulk
deformation is to but, for the strain concentration at the periphery, the
representative time is that for the edge of the contact to pass through the
Dugdale zone, i.e. t d = d I a, where the average value of a=ao I to and ao is
the contact size under Po' Hence the ratio of characteristic times governing
the strain rate in the two locations is
~=!!..
to a0
In the JKR regime d lao « 1, so that the characteristic times at which

viscoelastic relaxation will influence deformation in the bulk (to) and at the
edge of contact (td) are very different. Consequently we may treat the two
effects independently, in the same way that crack tip stresses and bulk
stresses are separated in elastic fracture mechanics. We will consider bulk
deformation first and ignore the presence of adhesion.
4. Viscoelastic Contact Mechanics
The contact of linear viscoelastic spheres in the absence of adhesion has

been studied in detail by Ting (1966) and Graham (1967). [For a summary
see Johnson (1985)]. If the two materials are different, their Poisson's ratios
must be assumed equal and interfacial friction neglected. We shall examine
the response to the ramp cycle ofloading and unloading shown in Figure 2.
The loading part of the cycle, i.e. during which the contact size a is
increasing is straightforward to analyse. The viscoelastic Correspondence
Principle applied to the Hertz elastic theory gives the variation of contact
size a( t) with load P(t) to be:
147
3R
a 3 (t) = - f
4 °
l d
<P(t - f)-pet )dt .
dt
(9)
Putting P = Uf and using the creep compliance function of equation (8) gives
Analysis of the unloading phase, during which a is decreasing, is more

difficult in view of the need to ensure that the traction on the contact surface
is everywhere compressive. The details of the procedure are given in the
references cited.
Computations have been performed for a 3-element solid with
k E: /E;
== =1/4. The variation in contact radius a( t) / ao through the cycle
is shown in Figure 4 for different values of cycle time 2fo, where ao is the
static value of a under the maximum load po. For a sufficiently slow cycle
a follows the Hertz 1/3 power law with E* = E: (shown dotted); for a fast
cycle a again follows Hertz with E* = E;. Note that, at intermediate speeds
the maximum contact size occurs after the load begins to decrease. This
asymmetry in the cycle reflects the energy dissipation which is a maximum
when the cycle time 2fo is of the same order as the relaxation time T
1.
C
IU
~O.B
fIl
::s
i3 0.6
f
o ~--~--~--~----~--~--~--~~--~--~--~
o 0.2 0.4 0.6 O.B 1 1.2 1.4 1.6 1.B 2
Time t/to = PIPo

Figure 4. Variation of contact radius with load during the ramp cycle of Figure 2
without adhesion
148
5. Viscoelastic fracture mechanics
As we have seen in the JKR elastic theory, the adhesion term can be
obtained by elastic fracture mechanics; in the present problem we tum to
viscoelastic fracture mechanics. The situation is illustrated in Figure 5,
which shows an interface crack with a Dugdale cohesion zone of width d.
The crack may be envisaged as being propagated at a steady speed V by a
frictionless pin in the crack mouth. First consider elastic solids. If the pin
moves to the left, since the stored elastic energy does not change, the work
done by the pin exactly balances the work done against the surface forces l1y.
So we can write:
F=K} 12E*= G =fly . (11)
For elastic solids this result is perfectly reversible and applies equally if the
pin moves to the right and the crack closes. A viscoelastic material, on the
other hand, involves energy dissipation D(V) as the strain field moves
through the solid. If we postulate that D(V) is a fraction a of the input work
G, for an opening crack we have:
or
(12)
In a closing crack the input work comes from the surface energy so that the
equivalent relationship becomes:
G
fJcl == ~
fly
= 1- a cl (V):::; 1 (13)
Equations (12) and (13) demonstrate the observed hysteresis effect in

performing a loading (crack closing) and unloading (crack opening) cycle
such as that shown in Figure 2.
5.1 The 'representative time' approach
The first mechanics analysis of an opening viscoelastic crack was by

Schapery (1975) followed by Greenwood & Johnson (1981) and for a
closing crack by Schapery (1989). In this approach an arbitrary, but finite,
149
distribution of surface force in a process zone of length 1* is assumed, which

is related to the 'representative time' t* by 1* = cVt*. Schapery suggested
c ::::: 3, while Greenwood and Johnson, on the basis of an inverse cube force -
separation law and a correspondingly different definition of 1*, chose
c = n12.
V(closing)
L(»d)
V(opening)
Figure 5. Representation of a steadily moving (opening and closing) interface crack with a
Maugis-Dugdale adhesive traction
Figure 6. Dugdale process zone for an opening crack; material points move from left to right
with velocity V.
150
With a Dugdale crack 1* equals the Dugdale length d which, in the
visoelastic case, is given by equation (7) with E* replaced by E: I ¢(t) , i.e.
d=!!.. E:h: _1_. (14)

4 /).y ~(t*)
Now d = cVt*, so that t* can be found from
I* 7r E* h2 7r 1
(t * / l).~(t * / T) =- ~(t*) =_....£..Q.... == - - , (15)
cVT 4c /).yVT 4c v
where c is the constant (1/3) or (2In). Schapery then showed that for an
opening crack the apparent work of adhesion is given approximately by:
(16)
The closing crack was examined by Schapery (1989). He showed that the
representative length and time were approximately the same as for the
opening crack (equation (14», but the apparent work of adhesion was given
by:
Pcl == ~; = ~(t*), (17)
which gives the neat result: Pop x Pcl ~ 1. Note that this corresponds to the
qualitative equations (12) and (13) if the dissipation coefficients in opening
and closing, a op and acl, are equal.
6. Exact viscoelastic analysis
The pioneering work described above, valuable that it is, has

approximations which call for verification by a more exact linear
viscoelastic analysis. This has recently been achieved for the three element
solid and a Dugdale process zone. Opening and closing cracks need to be
treated separately.
151
6.1. Opening crack
This outline follows Greenwood (2001, to be published) [see also Baney

and Hui (1999)]. A crack opening from right to left, with a Dugdale
distribution of interfacial stress, is shown in Figure 6. The opening shape of
the crack he(x) in an elastic solid is given by [Schapery (1975) or Greenwood
and Johnson (1981)]:
he(x) = 20"q fd{2

TfE * Jo
Ix -In ~x +~~AIx'
17 - x' U'
~l.~
E*
F(x /d)
Tf
(18)
For a 3-element viscoelastic solid with the creep function of equation (8), the
impulse response function rf>i(t) is given by:
'f',
d {l-k}
"'. (t) == -dt ~(t) =
-T exp( -t / 1) . (19)
If the crack opens with a steady speed V, the point now at a distance x ahead
of the current crack tip would have been a distance x, = x - Vt from the tip a
h(x)=~
20" d
TfE*
f
d/
0'
v
~(t)
,x-
time t ago. It follows that the relaxed (viscoelastic) shape is given by:
-
d
Vt) 20" d
dt+~~(O)F(x/d),
TfE*
(20)
where the second term is the instantaneous elastic response, for which
(J(O) = Eoo* / Eo* == k.
Now the crack opening displacement at the end of the Dugdale zone is
related to the surface energy by: h( d) = ho = dr /0"0' Inserting this
condition, together with the impulse response (19), into equation (20) and
changing the integration variable to u = 1 - Vtld gives an implicit equation
for the true length d of the cohesive zone: i.e.
VTdr ~l
v== E*h 2 =[r~(r,k)] , (21)
co 0
where
11 (r,k) [r
= ~ J~ (1- k) exp[- r(l-u)]F(u)du + kF(1) ]
and r = dlVT. This integral has been evaluated numerically and equation (21)
inverted to obtain d = VTr( v, k).
152
Returning to the contact stress distribution shown in Figure 6, it was first
shown by Barenblatt (1962) that, to maintain finite stresses at the tip of a
crack, the remote stress intensity KJ must be balanced by the stress o(x) in
the process zone according to:
K = f2 roo CJ(x) dx
J 1; Jo..rx '
which, for an opening Dugdale crack, gives
(22)
Since the stresses remote from the tip are relaxed, G is related to KJ by
G = K} / 2E:, whereupon
(23)
6.2. Closing crack
Now consider the crack in Figure 6 to be closing, i.e. moving from left to
right. Superposing instantaneous stress distributions to find the relaxed crack
shapes does not work in this case, since the need for contact of the surfaces
behind the crack front would be violated. Instead it is necessary to superpose
instantaneous crack shapes and to calculate the resulting relaxed stresses.
The required shape need not be known ab initio, but will be found in course
of the analysis. Following Schapery (1989) we shall refer to the stress
distribution which would produce this shape in an elastic solid of modulus
E: as the 'reference stress' and denote it by (TR(X]) = (T "g(x] / d), where d
remains to be found.
A point now at x would have been at x, = x + Vt a time t ago, so that the
relaxed stress at x is given by
(24)
153
where If/i(t) is the impulse relaxation function and 1fJ(0) = 11k. Making use of
the reciprocal nature of the relaxation and compliance functions, equation
(24) can be inverted to obtain the reference stress:
(:l(x/d)-=()og(x/d) = Jo
r(d-X)IV
¢i(t)()
(xd+ Vt) dt +¢(0)CJ(x/d).(25)
For a Dugdale crack relevant values of O"{ (x+ Vt)/d} are equal to 0"0 which,
when substituted into equation (25) and integrated, gives
g(xl d) = ¢{(d - x)IV}. (26)
The crack shape is found from the reference stresses by
~
hex) = 2(} fdg(X' /d){2
Jr E: 0
Ix -In ..rx -HU
1t..rx +H11dx' . (27)
At the end of the Dugdale zone h(x) = h(d)= ho = LirlO"o which, when
substituted in equation (27) gives an equation for the Dugdale length d.
Values of r -= d / VT as a function of v and k have been found numerically
by Greenwood (2001). It may be shown that the remote 'effective' stress
intensity factor is determined by the reference stress if rather than by the
true stress at the crack tip. Thus:
(28)
Hence we can write
(29)
Computations have been carried out for a three-element solid with

k -= E: / E: = 0.01, for both opening and closing cracks. The lengths d of
the Dugdale zone as a function of nonnalised speed v, plotted in Figure 7,
show large and similar increases in length with speed for both opening and
closing cracks [equations (21) and (28)]. The variation of effective work of
adhesion with speed, Pop [equation (23)] and Pel [equation (29)], are plotted
in Figure 8. As expected Pop increases with speed of opening and Pel
154
decreases with speed of closing. Plots of the Schapery equations (16) and
(17) are added to Figure 8 with the coefficients cop = 4.5 and Cel = 6 taken to
give a best fit with the exact results of equations (23) and (29). For the case
of a closing crack, computed values of the reference stress cI(x) compared
with the true stress a(x) in the vicinity of the crack tip for r = 10 and k =
0.01 are shown in Figure 9, from which it is seen that the reference stress
approaches the true stress away from the tip. Similar results have been found
by Baney and Hui (1999).
10 2 r----r----.----.----~--~----~--~----=
'tI
:S
....fi
CD
10 1
~
eq
=
CD
,
U
0
, ,
t ,
, ,
, ,
,,
10 1 10 2 10 3 10 4 10 5
Figure 7. Dugdale zone length as a function of non-dimensional crack speed.

k = 0.01; --opening [eq. (21)]; -.-.-.-.- closing [eq. (28)].
7. Viscoelastic adhesion cycle
We shall now attempt to model an adhesion test by considering the ramp

cycle in Figure 2, with a uniform loading and unloading rate
dP / dt = ±{Po / tJ. To simplify the calculation, we shall assume that this
elastic with a modulus E:.

rate of loading is sufficiently slow for the bulk deformation to be perfectly
Conditions for which the JKR theory is
appropriate (d « a) are assumed. In the purely elastic case the equilibrium
155
relationship between load and contact size is given by equation (4). In the
viscoelastic case the surface energy Liy is scaled by the factor Pel during
loading and by POP during unloading, for which Schapery's approximate
equations (16) and (17) will be used. In dimensionless variables equation (3)
becomes
- -3 r::=3
p= a -12/Ja (30)
or
Inverting the creep compliance function for a 3-element solid gives
* IT = In~ .
-k
t (31)
1- ¢J(t * I T)
-g.
'-'
ctl
tlo
.E!
10 1
~
g.
~
ci- 10 0
II
IQ.
A1VT u~ VT
Ivl = --.--z = - . -
EC1J ho Eo A1
Figure 8. Ratio Pof apparent / true surface energy for a moving crack.
k =0.01; --Exact, eq. (23) & (29); -.-.-.-.- Schapery, eq. (16) and (17)
with empirical shift factors cop = 4.5 and Cel = 6.0.
156
Substituting in equation (15) gives
.!!....l =.!!.... EJz o

2
= (t * / T)ln 1- k (32)
4c v 4c t1r VT ¢ 1- ¢(t * / T)'
in which, from equations (16) and (17),
¢(t* / T) = Pel = 11 Pop·

Now
d ~ = dP /(~ cia] ,
da dt a dt
where daldt is the sQ..eed of the edge of the contact, which corresponds to the
crack speed Vand d P / dt = p" / to = U, whereupon
(33)
1 "",'
/". \
I · . . .
. .t . .... :.........\ ....... :........ :...... .
I~~ ... ~........ ~.'~.~>,.~:
0.8
0.6
i :
.......~ .... _.. .
(1 : ..................... :
bO i· . .,....-.,;"..
"-
b 0.4 i ...... . :................ :........ : ..... -
;
gj ; .
Q)
.....r.. 0.2 i. . . . . . ... . . . . . . . . . . . . . . ... . . . . . . ... . . . . . . :-

I"Il
I-
I ..
0
-0.2 '-' .......................................... -
; I ;
-0.4
-1 -0.5 o 0.5 1 1.5
x
Figure 9. Closing crack, k = 0.01, VTld = 10. Reference stress d(x) [eq. (25)] compared with
true Maugis-Dugdale stress o(x).
157
3
lea
2.5
...
~
fIl
"(j
ea
....""CJ 2
....ea . . .
. fl-1- . .
§ 1.5 '" "f.~ ..: .... : .... : ..
CJ
"(j
Q)
fIl
. . .. . . . . . '.. . . . . . . . . . ... . .
~ 1 '.' '.' ~
e
~ ... ... ... ,' .... : .... ..
0.5 ,', ,', ',
o
-6 -4 -2 o 2 4 6 8 10 12
Normalised load P
(a)
6
'()
5 .. ·.
.... , .. · .
~ 4 .. .'.
~
CJ 3 ... .. .. ·.
lIS
...i
"(j
2 ..,. ..
"(j 1 ... :- ..... -: ...... :, ..... :....... :- ... .

Q)
fIl
~
0
e
ea
-1
~
-2 ..... ','. ..... '". ..... ','
. .... of·····
. '". ....
-3 ...... :- ..... -: ...... :, .... ': ...... :- ... .

-4L-~J----L--~ ____ ~ __ ~ __ ~ ____ ~ __ ~ __ ~
-6 -4 -2 o 2 4 6 8 10 12
Normalised load P
(b)
Figure 10. Simulation of ramp loading I unloading cycle. k =0.01; A =JO, 500.
(a) Variation in contact radius a; (b) variation in displacement (5.
158
The simultaneous equations (30), (31) and (33), can be integrated step-
by-step to give the variation of P with a and hence the variation of contact
size with load. During loading (crack closure) P= Pel given by equation (16);
during unloading (crack opening) P = Pop given by equation (17). In an
experiment, it is commonly the displacement 0 rather than the contact size
which is measured. It is given by equation (5) with the surface energy scaled
by the factor Pwhich, in dimensionless variables becomes
(34)
Computations have been carried out using the creep compliance function
in equation (8), with k = 0.01 and C = Cel = cop = 1/5, for various values of the
loading rate parameter A, where
(35)
The variation of contact size with load is plotted in Figure lOa. The elastic
(JKR) reversible relationship is shown by the chain line. At finite loading
rates adhesion hysteresis is clearly visible. When the surfaces first touch they
snap into contact at an initial size given by the elastic theory with E* = E; .
With increasing compressive load the contact grows; its size is less than the
JKR value since Pel < 1. When the load is first reduced, the contact size
continues to grow slightly until the surfaces begin to separate at the higher
apparent surface energy associated with an opening crack (Pop> 1). The
contact size then decreases until pull-off occurs at an enhanced pull-off
force. The contact size approaches its high speed values at the beginning and
end of the cycle. This is on account of the cube-root relationship between a
and P which, for dP/dt = constant, gives high values of dP/da when a is
small.
Having determined the variation of contact size a with load, the variation
of displacement ois given by equation (34), see Figure lOb.
8. Conclusion
Progress in modelling the contact mechanics and adhesion of a

spherically tipped probe and a viscoelastic solid has been reviewed. For
perfectly elastic solids and conservative surface forces, the process is
perfectly reversible and rate independent. This behaviour is modelled by the
elastic theories of Johnson et al (1972) and others. But for viscoelastic solids
159
more work is required to separate the surfaces than is extracted when they
come together and the process is rate dependent.
In the absence of adhesion, the contact mechanics of linear viscoelastic
solids is well established through the work of Ting (1966) and others. As in
the elastic case, the presence of adhesive forces at the interface introduces a
fracture mechanics problem at the edge of the contact area, where a Iprocess
zonel of length d can be identified. Provided d « a, where a is the contact
radius, the strain rates in the process zone are much greater than in the bulk.
This enables viscoelastic effects on the adhesive force to be separated from
those in bulk, as in elastic fracture mechanics. This condition is likely to be
satisfied with probes giving a contact size measured in ~, but may not be
with an Atomic Force Microscope in which a may be of order nm. The
analysis in this paper is simplified further by assuming that the bulk
deformation is carried out sufficiently slowly to show an approximately
elastic response.
Linear viscoelastic fracture mechanics developed from the work of
Schapery (1975) and (1989) who introduced the concept of a 'representative'
time t* and 'crack length' 1* (=Vt*), comparable in magnitude with the
process zone length d. He then showed how t* and the work of adhesion
G(V) for both opening and closing cracks could be found approximately
from the creep compliance function ¢i..t) of the material [equations (14), (16)
and (17)].
To investigate the accuracy of Schapery's results, Greenwood has
conducted a rigorous viscoelastic analysis of opening and closing 'Dugdale'
cracks, i.e. where the adhesive stress distribution has a constant value aD
over the process zone d, in which the separation of the surfaces is less than
h o • Good working agreement is found with Schapery's equations if 1* is
taken to have the value 4.5d when opening and 6d when closing, or even 5d
for both.
Finally, Schapery's equations have been used to simulate an adhesion test
in which a spherically tipped probe is brought into contact with a
viscoelastic solid, loaded at a slow uniform rate to a maximum compression,
and then unloaded at the same rate until the surfaces separate.
References
Baney, J.M. and Hui, C-Y (1999) J.AppI.Phys. 86,4232-4241.

Derjaguin, BY, Muller, V.M. and Toporov, Y.P. (l975)J.Coli. Inteiface Sci. 53,314.
Graham, GAC. (1967) IntJ.Eng.Sci. 5,495-514.
Greenwood, J.A and Johnson, KL. (1981) Phil. Mag. A43, 697-71l.
Johnson, KL., Kendall, K and Roberts, AD. (1971) Proc.Roy.Soc .Lond. A324, 301-313.
Johnson, KL. (1985) Contact Mechanics, C.U.P.
160
Johnson, K.L. (2000) in Microstructure and microtribology ofpolymer surfaces, Am. Chern.
Soc. Symposium Ser.741, 24-41.
Maugis, D.l (1992)J.Coll.Interface Sci. 150,243-269.
Schapery, R.A. (1975) Int.J.Fracture, 11,141-159,369-388.
Schapery, RA (1989) Int.J.Fracture, 39, 163-189.
Ting, I.C.I. (1966)ASMEJ.Appl.Mech. 33, 845-854.
A MODEL OF ADHESION
ADDED TO CONTACT WITH FRICTION
Christian Talon and Alain Curnier

Department of Mechanical Engineering
Swiss Federal Institute of Technology, Lausanne
Abstract A law of normal and tangential adhesion-decohesion is added to the

classical law of unilateral contact with threshold friction. That law
is formulated within the framework of standard generalised materials.
The model originality stems from the use of an irreversible adhesion gap
ga (similar to a plastic strain) as damage internal variable and a rate
independent decohesion law. The three laws of contact, friction and
adhesion are derived from their respective (quasi-) potentials and then
penalised before being implemented in a finite element code. Since the
law is developed to study debonding and damage in fibrous composites,
the standard pull-out test is used to demonstrate its validity.
1. Introduction
A simple and realistic law of normal and tangential adhesion (in-
cluding decohesion) is added to the classical law of unilateral contact
with threshold friction. The motivation for this extension is the study
of debonding and damage in fibrous composites by means of computa-
tional and experimental methods. The adopted framework is the con-
stitutive theory of standard generalised materials of Moreau (1973) and
Halphen and Nguyen (1975). More specifically, the model is inspired by
the pioneering work of Fremond (1987) and subsequent improvements
of Raous, Cangemi and Cocu (1997) on surface adhesion, as well as the
Lemaitre and Chaboche (1985) precedent on the analogous phenomenon
of damage in the bulk.
The proposed model differs from existing ones by the use of an ir-
reversible adhesion gap !f (similar to a plastic strain) as the damage
internal variable instead of the usual dimensionless fraction {3 of intact
adhesive links. By energetic duality, there corresponds a dual stress pa
to ga representing the adhesion intensity, which simultaneously derives
from an adhesion interface energy density q, and a dissipation power
161
162
density cP (adhesion and decohesion are mounted in series just as in

elastoplasticity). Regarding debonding, the elastic modulus of the bonds
is constant but these bonds undergo a permanent elongation ga (like a
locking law with increasing clearance). The maximum adhesive tension
PM decreases linearly with the adhesion gap over a finite distance gM.
A rate independent law is adopted for the evolution of decohesion. This
adhesion law is then mounted in parallel with the contact and friction
laws which are also derived from their respective (quasi-) potentials. The
laws are then penalised before being numerically implemented in the fi-
nite element code TACT (Curnier, 1994). The adhesive parameters PM
and gM are calibrated for epoxy-glass interfaces via novel experiments.
The proposed model is believed to be simpler and/or more realistic
(rate independent decohesion law, finite decohesion gap) than the ex-
isting models due to Fremond (adhesive tension increasing with gap) or
Cangemi (residual viscous decohesion). An augmented Lagrangian mul-
tiplier formulation of the law avoiding the artifacts of the penalty regu-
larisation is available but not reported here by lack of space. Several test
cases are used to illustrate and validate the model. The standard fibre
pull-out test is run for comparing this "plastically" controlled damage
model with existing ones and with experimental results.
2. Contact mechanics
2.1. Kinematic and static variables
Given two bodies 0 C R3 and 0' C R 3 , the gap vector g between a
striker and a target particles in regards, within the hypothesis of small
slips and strains, is simply defined together with its rate by:
g =y-y'
. .
g=y-y ,
., (1)
where y E Ot and y' E O~ are the current positions of the particles.

Dually, let p be the stress vector applied by the target on the striker
as it occurs in the action-reaction principle:
p = -p' . (2)
Classically, both the gap and stress vectors can be decomposed into
a normal and tangential components, defined using the unit outward
normal to the target surface n, as:
p=Pnn+pt, (3)
where gn is the normal gap or penetration, gt the tangential slip, Pn the

normal pressure or tension, and Pt the shear.
163
2.2. Tribology
A basic tribologicallaw, such as unilateral contact or dry friction, is a
relationship p between the kinematic variables g, g and the static stress
vector p, which typically is set-valued:
P E p[g,gl . (4)
Addition of secondary irreversible processes such as running-in, wear or
decohesion requires the introduction of additional internal state vari-
ables. Let gi be such a generic kinematic variable and pi its dual flux
(in the sense that pi . gi is a power). The tribological state law p and
the complementary law pi governing the evolution of gi take the form:
(5)
Adopting the framework of standard generalised materials, tribological
laws can be equivalently defined by a free energy potential w[g, gil, and a
dissipation power function <1>[g, gi; g, gil. For a broad class of materials,
the state and complementary laws are derived from these two potentials:
pi E -OgiW[g, gil
{ . . . . (6)
pl E Ogi<1>[g, gl; g, gl]
3. Contact, friction and adhesion laws

3.1. Unilateral contact - threshold friction laws
It is well known that normal unilateral contact is governed by a kine-
matic impenetrability condition, a static repulsion condition and an en-
ergetic complementarity condition:
gn ~ 0 , p~ ~ 0 , gnP~ = 0 . (7)
As unilateral contact is a reversible process, the contact pressure derives
from a free energy (the indicator of R+), and the dissipation is nil:
<1>C = 0 . (8)
It is also known that tangential Coulomb's friction law is governed by a
kinematical slip rule, a static friction criterion and a power complemen-
tarity condition:
Since friction is a dissipative process, the shear stress derives from a

friction dissipation (the scaled slip rate norm), and its energy is nil:
WS = 0 , (10)
164
For preparing its coupling with adhesion, friction can be equivalently
written as a rigid-friction assembly by introducing an artificial slip in-
ternal variable gZ = gt (g: = gt) in the previous potentials:
q,S(gt, gZ) = I{o}(gt -gZ), q,S(g:;p~) = I~(p~)(gZ) = J.lp~lIg:11 . (11)
3.2. Adhesion law

Modelling adhesion requires an additional kinematic state variable (of
gi type). Supposing the interface is composed of a multitude of adhesive
links, the new scalar variable can be of two kinds: either an adimensional
fraction of intact links (3, or an irreversible adhesion gap ff' which can be
interpreted as the irreversible cumulated stretching of each adhesive link
(similar to a plastic strain). Of course, a dual variable R(3 or pa must
also be introduced, representing respectively the energy of adhesion or
the adhesive stress.
Choosing the second approach, adhesion is governed by an irreversibil-
ity condition, a static adhesive criterion and a complementarity condi-
tion (figure 1):
9·a >
- 0 , (12)
and is coupled to the state variables and functions by an adhesive rule,
a clearance inequality and a complementarity condition:
In (12), s is a decreasing function describing debonding and in (13)

g = PR+(gn)n + gt the projection of the gap on the positive half space.
This model of adhesion can be derived from the following potentials:
q,a(g,ga) = IR-(llgll- ff') ,
(14)
PM
Figure 1. Law of adhesion and graph of function s(gG).

165
4. Coupled frictional contact with adhesion law

4.1. Potential formulation and force derivation
Adding the free energies and dissipation powers of unilateral contact
(8), tangential friction (11) and adhesion (14), we obtain the total free
energy and dissipation power for the three phenomena:
q,(gn,gt,ga,gt) = q,C(gn) + q,S(gt,gt) + q,a(g,ga)
= IR+(gn) + I{o}(gt - gt) + IR-(llgll- ga)
(15)
(ga, gti ga,p~) = + a(ga i ga)
S(gtiP~)
= JtP~llgtll + s(ga)ga + IR+(ga)

The state and complementary tribological laws which derive from the
potentials are:
Pn = p~ + p~ E + Ogn wa(g,ga) = Ogn w(gn, gt,ga, gt)
Ogn wC(gn)
{
Pt = p{ + pi E Ogt WS(gt, gt) + Ogt wa(g, ga) = Ogt w(gn' gt, ga, gt)
(16)
{ P:Pt E -Og:~:g~'sgt,:a'cgt)
E Og;(g ,gti9 ,Pn)
(17)
Therefore contact and friction are mounted in series as usual, whereas
adhesion is added in parallel.
4.2. Penalised formulation

Penalisation consists in replacing the exact non-differentiable free en-
ergy terms by an approximative differentiable function. The dissipation
remaining the same, the penalised free energy becomes:
~(gn,gt,ga,gt) = ~C(gn) + ~S(gt,gt) + ~a(g,ga)

(18)
= ~,~Dit+(gn) + ~'t(gt - gt)2 + ~,aDit_(llgll- ga) ,
where DR+ denotes the distance to R+. From the differentiable ana-
logues of (16) and (18), we get the normal contact and tangent friction
forces:
p~ = ,~PR- (gn) , p{ = ,%(gt - gn, pa = ,a P R+ (1Igll- ga)d , (19)
with pa = p~ n + pi and the unit vector d = g/llgll. The corresponding

graphs are displayed in figure 2. Using (15) and (18), the complementary
166
systems (17) take the following form:

S _ S( S) _ f pa = laPR+(llgll- ga) = Ilpall
{ Pt - It gt - gt - Pt {
pi E I-lp~allgi II pa E s(ga) + OIR + (ga)
(20)
The system (20) can be solved by resorting to an incremental formula-
tion. This consists in expanding the internal variables as gi = gio +gidt
and ga = go + gadt. Combining the solution of (20) with the forces ex-
pression (19), we get:
p{ = pC(p;,)bt(gt - gio)) ,
(21)
pa = pad = p[_oo,s(ga)]baPR+(llgll - go))d
p{
slip
adherence It f
gr
I
gi'
- - ~I I-lP~
PM~
Figure 2. Penalised unilateral contact, tangential friction and adhesion graphs.
Finally, the expression of the contact force results from summing the
contributions of each component:
p = p~ n + p{ + pa = I~PR - (gn) n
(22)
167
5. Benchmark problems
After deriving the appropriate Jacobian matrices for the generalised
Newton method of resolution (Curnier and Alart, 1988), the unilateral
contact with friction and adhesion law is implemented in the finite-
element code TACT. Numerical simulations of simple test problems of
unilateral contact and tangential friction coupled with adhesion are run.
The results displayed in figures 3 and 4 illustrate and validate the law.
30
Pn Pn
4
20
10
0.2
-10
-0.2 -0.1 0.4
-20
-2
-30
-40
-4
-50
Figure 3. Numerical simulations of penalised contact and penalised contact with adhesion.
60
Pt 60
Pt
40 40
20 20
.3 o. 0.1 0.1 0.2 .3 .3
-20 gt -20 gt
-40 -40
-60 -60
Pt
60
t~
40
20
.3 0.2 0.1 0.1 0.2 .3
-20
-40
-60
Figure 4. Numerical simulations of penalised friction, penalised adhesion and coupling.

168
6. Numerical simulation of a pull-out test
The standard pull-out test shown in figure 5 is used for illustrating the
overall model behavior. The model is able to trace the evolution of the
adhesion-gap-friction zones, as the load is increased; unlike the reference
analytical shear-lag model (Kim and Mai, 1996) which presumes a frozen
repartition.
z
r fibre
matr ix
(axisymmetric)
Ti 300
350
250
I weak load
) no debonding
~
200
150
100
a perfect adhesion
50 friction
~z
10 15 20 25 30
I
~
Ti 300
350 medium load
250
200 b) partial debonding
150
100 adhesion-decohesion
50 friction
10
~
15 20 25
----z
30
~ L_________
--_ ___._
strong load
c) complete debonding
Ti friction
5 10 15 20 25 30 Z
Figure 5. Pull-out of a fibre from a matrix. Analytical (plain lines) and FE simulation
(dots) of interfacial shear stress Ti = Pt versus z.
References
Curnier, A. (1994). Computational Methods in Solid Mechanics. Kluwer.
Curnier, A. and Alart, P. (1988). A generalised newton method for contact problems
with friction . J. Theor. Appl. Mech., 7(1):67- 82.
Fremond, M. (1987). Adhesion of solids. J. Theo. Appl. Meeh., 6(3) :383- 407.
Halphen, B. and Son, N. Q. (1975) . Sur les materiaux standards generalises. J. Meca,
14:39- 63.
Kim, J.-K. and Mai, Y-W. (1996). Modelling of stress transfer across the fibre-matrix
interface. In Numer. anal. and model. of compo mat., pages 287-326. Chap. & Hall.
Lemaitre, J. and Chaboche, J. (1985). Mecanique des materiaux solides. Dunod.
Moreau, J. (1973). On unilateral constraints, friction and plasticity. In Capriz and
Stampacchia, editors, New var. teeh. in math. phys., II-73, pages 175-322. Crem.
Raous, M., Cangemi, L., and Cocu, M. (1997). Un modele couplant adherence et
frottement pour Ie contact entre deux solides deformables. Cptes Rend. A cad. Sc.
Paris, 325:503- 509. Serie IIb.
ANALYSIS OF A 'WALKING' PUNCH
D. A. Hills
Department of Engineering Science, University of Oxford, Parks Road,
Oxford, OX1 3PJ.
A. Sackfield
Department of Mathematics, Nottingham Trent University,
Burton Street, Nottingham, NG1 4BU.
Abstract A rigid block rests on an imcompressible half plane, and is subject to

a constant shearing force, together with a normal force, constant in
magnitude but moving backwards and forward over the surface of the
block. The condition for the block to remain stationary, and for it to
start to creep along the surface, due to a moving regime of microslip,
are found.
1. Introduction
The problem to be studied is shown schematically in Figure 1. A
rigid punch rests on an incompressible half plane. A force, P, is applied
to the centre of the punch, and a cord, attached to the base of the punch,
is used to develop a constant shearing force, Q (Q < f P, where f is the
coefficient of friction). We assume that the position of the normal force,
s, then varies harmonically with time, so as to cause the punch to rock.
If the punch is of half-width, c, we will assume that the maximum extent
of the load offset, So, is less than c/2, so that complete contact is made
throughout. The object is to track out the stick/slip regime, shearing
traction distribution and slip displacement as a function of time. The
conditions under which 'walking' proceeds will then be found, and an
estimate of the size of the 'steps' found.
169
f.A. C. Manins and M.D.P. Monteiro Marques (eds.). Contact Mechanics. 169-178.
170
s p
-c x
Figure 1. Schematic showing the geometry of the problem
2. Formulation
2.1 Starting Conditions
First, push the punch normally onto the surface with s = 0, so that
the punch face is parallel with the half-plane, and the contact pressure
is given by
p
p( x) = ---;=~~ (1)
7rv'c2 - x2
Now apply a shearing force, Q. Providing there is complete adhesion,
this will lead to a shearing traction, Hills et. al. (1993)
Q
7rv'C2 - X 2
q(x) = (2)
Hence, at every point, providing I Q I / P = A < f
q(x) = A < f - c:::; x:::; c. (3)

p(x)
Now increase s to some finite value so that the punch tilts.
2.2 Pressure Distribution

In order to determine the pressure distribution we first note that the
slope of the punch profile will be given by
avo
ax = -q:>, (4)
where VO(x) is the normal displacement,and the connection between the

surface profile of the half-plane and the contact pressure distribution is
171
given by
.!. avo = _!
A ax 7r
l c
-c t -
pet) dt
x
(5)
where
A=1-v. (6)
J-L
Here J-L is the modulus of rigidity, and v is Poisson's ratio. In this

problem we are interested in a solution for moderate amounts of tilt,
so that full contact is maintained, and hence the pressure distribution
will be singular at both ends. The appropriate fundamental solution is
chosen which, with the Riemann-Hilbert inversion procedure, Muskhel-
ishvili (1953), gives the following
p(x) = v'c21_ x2 (: + ! x) . (7)
We now replace the central force and applied moment by a statically

equivalent offset force, at point s, Sackfield et. al. (in press) and hence
p(x;s)=.;)2P 2
2SX]
[ 1 +-
2 . (8)
7r C -x C
The pressure distribution is plotted out in Figure 2.
2.3 Partial Slip Problem

As the punch is rigid whilst the half-plane is incompressible, it follows
that Dundurs' second constant for the problem vanishes, so that the
problems for normal and shear loading are uncoupled. If we continue to
assume that the punch remains adhered, therefore, the shearing traction
distribution continues to be given by equation (2), and hence the traction
ratio is given by
q(x)
p(x) =
(Q)
P c2
c2
+ 2sx' (9)
For incipient slip, q(x)/p(x) ~ I, i.e.
q(x) ~
p(x)
(.9...)
IP
[1 + 2SX]
c
.
2
(10)
The most critical point is x/c ~ +1 if s < 0, and x/c ~ -1 if s > O.

In either case, stick will continue everywhere, i.e. there will be complete
172
cp(x)/P
3,0
2,5
"
,.:1
"
2,0
i\:
l,.,. 1.5
d'
sIc =-0,25,0,0,25"k:\ 1.0
\... .
"', = O~~C~-=='OC,,~ 0.5
-1.0 -0.5 0
x!c
Figure 2, Plot of the dimensionless normal pressure, cp( x) I P, along the contact for
various positions of the normal load, sic
adhesion while
!L
fP
< 1- 21s1
c . (11)
If this inequality does not hold, and assuming, from now on, that s <
0, we expect stick to occupy the region -c S x S b. The equation
connecting the slip displacement, g(x), to the shear traction distribution
is
(12)
and we let q(x) = fp(x) + q'(x), where q'(x) = 0 for b S x S c, so that

equation (12) becomes
~ dg _ 1+
A dx -
.!.l
7r
b
-c X - ~ ,
q'(~)d~ (13)
where
I-
-
.!.l
7r
c
-c 7rVc2 -
fP
~2
[1 + 2S~]
c
.-!!:L _ -2sf P
x- ~ -
2 7rC 2 .
(14)
After some algebra, the result is
q'(x)=2s f P
7rC 2
Jb-X,
c+ X
(15)
173
Q/fP
1.0
0.6
0.4
0.2
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
sIc
Figure 3. Plot of the initial extent of the slip zone (b/c) as a function of the position
of the normal load, s / c, and Q/ f P
Also, the connection between the shear force, Q, and the stick zone
extent, b, is given by the tangential equilibrium equation
Q=fP+1b q'(e)df,,=fP+ 281":1'

-c 7rC -c
Vb-;de.
+
C <"
(16)
The integral here is trivial, and provides the connection between the
dimensionless stick zone size, blc, and the load offset, 81c:
8 QlfP-l
(17)
c l+blc'
Note that, for complete contact I 8 I Ic < 1/2. If slip is incipient blc -+ 1
and equation (11) is recovered. As the value of QI f P is increased the
value of bI c decreases linearly, until, as Q I f P -+ 1, bI c -+ -1, which is
as required, for -1 I 2 < 81 c < O. The full stick regime and extent of the
initial slip zone are plotted out in Figure 3.
It may be verified that the shear traction in the stick region is every-
where less than the limiting value.
The next phase in the development of the solution is to establish the
relative slip displacement, g(x), in the slip region, b < x < c. Now,
equation (13) holds everywhere throughout the contact patch, although
the integral no longer assumes a Cauchy Principal Value. When evalu-
ated this gives
1 dg _ 28fP ~
A dx - - ---;;? ;+c V b< x < c. (18)
174
ltglx)/2fPA
2.5
blc
-0.75
2.0
- - QlfP =0.75 1.5

-_..- QlfP = 0.625
-0.5
-0.75 -0.5 -0.25 0 0.25 0.5 0.75 1.0

x1c
Figure 4. Plot of the dimensionless tangential displacement, 7rg(x}/2fPA, along

the contact surface, for various shear loads, Q/ f P, and stick zone sizes, b/c
The relative slip displacement is thus
b:S; x < c, (19)
which here is an elementary integral
2sfPA [ -..jx2+(c-b)x-bc-!(b+c)ln(b+c)+ 1
g(x)= 7rC2 !(b+c)ln(2..jx2+(c-b)x-bc+2x+c-b) '
(20)
and is plotted out in Figure 4. This, then, is the value of the relative
slip displacement in the slip zone, at any point during the first quarter
cycle of the loading history. To measure the 'step' length it is useful to
keep a record of the displacement at some particular point on the punch
surface. The most obvious point is the origin, so that, providing b < 0
g(O) = 2fPA [-J-bC+ (b+c) In (2J=bC+C-b)]. (21)

7rC 2 b+ c
3. Rocking Punch Back

We now wish to develop a solution when the position of the load
oscillates about the central point of the punch and this is schematically
tracked in Figure 5. Let the value of s increase from its most negative
value (-so), and consider the new stick and slip zone arrangement. The
punch has first been tilted in an anti clockwise direction, and, at the
175
stick-slip regimes Slip displacement (g(x))
() ~fflf#ffl#ffl#fflffffflfr 1
~ c
sdl
-so< s "I
-s.+ £ bfflfffflfffflffffflfff#fff'1
~ c
'---Ib,
~(X)l /
-~ .. -bo C
~
Figure 5_ Evolution of the stick/slip regimes
extreme position of the load, the stick region occupies (-C, bo ) and the
slip region (b o , c)_
Now rock the punch back, i.e. tilt the punch in a clockwise direction.
An infinitesimal rotation will cause the value of the contact pressure
to increase within the slip region, hereby causing stick, but adhesion
will still be maintained within the stick region, and hence the entire
interface is adhered. A further, finite, rotation will leave the problem
substantially unchanged until, at some new positive value of 8, slip will
start from the left hand end of the punch. Suppose that the value of
8 is increased still further, to a value 82, (0 < 82 < 8 0 ) and determine
the position of the new stick-slip transition point, b2 . The slip region
will now occupy the region (-c, b2) and stick will occupy the region
(b 2 , c). Within this new stick zone there will be a 'locked in' effect on
the relative displacements due to it previously being part of a slip zone.
Equation (8) continues to give the pressure distribution, p(x; 81), which
is uncoupled from the shearing traction, equation (12) continues to apply,
and therefore equation (13) becomes
~ dg = I+! t ql(~)d~. (22)

A dx 1T Jb 2 X - ~
176
Now, we require to form an integral equation for the corrective shearing

traction, q'(x), by writing down the value of the locked-in slip displace-
ment, left at the end of the first half of the loading cycle, in the stick
zone. Providing that b2 > bo , we see that the present stick region
is contained wholly within the previous slip region, for which the slip
displacement is given by equation (18). Thus equation (22) reads
.!. Je q'(~)d~ =
'Ir x- ~
2fP
'lrC
(81_Jx -+ b), b <
C X C
2 X < c. (23)
b2
We recall that q'(~) is required to be singular at x = c. It is not necessary

to solve this singular integral equation explicitly, because we intend to
use the solution to re-evaluate the slip displacement in the slip zone. We
then have, Gradshteyn and Ryzhik (1980),
} ]
(24)
Hence
(25)
where
(26)
and
(27)
Note that we must have b2 > bo in order for k2 < 1. This function
does not appear to be capable of analytical integration, but numerical
integration, to give the slip displacement is possible.
177
4. Walking Problem
The procedure described above gives the state of slip, shearing trac-
tions and slip displacement during the first three quarter cycles of the
rocking punch problem. Clearly, the solution is becoming increasingly
complex, because, with each rock, the expression for the 'locked in' slip
displacement is becoming more complex. It does not, therefore, seem
feasible to continue this solution towards a steady state. It is, how-
ever, possible to infer certain features from the solution obtained so far.
These are, first, the condition for the punch to remain stationary (which
is weaker than the condition for the punch to adhere). Secondly, an ap-
proximate measure of the 'step length' when the punch is walking may
also be found.
A good first approximation to the point representing the transition
from remaining stationary to walking is to determine the set of conditions
which cause the slip zones emerging from each corner of the punch just
to meet at the centre. Clearly, if the slip zones never touch, central
particles will remain adhered, and there can be no rigid body motion. It
will not be easy to determine the exact conditions for meeting, because
the problem will, inherently, contain a certain element of asymmetry,
but a sensible simple approximation would appear to be to determine
the condition where the slip zones move just as far as the centreline
(x = 0), from either side, i.e. Ib/cl - t O. The condition for this is that
walking cannot occur providing that
(28)
5. Conclusion
An analytical solution has been developed for the transient problem
of a punch, initially at rest and with a central normal load, subject to
an oscillatory moment of constant amplitude. It has been possible to
produce an analytical description of the behaviour of the punch for the
first three quarters of a cycle of loading. From this, the condition for
the punch to remain stationary, and an estimate of the initial step length
when slip occurs has been found. It is expected that these results will
form the basis of a fuller, numerical solution for further steps, against
which it may be calibrated.
References
Muskhelishvili, N.!. (1953) Some Basic problems of the Mathematical Theory of Elas-
ticity, Noordhoff, Gronigen.
178
Gradshteyn, 1. S. and Ryzhik, 1. M. (1980) Tables of Integrals, Series and Products,
Academic Press, New York.
Hills, D. A., Nowell, D., and Sackfield, A. (1993) Mechanics of Elastic Contact,
Butterworth-Heinemann, Oxford.
Sackfield, A., Truman, C. E., and Hills, D. A. The tilted punch under normal and
shear load (with application to fretting tests), Int. J. Mech. Sci., in press.
PHENOMENOLOGICAL MODEL
OF FRICTION ACCOUNTING FOR
SUBSURFACE PLASTIC DEFORMATION
IN METAL FORMING
Stanislaw Stupkiewicz
Institute of Fundamental Technological Research
Swi~tokrzyska 21, Warsaw, Poland
Zenon Mr6z
Institute of Fundamental Technological Research
Swi~tokrzyska 21, Warsaw, Poland
Abstract The paper deals with the effect of subsurface plastic deformation on
the evolution of the real contact area and on friction in metal forming
processes. A model of evolution of real contact area is derived within a
phenomenological framework by considering a thin homogeneous surface
layer representing the asperities and the layer of inhomogeneous defor-
mation induced by deforming asperities. The friction model is then
derived by assuming asperity flattening and ploughing mechanisms to
occur at different scales.
1. Introduction
Accurate modelling of contact interactions is crucial for reliable simu-
lations of metal forming processes. The real contact area fraction, which
is an important state variable governing the phenomena of friction, lu-
brication and heat transfer, is usually high in these processes due to
high contact pressures. An additional increase of the real area of con-
tact is caused by plastic deformation of the underlying bulk material
which significantly accelerates asperity flattening process. This effect is
observed experimentally and also predicted by micrcrmechanical models.
The corresponding models have been developed on the basis of the slip
line method (Sutcliffe, 1988), the upper bound approach (Wilson and
Sheu, 1988; Kimura and Childs, 1999) and finite element solutions {Ko-
179
l.A.C. Martins and M.D.P. Monteiro Marques (eds.J, Contact Mechanics, 179-186.
180
rzekwa et al., 1992). In these micro-mechanical models, highly idealized

process conditions and contact geometries are assumed, as required by
the applied solution techniques.
In the present work, the effect of bulk deformation is discussed within
the phenomenological approach. A thin surface layer is considered which
is weakened (as compared to the bulk) due to inhomogeneous plastic
deformations localized around the deforming asperities. The relations
between asperity flattening pressure, real contact area factor and its
rate, and plastic strain rate in the bulk are derived. Unlike the exist-
ing micro-mechanical models, the present model is applicable for general
stress/deformation states at the contact interface and can thus be ap-
plied for any kind of forming process. Furthermore, the general frame-
work of the model can be refined by adding the effects of roughness
anisotropy, visco-plasticity and strain hardening of the bulk material.
Based on the proposed evolution law of the real contact area factor, a
friction model is also proposed by assuming that two basic asperity defor-
mation mechanisms (flattening and ploughing) occur at different scales.
The model extends the dual asperity concept of Mroz and Stupkiewicz
(1998) by adding the effect of bulk plastic deformation on flattening of
workpiece asperities and evolution of real contact area.
2. Asperity interaction modes

The contact conditions in metal forming processes are characterized
by the contact of a relatively hard tool surface and a softer workpiece
surface. Consequently, two basic asperity interaction modes are dis-
tinguished, namely flattening of workpiece asperities and ploughing of
hard tool asperities through the workpiece. In the simple models of fric-
tion these two modes are associated with the two basic mechanisms of
friction, adhesion and abrasion, and approximated by simplified rough
workpiece-smooth tool (RW-ST) or rough tool-smooth workpiece (RT-
SW) interaction modes, Fig. 1.
v'
workpiece PN~
/
tool tool
(a) (b)
Figure 1. (a) RW-ST and (b) RT-SW interaction modes.

181
In the RW-ST interaction mode the tool surface is assumed to be

smooth and rigid and workpiece asperities are flattened due to the com-
bined action of normal and tangential contact stresses. Asperity flatten-
ing is accompanied by the increase of the real contact area fraction a so
that the sliding friction stress p¥
is
p¥ =mka, (1)
where mk is the adhesive friction stress at the real contacts, which is
assumed constant and expressed as the fraction m (O < m < 1) of the
workpiece yield stress in shear k. Since for high normal contact pressures,
a approaches unity, a threshold sliding friction stress p¥ = mk results
from this model. The real contact area fraction a in eqn. (1) is assumed
to depend on the normal pressure PN. Such relations can be obtained
from micro-mechanical considerations, for example using the slip line
field technique to solve the problem of flattening of interacting asperities,
cf. Wanheim et al. (1974) and Bay (1987). The resulting friction law
provides a nonlinear relation between the sliding friction stress and the
normal contact pressure, cf. Fig. 2.
Figure 2. Nonlinear friction law.
The abrasive friction mechanism results from ploughing action of hard

tool asperities through the workpiece surface, Fig. l{b). Here the fric-
tion force results from plastic deformation induced by ploughing and
also from adhesive forces at real contacts. The micro-mechanical models
assuming the RT-SW interaction mode have been studied extensively
in the literature using both slip line theory and upper bound method.
The latter method was used by Avitzur and Nakamura (1986) to de-
rive a nonlinear relation between normal and sliding friction stresses,
qualitatively similar to that shown in Fig. 2.
A model combining the two asperity deformation modes (flattening
and ploughing) has been proposed by Mroz and Stupkiewicz (1998).
This model is based on the assumption that the tool asperities are much
smaller than the workpiece asperities so that flattening of workpiece as-
perities is modelled according to RW-ST model, while the friction con-
ditions at the real contacts of workpiece asperities follow from the RT-
182
SW model, Fig. 3(a). In the model proposed by Mroz and Stupkiewicz

(1998) the effects of bulk plastic deformation on asperity flattening were
not accounted for. In the present work, this model is further developed
by incorporating the evolution law of the real contact area in the pres-
ence of bulk deformation proposed recently by Stupkiewicz and Mroz
(2001).
,,,,,,,,,,,,,,,
PN
- v
tool p'N p'N

(a) (b)
Figure 3. Dual asperity model: (a) asperity interaction, (b) average local stresses
at real workpiece asperity contacts.
3. Evolution of real contact area

In this section a phenomenological model of evolution of the real con-
tact area associated with flattening of workpiece asperities (RW-ST in-
teraction mode) is briefly introduced, cf. Stupkiewicz and Mroz (2001)
for details. The real contact area fraction a constitutes here the measure
of the asperity flattening process. Note that the dual asperity approach,
Fig. 3, implies that the actual real contact area is the product of the
real contact areas of workpiece asperities a w and tool asperities at, so
that a = awat, cf. Mroz and Stupkiewicz (1998). In this work we shall,
however, only consider the real contact area associated with flattening
of workpiece asperities.
Real asperities and a layer of inhomogeneous deformation localized
around the asperities are replaced by a thin homogeneous surface layer,
Fig. 4. The surface layer is weakened with respect to the bulk material
due to inhomogeneous plastic deformations resulting from interacting de-
formation fields of surface asperities. Several assumptions are adopted:
i) stresses and strains are uniform across the layer;
ii) stress and displacement continuity conditions are assumed at the
interface between the surface layer and the bulk material;
iii) strain hardening effects are neglected, so that local (pointwise)
material properties of the asperities are identical to those of the
bulk material;
183
iv) elastic strains are neglected and rigid-plastic, rate independent ma-
terial model is assumed for the bulk and the surface layer.
bulk material
Figure 4. Homogeneous surface layer.
The yield condition of the surface layer is assumed to depend on a

in addition to the (macroscopic) stress in the layer u l . In the matrix
notation it is written as
FI(uI, a) = J~(ul)TfJJl (a)P + !J2(a)P*]u l - 93 (a)uy = 0, (2)

where P is the standard projection matrix onto the deviatoric space
and the only non-zero components of P* are the diagonal components
of P corresponding to the components of the contact traction vector (a
coordinate system is adopted with the Oxa-axis normal to the nominal
contact surface). Functions 9i(a) are given by
9l(a) = 1, 92(a) = 1- a, 9a(a) = a2 + 2+11"

v'3 a(1 - a), (3)
so that for a = 1 the yield condition of the layer FI = 0 reduces to the

standard Huber-von Mises yield condition ofthe bulk material. Further,
an associated flow rule is assumed: el = )..1{jFI / {jul.
The flattening of the asperities is related to the deformation of the
layer by assuming a simple relation between the real contact area fraction
a and the volumetric strain in the surface layer i~ = £~l + £~2 + £~a,
namely
el = emaxa emax = 10'" + lItan8 ~ _ 1 - 1I 8, (4)

v v 'v g '" + tan 8 '"
where e~ax is the maximal volumetric strain in the layer, occurring for
a = 1, derived from simple geometric considerations. Parameters 8 and
1I depend on the initial roughness: 8 is the average asperity slope and
1 - 1I is the porosity of the layer between the highest asperity peaks and
the deepest asperity valleys (e.g. 1I = ~ for two-dimensional wedge-like
asperities). The phenomenological parameter 'fJ denotes the thickness of
the surface layer (relative to the characteristic asperity length).
184
Using the flow rule and the continuity conditions at the interface be-
tween the surface layer and the bulk material, the in-plane components
of the stress vector 0'1 and the out-of-plane components of the strain
rate vector £1 are eliminated, and the yield condition of the surface
layer is expressed in terms of dimensionless contact stresses PN = PN/k,
Fr = PT/k, real contact area fraction a and dimensionless bulk strain
rate Eo
p1(PN, Fr,Eo , a) =
92 [1 + 92 (1 _ R2) ( Eo ) 2] R2 + 91 + 92 R2 _ 1 = 0, (5)
39~ 91 T e~ax N 9~ T
where the effective bulk strain rate £b = Vi(£b)Teb is normalized by

the rate of the real contact area fraction
£b
Eo = -;- . (6)
a
The yield condition (5), also called the asperity flattenin9 condition,
can be resolved for Eo, so that Eo = Eo(PN,Fr,a). In view of the
definition (6) of Eo the following evolution law of the real contact area
fraction is obtained
(7)
This evolution law has successfully been verified against existing micro-
mechanical models and experimental data, cf. Stupkiewicz and Mroz
(2001). In Fig. 5 the predicted evolution of a as a function of plastic
strain in the bulk ell is compared to the results of asperity crushing
experiment of Sutcliffe (1988). In this experiment, wedge-like asperities
were compressed with a constant normal pressure and lateral tension
was applied, as depicted in Fig. 5.
4. Friction model
Let us now apply the dual asperity concept, Fig. 3(a), to derive a
friction model combining the two basic asperity deformation mechanisms
of flattening and ploughing. The average dimensionless local contact
stresses at workpiece asperity contacts, cf. Fig. 3(b), are given by
p~ = PN = PN , ~ = Fr = PT . (8)
a ak a ak
The effect of bulk plastic deformation on asperity flattening and the
related growth of real contact area a is discussed in the previous section.
185
0.025 0.05 0.075 0.1 0.125 0.15

£11
Figure 5. Real contact area fraction 0: as a function of lateral strain ell: model
predictions ("1 = 1.20) and experimental data (Sutcliffe, 1988).
Following the approach of Mr6z and Stupkiewicz (1998), the frictional

interaction at workpiece asperity contacts is assumed to be governed by
ploughing of tool asperities through the flattened workpiece asperities.
This interaction is assumed in the form of a simple nonlinear friction
law, cf. Fig. 2,
pa(p~,Pf) = IPfl-mtanh ('.t~~) ~ 0, (9)
where J.t is the friction coefficient and m is the friction factor. Combining
eqns. (9) and (8) the following limit friction condition is obtained
P(&,Fr,a) = IFrI-amtanh (:~) ~ o. (10)

This condition involves the state variable a with the evolution law speci-
fied by eqn. (7). Note that, unlike in the simple RW-ST models, a is not
a unique function of contact stresses, but depends on the history. Sur-
face roughening due to plastic deformations is not accounted for by the
evolution law (7), thus a increases monotonically. Figure 6(a) presents a
scheme of the evolution of the limit friction surface F = 0 with increas-
ing a. The limit friction surface P = 0 corresponding to a fixed value
of a is shown in Fig. 6(b) together with the asperity flattening surfaces
pi = 0 corresponding to different values of Ea. Assuming that the bulk
strain rate t b is fixed, the latter surfaces are the surfaces of constant a.
The evolution law (7) involves the effective strain rate in the bulk ma-
terial, which is a non-standard contact variable. Finite element imple-
mentation of the model is thus not a standard task. A possible approach
to treat this problem, using the extended contact element (Stupkiewicz,
2001) will be discussed separately.
186
(a) (b)
Figure 6. Limit friction and asperity flattening surfaces.
Acknowledgments
This work has been partially supported by the Polish State Committee
for Scientific Research (KBN), Grant No.8 T07A 022 20.
References
Avitzur, B. and Nakamura, Y. (1986). Analytical determination of friction resistance
as a function of normal load and geometry ofsurface irregularities. Wear, 107:367-
383.
Bay, N. (1987). Friction stress and normal stress in bulk metal forming processes. J.
Mech. Working Technol., 14:203-224.
Ike, H. (1995). Plastic deformation of surface asperities associated with bulk defor-
mation of metal workpiece in contact with rigid tool. In Raous, M., Jean, M., and
Moreau, J., editors, Contact Mechanics, pages 275-286. Plenum Press, New York.
Kimura, Y. and Childs, T. (1999). Surface asperity deformation under bulk plastic
straining conditions. Int. J. Mech. Sci., 41:283-307.
Korzekwa, D., Dawson, P., and Wilson, W. (1992). Surface asperity deformation dur-
ing sheet forming. Int. J. Mech. Sci., 34(7):521-539.
Mr6z, Z. and Stupkiewicz, S. (1998). Constitutive model of adhesive and ploughing
friction in metal forming processes. Int. J. Mech. Sci., 40:281-303.
Stupkiewicz, S. (2001). Extension of the node-to-segment contact element for surface-
expansion-dependent contact laws. Int. J. Num. Meth. Engng., 50:739-759.
Stupkiewicz, S. and Mr6z, Z. (1999). A model of third body abrasive friction and wear
in hot metal forming. Wear, 231:124-138.
Stupkiewicz, S. and Mr6z, Z. (2001). Phenomenological model of real contact area
evolution with account for bulk plastic deformation in metal forming. Int. J. Plast.
(to appear).
Sutcliffe, M. (1988). Surface asperity deformation in metal forming processes. Int. J.
Mech. Sci., 30(11):847-868.
Wanheim, T., Bay, N., and Petersen, A. (1974). A theoretically determined model for
friction in metal working processes. Wear, 28:251-258.
Wilson, W. and Sheu, S. (1988). Real area of contact and boundary friction in metal
forming. Int. J. Mech. Sci., 30(7):475-489.
THERMOELASTIC INSTABILITIES IN
AUTOMOTIVE DISC BRAKES -
FINITE ELEMENT ANALYSIS AND
EXPERIMENTAL VERIFICATION
Yun-Bo Yi and J.R.Barber

Department of Mechanical Engineering, University of Michigan
Ann Arbor, MI48109-2125, USA
D.L.Hartsock
Ford Scientific Research Laboratories
Dearborn, MI48121-2053, USA
Abstract The thermomechanical feedback process due to frictional heating in slid-

ing systems can cause thermoelastic instability (TEl), leading eventually
to localization of load and high temperatures at the sliding interface.
TEl in caliper/disc brake systems is an intermittent contact problem,
since material points on the disc experience periods of contact with the
pad alternating with periods of non-contact. The stability problem is
here solved numerically by setting up a frame of reference stationary
with respect to the pad and seeking a solution for the heat conduc-
tion and thermoelastic equations that varies exponentially in time. The
upwind scheme is introduced in the finite element formulation to avoid
possible numerical difficulties associated with the large convective terms.
A series of brake dynamometer drag tests was made to investigate ex-
perimentally the phenomenon of TEl in an automotive disc brake. The
temperature field on the rotor surface was measured with infrared (IR)
detectors and a high-speed data acquisition system. The Fast Fourier
Transform (FFT) method was used to determine the exponential growth
rate for various hot spot numbers and critical speeds. Linear extrapo-
lation was then used to determine the speed for zero growth rate - i.e.
the critical speed. The results for critical speed and the number of hot
spots show good agreement with the numerical predictions.
187
JAC. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 187-202.
188
1. Introduction
Automotive disc brakes are susceptible to a low frequency vibration
known as 'judder' or 'roughness', particularly during high speed engage-
ment (Kubota et al., 1998). Major improvements in noise reduction in
other areas of automotive design have reduced customer tolerance of
brake noise, resulting in a significant volume of warranty claims asso-
ciated with brake judder (Steffen and Bruns, 1998, Lee and Dinwiddie,
1998).
Various mechanisms have been proposed to account for brake jud-
der, including rotor imbalance, non-uniform accumulation of frictional
transfer films, core shift and disc thickness variation, but considerable
evidence has now accumulated to show that it is associated with an un-
stable interaction between thermoelastic distortion and frictional heat-
ing known as Thermoelastic Instability or TEl (Jacobsson, 1999, Yi et
al., 2000). Temperature measurements of noisy brakes typically show
a non-uniform circumferential temperature variation (Kreitlow et al.,
1985; Thoms, 1988, Little et al., 1998) and examination of brake discs
after engagement reveals evidence of focal hot spots distributed approx-
imately evenly around the circumference (Anderson and Knapp, 1989).
The feedback mechanism responsible for TEl is illustrated by the flow
diagram of Figure 1. Frictional heating during braking causes thermoe-
lastic distortion, which in turn modifies the contact pressure distribution
p(x, y, t) and hence the distribution of frictional heating q(x, y, t). Cou-
pling between the mechanical and thermal problems is introduced by the
energy balance relation
q(x,y,t) = jVp(x,y,t) , (1)
where j is the coefficient of friction and V is the sliding speed. It is
clear from Figure 1 that the product jV functions as the gain in the
feedback process and it follows that there will generally be a critical
speed Vcr above which any given sliding system will be unstable (Dow
and Burton, 1972). Above the critical speed, non-uniform perturbations
in the temperature field will grow, leading to a characteristic pattern of
hot spots or bands on the brake disc.
Ideally, we would like to be able to design the brake system so that
the operating conditions always lie below Vcr. However, Vcr is signifi-
cantly affected by the geometry of the system (Lee and Barber, 1993)
and hence it is generally necessary to use numerical methods (typically
the finite element method) to obtain appropriate design estimates. Two
broad categories of numerical solution have been attempted. In numer-
ical simulations, the system is discretized by the finite element method
and the nodal temperatures are evaluated after each of a succession of
189
Contact pressure p(x,y,t)

and contact area
/
Solution of contact
problem
~ Frictional heat generation
q(x,y,t) = jVp(x,y,t)
Unconstrained thennal Solution of heat

distortion conduction problem
v = sliding speed
f = friction coefficient
Figure 1. The feedback process for TEL
small time increments using an updating algorithm (Zagrodzki, 1990).

In the alternative eigenvalue formulation, Burton's method is general-
ized to determine the conditions under which a small perturbation in
the discretized temperature field can grow exponentially in time (Du et
al., 1997). In both methods, difficulties are encountered due to the rela-
tive motion of the brake components, which introduces convective terms
into the governing equations for at least one body. Peclet numbers are
generally high, leading to the confinement of the thermal disturbance
in a thin layer of the pad material and necessitating a very fine local
mesh (Yi et al., 2000). Yi showed that non-axisymmetric perturbations
on an otherwise axisymmetric brake or clutch system can be analyzed
by considering perturbations of Fourier form. In this case, analysis for
each Fourier number is restricted to the two-dimensional cross-sectional
plane, in which there is no convection. However, this method cannot be
used for typical caliper disc brakes, since the perturbations will generally
not then have Fourier form.
In the present paper, we shall extend Yi's method to non-axisymmetric
caliper disc brake systems, which will necessitate appropriate discretiza-
tion of equations with convective terms at high Peclet number. We shall
compare the predictions with a set of experiments on commercial disc
brakes using infrared (IR) temperature measurement to determine the
critical speed and dominant mode shapes.
190
2. The automotive disc brake

Figure 2 shows the main features of a typical automotive disc brake.
Two brake pads make contact with the two plane surfaces of the rotating
disc, the surface facing out from the vehicle being known as the outboard
surface and the other as the inboard surface. Most discs contain air vents
at the mid-plane for cooling purposes, but it should be remarked that
practical heat transfer coefficients are not sufficient to enable significant
cooling to occur during brake engagement. The principal function of
the cooling vents is to accelerate cooling of the disc in the usually longer
periods of time between engagements. The disc is connected to the axle
by a hat section and the precise way in which this is connected to the
disc can have a significant effect on thermoelastic distortion. The brake
pads are pressed against the disc surfaces by hydraulic pressure through
a caliper mechanism, designed to equalize the forces on the inboard
and outboard sides. However, this equalization is usually achieved by
a sliding mechanism and this may lock due to frictional effects during
loading, preventing the system from responding to changes in pad loads
during a single engagement. This can have a significant effect on the
stability of modes involving small numbers of hot spots.
pad
hat
section
....-
rotor
Figure 2. A typical automotive disc brake.
3. Finite element determination of critical speed

For the disc brake system, we use a cylindrical polar coordinate system
r, </J, z , fixed with respect to the caliper and brake pads. The eigenvalue
solution is then obtained by postulating the existence of a perturbation
191
in the temperature field of the form
T(r, </J, z, t) = R{ ebte(r, </J, z)} (2)

involving exponential growth with time t. If this form is substituted
into the various governing equations and boundary conditions of the
processes described in Figure 1, the exponential factor will cancel, leav-
ing a homogeneous problem defined in the spatial domain r, </J, z, with
the (possibly complex) growth rate b as a parameter. The finite element
method is then used to discretize these equations, leading to a linear
eigenvalue problem for b. The system is unstable if there exists any
eigenvalue b with positive real part, since any slightly perturbed initial
condition could be expressed as an eigenfunction expansion.
3.1. The heat conduction problem

The temperature perturbation must satisfy the heat conduction equa-
tion
2
Kr/v T - Pf3cf3
(aT
at + aT)
a</J =
wf3 0, (3)
where Kf3, Pf3, cf3 are the thermal conductivity, density and specific heat
respectively of material f' and wf3 is the angular velocity of body f' in
the given frame of reference. If we denote the rotor as f' = 1 and the
pad/ caliper assembly as f' = 2, then WI will be the rotational speed of
the rotor and W2 will be zero.
Substituting (2) into (3), canceling the common exponential factor
and discretizing the resulting equation by the finite element method
leads to the matrix equation
(K + C + bH)9 + Q = 0, (4)
where 9, Q are the vectors of nodal temperatures and nodal heat sources
respectively,
and N(r, </J, z), W(r, </J, z) are vectors of shape and weight functions re-
spectively, defined in the domain Of3.
192
The nodal heat sources Q are non-zero only at the contact interface,
where they result from frictional heating. For the disc brake geometry,
the sliding speed V varies with radius r, being given by V = wr. The
discrete form of equation (1) is therefore
Q = jVipP, (6)
where P is the nodal contact force normal to the contact interface which
is defined only at the Nc contact nodes, V is the diagonal Nc x Nc matrix
defined by
(7)
ip is an Nc x N matrix defined by
(8)
and I is the identity matrix of order Nc x N c.
3.2. Upwind finite element scheme

Numerical solutions of the convective-diffusion equation tend to ex-
hibit unacceptable oscillations when the Peclet number characterizing
the convective term is sufficiently large (Christie et al. 1976). This is
almost always the case for brake problems, where typical Peclet num-
bers are of the order of 105 . Galerkin-type finite element formulations
result in expressions similar to those of central difference, which become
unstable when the element Peclet number
Vh
Peel = k > 2, (9)
where h is the element dimension and k is the thermal diffusivity.

Christie et al. (1976) and Heinrich et al. (1977) showed that this
difficulty could be avoided by using linear shape functions Ni, but special
weighting functions of the form
(10)
where x E (0, h) is a local coordinate. This reduces to the Galerkin

form when the upwind factor a is zero, whereas a = 1 gives a fully
upwind scheme. Corresponding weighting functions for two and three-
dimensional problems can be obtained in the same way. For example,
193
for a bilinear quadrilateral element we obtain
( (I-e) _ 3a(l-e)(1+e)) (I-TJ)

242
( (I+e)+ 3a(l-e)(I+e)) (I-TJ)
242
(11)
( (I+e)+ 3a(1-e)(1+e)) (1+TJ)
242
( (I-Q _ 3a(1-e)(1+e)) (1+TJ)
242
e
where E (-1,1) and'fJ E (-1,1) are dimensionless local coordinates in
and perpendicular to the corivective direction respectively. The resulting
shape and weighting functions are compared in Figure 3 for a = 1.
w,
(a) bilinear shape function (b) upwind weighting function
Figure 3. Comparison of the shape function and the upwind weighting function for
a bilinear quadrilateral element.
3.3. The thermoelastic contact problem

To complete the feedback loop in Figure 1, we need to solve a ther-
moelastic contact problem to determine the contact pressure distribution
p(x, y, t) due to a given instantaneous temperature field T(x, y, z, t). As
long as full contact is maintained, this is a problem in linear thermoe-
lasticity which in discrete form corresponds to a matrix equation of the
form
P=A8, (12)
where A is an N x Nc matrix relating the vector of N nodal temperatures
8 to the vector of Nc nodal normal forces at the contact nodal pairs. Yi
et al. (2000) describe various methods of determining the matrix A, but
the most efficient is to develop a custom finite element description of the
thermoelastic contact problem. Details of this process are omitted here
in the interests of brevity.
194
3.4. The eigenvalue equation

Eliminating Q and P between equations (4,6,12), we obtain
[(K + C + jVcl>A) + bHJ8 = 0, (13)
which is a generalized linear eigenvalue equation for the exponential

growth rate b. The matrices are not symmetric and the eigenvalues are
generally found to be complex. Instability is indicated if at least one
eigenvalue has positive real part. The corresponding eigenfunction 8
then defines the form of temperature perturbation that will develop.
The factor w in equation (7) and hence (13) shows that the eigen-
values and eigenfunctions will depend on rotational speed. The critical
speed Wcr can be determined by iteration until the dominant eigenvalue
has zero real part. In practice it is found that the growth rate varies
approximately linearly with speed above the critical speed, so a good
estimate is obtained by making a single linear extrapolation through
growth rates for two super-critical speeds.
4. Experimental investigation
The evolution of hot spots in a caliper disc brake during drag braking
was monitored using an array of fiber optic cables connected to single-
point, two-colour, infrared (IR) detectors. The detectors were housed in
liquid nitrogen cooled dewars to maintain a constant temperature. The
IR system is described in detail by Hartsock et al. (2000).
Five sensors were arranged along a radial line on each of the inboard
and outboard sides of the rotor. During braking, each sensor sweeps
the entire circumferential distribution of temperature at a given radius
with every revolution, permitting a time history of the complete surface
temperature distribution to be reconstituted from the recorded data.
The IR system was set to take temperature readings at a frequency of
2000 Hz, which corresponds to one reading every 4.2 degrees of rotation
at the highest rotational speed used (1400 rpm). This was considered
to give adequate resolution of the spatial distribution of temperature for
typical hot spot patterns.
The use of two-colour detectors reduces the sensitivity of the system
to emissivity, but does not completely eliminate it. The system was
therefore calibrated against a set of thermocouples embedded 0.5 mm
below each surface of the disc. The rotor was subjected to 200 burnish
stops and several long drags to condition the surfaces. It was then heated
to 600°C by dragging at low speed and calibration readings were taken
during subsequent cooldown at 30 rpm with the brakes released and no
external cooling air. During this period, the surface of the rotor was
195
assumed to be the same temperature as the thermocouples due to the

high conductivity of the cast iron and the slow rate of cooling. A ninth
order poynomial was used to fit the data for each IR detector, a typical
calibration curve being shown in Figure 4.
o test data
- 9th order polyfit
500
~ 400
~
::J 300
~
Q)
~ 200
.Sl
-200L-----~-------L-------L------~----~
o 2 4 6 8 10
voltage (V)
Figure 4. Calibration curve for one of the IR sensors.
Figure 5 shows a typical instantaneous temperature distribution. The

temperature field is clearly periodic, so it is convenient to separate com-
ponents of different wavenumber in the digital data using Fast Fourier
Transformation (FFT). The amplitude of a typical Fourier mode is plot-
ted logarithmically against time in Figure 6. This particular component
only begins to rise out of the noise level after about 60 s of sliding. There
is then an approximately linear portion of the curve, corresponding to
exponential growth, after which the growth levels off, probably because
contact between the pad and the rotor becomes localized in hot spots,
resulting in a strong non-linearity (Zagrodzki et at., 2001). An expo-
nential growth rate was estimated for each Fourier mode by fitting a
straight line to the logarithmic plot in the growth phase, as shown in
Figure 6.
4.1. Test procedure

Whenever a new set of hardware was installed, it was subjected to
a program of 200 burnish stops to condition the pad/rotor interface.
196
Figure 5. Typical instantaneous temperature field due to TEl.
10'
straight line fit

T(C) for linear region
\
o
Qj)
50 100 150 200

time (s)
Figure 6. Logarithmic plot of average amplitude against time for the mode n = 7.
The slope of the straight line fit corresponds to a growth rate R(b) = 0.0368 S-1.
Temperature measurements were then made during a sequence of drag

braking (constant speed) tests in order of increasing sliding speed. Since
197
the probability of TEl increases with speed, this procedure eliminates

the possibility that TEl at a higher speed would precondition the rotor
at a particular wavenumber resulting in false indications of TEl at lower
speeds. A typical test sequence is shown in Table 1.
Table 1. Typical test sequence.
Sections Type Speed (rpm) Torque (Nm)

1 200 Burnish stops
2",,4 3 Drag tests 416 254
5 20 Burnish stops
6",,8 3 Drag tests 555 191
9 20 Burnish stops
10",,12 3 Drag tests 694 153
13 20 Burnish stops
14",,16 3 Drag tests 833 127
Notice that the friction torque was controlled so as to maintain the

same rate of power dissipation (and hence heat generation) in each test.
This ensures that the evolution of the mean temperature of the brake
system is approximately the same for each test and reduces the impact of
temperature-dependent material properties on the stability behaviour.
5. Results and discussion

For all the brake systems tested, hot spots were clearly identifiable
when the sliding speed was sufficiently high. The number of hot spots
developed was quite reproducible for a given brake design and the final
pattern usually involved a number of equally spaced focal hot spots lo-
cated approximately midway between the inner and outer radii of the
rotor. However, the scenario of hot spot development showed some vari-
ability. In some cases, hot spots started near the outer edge and pro-
gressed inwards, whilst in others they started near the inner edge and
progressed outwards. This movement was believed to be the result of
pad wear and warpage during the long drag. Although there was no
fixed location where the hot spots would start, there was a tendency to
develop a hot band near the outer radius of the rotor, probably as a re-
sult of the higher local sliding speed. This band then frequently started
to exhibit variations which would develop into hot spots.
198
Results from each drag test were used to obtain exponential growth
rates of all Fourier modes following the procedure illustrated in Figure
6. The critical speed for each mode was then estimated by linear extrap-
olation through the results for the speeds with identifiable exponential
growth.
5.1. Comparison of experimental results and

theoretical predictions
pad arc length

10
Figure 7. Dimensions of the brake system (in mm).
The dimensions of the basic brake design tested are shown in Figure 7.
The rotor contains 37 vents (not shown in Figure 7) of thickness 12 mm.
The vent:vane ratio is 4:1 and hence the vents occupy 80% of the total
circumference of the rotor at the mid-plane. For the standard brake, the
pad arc length <pp = 66°, but tests were also conducted on brakes with
reduced arc lengths of 53° and 40°. Other design modifications tested
included the reduction of the total rotor thickness dr (by machining a
standard rotor) from 28 mm to 21 mm and of the pad friction material
thickness dp from 10 mm to 5 mm.
Theoretical predictions of the critical speed were obtained using the
finite element method of section 3. In order to model the vents, 74
elements were used around the circumference of the disc, resulting in
the model of Figure 8, which contains about 3000 eight node solid el-
ements. Some approximation is involved in this representation of the
vents, since in reality the vent boundaries move in the given frame of
199
Figure 8. Finite element discretization of the brake system.
reference. However, numerical studies show that the temperature per-

turbation is largely concentrated near the rotor surface, so the resulting
error is not expected to be large. An alternative strategy would be to
estimate the properties of an 'equivalent homogeneous material' to re-
place the vent/vane region. A graded mesh was used near the sliding
surface of the friction material in order to capture the rapid temperature
gradients in the thermal skin. The mesh refinement in this region was
chosen to satisfy the criterion developed by Yi et al. (2000). The mate-
rial properties used in the numerical solution are given in Table 2 and
the coefficient of friction was taken as f = 0.4, based on experimental
measurements. The pad modulus was obtained by ultrasonic measure-
ments through the thickness direction, since this mode of deformation
dominates the pad deformation.
Table 2. Material properties.
E v a K k
CPa x10- 6 deg K- 1 Wlm deg K mm 2 /s
Rotor 112.4 0.25 13.25 57.0 17.2
Pad 2.03 0.35 30.0 0.93 0.52
Backing plate 200 0.3 12.0 42.0 11.9
Figure 9 compares experimental estimates of the critical speed for

each wavenumber with theoretical predictions for a brake with pad arc
angle <pp = 66°, pad thickness dp = 10 mm and rotor thickness dr = 21
mm. The dominant mode for this brake involves 7 hot spots, for which
the estimated experimental critical speed is Vcr = 348 rpm. The finite
200
Experimental result
3-D FEA model
600
•
_500
-
E
c.
....
- 0400 • 0 .
0
CI)
• 0 0 ~
~
!!!.co 300
(.)
• • •
0
:0=
.~
(.) 200
100
2 468 10 12
hot spot number
Figure 9. Comparison of experimental estimates and finite element predictions of

critical speeds for various wavenumbers.
element method also predicts 7 hot spots, with a critical speed of Yc~ =
316 rpm.
Similar comparisons were performed for brakes with other values of
the parameters <pp, dp, dr and the results are summarised in Table 3.
In all cases the predicted number of hot spots n* is within ±1 of the
experimentally observed value and the maximum error in the predicted
critical speed is 27%, which represents an exceptional level of agreement
for such a complex problem.
The experimental results show that reducing the pad arc angle <pp
from 66° to 53° has comparatively little effect on the critical speed, but
that a further reduction to 40° increases it by a factor of 2.4. This
increase is associated with an increase in the number of hot spots from 7
to 10, which is close to the minimum number that ensures that at least
one hot spot is located in the contact zone at all times (Lee and Barber,
1993).
Reducing the pad thickness dp reduces the critical speed significantly.
This is attributable to the increased stiffness associated with the thinner
layer of compliant friction material. It suggests that brakes are more
likely to exhibit judder when the brake pads are significantly worn.
These effects are clearly identified by the finite element analysis. Other
investigations, not reported here, have shown that the FEA also cor-
201
Table 3. Comparison of experimental results (Vcr) and theoretical predictions (Vcr)

of critical speed.
CPp dp dr V n b Vcr Vc*r n* Error

mm mm mm rpm S-l rpm rpm
66 10 21 555 7 0.022 348 316 7 9.2%
66 10 21 694 7 0.037
53 5 28 555 7 0.057 233 274 8 17.9%
53 5 28 694 7 0.067
53 10 28 555 6 0.007 342 435 7 27.1%
53 10 28 694 6 0.012
66 10 28 555 6 0.016 324 281 7 13.3%
66 10 28 694 6 0.026
40 10 28 833 10 -0.001 833 726 10 12.7%
40 10 28 971 10 0.043
rectly identifies the effect of changes in the geometry of the hat section
on susceptibility to judder.
6. Conclusions
We have shown that a finite element implementation of the eigenvalue
method gives generally excellent predictions of the critical speed for TEl
for a caliper disc brake and also successfully identifies the effect of minor
design changes on the critical speed. The analysis of one brake design
by this method takes about 5 hours CPU time on a workstation. By
contrast, no results have been reported for direct numerical simulation
of a brake system with anything like the present level of geometric detail.
We conclude that the eigenvalue method is the best numerical tool for
TEl brake design presently available.
Acknow ledgments
The authors are pleased to acknowledge support from the Ford Mo-
tor Company and from the National Science Foundation under contract
number CMS-9619527.
202
References
Anderson, A.E. and Knapp, R.A. (1990) Hot spotting in automotive friction systems,
Wear 135, 319-337.
Christie, I., Griffiths, D.F., Mitchell, A.R. and Zienkiewicz, O.C. (1976) Finite element
methods for second order differential equations with significant first derivatives,
International Journal for Numerical Methods in Engineering, 10, 1389-1396.
Dow, T.A. and Burton, R.A. (1972) Thermoelastic instability of sliding contact in
the absence of wear, Wear 19, 315-328.
Du, S., Zagrodzki, P., Barber, J.R. and Hulbert, G.M. (1997) Finite element analysis
of frictionally-excited thermoelastic instability, J. Thermal Stresses 20, 185-201.
Hartsock, D.L., Fash, J.W., Dalka, T., Smith, G., Yi, Y-B, Hecht, R. and Dinwiddie,
R.B. (2000) Development of a High-Speed IR System for Temperature Mapping of
a Rotating Target, 14th Annual International Symposium on Aerospace/Defense
Sensing, Simulation and Controls. Thermosense, Orlando, FL.
J.C.Heinrich, J.C., P.S.Huyakorn, P.S. and O.C.Zienkiewicz, O.C. (1977) An 'upwind'
finite element scheme for two dimensional convective transport equation, Interna-
tional Journal for Numerical Methods in Engineering, 11, 131-143.
Jacobssen, H. (1999) Analysis of brake judder by use of amplitude functions, SAE
99-01-1779.
Kreitlow, W. Schrodter, F. and Matthiii, H. (1985) Vibration and hum of disc brakes
under load, SAE 850079.
Kubota, M., Suenaga, T. and Doi, K (1998) A study of the mechanism causing high
speed brake judder, SAE 980594.
Lee, K and Dinwiddie, R.B. (1998) Conditions of frictional contact in disk brakes
and their effects on brake judder, SAE 980598.
Lee, K and Barber, J.R. (1993) Frictionally-excited thermoelastic instability in au-
tomotive disk brakes, ASME J. Tribology 115, 607-614.
Little, E., Kao, T-K, Ferdani, P. and Hodges, T. (1998) A dynamometer investigation
of thermal judder, SAE 982252.
Steffen, T., and Bruns, R. (1998) Hot spot formation on passenger car disks, ATZ
Automobiltechnische ZeitschriJt 100, 408-413.
Thoms, E. (1988) Disc brakes for heavy vehicles, Proc.Int. Conf. Disc Brakes for Com-
mercial Vehicles, Inst.Mech.Eng. (London), C464/88, 133-137.
Yi, Y-B., Barber, J.R. and Zagrodzki, P. (2000) Eigenvalue solution of thermoelas-
tic instability problems using Fourier reduction, Proc.Roy.Soc. (London) A 456,
2799-2821.
Zagrodzki, P. (1990) Analysis of thermomechanical phenomena in multidisc clutches
and brakes, Wear 140, 291-308.
Zagrodzki, P., Lam, KB., Al-Bahkali, E.and Barber, J.R. (2001) Nonlinear transient
behaviour of a sliding system with frictionally excited thermoelastic instability,
ASME J. Tribology, in press.
FRICTION COEFFICIENT EVOLUTION DURING
WEAR TESTS RELATED TO INFRARED
EMISSION
Nadine Stalin
Laboratoire de Mecanique des Solides, CNRS-UMR.7649, Ecole Polytechnique,
91128 Palaiseau cedex, France.
Presently at Direction des Constructions Navales, Etablissement S.c., Site du Mourillon,
BP 30, 83800 Toulon Naval, France.
Jean Christophe Eytard

Laboratoire de Mecanique des Solides, CNRS-UMR.7649, Ecole Polytechnique,
91128 Palaiseau cedex, France.
Abstract The purpose of this paper is to present an experimental methodology to

measure the friction coefficient in real time and to estimate temperature
changes during sliding wear contact. First, tribological and infrared
experiments are described. Second, experimental results consisting of
simultaneous estimations of friction coefficients and maximal temperatures are
presented. This approach has the advantages of being non-destructive and in
real time, and of requiring no additional contact.
1. Introduction
Contact problems occur frequently in engineering practice. Obvious

examples are roller-bearings, brakes, etc ... , where solid parts are in contact.
To predict their durability, it is desirable to know parameters such as the
contact pressures, the tangential tractions and the frictional heating. The heat
transfer in the vicinity of the contact area is often of critical importance.
If direct measurements are difficult, an experimental way to assess
thermal effects induced by friction is then the infrared thermography
method. For instance, Luong (1999) has used the infrared thermography to
detect the occurrence of damage and fatigue processes in metals. Sakagami
203
204
and al. (1997) determined the stress distribution by a thermography method.

More recently, Harish and al. (2000) used a thermography camera to study
heat exchange to evaluate stresses by measuring temperatures. The first
attempts to calculate the temperature due to frictional heat generation were
made by Block (1937) and by Jaeger (1943). Johansson (1992) studied
frictional heat generation under microslip conditions. He proposed a model
and a numerical algorithm to study the sliding contact between two elastic
half-planes with frictional heat generation and wear. Today, the finite-
element method is popular for solving contact problems. This numerical
method allows the estimation of mechanical variables such as stress or strain
near the contact zone with an elastoplastic constitutive model. For that
purpose, Barghava and al. (1985) considered an incremental translation of
the loading. Such an approach is rather involved and the authors are limited
to a small number of cycles. More recently, Dang Van and Maitournam
(1993, 1994, 1995) have developed a systematic and reliable finite-element
algorithm for thermoelastoplastic solids, which is well adapted for repeated
rolling and sliding contact problems. This method, which is based on a
steady state assumption for an observer moving with the load, was used by
Stalin (1997) to interpret sliding wear experiments. In that study, numerical
calculations were performed to estimate the mechanical stress and strain
fields in the vicinity of the wear region and to infer the dissipation
mechanism activated. In these numerical studies, the friction coefficient is a
crucial parameter. To make the calculations tractable, this parameter was
assumed to be constant. However the validity of this hypothesis was not
confirmed.
To check the validity of this assumption, wear experiments are now
conducted in which the continuous evolution of the friction coefficient is
shown. We measure in real time the evolution of the friction coefficient with
a classic tribological experiment and a thermal one. The latter is a new
approach based on infrared measurements. This paper is organised as
follows. First, tribological and infrared experiments are described. We
consider a half sphere sliding on a plate, both made of the same stainless
steel material. An infrared camera captures the thermographic emission
generated at the contact. Second, we present some experimental results
concerning the evolution of the global friction coefficient and the
temperature fields. During the tests, we measure directly and instantaneously
the values of the global friction coefficient, defmed as the ratio between the
tangential force and the normal force. All this time we measure indirectly the
evolution of the temperatures. These observations are made on a plane
parallel to the contact plane, using a precision infrared device. A comparison
between the measured evolutions of the friction coefficient and of the
maximal temperatures during the wear test is performed.
205
2. Experimental methodology
Observations are made on a specimen consisting of a plate of 45 mm

length, 35 mm width and I mm thickness, fixed on a rigid support. A half
sphere of 10 mm diameter slides on this plate. The roughness of the plates is
0.5 J.11Il. The average micro hardness measures are about 220 HV200 . The
plate and the half sphere are made of the same stainless steel (304L)
material. To increase the emissivity of the material we covered the observed
surfaces with a mat colour. Continuous contact and repeated loading are
made with a gauge and a spiral spring. This mechanical device applies to the
plate a force that is perpendicular to the surface. Observations are made on a
surface, parallel to the plane of contact where the temperature increases
because of conduction through the thickness of the specimen. The
temperature change is observed by an infrared thermal device. The reference
temperature is the room temperature. The objective of the infrared thermal
device, situated at 10 cm away from the plane, allows the observation of a
part ofthe plate with a wide angle lens (40°). The resolution is about 2 pixels
per millimetre. The observed area is large enough to include the
displacement of the sphere on the plate. The global friction coefficient is
measured with appropriate sensors. Figure 1 presents the tribological
experimental device and the scheme of experiments.
LOADING
t
BALL
INf'RARED THERMOGRAPHY
Figure 1. Experimental device and the scheme of experiments
Today the power of infrared technology is capable of resolving

temperatures finely both spatially and temporally. The following figures
show the maps of temperatures which have been observed during a test. The
normal loading is 37 N and the sliding velocity is 17 mm/s. Figure 2 shows
one sequence chosen between two thousand thermal maps which have been
206
recorded during the test. The frequency is about ten pictures per second. The
sequence ofthermal maps shows the evolution oftemperature distribution.
==l MUlIIINOX PI"" "~E'J

156 208 456 506 556 606
15:06:18 «15:06:22 .441 5:06: 26 4C'I 5:06:30.4115:06:34.4:1 5:06:38.4415 06:42 «15.06:46 4Sl5:06:5CI.«15.06:54 41
.!l .J -!..l
Figure 2. Sequence of thennal maps
Figure 3 shows a zoom of a map and a profile of temperatures.
L" ' PHUtlllINU X 11"JL'PI-, . . r-I
Figure 3. A map and a profile of temperatures
The camera zoomed in the observed surface. We can localise the

evolution of temperatures in real time. The maximal temperature reached an
average of 7°C. The temperature fields observed with this set-up are
influenced by both frictional heating, due to the relative motion at the
contact interface, and the coupled thermomechanical strains in the material.
3. Results of the wear tests
The method discussed here consists of estimating simultaneously the

evolution of the friction coefficient and the maximal temperature during the
sliding tests. We made several attempts with different loads and relative
velocities. Three of them are presented in this paper. The experimental
conditions (t : time, F : normal loading, v: sliding velocity, d : amplitude of
imposed displacement and N : frequency of loading) are given in table 1.
207
Table 1. Experimental conditions of wear tests
t (s) F(N) V (mmls) d(mm) N(Hz)

220 17 6 9.5 0.3
225 18 17 8. 1
100 27 120 10.4 6
The following figures 4, 6 and 7 show the evolution of the friction

coefficient f and the variation of maximal temperature I1Tmax measured on
the parallel surface to the contact plane. The sliding velocities are 6, 17 and
120 mmls respectively. The friction coefficient f is a global parameter. The
variation of maximal temperature I1Tmax is the temperature measured in the
: r~-~ -.--..- -·=~-=-I~·=-I :

observed area, the reference temperature being the room temperature. The
delay I1t is due to the conduction through the plate thickness.
~
~
0,7
0,6
1;'i 05
e '
- - - ..:
- - - - - - - - - - - - - - i1
---- -_1-
:
!:.
~: ~~
-*------~-
--------
0,7
0,6
0,5 ....
~ 04 +-___ d __ :I
I:
.
.. : ····1
!!!'(*i rl_+-~'-II--\
l:'
, ,:il'! r
0,4
:~! lif \:;;i1i'~ii.

0,3
0,1 -
o
o 20 40 60 80 100 120 140 160 180 200 220 240
time(s)
Figure 4. Evolution off - - and IlTmar------------ and average IlTmax _
F = 17 N, v = 6 mm/s and N = 0.3 Hz
A comparison between the measured evolution of the friction coefficient

and the maximal temperature during the wear test shows a similar behaviour.
We remark that the friction coefficient f varies continuously in the range of
0.1 to 0.5. The increase in temperature is about 0.3 °C. The thermal variation
is insignificant. The temperature changes are small due in part to the load
level and the thermal properties of stainless steel 304L. We observe a sudden
drop in temperature in time after 150 s. An observation of the load history
shows that the sudden decrease of the temperature corresponds to a decrease
of the loading. Figure 5 shows the evolution of the loading during the test.
208
26
24
22 +--lMH~~,,~--~---~
20
18
16
~ 14
'" 12
10
8
6~------------------~----~~~~~1~~H+-----
4~------------------------------~mm~
2~--------------------------~~~
O~r~~~~--''-T~~~~T~~~~~~~~~~~~~~-'.4-,~-c~
o 20 40 60 80 100 120 140 160 180 200 220 240

time (s)
Figure 5. Evolution ofloading

F = 17 N, v = 6 mm/s and N = 0.3 Hz
The loading dropped because there were some debris in the contact
interface and this probably modified the control of the loading. When the
sliding velocities are larger, the loading is almost constant: see Figures 6 and
7, where v=17 mm1s and v=120 mm1s respectively.
4,5 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 0,9
0,8
3,5 - 0,7
0,6
0,5 ...
0,4
1,5 ---------------"-----. ------- 0,3
0,5 0,1
°° 20 40 60 80 100 120
time(s)
140 160 180 200 °
220
Figure 6. Evolution of f _ _ and IlTmax _

F= 18N, v= 17 mm/s andN= I Hz
Transient and stabilised phases are observed on both curves. Large

plastic deformations are supposed to occur during the first loading cycles
because we observe large variations of the friction coefficient (maximal
variation is 0.5) during this phase.
209
28
26
0,9
24
22 0,8
20 0,7
18
E 16
0,6
....
~ 14 0,5
f-<12
<I 0,4
10
8 0,3
6 0,2
4
0,1
2
0 0
0 10 20 30 40 50 60 70 80 90 100 110
time (s)
Figure 7. Evolution off _ _ and IlTm a x _

F = 27 N, v= 120mmls and N = 6 Hz
We can observe a very short transient phase and a large stabilised phase
for both evolution curves. The friction coefficient f increases in the
beginning then it becomes stabilised (average value is 0.6) very quickly
(only 20 seconds after the beginning of the test). The curve corresponding to
!1Tmax has the same evolution although with some delay. The similarity
between the evolution of the friction coefficient and that of the temperature
is quite remarkable.
Other different tests have been conducted with various values of

velocities (50 mmls, 80 mmls ... ) and the same behaviour has been observed.
4. Conclusion
Numerical friction studies require a friction coefficient which is generally
assumed to be constant. The objective of this paper is to check this
hypothesis and it is shown experimentally that instead the friction coefficient
varies continuously in the range of 0.1 to 0.6.
Technological progress in infrared thermography has changed the way
one studies thermal effects in mechanics and, in particular, in frictional
contact problems where direct measurements are difficult.
A new approach based on indirect infrared measurements of temperature
fields and on direct measurements of the friction coefficient has been
210
proposed. The thennal emissions, observed in real time on a surface parallel
to the contact zone, pennit an understanding of the evolution of the friction
coefficient under the contact zone. These results show that the friction
coefficient varies continuously during the tests. Transient and stabilised
phases are observed and we have measured large variations of the friction
coefficient at the beginning of the tests. It varies in the range of 0.2 to 0.5 in
Figure 6 and 0.3 to 0.6 in Figure 7. The nonnalised pressure probably
exceeds the elastic limit which is approximately equal to 3, as shown by
Johnson (1992). In the wear tests the changes in the contact area arise from
indentation, wear or plastic defonnations of the layers near the surface
during repeated sliding contact. The proposed method opens up new
horizons for evaluating the local friction coefficient and for interpreting wear
phenomena.
References
Barghava V., Hahn G. T. , and Rubin C. A. (1985), A.S.ME. 1. Appl. Mech., 52,67-82.
Block H. (1937), Theoretical Study of Temperature Rise at Surfaces of Actual Contact under
Oiliness Conditions, Proc. Inst. of Mechanical Engineers General Discussion of
Lubrication, 2, 22-235.
Dang Van K., and Maitoumam H. (1993), Steady-state flow in classical elastoplasticity :
Applications to repeated rolling and sliding contact, 1. Mech. Phys. Solids, 41,1691-1710.
Dang Van K., and Maitoumam H. (1994), Elasto-plastic calculations of the mechanical state
in reciprocating moving contacts : Application to fretting fatigue, Mechanical
Engineering Publications, London, 161-168.
Dang Van K., Maitoumam H. and Prasil B. (1995), Elastoplastic analysis of repeated moving
contact, application to railways damage phenomena, Wear, 196, 77- 81.
Harish G., Szolwinski M. P. , Farris T. N. and Sakagami T. (2000), Evaluation of fretting
stresses through full-field temperature measurements, Fretting Fatigue: Current
Technology and Practices, edited by Hoeppner D. W. , Chandrasekaran V. and Elliott C.
B. for Testing and Materials, West Conshohocken, PA, ASTM STP, 1367, 423-435.
Jaeger J. C. (1943), Moving Sources of Heat and the Temperature at Sliding Contacts, Proc.
R. Soc. N. S.w., 76, 203-224.
Johansson L. (1992), Elastic and Thermoelastic Contact Problems with Friction and Wear,
Linkoping Studies in Science and Technology, 266.
Johnson K. L. (1992), The application of shakedown principles in rolling and sliding contact,
J. Mech. A. Solids, 11,155-172.
Luong M. P. (1999), Infrared thermography of macrostructural aspects of thermoplasticity,
IUTAM Symposium on Micro and Macrostructural Aspects of Thermoplasticity, 0. T.
Bruhns and E. Stein, Kluwer Academic Publishers, 20, 437-446.
Sakagami T., Ogura K., Kubo S., Lesniak J. R., Boyce B. R. et B. I. Sandor B. I. (1997),
Visualization of Contact Stress Distribution Using Infrared Stress Measurement System,
SPIE, 3056, 250-259.
Stalin N., and Dang Van K. (1997), Numerical simulation of the sliding wear test in relation
to material properties, Wear, 203-204, 180-186.
A CONTACT FORMULATION FOR ELECTRICAL
AND MECHANICAL RESISTANCE
G. Zavarise
Department o/Structural Engineering and Geotechnics
The Turin Polytechnic
D. Boso and B. A. Schrefler

Department o/Construction and Transportation Engineering
University 0/ Padua
Abstract In this paper a formulation to deal with the Finite Element modelling of
coupled electric-mechanical contact resistance is presented. The electrical
contact resistance is based on a micro-macro approach, i.e. the contacting
surfaces are not considered as perfectly smooth, but are characterised at the
microscopic level by the presence of valleys and peaks. Such roughness
reduces the real contact area to a very small percentage of the apparent one.
The microscopic geometry thus plays a very important role in perturbing any
type of field (mechanical, thermal, electrical etc.) in the contact zone. Some
macroscopic models, based on microscopic characterizations, have been
proposed to describe the normal and tangential contact stiffness and the
thermal contact resistance. Within such models, the surface microrugosity of
the asperities in contact is taken into account by replacing each surface by a
statistically equivalent one. The load-flattening mechanical behaviour of the
contacting asperities is modelled by choosing a perfectly plastic constitutive
law as well as an elastic one. In this way the mean plane approach is obtained
as a function of the applied load variation.
The smallness of the real contact area causes a perturbation also of the electric
field. The flux lines must concentrate where the asperities are in contact. In
analogy with the thermal contact resistance this electrical constriction
resistance is studied, supposing a flux tube around each asperity, and choosing
a suitable geometry for its narrowing at the contact zone. The contributions,
due to each flux tube, are added in parallel to determine the electric resistance
of the apparent contact area. The contact constitutive law so obtained, is then
consistently linearized, in order to keep the quadratic convergence of the
Newton-Raphson iterative scheme.
211
212
1. Introduction
During the last few decades the family of contact problems has been
growing in complexity more and more, due to the progresses made in
advanced fields such as fusion technology, astronautics, metal forming
processes etc ..
In several applications, contact conditions have to be satisfied very
accurately, i.e. global parameters like the Coulomb friction coefficient have
to be replaced with more accurate relationships. This means that the global
contact behaviour, expressed through contact constitutive laws, has to be
based on the real physical behaviour of the contacting surfaces.
In this study attention is focused on the development of a constitutive law
suitable for contact problems with coupled electric and mechanical fields.
2. Model set up
Several theoretical models are available to deal with mechanical and

thermal aspects. Although, in their original formulation, they can be used
only for closed form solution of very simple problems, they constitute a
good starting point for building a FE contact element.
An interesting theoretical contact model for rough surfaces was
developed by Cooper, Mikic and Yovanovich (1969) and successiveJy
improved by Mikic (1971), (1974), Yovanovich (1981), Song and
Yovanovich (1987). The model is based on the mechanical hypothesis of
perfect plastic behaviour of the contacting asperities, and is devoted to the
analysis of thermal contact resistance. This approach has shown a good
correspondence with several experimental tests. Due to the close analogy
between the thermal and the electrical fields, such approach can be adapted
to the analysis of the electric contact behaviour as well.
2.1. Geometrical and mechanical aspects
In Cooper et al. (1969), a geometrical relationship is obtained where the

ratio between the real and the apparent contact area is related to the ratio
between the sum of the lengths of the real contact zones and the total length
of a two-dimensional representative profile. Assuming that, from a statistical
point of view, the contact between two ideal asperities represents that of any
asperity in the whole contact surface, the mean radius of the asperities and
the number of contacts per unit area can be obtained as a function of some
statistical parameters. With a Gaussian distribution of the asperity heights,
the number of contacts per unit area obtained by Cooper et al. (1969), is
213
_ Nse _ 1 m 2 exp - d 2 / a 2
n,e - A - 16 a 2 eric d/,[ia '
(1)
where d is the mean plane distance, m is the mean asperity slope and
a is the surface roughness.
The mean radius of the circular contact zone is
-a =V;
[8 m
a exp[ 2a2
d 2} dJ
ric ( ,[ia . (2)
These equations involve only geometrical terms. In order to predict the

contact conductance we need a relationship between the pressure applied at
the interface and the actual contact area Ar , or the ratio AJAa. The model
is developed assuming that all the contact spots are in a plastic state, hence
the following linear relationship holds
(3)
where p is the pressure and H the hardness. From eqs. (2) and (3) the
ratio between real and apparent contact area can also be written as
(4)
Comparing eqs. (3) and (4) we obtain a relationship between the applied
pressure and the mean plane approach
The relationship is characterised by the presence of the term eric, which

does not have a closed form. A simpler relationship has been proposed in
Song and Yovanovich, (1987). However, within the framework of the FE
method it is possible to deal directly with the original equation and to
integrate it numerically. The advantages of such a strategy are the improved
quality of the solution, and the increased range of values for which the law
can be applied.
214
Surface finishing leads to the surface having a different hardness

modulus from the bulk material. This modulus can be related to the pressure
under the indenter in an indentation test, which depends also on the size and
shape of the indenter. Hegazy (1985) proposed the following relationship
(6)
where d v is the diagonal of the indenter and C/' C2 are two parameters
depending on the material. Assuming that the contact micro hardness of the
surface being penetrated by the asperities of the harder surface is the same as
the Vickers micro hardness (cOrre.10nding to the equivalent Vickers
indentation diagonal) we have d v = 2Jra, which combined with eqs. (2)
and (6) gives the asperity hardness
(7)
By substituting eq. (7) in (5) we obtain a relationship which links the

apparent contact pressure to the surface approach, i.e. a pressure-mean plane
distance relationship which is the basis for a mechanical contact law:
2.2. Electrical and thermal aspects
The electrical constriction resistance can be studied like the thermal one,
by supposing a flux tube around each asperity in contact, and choosing a
suitable geometry for its narrowing at the contact zone. The contribution of
each tube flux is then added in parallel to determine the electric resistance of
the apparent contact area. The most suitable narrowing is the flat circular
one, and for a disk and a half-space the constriction resistance can be
expressed as (see Holm (1981))
1
R=-- (9)
I 4k*a'
where k* is the mean harmonic conductivity of the joint if two different

materials are in contact: k * = (2k/k2 )/(k/ + k)). To deal with a flux tube the
215
conductance is half the previous one, and a correction rule, 'If;, as a function
of the narrowing radius, a;, and the flux radius b; must be applied
c =_2_k_*a_; = 2k*a i
c, 'If; (I -a;/bi ys . (10)
Applying this result to each ideal tube, we obtain the conductance of the
joint:
(11)
and by recalling eq. (1), (2) and (3) we obtain, after some algebra
h.
(
= k' m
2(5
(1- VnrP)-]'S 2
exp(- d /(2(52 )) .
Jii (12)
Again due to the complexity of this equation, Yovanovich (1981)

proposed an alternative correlation function.
3. Formulation for FEM computation
We need to discretise the contact constitutive law
(13)
where u is the vector of the mechanical unknowns of the contact

element, K is the non-linear stiffness matrix and F is the vector of applied
loads. Indices M and E stand for the mechanical and the electrical field
respectively. The following formulation is based on the relationships (8) and
(12).
3.1. Contact normal stiffness
To develop the constitutive law, the mean plane distance, d , is replaced

by the difference between the maximum asperity height, ;, and the current
approach, gN' i.e. d =; - gN' The normal force FN is obtained as a
216
product of the apparent contact area and the apparent pressure. Recalling eq.
(8) we have
Collecting numerical constants and constant parameters in a suitable

form, we find
(15)
Equation (14) can be written as
3.2. Electric contact resistance
Substituting equation (5) in eq. (12) we obtain the exact form for the
electric current
=A k*m (1-
a 2a
where gv is the contact voltage jump. Note that the relationship depends
on the apparent contact area, the mechanical and the electrical gap, i.e. it is a
function which may be easily expressed in terms of primary variables.
217
3.3. Contact geometry
The mechanical and thermal analysis of the contact zone is obtained by

integrating the contact constitutive law along the contact surface. Due to the
required accuracy, in some cases the node-to-segment formulation is not
suitable for this purpose. To achieve a better result, we have to increase the
number of sampling points. This can be obtained by considering a suitable
number of sampling points on the segment between two slave nodes. If the
slave surface is used to define the integration line, then the optimal choice is
to associate the sampling points to the Gauss ones. In this way, each Gauss
point corresponds to a virtual slave node (see Fig. 1). All the data belonging
to a virtual slave node are associated with the two slave nodes that define the
slave segment through classical shape functions. The possibility of using as
many sampling points as we need is a noticeable feature that can prevent the
classical problem with unchecked penetrations.
The geometrical formulation of this method has been implemented and
checked with the classical penalty method. The implementation of the
contact constitutive law is now in progress and results will be presented in a
forthcoming paper.
Virtual slave node Slave segment
Master segment
Figure 1. Geometrical contact scheme
Acknowledgments
The authors wish to thank the support of MURST Res. Contr. "40% -
Contact mechanics: constitutive laws for interfacial phenomena and discre-
tization techniques", and EURATOM/ENEA Res. Contr. "EFDAlOO-521".
218
References
Cooper, M. G., Mikic, B. B. and Yovanovich, M. M. (1969) Thermal contact conductance,

Int. J of Heat and Mass Transfer 12, 279-300.
Hegazy (1985) Thermal Joint Conductances of Conforming Rough Surfaces: Effects of
Surface Microhardness Variation, Ph.D Thesis, Dept. of Mechanical Engineering,
University of Waterloo, Canada.
Holm, R. (1981) Electric Contacts: Theory and Applications, 4th edn. Springer Verlag, Berlin
Heidelberg New York.
Mikic, B. B. (1971) Analitical studies of contact of nominally flat surfaces; effect of previous
loading, Journal of Lubrification Technology 93,451-459.
Mikic, B. B. (1974) Thermal contact conductance; theoretical considerations, Int. J of Heat
and Mass Transfer 17,205-214.
Song, S. and Yovanovich, M. M. (1987) Explicit Relative Contact Pressure Expression:
Dependence Upon Surface Roughness Parameters and Vickers Microhardness
Coefficients, AIAA 25th Aerospace Sciences Meeting, Reno Nevada.
Yovanovich, M. M. (1981) Thermal Contact Correlation, AIAA Paper, 83-95.
THE REAL AREA OF CONTACT-
A COMBINATION OF EXPERIMENTAL AND
NUMERICAL APPROACHES
FHBucher
Institutfor Luft- und Raumfahrttechnik. Technische Universitiit Berlin.
Marchstr. 12. 10587 Berlin
R S Dwyer-Joyce
Department of Mechanical Engineering. University of Sheffield.
Sheffield. S1 3JD
Abstract In this work the real area of contact between two rough surfaces has been in-
vestigated by means of ultrasonic experiments and numerical modelling. A
wave of ultrasound is partially reflected from a rough surface interface. The re-
flection can be related to the stiffness of that interface. However, the stiffness
does not uniquely define the real area of contact within the interface. The first
loading of the rough pair was shown to occur plastically. A numerical contact
model has been adapted to approximate plasticity at the asperity contacts. The
contact pressure at the asperity peaks is truncated and the plastically flowed
material distributed into nearby valleys. Predictions of the interface stiffness
from this model compare favoumbly with those measured in the ultrasonic ex-
periments. This combined approach provides a way of either validating rough
contact models or to help in the interpretation of measured stiffness results.
1. Introduction
When engineering components are loaded together, contact is made at the

peaks of the surface roughness. The real area of contact can be significantly
less than the nominal area over which the bodies would be expected to come
into contact. So far, it has proved difficult both, to experimentally measure
this real area of contact, and also to theoretically predict it.
219
JA.C. Martins and M.D.P. Monteiro Marques (eds.). Contact Mechanics. 219-228.
220
One modelling approach is to consider the surface as an array of geomet-
rically shaped asperities with some statistical height distribution (for exam-
ple Greenwood and Williamson 1966). An alternative is to use a discretized
representation of the contact surfaces in an elastic model. An early approach
along these lines was that of Webster and Sayles (1986). The load on each
node point is related to the normal deflection of that point and all its neigh-
bours (i.e. the Bousinnesq solution). A matrix relating the complete array of
contact pressures to the total surface deformation is created. This is then
solved until the load applied to the contacting bodies is consistent with the
deformed shape of the surfaces. Further developments have incorporated
tangential tractions and creepages (Kalker 1967, Wang 1991) and advanced
rapid solver techniques (Venner and Lubrecht 2000).
Surface discretization using a half space assumption has inherent prob-
lems. Firstly, it is only suitable for elastic deformation, since regions of
maximum plasticity occur sub-surface. Secondly, as the discretization inter-
val is reduced, more detail of the surface roughness is revealed. The numeri-
cal calculation then leads to lower real areas of contact and commensurate
high local stresses typically above the elastic limit (thereby violating the ini-
tial elastic assumption). In this work an approximate way of handling this
asperity plastic tip deformation has been employed.
A reliable experimental measurement of real area of contact has also
proved elusive. Optical methods can be used to directly observe contact re-
gions (but generally at a low resolution and only if one body is transparent).
The conduction of electricity or heat across an interface can give information
about the conformity of the contact. A promising method is the use of ultra-
sonic reflection; first demonstrated by Kendall and Tabor (1971). The pres-
ence of an incomplete interface causes the partial reflection of an ultrasonic
wave. This reflection can be processed to give a measurement of the stiffness
of the interface (Drinkwater et al. 1996). The stiffness varies from zero to
infinity as the real area of contact (and the proportion of the ultrasonic wave
reflected) varies from 0% to 100%. However, stiffness cannot be uniquely
related to the real area of contact since it depends not only on the number
and size of the individual contact regions, but also on their distribution.
In this study a numerical contact procedure has been adapted to incorpo-
rate an approximation of plasticity at the asperity tips. To do this effectively
requires the specification of a number of parameters. Results of experiments
using ultrasound have been used to set these parameters. The experimental
method is used in this way to tune the operation of the numerical model.
221
2. Ultrasonic Measurement of Interface Properties
2.1. Principle of Operation
When an ultrasonic wave is incident at an interface it will pass through

regions of contact and be reflected back at any air gaps. The proportion of
the wave amplitude reflected back is known as the reflection coefficient, R.
In the low frequency regime (i.e. the ultrasonic wavelength is much lar-
ger than the size of the contact gaps) this reflection coefficient can be shown
to be (Tattersall 1973) a function of both the frequency of the wave, ro and
the stiffness of the interface, K:
R = ~ZIZ2f +K2(z1 -z2f (1)

~ZIZ2f +K2(ZI +z2f '
where z is the acoustic impedance of the material (the product of the wave
speed and the density) and 1 and 2 refer to the materials on either side ofthe
interface.
The stiffness of an interface (in the above equation expressed per unit
area) is defmed as the pressure required to cause an unit approach of the sur-
faces.
K=I~· (2)
We use the term nominal pressure,Pnom (i.e. load divided by nominal area
of contact) to distinguish from the real contact pressure, Pr (the load divided
by the real area of contact) which develops at the contact regions.
The measurement of the reflection coefficient thus gives a method for de-
termining the stiffness of an interface. In parallel we use a numerical contact
model to determine the relationship between the applied load on the surfaces
and their approach and hence the stiffness of the interface according to equa-
tion (2).
2.2. Apparatus
Figure 1 shows a schematic of the apparatus used to record the reflection

of ultrasound from an interface. The transducer emits a focused regular
pulse. The pulse reflects back at the interface and is received by the same
transducer. The return signal is captured using a digital scope and passed to a
PC.
222
The signal from the loaded interface is divided by that from an unloaded
interface. This gives the reflection coefficient as it varies with frequency.
The spring model (equation I) is then used to determine the stiffness of the
interface, K (which is independent of frequency). Further details of this ex-
perimental procedure along with the development and validation of the ap-
proach is given in Dwyer-Joyce et al. (2001).
applied load
Fig. I: Schematic of the ultrasonic reflection apparatus
In this work the lower specimen is grit blasted aluminium (H30TF). The
upper specimen is hardened steel (EN24) which has been ground. The di-
ameter of the nominal contact patch is 9 mm. Here, the idea is to load the
rough aluminium against the smooth hard steel such that the majority of the
deformation will occur on the aluminium surface.
3. The Theoretical Approach
3.1. Discretization of the rough surfaces
A possible way of simulating the contact of rough surfaces with a deter-

ministic model is to consider them as elastic half spaces. A simplification
leads to the contact of one elastic half space with a rigid counterpart. The
combined surface roughness is then applied to the elastic part as a deviation
to the smooth macro contour.
The surfaces have to be discretized. The number of nodes with which the
surface is represented depends on the shortest wavelength to model, Af.
Hereby the asperities are thought to be part of the half space, thus they them-
selves have to be modelled with a sufficient number of nodes. Results from
earlier computations show that ten points for the smallest wavelength in the
profile in each direction is a good choice. Thus to compute the contact of
223
rough surfaces including wavelengths down to one hundredth of the discre-

tized length, one must take at least 1,000 nodes.
Since this is true for each direction of discretization, a three dimensional
contact model leads to 106 nodes representing the rough surface. In a plain
strain model on the other hand every asperity like the bulk itself is taken to
be infinitely wide in the y-direction. Since one usually gets the surface
heights data via measurements taken with stylus profilometer, the arbitrarily
chosen line need not necessarily represent the roughness of the surface. To-
gether with another disadvantage of that approach, which also appears in the
3d contact model, the fact that one has to arbitrary position the two surface
profiles to each other, makes it hard to compare the results of the simulation
to what might be seen in practice.
3.2. Numerical aspects
If one nevertheless accepts those two weaknesses in the approach, the

plain strain model leads to very fast numerical methods to compute the con-
tact of rough surfaces including very low roughness wavelengths, i.e. very
high spatial frequencies.
In this case the normal contact problem reads:
uz+h-u z
I:: {= 0 point is in contact ~ p >0
(3)
> 0 point is not in contact ~ p =0
with Uz as the normal displacement, h as the surface geometry and Oz as the
rigid body displacement. The system of equations in the plain strain contact
model for the contact points is given by:
_I '" IT/'" . . _ K .. 1--.

E * £.. ~'O' lj JP,
=ii . - ii .
Z,IO Z,I
(4)
with the influence coefficients K ij obtained from the Boussinesq-equation

for piecewise linear distributions of unit pressure values.
For this work we used the numerical program VISeON (see Wang 1991)
for the simulation of the contact. VISeON is a program based on a direct
method to solve the system of equations (4). Thus due to the computer ca-
pacity it is limited to approximately 5,000 nodes for the representation of the
rough surface. Nevertheless because of the large number of iterations in the
approximate asperity plasticity algorithm described below, the number was
restricted to ~I,OOO nodes.
224
3.3. Material behaviour
When using the half space assumption for modelling the contact of rough
bodies, one is restricted to pure elastic behaviour of the material.
Considering higher spatial frequencies of the rough surfaces, leads to unrea-
sonably high surface stresses and lower real area of contact (see figure 3).
Even at low loads the material close beneath the surface will plastically de-
fonn due to the very high local pressures. Assuming that every local contact
spot is a small Hertzian contact, one can introduce a maximum value for the
local nonnal pressure (i.e. a shakedown limit), above which the material in
the vicinity of the asperity is assumed to plastically defonn.
Bucher et. al (2000) introduced an iterative algorithm to keep the nonnal
pressures below that maximum pressure value and still maintain equilibrium.
However this does not take into account the flow of material from the plasti-
cally defonned asperity tips into the neighbouring valleys (as observed in the
experiments of Pullen and Williamson 1972).
.....\-----.-..--------.-..
PI--------~~ ..~---
Pm..
/: pressure
o ••• (' i .............J....:..

elastically deformed
\ \ surface profile
undeformed
\ surface profile
local heights to be added

to the undeformed surface
~ ~ " profile for the next iteration
U Xo f, f, = c·~h·cos(1t·x/a,,,,,)
f,= O.5·c·~h·cos(1t·x/a ...,)
Fig. 2: Geometrical changes of an asperity in the "approximately plastic" algorithm
For this work, a new procedure for approximate consideration of the mi-
cro plasticity has been developed. The changes in the local contact spots that
had elastic pressure values larger than Proax were considered in the following
way: a very small change in the asperity height was applied; the curvature
radius was increased and the "defonned mass" was distributed amongst the
neighbouring nodes (see Figure 2). This was done for every load considered
until the solution contained no local pressures larger than Pmax.
4. Results and Discussion
Figure 3 shows stiffness measurements recorded during three cycles of

loading and unloading. As the specimens are loaded together the surfaces
225
confonn and hence the interface becomes stiffer. The first loading takes
place with plastic defonnation occurring at the asperity tips. Unloading then
takes place elastically; now the stiffness at a given load is higher because the
surfaces have been previously plastically defonned.
Subsequent re-Ioading and unloading cycles follow closely to the last
unloading. This suggests that these loadings are largely elastic. However,
there is some further plastic flow (indicated by the continuing rise of the
stiffness at the final load point) and a hysteresis loop on each cycle. It is
suggested that this hysteresis is the result of adhesion between the surfaces
(Dwyer-Joyce et al. 2001).
1.' ,..----------------=__.
1.4
I 1.2
I
I
1
i
0'
~
Fhi;;;;.;-l
0.' -+-FIrltUnIoad
0.4
..•.. '."""'lDod
··.··SecondU""*'
- .. ·ThinlLoId
.j
0.2
-0-. Third UnIoacI
~ ~ w ~ ~ M ~ U
NomIMI Contact P.....un, GP.
Fig. 3: Interface stiffness for an aluminium steel contact as it undergoes three cycles ofload-
ing and unloading.
Surface profIles were recorded from the specimen before loading took
place. The numerical contact procedure described above was perfonned.
Figure 4 shows the pressure distribution predicted for two elastic cases (with
differing threshold roughness wavelengths) and a plastic case. Figure 5
shows the original and defonned surface profIle after the plastic loading.
The stiffness determined from the numerical model is shown in figure 6
compared with the first loading cycle extracted from figure 3.
The agreement between the experimental stiffness and the modelled re-
sult for Pmax = 1.2 GPa and considered maximum spatial frequency of
30 microns is good. The selection of maximum spatial frequency is not par-
ticularly significant. The stiffness of the interface is dominated by the con-
tact of the taller peaks; the small wavelengths (high frequencies) have little
contribution to stiffness. It is clear that significant plasticity is taking place,
and that this is modelled reasonably well by the approximate procedure
adopted.
226
li-fU,-ll,lk,---;fl-;--~-~-;---~~i~ -
8 Pnom = 0.24 GPa - - At = 8 ~m. elastic case
-1--!J-l-.
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
8
- - A t = 30 ~m, elastic case
~6
8:.
4
Q.2
~ 0~-4~~-.~~-.~~~~~-.~~~~~~L-~
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
- - At = 30 ~m, Pma, = 0.9 GPa
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6

x [mm]
Fig. 4: Computed nonnal pressures for the elastic and approximately plastic case
+0.02 , - - - - - - - - - - - , - - - - - - - - - - - - - ,
undeformed
+0.01 deformed after full loading with Pnom =0.333 GPa
E
oS +0.00
N
-0.01
-0.02 ' - - - - - - - - - - - - ' - - - - - - - - - -....

-0.75 0.0 0.75
x [mm]
Fig. 5: The original and defonned shape of the rough surface

___ measurement loading -e- measurement unloading
..... - A. f " 16 ~m, elastic case --0-- ).., '" 30 ~lm, elastic case
--+- A-,'" 16 IJom, P,-, .. 1.2 GPa, loading
--,..--). ,=30 ~m. p....... 1.2 GPa, loading
--V--A,= 30 ~m, p..... = 1.2 GPe, unloading --II.--A.,= 16 j..im, Pm.. :: 0.6 GPa, loading
---.t.--),., = 30 j..im, P"",. =0.6 GPa, loading
2.00
E:1. 1.75
1.50
CO
a.
Q, 1.25
''c:""
Q)
1.00
0.75
~
Q) 0.50
.g" 0.25
~
.5 0.00
0.0 0.4
0.1 nominal prgJsure [GPa] 0.3
Fig. 6: Stiffness detennined from the numerical model - different maximum pressures and
maximum spatial frequencies.
The ultrasonic method gives only interface stiffness, which does not
uniquely determine the real area of contact. This interpretation of the results
using a numerical model provides a way of calibrating the experimental ap-
227
proach. Figure 7 shows the predicted real area of contact and local pressure
for the model with plasticity parameters chosen to best-fit experimental
stiffness measurements.
1.0 1.0
0.9
A,= 30 ~m, p.... = 1.2 GPa 0.9
0.8 __ real area of contact 0.8
: : 0.7 0.7~
III
~ 0.6 0.8£1.
E
c:
0.5 0.5 ~
8 0.4 0.4 I!!~
a.
'00 0.3 0.37ij
I!!
""~ 0.2 0.2
0.1 0.1
M M
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
nominal pressure [GPa)
Fig. 7: Real area of contact and real pressure as determined from the numerical model.
5. Conclusions
(i) A numerical model has been developed to predict the real area of contact
between rough surfaces undergoing elastic-plastic deformation. Plastic-
ity is modelled approximately using a limited pressure and surface flow
algorithm.
(ii) Experiments have been performed to investigate some rough surface
contacts using ultrasonic reflection. The method can be used to deter-
mine the stiffness of the rough surface contact.
(iii) The model and experiments agree well for appropriate selection of plas-
ticity parameters. This suggests that this simple approach to plastic flow
gives a realistic approximation.
(iv) The interpretation of stiffness measurements with a numerical model
provides a method for determining real area of contact.
6. Acknowledgements
The authors would like to thank: Dr Bruce Drinkwater and Mrs Amy
Quinn of the University of Bristol who carried out the experimental work.
Thanks also to Professor Knothe of the Technical University of Berlin who
organised the exchange programme which facilitated this joint research
work.
228
References
Bucher, F. H., Theiler, A., Knothe, K., 2000, ''Nonnal and tangential contact problem of sur-
faces with measured roughness", Contact Mechanics and Wear of Rai/lWheel Systems, 5th
International Conference Tokyo, Japan, 25-28th July 2000, to appear in: WEAR
Dwyer-Joyce, R. S., Drinkwater, B. W., and Quinn, AM., (2001), "The Use of Ultrasound in
the Investigation ofRough Surface Interfaces", ASME Journal ofTribology, Vol. 123, pp.
8-16.
Drinkwater, B. W., Dwyer-Joyce, R. S., and Cawley, P., 1996, "A Study of the Interaction
between Ultrasound and a Partially Contacting Solid-Solid Interface", Proc. Royal Soc. A
London, Vol. 452, pp. 2613-2628.
Greenwood, 1. A. and Williamson, 1. B. P., 1966, "Contact of Nominally Flat Surfaces",
Proc. Royal Soc. A London, Vol. 295, pp. 300-319.
Kalker, J. 1., "On the rolling contact of two elastic bodies in the presence of dry friction", PhD
thesis, Delft.
Kendall, K. and Tabor, D., 1971, "An Ultrasonic Study of the Area of Contact between Sta-
tionary and Sliding Surfaces", Proc. Royal Soc. A, Vol. 323, pp. 321-340.
Pullen, 1. and Williamson, J.B.P., 1972, "On the Plastic Contact of Rough Surfaces", Proc.
Royal Soc. A, London, Vol. 327, pp.l59-173.
Tattersall, H.G., 1973, ''The Ultrasonic Pulse-Echo Technique as Applied to Adhesion Test-
ing",1. Appl. Phys. D6, pp. 819-832.
Venner, C. H., Lubrecht, A A, 2000, "Multilevel Methods in lubrication", Elsevier
Wang, G., 1991, "Rollkontakt zweier viskoelastischer Walzen mit Coulombscher Reibung",
PhD thesis, TU Berlin
Webster, M. N. and Sayles, R. S., 1986, "A Numerical Model for the Elastic Frictionless Con-
tact of Real Rough Surfaces", Trans. ASME, 1. Trib., Vol. 108, No.3, pp. 314-320.
PREDICTION OF WHEEL WEAR FOR RAIL
VEHICLES - METHODOLOGY AND
VERIFICATION
Tomas lendel
Div. ofRailway Technology, KTH
S£-10044 Stockholm, Sweden
Mats Berg
Div. ofRailway Technology, KTH
S£-/0044 Stockholm, Sweden
Abstract This paper presents a simulation tool for systematic prediction of consecutive
wheel wear of rail vehicles. The methodology includes steps like load
coJlective design, vehicle-track interaction simulation, wear modelling, wheel
profile updating and in-field verification.
1. Introduction
In the railway field the vehicle wheels wear when they run on the rails.
The degree of wear depends on horizontal curve radii, type of vehicle
bogies, lubrication etc. Wheel (and rail) wear is not only a maintenance
problem but also significantly affects the vehicle dynamic performance in
terms of safety against derailment, dynamic stability, ride comfort etc.
It is therefore of interest to investigate how wheel rim shapes (profiles)
change due to wear over the time for various railway applications. In
particular, simulation of such consecutive wear has a great potential. Such
simulation tools are rare today, but see for instance Pearce and Sherratt
(1991) and Li and Kalker (1998), partly due to rather long simulation times
but also due to a lack of methodology and verification. In this paper both a
methodology and in-field verification are presented, see also lendel (2000).
229
J.A.C. Martins and M.D.P. Monteiro Marques (eels.), Contact Mechanics, 229-236.
230
2. Methodology
The methodology of the present wear simulation tool is illustrated in

Figure 1. Studying a single vehicle wheel we start from a given wheel
profile, e.g. the nominal unworn one. Then a so-called load collective is
designed giving a limited set of dynamic time domain simulations
representative for the wheel's contacts with the rails. In the load collective
the railway network is primarily discretized based on the curve radius
distribution (design geometry). Other key parameters are rail profiles, track
irregularities and wheel-rail coefficient of friction. See also Jendel (2000).
design geometry
Track data: { rail profiles
track irregular.
Coefficient of friction
Braking and acceleration
Vehicle traffic conditions
1----------f"IIIr""""""'1 Vehicle-track interaction model

Wheel-rail contact response
Wear model, wear data

Environmental parameters
'------~ Desired distance attained?

No (new wear step) I
• Yes (finished)
Figure 1. Flow chart of the proposed wheel profile wear prediction tool.
The vehicle-track simulations are based on multibody models of the

vehicle and track and, in particular, a proper modelling of the wheel-rail
contact. As output the simulations give wheel-rail contact forces and motions
at certain time instants. See Section 3 below.
Then, these quantities are input to the wear modelling and calculations.
The calculated wear distribution is smoothed and the wheel profile updated
due to material removal. See Section 4. Since the actual wheel profile affects
the contact response, the successive change of wheel profile must be divided
into several wear steps, as implied by the loop in Figure 1.
Section 5 shows an in-field verification of the present simulation tool.
231
3. Wheel-rail modelling and vehicle-track simulation
The wheel-rail contact modelling is crucial since we are interested in

wheel (and rail) wear. Still, the computational effort must be reasonable
since the present wheel-rail model should be used in extensive vehicle-track
simulations.
First the geometrical wheel-rail interaction needs to be described. Figure
2 shows an example of how a wheel profile and a rail profile match each
other. The straight lines connect pairs of wheel and rail points (surfaces) that
will come into contact. Each lateral wheel-rail displacement, Ay, gives a
straight line.
In this transverse plane the wheel radii span from, say, 15 mm to very
large radii. For the rail the radii are usually in the span of 13-300 mm. The
wheel-rail contact area is usually in the span of 1-2 cm2• In Figure 2 wheel-
rail contact elasticity is considered.
'"?' I
~
:"
I
~ I
dy006. :-' ,d1 ·6.
dyo·2.
dyoo.
I
I Wheel tread
I I
I
I
I
i I t£
lI!!I~ VI/II/~'./II/!& 7/:i r.-r-.

I
'32. -2~. 13 4 • 4 , 56
WI/~ //I, 1/// / p/

~ "'~
"
~ t --- ~ Wheel
I Rail top flange
i\
~
11"< , -
~-
i I ' I
II
I I
I I
~
;-
Rail gauge
II
comer "-
I
I I
I
I
f'--
I
~ I
i
I
I I
I
I
I
I
I
I
I
I ~ I
Figure 2. Example of wheel-rail match: nominal VIC S I 002 wheel profile and nominal
UIC60 rail profile, inclined 1:30. dy =£\y. Units in mm. Results from the MBS software
GENSYS, DEsolver (2000).
232
Then we need to model the normal and tangential contact problems. Here
Hertzian theory respectively Kalker's simplified theory, Kalker (1990) and
the software FASTSIM, are used. This choice implies, among other things,
that the contact area is elliptical and small compared to the overall size of the
bodies and their radii in the contact vicinity. For wheel tread - rail top
contacts, the chosen contact model is good. However, for wheel flange - rail
gauge comer contacts the choice may be more questionable. But since we
will include thousands of contact problems for each wear step, cf. Sections 2
and 4, an averaging effect will appear. Methods like BEM (software
CONTACT, Kalker (1990)) and FEM are not realistic in the present context
with today's computers.
The chosen wheel-rail contact model as well as multibody models of the
vehicle and track at hand, define a dynamic model for vehicle-track
interaction simulations. Note that due to the "stiff' wheel-rail contact, time
steps are usually only about I ms. This means that even a single simulation,
e.g. describing the vehicle negotiating a curve of a certain radius, is not a
simple computational task. The present vehicle-track simulations are
performed with the multibody dynamics software GENSYS, DEsolver
(2000). See also, for instance, Andersson et al. (1999) for general
information on rail vehicle dynamics.
4. Wear modelling and wheel profile updating
In principle, the wheel-rail contact response may be obtained as output

data for each time step in the simulations. This will give output at least every
0.1 m of travelled distance. Ifwe want to study the wheel wear development
during 200000 km of travel, like in Section 5, the load collective design may
reduce the simulated distance to about 1000-2000 km, but still the output
step may be extended.
Parametric studies have shown that, at least for the example in Section 5,
an output step of about 3 m or one wheel revolution is sufficient. This further
simplification makes it computationally realistic to study the contact
response in more detail to allow for wear calculations.
Here Archard's wear model, Archard (1953), is used as basis for the wear
calculations. This model suggests that the wear volume is proportional to the
product of normal contact force and sliding distance. An analogous model is
used here and related to the wheel-rail quantities of Figure 3.
According to Archard, wear only occurs in the slip zone of the contact
area in Figure 3. By means of FASTSIM, and at each output step, the slip
zone is divided into small elements ~~ll for which the longitudinal and
lateral (rigid) slip velocities, v~ respectively vll , can be determined.
233
Figure 3. Definition of quantities in the contact ellipse for wear calculations. The vehicle
travels in the positive ~-direction.
It can be shown that the wear depth ~S of an element is then given by
~s = kp(ru)/H (1)
where k is the wear coefficient, p the contact pressure (here given by the
Hertzian solution) and H the hardness of the softer material. ru is the sliding
distance experienced by a material point on the wheel profile when passing
through an element. This sliding distance is given by
(2)
where VsJip is the magnitude of the slip vector VsJip, M the time it takes to pass
through the element, and Vvehic/e is the vehicle speed.
The wheel wear due to one wheel revolution is found by adding the
contributions from the elements along each longitudinal strip as implied in
Figure 3.
The wear coefficient k in (1) is generally a function of slip velocity,
pressure, temperature, contact environment etc. Laboratory tests have been
made on wheel and rail steels to determine the wear coefficient, Telliskivi
and Olofsson (2001). The tests were performed under dry, room temperature
conditions for various slip velocities and pressures. Both disk-on-disk and
pin-on-disk machines were used in order to cover a wide range of velocities
and pressures. Results from these tests, and measurements on similar steels,
gave the wear chart according to Figure 4. As can be seen from Figure 4 the
wear chart has been divided into four different regions, with the most severe
wear occurring in the upper region withp>0.8H.
234
k] = 300-400' 10-4
0.8H
k2 = k4 =
k3 = 30-40 .10-4
1-10,10-4 1-10,10-4
Vslip (m/s)
o 0.2 0.7
Figure 4. Wear chart for wheel and rail steels. Wear coefficient k for different regions of slip
velocity and pressure. Dry, room temperature conditions. Telliskivi and Olofsson (2001).
The span for each wear coefficient, k, to k4' implies scatter in the
measurements. Still an average k-value has been chosen for each region.
Note that wheel tread contacts usually occur in the lower left comer, whereas
wheel flange contacts can be found in all four regions.
The effect on the wear coefficient due to natural lubrication (weather
etc.) and man-made lubrication, on the high rail in curves with radii less than
600 m in the application of Section 5, has been estimated based on different
field measurements. In this way the coefficients of Figure 4 have been
reduced. Laboratory tests for lubricated conditions will soon be carried out.
When the wear in a wear step - corresponding to, say, 1500 km of travel
or a maximum wear depth of 0.1 mm - is determined, the wear distribution
along a wheel profile must be smoothed somewhat to avoid difficulties in the
following wear steps. The smoothed wear distribution is subtracted from the
wheel profile giving an updated wheel profile, which in tum needs some
additional smoothing. A cubic spline interpolation algorithm has been used
for both smoothings and tested out carefully.
5. In-field verification
The present wheel wear simulation tool has been verified by comparing
simulated and measured wheel wear on a commuter vehicle trafficking the
commuter railway network in Stockholm. A vehicle XlOB, with two non-
powered and rather track-friendly two-axled bogies, has been selected for the
main comparisons. The network comprises about 200 km of track and has
many curves with radii in the span of 300-2000 m. A load collective have
235
been designed with seven curve radii and three rail profiles (small, medium
and large wear). The measurements have been carried out by Nilsson (2000).
The present vehicle and track have been properly modelled and the MBS
software GENSYS, DEsolver (2000), has been used for the simulation. The
well-known software MATLAB has been used to administrate the entire
simulations, cf. Figure 1.
Figure 5 shows an example of the simulation-measurement comparison.
Here we can see how a start wheel profile, with a straight tread, is worn after
200000 km of travel. Simulated and measured wheel profiles agree well.
Scalar wheel wear measures such as flange thickness, flange height, flange
inclination and decrease in cross-sectional area can also be shown to give
good agreement.
-10
-20
-30
~------~------~------~----~~----~~----~
-60 -40 -20 o 20 40 60
10.-------,-------,-------,-------,-------.-------~
8
6
4
2
o
_2L-------L-------~------~------~------~------~
-60 -40 -20 o 20 40 60
Figure 5. Simulated and measured wheel profile (top) and wheel wear (radius decrease,
bottom) after 200000 km running distance for Stockholm commuter vehicle XIOB-3149.
Simulation (solid line), measurement (dashed line), start profile (dash-dotted line, in top
figure only). Simulated results are for wheel set I and 4. Measured results are for the left
wheel of wheel set I. Units in mm.
236
6. Concluding remarks
A wheel wear simulation tool, based on a systematic methodology and

verified by in-field measurements, has been proposed. Even though the
simple wheel-rail modelling of Hertz and FASTSIM has been used, it is
shown that good agreement between simulated and measured wheel wear
can be obtained.
The wheel wear prediction tool can also be utilised to estimate
maintenance intervals, design nominal wheel profiles and optimise vehicle
suspensions. The time and costs for wheel profile measurements can also be
reduced.
Acknowledgments
The financial support by Adtranz Sweden, Banverket, SJ, SL and

NUTEK is gratefully acknowledged.
References
Andersson, E., Berg, M., and Stichel, S. (1999) Rail vehicle dynamics - fundamentals and
guidelines, Compendium, Div. of Railway Technology, KTH, Stockholm.
Archard, 1. F. (1953) Contact and rubbing of flat surfaces, Journal of Applied Physics 24,
981-988.
DEsolver (2000) GENSYS User's Manual, Release 0003, Ostersund.
Jendel, T. (2000) Prediction of wheel profile wear - methodology and verification, Licentiate
Thesis, Report TRITA-FKT 2000:49, Div. of Railway Technology, KTH, Stockholm.
Kalker, 1. 1. (1990) Three-dimensional elastic bodies in rolling contact, Kluwer Academic
Publisher, Dordrecht.
Li, Z. L., and Kalker, J. J. (1998) Simulation of severe wheel-rail wear, Computers in
Railways, 393-402.
Nilsson, R. (2000) Wheel and rail wear - measured profile and hardness changes during 2.5
years for Stockholm commuter traffic, Railway Engineering 2000, London.
Pearce, T. G., and Sherratt, N. D. (1991) Prediction of wheel profile wear, Wear 144, 343-
351.
Telliskivi, T., and Olofsson, U. (2001) Wear and plastic deformation of two rail steels - full
scale test and laboratory study, 2nd World Tribology Congress, Vienna. To appear.
MULTI-BODY MODELING OF PAPER
CALENDERING UNIT BY CONTACT DYNAMICS
FORMULATION
Emo Keskinen
Tampere University of Technology, Laboratory ofMachine Dynamics
P.O. Box 589, FIN-33JOJ Tampere, Finland
Sirpa Launis and Juha-Matti Kivinen

Tampere University of Technology, Laboratory ofMachine Dynamics
P.D. Box 589, FlN-33JOJ Tampere, Finland
Abstract A multi-body contact dynamics fonnulation is introduced to model the

dynamic motion of roll mechanisms typical for paper manufacturing systems.
Contacts between rolls manipulating the paper web are modeled as physical
contact line loads in nonnal pressing and tangential traction directions. In the
case study the new loading mechanism of a laboratory calender has been
dynamically tested by means of system level simulations.
1. Introduction
Multi-body contact dynamics (MCD), Keskinen (1994), is a formulation

that connects the movements of machine parts together using physical
contact forces while most of the existing algorithms eliminate joint forces,
Shabana (1998), or handle them as constraint forces for satisfying
kinematically the conditions of connectivity Haug (1989). In MCD one
operates with absolute link position coordinates to get one differential
equation group for link movements and another explicit algebraic equation
group for physical contact forces acting in joints. A representative example
of the application ofMCD is a 'Japer finishing unit, in which two cylindrical
237
238
Figure J. Mechanical lay-out of industrial calender unit.
shells are in rolling contact for calendering, see Smook (1992), the axially
moving web material. In calendering the paper web is manipulated between
two compressed rolls in the contact zone called nip. In order to extend the
nip manipulation time the rolls are covered by polymer layer providing a
wide contact zone.
2. Roll dynamics
Paper machine rolls as circular tubes behave like elastic cylindrical

shells. As the length of rolls is much bigger than the diameter a good
approximation for thick walled rolls is to describe the bending motion using
the theory of an elastic beam. The kinematics of roll cross-section locating at
axial coordinate point z is illustrated in figure 2a.
a) b)
Figure 2. Angular degrees of freedom related to roll spin motion (a) and orientation (b).
239
Roll vibration is described in local rotating body coordinate system while

the rigid body motion is fixed to the global stationary frame. Under these
assumptions the material point locating originally at roll surface in
coordinate point (r, e!>, z) has a fmallocation (X, Y) evaluated by vector
(1)
where s=(u'vl is the rigid body position of the roll c.o.g. Angles Q x and
Q y are small roll orientation inclinations around global coordinate axis, as
illustrated in figure 2b. Vector s=(u, v/ is the bending displacement vector of
roll cross-section and vector er=(e(e!»,s(e!»Yhas elements
e(e!» = eDse!> , s(e!» = sine!> . Matrix
A = [e(e) -s(e)] (2)

s(e) e(e)
is the coordinate transformation matrix from local to global system, in which

e is the roll angular position in the spin motion. As roll ends are mounted on
moving arms bending motion is expanded instead of Ritz-like assumed
modes in terms of a complete set of free-free eigenmodes Meirovitch (1967)
u =u(t/ g(z) ; v =v(t/ g(z) (3)
a) b)
Figure 3. Roll loaded by line loads (a) and roll interaction forces in calendering (b).
240
Suppose now the roll is loaded on its outer surface by a set of distributed line
loads fi.(z) = pJz)er(Vi)+qJz)e'(Vi) and line moments t;{z) together
with concentrated bearing reaction forces Fj acting on the roll ends.
A relatively long but straightforward equation derivation, see Keskinen
(200 1), produces the rigid body equations of motion in form
e/2
MU= L f [C(Vi)Pi(Z)-S(Vi)qi(Z)}iZ+ LC(Vj)Fj
i_~ j
e/2
MV= L f ~(Vi)P;(Z)+C(Vi)qi(Z)}iZ+ LS(Vj)Fj -Mg
i -e/2 j
e/2
IQ x =-L f [S(Vi)Pi(Z) + c(Vi)q;(z)]zdz- LS(Vj)Fjz (4)
i_~ j
e/2
IQ y =L f [c(V;)p;(Z)-S(Vi)qi(Z)]zdz+ LC(Vj)Fjz j
i -e/2 j
e/2
I i = L J[q/z)R + Ii (z)}tz + T
i -e/2
where they are dynamically decoupled from the bending equations of motion
e/2
mii-2mev+(k-e 2m)u= L J[C('IIi -9)Pi(Z)-S('IIi -9)qi(Z)g(z)dz
i -e/2
+ LC('II j )Fjg(z j)
j
(5)
e/2
mv + 2meu+(k-e 2m)v = L J[S('IIi -9 )Pi(Z)+C('IIi -9 )q;(z)g(z)dz
i -e/2
+ LA'll j )Fjg(z j)
j
e/2 e/2
with m=pAfg(z)g(z)Tdz and k=El x f g"(z)g"(zldz . Eqs. (4)
-e/2 -e/2
and (5) have similar uncoupled structure than the ones obtained for a
vibrating unrestrained airplane structure in Bisplinghoff et al. (1955), p. 114.
This unrestrained body formulation is new in the dynamics of elastic rotors.
241
3. Nip dynamics
The unit vector from center of roll A to center of roll B is
(6)
where d ( Z) = ~( RB - RA)( RB - RA) is the distance between roll centers.

The global angular positions of contact point are then 'I' A = Arc tan (E 2/ E 1)
and 'I' B ='1' A +1t corresponding to local angular positions <\> A ='1' A -0 A and
<\>B ='1' B -OB . Nip penetration is a nonnegative quantity
E(Z) = R A( z) + R B( z) + s + h(z,t) + Il(Z'<\>A) - d(z) (7)
where h(z,t) is the thickness variation of paper web entering the nip, s is the
thickness of roll cover and Il is the out-of-roundness profile of permanent
cover deformation. The rate of penetration is now
E( z) = h( z, t) - ell'( z '<\> A) - d(z ) (8)
where the rate of roll distance is d = (RB - RA ). E . Nip shear motion

associated with tangential nip traction is
't( z) = (R A + s)3 A + RBa B

(9)
- s(1» A )u~g( z) + c(1» A )v~g( z) - s(1» A )u~g( z) + c(1» A )v~g( z)
Nip force distribution in normal direction gets form
n(z)=aE(Z/ +C~E(Z)E(Z) (10)
If one of the rolls, say roll A, has been covered by soft polymer of thickness
s and characterized by its elastic and viscous moduli E and 11 corresponding
242
to the classical Kelvin-Voigt material model, the theory of elastic
foundation, Johnson (1985), is available giving contact parameters
i
a = E ~ ; b =!.- ; c =
3 S v~ 2 S v~
2rlf!o
with 2~ =(RA + S 'j1 + RB-1.
Tangential traction force may be evaluated by equation
where the maximum possible sticking shear deformation has expression

't max (z) =3Jl/2b ·e{z). It can be shown also that rolling resistance between
the rolls is generating torque distributiont(z) = -1lE- 1n {z)(R A + s)8 A •
4. Case study
A new calender configuration Kivinen (2001) includes machine frame

and two rolls of which the upper one is mounted on the frame while the
lower one is fixed on to the hydraulically actuated loading arm (see Fig. 4a).
b)
Figure 4. a) Main elements of test calender: Frame I, loading ann 2, lower roIl 3, upper roIl 4,
locking ann 5, loading cylinder 6, damping cylinder 7, locking cylinder 8, loading linkage 9,
load ceIl 10 and. b) System diagram of multi-body simulation model.
A complete simulation model of the unit is based on system diagram 4b.

243
Observations show that periodical thickness variations of paper web

excite the rolls to execute relative oscillatory motions in the normal direction
of the contact plane. In the case study the system responses of a unit with
f = 4.4m have been solved for two different fixed <p values, namely 10°
and 20°. The response curves of the relative roll approach during nip
closing maneuver are plotted in Fig. 5a. The fluctuations of the same
quantity are shown in Fig. 5b when periodic paper thickness variation of
amplitude 1.lO-5 m is exciting the roll nip over frequency range 25 ... 75 Hz.
e [mm] 0.1.----,----,-----,--r===:===ll
o :~~~
··············j················T················[·····...
-0.1 . \...j..... ..+..........;. ........... ;
=
-0.2
\:
····\·j""'· ......·..·t:. . ·. ·. . ·!"". ..·...... :·. ..· .. ··
~: -=~m---r--
-0.6 ............. -!--.......... + .........: ....... j
o 2 4 6 8 10
t (s]
a)
e[mm]-0.5r-----,----:---,-r===~
10 12 14 16 18 20
t (s]
b)
Figure 5. Relative roll approach during nip closing maneuver (a) and under periodic paper
thickness variation (b).
The calculated responses show that the new loading mechanism is able to
limit the roll vibrations in the excitation amplitude level. Numerical time
integration of equation system gave satisfactory results for time step 1([4 S.
244
5. Conclusions
A two roll paper finishing unit has been modeled using multi-body
contact dynamics approach. This work contains the development of a roll
element that can be flexibly connected to mechanism links, bearings, drives
and hydraulic loading actuators. The complete system model consists of
dynamic equations of motion for the multi-body system. Algebraic contact
equations represent the coupling of neighbouring rolls in terms of contact
force components depending explicitly on the relative positions of the roll
surfaces. Dynamic excitation arising from periodic paper thickness variation
makes the system to vibrate the effect of which has been investigated by
means of a new kind of nip loading mechanism. The results show that the
new modeling approach is able to describe the contact dynamical
phenomena of cylinders in rolling contact. The system model has been
utilized in the design of the new calendering mechanism.
References
Bisplinghoff RL, Ashley H. (1962) Principles ofAeroelasticity. New York: Dover.
Bisplinghoff RL, Ashley H., Halfman RL, (1955) Aeroelasticity. New York: Dover.
Haug EJ. (1989) Computer-aided kinematics and dynamics of mechanical systems. Volume I:
Basic methods. Boston: Allyn and Bacon.
Johnson KL. (1985) Contact mechanics. Cambridge: Cambridge university press.
Keskinen EK. (1994) A contact dynamics formulation for distributed simulation of flexible
hydromechanical systems. Ph.D. Thesis. Tampere: Tampere university of technology
publications 134.
Keskinen EK. (2001) Multi-body contact dynamics modelling of roll mechanisms in paper
manufacturing systems. Proc. of ASME International Mechanical Engineering Congress
and Exposition, Symposium on the Vibration and Control of Mechanical Systems, Nov
11-16,2001, New York.
Kivinen J-M. (2001) A variable parameter facility for dynamic testing of polymer covered
paper machine rolls. Ph.D. Thesis. Tampere: Tampere university of technology
publications 347.
Lubkin JL. (1962) Contact problems. In: Fluegge W, editor. Handbook of engineering
mechanics. New York: McGraw-Hill.
Meirovitch L. (1967) Analytical methods in vibrations. New York: Macmillan.
Shabana A.A. (1998) Dynamics of multibody systems. Sec. ed. Cambridge: Cambridge
university press.
Smook GA. (1992) Handbook for pulp and paper technologists. Second ed. Vancouver:
Angus Wilde Publications.
EXISTENCE AND UNIQUENESS FOR
QUASISTATIC CONTACT PROBLEMS
WITH FRICTION
Lars-Erik Andersson
Department of Mathematics
Linkoping University, SE-581 83 LINKOPING, Sweden
Anders Klarbring
Linkoping University, SE-581 83 LINKOPING, Sweden
Abstract This paper deals with mathematical results of existence and uniqueness
of solutions to problems of linear elasticity involving friction, in partic-
ular for the quasistatic case when inertia forces are neglected. Both the
case of a continuum body and that of a space discrete one are treated.
The intention is to describe the state of the art for these problems.
1. Introduction
The present paper reviews results, some of them recent, concerning
existence and uniqueness of solutions for the problem of a linearly elastic
body in frictional contact with a rigid obstacle.
The review is limited to quasistatic problems where inertia terms are
assumed negligible. Both the case of a continuum body and that of a
space discrete one are treated. We also restrict our attention to so called
coercive problems, where some part of the boundary of the elastic body
has a prescribed motion, excluding rigid body motions.
The mathematical analysis of frictional contact problems is very dif-
ferent when the coefficient of friction, fJ, is small or large, respectively.
It is well known that for large fJ one cannot expect any general existence
or uniqueness results for quasistatic problems, not even for the simplest
possible case with a single particle system with two degrees of freedom.
There is a well known counterexample, due to Klarbring (1987, 1990),
which is presented in Section 2. Otherwise we focus on existence and
245
J.A.C. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics. 245-260.
246
uniqueness results which can be obtained under the assumption that fJ
is sufficiently small. Furthermore, we consider only the case of small
deformations and assume that the equations of linearized elasticity can
be used.
The rest of the article is organized as follows. In Section 2 we give
the above mentioned counterexample. In Section 3 we first give classical
formulations of the quasistatic problem for a continuous elastic system,
including the s~called normal compliance approach. By classical we
mean partial differential equations and boundary contact conditions,
assuming sufficient regularity of displacement and stress fields. Next
we reformulate the problems as a variational time-dependent problem in
appropriate function spaces and present some recent results, in particular
for the case with Signorini contact conditions.
In Section 4 we specialize to the space-discretized problems, for which
the mathematical analysis is quite different. Contrary to the case with
a continuous elastic system, it is here possible to obtain some rather
general uniqueness results, provided that one imposes some condition of
time-regularity for the exterior force-field.
2. Elementary example of non-uniqueness and

non-existence
We consider a structure, condensed into a single contact node, in
two-dimensional contact with a rigid obstacle. The external load and
previous evolution of the structure is such that the normal component
of the contact force PN < 0 satisfies PT = fJP N, where PT is the tangential
component of the contact force and fJ is the friction coefficient. That
is, the contact force state of the node is on the boundary of the friction
cone and two qualitatively different further continuous evolutions are
possible:
• the node may evolve into a stick state which means that
(1)
where PT, PN and WT are right hand time derivatives of the contact
forces and the tangential displacement;
• the node may evolve into a slip state which means that
(2)
The previously introduced time derivatives PT' PN and WT' and the time
derivative of the normal displacement, WN' are linearly coupled through
247
a stiffness equation:
[~:: Z: 1[: 1= [ ~ 1+ [ ~; 1' (3)
where kNN' etc. are real numbers such that the system matrix is positive
definite and j N and iT are time derivatives of the external forces. Since
PN < 0 it holds that WN = 0 so we may conclude from (3), by multiplying
the first equation with /-t and subtracting, that
(4)
From (1) and (4) it follows that ifthe node goes into a stick state, then
/-tiN ~ iT.
From (2) and (4) one finds that if the node goes into a slip state, the
condition on derivatives of external forces depends on the sign of the
"effective stiffness" in (4):
if -/-tkNT + krT > 0 then /-tiN ~ iT
if -/-tkNT + kTT = 0 then /-tiN = iT
if -/-tkNT + kTT < 0 then /-tfN ~ fT.
We conclude the following for the problem of determining the right hand
time derivative of contact forces and displacements when given a rate of
change of the external forces:
• if -/-tk NT + kTT > 0, then there exists a unique solution;
• if -/-tk NT + kTT :::; 0, then, depending on the direction of change
of external loading, there does not exist any solution or there are
multiple solutions.
Note that /-t > 0 and kTT > 0 while kNT may have any sign. This later
constant represent a stiffness coupling between normal and tangential
degrees of freedom, so, clearly, for large such couplings we may experi-
ence non-uniqueness or non-existence. The non-existence of right hand
time derivatives means that the quasistatic problem cannot have any
solutions that are continuous in time.
3. A quasistatic problem for the continuum case

3.1. Classical formulation of the quasistatic
frictional contact problem
We will give a classical formulation of the quasistatic contact problem
with Coulomb friction. A linearly elastic body which may come into
248
contact with a fixed rigid obstacle is considered. Two mathematical for-

mulations of the physical conditions of impenetrability and non-adhesion
are given. The first one is the Signorini-Fichera complementarity con-
dition. The second one is the so-called normal compliance law. The
u(t)lsu = ii.
Su ~"""""'rl
-divO"(u(t)) = f(t) in n
O"(u(t))ft = t(t)
Figure 1. Elastic body in contact with a rigid obstacle.
classical quasistatic formulation is obtained from the dynamic formula-

tion by neglecting inertial terms, which means that we are looking at a
sequence of equilibrium states.
The linearly elastic body occupies a bounded Lipschitz domain
in R3 (or R2). The body is subjected to body forces / = (fl, 12, fa)
°
and to prescribed tractions t = (tl' t2, t3) and displacements u, on the
parts St and Su of the boundary 00, respectively. The potential contact
boundary is Sc. Furthermore, Sc, St and Su will be mutually disjoint,
relatively open subsets of 00. The following classical equations of linear
elasticity are valid:
-+
a(Jij f i= 0 in 0, (5)
ax;
(Jij = aijkl a
aUk
Xl
in 0, (6)
(Jijnj = ti on St· (7)
u=u, on SUo (8)
Here u = (Ul' U2, U3) is the displacement vector, U = {(Jij} is the stress
tensor and nj are the components of the outward unit normal vector.
249
The elasticity coefficients aijkl satisfy the usual symmetry and ellipticity
conditions. Furthermore, i, j, k, I = 1,2,3, the summation convention is
used and (0, Xl, X2, X3) is the cartesian reference frame.
To state the laws of contact and friction we decompose the displace-
ment and traction vectors on Se into normal and tangential components:
The classical Signorini-Fichera contact law may now be stated as
UN ~ 0, UN - 9 ~ 0, UN(UN - g) = ° on Se, (9)

where 9 is the initial gap between the body and the rigid support. Note
that there is no sign restriction for g.
The friction law is that of Coulomb, which can be written as
IUTI ~ -J.tUN, and (10)

IUTI < -J.tUN ::} UT = 0, °< IUTI = -J.tUN ::} UT = -AUT, A ~ 0,
where J.t is the friction coefficient and a superposed dot denotes time
derivative.
Relations (5)-(10) constitute the quasi-static frictional contact prob-
lem. The problem is time dependent via the time derivative in Coulomb's
friction law. Therefore the external forces / and t should be taken as
time dependent.
The Signorini-Fichera conditions (9) are approximations to the be-
havior of contacting surfaces. Although these conditions are very useful
in many situations an alternative which somewhat reflects the physical
nature of contacting surfaces has been suggested and analyzed in the
literature. This is the so called normal compliance law
(11)
where CN and mN are positive parameters representing the physical char-

acteristics of the interface and (z)+ = max(O, z). Since (11) is a local
relation between UN and UN we can eliminate the fonner from the friction
law (10). However, one can also generalize the friction law somewhat in
the following way
IUTI ~ CT(UN - g)~T, and (12)

IUTI < c.r(UN - g)~T ::} UT = 0,
°< IUTI = c.r(UN - g)~T ::} UT = -AUT, A ~ 0,
250
where Or and mT are new physical parameters. One retrieves (10) from
(11) and (12) by putting Cr = I-'CN and mT = mN'
Relations (5)-(8), (11) and (12) constitute the quasistatic frictional
contact problem with normal compliance.
3.2. Existence results for quasistatic friction

problems
We will now formulate a variational problem corresponding to the
classical formulation (5)-(10) in Subsection 3.1. Using Green's formula,
we are led to the following time dependent variational problem, con-
sisting of two simultaneous inequalities (for additional notation, see the
appendix).
VPQ: Given / E WI(O, Tj (L2(O))3) and t E WI(O, Tj (HI (80))3), find
a mapping U E WI (0, Tj K) such that for almost all t E [0, T) we have
a(u,l1 - u) - ('f/;cUN(U),VN - UN)
-('f/;cl-'UN(U), I11TI-luTI)) ~ (/,11 - u) + (t,l1 - u) (13)
for all 11 E V and such that
('f/;cUN(U), WN - 'UN) ~ ° (14)
°
for all t and all w E K. Here 'f/;c E Cr(Rn ) is a cut-off function which
equals 1 on a neighbourhood of Sc and on a neighbourhood of Sa U Su.
For the initial state u(O) = Uo we have the following necessary com-
patibility condition
a(Uo,l1 - uo) - ('f/;cl-'UN(U), I11T - uOTI)
~ (/,11 - uo) + (t,l1 - uo) (15)
for all 11 E K. Formally, the latter condition expresses that the initial
state is in equilibrium, satisfies the contact law (9) and that IUT(UO)I ~
-I-'UN(UO) on Sc·
Cocou et al. (1996) gave an existence result for a regularized version
of this problem, when UN is replaced by u~, where * denotes a bounded
linear mapping
It was also assumed that the coefficient of friction is chosen sufficiently

small. We may also note that, in their work, I-' is allowed to be time
dependent. Their proof is based on an incremental formulation of the
problem VPQ and employs a technique for passing to the limit which
251
is similar to that in Andersson (1991), where a quasistatic problem with

normal compliance was continued.
Another way of regularizing VPQ is by normal compliance. The
appropriate variational formulation is then
VPQNC: Find '1£ E Wl(O, T; K) such that for almost all t we have
a(u,v - ti) + CPnc(u,v- ti) + inc ('1£, v) - inc(u,ti) ~ (F,v- ti) (16)
for all v E V.
Here we have introduced the notation CPnc( '1£, v) = Isc CN (UN -g )~N UN dS
and inc ('1£, v) = Isc CT(UN - g)~ IVTI dS, and (F, '1£) = In J . v dx +
fS t t . v dS.
The problem VPQNC and variants thereof was investigated by Klar-
bring et al. (1988, 1989, 1991).
In Andersson (1999a) the problem VPQNC was treated with mT =
mNand
and
inc ('1£, v) = A r Il( UN -
lsc
g)~N IVTI dS
where A is a, possibly large, penalization parameter. It was shown that

if
CO 1 Co 1
1I1l1i00 < Co IItrolillEoll and IIIlIl < Co IItrllIlEII' (17)
then there exists a solution '1£ of VPQNC (for notation see the ap-
pendix). Further it was shown that
Ilti(t)1I ~ C(lIi(t) II + IIt(t)ID (18)

for almost all t, with the constant C independent of the penalty param-
eter A. We note also that the upper bounds on the norms of Il given
in (17) are also independent of A.
Using the uniform (in A) estimates (18) and the shifting technique of
Fichera the following result was proved in Andersson (2000).
Theorem 1 Assume that the assumptions of regularity etc. for 0, Sc,
Su, aijkl and g given in the appendix are valid and that 0 < a < f3 ~
1, a ~ 1/2. For the initial state '1£0 we require that O"N(UO) = 0 on
Sc. Also assume that Il satisfies {17}. Then VPQ has a solution '1£
satisfying an inequality {18} almost everywhere, with the constant C
depending only on IIIlIl and 1I1l1i00. Further, on Sc we have the regularity
O"N(U) E H lod2+0 (Sc).
252
Later Rocca (20(0) gave a slightly different version of the same the-
orem, valid for the more general initial condition (15). Rocca used a
time-incremental version of VPQ, and with the aid of similar a priori
estimates as in Andersson (2000), he proved existence by taking limits
as the time-step tends to zero.
4. A quasistatic problem for a system with

finitely many spatial degrees of freedom
In this section we will consider the question of existence and unique-
ness for the quasistatic problem with finitely many spatial degrees of
freedom. The counter-example presented in Section 2 shows that for
large coefficients of friction we cannot expect any general existence or
uniqueness results for this problem. Furthermore, Ballard (1999) has
given a remarkable counter-example to uniqueness for a 3 DOF system
with a single contact node confined to a half-space. His counter-example
is valid for an arbitrarily small coefficient of friction. The applied force in
his example is in W1,OO{0, Tj R 3) but has an oscillating direction close to
the bifurcation point. Therefore the question remains whether one can
obtain some uniqueness results if additional time-regularity is required
of the applied force-field.
In the following subsections we will review the recent work of Ander-
sson (1999b). An upper bound ii for the coefficients of friction which
guarantees the existence of a solution is provided, and also a rather gen-
eral uniqueness result for forces which are right piece-wise real analytic
in time is given. Only the coercive case is treated.
4.1. Formulation of the problem

We consider a particular type of discrete structure which is com-
posed in such a way that its displacement state can be represented by
three-dimensional (or two-dimensional) geometric displacement vectors
associated with non-coinciding points in the physical Euclidean three-
dimensional space. These points are called (displacement) nodes. There
are two classes of nodes: the contact nodes which upon deformation may
come into frictional contact with rigid obstacles, and the rest, which are
called interior nodes. There are l contact nodes. From the displace-
ment vectors Ui and the reaction forces ri, 1 ~ i ~ I, associated with
these nodes, we form a global contact displacement vector and a global
253
fixed nodes
external forces
contact nodes and ob tacles

Figure 2. Triangulation of a plane elastic body, resulting in the discrete quasi-static
friction problem.
reaction force vector
Similarly, the displacement of the interior nodes are collected in a vector

UJ. External prescribed forces are denoted by F = [ 'l ], f acting on
contact nodes and II
on interior nodes respectively.
For a linear elastic structure under a small displacement assumption
and ignoring inertia we have the following stiffness relation
[f II]
+r = [Kll
(KJ)t (19)
where
Kll KJ]
[ (KJ)t K '
is a symmetric stiffness matrix which, if the structure is sufficiently an-
chored, is positive definite and a superscript t denotes the transpose of
254
a matrix. Solving equation {19} for u we obtain the structural equation

in flexibility form:
u=Lr+MF, {20}
where L is a symmetric positive definite flexibility matrix and M a ma-
trix. In the following we will be interested in time dependent problems.
The external forces F{t} are then prescribed functions oftime.
Collecting the structural equation, the non-penetration contact con-
ditions and Coulomb's friction law we then have the following space-
discrete quasistatic evolution problem. Here u+ denotes the time deriva-
tive to the right and displacements and contact forces are decomposed
into tangential and normal vectors: Ui = UiT + UiNni, UiT . ni = 0,
ri = riT + riNni, riT· ni = 0, where a central dot indicates the standard
scalar product of geometric vectors and ni is a normal vector of obstacle
i.
DQP For a given exterior force vector F = F{t}, find displacements
U = u{t} and reaction forces r = r{t} satisfying {20}, Signorini contact
conditions:
{21}
and Coulomb's friction law:
UiN = 0 =} IriTI ~ -Jl.iriN,

o< IriTI = -Jl.iriN =} it.t;. = -Ai{t)riT' Ai{t} ~ 0,
IriTI < -Jl.iriN =} it.t;. = o.
By taking the right-hand time derivative of the inequalities and equa-

tions of DQP, it is possible to derive a so-called rate problem which,
given an initial state, concerns finding right-hand time derivatives of
unknown variables. This rate problems turns out to be a Nonlinear
Complementarity Problem, which we denote NLCP. In this way it is
possible to conclude that the full problem DQP can be divided into two
subproblems:
• Rate Problem: Given rand F, is there a unique solution f+ to
the NLCP, so that we may write
f+ = :F{r, F} ?
• Integration Problem: If so, can we integrate this integral equa-

tion for a {unique?} solution r{t} with
f+{t) = :F{r{t), F{t)) ?
255
4.2. A fundamental frictional parameter and
existence results
In the following we give, in the case of a small friction coefficient,
an existence and uniqueness result for the rate problem as well as an
existence result for DQP. The smallness of the friction coefficient is
expressed by a fundamental parameter which we first define:
Definition 1
cp(L, n) = max
ufO
{m~n
l~t:Sl
rpi(L, n; u)} ,
Ui.lni
jj = jj(L, n) = cot cp(L, n),

where
and Pi is the orthogonal projection on span {ui,ni}.

The parameter cp has the following simple mechanical interpretation: If
the system is acted upon by an arbitrary force vector r on the contact
nodes (F = 0) such that all displacements are tangential, then for at
least one node we have
rpi ~ cp.
In Andersson (1999b) the following theorem of existence for the dis-
crete quasistatic problem DQP is given.
Theorem 2 If r(O), u(O) satisfy the natural compatibility conditions,
F(t) is absolutely continuous and if F E £<)0(0, T), then there exists a
solution r to the quasistatic problem DQP so that, a.e.,
r+(t) = F(r(t), F(t)).
Moreover, there exists a constant C = C (J.t) so that, a. e.,
Ilr(t)1I ~ CIIF(t)lI·
The main steps in the proof of this existence theorem are the following.
• First prove unique solvability of the Rate Problem.
• Then exploit the theory of differential inclusions and differential
equations with set-valued mappings, as given, e.g., in Filippov
(1988).
For a system with one node and 2 DOF it is possible to show that
the solution given in the previous existence theorem is unique, see, e.g.,
Andersson (1999b).
256
4.3. Piecewise real analytic forces

For piecewise real analytic forces it is possible to obtain a partial
uniqueness result for the full problem DQP.
Definition 2 The mapping [0, T) 3 t I-t F{t) E R31 is right piecewise
°
real analytic (RPRA) if it is continuous and if for every to E [0, T) there
exists f > such that
L Fk{t -
00
F{t) = to)k, for t E [to, to + f).

k=O
Further, F is said to be absolutely continuous (AG) if
loT IIF{t) II dt < 00.
Right piecewise real analyticity implies that v{t) = MF{t) has at most
countably many discontinuities.
We first formulate an existence and uniqueness result for the case that
we have two degrees of freedom for each node.
Theorem 3 Assume that J.ti < fl., that F : [0, T) -+ R21 is RPRA and
AG and that the initial values r{O) and u{O) satisfy the natural compat-
ibility conditions. Then there exists a unique RPRA mapping
[0, T) 3 t I-t r{t) E R21
solving our quasistatic friction problem (for all t). Further there exists
a constant C = C{J.t) such that
for all t, i.e., r{t) is AG.
Sketch of proof:
1. For a given to construct a formal power series solution r{t)
00
E rk{t - to)k. The coefficients rk are uniquely determined by a se-
k=O
quence (tree) of NLCP(k), for k ~ 1. These NLCP:s are similar to
that for the Rate Problem, although a little more complicated.
2. Prove that the power series converges in some interval [to, to + f),
giving a local solution.
3. Repeat with new initial point.
257
The proof does not go through completely for the case with 3 DOF
at each node. However we can still construct unique formal power series
solutions
00
ri(t) = L r~(t - to)k, r:' 0:1 0

k=s,
00
Ui(t) = u? + L u~(t - to)k, u1' 0:1 0

k=q,
and convergence is obtained if for all nodes in contact we have
The exceptional case, Si > qi, should be interpreted as a case with

grazing tangential contact, meaning that (formally) if the obstacle were
removed then the velocity vector would be perpendicular to the normal
vector. We may therefore formulate the following corollary
Corollary 1 We have uniqueness for the case with 3 DOF at each node,
provided that grazing tangential contact does not occur.
Summarizing we have the following:
Existence if f..'i < ji. for all contact nodes and with "relatively arbi-
trary" external forces: FE L OO •
Existence and uniqueness if in addition
• "Arbitrary forces" , one single node, 2 DOF.

• Regular forces, F in RPRA and AC.
a Many-particle systems, 2 DOF at each node.
b Many-particle systems, 3 DOF at each node, provided that
Grazing Tangential Contact does not occur!
Acknowledgments
This work was supported by the Swedish Research Council for Engi-
neering Sciences under contract no. 223-99-676.
258
Appendix: Notation and definitions
For the subsets Sc, Sa and Su c an
introduced earlier we make the further as-
sumption that they have mutually disjoint closures.
The (Hilbert space) Sobolev norm of order p over 0. or its boundary is denoted by
II· 11",0 or II· 11",110 respectively. The same notation is also used when the functions
are vector-valued. So we write, e.g., II· II(L 2 (0»ft =
11,110,0 and II· II(H1(or=II ·Ih.o.
Similarly, the dual pairing, e.g., between the spaces H- 1/ 2 (an) and H1/ (an)
is writ-
ten (', ')-1/2,1/2(80). When no confusion is likely to appear, the sub-indices denoting
the particular spaces are omitted.
To denote the dual pairing between a Hilbert space H and its dual H' we use
(-,.) H' ,H' The inner product and the norm in H will be denoted by (',')H and II· IIH'
We also introduce a gap function 9 E H 1 / 2 an defining the initial gap between
the elastic body in the reference configuration. Note that we do not assume that
9 ~ 0, which means that the body may be pre-loaded by the obstacle in the absence
of external forces f and t.
The trace operator is denoted by
(A.i)
Further we use the same notation for the trace operator tr : (HI (0.» n -+ (HI/2 (an»n
and denote its norm by IItrll. We also use that there exists a linear bounded extension
operator
=
with norm denoted by IICII, such that troC id(H1/2(IIO»ft. It is clear that IIClilltrll ~ 1
and that an optimal value of this product depends only on the geometry of n. For
the case that 0. = R+ = {(x, Xn) : Xn > O} these mappings are denoted by tro and
Co.
Next, let us introduce the affine subspace
v = {1£ E (H1(n)t : 1£15.. = u}

ofthe Sobolev space (HI(n)t and its closed convex subset
K = {1£ E V: 1£15.. = 0 and uNls. $ g}.

Here, and subsequently, restrictions of functions are interpreted in the trace sense.
The bilinear elastic energy form is given by a(1£, v) = 10
aijkl ~ ~ dx, where the
= =
stress tensor is Uij Uij(1£) aijkl ~:~. For the coefficients aijkl ofthe elasticity ten-
sor we require, besides the usual symmetry conditions, that aijkl E Loo(n). Further
we assume that there exists a constant 00 > 0 such that
aijkl{ij{kl ~ 001{1 2 = OO{ij{ij

for all {ij such that {ij = {ji. Here we have used the summation convention. The
1
mapping
(1£, v) t-+ a(1£, v) = o
auk aVi
aijkl(X)~~ dx
UXI uXj
then defines a bounded, symmetric bilinear form on V x V. By Korn's inequality and
the boundedness of aijkl there exist constants Co = co(o, 0., Su) and Co such that for
alluEV
Coll1£ll~,o < a(1£,1£) =a(1£) < Coll1£ll~,o. (A.2)
259
For convenience we may let the non-negative coefficient of friction I' be defined on
the whole of the boundary and we assume that I' E L oo with sup-norm 111'1100'
1. Assumptions made in Subsection 3.2

For the coefficients aijlc/ of the elasticity tensor we require in addition that they
are locally in CO,P in a neighbourhood of Se i.e., that each point Xo E Se has a
neighbourhood Uo 3 Xo such that for some constant Lo and all x, y E Uo we have the
inequality
(A.3)
For the coefficient of friction I' we assume further that it is a multiplier on H 1 / 2 (8n),
i.e., that the mapping
Hl/2(8n) 3 v ~ I'v E Hl/2(OO)
is bounded with norm 111'11, so that IIl'vlh/2,8n ~ 1II'IIIIvlh/2,8n. By duality, it follows

that I' is a multiplier on Hl/2(8n) if and only if it is a multiplier on H- 1/ 2 (OO), and
that the respective norms are equal. We note that I' is such a multiplier if I' is a
Lipschitz function on 80.. For more information on multipliers and for more general
sufficient conditions that I' is a multiplier, we refer to Maz'ya & Shaposhnikova (1985).
We also assume that I' E L oo (8n) with the norm 111'1100'
For a Hilbert space H, L 2 (0, Tj H) is the class of mappings [0, T] 3 t ~ I(t) E H,
where I is weakly measurable and 1I/l1~2(o,T;H) = JOT II/(t)lI~ dt < 00.
For the set Se we assume that it is locally in C1,p for some /3 E (0,1].
Further, for the sets S ... and Se we make the assumption that their relative bound-
aries 8S... and 8Se have Hl/2(OO)-capacity equal to zero,
Here the capacity of a compact subset e c 80. is defined as
A sufficient condition for this is for example that 8S... and 8Sc are Lipschitz curves.
We finally assume that 9 E Hl1fc2+Q(Se) n Hl/2(OO).
The parameters 0: and /3 should satisfy the inequalities, 0 < 0: < /3 ~ 1 and
0: ~ 1/2.
References
Andersson, L.-E. (1991) A quasistatic frictional problem with normal compliance,
Nonlinear Analysis, Theory Methods & Applications 16,347-369.
Andersson, L.-E. (1999a) A quasistatic frictional problem with a normal compliance
penalization term, Nonlinear Analysis, Theory Methods & Applications 37,689-
705.
Andersson, L.-E. (1999b) Quasistatic frictional contact problems with finitely many
degrees of freedom. LiTH-MAT-R-1999-22.
Andersson, L.-E. (2000) Existence results for quasistatic contact problems with Coulomb
friction, Applied mathematics and Optimization 42 (2000), 169-202.
260
Andersson, L.-E. and Klarbring, A. (2000) Quasi-static frictional contact of discrete

mechanical structures. European Journal of Mecbanics/A 19, 61-77.
Ballard, P. 1999 A counter-example to uniqueness in quasistatic elastic contact prob-
lems with small friction. Int. J. Engng. Sci. 22, 163-178.
Cocou, M., Pratt, E. and Raous, M. (1996) Formulation and approximation of qua-
sistatic frictional contact. Int. J. Engng. Sci. 34, 783-798.
Klarbring, A. (1987) Contact problems with friction by linear complementarity. In
Unilateral Problems in Structural Analysis, Vol. 2 (CISM Courses and Lectures,
No. 304), (eds. G. Del Piero, F. Maceri, Springer, Wien,pp. 197-219.
Klarhring, A. (1990a) Derivation and analysis of rate boundary-value problems with
friction. European Journal of Mecbanics/A 1, 211-226.
Klarbring, A. (1990b) Examples of non-uniqueness and non-existence of solutions to
quasistatic contact problems with friction. Ingenieur-Archiv 60, 529-541.
Klarbring, A., Mikelic, A. and Shillor, M. (1988) Frictional contact problems with
normal compliance. Int. J. Engng. Sci 26, 811-832.
Klarbring, A., Mikelic, A. and Shillor, M. (1989) On friction problems with normal
compliance. Nonlinear Analysis, Theory, Methods & Applications 13, 935-955.
Klarbring, A., Mikelic, A. and Shillor, M. (1991) A global existence result for the
quasistatic frictional contact problem with normal compliance. International Series
of Numerical Mathematics 101, 85-111.
Rocca, R. (2000) Analyse matMmathique et nummque de problemes quasi statiques
de contact unilateral avec frottement local de Coulomb en elasticite. Ph.D. thesis,
I'Universite d' Aix-Marseille I.
AUGMENTED LAGRANGIAN
METHODS FOR A CLASS OF
NONCONVEX CONTACT PROBLEMS
IN STRUCTURAL MECHANICS
W. R. Bielski
Institute of Geophysics, Pol. Acad. Sci.,
01-452 Warsaw, ul. Ksi~cia Janusza 64, Poland
A. Galka
Institute of Fundamental Technological Research, Pol. Acad. Sci.,
Swi~tokrzyska 21, 00-049 Warsaw, Poland
J. J. Telega
Institute of Fundamental Technological Research, Pol. Acad. Sci.,
Swi~tokrzY8ka 21, 00-049 Warsaw, Poland
Introduction
In Bielski and Telega (1998) the study of unilateral contact problems
for elastic plates was started. Such a study was next extended to the
linear Koiter shell model in Telega et al. (2001). The aim of the present
contribution is to develop augmented Lagrangian methods applicable to
nonconvex unilateral contact problems arising in structural mechanics.
To this end we have extended the approach proposed by Ito and Kunisch
(1990, 1995) for convex problems. An illustrative example pertains to a
geometrically nonlinear elastic beam.
1. Von Karman plates

The strain measures are defined by, cf. Lewinski and Telega (2000),
261
I.A. C. Martins and M.D.P. Monteiro Marques (eds.). Contact Mechanics. 261-268.
262
where Ea,B(U) = u(a,,B) = H~+ ~:~). We note that only the first strain
measure is nonlinear.
The constitutive equations have the form
As previously, N and M are the membrane forces tensor and moments

tensor, respectively. In the absence of the obstacle the equilibrium equa-
tions are given by
Na,B,,B + Pa = 0, Ma,B,,Ba + (Na,Bw,,B ),a +p = 0 in O. (3)

We impose the following boundary conditions
U = 0 on ro , w
aw
= -an = 0 on r .
An appropriate space for displacements is
The functional of the total potential energy is given by

1 ( 1 1
J(u, w) = i In[A a,B>..P,(c:a,B(U) + iw,a w,,B ) (c:>..p,(u) + i W'>" w,p,)
+Ba,B>..p,ll:a,B(w)Il:>..p,(w))dz - { (Paua + pw)dz - (rauadf'. (5)

ln lrl
The nonlinear strain measure renders the functional J nonconvex on
Hl(O)2 x H6(O), and particularly on V. This functional is weakly lower
semicontinuous and bounded from below. For the obstacle problem the
set of kinematically admissible fields is specified by
K = {(u, w) E VI w(z) + h ~ f(xa + ua(z) - hW,a(z)), z EO}.
If K is non-empty, then in general it is a non-convex set. K is a convex

provided that f is a concave function. We assume that K =f=. 0 . We can
now formulate the obstacle contact problem.
Problem (P) Find
inf{J(u,w) I (u,w) E K}.
The problem of existence of a solution to problem (P) was solved in

Bielski and Telega (1998).
Remark 1.1. In Telega et al. (2001) we have formulated a general
obstacle problem for the Koiter model of shell.
263
2. Augmented Lagrangian methods for
nonconvex problems
In this section we propose augmented Lagrangian methods applicable
to nonconvex contact problems. To this end we extend the approach
developed by Ito and Kunisch (1990, 1995).
Nonconvex set of constraints Ito and Kunisch (1990) carefully

studied the augmented Lagrangian method directly applicable to geo-
metrically linear problems in the case of convex sets of constraints.
This approach is now extended to geometrically nonlinear contact
problems in the presence of nonconvex constraints. First, we consider
the case where only the set of constraints is nonconvex. The algorithm
we are going to present is applicable to geometrically linear structures
where constraints are nonconvex.
The problem under investigation is
(P) min{~a(u, u) -l(u) I g(u) ~ 0, u E B}.

Here the following spaces and mappings are used: V is a Hilbert space;
B is a reflexive Banach space continuously embedded into V; H is a
Hilbert lattice with inner product (-, .); see Ito and Kunisch (1990).
a(·,·): V X V is a bilinear and continuous, V-elliptic form, with
a(u,u) ~ Collull~, for some Co > 0; l: V ~ lR is a continuous linear
functional; 9 : B ~ H is in general a nonconvex, continuous, Gateaux
differentiable mapping.
From a practical point of view, the expression" Hilbert lattice" merely
means that the constraint g(u) ~ 0 appearing in problem (P) is mean-
ingful. We assume that g(u) = G(u) - G1(u), where the mapping G is
convex whilst Gl is nonconvex. Ito and Kunisch (1990) procedure can
be extended by combining their augmented Lagrangian technique with
an iterative procedure:
mth step.
(6)
Then the set
(7)
is convex. At each step m we define a family of augmented Lagrangian
problems by
(P)m,c,),m, Lm,c(um, Am) = min{Lm,c(u, Am)1 u E B}
where Am is the Lagrangian multiplier at step m and
Lm,c(u, Am) = ~a(u,u) -l(u) + (Am,gm(U,Am,C)} + ~lIgm(U,Am,c)lIt,

264
and A E H, c > 0, c E jR+. Moreover
9m(U, A, c) = sup (9m(U), -~) .

The mapping 9m is defined by 9m(U) = G(u) - G(u m- l ).
The Algorithm
(1) Choose Ar E H, Ar ~ 0, and c > 0,
(2) put n = 1,
(3) solve (P)m,c,.\w for u~,
(4) put A~+1 = A~ + cg(u~, A~, c) = sup(O, A~ + C9(u~)),
(5) put n = n + 1 and return to (3).
We observe that the parameter c may also depend on m.
00
Applying Ito and Kunisch results we get Go 2: lIu~ - umll} ~ 2lc IIAr-
n=l
Amllk ~ sup 2lcllAr - A*mll k < 00, since c may be taken sufficiently
m>l
large, such that for each mEN we have
;cllAr - Amllk < Gl , Gl > O. (8)

Let us pass to an example.
Example 2.l. To cope with geometrically nonlinear plates we addi-
tionally introduce a sequence of bilinear forms. For instance, in the case
of von Karman plates we take
. _r 1 m-l m-l 1 m-l m-l
am(u,w,u,w) - In[Aa,B.\p,(Ea,B(u)+2 w ,a w,,B )(E.\p,(U)+2"w,.\ w,p, )
+Ba,B.\p,Ka,B{W)K.\p,{W)] dx; m = 1,2, ...
Another possibility is to introduce the following sequence of bilinear
forms
am{u,w;u,w) = !n[Aa,B.\p,{EafJ{U) + ~W,aW:a-l)(E.\p,{U) + ~w,.\w:;!-l)

+Ba,B.\p,Ka,B{W)K.\p,{W)] dx; m = 1,2, ...
Then, instead of problem (P), we have a sequence of problems
(Pm) min{~am{u,w;U,w) -l{u,w) I 9{U,W) ~ 0, (u,w) E K2};
for m = 1, 2, ... , and similarly in the case of am.

Here l (u, w) is a loading functional. If K 2 is a nonconvex set, in order
to use the augmented Lagrangian method outlined previously, we have
to replace K2 by a sequence of convex set of constraints K'!J".
265
Nonconvex extension ofIto and Kunisch augmented Lagrangian

method Ito and Kunisch (1995) investigated an augmented Lagrangian
method for a significant class of nonsmooth, convex optimization prob-
lems in infinite dimensional Hilbert spaces. More precisely, let X, H be
real Hilbert spaces and K a closed convex subset of X. Consider the
minimization problem
(Q) min{J(u) + cp(Au) I u E K},
where J : X -+ lR is a lower, semicontinuous differentiable, convex func-
tion, A E L(X, H) and cp : X -+ lR is a proper, lower semicontinuous
convex function. The convex functional cp is not necessarily smooth;
in applications it may be an indicator function of a closed convex set.
Several examples of the linear and continuous operator A are provided
in Ito and Kunisch (1995). For instance, in unilateral contact problems
with constraints imposed on the boundary, A is a trace operator (in the
sense of value of a function on the boundary).
A smooth approximation of cp yields the following problem:
min{Lc(u, A) I u E K},
where
Lc(u, A) = J(u) + CPc(Au, A), (9)
CPc(v, A) = inf{cp(v - u) + (A, u) H + ~lIull~}. (10)
Here (c, A) E lR+ X H. We observe that cp(., A) is (Lipschitz) continuously
Fh~chet differentiable.
Ito and Kunisch (1995) developed the following augmented Lagrangian
method involving a sequential minimization:
A ugmented Lagrangian Algorithm
1: Choose a starting value Al E H, a positive number c and set k = 1.
Step 2: Given Ak E H find Uk E K by
Lc(Uk, Ak) = min{Lc(u, Ak) I u E K}.
Step 3: Update Ak by Ak+! = CP~(AUk' Ak), where cp' denotes the Frechet
derivative of the functional cp(., A).
Step 4: If the convergence criterion is not satisfied then set k = k + 1
and go to Step 2.
Under suitable, physically plausible assumptions, the augmented La-
grangian algorithm just sketched converges.
Obviously, this algorithm is not directly applicable to nonconvex con-
tact problems of say finitely deformed elastic bodies and geometrically
nonlinear structures. There are three basic sources of nonconvexity: (i)
266
a nonconvex functional J, (ii) a nonconvex functional cp, (iii) a nonlin-
ear operator appearing in the functional cpo Such an operator is denoted
by N. Obviously, in practice various combinations of cases (i)-(iii) are
important.
For geometrically nonlinear problems the functional cp is usually an
indicator function of a (weakly) closed and nonconvex set, cf. Example
2.1 and He et al. (1996). We already know how to generate a sequence
of convex sets of constraints.
A large class of geometrically nonlinear problems leads to the functional
J of the form,
J(u) = G(Au) + F(u),
where G represents the functional of the total internal energy whilst
F is a loading functional, usually a linear one. The functional G is
nonconvex. For nonlinear structures it may often be written as follows
(11)
where the functional G(·, A2U) is convex whilst G(A1u,·) is nonconvex.

To use the augmented Lagrangian method we combine the approach of
Ito and Kunisch (1995) with the iterative procedure. To this end we set
Gm(Au) = G(A1u,A2 u m- 1), m = 1,2.... (12)
and consider a sequence of regularized minimization problems
min{Gm(Au) + F(u) + CPc(Au,,x) I u E K}.

Now we have a sequence of convex problems (Q)m,c,~, m = 1,2, ... , to
which we can apply the augmented Lagrangian method developed in Ito
and Kunisch (1995).
3. Specific one-dimensional nonconvex contact

problem
In this section we are going to study a simple one-dimensional non-
convex contact problem. Consider the following minimization problem.
Find (u, w) E Ku x Kw such that
J(u,w) = inf J(u,w)

(u,W)EK,.xK,.,
where
J(u,w) = 0 10.5 (
-0.5
1
u,x + 2"W'XW 2H
,x )2 dx + -3-0
3 1°·5 (w,xx)2dx
-0.5
267
{0.5
+ i-O.5 bw{x)dx,
Ku = {u I U E W 1,4{-O.5,O.5), u{-O.5) = u{O.5) = O},
Kw = {w IwE W 1,2(O.5, 0.5), w( -0.5) = w(0.5) = w,x( -0.5) =
w,x(0.5) = 0, g(w) ~ O}.
and C > 0, Particular forms of the function 9 are given below. Anyway,
we assume that the set Kw is convex. To solve this problem we introduce
the sequence of functionals, cf. the previous section,
Jm{u, w) = C 10.5
-0.5
{u,x
1
+ -2W,xW~-1)2 dx + -3-C
2 H3 1°·5 (w,xx)2 dx
-0.5
+ 10.5
-0.5 bw(x) dx, m= 1,2, ...
Lm,c,>.(u, w) = Jm(u, w) + 21
C
1°·5 [(sup{O,). + cg(w)})2 ]dx.
-0.5
_).2
To apply the augmented Lagrangian method we consider the following

constraint: g(w) = -w(x) - ~~ - V1 - x 2.
To proper the calculation we take:
b 2 104 H 1 o() 0),

C = -. , = 192' w x = , ° = {10 for x E (-0.2; 0.2)
otherwise
and c = 10000, where c, ). are the parameters in the augmented La-

grangian. The results of calculations are presented in Figs. 1, 2. They
have been obtained by using FEM.
0.1 0.2 0.3 0.4 0.5
Figure 1. The transverse displacement w(x) , steps 1,2,3,4; c = 10000,

268
0.0025
0.002
0.0015
0.001
0.0005
o 0.1 0.2 x 0.3 0.4 0.5

Figure 2. The axial displacement u(x) , steps 1,2 ; c = 10000;
Acknowledgments
The first author was partially supported by the State Committee for
Scientific Research (KBN, Poland) through the grant No 6 P04D 039
15.
References
Bielski W. R., and Telega, J. J. (1998) Existence of solutions to obstacle problems for
linear and nonlinear elastic plates, Math. Compo Modelling, 28, 55-66.
Telega, J.J., Galka, A., and Bielski, W. R. (2001) Augmented Lagrangian methods
for a class of convex and nonconvex problems, J. Theor. Appl. Mech., 3, 741-768.
Ito, K., and Kunisch, K. (1990) An augmented Lagrangian technique for variational
inequalities, Appl. Math. Optim., 21, 223-241.
Ito, K., and Kunisch, K. (1995) Augmented Lagrangian methods for nonsmooth,
convex optimization in Hilbert spaces, in: Control of Partial Differential Equations
and Applications, ed. by E. Casas, pp. 107-117, Marcel Dekker.
Lewinski, T and Telega, J. J. (2000) Plates, Laminates and Shells: Asymptotic Analy-
sis and Homogenization, Series on Advances in Mathematics for Applied Sciences,
vo1.52, World Scientific, Singapore.
He, Q.-C., Telega, J. J., and Curnier, A. (1996) Unilateral contact of two solids subject
to large deformation and existence results, Proc. R. Soc., London, A452, 2691-
2717.
APPROXIMATION OF QUASISTATIC
SIGNORINI PROBLEMS WITH LOCAL
FRICTION BY A MIXED METHOD
Marius Cocou
Laboratoire de Mecanique et d'Acoustique - C.N.R.S., 91 chemin Joseph Aiguier,
19402 Marseille Cedex 20 and Universite de Provence, Marseille, France
Remi Rocca
Laboratoire de Mecanique et d'Acoustique - C.N.R.S., 91, chemin J.Aiguier,
19402 Marseille Cedex 20, France
Abstract The approximation of the quasistatic two-body unilateral contact prob-

lem with local Coulomb friction by a mixed finite element method is
studied. Continuous and discrete variational formulations are stated.
Using a regularity result for the normal component of the stress vector
for an auxiliary problem with given friction and some error estimates,
convergence to the continuous quasistatic solution is proved when the
discretization parameters tend to zero.
1. Introduction
This paper deals with the numerical analysis of the quasistatic prob-
lem of unilateral contact with local Coulomb friction between two elas-
tic bodies. The approach we use here follows that of Rocca and Cocou
(2001b), where the approximation of the contact problem for an elastic
body by a mixed finite element method is studied.
Contact problems with local friction law were first considered in the
static case by Necas, Jarusek and Haslinger (1980). They proved an
existence result in the case of a strip by a shifting technique when the
friction coefficient is sufficiently small. This result was extended by
Jaruaek (1983) for a domain with smooth contact zone and for two elastic
bodies.
269
lAC. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 269-276.
270
Using a mixed finite element method, Haslinger (1983) proved the

existence of discrete solutions converging towards a solution of the static
unilateral contact problem with local friction.
Existence results for the quasistatic unilateral contact with local fric-
tion have been obtained by Andersson (2000) and by the authors, see
Rocca (1999), Rocca and Cocou (200la), who have proved that there ex-
ists a solution if the friction coefficient is sufficiently small and smooth.
An extension to a quasistatic problem coupling unilateral contact, local
friction and adhesion has been also studied by Cocou and Rocca (2000).
This paper is organized as follows. In section 2, we present the con-
tinuous variational formulations for the initial problem.
In section 3, we present the discrete variational formulations and their
approximation.
In section 4 we prove the existence of a saddle point for the incremental
regularized discrete problem. Using some error estimates we prove that
the Lagrangian multiplier is bounded in H- 1/2+(3 independently of the
spatial discretization parameter provided that the friction coefficient is
small enough. Section 5 is concerned with convergence results.
2. Classical and variational formulations

Let us consider two linear elastic bodies which occupy the bounded
domains 0.8, (3 = 1,2, of Rd, d = 2, 3. Let rf, r~ and r3 be three
open disjoint parts of r.8 = ao.8 such that r.8 = rf u r~ u r3 and
mes(rf) > O. We suppose that the two solids are initially in contact
with local friction on r3, which is considered as being a bounded zone of
contact. We denote by n.8 the outward normal unit vector to r.8, u.8 =
(uf, ... , u~) the displacement field, £.8 = (fij (u.8» the strain tensor and
by 0'.8 = (of;(u.8» = (afjkl fkl(u.8» the stress tensor. We adopt the
notations u~, U?, O'?
u~, for the normal and tangential components of
the displacement vector and of the stress vector with respect to n 1 •
The problem consists in finding the displacement fields solution of the
equilibrium equations and boundary conditions. We suppose that a~kl
are of class 0°,1/2+0 with 1/2 > & > 0 in 0.8, 1 ~ i, j, k, I ~ d, with the
usual conditions of symmetry and ellipticity.
Let us denote by q, = (q,I, q,2) and ..p = (..pI, ..p2) the given body
forces and tractions. The initial displacements of the bodies are denoted
by uO,(3 and a displacement U.8 = 0 is prescribed on rf, (3 = 1,2. We
suppose that the friction coefficient fJ belongs to LOO(rl) and to the set
of the multipliers of Hl/2(rl) denoted by M(Hl/2(rl».
271
The classical problem is as follows: find u = u( t, x) = (u 1 , u 2) such

that u(o) = (UO,l, UO,2) in 0 1 x 0 2 and for all t E]O, T[, (3 = 1, 2,
div ooP(uP) = -qi, O"Z(u P) = a~klfkl(UP) in OP, (1)

= 0 on rf, ooPnP = 1/JP on rg, (2)
°on
uP
u}y - u~::; 0, O"}y(u l )::; 0, (u}y - u~) O"}y(u l ) = r 3, (3)
ooini = _002n 2, Iootl::; -p, O"}y on r3 and (4)
{ Iootl < -p, O"}y ==> tit - 14 = 0,
Iootl = -p, O"}y ==> 3A ~ 0, ti} - 14 = -A oot. (5)
We define V = VI X V 2 , where vP = {v p E [HI (OP)]d; v P = 0 a.e. on
°
rf}, {3 = 1,2, and K = {v E V; v}y - v'h ::; a.e. on r3}. The dual-
ity pairings on H l / 2(r P), H- 1/ 2(r P) and on [H I/ 2(r P)]d, [H- 1/ 2(r P)]d
are denoted by ( , ). We suppose that uO = (uo'l, UO,2) E K, lPP E
W I ,2(0, T; [L 2(OP)]d) and that 1/JP E W 1,2(0, T; [L2(r~)]d) such that its
support is strictly included in r~. Let us introduce the cut-off function
() E COO (Rd ) , 0::; () ::; 1, such that () = 1 on I'3 and () = ° on I'~,o, where
r~'o is an open subset having the property that for all t E [0, T],
supp 'l/i(t) c r~'o c I'~,o c r~.
Using Green's formula we obtain a variational formulation of the qua-
sistatic unilateral contact problem with local friction as follows.
Problem Po : Find u E W 1,2(0, T; V) such that u(O) = (UO,l, uO,2)
and for almost all t E]O, T[ u(t) E K and
a(u, v - til + jo(u, v) - jo(u, til (lP, v - til + (1/J, v - ti)
~
+(()O"}Y(u 1), [VN] - [UN]} 'V v E V, (6)
(()O"}y(u l ), [ZN] - [UN]} ~ ° 'V z E K, (7)
where a(u, v) = Ep aP(u P, v P ), aP(uP, v P) = Jn~ a~klfij(UP)fkl(VP),
jo(u, v) = -(p,0"}y(u 1 ), l[vT]I}, [1]] = 1]1 _1]2, (lP, v) = Ep(lPP, v P),
(lPP, v P) = Jn~ lPP.vp, (1/J, v) = Ep (1/JP , v P} = Ep Jr~ 1/Jp.v f1 •
2
Problem Po is equivalent to the following mixed variational formu-
lation.
Problem P: Find two mappings t H u(t) and t H A(t) such that
u E W 1,2(0, T; V), A E W 1,2(0, T; H- l / 2(r l )), u(O) = uO, A(O) = AO =
()O"}y(UO,l) and for almost all t E]O, T[ A(t) E C*- and
a(u, v - til + j(A, v) - j(A, til ~ (lP, v - til + (1/J, v - ti)

+( A, [VN] - [UN]) 'V v E V, (8)
(11" - A, [UN]) ~ ° 'V 11" E C*-, (9)
272
where C*- = {11" E H-1/2(r1); supp 11" C r 3 } n B_(r1), B_(r1) is the
set of bounded negative measures on r 1 and j(>.., v) = -(1-'>", I[VT]I). The
Lagrange multiplier>.. satisfies>.. = (Iu1(u 1).
3. Discrete formulations
In order to prove the existence of discrete solutions converging towards
u, we consider the following finite element approximations.
Let (r'/)h be a set of regular triangulations of Of3, {3 = 1,2, see
Ciarlet (1978). Each triangulation is a collection of elements (Or) such
that Of3 = UiEIf O~ with O~ n Or = 0 for all k,1 E If, k =f. I. We
assume that each triangulation is compatible with the partition of r f3 ,
that is, each point where the boundary condition changes is a node of
a set O~ and we require also that the nodes lying on r3 belong to both
triangulations. We shall use the notation Th = {T,/, 1h2}. Let (r~)H be
a set of triangulations of r3 = UiE.1H r 3 ,;. We suppose that the elements
r 3I i, i E .JH, are segments for d = 2 .and rectangles
..
for d = .
3. We
.
set
HI = length (r3,i) for d = 2, and HI = (HI ; H~), where HI, H~ are
the lengths of the edges of r 3 ,i for d = 3. We assume that (r~)H are
regular that is there exists a constant '1 > 0 such that ~in(Hi)/H ~ '1,
IE.1H
where H = ~ax(Hi) for d = 2, and ~in(HLm)/H ~ '1, where
IE.1H IE.1H
H = ~ax(HL H~) for d = 3.
IE.1H
We introduce the following sets:
Vh = {Vh = (vk,v~) E [CO(01)]d x [CO(02)]d ;v!lflj.8 E [P1(Or)]d

'ViEIf, v~=O on rf, {3=1,2},
Vh {VhEVh; V~T=O on r f3 , V~N=O on r f3 \r3, {3=1,2},
LH {1I"H E L 21
(r ); 1I"H = 0 on r 1-
\r3 ,1I"Hlrs,i E PO(r3,i) 'Vi E.JH } ,
IIH {1I"H E LH; 1I"H ~0 on r 3,i 'Vi E .JH },
KhH {Vh E Vh; (1I"H, [VhN]) ~ 0, 'V 1I"H E IIH},
where Pk(W) denotes the space of polynomials of degree lower or equal
to k on w.
We consider the following semi-discrete variational problem.
Problem PhH: Find two mappings t -+ Uh(t) and t -+ >"H(t)
such that Uh E W 1,2(0, T; Vh), >"H E W 1,2(0, T; LH), Uh(O) = u~,
>"H(O) = >..~ and for almost all t E]O, T[ >"H(t) E IIH and
273
a(Uh, vh - Uh) + j(>'H, Vh) - j(>'H, Uh) ~ (q" Vh - Uh)

+(1/J, Vh - Uh) + (>'H, [VhN] - [UhN)) vVh E Vh, (10)
( 1rH - >'H, [UhN)) ~ 0 V 1rH E IIH. (11)
We adopt the following time discretization. For all n E N* we use the
notation q,i = q,(i6.t) and 1/Ji = 1/J(i6.t) for i = 0, ... , n, where 6.t = TIn.
We consider the following incremental problem and its regularization for
i = 0, ... ,n-l.
Problem P~H: For uk E K hH , find u~+1 E K hH , >.~1 E IIH such
that
a(u~+1, Vh - u~+1) + j(>.~l, Vh - uk) - j(>.~I, u~+1 - u~)
~ (q,i+1, Vh - u~+1) + (1/Ji+1, Vh - u~+1) (12)
'+1 '+1
+(>.y ,[VhN] - [uhN)) V Vh E Vh,
( 1rH - >.~1, [u~jJ]) ~ 0 V 1rH E IIH. (13)
Problem P~iI : For u~ E K hH , find Uhll E KhH, >'HII E IIH such that
a(uhll' Vh - Uhll) + jll(>'HII' Vh - uk) - jll(>'HII' Uhll - uk)

'+1 '+1
~ (q,' ,Vh - Uhll) + (1/J' ,Vh - Uhll) (14)
+(>'HII' [VhN] - [UhIlN]) V Vh E Vh,
( 1rH - >'HII' [UhIlN]) ~ 0 V 1rH E IIH, (15)
where jll(>" v) = -(I-'>',1'/II(I[VT]I)) with (1'/11)11 a family of functions such
that Vv > 0, 1'/11 is convex, 1'/11 E Cl(R), 0 ~ 1'/~ ~ 1, 1'/11(0) = 1'/~(0) =
oand 0 ~ s -1'/II(S) ~ v, Vs E R.
4. Existence of incremental solution
In order to prove the existence of a solution for problem (12,13) we
consider the following auxiliary problem associated to (14,15).
Problem P~iI, 9H : For gH E IIH, find Uh = Uh(gH) E KhH, >'H E
II H such that
a(uh' Vh - Uh) - (l-'gH, 1'/11(1 [VhT - u~T]I) -1'/II(I[UhT - u~T]I)) (16)

~ (q,i+l, Vh - Uh) + (1/Ji+1, Vh - Uh) + (>'H, [VhN] - [UhN]) VVh E Vh,
(1rH - >'H, [UhN]) ~ 0 V 1rH E IIH. (17)
Theorem 1 For all h, H there exists a solution (Uh' >'H) of the problem
P~iI, 9H and Uh is unique. Moreover, >'H is unique if we assume that
the following condition holds
(>'H, [VhN)) = 0, V Vh E Vh :=:} >'H = 0, (18)
274
Proof. Let J(v) = ~a(v,v) - (JLgH,1]v(I[VT - UkT] I)) - (qi+1,v)-

("pi+l, v) and ,c(v, 7rH) = J(v) - (7rH, [VhN]). The Lagrangian functional
,c(.,.) has a saddle point (Uh, AH) on Vh X IIH (see for example Ekeland
and Temam (1974)). From the coerciveness of a(.,.) and relation (16) it
follows that Uh is unique. By condition (18), we obtain that AH is also
unique.
Condition (18) is fulfilled if 2h ~ H and we will suppose that this
relation holds in the following. Theorem 1 enables us to define the
mapping ~H as follows: IIH 3 gH -+ ~H(gH) = AH E IIH, where AH
is the solution of (16,17). The end of the section is devoted to prove
that ~H has a fixed point with an estimate independent of h, H, for the
Lagrange multiplier AH.
Theorem 2 Under the above hypotheses there exists a solution of prob-
lem P~~ for any value of the friction coefficient.
By Schauder's second fixed point theorem (see Zeidler (1993)) it fol-
lows that the mapping ~ H has a fixed point which is a solution of (14,15).
Theorem 3 Let Ti", ~ be such that there exist 71 > 0,72 > 0 indepen-
dent of hand H with 71 ~ h/ H ~ 72 and a E]O, 1/2[. Then there exists
a value JL* > 0 such that the following relation holds
IIAHIIH-l/2+<>(r1) ~ IIJLllvlO(rl)/JL* IIgHIIH-1/2+<>(r1)
+Cl (lIqi+1,1 II [L2(!V))d + l11/1i+1,II1[L2(r1))d + IIU~II1[Hl(!ll))d ).(19)

Thus for JL such that IIJL IIvx> (rl) < JL*, there exists TO > 0 such that
the mapping ~H has a fixed point in {7rH E IIH; I17rHIIH-1/2+<>(r1) ~ TO}.
Consequently one obtains once again that problem P~~ has a solution.
Theorem 4 There exists a solution (u~+1, A~ 1 ) to problem P~H
satisfying the following estimate
.+1
IIAk IIH-l/2+<>(r1) ~ C2 ( IIcp'+11, II [L2(!l1))d, ·+11
l11/1z , 1I[L2(r 1))d,
IIcpi,IIi[L2(!ll ))d, lI"pi,II1[L2(r1 ))d ). (20)

This enables us to select a subsequence still denoted by (v) such that
v -+ 0+, Uhv -+ U~+1 in V, AHv -+ A~1 in H- 1/2+0(r 1). It follows that
·
11m a(Uhv, Uhv ) = a (i+
U h 1 , U i+
h
1) '
v-+o+
· Jv
11m . ('AHv, Uhv - Uhi ) = JO( AH
d+l , Ui+1 - Uh
i ),
h
v-+o+
lim (AHv,UhvN) = (A~I,u~1J).
v-+o+
275
All the previous relations, as well as other similar ones concerning the
limits with respect to the parameter II enable us to pass to the limit in
(14,15) and to obtain (12,13). Relation (20) is obtained from (19).
5. Convergence results
In this section we prove that there exists a sequence of discrete solu-
tions of problems (P~H)i which converges towards (Uh, AH), solution
of problem PhH' If IIJ.tIIM(Hl/2(r)) < ji., then there exist two constants
C3 > 0, C4 > 0 such that
lI~ukllv ~ C3(II~qillv + 1I~1/Jill[H-l/2(rl)ldx[H-l/2(r2)ld),

II~AkIlH-l/2(rl) ~ C4(1I~4>illv + 1I~1/Jill[H-l/2(rl)ldx[H-l/2(r2)ld).
These two estimates allow us to construct some sequences of discrete
solutions for the displacement fields and the Lagrange multipliers which
converge towards some absolutely continuous mappings with respect to
time, as follows. For i = 0, ... , n we define
Uhn(t) = u~+1, Uhn(t) = uk + (t - ti)(U~+1 - uk)/ ~t,

'+1 - , '+1'
AHn(t) = Ak , AHn(t) = An + (t - ti)(Ak - An)/~t,
Uhn(O) = Uhn(O) = u~, AHn(O) = XHn(O) = A~
for t E]ti' ti+1] , ti = i~t. There exist two elements Uh, AH and a subse-
quence (nk)kEN such that
Uhnle --+ Uh III W 1,2(O, Tj V),
XHnle --+ AH III W ,2(O, Tj H- 1/ 2(r 1 )),
1
Uhnle (t) --+ Uh(t) in V \It E [0, T],
AHnle (t) --+ AH(t) in H-l/2(rl) \It E [0, T].
The proof is similar to the ones given in Cocou and Rocca (2000) or
Rocca and Cocou (2001) in semidiscrete cases. We still denote by
(Uhn) , (AHn), (Uhn) and (X Hn ), the corresponding convergent subse-
quences. Using the previous estimates, one can prove the
Theorem 5 Let (1hj )jEN and (r~j)jEN be such that for all j EN we
have 71 ~ hj/Hj ~ 72. Let I' E M(Hl/2(rl)) be such that IIJ.tllv"'(r1) <
1'* and IIJ.tIlM(Hl/2(r1)) < ji.. Then there exists a subsequence (jk)kEN such
that (Uh" Ie , AH" Ie )kEN, the solutions of problems (Ph'11e H,' Ie ) , converge
weakly in W 1,2(0, Tj V)XW 1,2(0, Tj H-l/2(r 1 )) towards (u, A = 8UN(U))
solution of problem P. Moreover, we have
276
The result follows by using the previous estimates and by passing to the
limit in (10,11).
6. Conclusions
In this paper we have considered the unilateral contact with local fric-
tion between two elastic bodies. An implicit Euler scheme with respect
to time and a mixed finite element method for the space discretization
have been adopted. The fully discrete problem was solved by a fixed
point approach. This result and some estimates independent of the dis-
cretization parameters enable us to pass to the limit with respect to the
mesh size and to time. It would be interesting to consider the numeri-
cal analysis of unilateral contact problems with local friction for curved
contact zones. The problem of obtaining rates of convergence for the
quasistatic contact problem is open.
References
Andersson, L. E. (2000) Existence results for quasistatic contact problems with Coulomb
friction, Appl. Math. Opt. 42, 169 - 202.
Ciarlet, P. G. (1978) The finite element method for elliptic problems, North-Holland,
Amsterdam.
Cocou, M., Pratt, E., and Raous, M. (1996) Formulation and approximation of qua-
sistatic frictional contact, Int. J. Engrg. Sci. 34, 783-798.
Cocou, M., and Rocca, R. (2000) Existence results for unilateral quasistatic contact
problems with friction and adhesion, Math. Modelling and Num. Analysis 34,981-
100!.
Ekeland, 1, and Temam, R. (1974) Analyse convexe et problemes variationnels, Dunod,
Paris.
Haslinger, J. (1983) Approximation of the Signorini problem with friction, obeying
the Coulomb law, Math. Meth. Appl. Sci., 5, 422-437.
Hlavacek, J., Haslinger, J., Neeas, J., and LoviSek, J. (1988) Solution of variational
inequalities in mechanics, Springer-Verlag, New-York, Berlin, Heidelberg.
JaruSek, J. (1983) Contact problems with bounded friction - coercive case, Czechoslo-
vak Math. J. 33, 237-26!.
Netas, J., JaruSek, J., and Haslinger, J. (1980) On the solution of the variational
inequality to the Signorini problem with small friction, Bolletino U.M.l. (5) 17-
B,796-81!.
Rocca, R. (1999) Existence of a solution for a quasi static contact problem with local
friction, C. R. Acad. Sci. Paris 328 Serie I, 1253-1258.
Rocca, R., and Cocu, M. (2001) Existence and approximation of a solution to qua-
sistatic Signorini problem with local friction, Int. J. Engrg. Sci. 39, 1233-1255.
Rocca, R., and Cocou, M. (2001) Numerical analysis of quasistatic unilateral contact
problems with local friction, SIAM Journal of Numerical Analysis, to appear.
Zeidler, E. (1993) Nonlinear functional analysis and its applications, Vol I, Springer-
Verlag, New-York.
A BEAM IN ADHESIVE CONTACT
W.Han
Department of Mathematics,
University of Iowa, Iowa City, IA 52242, USA
K. L. Kuttler
Brigham Young University, Provo, UT 84602, USA
M. Shillor
Department of Mathematics and Statistics,
Oakland University, Rochester, MI 48309, USA
M. Sofonea
Laboratoire de Theorie des Systemes,
Universite de Perpignan, 66 860 Perpignan, France
Abstract A quasistatic process of contact with adhesion between an elastic beam

and a foundation is considered. The contact is modeled with the Sig-
norini condition when the foundation is rigid, and with normal compli-
ance when it is deformable. The adhesion is modeled by introducing the
bonding function (3, the evolution of which is described by an ordinary
differential equation. The existence and uniqueness of the weak solution
for each of the problems is established. A fully-discrete scheme for nu-
merical solutions of the problem with normal compliance is described.
1. Introduction
Processes of adhesion are very important in industry, especially when
composite materials are involved. There exists extensive engineering
literature on various aspects of the subject, however, general mathemat-
ically sound models are very recent. A novel approach to the modeling
277
lAC. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 277-284.
278
of contact with adhesion, based on thermodynamic derivation, can be

found in F'remond (1982, 1987). There, the adhesive contact process has
been modeled by the introduction of an internal variable, the adhesion
field (3, which measures the fraction of active bonds. Recent modeling,
analysis and numerical simulations of adhesive contact can be found in
Raous et a1. (1999) and references therein, and in Rojek and Telega
(2001), Rojek et al. (2001) and Han et al. (2001).
This work describes the adhesive contact between a beam and a foun-
dation, or an obstacle. The beam is assumed to be elastic, and the
obstacle either rigid or deformable. Our interest lies in the description
and analysis of the quasistatic processes of contact with adhesion. We
also describe a convergent fully-discrete numerical scheme for the prob-
lem with deformable foundation.
The paper is organized as follows. In Section 2 we describe the classi-
cal model for adhesive quasistatic contact between an elastic beam and
a rigid foundation. In Section 3 we present the variational formulation
of the problem, list the assumptions on the problem data, and state
our existence and uniqueness result in Theorem 3.1. In Section 4 we
present the model for adhesive contact between an elastic beam and a
deformable foundation. The reaction force is modeled with the 'normal
compliance' condition. We state the result on the existence of the unique
solution, and indicate that when the foundation becomes stiffer the so-
lutions converge to the solution of the problem with a rigid foundation.
In Section 5 we consider a fully-discrete approximation of the problem
with normal compliance. The discretized approximations converge to
the solution and, under additional regularity assumptions, the conver-
gence rate is obtained. The detailed proofs of the results presented here,
as well as additional results on dynamic and quasistatic adhesive contact
involving elastic or viscoelastic beams can be found in Han et a1. (2001).
2. The model
We consider an elastic beam of length L that is clamped at its left
end while the right end is free. The beam is being acted upon by an
applied force of (linear) density f, and it may come in adhesive contact
with a rigid foundation below it. The setting is depicted in Figure 1.
We denote OT = (0, L) x (0, T), for T > 0, and let u = u(x, t) represent
the vertical displacement of the beam, at (x, t) E flT. We assume that
the obstacle is described by the function y = 4>(x), for 0 ~ x ~ L. We
denote A = El, where I is the beam's moment of inertia and E the
Young modulus. We assume that the acting forces vary slowly in time
and the process is quasistatic. Then, the equation of motion for the
279
beam is
in Or, (1)
e
where = e(x, t) denotes the reaction force of the foundation and the
adhesion force. The displacements of the beam are restricted by the
rigid foundation to be above it, thus
(2)
which represents a nonpenetration condition. When contact takes place
e
the foundation's reaction force is directed upward,
in 0T. (3)
• !
f
!
0r---~----------~----~_
L x
y=~
Figure 1. The setting of the problem.

We now describe the adhesion process, following F'remond (1982, 1987).
We introduce the internal state variable {3 = (3(x, t), the 'bonding field,'
which measures the fraction of the active bonds between the beam and
the foundation. When {3 = 1 at a point the adhesion is complete; f3 = 0
means that all the bonds are severed and there is no adhesion, and
o< {3 < 1 represents the state of partial bonding. We suppose that the
adhesive resistance is active only when the force is directed upwards, try-
ing to separate the beam from the foundation, and this restoring force
is proportional to the distance from the obstacle and to f32. Therefore,
in OT. (4)
Here, K. > 0 represents the interface stiffness when the adhesion is com-
plete, and K.{32 is the 'spring constant' of the bonding field. Now, condi-
tions (2)-(4) may be written in the following complementarity form:
u ~ ~, e+ K.(u - ~){32 ~ 0, (u - ~)(e + K.(u - ~)(32) = 0 in OT. (5)
280
Next, following Raous et al. (1999), the evolution of the adhesion field
is given by
in flT, (6)
where 'Y is the adhesion rate constant, and r + = max{r,O} denotes the
positive part of r. We note that re-bonding is excluded in this model,
however, our methods can be used in the case when it is included. In (6),
and everywhere in the sequel, a prime represents the time derivative.
To complete the model we prescribe appropriate initial and boundary
conditions. The initial condition takes the form
{3(x, 0) = f30(x) for x E [0, L], (7)

where (3o represents the initial bonding field. The beam is rigidly at-
tached at its left end and there are no moments acting at the free end
of the beam, thus,
u(O, t) = ux(O, t) = uxx(L, t) = uxxx(L, t) = ° for t E [0, T). (8)

Here and below, subscripts x, xx and xxx denote the first, second and
third partial derivatives with respect to x, respectively.
The classical statement of the problem of quasistatic adhesive contact
of a beam with a rigid obstacle is as follows:
Problem P. Find a displacement function u : flT --t R and an adhe-
sion function (3 : flT --t R such that (1), (5)-(8) hold.
3. Existence and uniqueness

We proceed to derive a weak or variational formulation for the prob-
lem. We use standard notation for LP and Sobolev spaces and let V be
the closed subspace of H 2 (0,L) given by
= { v E H 2 (0,L) I v(O) = vx(O) = 0 }.

V
We denote by H the space L2 (0, L) and by h ·)H, I·IH its inner product
and the associated norm, respectively. Let K denote the convex subset
of V defined by
K = { v E V I v ~ ¢ on [0, L] }.
If (X, I . Ix) is a real normed space, C(O, Tj X) and C1 (0, Tj X) denote

the spaces of continuous and continuously differentiable functions from
[0, T] to X, with the usual norms.
In the study of Problem P we assume the following on the data:
A E DXl(O,L) and 3Ao °

> such that A ~ Ao a.e. on (O,L), (9)
281
f E C(O,T; L2(O, L)), (10)
'Y = const. > 0, (11)
¢ E C 1 (O, L), ¢ 5:. 0, (12)
E Loo(O, L), ~ ° a.e. on (0, L), (13)
°<
K, K,
(3o E Loo(O, L), {3o 5:. 1 a.e. on (0, L). (14)
Let a : V X V -+ R be the functional a( u, v) = IOL Auxxvxxdx for

u, v E V. Using a standard procedure we obtain the variational formu-
lation of the mechanical problem P.
Problem Pv. Find a displacement function u : [0, T] -+ V and an
adhesion function (3 : [0, T) -+ Loo(O, L) such that, for all t E [0, T),
u(t) E K, a(u(t), v - u(t)) + (K,{32(t)(U(t) - ¢), v - U(t))H ~

(J(t),v-u(t))H VvEK, (15)
(3'(t) + 'YK,(u(t) - ¢)2({3(t))+ = 0, (16)

(3(0) = {3o. (17)
Our main result is the following.

Theorem 1. Assume that (9)-(14) hold. Then, Problem Pv admits
a unique solution {u, {3} . Moreover, the solution satisfies
u E C(O, T; V), (3 E C 1 (0,T;L oo (0,L)). (18)
We conclude that, under the assumptions (9)-(14), the mechanical

problem P has the unique weak solution {u, {3}.
4. The case with deformable support

We describe and analyze a version of the model in which the support
is deformable. We use the so-called 'normal compliance' condition to
describe the reaction of the foundation when u < ¢. For the sake of
simplicity, and without loss of generality, we assume that ¢ == 0. All the
results below apply to the case when ¢ satisfies (12).
We consider the same setting as in Section 2, but the foundation now
is deformable and its reaction force depends on the displacements. We
use a 'normal compliance' condition to describe it (see, e.g. Kikuchi and
Oden (1988) or Klarbring et al. (1988)), thus,
(19)
282
Here, p = p(.) ~ 0 is a given function which vanishes for nonnegative
values of its argument. We note that when u > 0 then = -1i:(u+)/32e
e
represents the adhesion force and when u ~ 0 then = p(u) represents
the reaction force. Next, the evolution equation for the adhesion field
(6) is modified as follows:
/3' + 'Y1i:((u)+)2({3)+ = 0 in !IT. (20)
We use (u)+ to ensure that only tension contributes to debonding, since
compression does not affect it.
The classical formulation of the contact problem between a beam and
a flexible foundation with adhesion is:
Problem PNC. Find a displacement function u : !IT ~ R and an
adhesion function /3 : !IT ~ R such that (1), (19), (20), (7)-(8) hold.
We have the following weak formulation of problem PNC.
Problem PNC-V. Find a displacement function u: [0, T] ~ V and
an adhesion function /3 : [0, T] ~ LOO(O, L) such that, for all t E [0, T],
a(u(t),v) + (1i:/32(t)u(t),V)H = (f(t) +p(u(t)),v)H \::Iv E V, (21)
/3'(t) + 'Y1i: (u+)2(t)(/3(t))+ = 0, (22)
/3(0) = /30. (23)
We have the following existence and uniqueness result.

Theorem 2. Assume that (9)-(11), (13), (14) hold and that p
R ~ R+ is a decreasing, globally Lipschitz function which vanishes on
R+. Then, there exists a unique solution {u, /3} of Problem PNC- v.
Moreover, the solution satisfies (18).
We conclude that, under the assumptions (9)-(11), (13), (14), the
mechanical problem PNC has a unique weak solution {u,/3}.
We consider next the manner in which solutions of the problems with
normal compliance approach the solution of the problem of Section 3, in
which no penetration is allowed. To this end we denote by {uc' /3c} the
solution of the problem in which the normal compliance is modified by
mUltiplying it with 1/c, and let {u, /3} be the solution of the problem Pv.
Let W be any space for which the embedding of V into W is compact.
In Han et al. (2001) it has been shown that the sequence {uc } converges
strongly in C(O, T; W) and weak* in LOO(O, T; V) to u as c ~ O. The main
idea of the proof was to establish the equicontinuity of the sequence of
the solutions {uc } for the problems with normal compliance, and using
the uniqueness of the limit. This convergence result guarantees that as
the support becomes more rigid the solution gets closer to that of the
problem with a rigid obstacle.
283
5. Fully-discrete approximation
We now turn to numerical approximations of the quasistatic elastic
problem with normal compliance studied in the previous section. We
introduce a partition of the spatial domain [0, L] : 0 = Xo < Xl < ... <
XM = L. Denote Ii = [Xi-I, xd and hi = Xi - Xi-l for i = 1, ... , M, and
h = maxl~i~M hi the meshsize. We define the finite element spaces
Vh = {v h E V I vhlIi is cubic, 1 ~ i ~ M},
Qh = {qh E D)O(O,L) I qhl Ii is constant, 1 ~ i ~ M}.
Thus, Vh consists of piecewise cubics, and Qh of piecewise constant
functions. We define a piecewise averaging operator ph : Ll (0, L) -+ Qh
by
h
p UIIi
1 r
= IIil iIi udx, (24)
We use the following partition of [0, T] : 0 = to < tl < ... < tN = T,

with step-size kn = tn - tn-I, for n = 1, ... , N. Since a non-uniform
partition is allowed, we denote by k = maxn k n the maximal time step-
size. For a continuous function w(t), the notation Wn = w(tn ) will be
used.
A fully-discrete scheme for the quasistatic problem with normal com-
pliance is:
Problem P~'C. Find {u~k,f1~k};:=o C Vh X Qh such that for n =
1, ... ,N,
a(u~k,vh) + (~(f1~k)2u~k,vh) = Un + p(u~k),vh) 't/v h E vh, (25)
f1~k - f1~~1 + ,kn 'Ph[~ (u~~d!](f1~~d+ = 0, (26)

and
uohk -_ uo, ahk -- fJO
h fJO ah on (0 , L) , (27)
where f18 E Qh is an approximation of f10 and uS E V h.
Here, uS E Vh is an artificial initial value required by the fully-
discrete scheme. This value is needed in (26) when n = l.
The fully-discrete solution exists and is unique. Similar to the ar-
guments in Han and Reddy (1999), Han and Sofonea (2000), we have
the following result concerning convergence and convergence order of the
fully-discrete scheme.
Theorem 3. Assume I/uSI/H is uniformly bounded with respect to h
and the initial value f18 is chosen so that
(28)
284
Then, the fully-discrete method converges,

max [IIu n - u~kllv + lI,8n - ,8!kIl H ] -+
1~n~N
° as h, k -+ 0.
If, in addition, we assume K. E H 1(O,L), u E DXl(O,Tj H 3 (O, L)), (u+)"

(,8+), E W1,OO(O, Tj H), Ilu~IIH is uniformly bounded with respect to h
and
11,80 - ,8glIH ~ Ch, (29)
then, we have the error estimate
max [IIu n
l~n~N
- u~kllv + lI,8n - ,8!kIl H ] <
-
C(h + k).
We note that if we take ,8e = ph,Bo, then conditions (28) and (29) are
easily satisfied.
Acknowledgments
The work of the first author was supported by NSF under Grant
DMS-9874015.
References
Fremond, M. (1982) Equilibre des structures qui adherent a. leur support, C.R. Acad.
Sci. Paris 295, Serle II, 913-916.
Fremond, M. (1987) Adherence des solides, Journal de Mecanique Theorique et Ap-
pliquee 6, 383-407.
Han, W., and Reddy, B. D. (1999) Plasticity: Mathematical Theory and Numerical
Analysis, Springer-Verlag, New York.
Han, W., Kuttler, K. L., Shillor, M., and Sofonea, M. (2001) Elastic beam in adhesive
contact, preprint.
Han, W., and Sofonea, M. (2000) Evolutionary variational inequalities arising in vis-
coelastic contact problems, SIAM J. Numer. Anal., 38, 556-579.
Kikuchi, N., and Oden, J. T. (1988) Contact Problema in Elasticity, SIAM, Philadel-
phia, 1988.
Klarbring, A., Mikelic A., and Shillor, M. (1988) Frictional contact problems with
normal compliance, Int. J. Engng. Sci., 26, 811-832.
Raous, M., Cangemi, L., and Cocu, M. (1999) A consistent model coupling adhesion,
friction, and unilateral contact, Computer Meth. Appl. Mechan. Engng., 177,383-
399.
Rojek, J., and Telega, J. J. (2001) Contact problems with friction, adhesion and
wear in orthopaedic biomechanics. Part I: General development, J. Theoretical
and Applied Mechanics, 39, to appear.
Rojek, J., Telega, J. J., and Stupkiewicz, S. (2001) Contact problems with friction,
adhesion and wear in orthopaedic biomechanics. Part II: Numerical implementation
and application to implanted knee joints, J. Theoretical and Applied Mechanics,
39, to appear.
COULOMB FLUID-SOLID
INTERFACE LAW IN LUBRICATION
Guy Bayada
CNRS-UMR 5585 and 5514 LMC-INSA
401 Math, Villeurbanne 69621 Prance.
Mahdi Boukrouche
CNRS-UMR 5585 and UPRES 3058
E.A.N. Saint-Etienne 42023, Prance.
Abstract The asymptotic behaviour of a Stokes flow with Coulomb free boundary
friction condition when one dimension of the fluid domain tends to zero
is studied. The specific Reynolds equation associated with variational
inequalities is obtained and uniqueness is proved.
1. Introduction
An accurate choice of the boundary conditions on the fluid-solid inter-
face is of particular interest in lubrication area, which is concerned with
thin film flow behavior. In that case, the difference of velocities between
the surrounding surfaces is the governing phenomenon that allows the
pressure in the fluid to build up and prevent the solid surfaces to collapse
which is the main objective of the lubrication. In operating conditions,
no slip condition is induced by chemical bounds (cf. Pit (1999)) between
the lubricant and the surrounding surfaces and by the action of the nor-
mal stresses, which are linked to the pressure inside the flow. On the
contrary, tangential stresses are so high that they tend to destroy the
chemical bounds and induce slip phenomenon. This induces a transposi-
tion of the well-known Coulomb law between two solids (Duvaut (1980))
to the fluid solid interface. Although being implicitly used in numeri-
cal procedures for lubrication problems, a Reynolds thin film equation
taking account of such slip phenomenon seems not to have been stud-
ied in a somewhat mathematical aspect until recently for the simplified
Tresca interface condition in Bayada-Boukrouche (2001). The goal of
285
1.A. C. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 285-292.
286
the present paper is not only to give existence and uniqueness for this
problem but also to obtain rigorously the equation describing such a
phenomenon in a thin film flow by way of an asymptotic analysis in
which a small parameter is the width of the gap. The departure point
is the Stokes equation with the Coulomb boundary conditions and fall
within the scope of the work of Consiglieri (1992). The related formula-
tion is given in section 2. Following the same idea as in Assemien et al.
(1994) Bayada-Boukrouche (2001), Bresh et al. (1997), we introduce a
new scaling taking into account a small parameter. It is to be noticed
that we are led in a natural way to express the Coulomb coefficient in
terms of this small parameter in a way that induces the uniqueness of the
related Stokes problem. Then in section 3 we get estimates for velocity
and then pressure. Passing to the limit in the initial inequalities using
the results of Bayada-Lhalouani (2001), we are able to obtain the limit
problem, which can be expressed in terms of the (unknown) velocity and
tangential stresses. In section 4 a specific decomposition based upon
a new div-curl decomposition allows us to study the uniqueness of the
limit problem.
2. Basic equations and assumptions

Let w be a fixed bounded domain of the JR2 plane (x = (Xl, X2)). We
suppose that w has a Lipschitz continuous boundary and is the bottom
of the fluid domain. The upper surface fI is defined by X3 = H(x) =
H(XI, X2). Assuming that the film of fluid between the surfaces is thin,
we introduce a small parameter c, that will tend to zero, and a positive
smooth and bounded function h such that H(x) = ch(x). We denote by
nc = {(x, X3) E JR3 : x E wand 0< X3 < ch(x)},
Let fC be the boundary of nc. We have fC = wu f'I u f't where ft is
the lateral boundary. The motion in the fluid is described by
• The basic Stokes system of equations :
-v.D.uc + \1pc = 0 in Oc (1)
where pC, u c , and v are the pressure, the velocity field, the viscosity.
• the incompressibility equation
(2)
To describe the boundary conditions, let us introduce first a function gc

in (H ~ (fc))3 such that
(3)
287
Due to (3) it is well known cf. Girault-Raviart (1979) (lemma 2.2 p.24)
that there exists a function Gc such that Gc E (Hl(Oc))3 with
div(Gc)=O in oc, and GC=gC on rc. (4)
The actual velocities on the boundary are the following. On ry, a no

slip condition is given. The upper surface is assumed to be fixed so that:
(5)
On ri, the velocity is known and is parallel to the w- plane :
(6)
On w, there is a no flux condition across w so that
U3 = g~ = 0 on w. (7)
The tangential velocity is unknown and satisfies the Coulomb friction
law with k c the ratio of the tangential stress and the normal stress
where 1.1 here is the JR2 Euclidean norm, n = (nl' n2, n3) is the unit
outward normal to f c , and
are, respectively, the normal and the tangential velocity, the components
of the normal and the tangential stress tensor :
8uEC 8uj
(Ji,j = _pc 6i,j +v(8x; + 8Xi) (1~i,j~3),
and S is a regularization operator from H- ~ (w) into L~ (w) defined by :
'iT E H-!(w), S(T) E L!(w), S(T)(X) = I < T(t), <p(x-t) > I 1 1,

H-'}, (w),H6o(w)
for all x in w, where <p is a given positive function of class Coo with
support in w. We will denote in the following s := gc on w. To get a
weak formulation, we introduce
288
L6(ft) = {q E L2 (nc) { qdxdx3 = 0, where dx = dX1dx2}.
Joe
A formal application of Green's formula, using (1)-(3) and (5)-(8) leads
to the weak formulation: For Gc as in (4), find u C = (uj, u~, u~) in
K(nc) and pC in L5(n c), such that for all (q, cp) E L5(Oc) x K(Oc):
( qdiv( uC)dxdx3 = 0, (9)

Joe
Theorem 1 Assuming that the friction coefficient kc is a non negative

function in LOO(w), then there exists (uc,pc) in K(w c) x L5(Oc) which
is a solution to the problem (9),(10).
Proof. Using an idea of Consiglieri L.(1992), we use the Tichonov's
fixed point theorem to deduce the existence of u c , then as in Bayada-
Boukrouche (2001) the existence of pC is obtained using a duality result
of convex optimisation cf. Ekeland-Temam (1974).
Remark 1 For a small kc , it is possible, exactly as for solid-solid
Coulomb interface law, to prove uniqueness of u C and then of pC (up
to an additive constant).
3. Study of the limit problem

According to the change of variables :y = X3/£, we define the fixed
domain: 0 = {(x, y) such that x E w, and 0 < y < h(x)}, and
r r
we denote its boundary by r = wu L U 1. We define the following
functions in 0 : uf{x, y) = uf{x, X3) 1 ~ i ~ 2, and
Let us assume the following dependence of the data (with respect to £):
(11)
The second assumption means that the friction coefficient is the ratio
of the tangential stress inside the film (which has the same order of
magnitude as the ratio between the tangential velocity and the gap: c~)
and of the normal stress, which is close to the actual pressure which has
order of magnitude of £-2.
289
Let us define the c-independent vector G(x,y) = (G 1,G2,G3)(X,y) :
8G1 + 8G2 + 8G3 _ 0 G= 9 on 80,

8X1 8X2 8y - ,
and recalling that 93 = 0 on 80, then we can choose for Gc the lift
defined by GI(x, X3) = Gi(x, y) for i = 1,2 and GHx, X3) = cCh(x, y).
Then the problem (9)-(10) leads to the following form: Assuming (11),
there exists a unique u in K(n) and fl in L6(0), such that
C
In q div( uc ) dxdy = 0 Vq E L6(n) (12)
2
L ro-i,j 88,x) (rpi -
i,j=l in
uDdxdy + 1 w
kS(o-;)(lrp - sl-Iuc - sl)dx+
(13)
where for 1 :::; i,j :::; 2

,c 2 c
8 'c 8uc) ) 'c s:
2 (U i 'c
c 8'ci
(U 8'c
2 U3)
(Ji)' = C; (Ji)' = C; l/ -8 + -8 -p Vi,j, l/ -8 +C; -
(Ji 3 = C;(Ji 3 =•
, , Xj Xi "y 8Xi
Theorem 2 Assuming (11), then there exists ui in Vy (i = 1,2), and
p* in L6(0), where Vy = {7P E L2(0) : ~ E L2(0)}, such that as c; - t 0
we have:
ui -" ui (1:::; i :::; 2) weakly in Vy (14)
(15)
8u§ -" 0
c;- (16)
8y ,
fl -" p* weakly in L6(0) (17)

Proof First we get estimates on the velocity by choosing rpi = Gi for
i = 1,2 and rp3 = C;G3, in (13), so we obtain:
c;211 8ul 112 + c;211 8ul 112 + c211 8u~ 112 + II 8ul 112 + II 8u~ 112 +
8Xl 8X2 8X1 8y 8y
8'c 8'c 8'c 8'c
+c;211~112 + c;211 u3 112 + c;411 U 3 11 2 + c;411 u3 11 2 :::; Co. (18)
8X2 8y 8Xl 8X2
290
where 11.11 denotes here the L2-norm in H, and Co does not depend on E.
Then choosing 'ljJ E HJ(O), we get:
ail
(i=I,2), II-a I
y H-l(fl)
~ E.C2. (19)
Theorem 3 Assuming (11), we have
02U~ op*
_ v _l + - = 0 (i = 1,2) in L2(0), (20)
oy2 OXi
moreover the traces s* := u*(x, 0) and 7* := &&~ (x, 0) satisfy
L(ks( -p*)(I'ljJ + s* - sl-Is* - sl) - V7*'ljJ) dx 2: 0, V'ljJ E (L2(w))2

(21)
V17*1 = ~S( -p*) ===> :3'\ 2: 0 s* = s+ '\7* } a.e. in W (22)

v17*1 < kS( -p*) ===> s* = S
h2
2""\lp* + vs* + Vh7* = 0 a.e. III w. (23)
where
rh(x)
jJi(x) = Jo fJi(X, y)dy.
Proof. See Bayada-Boukrouche (2001).
4. Study of the uniqueness

In this section, we will give another formulation of the limit inequalities
for s* and 7* on w which enables us to express s* as a solution of a
"non linear" variational inequality of the second kind with a convenient
decomposition.
Theorem 4 Let h E LOO(w) n Hl(w), then there exists k* given by (31)

such that for Ilkllvx,(w) ~ k*, the solution s* is uniquely given by
(25)
291
where U = (C, D) is the solution of the following "non linear" variational
problem: Find U E Hl(w) X H6(w) such that
a(U, ¢ - U) + J(¢) - J(U) ~ £(¢ - U) V¢ E Hl(w) x HJ(w), (26)
a(U, ¢) = 14vh3VCVCPdx + 1 vh-3curl(D)curl(8)dx, ¢:= (cp,8),
=j r 6vg.ncp.
J(¢)
wkS( -p*)lh Vcp + h- curl(8) - sldx, £¢ =
2 1
kw
Proof. It follows the general line of the one of theorem 5.1 in Bayada-
Boukrouche (2001). The first step is to prove that any function in
(L2(w))2 can be written as (25) so that (C, D) satisfies (26). To prove
uniqueness of s* ( and then of p* and T*) we have to prove uniqueness
of C and D. Let U1 , U2 be two solutions to (26), then:
then, there exists a constant A!, which depends on the regularization

1
operator S and the continuous injection from H- 2 onto L2 such that
where hmax = maxw h, hmin = minw h. And from (23)-(24), we have
Choosing cp = P2 - pi in (28), we get
IIV(P2 - pi) 11£2(w) ~ h~v hmaxmax(h~ax' h~~n) IIV(U2 - U1 11£2(w) (29)

mzn
Using the Poincare's inequality in Hl(w), and the fact that pi E L6(w)
for i = 1 or 2, there exists a constant A2, which depend only on w, such
that
(30)
Combining (27), (30), and (29), we deduce that if Ilkllux,(w) ~ k*, where
(31 )
292
then
therefore s~ - si = O.
Remark 2 For IlkllL'xl(w) ::;
k*, under the same hypothesis of theorem
4, the uniqueness of p* follows from
1(W
h3 \lp* - '3:. s*) \ltpdx = [ tpg.n. \;ftp
1211 2 Jaw
E Hl(w), (32)
which can be obtained from (23) and (24). And then 7* is also unique,
from the uniqueness of p* and s*, by using (23).
References
Assemien, A., Bayada, G., and Chambat, M., Inertial effects in the asymptotic be-
haviour of a thin film flow. Asymptotic Analysis, 9, (1994), 117-208.
Bayada, G., Boukrouche, M., On a free boundary problem for Reynolds equation
derived from the Stokes system with Tresca boundary conditions. 325, (2001), on
http://numerix.univ-Iyon1.fr / publis / publi v/ publis.html
Bayada, G., Lhalouani, K. Asymptotic and numerical analysis for unilateral contact
problem with Coulomb's friction between an elastic body and a thin elastic soft
layer. Asymptotic Analysis, 25 (2001) 329-362.
Bresh, D., Lemoine, J., and Simon, J., Ecoulement engendre par Ie vent et la force de
Coriolis dans un domaine mince: I Cas stationnaire. C.R.A.S. Paris, t.325, Serie
I, (1997), 807-812.
Consiglieri, L., Stationary Solutions for a Bingham Flow with Nonlocal Friction. Pit-
man Research Notes in Mathematics Series 274, (1992), 237-243.
Duvaut, G., Equilibre d'un solide elastique avec contact unilateral et frottement de
Coulomb. C.R.Acad.Sc.Paris, t 290, (1980), 263-265.
Ekeland, I., Temam, R., Analyse convexe et problemes variationnels. Dunod et
Gauthier-Villars, Paris, {1974}.
Girault, V., Raviart, P.A., Finite element Approximation of the Navier-Stokes Equa-
tions. Springer-Verlag, {1979}.
Pit, R., Mesure locale de la vitesse a l'interface solide-liquide simple: Glissement et
rle des interactions. These Physique, Univ. Paris XI (1999).
Strozzi, A., Formulation of three lubrication problems in term of complementar-
ity. Wear, 104, (1985), 103-119.
ELASTODYNAMIC FRICTION PROBLEM
WITH A "SURFACE INERTIA"
PERTURBATION
Jean-Claude Paumier
LMC-IMAG, BP 53X
38041 Grenoble, France
Yves Renard
MIP, INSA de Toulouse
31077 Toulouse, France
Abstract We consider the dynamic process of the frictional sliding of an elastic

body on a rigid foundation. Friction is modeled with a Coulomb law
with a possible dependence on the slip velocity. To keep the possibil-
ity of occurence of velocity discontinuities we use the elastodynamic
(hyperbolic) framework. Nevertheless this does not allow to get a well-
posed problem. To make up for this we perturb the solution of the
elastodynamic problem in a thin layer next to the contact boundary.
This is a generalization of an approach precedently given in a one-
dimensional case. We establish existence and uniqueness results for
this perturbed problem and give an a priori estimate and results for
some non-perturbed problems.
1. Introduction
Here, we present an approach of the elastodynamic friction problem
(Le. not in the visco-elastic or quasi-static framework) which considers
a regularization in a layer of small thickness localized on the contact
zone. We called this regularization a surface inertia perturbation. A
simple geometry is considered and also regularized friction and unilateral
contact conditions.
The main motivation is to try to generalize theoretical and qualitative
results established in Ionescu and Paumier (1994) and in Renard (2000)
for the one-dimensional problem to the multi-dimensional case. In the
one-dimensional case, the analysis of the problem showed that the use
293
JA.C. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 293-300.
294
of a non-monotone slip dependent friction coefficient in the purely elas-
to dynamic problem introduces a multiplicity of the solutions and shocks
in velocity. This non-uniqueness is still present if the friction and unilat-
eral contact condition are regularized. The one-dimensional case has the
interest to show precisely the regularity which can be expected for the
solutions. In this case, it has been proved that the surface inertia allows
to recover the uniqueness of the solution and, when the perturbation
parameter goes to zero, the solution tends to a particular solution to the
non-perturbed problem which is related to the perfect delay criterion in-
troduced for this problem in Ionescu and Paumier (1994). Unfortunately,
this criterion has not a clear extension to the multi-dimensional case, and
the discussion is still open to know whether or not non-uniqueness can be
observed with a non-decreasing friction coefficient (this is not the case in
the one-dimensional case) and whether or not stick-slip instabilities can
be observed with a constant friction coefficient (positive elements are
presented in Renard (1998), Moirot and Q.-S. Nguyen(2001), Simoes
and Martins (1998) and in Martins et al. (1999)).
2. Description of the problem

°c ~nis a domain representing a linearly elastic body submitted to
a Neumann condition on r N' a Dirichlet condition on r D and a unilateral
contact with Coulomb friction condition on r c between the body and
a Hat rigid foundation, where r N' r D and r c are non-overlapping parts
of 80, the boundary of 0. The evolution of u(t, x), the displacement of
the body, is submitted to the following equations:
pu - div u(u) = j, in jO,Tj x 0,

u(u) = Ae(u), in jO, Tj x 0,
u{u)n = g, on jO, Tj x r N'
u=u D ' on ]0, Tj x r D'
uN(u) E -IN(U N), on jO,Tj x r c '
uT(u) E UNIt(luT - 7"T I) DirT (U T - 7"T)' on jO,Tj x r c '
I
lisboabstract.tex using the following multi-valued functions:
{O} if ~ < 0,
{I~:I}' if vT # 0,
e
[0, +00[, if = 0, and DirAv) =
{Wi Iwi ~ l'~N = O},
¢, if ~ > 0, If vT = 0,
295
and with p the density, u(u) the stress tensor, e(u) the linearized strain
tensor, A the elasticity tensor which satisfies usual conditions of sym-
metry and coercivity, n the outward unit normal to 0 on 80, u(u)n =
uN(u)n+uT(u) and u = uNn+uT on ro the usual decompositions in
normal and tangential part on r 0' u D, g, f are given, TT the tangential
velocity ofthe rigid foundation, and J.t(.) the friction coefficient which is
allowed to depend on the sliding velocity (itT - T+). It is assumed that
this dependence is such that J.t(') lies in W1,OO(lR ), but no assumption
is made for the friction coefficient to be small enough.
The following regularizations of JN and DirT correspond to classical
regularizations of unilateral contact and friction conditions:
0 if ~ < 0,
{
~(~) = ~
- if ~ ~ o.
."
This corresponds to a Yosida regularization of the two monotone maps

I N and DirT. Denoting, for shorter notations, J.t l1 (v) = J.t(lvI)Dir~(v),
we introduce the regularized unilateral contact and friction conditions
on r a as:
UN -~(UN)'
UT -~(UN)J.tl1(itT - TT),
where j~ is equal to J~ in the two-dimensional (n = 2) case, and to
j~ (UN) = min(J~ (UN)'!)'

."
in the three-dimensional case, to avoid technical difficulties in the exis-
tence and uniqueness proof.
3. Simplified problem
For the sake of simplicity, and as a first approach to avoid difficulties
coming from the geometry, it is assumed that 0 is a cylinder (see fig. 1 ),
i.e. 0 = wx]O, D[ where w is a domain of lRn - 1 and D > 0 is the height.
We assume that r D = W x {D}, r N = 8wx]0, D[ and r a = w x {O}. We
denote x = (x',x n ) where x' E wand Xn E ]O,D[.
It is also assumed for simplicity that p == 1, and even if it means a
translation of the solution that TT(t) = 0, UD = 0, u(O) = it(O) = o.
296
Xn =
Xn = 0
Figure 1. Simplified geometry.
Setting the following spaces:

Ho = L2(O;lin), Va = {v E Hl(O;lin);v = ° on r D}'
with their classical norms denoted by IlvllHo and IIvllvo' we will say that
the displacement u is a solution to the regularized friction problem if it
is sufficiently smooth (in a sense which has to be defined) and if it is
solution (in the sense of distributions in time) to the following equations:
Find u : ]0, T] ---+ Va such that
u(t) + Au(t) = B(t) +.ry, (t) + Fi(t), in V~,
(.ry,(t),v)vJ,vo = - { J2(u N (t))v N du, (1)

ire
(.r;!(t),v)vJ,vo = - ( J2 (UN (t))/-t17(uT (t)).vT du,
ire
u(o) = u(o) = 0,
where
(Au, v)vJ,vo = a(u, v) = In u(u) : e(v)dx,
and B(t) contains all the other data. It is assumed that B(t) satisfies
B E LOO(O, T; V~), BE LOO(O, T; V~). (2)
297
See Duvaut and Lions (1972) for details concerning convenient regu-
larities to obtain this condition. A major difficulty for Problem (1) is
that the relations linking tangential and normal forces on the contact
boundary to normal displacement and tangential velocity are not a priori
well-posed in the proposed functional framework.
4. Surface inertia perturbation

What we call a surface inertia perturbation is in fact the generalization
of the perturbation introduced for the one-dimensional case in Renard
(2000). In that case, the perturbation can be viewed as the addition
of a density of mass on the contact surface. The idea is to perturb the
problem in a thin layer of thickness c
Ec = wx]O,c[ CO,
and to introduce the spaces
where !l0 is taken in the sense of distributions. It corresponds to

uXn
functions which are independent of Xn in Ec (It is possible to make a
similar construction with higher degree polynomials, see Paumier and
Renard (1999)). Spaces Hc and Vc are closed sub-spaces of respectively
Ho and Vo and Vc is densely included in Hc. The most interesting
property of spaces Hc and Vc is:
Lemma 1 The following maps
Ic1 : TT
Vc ----->..
---r Hl (rc.' lIl>n)
~ 2
Ic: -----'- L2(rc·,lIl>n)
H c ---r ~
and
V 1---7 V 1---7
are linear continuous such that
This can be assimilated as reinforced traces on r c' and gives some

sense to contact and friction relations. The perturbed problem of friction
298
is Problem (1) but set in the space Ve:
Find ue'T/ : ]0, T] ~ Ve such that

ue'T/(t) + Aue'T/(t) = B(t) + F~'T/(t) + F;'T/(t) , in V;,
(F~'T/(t),v)V;,Ve = r -JZ(u~'T/(t))vNd(T,
ire
(3)
(F;'T/(t),v)V;,Ve = r -jZ(u~'T/(t))J.t'T/(u~'T/(t)).vTd(T,
ire
ue'T/(O) = ue'T/(O) = 0,
We will also use in the following:
The term "surface inertia" is motivated by the fact that this pertur-
bation can be viewed as the addition on the contact surface of a surface
operator (Submitted paper).
5. Existence and uniqueness result for the

perturbed problem
°
Theorem 1 Under assumptions J.t(') E W1,00(lR+) and (2) Problem (3)
°
has a unique solution ue'T/(t, x) for all c > and'fJ > which belongs to
Loo(O, Tj Ve) n W1,00(0, Tj He).
The principle of the proof comes from the classical existence and
uniqueness result for initial value problem of differential equations
(Cauchy-Lipschitz theorem). We show, thanks to an a priori estimate
and a reformulation of the problem in terms of fixed point, that a p-
iterated operator is a contraction in a certain closed ball of Loo(O, Tj Ve) n
W1,00(0, Tj He). The reinforced traces on replay an important role.
Detailed proofs an be found in Paumier and Renard (1999) and in the
submitted paper.
6. A priori estimate
Proposition 1 Under the same assumptions as Theorem 1, ue'T/(t) be-
ing the unique solution to Problem (3) belonging to Loo(O, Tj Ve) n
W1,00(0, Tj He), there exists a constant C > °
independent of c and
299
1} such that for almost all t in (0, T) one has
IIu el1 (t)IIHo + Iluel1 (t)llvo + [ 21 (u~l1(t))~dO"

1re 1}
+ [t [ J~ (u~l1(t))hl1(lu~l1(t)l)dO"ds ~ C,
10 1re
where hl1 : R+ ~ R+ is defined by hl1 (e) = J.t(e) e
1}
if e< 1}, and hl1 (e) =
e
J.t(e)e if ~ 1},
In particular the {double} sequence u el1 is bounded in Loo(O, Tj Vo) and
the sequence u el1 is bounded in Loo(O, Tj Ho).
The proof is based on energy conservation. Due to the lack of regu-
larity, a finite dimensional approximated problem is first considered.
7. Dynamic Signorini Problem with penalization

What we call the dynamic Signorini problem with penalization is
Problem (1) without the friction term (i.e. with J.t == 0):
Find U : ]0, T] ~ Vo such that
{ u(t) + Au(t) = B(t) + FN (t), in V~,
(4)
(FN(t),~)V&,Vo = fre -Jl1(u N(t))v NdO",
u(O) = u(O) = 0,
From section 4, we know that this problem, if posed in Ve (Le. Prob-
lem (3) with J.t == 0), has a unique solution u el1 (t). From section 5, we
know that u el1 (t) is uniformly bounded in Loo(O, Tj Vo) and u el1 (t) is uni-
formly bounded in Loo(O, Tj Ho). As a consequence, u~l1(t) is uniformly
bounded in Loo(O, Tj Hl/2(r e)) and F~ also, because J~ maps continu-
ously Loo(O, Tj Ho) into itself. Thus passing to the weak star limit on c
(on a subsequence), we establish the following result.
Theorem 2 Problem {4}, under the same assumptions as Theorem 1
has at least one solution belonging to Loo(O, Tj Vo) n W1,00(0, Tj Ho).
Remark: A similar approach seems to be possible for the dynamic Sig-
norini problem without any penalization. But a difficulty is still present
for the complementarity equation.
8. Conclusion
The proposed perturbation in an arbitrary thin layer near the contact
boundary, allows to regain the existence and uniqueness of the solution of
300
the dynamic problem of friction, at least for the simplified geometry we
considered, and for an arbitrary large friction coefficient. It is known, for
such a problem, that uniqueness is lost for large coefficients of friction.
So it should be interesting to know if, as it is the case in dimension one
(see Renard (2000)), the solution of the perturbed problem tends to a
particular solution of the non-perturbed problem related with a certain
criterion.
In any case, this perturbation seems to be a useful tool to deal with
dynamic contact and friction problems.
References
G. DUVAUT, J.L. LIONS (1972), Les inequations en mecanique et en physique, Dunod
Paris.
I.R. IONESCU, J.-C. PAUMIER (1994), On the contact problem with slip rate dependent
friction in elastodynamic, Eur. J. Mech., A/Solids, N° 4, pp 555-568.
J.A.C. MARTINS, S. BARBARIN, M. RAous, A. PINTO DA COSTA (1999), Dynamic
stability of finite dimensional linearly elastic systems with unilateral contact and
Coulomb friction, Compo Meth. Appl. Mech. Engng., vol. 177, pp 289-328.
F. MOIROT, Q.-S. NGUYEN (2000), An example of stick-slip waves, C.R. Acad. Sci.
Paris, II B-Mec., 328 (9), pp 663-669.
J.-C. PAUMIER, Y. RENARD (1999), F'rottement Elastodynamique avec Inertie de Sur-
face. Rapport IMAG RT 179, Universite Joseph Fourier, Grenoble, France.
J.-C. PAUMIER, Y. RENARD (2001), Existence and uniqueness result for a perturbed
purely elasto-dynamic problem with friction. Submitted to the European Journal
of Applied Mathematics.
y. RENARD (1998), Modelisation des instabilites liees au frottement sec des solides
elastiques, aspects thioriques et numeriques., These de doctorat, LMC-IMAG Greno-
ble.
y. RENARD (2000), Singular perturbation approach to an elastic dry friction problem
with a non-monotone friction coefficient, Quarterly of Applied Mathematics, vol.
LVIII, W 2, pp 303-324.
F.M.F. SIMOES, J.A.C. MARTINS (1998), Instability and ill-posedness in some friction
problems, Int. J. Engng. Sci., 36, pp 1265-1293.
SOLVABILITY OF THERMO-VISCO-
ELASTIC CONTACT PROBLEMS
WITH COULOMB FRICTION AND
NONLINEAR HEAT CONDUCTIVITY
Christof Eck
Institute of Applied Mathematics,
University Erlangen-Niimberg
Jifi J arusek
Mathematical Institute,
Academy of Sciences of the Ozech Republic
Abstract The solvability of a coupled thermo-viscoelastic system including con-

tact and friction is outlined. The nonlinear growth of the viscous, fric-
tional and deformation heat occuring in the system is compensated by
a certain superlinear growth of the solution--dependent diffusion coeffi-
cients.
1. Introduction
In models for contact problems with friction it is often necessary to
include the aspect of heat transport, because friction may generate a
non-negligible amount of heat. Since the solvability of unilateral dy-
namic problems with Coulomb friction seems to be up-to-now estab-
lished only for material having some viscous behaviour (cf. Jaruaek and
Eck (1999)), a suitable model for such a problem consists of a viscoelastic
contact problem and a heat equation. As in the cited paper, the approx-
imate contact condition in velocities is employed here. This limits the
applicability of the presented result in the manner mentioned there.
Let us formulate the problem to be considered. We assume a bounded
domain n whose Lipschitz boundary r is composed of three measurable
mutually disjoint parts ru, rT and re. Let Irr = (O,!) be a bounded
time interval. Let Qrr == Irr )I n be the appropriate time-space domain
301
I.A. C. Martins and M.D.P. Monteiro Marques (eels.), Contact Mechanics, 301-308.
302
and let S'! denote its lateral boundary consisting of the parts SX,'! ==
I,! x rx for X = U, T, C. For TEl,! an analogous notation In Q-r, S-r
will be used. We shall employ a usual summation convention, dots for
time derivatives, ',i for the derivatives with respect to i-th variable, u for
a displacement, 6 for temperature, n for the unit outer normal vector
and subscripts n and t for normal and tangential components of vectors.
We assume the linear strain-stress relation of the Kelvin-Voight type
(1ij == (1ij(U, 8) = a~nlekl(u) + aUklekl(U) - bij6, i,j = 1, ... , N, (1)
with eij(u) == ~(UiJ + Uj,i)' With this notation the problem has the
following classical formulation:
Find a couple [u,8] such that the following relations are satisfied:
Uj - (1ijJ(U, 6) = Ii, i= 1, ... ,N, in Q'!, (2)
u=U on SU,,!, (3)
T(u) = h on Sr,'!, (4)
Un ~O, Tn ~O, Tnun = 0,
Ut = 0
Ut7~O
=}
=}
ITtl ~ ~ITnl,
Tt = -~ITnllutl
Ut } on Sc,'!, (5)
u(O, x) = uo(x), u(O,x) = Ul(X) for xEn, (6)

e- (Cij8J),i = a~;lleij(u)ekl(u) - in Q'!,bij 8uiJ (7)
Cij6Jni = K(T - 8) - R(6) on SU,,!U Sr,'!, (8)
Cij6Jni = ~ITnIIUtl + K(T - 6) - R(6)on Sc,'!, (9)
6(0, x) = 8 o(x) for xEn. (10)
The generalized Lame system and the heat equation are here coupled by
the temperature-dependent stress tensor, by volume terms describing
the generation of heat by viscous deformation and by thermal stresses
in the viscoelastic material (in (7», and by boundary terms modeling
the generation of heat by friction in (9). Here and in the sequel T
denotes the boundary traction. The tensors {a~Jkl} and {a~;kl} are
assumed to depend Lipschitz-continuously on the space variable and to
· ·I.e. aijld
be symmet rIC, (£) as we11 as b ounded and e11'Ipt'IC,
(£) = a(£) = aldij'
jikl
i.e.
a(£)C .. C.. < a(£) C . • ckD < .d(£)c . .C.. (11)
o "'3"'3 - ijkl..'3 .... -""'"0 "'3"'3
for all symmetric tensors {eij} E RN,N with real constants 0 < a~£) ~
4£), " = 0,1. The tensor {bij} of thermal expansion is symmetric,
303
Lipschitz with respect to the space variable and globally bounded. The
tensor of thermal conductivity Cij is assumed to be symmetric and locally
Lipschitz-continuously solution dependent. Two different models of this
dependence are studied:
Case A (cf. Eck (2000)) Cij == Cij(9) such that
eo(l + lel'Y)eiei ~ Cij(e)eiej ~ Co (1 + lel'Y)eiei (12)
for all vectors {ei} E]RN and all x E n with 0 < eo ~ Co < +00.
Case B (cf. JaruSek (2000)) Cij == Cij(Ve) satisfies the growth condition
Cl (1 + IV91'Y) eiei ~ Cij(Ve)eiej ~ C2 (1 + IV91'Y) eiei, e E JRN, (13)
the strong monotonicity
(cij(V9)9,j - Cij(VS)S,j, e,i - S,i}Q$
~ c311V(9 - S)III~!2(Q$) + c411V(9 - S)IIL(Q$) (14)
for each e, S E L'Y+2 (i'I'j W~+2(n)), and the continuity relation
Cij(Ve(k»)9~) --+ Cij(V9)9,j in L'Y*(Q'I'), i = 1, ... ,N, (15)
for 9(k) --+ 9 strongly in L'Y+2 (I'I'j W~+2(n)) and '1* = (-y+2)/(-y+
1).
In both cases '1 is a positive number. The aim is to minimize '1 with
respect to the dimension N.
2. Variational formulation of the problem

In the sequel W;(M) denotes the Sobolev-Slobodetskii space on a
domain (or a sufficiently smooth manifold) M for k ~ 0 and p ~ l.
For k = 2, Hk(M) == Wf(M) and W;(M) == W;(Mj ]RN). The duals
of these spaces are marked by asterisks. The brackets (.,.) M denote
a generalized L2(M)-duality pairing. For 0, fJ ~ 0 and an interval I,
HQ,f3(I x M) == HQ(IjL2(M)) nL2(IjHf3(M)). By Bo(IjX) we denote
the space of bounded and weakly continuous functions I --+ X for a
Banach space X. The following spaces and sets are important for the
variational formulation of the problem
U:= {v E H 1/ 2 , 1 (Q'I')j V = 0 on Su,d (16)
.ft:= {v E iT + Uj Vn ~ 0 a.e. on Sa,d (17)
9J := (L2(hj Hl(np in Case A, (18)
L2+'Y(I'I'j w2+'Y(n)) in Case B.
304
We introduce the linear form ,£ : v t-+ i.Q

~
/ivi dx dr + Is F,~
hiVi dx s dr
and for t = 0,1 the bilinearforms a(~) : [u, v] t-+ IQ~ a~j~leij( u)ekl( v )dx dr.
Definition 1 A couple [u, e] satisfying the initial condition (6), (10)
and such that u E Bo(I't; HI (0)) with U E Bo (I't; L 2(O)) n~, u E
L2(h;H- I (0)) nH I / 2 (h;L2(O))* and e E mnBo(I't;L2(O)) will be
called a weak solution of the problem, if for each [v, cp] E ~ x 2lJ the
following relations are valid:
{Ui' Vi - Ui)Q~ + (Uij(U, e), eij(v - u))Q~ (19)
+ {JITn(u,e)I,lvtl-lutl)sc,~ ~ £(v-u),
(8, cp)Q~ + {Cije,j, CP,i)Q~ + {bij eUi,j, cp)Q~ (20)

+ {K(e-T)+R(e),cp)s~
= (amleij(u)ekl(u), cp)Q~ + (JITn(u, e)llutl, cp)sc,~·

Here 2lJ = Loo(h; W~(O)) for Case A while 2lJ = mfor Case B.
3. Result
For the generalized Lame system we assume:
Assumption 1 In addition to the above mentioned conditions on the
domain 0, its parts of boundary rx, X = U, F, C and the coefficient
functions a~j~l' bij and Cij, we assume re E c2+{3 for some f3 > 0 ar-
bitmrily small, f E L 2(Q't) UHI(I't;HI(O)*), hE HI (h; H I / 2(rT)*) ,
UO,UI E H3/2(O) and U E H2(Q't) such that U E L2(I't; HI(O)). More-
over, U satisfies the compatibility conditions U = 0 on Se,'t, U(O,') = Uo
and U(O,.) = Ul on O. The coefficient of friction J =
J(x,u) is a
nonnegative function satisfying the Camtheodory conditions, supp J c
=
re,w x ~N with re,w {x Ere; dist(x, are) ~ w} for some w > 0 and
IIJIIL oo (rc) < C'J, where the constant C'J (always dependent on the viscos-
ity of the material only) is given in Jarusek and Eck (1999), Proposition
4 and formula (4.23) for an anisotropic material and in Eck and Jarusek
(1998) for an isotropic material in two dimensions.
For the heat equation we assume:
Assumption 2 The input data satisfy eo
~ 0, 0:$ T E L2(S't), Ko :$
K E Loo(r) with a constant Ko > O. The mdiation R R(x, e) satisfies =
the Camtheodory conditions, R(·,O) = 0, R =
0 or R(x,') is increasing
on [0, +(0) and satisfies the growth condition
cller R :$ R(e) :$ c21erR + C3 for e~0
305
for an exponent rR > ro ~ 1 and'Y > 'Yo. In Case A we assume eo E

L2(0), R = 0 and 'Yo = 1- *" - *"
N~l or ro = 2 + and 'Yo = 1- j. In
Case B we require eo E L2+'Y(I~; Wi+,iO)) n Hl/2(Q~), for N = 2 we
assume R = 0 and 'Yo = (v'12.2 - 3)/4 or ro = ~ and 'Yo = 0 and for
N = 3 we assume R = 0 and 'Yo = ~.
We remark that the last requirement for 'Y in the case N = 3 cannot be
weakened by any radiation condition. Moreover, we assume that for the
Lame system the following estimate holds
(21)
where the constants depend on the input data only. The localization
and shift technique with the help of some interpolation proves such an
estimate e.g. if r = re. It holds also along possible other parts of S if
r E C2+f3. Here, however, the result needs some additional requirements
concerning the character of the common parts of boundaries of rx, X =
U,T,C.
The main result is
Theorem 1 Let Assumptions 1 and 2 and the estimate (21) be satisfied.
Then there exists a weak solution of the problem (cf. Definition 1).
Remark 1. The strong assumptions about U and £ are related with the
"energetical" method of the proof, in particular with the estimate (26)
below. These assumptions can be weakened, but then a larger exponent
'Y (i.e. a higher growth for the heat energy) must be required.
2. It is easy to see that for ~ which satisfies ~(x, Un, 0) ~ ~o{x, Un, 0) ==
.lim ~(x, Un, Ut) for any x E n and Un ~ 0, every solution to the problem
'l£t-+O
with the coefficient of friction ~o, defined as ~ for Ut =I 0, is a solution
of the problem with the coefficient ~, if ~o satisfies Assumption 1. This
allows to have a coefficient of friction of stick greater than the corre-
sponding coefficient of friction of slip within the presented result.
4. Scheme of the proof

Three approximations are employed in order to prove Theorem 1. The
Signorini contact condition is replaced by
O"n{u) = -i[un]+ with [.]+ := max{·, O} and 0> O. (22)
The Euclidian norm in the friction term is replaced by
lxi, Ixl ~ 17,
«PI) : x I--t ( Ixl 4 31xl 32 (23)
- 817 3 + 4;J + 817, Ixl < 17,
306
which is convex, twice continuously differentiable and satisfies 4>1'/ ~

4>0== I. I in any Co. (JRN ), n E [0, 1) for 1] ~ O. Finally, the growth of the
"coupling" terms is limited with the help of a function
¢11 : y H min{y, 'I?}, y ~ O.
Then the original problem (19, 20) is approximated by the following
auxiliary problem:
Find a couple [u,9] with [u(O, .), 9(0, .)] = [uo,9 0 ], u(O,·) = Ut, U E
U+11,9 E 21, e
E 21*, U E 11*, such that for all [v,rp] E 11 x 21 there
holds
(Ui,Vi}Qx + a(O)(u,v) + a(l)(u, v) - (bij¢11(9), eij(v)}Qx (24)
+ «(l/O)[un]+, vn}sc,x + «(~/8)[un]+ V<PI'/(Ut), Vt}Sc,x = ,c(v),
(9, rp)Qx + (Cij9J, rp,i}Qx + (bij¢11(9+)UiJ, rp}Qx (25)
+ (K(E> - T) + R(9), rp} Sx
= (¢11 (aUlt eij(u)ekl(U)) , rp) + (~¢11([un]+<PI'/(ut)/8), rp)s .
Qx c,x
This problem is solved via the Galerkin approximation. Due to the
boundedness of the "coupling" terms it is not too difficult to prove cer-
tain standard a priori estimates both for the system (24, 25) and for its
finite-dimensional version and to perform the limit procedure to obtain
the existence theorem for this auxiliary problem.
The crucial non-negativity of the temperature is then proved testing
(25) by rp = XT()~L with 9_ := min{O, 9}, Xro == 1 on Iro and XT()(r) = 0
for r f/. Iro. This proves immediately E>_ = 0 (the signs of K, T, E>o etc.
are important here). Then we put [v, rp] = XT()[u, 1] in (24, 25). After
application of the Gronwall lemma this yields the energy estimate
u
lI IIL,,(Ix;L2(n» + + 11 9 I1L oo (Ix;Ll(nn + 11 9 I1 L l(sx)
lI u lli oo (Ix;H 1 (n»
+ 11911~~R(Sx) ~ C7 (26)
with a constant independent of the approximation parameters 'I? , 1] and
6, because all "coupling" terms disappear. We remark that the last
term on the left-hand side of (26) only occurs if the nonzero boundary
radiation is assumed.
This very weak estimate, clearly insufficient for the limit procedures,
is, however, a starting point for further estimates. In Case A these
estimates start with the help oftest function rp = XroE>ao with no = 1,
no = N~l in the case with and without radiation, respectively. The
choice of the exponent is such that the estimate (21) can be employed.
The following interpolation-imbedding lemma is used in the estimates:
307
Lemma 1 (i) For v E H 3/ 4,3/2(Qrr) there holds
IIV'vIILp(I'X;Lq(Sl)) ~ cllvII H 3/4,3/2(Q'X)'

if p, q ? 2 and ~ + 2: ? N + l.
(ii) Let cp E H 1/ 2 ,l(s,:r) and p, q ? 2 with ~ + 2(~-1) ? N - 1. Then
IIcpIILp(h;Lq(r)) ~ cllcpIlHl/2, 1(8'X).
(iii) For cp E Wi(O) with J~l ~ p <N there holds
IIcpIIH1/2(Sl) ~ cllcpII£l (Sl) IIcpII~J~Sl)

with), = ~ - 2(N(p'!1)+P).
The proof of the lemma is a combination of a series of interpolations and

the imbedding theorem.
The main task is to estimate the "coupling" terms with the help of
(21) only (disregarding the artificial limitation oftheir growth). There is
, not far from ,0

no space here to describe these technicalities. This procedure yields for
a certain uniform estimate which is further improved
by a new test function cp = Xroe1--y (for a larger, it is the only test
function employed). Finally the estimate
lI e Il Loo (I'X;L2_-Y(Sl)) + II ell L2 (I'X;H1 (Sl)) + lI e II LrR +l--y(8'X) ~ c, (27)

11 8 I1 Loo (I'X;L(N+2)/(N+1)(!1)) + 118I1L2(hjWJ(!1)) ~ c (28)
(with p = 2N~i~~~+1) in the second relation) for the case with and
without radiation, respectively, is obtained with a constant indepen-
dent of the approximation parameters {j, 'fJ and~. These and the ap-
propriate dual estimate derived with the help of test functions from
If
Loo (h; if" (0)) with '" > 1 + finally leads to the a priori estimate
lIuIlHi'!(Q'X) + lIellLoo(I'XjL2_-Y(Sl)) + II e ll H8,1(Q'X)
+ lIeIlLrR+l_-y(8'X) ~c
with 0 < s < 4~N and a constant independent of {j, 'fJ and~. This
estimate enables to perform successively the necessary limit procedures
for all approximating parameters and Theorem 1 is proved in this case
for, E (To, 1). The case, ? 1 is not so difficult.
In Case B the scheme of estimation is similar. For N = 2 the most
difficult term to be estimated is the frictional heat. For R = 0 we put
308
cP = X7b 8ao with 00 = ! - ~, 'Y = ! for'Y E ('Yo, ~), 'Y ~ ~, respectively.

The choice 00 = !
is sufficient also for the case with radiation with
'Yo = 0 and TO = ~. This leads after a careful estimation of the coupling
terms with the help of a series of interpolations and applications of
the imbedding theorem to an estimate enabling to employ (21). This
estimate is then improved by putting cp = Xro 8 to prove the a priori
estimate
(29)
with Cs independent of the approximating parameters. A certain time-

difference technique then proves the estimate
(30)
The estimates (29) and (30) are independent of the approximating pa.-
rameters. Together with (21) and (13-15) they enable the limit proce-
dure for those parameters and Theorem 1 is proved for this case.
For N = 3 the estimate of the frictional heat and the viscous heat
yield the same value for the bound 'YO. For 'Y > 1 the test function
cp = X7b8 is sufficient to derive successively the estimates (29) and (30).
For'Y E (~, 1) we start with cp = Xro 8 1/ 3 and then we employ cp = Xroe.
In such a way we again derive the estimates (29) and (30) and with them
and the limit procedures for the approximating parameters we prove
Theorem 1 for this case, too.
Acknowledgments
The second author gratefully acknowledges support by the Grant
Agency of the Acad. Sci. of the Czech Republic, grant No. 107 5005.
References
Eck, c., and JaruSek, J. (1998), Existence of solutions for the dynamic frictional con-
tact problem of isotropic viscoelastic bodies. Preprint Nr. 242, Inst. Appl. Math.,
Univ. Erlangen-Niirnberg.
Eck, C. (2000), Existence of solutions to a thermo-viscoelastic contact problem with
Coulomb friction and heat radiation. Preprint Nr. 271, Inst. Appl. Math., Univ.
Erlangen-Niirnberg.
JaruSek, J., and Eck, C. (1999), Dynamic contact problems with small Coulomb fric-
tion for viscoelastic bodies. Existence of solutions. Math. Models Meth. Appl. Sci. 9,
11- 34.
Jarusek, J. (2000), Solvability of nonlinear thermo-viscoelastic contact problem with
small friction and general growth of the heat energy. Preprint Nr. 273, Inst. Appl.
Math., Univ. Erlangen-Niirnberg.
A SIMPLIFIED MODEL OF IMPACT
Jeongho Ahn
University of Iowa
David E. Stewart
University of Iowa
Abstract The standard Signorini contact condition is integrated against a given

function 'IjJ over the boundary to obtain a simplified model of contact
suitable for impact problems. An implicit method (implicit mid-point
rule for elasticity, and implicit Euler for the contact conditions) is pro-
posed to numerically solve the simplified model, and some properties of
the solution are obtained. The results are only partial at this stage, but
they seem to indicate that contact forces in elastic impacts are consid-
erably more regular than general measures.
1. Introduction
In this paper we consider a simplified model for impact where the
standard pointwise Signorini contact condition is replaced by an inte-
grated version which corresponds to an a priori normal contact force
distribution. Comparison is also made to some recent work of the au-
thors for the standard Signorini contact conditions. Proofs are omitted
for conciseness.
The simplified contact conditions have the form
o~ r 'IjJ(x)(n(x)· u(x, t) - g(x)) dS(x)

Jan
1.. N(t) ~ 0 (1)
where the contact force (or traction) on the boundary is - N (t) 'IjJ (x) n( x).
There is no friction in this model. The reason for this kind of contact con-
dition is that it results in a scalar (dynamic) complementarity problem
while keeping all the infinitely many elastic modes of vibration. Since
simulations appear to indicate that these elastic vibrations are a major
309
J.A.C. Martins and M.D.P. Monteiro Marques (eds.). Contact Mechanics. 309-316.
310
sink of energy in impacts, this should bring us closer to understanding

quantities such as coefficients of restitution in impacts.
First, we introduce the standard Signorini contact condition in the
case of linear elasticity (using density p == 1):
8 2u
{)t2
V . O"[u] + f(x) inn (2)
O"[u]· n N·n on8n (3)
O?u·n-g ...L N 5, 0 on 80. (4)
faan N(u·n-g)ds 0 for all t (5)
We use the following notation: n c R3 is the domain of the (unde-

formed) material body, u(x, t) is the deformation vector field for x E n,
0"[.] is the stress tensor operator, f(x) the body forces, N(x, t) n(x) the
contact force, and g( x) the gap function which is a measure of the "gap"
between the elastic body and the rigid foundation (or obstacle). The
stress tensor operator is given in linear elasticity by O"[U]ij = Eijklc[U]kl
using the Einstein summation convention, and cluj is the strain ten-
sor: c[WJij = !(Wi,j + Wj,i), where Wi,j = 8wi!8xj. The Hooke's ten-
sor Eijkl characterizes the material. Note that Eijkl has the symmetry
properties: Eijkl = Ejikl = Eijlk = Eklij. We assume that Hooke's ten-
sor Eijkl satisfies the following conditions (Oleinik et al., 1992, §1.3.1):
Eijkl E LOO(n), and there is a constant m > 0 such that almost every-
where in 0., EijklCijCkl ? mCijCij.
Note that since we are using linear elasticity, we are assuming that we
only have small displacements. Since we have no "prescribed displace-
ment" conditions, the body will eventually move far from the original
position and orientation; therefore, this model only has validity for small
time scales. However, that is sufficient for dealing with the behavior of
a body in impact.
The problem of elastic impact with the simplified contact conditions is
superficially similar to one solved in (Schatzman, 1973). In (Schatzman,
1973), an abstract version is considered where u(t) EVe H = H' = V'
with the condition that u(t) E K, a half-space in H. However, for the
elasticity operator H = L 2 (n), and the condition Jan 'I/J(n· u - g)dB 5, 0
defines a half-space in V = H 1 (n), not H.
Here vectors 0 5, a ...L b ? 0 means that a, b ? 0 componentwise and
that aTb = O. Note that Jan'I/J (u . n - g) dB and N in (1) are scalar
functions of time only.
311
2. The simplified contact condition and a

numerical discretization
The simplified contact condition replaces (3-5) above with
a[u](x, t) . n(x) 'I/J(x) n(x) N(t) for all x E an, and t, (6)
02: N(t) 1- ( 'I/J(x) (u(x, t) . n(x) - g(x)) dS(x) sO. (7)
Jan
If we use the mid-point rule for the elasticity operator and the implicit
Euler rule for the simplified contact conditions using ul(x) ~ u(x, tl) and
vl(x) = u(x, tl) for tl = to + 1h, we get
vl+ 1 _ vl l l
h
= \1·a [u +12+ u ] +1 in 0., (8)
ul+1_ ul 1
= _(vl+ 1 + vl) in 0., (9)
h 2
a[ul+ 1]. n 'l/JnNl on an, (10)
02: N l 1- 1an
'I/J(u1+1. n - g)dSSO. (11)
If A = - \1 . a [.], then 1 + A is a linear elliptic operator . Note that we

implicitly assume the zero traction boundary conditions for defining A:
(1 + A)u = 1 implies n . a[u] = 0 on an. Solving the above equations
for u1+1 and v l+1 in terms of ul and vl gives:
Now L:~1 (JjU~ 2: 0 by induction on l. where (J E H- 1(n) is the sur-

face measure 'I/J Van with Van the standard (d - 1)-dimensional Lebesgue
measure on an. In fact, if 'Yo : Hl(n) -4 Hl/2(an) is the usual trace
operator, then (J = 'Yo('l/Jn) where 'Yo : H- 1/ 2(an) -4 H- 1(n) is the
adjoint of 'Yo.
2.1 Modal formulation

We use a modal approach to modelling the elastic vibrations: Acpj =
)..j¢j where 0 S )..1 S )..2 S ... and {cpj }~1 is an orthonormal basis
in L2(0), with the boundary conditions a[cpj] . n = 0 on ao. Suppose
that 0 = )..1 = )..2 = ... = )"J < )..J+1 S )..J+2 S .... Then the first
312
J modes correspond to rigid-body modes. Then we can write u(x, t) =

2::~1 Uj(t) <pj(x) where Uj(t) = Jn u(x, t) . <pj(x) dV(x). Then we can
integrate <Pj against the equations of elasticity to obtain equations for
the Uj(t) coefficients:
Uj = r
Jan
N(t)'l/J(x)n(x)· <Pj dS - Aj Uj + fj(t) for all j. (13)
Let wJ
= Aj and Wj ~ o.
The simplified contact law (1) can be written in terms of the modal
coefficients Uj as
o~ L
00
{3jUj(t) + 'flo 1. N(t) ~ 0 for all t (14)

j=l
where {3j = Jan <pj(x) ·n(x)'l/J(x) dS(x) and 'flo = Jan 'l/J(x)'fl(x) dS(x). We
assume without loss of generality that 'flo = o.
3. Numerical approximation
The system that we want to solve is
(15)
L {3jUj(t) ~ 0,
00
o ~ N(t) 1. (16)
j=l
for all j = 1,2,3, .... Numerical approximations are needed to prove ex-
istence of solutions to this system of differential equations and comple-
mentarity conditions. Let v = U and Vj = Uj. Let h > 0 be the step-size
and tl = l h. Given u; ~ Uj(tl) and v; ~ Vj(tl) we wish to compute u;+1
and v;+l. To explicitly show the dependence on h > 0 we write these as
ul+l;h
j , Vjl+l;h ,etc. L et UI -_ (UI'
I I
u2' I
u3'··· )
an d vI -_ (VI'
I I I
V2' V3' ... ) ,etc.
As usual, £2 is the Hilbert space of sequences x = (Xl, X2, X3, •.• ) where
Ilxll = J2:: j IXjl2 is finite.
Now all we need to check is that the complementarity condition (11)
can be made to hold, where (using (12)):
00
L{3jU;+1
j=l
313
Let us assume that the energy in (u l , vl) is finite, and that fl is in £2.
Then as long as the sequence ((3jl(1 +Wj) I j = 1,2, ... ) is in £2, then all
sums above are finite. Also assuming f3 =1= 0 (that is, f3 = ((31, (32, (33,· .. )
contains a non-zero entry) makes l:: j (3J I (1 + h2 Aj 14) a finite, positive
number for any h > o. The complementarity condition (11) can then be
seen to have the form
o ~ N l .1 cl(h) + c2(h)Nl ~ 0
where c2(h) is positive. This is a simple one-dimensional complemen-
tarity problem and has a finite solution N l ~ O. Substituting this value
for N l shows that there are solutions to (8-11) provided f3 =1= 0 and
((3jl(1 + wj)li ~ 1) is in £2.
The energy E(u,v) =! l::~1(V] + AjU] + 2/juj) is dissipated in the
scheme (8-11); that is, E(u l+1,vl+1) ~ E(ul,v l ), provided L-j(3jU; +
'flo ~ o. Thus we can obtain numerical trajectories uh(t) by piecewise
linear interpolation of Uh(tl) = ul;h and vh(t) by piecewise linear inter-
polation of Vh(tl) = vl;h. The function Nh(t) = Nl for tl ~ t < tl+1 is
the numerical force function for this step size: Nl = (1Ih) Jt~l+1 Nh(t) dt.
4. Convergence theory
Taking (Stewart, 1998) as a guide, it would be desirable and appro-
priate to build on these results to give a complete convergence theory
for this simplified impact problem. Because we are dealing with an infi-
nite dimensional problem, some convergence issues become considerably
trickier than in the finite-dimensional case in (Stewart, 1998).
One issue that remains for the finite-dimensional case is that the func-
tions Nh do not converge to an ordinary integrable function, but rather
to a measure, such as a Dirac-8 function. It can be proven that, pro-
vided E(uO,vO) < 00, limsuPhloJoT Nh(t)dt <00. Since Nh is a scalar-
valued function, we can use the Helly selection theorem (Natanson, 1961,
Ch. VIII, §7, pp. 222-223) or Alaoglu's theorem (Lang, 1993, p. 71) to
conclude that there is a subsequence Nhk which converges weak* in the
space of measures to a measure N on [0, T] with scalar non-negative val-
ues. In the finite-dimensional case, this is sufficient to obtain convergence
of u hk and v hk • However, it is not generally so in infinite-dimensions.
Convergence of subsequences of (u h , v h ) in the following weak senses can
be shown. Since Ilvh(t)11 is uniformly bounded by the energy bound, it
follows that v h is uniformly bounded in V(O, T; £2) for 1 ~ p ~ 00,
the space of functions 'l/J : [0, T] ~ £2 where J[ I 'l/J (t) liP dt < +00.
Then by Alaoglu's theorem, there are weakly convergent subsequences
in V(O, T; £2) for 1 ~ p < 00, but only weakly* convergent subsequences
314
in LOO(0,T;£2) = L1 (0,T;£2)*. In the search for strong convergence,
there are two cases to consider: (a) the "easy" case where f3 E £2, and
(b) the "hard" but physically appropriate case where f3 ¢ £2.
4.1 The "easy" case

In this case, f3 E £2. In this case, we can obtain uniform bounds on
the amount of energy in the high-frequency modes. If f3 E £2, then for
sufficiently small ho > 0, limk-->oo sUP1,h l::j:j?k (( vt)2 + Aj (ut)2) = 0
where the supremum is taken over all l and ho ~ h > 0 where 0 ~ l h ~
T.
In this case, the "impact law" that this method represents is neither
perfectly inelastic nor perfectly elastic; it appears to have a "random"
coefficient of restitution ranging from zero to one. This affects many
numerical ODE schemes for impact problems (see, e.g., Stewart, 2000).
The reason for this is that (8-11) is a position-based scheme. In order
to obtain a particular coefficient of restitution, a velocity-based scheme
rather than a position-based scheme. See, for example, (Stewart, 1998)
and (Anitescu et al., 1999) which use velocity-based schemes rather than
position-based schemes.
4.2 The "hard" case

The "hard" case (f3 ¢ £2) arises naturally in the problem as originally
given. This can be most easily seen in the one-dimensional problem
since then f31 = 7r- 1/ 2 and f3j = (2/7r)1/2 for j ~ 2. Clearly in this case
f3 ¢ £2. Later, more precise descriptions of what space(s) f3 belongs to
will be given.
In this case, the differential equations Vj = f3j N(t) - AjUj + /j(t)
and Uj = Vj cannot allow impulses in N(t), because then v(t+) would
no longer be in £2 where t is the time of the impulse, and the system
would then have infinite energy. Nevertheless, some sort of singularity
is expected when there is an impact. A crucial quantity in controlling
what kind of singularity occurs is p(h) = l::~l f3; /(1 + h2 Aj/2). If
(f3j/(1 +Wj) I j = 1,2,3, ... ) is in £2, then p(h) is finite for all h > 0,
but we will have limhto p( h) = +00 if f3 ¢ £2. As we will see, p( h) rv
const h- 1/ 2 as h 1 o.
4.3 Spectral asymptotics

In order to obtain more precise bounds on the normal contact force,
we need to use asymptotics. The proof of these results uses the follow-
ing generalization of a Karamata-Tauberian theorem: If IL is a positive
315
measure on [0,00), a E (0,00), then 1;0 e-t >" dJ.L(>") rv at-a as t ! 0

10 10
implies 00 g(t>..) dJ.L(>") rv b 00 g(t>..) >..0.-1 d>" as t ! 0 provided 9 is a
continuous function where g(>..) = o(>..-a-E), >.. i 00, for some € > 0,
1000 g(z) za-l dz t= 0, and b = a/f(a + 1).
To apply this asymptotic theory we also need to obtain some asymp-

totic results for certain solutions of the "diffusion" version of the elas-
ticity equations. In any case, the following result can be easily verified
for the Laplacian operator and 'ljJ E £2(80):
Lf(h>'j)f3J rv const(f) h- 1 as h! o. (17)

j
This result can be extended to the full elasticity operator with more
work. These are now used to bound Nl;h.
First we note that l:~1 f3jU~+1 must be finite. Writel:~1 f3jU~+1 =
l:f=l f3jU~+1 + l:~J+1(f3j/Wj)(WjU~+l). Since wou l+1 E £2 and (f3j/Wj I
j > J) E £2 (0 is the Hadamard or componentwise product of vec-
tors in £2/' we see that the sum on the right is finite, as desired. If
l:~l f3ju/ 1 < 0 then N l = o. So for N l < 0 we have l:j f3jU~+1 = 0
but 'E j f3jU~ :::; OJ thus INII (h 2/2) 'Ej f3J/(1 +h2>'j/4) is bounded by the
absolute value of l:j( -h2>.jf3juV2 + hf3jv~ + (h 2/2)f3jfj )/(1 + h2>'j/4).
For the term 'E j -h2wjf3jwju;/(2(1 + h2>'j/4)) we bound the £2 norm
of the vector h 2diag(2 + h2>../2)-I(w 0 (3). Using (17) this can be es-
10
timated by the square root of 00 h 4 >.>.-1/2/(2 + h2>'/2)2d>' = 8(h).
Thus Ilh 2diag(2 + h 2>../2)-1(w 0 (3)11 = 8(hl/2). Similarly, Ildiag(1 +
h2>"/4)-1 (h{3) II = 8(hl/2), and II(h2/2)diag(1 + h2>"/4)-1{311 = 8(h 3 / 2).
Combining these asymptotic results, we obtain,
Nl;h = O(h-l/2).
This result is indicative of the fact that the contact force functions Nh(t)
are not only uniformly bounded as measures, but are bounded under
stronger norms.
5. The original Signorini contact condition

Similar results can be obtained for the original Signorini contact con-
ditions. From the dynamic contact continuous formulation, we establish
a numerical formulation for (2-5):
~ (vl+ 1 - vI) \l . u[ul+ 1] + fl in 0, (18)

u[u l +1] . n = N1n on 80, (19)
o ~ u l+1 . n - 9 ..L N l :::; 0 on 80. (20)
316
This can be used to derive a variational inequality for the discretized
time problem. Dissipativity results can be proven for this numerical
scheme: E( u l +1, vl+ 1) ::; E( u l , vI). In this section we estimate for N
in the H-1/2(arl) norm. Let tr be the standard trace operator of rl
onto arl defined by tr(w) = wlan for wE COO(rl) (Renardy and Rogers,
1993, pp. 218ff, §6.4.8). Then tr can be extended to a linear bounded
operator, also denoted tr : H1(rl) ~ H1/2(arl), where tr is onto. In
fact there are extension operators extk : H1/2(arl) ~ H1(rl) which are
the right inverses of the trace operator tr : H1(rl) ~ H1/2(arl). That
is, tr 0 extk = id Hl/2(an)' However, extk 0 tr =f:. idHl(n)' The extension
operators can be parameterized by k so that for large k the function
extk(W) has a support within distance 11k of arl. There is such a family
of parameterized extension operators where:
lextk(W)IH1(n) < C1 VI + kllwIIHl/2(an), (21)
lIextk(W)II£2(n) < C2 k - 1/ 2 I1 w IIHl/2(an) , (22)
and lul1-1(n) = In IVul 2 dV. Then it can be shown that as h ~ 0,
II NI II H-1/2(an) = O(h- 1/ 2 ). (23)
References
Anitescu, M., Potra, F. A., and Stewart, D. E. (1999). Time-stepping for three-
dimensional rigid body dynamics. Compo Methods Appl. Meeh. Engineering,
177:183-197.
Lang, S. (1993). Real and Functional Analysis, volume 142 of Graduate Texts in Math-
ematics. Springer, Berlin, Heidelberg, New York, 2nd edition.
Natanson, I. P. (1961). Theory of Functions of a Real Variable. Frederick Ungar Publ.
Co., New York. Revised edition. Original translated from the Russian by L.F.
Boron and annotated by E. Hewitt.
Oleinik, O. A., Shamaev, A. S., and Yosifian, G. A. (1992). Mathematical Problems
in Elasticity and Homogenization, volume 26 of Studies in Mathematics and ist
Applications. North Holland, Amsterdam, London, New York, Tokyo.
Renardy, M. and Rogers, R. C. (1993). An Introduction to Partial Differential Equa-
tions, volume 13 of Texts in Applied Mathematics. Springer Verlag, New York,
Berlin, Heidelberg.
Schatzman, M. (1973). Sur une classe de problemes hyperboliques non liw3aires. C.
R. Acad. Sci. Paris Ser. A-B, 277:A671-A674.
Stewart, D. E. (1998). Convergence of a time-stepping scheme for rigid body dy-
namics and resolution of Painleve's problems. Archive for Rational Mechanics and
Analysis, 145(3):215-260.
Stewart, D. E. (2000). Rigid-body dynamics with friction and impact. SIAM Review,
42(1):3-39.
CONVERGENCE FOR A TIME DISCRET-
IZATION OF DYNAMIC CONTACT PRO-
BLEMS WITH FRICTION
E. Pratt
Universite de Provence and
Laboratoire de Mecanique et d'Acoustique, C.N.R.S.,
31 chemin J. Aiguier, 13402 Marseille cedex 20.
J.-M. Ricaud
Laboratoire de Mecanique et d'Acoustique, C.N.R.S.,
31 chemin J. Aiguier, 13402 Marseille cedex 20.
Abstract A certain number of frictional dynamic contact problems can be formu-

lated as implicit variational inequalities. A time discretization of such
inequalities is given. Sequences of mappings interpolating the discrete
solution are then shown to converge towards the unique solution of the
variational inequality.
1. Introduction
The models adopted to describe the physical phenomena of frictional
contact induce numerous difficulties, such as strong non-linearities or the
implicit status of the constraint convex. Most of the results obtained are
issued from the study of various operators coupled with an unilateral
constraint (see Brezis (1971), Moreau (1983)). From a frictional contact
mechanics point of view, Duvaut and Lions (1972) elaborated variational
formulations through which they obtained some first results. Confronted
with the difficult task of formulating dynamical contact problems, Mar-
tins and Oden introduced the so called normal compliance model for
which they obtained an existence and uniqueness result in Martins and
Oden (1987). Recently, Jarusek and Eck (1999) obtained some results
concerning dynamical unilateral contact problems with local Coulomb
friction, using a velocity formulation for the contact.
We begin in section 2 by setting the generic implicit variational in-
317
318
equality which models the viscoelastic contact problems we consider. It

is generic in the sense that a certain number of different contact and
friction laws fit in to this formulation.
In section 3 the discretized scheme is given and the proof of the con-
vergence theorem is outlined. Finally a numerical example is to be found
in section 4.
2. Variational formulation
We consider a viscoelastic body that initially occupies a domain 0 in
IRd , d = 2,3. The boundary of 0 denoted by r is sufficiently regular.
This boundary is divided into three disjoint and open parts: r =
r u u r fur c. The body is subjected to volume forces of density F I ,
prescribed zero displacements and tractions F2 on the parts r u and r f
respectively. On r c, the body may come into contact with a foundation.
We shall denote by u = (Ui) the displacement field, by (J=((Jij) the stress
tensor, by € = (€ij) the strain tensor, for i,j E {1, .. ,d}. We denote by
p the mass density of the material, and respectively by A = (aijkl) and
B = (bijkl) the elasticity and viscosity tensors with the usual properties
of ellipticity and symmetry. The strong formulation of the problem is as
follows.
Find u = u(t,x) such that u(O,x) = ua,u(O,x) = ua and for all t E]O,T[,
div (J + FI = P ii. in 0, (J = A €(u) + B €(u),
U = ° on r u and (J.n = F2 on r f.
As frictional contact can occur on the part r c of the boundary, some
extra conditions depending on the models of contact and friction adopted
should be added to this formulation. In the following generic weak for-
mulation of the problem, these additional boundary conditions are taken
into account by the mapping ¢.
Find u E K such that for almost all t E]O, T[ and for all v E V:
{ (ii.(t) , v - u(t)) + a(u(t), v - u(t)) + b(u(t), v - u(t)) (1)

+¢(t, u(t), v) - ¢(t, u(t), u(t)) ~ ((f(t), v - u(t))).
where
H °
= [L2(O)]d, V = {v E [HI(O)]d, v = a.e. on r u} ,
K = {v E W I ,2(O, T; V) n W 2,2(O, T; H); v(O) = Ua E V, v(O) = ua E V} ,
V u,v E V, a(u,v) = k
aijkl Ui,j vk,l dx, b(u,v) = k
bijkl Ui,j Vk,l dx,
Vu, v E H, (u, v) =< pu, v > where <.,. > is the inner product in H,
FI E L2(O, T; H), F2 E L2(O, T; [L2(r f )]d),
"Iv E V, ((f'v)) =
inrFI v+ ir,
r F2 v.
319
According to the model one adopts for contact and friction, the map-
ping ¢ is given by:
Normal compliance and Coulomb friction law:
¢(t,u,v) = r CN (UN)+ vNds + irer CT (UN)+ IVTlds

ire
Tresca friction law with bilateral contact:
¢(t,u,v) = r g(t) IVTlds in V = {v E V,VN =

ire ° a.e. on rc}
Prescribed normal constraint:

¢(t,u,v) = r J.t (IN IVTlds with ((j,v)) = ((f,v)) + kr (IN VN
k
The different models considered above and the usual properties of the
elasticity and viscosity tensors allow us to assume (see Ricaud (1999)):
- a, b: V x V --t 1R are bilinear, symmetric, continuous and coercive.
- ¢ : [0, T] x V x V --t 1R is the sum of two mappings ¢I and ¢2 so
that for all (t,u,v) E [O,T] x V x V, ¢(t,u,v) = ¢1(U,V) + ¢2(t,u,V)
with ¢I, ¢2(t,.,.) weakly sequentially continuous for all t E [0, T] and
satisfying the following conditions:
for all sequences (ukheIN such that Uk 'U in W I ,2(0, Tj V),

k--++oo
ior
liminf T ¢(t,Uk(t), Uk (t))dt 2:: r T ¢(t,u(t),u(t))dt,
k--+oo io
for all u E V, ¢l (u, .) is linear,
¢2(0, uo,.) = 0, ¢2(t, u,.) is a semi norm,
3 'fJ > 0, 'v'Vb V2, Wb W2 E V, such that 'v'tb t2 E [0, T],
I¢(tb VI, wt} - ¢(tb vb W2) + ¢(t2,v2,W2) - ¢(t2,v2,WI)1
~ 'fJ {IIVI - v211 + It I - t21} IIWI - w211·
- f is a Lipschitz mapping from [0, T] into V.
- Uo and Uo satisfy the following compatibility condition:
there exists 1 E H such that

(2)
a(uo,v) + b(uo, v) + ¢(O,uo,v) = ((f(O),v)) - (l,v) 'Iv E V.
Remark: Assumption 2 consists in supposing that the initial conditons

are sufficiently regular to ensure that no initial shock occurs, however
an initial acceleration is not forbidden.
320
3. Convergence of the discrete problem

For all n E 1N*, we set At = Tin and for all i E {O,···, n}, t i =
iAt, fi = f(t i ). We set ui as the approximation of u at t i and lJi and 'Yi
as the approximations of the velocity and the acceleration:
We consider the following discrete problem:
For all i E {1, .. ·, n - 1}, find ui+l E V such that

hi, v - cf) + a(ui+1, v -lJi) + b(lJi, v -lJi) + ¢(ti, u i+1, v) (3)
{ -¢(t i , ui+1 ,lJi) ~ «(ji, v - lJi)) 'Vv E V,
uo = uo, 8° = uo, u l = uo + At 8° = Uo + At uo.
Using classical results on implicit elliptic inequalities, problem 3 is

shown to have a solution which is unique if At is sufficiently small.
3.1. Stability results

The following stability results for the solution of the discrete problem
3 are established (see Ricaud and Pratt (2001)).
Lemma 1
There exists no E 1N* and c E R+ such that :
'Vn ~ no, l'YllH ~ c.
Lemma 2
There exists no E 1N* and C E R+ such that:
'Vn ~ no, Ildlllv ~ c.
Lemma 3
There exists no E 1N* and Cll C2, C3 E R+ such that 'Vn ~ no:
'Vi E {1,··· ,n -1}, Iluiliv ~ Cll IlfJillv ~ C2, l'YilH ~ C3, Ilunllv ~ Cl.
The proof of these lemmas rests uniquely on the assumptions of sec-
tion 2 and the constants exhibited depend only on the initial data. Let
us point out that the presence of viscosity is essential in these proofs
nevertheless these lemmas hold however little viscosity is present and
also however stiff the contact and friction conditions are.
321
3.2. Convergence results

Let us define the following sequences of mappings:
For all i E {I,··. ,n - I} and t E]ti, ti+I],

fn(t) = J(ti ), <Pn(t,.,.) = <p(ti ,.,.), Un(t) = ui+1,
. 2
un(t) = ui + (t -t i ) cf, un(t) = ui + (t - ti)J- + ¥ , i
and for t E [to, tl],
In(t) = J(O), <Pn(t,.,.) = <p(0,., .),
un(t) = uO = Uo, un(t) = uO + (t - to)6° = Uo + (t - to)uo,
un(t) = uO + (t - t o)60 + (t -2to )2 1 = Uo + (t - to)uo + (t -2tO )2 1.
The constants ui , ui and Ji are determined to obtain a continuous

mapping for Un and a continuously differentiable mapping for Un. By
choosing un(t) = uO = Uo on the first time interval [to, tl] the compati-
bility condition (2) ensures that the following inequality (4) is satisfied
for t E [to, tl] and as (U i )iE{2,,,.,n} are solutions of (3), (4) is also satisfied
for t EW, t n], so that we have: 'Vt E]O, T[, Ir:/v E L2(0, Tj V)
(~un(t)'V(t) - ftun(t)) +a(un(t),v(t) - ftun(t))

+b(ftun(t),V(t) - ftun(t)) + <Pn(t,un(t),v(t)) (4)
-<Pn(t, un(t), ftUn(t)) ~ ((fn(t), v(t) - ftun(t))) .
The stability results enable us to prove the following lemma:
Lemma 4
There exists a subsequence (nk)kElN such that:
Ir:/t E [0, T], UnA: (t) 'u(t) in V,

k-++oo
Ir:/t E [0, TJ, 4t unA: (t) , dd u(t) in H,
aT; k-++oo t
unA: ----->O'u in W I,2(0, Tj V),
k-++oo
unA: 'U in W 2,2(0, Tj H).
k-++oo
°
The convergence result is now obtained in the following way. We be-
gin by integrating (4) between and T and going to the limit in the
resulting inequality. The limit u obtained in lemma 4 is thus shown to
satisfy the following inequality:
322
Vv E L2(O, Tj V) :
loT (U(t), V(t) - u(t))dt + loT a(u(t), V(t) - u(t))dt

a TaT
+ 10 b(u(t), v(t) - u(t))dt + 10 ¢(t, u(t), v(t))dt
aT T a
- fa ¢(t,u(t),u(t))dt ~ fa ((j(t),v(t) - u(t)))dt.
By a judicious choice of v E L2(O, Tj V), u is shown to be a solution

problem (1). Finally Gronwall's lemma ensures the uniqueness of the
solution u and therefore the convergence of the entire sequences towards
u. We thus obtain the following result:
Theorem 1
The sequences Un, Un and Un converge weakly (in the sense of lemma 4)
towards the unique solution of problem (1).
4. Numerical results
In this section we give the numerical results we have obtained by com-
puting a Q2 finite element approximation of the incremental solution
given by (3). Contact is modelized by a normal compliance law, friction
by a Coulomb law. The discretized problem thus obtained becomes a
finite dimensional implicit inequality that can be set as a minimization
problem by using a fixed point procedure.
The mechanical benchmark consists of a two dimensional rectangu-
lar solid, 20mm by 160mm which may come into contact with a plane
obstacle on one of its long sides and has some prescribed displacement
on its two sides. The prescribed force is sinusoidal in time. Such a me-
chanical test was studied in Martins and Oden (1985) but they used a
different normal compliance law and a regularization of the friction law,
instead of the Coulomb law.
The geometry of the problem with the data concerning the loading,
the prescribed displacement and the contact zone are described in Fig-
ure 1, because of the symmetry we consider only the left hand side of
the solid.
323
.n'" .n
~
Scm
Fig. 1- The mechanical benchmark.
The Young modulus equals 210000 mPa, the Poisson ratio equals
0.25 and the mass density equals 7900kg/m 3 •
A Rayleigh damping method is used to take the viscosity into account.
The evolution of the normal and tangential displacements of a node at
the centre of the contact zone are plotted in Figure 2. The evolution of
the ratio UT / UN and the evolution of the tangential velocity of the first
node of the contact zone are plotted in Figure 3.
Fig. 2- Normal, tangential displacements.
0.3
l l
0.2
0.1
-0.1 l, r
0.0
0.0 , 0.00 , 0.00 , 0.0 01
-0.2
-0.3
Fig. 3- Evolution of the ratio UT/UN, tangential velocity.

324
References
BR Brezis, H., Problemes unilateraux. Master Thesis, Universite de Paris VI, fevrier
1971.
Cocu, M., Existence of solutions of Signorini problems with friction. Int. J. Engng
Sci., Vol. 22, p. 567-575, 1984.
Duvaut, G., and Lions, J.-L., Les inequations en mecanique et en physique. Dunod,
Paris, 1972
JaruSek, J., and Eck, C., Dynamic contact problems with small Coulomb friction for
viscoelastic bodies. Existence of solutions. Math. Models and Meth. Appl. Sci.,
Vol. 9, p.11-34, 1999.
Martins, J.A.C., and Oden, J. T., Existence and uniqueness results for dynamic contact
problems with non-linear normal friction and interface laws. Nonlinear Analysis,
Theory Meth. Applic., Vol. 11, p. 407-428, 1987.
Moreau, J.-J., Liaisons unilaterales sans frottement et chocs inelastiques. C. R. Acad.
Sci. Paris, Vol. 286, p. 1473-1476, 1983.
Oden, J.T., and Martins, J.A.C., Models and computational methods for dynamic
friction phenomena. Comput. Meth. Appl. and Engng., Vol. 52, p. 527-634, 1985.
Ricaud, J.-M., and Pratt, E., Analysis of a time discretization for an implicit varia-
tional inequality modelling dynamic contact problems with friction. Mathematical
Methods in the Applied Sciences, Vol. 24, p. 491-511, 2001.
Ricaud, J.-M., Etude d'une classe d'inequations d'evolution implicites et application
a des problemes dynamiques de contact avec frottement. Ph. D. thesis, Universite
de Provence, novembre 1999.
SOLUTION METHODS
FOR STRUCTURAL OPTIMIZATION
IN CONTACT ROD PROBLEMS
Isabel N. Figueiredo
Departamento de Matematica, Universidade de Coimbra
Apartado 3008, 3001-454 Coimbra, Portugal
Joaquim J. Judice
Departamento de Matematica, Universidade de Coimbra
Apartado 3008, 3001-454 Coimbra, Portugal
Silverio S. Rosa
Departamento de Matematica, Universidade da Beira Interior
6201-001 Covilha, Portugal
Abstract We consider a material and shape optimization problem which involves

a composite rod in contact with a rigid foundation. It is shown that the
problem can be reduced to an optimization problem in some indepen-
dent variables and some variables defined implicitly by the solution of a
linear complementarity problem (LCP) with a positive definite matrix.
A projected-gradient algorithm is proposed that incorporates an efficient
LCP solver to compute function values and gradients. An example is
included to illustrate the suitability of the proposed methodology.
1. Introduction
We address a structural optimization problem which involves a unidi-
rectional fiber reinforced composite rod in contact with a rigid founda-
tion. The purpose of the model is to find the fiber volume fraction and
the value of some geometric parameters of the rod in order to minimize
the compliance of the structure. The variables are continuous and vary
in an admissible set.
325
l.A.C. Martins and M.D.P. Monteiro Marques (eels.), Contact Mechanics, 325-332.
326
The discretization of this optimization problem by using appropriate

finite elements leads to a mathematical programming problem with equi-
librium constraints (MPEC), in which the objective function depends
on the so-called outer independent variables (material and geometric
parameters) and on inner dependent variables, that are the solution of
a linear complementarity problem (LCP), representing the contact rod
problem. Due to the definition of the compliance function, it is possible
to reduce the problem to an optimization problem in the independent
variables. Furthermore the constraints on these variables are simple
lower and upper bounds, and the objective function is continuously dif-
ferentiable in its constraint set. A projected-gradient algorithm is pro-
posed for the solution of such an optimization problem. In addition it
is shown that the values of the objective function and of the gradient
in each point used by the algorithm can be obtained by solving special
LCPs with positive definite matrices.
The outline of the upcoming sections is as follows. In section 2 the
optimization rod problem is introduced. The algorithm and its imple-
mentation are discussed in section 3. An example of the application of
the algorithm is included in section 4.
2. Notations and Description of the Problem

Let w be an open, bounded and connected subset of R2, and L > 0
be a constant. We denote by w x [0, L] the set occupied by the rod,
in its reference configuration, with length L and cross section w. We
assume that the material of the rod is an unidirectional fiber reinforced
composite material. We denote by x = (Xl,X2, ••• ,Xs ) E R S the vector
whose s components specify the type and the number of material and
geometric features of the rod under consideration. Moreover we sup-
pose that the rod is clamped at its extremities and is subjected to the
action of applied forces that force a part of the lateral surface of the
rod to be in contact with a rigid foundation. In addition we assume
that the candidate lateral contact surface is plane and perpendicular to
one of the inertia axes of the rod. For a fixed vector x, the continuous
one-dimensional equilibrium model describing this contact rod problem
is a generalization of the Bernoulli-Navier model and it can be mathe-
matically justified by the asymptotic expansion method, as in Trabucho
and Viano (1996) chap.6, for the homogeneous and isotropic case. Using
the finite element method, the discrete formulation, for each x, of the
one-dimensional contact rod model under consideration constitutes the
327
following discrete variational inequality
Find uEU={VERn : Vj=O, VJ~9J}' such that

{ (1)
(v - uf(B(x)u - F(x)) ~ 0, \tv E U.
In (1) n denotes the number of global degrees of freedom of the finite
element mesh of the rod axis [0, L]. The matrix B(x) is the stiffness
matrix and F(x) is the vector associated to the applied forces. B(x)
depends explicitly on x and F may also depend on the components of
x. The vector u is the solution of the contact rod model and represents
the approximate displacement of the rod. We remark that u depends
implicitly on x. The notation (v - u) T stands for the transpose of vector
(v - u). The set U is the set of admissible displacements and the sets
j and J are subsets of the global degrees of freedom {1, 2, ... , n}. The
vectors Vj and VJ are subvectors of v, with components (Vj)iEi and
(VJ)jEJ, respectively. The condition Vj = 0 corresponds to the clamped
rod condition. The vector 9J = (9j)jEJ is independent of x and defines
the gap between the rod and the rigid foundation at the nodes j E J.
The condition VJ ~ 9J means that Vj ~ 9j, for j E J, and states that
the rod can touch but not penetrate the rigid foundation at node j E J.
In order to clarify the dependence of B and F on x, we describe
next the structure of the element stiffness matrix and of the element
vector force. We denote by hi the amplitude of the generic finite ele-
ment [Yi, YHl] subset of [0, L]. Then the corresponding element stiffness
matrix Bi is
~ 0 0 -~ 0 0
hi hi
121 61 121 61
0 0 --.:z-
hf h'f hi h'f
61 41 61 21
0 0 --.:-2"
Bi(X) =E h'f hi hi hi
(2)
-~hi 0 0 ~
hi 0 0
121 61 121 61
0 --.:z- --.:-2" 0 --.:-2"
hi hi hf hi
61 21 61 41
0 0 --.:-2"
h'f hi hi hi
where E, Iwl and I depend on x and represent respectively the longi-

tudinal modulus of the material, the area of the cross section and the
moment of inertia. Assuming now that q and p are the uniformly dis-
tributed forces per unit of length in the direction of the rod axis and
in the direction perpendicular to the rod axis, respectively, then the
element vector force F i (x) is defined by
ph~
12
- Ph~l
12 . (3)
328
We remark that (1) is an obstacle problem. In particular it can be re-

formulated as a mixed complementarity problem. To see this, we denote
by K and H the subsets of indices defined by K = {1, 2, ... n} \ {j U J}
and H = K U J, respectively. By performing the change of variables
VEU<==}v-gEU={vER n : Vj=O, vJ~O} (4)
where the vector 9 ERn is defined by
9 = {gj)jElV' and gj = 0, if j ~ J, gj = gj, if j E J, (5)

then problem (1) is equivalent to the following mixed complementarity
problem
Find u E RIHI, wE RIHI such that
{ A(x)u - F{x) = W
(6)
UJ ~ 0, WJ ~ 0, WK = 0,
uJWJ =0.
The number IHI is the cardinal of H, A is a submatrix of Band F is a
subvector of F - Bg, whose elements have indices in H, that is,
and F{x) = (F{x) - B(x)g)H. (7)
We remark that u is a solution of (6) if and only if u + 9 is a solution of
problem (1).
The structural optimization problem considered in this paper consists
of finding an equilibrium point of the contact rod problem that mini-
mizes the compliance function and therefore maximizes the stiffness of
the structure. Due to the equivalence shown above, this problem can
be written as the following mathematical programming problem with
equilibrium constraints (MPEC):
minO(x,u) = minluTA(x)u
subject to:
u E RIHI, W E RIHI
MPEC (8)
xEX and { A(x)u - F(x) = W
uJ~O, wJ~O, WK=O
U}WJ = O.
The set X is the set of admissible material and geometric parameters

defined by X = {x = (Xl, ••. ,X s ) E RS: xiin ~ Xi ~ xiax, i = 1, ... ,s},
where xiin and xiax are real constants. The objective function O(x, u)
satisfies
329
because of the complementarity condition uT w = O. Therefore, for each

x, 8{x, u) is the compliance of the rod constrained by the zero obstacle
and subjected to the action of applied loads represented by the vector
(F{x) - B{X)g)H. If 9 = 0 then 8{x,u) is exactly the compliance of
the rod with applied forces F(x) and constant zero obstacle. We refer
to Petersson (1995) for a justification of other definitions of stiffness
measure in structural optimization.
3. A Projected-Gradient Algorithm for the

MPEC
Consider the inner complementarity problem of the MPEC. Since for
each x E X the matrix A(x) is symmetric positive definite, then this
complementarity problem has a unique solution. Hence it is possible to
write the MPEC as the following optimization problem in the variable
x
[ minl(x) = min8(x,u(x)) (10)
subject to x E X
where u depends implicitly and uniquely on x through the complemen-
tarity problem. In general, the non-smoothness of u(x) with respect
to the variable x may originate the non-smoothness of the objective
function I. In fact, u is a Lipschitz function on the feasible set X,
the directional derivative u'(x, x) of u at x in the direction x exists,
but the gradient 'V xU(x) of u at x does not exist if the coincidence
set {j E J: Wj(x) = 0, Uj(x) = O} is not empty (see, Harker and
Pang (1990) or Haslinger and Neittaanmaki (1997) for a justification
of these statments). Thus non-smooth optimization algorithms such as
subgradient and bundle methods (see Outrata et al. (1998)) should be
recommended to solve (10), in general. Nevertheless for the particular
objective function I defined in problem (10) the gradient of I, 'Vxl,
exists and is defined by
1
'V xl(x) = 'V xF(x)T u(x) - "2u(xf'V xA(x)u(x). (11)
For each x, the vector u(x) is the solution of the complementarity prob-
lem and 'V xF and 'V xA are the gradients of F and A defined by
with Fi and Aij the elements of F and A respectively. Assuming that

F and A are of class C 1 with respect to x, the gradient 'V xl is of class
Co. We observe that formula (11) can be obtained by calculating the
directional derivative I'(x, x) of I at the point x in the direction x. In
330
fact, it follows from the definition of f and the complementarity problem

(6) that
f'(x, x) = u(x)T A(x)u'(x, x) + ~u(x)T A'(x, x)u(x) (13)
where A'(x,x) is the directional derivative of f at x in the direction x

and u'(x,x) satisfies
A(x)u'(x, x) = w'(x, x) + F'(x, x) - A'(x, x)u(x)
(14)
u(x)T w'(x, x) = 0
with w'(x, x) and F' (x, x) the directional derivatives of wand F at x in

the direction x, respectively. Introducing (14) in (13), the term u'(x, x)
disappears. Since A and F are of class c1, then A'(x, x) = V xA· x and
F'(x,x) = VxF· x, where the dot means the usual Euclidean product
in R S • It is now easy to obtain the expression (11).
Therefore for the specific problem (10) it is possible to apply a classical
projected gradient method. The steps of this algorithm are described
next, where Px represents the projection on the set X.
Projected-Gradient Algorithm
• Let xO E X and I: > 0 be a given tolerance.
• For k = 0, 1,2, ...
- Compute Vxf(x k), yk = Px(xk-Vxf(x k)) andpk = yk_xk.
- If IIpk II < 1:, stop with (xk, u(xk)) a solution of the MPEC.
- Compute the stepsize Ok E]O, 1] using the Armijo Criterion
f(x k + Okpk) ~ f(x k) + COk V xf(xkf pk (15)
with 0 < C < l.
• Update xk+l = xk + Okpk.
As discussed in Nocedal et al. (1999) this algorithm possesses global
convergence to a stationary point of the function f (x) on the convex set
X. In order to compute objective function and gradient values and to
employ the Armijo criterion, a complementarity algorithm is required
to evaluate u(xk) and u(x k + Okpk), that are the solutions of the com-
plementarity problem (6) for x = xk and x = xk + Okpk, respectively.
Since the matrix A(x) is symmetric positive definite for each x E X,
the block pivoting or interior-point algorithms should be recommended
to process these LCPs (see Fernandes et al. (2002)). Furthermore, for
this problem the projection is quite simple to obtain because the set X
consists of simple bounds on the variables Xi.
331
4. A Numerical Example
We have tested the previous algorithms in an example. The material is
an unidirectional fiber reinforced composite material, whose longitudinal
modulus is E = EIVI + Em(l- VI) with EI the modulus of the fiber,
Em the modulus of the matrix and VI the fiber volume fraction which
belongs to [0,1]. The optimization variable is VI and the data of the
problem are displayed in the table below.
Table 1. Data of the Example
Parameter Value
Em (OPa) - modulus of the matrix 3.45
E, (OPa) - modulus of the fiber 86
Vr Vrin , ax -lower and upper bounds for V, 0.01,0.99
L (m) - length of the rod 10
q (N) - distributed force in the direction of the rod axis -18000
p (N) - distributed force in the direction perpendicular to the rod axis -200
9 (m) - obstacle (constant) -0.001
Iwl (m 2 ) - area of the cross section 0.004
J(m4) - moment of inertia 2.1e- 6
The symbols (GPa), (N) and (m) denote the units Giga Pascal, New-
ton and meter, respectively. The interval [0,10] has been discretized
with 50 finite elements, whose length hi is constant and equal to 0.2.
The projected-gradient algorithm has successfully found an optimal
solution VI = 0.3518 in 7 iterations. We have employed the block pivot-
ing algorithm described in Fernandes et al. (2002) to process the LCP's
required by the projected-gradient method. The block pivoting algo-
rithm has required a total of 150 iterations to process all the 44 LCP's
needed by the projected-gradient algorithm. In these tests the subrou-
tine beam2e of the CALFEM toolbox of MATLAB has been used to
evaluate the stiffness matrix A and the force vector F. The comple-
mentarity algorithm has been implemented in MATLAB. The figure 1
shows the plot of the equilibrium bending displacement of the rod, at
the optimum value VI = 0.3518.
As stated in this section, the projected gradient algorithm has per-
formed well for solving this example of MPEC problem discussed in this
paper. Further computational investigation is required to evaluate the
performance of the algorithm in practice. A description of this experi-
ence will be reported in the near future.
332
X 10-3
2 4 5
Figure 1. Bending displacement of the rod for V, = 0.3518
References
Fernandes, L.M. ; Figueiredo, I.N. and Judice, J.J. (2002) On the solution of a finite
element approximation of a linear obstacle plate problem, International Journal
of Applied Mathematics and Computer Science (to appear).
Harker, P.T. and Pang, J.S. (1990) Finite-dimensional variational inequality and non-
linear complementarity problems: a survey of theory, algorithms and applications,
Mathematical Programming 48,2, 161-220.
Haslinger, J. and Neittaanmi:i.ki, P. (1997) Finite Element Approximation for optimal
Shape Design, Theory and Applications, Wiley, Chichester.
Nocedal, J. and Wright, S. (1999) Numerical Optimization, Springer, New York.
Outrata, J. ; Koevara, M. and Zowe, J. (1998) Nonsmooth Approach to Optimization
Problems with Equilibrium Constraints, Kluwer Academic Publishers, Dordrecht.
Petersson, J. (1995) Optimization of Structures in Unilateral Contact, Linkoping Stud-
ies in Science and Technology, Dissertations n.397, Linkoping University, Sweden.
Trabucho, L. and Viano, J.M. (1996) Mathematical Modeling of Rods, in P.G. Ciarlet
and J.L Lions (edts), Handbook of Numerical Analysis, VoI.IV, 487-974, North-
Holland, Amsterdam.
UNILATERAL CONTACT, FRICTION AND
ADHESION: 3D CRACKS IN COMPOSITE
MATERIALS
Michel Raous
Laboratoire de Mecanique et d'Acoustique - CNRS
Marseille - France
Yann Monerie
Laboratoire de Mecanique et d'Acoustique - CNRS
Marseille - France
Abstract The model coupling adhesion, unilateral contact and friction, devel-
oped by M. Raous, L. Cangemi and M. Cocou, is extended here for
studying propagating cracks in fiber-reinforced composites. The brittle
fracture of the bulk materials (fiber and matrix) is described by a time-
independent version of the model. In this case, jumps in the evolution of
the solution may occur and a dynamic formulation has to be used. This
is performed by adapting a dynamic formulation due to J.J. Moreau
and M. Jean to this problem. Numerical simulations of 3D fracture of
composite materials are then presented.
1. Introduction
The present work deals with interface problems where an initial ad-
hesion due to chemical or thermal treatments can disappear during the
loading and be replaced by frictional sliding. A model taking into ac-
count adhesion, friction and unilateral contact was proposed by
M. Raous, L. Cangemi and M. Cocou (the RCC model) with a quasi-
static formulation in Raous et al. (1997, 1999). It was used to simulate
the micro-indentation of a fiber in a composite material. A comparison
between this model and those developed by Tvergaard (1990), Allix et
al. (1995, 1998), Ladeveze (1995), Michel and Suquet (1994), Michel et
al. (1994) and Chaboche et al. (1997, 2001) was presented in Monerie
et al. (1998). In the RCC model, using the adhesion intensity vari-
333
I.A.C. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 333-346.
334
able introduced by F'remond (1987, 1988) and Trong Dinh Tien (1990),
adhesion, friction and unilateral contact are described via an unilateral
elastic behaviour of the interface which fulfills the non penetration con-
ditions and depends on the intensity of adhesion. Viscosity was taken
into account in the evolution of the adhesion intensity.
In the present work, the model is extended and a new formulation is
proposed in order to describe the propagation of a crack in a composite
material and to study the interactions occurring between the crack and
the fiber-matrix interfaces. To study the crack propagation, a model
without interface viscosity has to be used (and friction will be neglected).
In this case, rapid changes in the evolution of the solution may occur
and a dynamic formulation has to be used. We present here :
- a new form of the RCC model introducing a new term which gives a
better transition from adhesion to friction,
- a dynamic formulation, which is an extension to the RCC model of the
Non Smooth Contact Dynamics method (NSCD method) developed by
Jean (1999) and Moreau (1988, 1999),
- the simulation of 3D fracture in a composite material.
2. The model coupling adhesion, friction and

unilateral contact
The model given by relations (1) to (3) differs from the standard RCC
model (see Raous et al. (1999)) because of the term (1 - (j) in the fric-
tion law. This means that friction is not active when adhesion is still
complete and that the friction threshold progressively increases as the
adhesion decreases. In this model :
- adhesion is characterized by the variable {j (see Fremond (1987, 1988)
and Trong Dinh Tien (1990)), where 1 ~ {j ~ 0, and {j = 1 corresponds
to total adhesion, and {j = 0 to no adhesion;
- unilateral stiffnesses {j2CN (normal) and {j2CT (tangential) ofthe inter-
face, depending on the adhesion intensity, ensure a continuous behaviour
during the competition between friction and adhesion;
- a viscosity b is included in the evolution of the adhesion intensity {j
(this viscosity is neglected for modeling crack propagation and the for-
mulation and the numerical treatment will be adapted to this case in
the following);
- in (3), the Laplacian term kA{j introduces a non local adhesion, but it
will not be taken into account here.
335
2.1 The model
The behaviour of the interface is given by the following relations,
where (1) characterizes the unilateral contact with adhesion, (2) the
Coulomb friction with adhesion, and (3) the evolution of the adhesion
intensity (3.
-~ + (32CNUN ~ 0, UN ~ 0, (-R~ + (32CNUN) .UN = 0, (1)
(2)
with
if IIRT - R~II < 1-'(1- (3) IRN - (32CNUNI =} uT = 0,
if II~ - R~II = 1-'(1- (3) I~ - (32CNUN =} 3'\ ~ 0, u
T = ,\(~ - ~),
(3)
where RN and UN are the algebraic values of the normal components of

the contact force and of the relative displacement between the two bodies
(occupying the domains 0 1 and 0 2 ) defined on the contact boundary r c ,
and ~ and U T are the tangential components of this contact force and of
this relative displacement. The constitutive parameters of this interface
law are as follows :
- CN and CT are the initial stiffnesses of the interface (when adhesion is
complete),
- I-' is the friction coefficient,
- w is the decohesion energy,
- p is a power coefficient (p = 1 in what follows),
- k = 0 in what follows.
Figure 1 gives the normal behaviour of the interface during loading
and unloading (in Figure 1, u~ = ..jw/CN and R~ = v'wCN ) . It
should be noted that strict unilateral conditions are enforced (Signorini
conditions), and that neither penalization nor compliance are used for
the unilateral contact (no extra parameters are needed).
Figure 2 gives the tangential behaviour (in Figure 2, u~ = ..jw/CT
and ~ = v'WCT ). When the stresses and displacements are very small,
the behaviour of the interface is elastic. When the local elastic energy
reaches the threshold w (decohesion energy), the intensity of adhesion
336
begins to decrease, the apparent stifi'nesses (32CN and (32CT also begin
to decrease (softening) and the friction begins to act. When (3 tends to-
wards zero, classical Coulomb friction behaviour is obtained (the limit is
asymptotic). This behaviour can be compared with that of the standard
RCC model given in Figure 3, where the dotted line gives the evolution
with infinitely small velocities. Figure 3 gives a clearer picture than
Figure 2 of the energy exchanges. The interface energy is dissipated by
viscosity, friction or adhesion (when the intensity of adhesion begins to
decrease). A reversible elastic part is stored at the interface.
i
I
l -------------+~
i
1
·2
-5
----
. 1+ - - - - - --11
~ cydo •• p.-!b«I
-4
~ty
-f-.--r-'..:...
-3
,.-.-....""
...-.-
-2 -I
...
~r__+_._+_~~r___.,....._+
0
Utr/UON
2 3 -4
Figure 1. Normal behaviour of the interface.
At the point P :
• Energy dissipated
by viscosity
1+ ~ I RN I Energy consu med
Ro by the damage
• Energy stored In the
interfa.ce
$ Stored complementary
energy (viscosity effect)
~ Energy diBBipated
by friction
1 2 3 4
Figure 2. Thngential behaviour of the interface (present model) .

337
At the point P ,
• Energy dissipated
1 + J'IRNI by viscosity
Ro
Energy consumed
by the damage
• Energy stored in the
interfacial bond
ID Comple menta ry stored

energy
J'IRNI
Ro
1 2 3 4
Figure 3. Tangential behaviour of the interface (RCC model) .
2.2 Thermodynamic basis

The interface is considered as a material surface and the following
thermodynamic variables are introduced : the relative displacements
(UN n, u T ) and the adhesion intensity (3 are chosen as the state variables,
and the contact force R and a decohesion force G, as the associated
thermodynamic forces. The thermodynamic analysis given for the RCC
model in Raous et al. (1999) can be extended to the present model in
order to obtain the model given by relations (1) to (3). The following
choices (4) and (5) are made for the free energy ~(UN' Un (3) and the
potential of dissipation q, (UT'~) (in the case when k = 0). Because of
the lack of differentiability or convexity of some terms, the state laws and
the complementary laws have to be expressed as differential inclusions
(see Moreau (1988), Raous (1999)). In (4), the indicator function Ii(
(where K = {v / v ~ O} ) imposes the unilateral condition UN ~ 0
and the indicator function Ip (where P = b / 0 ~ 'Y ~ I}) imposes
the condition (3 E [0,1]. In (5), the indicator function Ic- (~) (where
C- = b / 'Y ~ O}) imposes that ~ ~ 0 : the adhesion can only decrease
and cannot be regenerated (it is not reversible) in the present model.
338
3. The quasi-static problem

The quasi-static formulation associated with the model including in-
terface viscosity can be set as two variational inequalities (one of which is
implicit) and a differential equation (extension of the formulation given
in Cocou et al. (1996) and Raous et al. (1999)).
Problem (Pt) : Find (u,,8) E W 1,2(0, T; V) x W 1,2(0, T; H) such that

U(O) = Uo E K, ,8(0) =,80 E Hn[O, 1[ and for Vt E [O,T), u(t) E K, and
a(u, v - 11) + j(,8, UN, vT) - j(,8, UN, UT) +
! rc ,82 CTUT . (VT - uT)ds ~ (F, v - 11) - (RN' V N- UN) Vv E V,
-(RN' Z - UN) +! ,82 CNUN.(Z - uN)ds ~0 Vz E K, (6)

rc
~ = -l/b [w - (C N u~ + CT lIu T II2),8] - a.e. on rc, (7)
where:
-u = (uI, u2) where u1 and u2 define the displacements in 0 1 and 0 2,
- V = (VI, V 2), va = {va E [H 1(oa)]3;va = 0 a.e. on a = 1,2 ru},
(ru are the parts of the boundaries where the displacements are pre-
scribed),
- H = LOO(re ),
- K = {v = (vI,v 2) E VI X V2;VN ~ Oa.e. onre}, where re is the
contact boundary between the two solids 0 1 and O2,
- a(.,.) is the bilinear form classically associated to the elasticity map-
ping,
- j(,8,UN,VT) =! J.t(1- ,8)lu(RN) - ,82CN uN lllvT llds , where U(RN) is
rc
a linear and compact mapping, u(.) : H-l/2(r e) ~ L2(re) (non local
friction),
- F = (PI, p 2 ) are the given force densities applied respectively to solid
1 and to solid 2.
In Raous et al. (1999), an incremental formulation was given and the
problem was solved by associating minimization techniques and a fixed
point method on the sliding threshold for the contact, and a O-method
for the differential equation. Details on the computational methods are
presented extensively in Raous (1999). This quasi-static formulation was
used to simulate the micro-indentation of a fiber in a SiC/SiC composite
and identify the constitutive parameters of the fiber-matrix interface.
This formulation has been extended to hyper-elastic bodies in Bretelle
et al. (2001).
339
4. The model with no viscosity : dynamic

formulation
In some cases, there is no mechanical reason to introduce a viscous
dissipation in the interface behaviour. This is the case when this model
is used to describe the propagation of a crack in a brittle bulk material,
as we will now do. In order to study the influence of the characteristics
of the fiber-matrix interface on the propagation of a crack in a composite
material, we will suppose that the plane of propagation of the crack is
given (which is of course a restrictive assumption) and we will use the
adhesive model (without friction and without viscosity) on that plane
considered as an interface (matrix-matrix and fiber-fiber "interfaces"
because the crack propagates in both materials).
4.1 The adhesion model with no viscosity

The quasi-static problem for an adhesive contact without friction, con-
sidering a rate-independent cohesive model (no viscosity), can be written
as presented in Problem (P2).
Problem (P2 ) : Find uE K such that:
a{u, v - u) - (F, v - u) + { R ([U] ,,8) . ([v] - [U]) ds ~ 0 'Iv E K,(8)

ire
,8 = h([U]) a.e. onrc , (9)
where:
- [u] is the relative displacement on r c, [U] = uNn + UT
- the contact force R ([U] ,,8) is given by :
R ([u],,8) = ,82 (CNuNn + CTUT) (10)

- h([U]) is a given func.tion deduced from (3) and satisfying the condi-
tions 1 ~ ,8 ~ 0 and ,8 :::; 0 ; with an incremental formulation associ-
ated to Problem (P2), ,8k+l will be defined, at each time step t k+1, by
,8k+l = Min [l,,8k, w/(CN (u N k+1)2 + CT (lIuT ll k+1)2)].
In Cocou et al. (1997) and Raous et al. (1999), we gave a condition for
the existence of a solution of the incremental form associated to Problem
{Pt} when a non local friction is considered. In Cocou and Rocca (2000),
a similar condition was obtained on the quasi-static Problem (Pt) itself
(considering local friction). Here, for the frictionless problem with no
viscosity for the adhesion evolution, existence is obtained and a condi-
tion ensuring uniqueness is established (Monerie (2000)).
340
Theorem
There exists a fixed point, k, for the application s :
s : k -+ 1/ ([u(k)]) = R([u(k)] , h([u(k)]))
where u is the solution of :
a(u, v - u) - (1', v - u) + r k. ([v] -
ire
[U]) ds ~0 'Iv E K. (11)
and
where:
- CO is the Lipschitz constant for the application
R(.,.) : L2(rc)3 -+ L2(rc)3,
- I is the constant of coercivity of a(., .),
- >. is a constant related to the continuity of the injection from Hl/2 (r c)3
to L2(rc)3 and to the one of the trace mapping from Hl(OOi)3 to
Hl/2(rc)3 where a = 1,2.
The uniqueness condition (>. 1/ CO ~ 1) depends on the characteristics

of the interface and on the elastic properties of the materials. An analysis
of this condition on a simplified problem is given in Monerie (2000).
Because of the softening character of the interface law, the quasi-static
problem is ill-posed, and rapid changes in the evolution of the solution
may occur. A dynamic formulation of the problem is needed.
4.2 The 3D dynamic formulation

In order to deal with the non-viscous model used to describe the
fracture propagation, a dynamic formulation is developed on the basis
on the Non Smooth Contact Dynamics method given by Jean (1999)
and Moreau (1988, 1999) for granular media. To treat correctly the non-
smooth character of the contact law, the dynamic problem is written in
terms of differential measures and adapted to the adhesive and frictional
law.
For the contact of a single solid with a rigid obstacle (q is the dis-
placement and r the contact force), the equation of motion
M(q, t)ij = F(q, q, t) - r, (13)
is then written as :
M(q, t)dq = F(q, q, t)dt - rdv, (14)
where dq is a differential measure associated to q(t) :
rt2
it!
dq = q(tt) - q(tl) 'it2 > tl , (15)
341
and then:
[ t2 M(q, t)dq = [t2 F(q, q, t)dt - 1 rdv, (16)
l
~ ~ l~,bl
t2
q(t2) = q(tt} + q(r)dr. (17)
tl
The formulation due to Jean (1999) and Moreau (1988, 1999) is ex-
tended here to the adhesive frictional model (Monerie (2000)). The non
linear problem is then solved by using a generalization of a Newton-
Raphson algorithm to non-smooth functions due to Alart and Curnier
(1991). The implementation is performed in the 3D finite element code
LMGC (see Jean (1999) and Monerie and Acary (2001)).
5. Fracture of composite material

5.1 The problem
When a crack propagates in a composite material, various mecha-
nisms can occur depending on the characteristics of the interface : crack
trapping, crack bridging or fiber breaking as shown in Figure 4. Various
contributions can be found in He et al. (1989), Leguillon et al. (2000),
Mathur et al. (1996), Needlemann and Tvergaard (1998), Tvergaard
(1990), Siegmund et al. (1997), and Xu et al. (1997, 1998).
'-
P=
r
Figure 4. Crack trapping, crack bridging and fiber breaking.
The following simulations show the influence of the interface charac-

teristics on the crack propagation.
As presented in Figure 5, a cell composed of a single fiber surrounded
by a matrix domain is considered. The boundary conditions are chosen
in order to respect the symmetry and to approximate a semi-infinite
domain. An increasing displacement is prescribed on the top of the cell
and the crack is assumed to propagate along the symmetry plane at the
bottom, both in the matrix and in the fiber.
342
INITlAL
_ MFrH!..<. Cn~ rt>-U. GFaiu~ CRACK
1 '111 ... 4 .~'t:~ .
Figure 5. Mesh of the fiber and the surrounding matrix (12246 ddl).
5.2 Identification of the parameters

The interface model with no friction and no viscosity is used to simu-
late the crack propagation in the matrix and in the fiber (matrix-matrix
and fiber-fiber "interfaces"). The parameters C::" and (CT = 0 be- ct
cause of the symmetry of the problem) and wm and wI for both materials
(subscripts m and f relate respectively to the matrix and to the fiber)
are evaluated by considering :
- the stiffnesses of the materials (the Young moduli are Em = 350GPa
for the matrix, and EI = 200GPa for the fiber) and the maximum de-
formation before collapse, to identify the contact stiffnesses C::" and ct
(C::" = O.044Njp,m 3 , ct = O.025Njp,m 3 ),
- the decohesive energy and the restitution energy, to identify the thresh-
olds w m and wI (w m = 4Jjm 2 , wI = 20Jjm2 ) .
We introduce jj, which is the maximum stress supported by the in-
terfaces, defined by jj = JCNw. Thus, we have: jjl = 707MPa and
jjm = 418M Pa.
To simulate the fiber-matrix interface, the results presented here were
based on the same model, with no interface viscosity and no friction. As
shear is dominant for the present loading, the maximum stress jji for the
fiber-matrix interface depends mainly on the sliding effect and therefore
jji = J C~ wi. Various values of jji have been tested in the numerical
343
simulations in order to determine the influence of this parameter on the

crack propagation process.
5.3 The crack propagation

In Figures 6 and 7, the propagation of the crack is presented for two
kinds of interfaces :
- a strong interface in Figure 6 where iji = 247MPa,
- a weak interface in Figure 7 where iji = 3.5M Pa.
Figure 6 and 7 give the evolution of the reaction force on the top of the
cell when the prescribed displacement increases. The inserted pictures
show the solutions at different steps of the loading. The dark zones
correspond to the parts where there is no more adhesion, i.e. where the
crack is open or where the fiber slides along the matrix.
In Figure 6 (strong interface), it can be noted that the composite be-
haves like a single material with variable characteristics (corresponding
to those of the matrix and of the fiber) and it turns out to be rather
brittle: both the matrix and the fiber are broken when the prescribed
displacement is about O.ll"m.
Figure 7 shows that a weak interface protects the fiber from breaking
for a larger loading: part of the energy is dissipated into the fiber-
matrix interface and the total collapse of the cell occurs only when the
prescribed displacement is about OAl"m (i.e., when it is four times larger
than in the previous case).
In this work, the RCC model is extended for modeling crack propa-
gation in composite materials. A dynamic formulation is used because
viscosity is not considered for the evolution of the adhesion intensity.
The 3D simulations of the propagation of a crack taking into account
the debonding on the fiber-matrix interfaces shows the dependence of
the process on the properties of these interfaces. The results should help
to improve the toughness of fiber-reinforced composites by choosing ap-
propriate characteristics of these interfaces.
344
z
::::;0.08
1(i/~f = 0.35 1
...-
Q)
u
Q)
-50.06
s:::
o
s:::
o
·ZO.04
~
~
0.02
0.1 0.2 0.3 0.4 0.5

Prescribed displacement (J.Lm)
Figure 6. Propagation of the crack and of the breakdown area of the interface
(strong interface).
,---..
Z
"--'0 . 08
:::l
Q)
u
Q)
-50.06
s:::
o
s:::
o
'ZO . 04
~
~
0.02
0.1 0.2 0.3 0.4 0.5

Prescribed displacement (J.Lm)
Figure 7. Propagation of the crack and of the breakdown area of the interface (weak
interface) .
345
References
Alart, P., and Curnier, A. (1991) A generalised Newton method for contact problems
with friction, J. Mech. TMor. Appliq. 7, 67 - 82.
Allix, 0., Ladeveze, P., and Corigliano, A. (1995) Damage analysis of interlaminar
fracture specimen, Composite Structures 31(1), 61 - 74.
Allix, 0., Leveque, D., and Perret, L. (1998) Identification and forecast of delam-
ination in composite laminates by an interlaminar interface model, Composites
Science and Technologie, 58,671 - 678.
Bretelle, A.-S., Cocou, M., and Monerie, Y. (2001) Unilateral contact with adhesion
and friction between two hyperelastic bodies, Int. J. Engrg. Sci. 39, 2015 - 2032.
Cangemi, L. (1997) Frottement et adherence: modele, traitement numerique et appli-
cation al'interface fibre/matrice, Thesis, Universite de la Mediterranee, Marseille.
Chaboche, J.-L., Girard, R., and Levasseur, P. (1997) On the interface debonding
model, Int. J. Damage Mechanics 6, 220 - 257.
Chaboche, J.-L., Feyel, F., and Monerie, Y. (2001) Interface debonding model: a
viscous regularization with a limited rate dependency, Int. J. Solids Structures
38(18), 3127 - 3160.
Cocou, M., Pratt, E., and Raous, M. (1996) Formulation and approximation of qua-
sistatic frictional contact, Int. J. Eng. Sci. 34(7), 783 - 798.
Cocou, M., Cangemi, L., and Raous, M. (1997) Approximation results for a class
of quasistatic contact problem including adhesion and friction, in Variation of
domains and free-boundary problems in solids mechanics, edited by A. Argoul,
M. Fremond and Q.S. Nguyen, Solid Mechanics and its Applications 66, Kluwer,
211 - 218.
Cocou, M., and Rocca, R. (2000) Existence results for unilateral quasistatic contact
problems with friction and adhesion, Math. Model. Num. Anal. 34(5),981 -1001.
Fremond, M. (1987) Adherence des solides, J. Mec. The. Appl. 6(3), 383 - 407.
Fremond, M. (1988) Contact with adhesion, in Nonsmooth Mechanics and Appli-
cations, edited by J.J. Moreau and P.D. Panagiotopoulos, CISM - Courses and
Lectures 302, Springer Verlag, 177 - 221.
He, M.Y., and Hutchinson, J.W. (1989) Crack deflection at an interface between
dissimilar elastic materials, Int. J. Solids Structures 25, 1053 - 1067.
Jean, M. (1999) The Non Smooth Contact Dynamics method, Computer Meth. Appl.
Mech. and Engrg. 177(3-4), 235 - 257.
Ladeveze, P. (1995), A damage computational approach for composites: basic aspects
and mircomechanical relations, Computational Mechanics 17, 142 -150.
Leguillon, D., Lacroix, C., and Martine, E. (2000) Interface debonding ahead of a
primary crack, J. Mech. Phys. of Solids 48, 2137-2161.
Mathur, K.K., Needleman, A., and Tvergaard, V. (1996) Three dimensional analysis
of dynamic ductile crack growth in a thin plate, J. Mech. Phys. Solids 44(3), 439
- 464.
Michel, J.-C., and Suquet, P. (1994) An analytical and numerical study ofthe overall
behaviour of metal-matrix composites, Model. Simul. Mater. Sci. 2,637 - 658.
Michel, J.-C., Suquet, P., and Thebaud, F. (1994) Une modelisation du rOle des inter-
faces dans Ie comportement des composites a matrice metallique, Rev. Eur. Elem.
Finis 3(4), 573 - 595.
Monerie, Y. (2000) Fissuration des materiaux composites: role de l'interface fibre-
matrice, Thesis, Universite de la Mediterranee, Marseille.
346
Monerie, Y., Raous, M., Leroy, F.-H., Sudre, 0., Feyel, F., and Chaboche, J.-L. (1998)
Comparaison de lois d'interface fibre/matrice sur la base d'un modele uni-axial
d'experience de micro-indentation, in Comptes Rendus des Onziemes Jounees Na-
tionales sur les Composites, edited by J. Lamon and D. Baptiste, AMAC publisher,
565 - 574.
Monerie, Y., and Raous, M. (2000) A model coupling adhesion to friction for the
interaction between a crack and a fibre-matrix interface, Zeitschrift fUr Angewandte
Mathematik und Mechanik 80, 205 - 209.
Monerie, Y., and Acary, V. (2001) Formulation dynamique d'un modele de zone
cohesive tridimensionnel couplant endommagement et frottement, Rev. Eur. Elem.
Finis 10, 489 - 504.
Moreau, J.-J. (1988) Unilateral contact and dry friction in finite freedom dynamics, in
Non Smooth Mechanics and Applications, edited by J.J. Moreau and
P.D. Panagiotopoulos, CISM Courses and lectures 302, 1 - 82.
Moreau, J.-J. (1999) Some basics of unilateral dynamics, in Unilateral Multibody Con-
tacts, edited by F. Pfeiffer and Ph. Glocker, Solid Mechanics and its applications
72, Kluwer, 1 - 14.
Needleman, A. (1990) An analysis of tensile decohesion along an interface, J. Mech.
Phys. Solids 38(3), 289-324.
Needleman, A., and Tvergaard, V. (1998) Dynamic crack growth in a nonlocal pro-
gressively cavitating solid, Eur. J. Mech. A/Solids 17(3), 421 - 438.
Raous, M. (1999) Quasistatic Signorini problem with Coulomb friction and cou-
pling to adhesion, in New developments in contact problems, edited by P.Wriggers-
P.Panagiotopoulos, CISM Courses and Lectures 384, Springer Verlag, 101 - 178.
Raous, M., Cangemi, L., and Cocou, M. (1999) Consistent model coupling adhesion,
friction and unilateral contact, Compo Meth. in Appl. Mech. and Engrg. 177(3-4),
383 - 399.
Raous, M., Cangemi, L., and Cocou, M. (1997) Un modele couplant adherence et
frottement pour Ie contact unilateral entre deux soli des deformables, C. R. Acad.
Sci. Paris, Serie II b 329, 503 - 509.
Siegmund, T., Fleckand, N.A., and Needleman, A. (1997) Dynamic crack growth
across an interface, Int. J. of Fracture 85, 381 - 402.
'Truong Dinh Tien, J.-M. (1990) Contact avec adherence, Thesis, Universite de Paris
VI, Paris.
Tvergaard, V. (1990) Effect of fibre debonding in a whisker-reinforced metal, Mat.
Sci. Engrg. A125, 203 - 213.
Tvergaard, V., and Needleman, A. (1995) Effects of nonlocal damage in porous plastic
solids, Int. J. Solids Structures 32, 1063 - 1077.
Tvergaard, V., and Needleman, A. (1997) Nonlocal effects on localization in a void-
sheet, Int. J. Solids Structures 34(18), 2221 - 2238.
Vola, D., Pratt, E., Jean, M., and Raous, M. (1998) Consistent time discretization for
dynamical frictional contact problems and complementarity techniques, Rev. Eur.
Elements Finis 7, 149 - 162.
Xu, X.-P., Needleman, A., and Abraham, F.F. (1997) Effect of inhomogeneities on
dynamic crack growth in an elastic solid, Model. Simul. Mater. Sci. Engrg. 5, 489
- 516.
Xu, G., Bower, A.F., and Ortiz, M. (1998) The influence of crach trapping on the
toughness of fiber-reinforced composites, J. Mech. Phys. Solids 46(10), 1815 -
1833.
THE UNILATERAL FRICTIONAL
CONTACT OF A PIEZOELECTRIC BODY
WITH A RIGID SUPPORT
Dedicated to the memory of Professor P. D. Panagiotopoulos.
Paolo Bisegna
Dipartimento di Ingegneria Oivile, Universitd di Roma "Tor Vergata", Rome, Italy
Frederic Lebon
Labomtoire mecanique materiaux, Universite Lyon 1, Villeurbanne, I+ance
Franco Maceri
Dipartimento di Ingegneria Oivile, Universitd di Roma "Tor Vergata ", Rome, Italy
Abstract This paper is devoted to study the frictional contact of a linearly piezo-
electric body with a rigid foundation, under a quasi-static monotone
loading. Contact and friction are modelled by the laws of Signorini
and Coulomb. Primal and dual formulations of the contact problem
are presented and condensed on the boundary. A new numerical al-
gorithm, which is the dual version of the celebrated PANA algorithm
by P. D. Panagiotopoulos, is introduced to solve the dual discretized
formulation by a block-relaxation technique. Numerical examples are
presented, showing the effectiveness of the proposed algorithm.
1. Introduction
The problem of a piezoelectric body in frictional contact with a rigid
support, under a quasi-static monotone loading, is considered. This
is an extension of the classical Signorini-Coulomb problem. It has a
perspective technological importance: as an example, the relative motion
of two bodies may be detected by a piezoelectric sensor in frictional
contact with them.
In the first part of the paper, the governing equations are presented
in the framework of linear piezoelectricity (Tiersten, (1969». Two weak
347
J.A.C. Martins and M.D.P. Monteiro Marques (eels.), Contact Mechanics, 347-354.
348
formulations are established taking into account the non-associated char-
acter of the friction. The first one, extending to piezoelectric materials
the classical primal approach (see, e.g., Lebon and Raous, (1992)), is
written in terms of a variational equation coupled with an implicit varia-
tional inequality, and involves the displacement and electric-displacement
fields as unknowns. The second one, generalizing the dual formulation (
Telega, (1988); Klarbring et al., (1989)), is written in terms of a varia-
tional equation coupled with a quasi-variational inequality, and involves
the stress and electric potential fields as unknowns.
The second part of the paper is concerned with the numerical so-
lution of the nonlinear problem associated with the dual formulation
presented above. The nonlinear discretized problem is solved by using
an iterative algorithm based on a condensation procedure coupled with
a block-relaxation technique. The idea of this technique is to solve, at
the k-th step, two sub-problems: the first is a problem of friction with
given normal forces, and the second is a problem of unilateral contact
with prescribed tangential forces. This is the dual version of the PANA
algorithm due to P. D. Panagiotopoulos (Panagiotopoulos, (1975)).
In the last part of the paper a numerical application is presented,
in order to show the effectiveness and robustness of the proposed algo-
rithm. The frictional contact of a linear piezoelectric plane body with
a rigid support is analyzed. Different values of the friction coefficient
are considered and the influence of an actuation electric potential on
the detachment zone and on the distribution of the contact forces is
investigated.
2. Formulations
2.1. Notations
Let R be the set of real numbers. Let (.) and (x) denote the inner
and vector products, respectively. A body, comprised of linear piezoelec-
tric material, occupies an open bounded region 0 of R3 with sufficiently
smooth boundary 00. The body, subject to monotone slowly increasing
loads, is in unilateral frictional receding contact with a rigid obstacle
along the part aco of 00. This problem is quasi-static both in the
sense of electromagnetism (essentially, electromagnetic waves uncouple
from elastic waves), and in the sense of mechanics (Le., inertia forces
are negligible). The following fields are introduced: (i) the electric po-
tential cp, the electric field E, the electric-displacement field D, (ii) the
displacement field u, the strain field e and the stress field u.
The body 0 is subject to: (i) a volume charge p, a surface charge w
on awo, a prescribed electric potential CPo on al"0' where {awO, al"0}
349
constitutes a partition of an, (ii) body forces F, surface forces f on

aln and a prescribed displacement Uo on ado., where {aln, ado., aen}
constitutes a partition of an. With usual notations, the equilibrium and
compatibility equations are (Tiersten, (1969))
div D - p = 0 in 0., diva + F = 0 in 0.,
D . n = -won awn, an = f on aln,
(1)
E = -grad cP in 0., e = grads U in 0.,
cP = CPo on acpn, U = Uo on ado..
On the part aen of the boundary, where an unilateral frictional contact
with a rigid support may occur, the following equations prevail
Signorini's law UN::; 0, aN ::; 0, uNaN = 0
Coulomb's law IlaTII::; -p,aN (2)

if IlaT11 < -p,aN then UT = 0
if IlaT11 = -p,aN then UT = -AaT, A ~ O.
Here p, is the friction coefficient, UN = u'n, UT = U-UNn, aN = (an)·n,
aT = an-aNn and n denotes the external normal unit vector to an. The
Coulomb friction law is written in terms of displacements, because quasi-
static monotone loadings are considered (Duvaut and Lions, (1976)).
The rigid support is assumed to be an insulator (thus awn ;2 aen).
The free energy density cI>(e, D) of the material comprising the body
and its Legendre conjugate W(a, E) are
cI>(e, D) = lCe·e-he·D+lf3D·D
W(a, E) = ISa.a+da.E+ }fE.E (3)
where C (resp. S) is the fourth-order elasticity (resp. compliance) ten-
sor, f3 (resp. f) is the second-order impermeability (resp. dielectric)
tensor, hand d are third-order piezoelectric tensors. The constitutive
relationships derived from cI> and Ware, respectively
a=Ce-htD, E=f3D-he (4)
e = Sa + Of E, D = fE + da
Finally, we introduce the spaces of kinematically or statically admis-
sible fields
V - {v E (Hl(n))3, v = Uo on adn},
Kd {v E V;VN::; 0 on aen},
W {cp E H 1 (n), cp = CPo on acpn},
H {Tij E L2(n); Tij = Tji; div T + F = 0 in 0.; Tn = f on aln},
{T E H;TN::; 0, 117']'11::; -Wy on aen},
= {D E (L2(n))3;div D - p = 0 in n;D· n = -won awn}.
(5)
350
2.2. Primal formulation
The primal formulation of the contact problem leads to an implicit
variational inequality (Duvaut and Lions, (1976)) coupled with a varia-
tional equality
Find (u,D) E /Cd X K such that 'v'(v, <5) E /Cd X K

al(u,v - u) + hl(D,v - u) + j(u,v) - j(u,u) - Ll(v - u) ~ 0,
a2(D,6 - D) + h2(u, 6 - D) - L2(6 - D) = 0
(6)
with
at{u, v) - f C grads u . grads v dx,
0
Ll(V) - f F . v dx + f f· v dl,
0 Blo
j(u, v) - f I-'UN(U) IIVTII dl,
BcO
a2(D, 6) f {jD· 6 dx, (7)
0
L2(6) f rpo 6· n dx,
B<pO
hl(D,v) - f hgrad s V· D dx,
0
h2(U, 6) - - f hgrad s U· 6 dx.
0
2.3. Dual Formulation

The dual formulation of the contact problem leads to a quasi-variational
inequality (Telega, (1988); Panagiotopoulos and Haslinger, (1992)) cou-
pled with a variational equality
Find (u, rp) E /CC(UN) X W such that 'v'(T, X) E /CC(UN) X W

bl (u, T - u) + h (rp, T - u) - Ml (T - u) ~ 0, (8)
~(rp,X - rp) + h(u,x - rp) - M 2 (X - rp) = 0
with
bl(u,T) f S U· T dx,
0
Ml(T) - f uo· Tn dl,
Bd O
~(rp,X) f f grad rp. grad X dx,
0 (9)
M2(X) - f p X dx - f w· X dl,
0 Bw O
h(rp, T) - f d T . grad rp dx,
0
l2(U, X) - f d U· grad X dx.
0
351
Remark Usually, the formulations presented above are condensed on

the contact zone. In the case of the dual formulation, the condensation is
a generalization to the piezoelectric case of the one established by Telega
(Telega, (1988» and is based on the Mosco duality (Capuzzo-Dolcetta
and Matzeu, (1980), see also Klarbring et al., (1989».
3. Numerical procedures
3.1. Introduction
Classically, the primal formulation is equivalent to a fixed point prob-
lem on the sliding limit (Duvaut and Lions, (1976». In order to compute
the solution, the problem is discretized by finite elements and usually
condensed on the contact zone, on the degrees of freedom corresponding
to the normal and tangential displacements.
Here we focus on the dual formulation. The discretized problem is
condensed on the contact zone on the degrees of freedom corresponding
to the normal and tangential reactions, and solved by a block-relaxation
method.
3.2. Block-relaxation technique: D-PAN A

algorithm
After discretization and condensation, the vector of the degrees of
freedom corresponding to the normal part of the contact reactions is
denoted by u and the one corresponding to the tangential part is denoted
by T.
The proposed block-relaxation technique consists of the successive
solution, at each iterate, of two sub-problems: the first one is a problem
of friction with given normal forces, and the second one is a problem
of unilateral contact with prescribed tangential forces. This algorithm,
being the dual version of the PANA algorithm (Panagiotopoulos, (1975);
Panagiotopoulos, (1985)), is named D-PANA.
The compliance matrix K is decomposed in four blocks as follows
(10)
where aUu (resp. aTT ) is the sub-matrix corresponding to the normal

(resp. tangential) degrees of freedom. Analogously, the generalized dis-
placement vector b is decomposed in two blocks
(11)
352
x
3
o v
l:S::,ssS::Ss:S;ssiSss~s;s:sssss:S;ssiSs~ss;s:sssss:S;ss~sss----",x
2
Figure 1. Piezoelectric body in contact with a rigid support. Elastic, piezoelectric,

and dielectric constants of lithium niobate: Ou = 219.2 GPa, Oss = 251.6 GPa,
044 = 95.2 GPa, 014 = -14.8 GPa, hSI = 0.778 GV 1m, h33 = 5.06 GV 1m, h22 = 6.43
GV1m, hl5 = 9.51 GV1m, {3u = 2.57 Gm/F, {333 = 3.89 Gm/F
where bCT (resp. bTl corresponds to the normal (resp. tangential) degrees
of freedom.
At the k-th step of the proposed algorithm, the first su~problem
allows to find Tk, solution of:
mi~nTtaTT l' - Tt(bT - aTCT uk-d, 11'1 ~ -I-'Uk-I}, (12)
where Uk-I ~ 0 comes out from the step k - 1. Since aTT is positive
definite and the minimization set is nonempty closed convex, this sub-
problem has a unique solution.
The second sub-problem allows to compute Uk, solution of:
minqHutaCTCT u - Ut(bCT - (aTCT)tTk)' u ~ o}. (13)
As before, since aCTCT is positive definite and the minimization set is
nonempty closed convex, this su~problem has a unique solution.
In Bisegna et al., (2001) it is shown that, for a sufficiently small friction
coefficient 1-', the proposed D-PANA algorithm is convergent.
In this section, a numerical application is presented. A Cartesian
coordinate frame (O,XI,X2,X3) and a piezoelectric body extending in-
definitely in the Xl direction are introduced. The body has a rectangular
cross section, with dimensions 0.1 m and 0.05 m along the X2 and X3
axes, respectively (figure 1). It is initially in contact with a rigid plane
support along the side X3 = O. It is acted upon, on the opposite side,
by a uniform normal pressure of 100 N/mm2 distributed over the in-
terval 0.0625 m ~ X2 ~ 0.1 m, and by a difference of electric potential
V between its vertical sides. The electric and mechanical loadings are
constant along Xl.
353
(a) (b)
2OOr.====:::::;-~~-~---"I 100 r;=:====:::;-~-~-""7---'
-&- 11=0 -&- 11=0
150 -+- 11=0.2 -+- 11=0.2
~
..
~ --- 11=1 50 ---11=1
!:l 100 !!!

!!! 'liS
'liS iii
~ 50 ""i
(;
t: S'" -50
-so0 0.02 0.04 0.06 0.08 0.1
-100'---~-~-~-"">""'--'
o 0.02 0.04 0.06 0.08 0.1
contact zone (m) contact zone (m)
(c) (d)
200 ,
1'_' '02: 1 ,,
~
100
- OV
~ 150 - - 2V ~ 50 --2V I
'"
!!:.
.. 100
.. ../... I
~ 0 ,.' I
~
"., "
iii I----~c\,
,
iii 50 ~ \'
E ~ (
0
t:
0
S -50 \
,, \
, \
-50 -100'---~~~-~""""'~....L......I
0 0.02 0.04 0.06 0.08 0.1 o 0.02 0.04 0.06 0.08 0.1
contact zone (m) contact zone (m)
Figure 2. Influence of the friction coefficient J! and of the applied voltage V on the
contact forces: (a) normal and (b) tangential forces for three values of the friction
coefficient: J! = 0, J! = 0.2, J! = 1, the sliding limit is presented in the case J! = 0.2.
(c) normal and (d) tangential forces for three values of the applied voltage: V = -2
V, V = 0 V, V = 2 V, the sliding limit is presented in the case V = 2 V
The body is made of lithium niobate, whose material parameters, in

usual notations (Tiersten, (1969)), are given in the caption of figure l.
An equilibrium piezoelectric plane finite element (Maceri and Bisegna,
(1998)) is used to discretize the continuous problem.
The first numerical results investigate the influence of the friction co-
efficient I-' on the contact zone and on the distribution of contact normal
and tangential forces (figures 2a and 2b). Three different values of I-' are
considered: I-' = 0, I-' = 0.2, I-' = 1, and the applied voltage V is kept
fixed to zero. In this example, it turns out that the friction coefficient
has a weak influence on the contact force distribution.
Other numerical results investigate the influence of the applied voltage
V (figures 2c and 2d). The distribution ofthe contact forces is computed
for three different values of V: V = -2 V, V = 0 V, V = 2 V, when
the friction coefficient is kept fixed to 0.2. The sensitivity to the applied
354
voltage V appears to be significant and could be promisingly exploited
in applications.
Acknowledgments
The financial supports of CNR and AS! are gratefully acknowledged.
This research was developed within the framework of Lagrange Labora-
tory, an European research group between CNRS, CNR, University of
Rome "Tor Vergata", University of Montpellier II, ENPC and LCPC.
References
Bisegna, P., Lebon, F., and Maceri, F. (2001) D-PANA: a convergent block-relaxation
solution method for the discretized dual formulation of the Signorini-Coulomb
contact problem, C. R. Acad. Sci. Paris Ser. I Math., to appear.
Capuzzo-Dolcetta, I., and Matzeu, M. (1980) Duality for implicit variational problems
and numerical applications, Numer. Punct. Anal. Optim., 2, 231 - 265.
Duvaut, G., and Lions, J. L. (1976) Inequalities in mechanics and physics, Springer-
Verlag, Berlin-New York.
Tiersten, H. F. (1969) Linear piezoelectric plate vibrations, Plenum Press, New York.
Klarbring, A., Mikelic, A., and Shillor, M. (1989) On friction problems with normal
compliance, Nonlinear Anal., 13, 935 - 955.
Lebon, F., and Raous, M. (1992) Friction modelling of a bolted junction under internal
pressure loading, Comput. & Structures, 43, 925 - 933.
Maceri, F., and Bisegna, P. (1998) The unilateral frictionless contact of a piezoelectric
body with a rigid support, Math. Comput. Modelling, 28, 19 - 28.
Panagiotopoulos, P. D. (1975) A nonlinear programming approach to the unilateral
contact and friction boundary value problem in the theory of elasticity, Ing.-Arch.,
44, 421 - 432.
Panagiotopoulos, P. D. (1985) Inequality problems in mechanics and applications.
Convex and nonconvex energy junctions, Birkhuser, Boston.
Panagiotopoulos, P. D., Haslinger, J. (1992) On the dual reciprocal variational ap-
proach to the Signorini-Fichera problem. Convex and nonconvex generalization,
Z. Angew. Math. Mech., 72, 497 - 506.
Telega, J. J. (1988) Topics on unilateral contact problems of elasticity and inelastic-
ity, in Nonsmooth mechanics and applications, edited by J.-J. Moreau and P. D.
Panagiotopoulos, CISM Courses and Lectures 302, Springer-Verlag, Vienna, 341 -
462.
Dostal, Z., Haslinger, J., and Kucera, R. (2001) Implementation of the fixed point
method in contact problems with Coulomb friction based on a dual splitting type
technique, Preprint.
Note. The D-PANA algorithm has been presented for the first time during the
CMIS 2001 Conference. Prof. J. Haslinger, attending the Conference, drew to the
authors' attention a preprint (Dostal et al., (2001)) presenting the same algorithm
named FPMII.
CONTACT BETWEEN 3D BEAMS WITH
RECTANGULAR CROSS-SECTIONS
Przemyslaw Litewka
Institute/or Structural and Computational Mechanics, University Hanover
Appelstr. 9A, 30167 Hanover, Germany
Peter Wriggers
Institute/or Structural and Computational Mechanics, University Hanover
Appelstr. 9A, 30167 Hanover, Germany
Abstract A frictionless contact between 3D beams is analysed. The contact element

allows for large displacements but small strains. An effective contact-search
algorithm, which is necessary to determine an active set for the contact
contribution treatment, is developed. The contact element uses the physical
shape functions of Timoshenko beam element. A consistent linearisation of the
contact contribution within the penalty approach is derived. One of many
solved numerical examples shows the efficiency of the prepared approach.
1. Introduction
Contact between beams undergoing large displacements can be met often

in practice. Examples of problems where beams get in contact while moving
include machines, robots, woven fabrics, tennis raquet stringing, etc.
The scientific effort concerning contact problems between 2D or 3D
bodies, with or without friction is enormous and numerous papers could be
mentioned. Contrary to this, the literature concerning beam-to-beam contact
is not so broad. There exist only few contributions: Maker and Laursen
(1994) concerning the contact between beam and 3D body and Wriggers and
Zavarise (1997), Zavarise and Wriggers (2000) dealing with the contact
between beams with circular cross-section. In this work a more challenging
case of beams with rectangular cross-section is considered. The results from
355
356
Wriggers and Zavarise (1997) concerning the formulation and linearisation

of frictionless contact contributions are used and further developed herein.
Different cases of pointwise contact between edges of beams must be
considered. This requires complete modification of the contact-search
algorithm. The displacement field is approximated using the physical shape
function concept (Rakowski (1990)). The plane rectangular cross-sections
remain plane but may not be perpendicular to the deformed beam axis.
In Section 2 details of the contact search routine are discussed. The weak
form of contact contribution within the active set strategy using penalty
method is given in Section 3. Section 4 presents the matrix formulation of
the linearised contact contributions to be used within FEM. A typical
numerical example, one of many solved, confirming the effectiveness of the
formulation is shown in Section 5. Section 6 contains concluding remarks.
2. Contact search
For the beam-to-beam contact considered herein it is assumed that:

a) beams contact pointwise, contacting parallel beams are not considered,
b) beam cross-sections undergo rigid body movements without deformation,
c) contact occurs between beam edges (max. two edges per beam, Fig. 1).
l'~l~
1 edge to 1 edge I edge to 2 edges
2 edges to 2 edges
not allowed
Figure 1. Possibilities of contact between beams
The first step of contact search is to find the pair of closest elements. The
pair of elements with the shortest distance between fictitious middle points
defmed by the current positions of real element nodes is to be located.
Further on pairs edge-edge or edge-axis are considered. They have initial
position vectors X, and X2 , displacements u, and U2 and current position
vectors: Xj =X j + Uj, i =1, 2. Orthogonality conditions (1) are used to fmd a
pair of closest points on edges, C, and C2, given by co-ordinates rCl and rC2:
i·x1,1.. =0, (1)

357
x
where =X2 - XI and O,i =dO,i / dri'
Generally (1) represents a set of non-linear equations. To solve it
iteratively by Newton's scheme its linearisation is required:
Functions of displacements Ui are expressed in terms of initial nodal co-

ordinates and nodal displacements. The approximation based on physical
shape functions is used. It is obtained considering displacements of comers
or centres of rectangular cross-sections subjected to rigid body movements.
In the second and third stages the closest edges are determined. For each
beam two edges are found, which are the closest ones to the axis of the
second beam. In this process Newton iteration scheme (2) is used. The edges
are labelled 1.1, 1.2 and 2.1, 2.2, respectively.
The fourth stage considers the four closest edges found previously. A
search for the closest points for each of the four edge pairs (1.1-2.1, 1.1-2.2,
1.2-2.1 and 1.2-2.2) is performed according to (2).
Vcc .---------
VCA2
Figure 2. Vectors and angles for penetration check
With the closest points for each pair of edges located, a penetration check
is done. To this end the following three vectors are defined (see Fig. 2):
• vector between closest points C I and C2: VCC =XCI - XC2,
• vector between C I and its counterpart on beam axis AI: VCAI =XCI - XAh
• vector between C2 and its counterpart on beam axis A 2: VCA2 =XC2 - XA2.
358
Further on the values of angles u, =L(vee , VeAl) and U2 =L(-Vee, VeA2)
are considered. If there is penetration both angles are acute, in the opposite
case they are obtuse, see Fig. 2. Hence, the condition for penetration reads:
(3)
3. Weak form of contact contribution
Solving contact problems in mechanics involves the minimisation of an

objective functional with inequality constraints. Using the active set strategy
one turns them into equality constraints. One of the available approaches is
the penalty method known from optimisation theory (e.g. Luenberger
(1989». In our case minimisation of the strain energy IT can be split into
three parts corresponding to both beams and to the contact contribution:
(4)
where E is penalty parameter and gN= II XCI - xC211 is the penetration.

Its linearisation required for the Newton iteration scheme has the form:
Expressions for L\gN and 3gNwere derived in Wriggers and Zavarise (1997):
Yn,
L\g N =(L\U C2 - L\uCl
(6)
3g N = (3u C2 - 3UCl Yn,
where n =Vee / gN is a unit vector normal to both contacting edges.

The value of L\3gN is refined by adding the term due to L\OUi which is
non-zero for the adopted displacement approximation. Its full form reads:
L\og N = (OU C2 ,2L\fc2 - OUCl,lNCI Yn + (L\UC2,20fC2 - L\UCI,IOfCl)T n +

+ (X C2 ,22L\fc2 0fC2 - XCI ,llL\fCl OfCl Yn + (L\OU C2 - L\Ou Yn +
Ci
(,
+ \OU C2 + XC2 0f 2 oU CI - XCI1 0fCl
C2 -
Y(l-n(8)n) . (7)
, , gN
. (L\U C2 + XC2 ,2L\fc2 - L\uCl - xcl,lL\rcl ),

359
4. Finite element discretisation of contact contribution
Here the contact contributions to (4) and (5) are expressed in terms of
nodal displacements for a pair of contacting elements UM) and UM2. This is
done by using the representation of edge functions X;, evaluated at the closest
points C) and C2 • In this way the extra terms to be included in the tangent
stiffness matrix and the residual vector for the Newton iterations scheme are
obtained.
Values of L\uc;, oUc;, L\uc;,; and oUc;,; present in expressions (6) and (7)
are determined by the partial differentiation of displacement approximations
Uj with respect to nodal displacements and local co-ordinates rio This yields
the following matrix relations:
L\UCj =G; . L\u Mi '

OU c; =G; 'OU Mi '
(8)
L\u c;;=H; . L\U Mi '
oU c;; =H; . oU Mi '
Variations and linearisations of co-ordinates of contacting points rc; also

must be expressed in terms of nodal displacements. Variation of
orthogonality conditions (I) yields the following relation:
(9)
where A is the matrix of coefficients at L\rj in equation (2), evaluated at the

contact points.
To shorten notation one more matrix is introduced:
(10)
Values of L\ou; can be determined by double differentiation of

displacements Uj. Yet, this would lead to very complicated expressions.
Instead the perturbation method with respect to matrices G j and finite
difference scheme are applied. Matrices G j with dimensions 3x12 are split
360
into three vectors: G;=(G;J, G a , G;3) and the perturbation method yields six
(12xI2)-matrices Gdij,j= 1, 2, 3. They are used to get:
(11)
With matrix relations (8), (9), (10) and (11), variation, linearisation and
linearisation of variation of gap function (6) and (7) can be expressed by
nodal displacements. Their substitution into (4) and (5) yields the contact
contributions to variation and linearisation of variation of strain energy:
80IT e = (OU M2 T ,OU MJ T). [K e 14X24 . (8U M2 T,8UMJT)T

(12)
oITe = (OU M2 T,ou MJ T). [Re 14XJ '
Finally the contact contribution to the residual vector has the form:
(13)
And the tangent stiffness matrix due to contact is:
o
3
- LnjG djJ
j=J
where nj are the components of the normal vector B.

361
5. Numerical example
Three beams, shown in Fig. 3 are analysed. Two of them, crossed at the
angle 15.40 in the plane XY, are of cantilever type with displacements
d3 = At = 10 along Z-axis moving their tips against each other. The third
beam, the "free" one, is moved as a rigid body towards two former ones by
displacements dl = d2 = 100 along X-axis applied at its tips. Displacements
are applied in 80 increments, all beams are divided into 10 elements and
have the same data: E=20000, v =0.3, dimensions 5x5xlOO. Penalty
parameters are: for contact between the crossed beams E= 22500, for contact
between the "free" beam and two other ones E= 1200.
Results of the convergence study for four selected load increments are
presented in Table 1. The energy tolerance is taken as (etol)= 10-8• The
corresponding stages of a deformation process are also shown in Fig. 3.
A very fast convergence is observed throughout the whole process of
deformation. Only 3 to 4 iterations per increment suffice to get eqUilibrium.
The maximum penetration value is about 1% of the size of beam cross-
section. Reaching this level was the objective of penalty parameter tuning.
XYVIEW ~. XZVIEW YZVIEW ~.
3D VIEW
T= 1.00
Figure 3. Initial configuration and four stages of defonnation process

362
Table 1. Convergence study and penetration.
Increment Number of iterations gN ("free" beam) gN (crossed beams)

20 3 open 0.0270
40 3 0.0038 0.0463
60 4 0.0151 0.0476
80 3 0.0566 0.0542
6. Concluding remarks
In the present paper frictionless contact between beams with rectangular

cross-section is considered. The penalty method is used to include the
contact constraints in the strain energy formula. Contact between beams is
assumed to be pointwise, beams are contacting at their edges. The consistent
linearisation of contact contributions to be applied in the Newton iteration
scheme is derived.
Out of several solved numerical examples one typical is presented. It
concerns the multiple beams contacting one with another and depicts the
efficiency of the derived beam-to-beam contact element. A very good
convergence rate is observed.
The future work will focus on the inclusion of friction in the contact
model as well as on the refinement of the approximation of beam edge
function to simplify the FEM formulation.
Acknowledgment
Przemyslaw Litewka acknowledges the grant of Alexander von

Humboldt Foundation.
References
Luenberger, D.G. (1989) Linear and Nonlinear Programming. Addison-Wesley Publishing

Company: Reading, Massachusetts.
Maker, B.N., and Laursen, T.A. (1994) A finite element formulation for the rod/continuum
interactions: the one-dimensional sliding, Int. 1. Num. Meth. Enging 37, 1-18.
Rakowski, 1. (1990) The interpretation of the shear-locking in beam elements, Compo Struct.
37, 769-776.
Wriggers, P., and Zavarise, G. (1997) On contact between three-dimensional beams
undergoing large deflections, Comm. Num. Meth. Engng 13, 429-438.
Zavarise, G., and Wriggers, P. (2000) Contact with friction between beams in 3-D space, Int.
1. Num. Meth. Enging, 49, 977-1006.
ALGORITHMS FOR THERMOELASTIC
WEAR PROBLEMS
Peter Ireman, Anders Klarbring and Niclas Stromberg

Linkoping University, Sweden
Abstract This contribution concerns the numerical treatment of discrete ther-

moelastic wear problems. Two different approaches, both utilizing a
non-smooth Newton method as non-linear equation solver, are outlined
and compared. Furthermore, a numerical example shows how the pre-
dicted wear gap is influenced by the bulk properties of the contacting
bodies.
1. Introduction
The present paper concerns the numerical treatment of the ther-
moelastic model of contact friction and wear developed in Stromberg
et al. (1996). Here two different approaches are presented and com-
pared. Both are based on the Newton method for non-smooth equations
suggested by Pang (1990). The first idea is to apply the Newton method
directly to the system of coupled thermomechanical equations. This ex-
tends the approach developed in Stromberg (1998) for one thermoelastic
body in frictional wearing contact with a rigid support to the case of
two-body contact. The second idea is to use a strategy where essen-
tially mechanical and thermal subproblems, respectively, are identified
and solved in an iterative process, which from an algorithmic point of
view is a Gauss-Seidel scheme. Also in this case the non-smooth New-
ton method is utilized as solver for the mechanical subproblem, while
the thermal subproblem is simply a system of linear equations.
2. The discrete thermoelastic wear problem

Let us consider two thermoelastic bodies in unilateral, frictional wear-
ing contact. The governing equations for the thermomechanical contact
problem are the equilibrium equation, the law of action and reaction,
363
l.A.C. Martins and M.D.P. Monteiro Marques (eds.J, Contact Mechanics, 363-370.
364
the constitutive law of linear isotropic thermoelasticity, Signorini's con-

tact conditions, Coulomb's law of friction, Archard's law of wear, the
balance of energy, Fourier's law and an assumption that the thermal
contact conductances depend linearly on the normal contact pressure.
These were formulated as a generalized standard model in Stromberg et
al. (1996).
A formal discretization of the problem is obtained through a vari-
ational formulation of the governing equations, introducing finite ele-
ment approximations, evaluating integrals over the contact surface by
an appropriate quadrature rule and approximating time rates by Euler
backward finite differences. These steps are described in detail in Ireman
et al. (2001), see also Johansson and Klarbring (1993). Furthermore, the
variational inequalities equivalent to Signorini's contact conditions and
Coulomb's law of friction are reformulated as equations by means of pro-
jections, see Klarbring (1992) and see also Stromberg (1997) and (1998).
For two-dimensional problems, the result is the following system of
non-smooth equations for each time interval [tk, tk+l]:
= 0, (1)
where Yc = (dl,d 2,T 1 ,T2,Pn,Pt). The first two equations are the
discrete equilibrium equations for each body, respectively, where, letting
index l = 1, 2 refer to the two bodies, Klare the stiffness matrices,
KI are the dilatation matrices, d l are vectors of nodal displacements,
TI are vectors of nodal temperatures, FI are external nodal forces, P n
is a vector of the nodal normal contact forces and P t is a vector of
nodal tangential contact forces. The matrices en and C t are kinematic
transformation matrices. The next two equations are the discrete bal-
ance of energy for each body respectively, where MI are the capacity
matrices, 0 1 the conductivity matrices and QI = MITI(tk). Assuming
that the nodal values are ordered such that indices corresponding to con-
tact nodes are put first, the vectors LI, representing the heat generated
by friction and wear and the heat exchange between the two bodies, have
the form LI = (Lf, 0). The entries of L} and L~;' respectively, are given
by
365
(2)
and
(3)
where (x)+ = max(O, x), {}l are the thermal contact conductances, p, is
the coefficient of friction, k is the wear coefficient, {wf} = C t (d1 - d2)
and Wt = wt(tk+d - Wt(tk). Here and in the following, index M refers
to contact nodes. One might also note that the assumption that the
overall contact conductances depend linearly on the contact pressure is
included explicitly in (2) and (3), {}l are treated as constants.
Finally, lIn is the projection equivalent to Signorini's contact con-
ditions with Archard's law of wear inserted and lIt is the projection
equivalent to Coulomb's law of friction. These projections are defined
by
and
where r > 0 is an arbitrary parameter, {w~} = Cn(d 1 - d2) and Wo =

W(tk)'
It can be shown that the system of equations (1) is B-differentiable (B
for Bouligand), i.e. Lipschitz continuous and directionally differentiable
in all directions, implying that the Newton method suggested by Pang
(1990) and described in the next section can be applied directly to (1).
3. Algorithms
In this section we present two different strategies for solving H(y) = 0
where H(y) is B-differentiable.
366
The first strategy is to solve H (y) = 0 directly using a modification

of the Newton method for B-differentiable equations suggested by Pang
(1990).
The second strategy is to decompose the system of equations according
to y = (Yl' Y2),
(4)
and to treat these in an iterative process, which from the algorithmic

point of view is a Gauss-Seidel scheme. As will be seen later this is
particularly attractive when solving (1).
3.1. Pang's Newton Method

Pang (1990) suggested the following algorithm for solving H(y) = 0
where H (y) is B-differentiable:
Algorithm BN: Let f3, , and f be given scalars with f3 E (0,1)" E

(0,1/2) and f small. Repeat the following steps for each time increment
k + 1:
o Let yO be given from the previous time step k and set j = O.
1 Find a search direction z such that
H(yj) + H'(yj; z) = 0, (5)

where H' (yj; z) is the directional derivative.
2 Let a j = f3 mj , where mj is the smallest integer m ~ 0 for which
the following decrease criterion holds:
3 Set yj+l = yj +ajz.
4 If 1>(yj+l) ~ f, then terminate with yj+l as an approximate zero

of H(y). Otherwise, replace j by j + 1 and return to step l.
Since (5) is generally non-linear, the directional derivative is modified
in such a manner that this equation always becomes linear, for details see
Ireman et al. (2001). In addition an upper bound for mj is introduced
in order to prevent the algorithm from stalling. The direct application
of the Newton method is referred to as algorithm MBN.
367
3.2. Iterative strategy

Instead of solving H(y) = 0, one can use the decomposition (4) and
treat the different parts using the following Gauss-Seidel-type algorithm:
Algorithm GS: Repeat the following steps for each time increment
k + 1:
o Set y~=o = Y2(tk)
1 Solve if I (yi+ l ) = HI (yi+l, y~) = 0 to get yi+l.
2 Solve if2(y~+I) = H2(Yi+l,y~+I) = 0 to get y~+I.
3 Terminate if !H(yi+I)H(yi+l) ~ f. Otherwise replace i by i +1
and return to step 1.
When this strategy is applied, we identify HI and H 2 with the sub-
problems defined by
} =0,
and
respectively, where YM = (db d 2, P n , P t ) and YT = (TI' T2)' Note

that this can be done in two different ways, where either the mechanical
subproblem or the thermal subproblem is solved first.
When the iterative strategy is applied, the mechanical subproblem
is solved by Pang's Newton method, while the thermal subproblem
is simply a system of linear equations. This fact makes the iterative
strategy attractive if the number of Gauss-Seidel iterations is kept small.
The application of this approach is referred to as algorithm GSMT,
when the mechanical subproblem is solved first and algorithm GSTM
when the thermal subproblem is solved first.
4. Numerical example
Consider the two thermoelastic bodies, denoted punch and founda-
tion, shown in figure 1. The potential contact surface is defined by the
lower end of the punch.
368
u(t)
Punch
Foundation
N ~ 7-J-
Figure 1. Geometry and boundary conditions of the example.
The dimensions of the punch are 20 x 4 [mm2] and the dimensions

of the foundation are 40 x 20 [mm2]. The upper end of the punch is
subjected to uniformly distributed tractions t(t) = -200el [N/mm2] and
prescribed cyclic displacements u(t) = 0.15e(t)el [mm], where one cycle
of e(t) is shown in figure 2. Furthermore the lower end of the punch
is locked in the e2-direction and the midpoint is also locked in the el-
direction. All boundaries except the contact surfaces are assumed to be
thermally insulated.
1.0
Figure 2. One cycle of the prescribed displacement.

369
For the bodies, constitutive parameters corresponding to steel or alu-
minum are used, while for the contact interface we choose jJ = 0.3,
k = 10- 11 [mm2 /N] and ,/?l = ,/?2 = 10- 3 [W / N . K]. The initial contact
gap is of Hertz-type with maximum depth 0.05 [mm].
4.1. Comparison of the three algorithms

Let us compare the performance of the algorithms by considering the
execution statistics from 100 cycles using steel as bulk properties. Table
1 shows the average values of the number of Gauss-Seidel iterations per
time increment (GS/inc.), the number of Newton iterations per time
increment (Newt./lnc.), the number of line searches per Newton iter-
ation (Arm./Newt.) and the measured CPU time per time increment
(CPU/inc.) normalized to the CPU time obtained for algorithm MBN.
Note that the same stopping criterion: !Hb(ye)H c(ye) ~ € is used in
all algorithms.
Table 1. Execution statistics for the example.
Algorithm GS/inc. Newt/inc. Arm/Newt. CPU/inc.

MBN 5.27 1.46 1.00
GSMT 3.80 8.92 1.27 0.98
GSTM 3.85 8.92 1.27 0.97
The execution statistics indicate that algorithms GSMT and GSTM

are slightly more efficient than algorithm MBN. This difference is ex-
pected to be more significant when the number of degrees of freedom is
increased.
4.2. Results for different constitutive settings

Let us now discuss the predicted wear gap, shown in figure 3, for dif-
ferent settings of the bulk properties. The wear gap is evaluated after
100 cycles of the prescribed displacement for different combinations of
steel and aluminum but using the same wear coefficient, coefficient of
friction and contact conductances. In addition the solution for the ther-
moelastic problem is compared to the one for the purely elastic problem,
i.e. when the thermal dilatation is removed from the model.
The major point illustrated by figure 3 is that the calculated wear gap
might be quite different in shape as well as in maximum depth depending
only on whether the thermal dilatation of the bodies are taken into
account or not. One might also conclude that less material is worn away
in the cases when the foundation is made of aluminum. This is due to
370
steel on steel 0.14

steel on alum.
0.14
0.12 0.12
S 0.1
,, S 0.1
..§, 0.08 /
/
/
S
~ 0.08
3 0.06
3 0.06
0.04 0.04
-10 -5 0 5 10 -10 -5 0 5 10
x [mm] x [mm]
0.14
alum. on steel 0.14
alum. on alum.
0.12 0.12
-----
S 0.1
S 0.1
..§, 0.08 /
/ "-
S
~ 0.08
/
3 0.06
3 0.06 ...... ------
0.04 0.04
-10 -5 0 5 10 -10 -5 0 5 10
X [mm] x [mm]
Figure 3. Wear gap after 100 cycles when thermal effects are included (solid line)
and excluded (dashed line)
the larger elastic deformations of the foundation compared to the cases

when it is made of steel.
References
Ireman, P., Klarbring, A. and Stromberg, N. (2001) Finite element algorithms for
thermoelastic wear problems. Submitted for publication.
Johansson, L. and Klarbring, A. (1993) Thermoelastic frictional contact problems:
modelling, FE-approximation and numerical realization. Compo Meths. in Appl.
Mech. Engng. 105, 181 - 210.
Klarbring, A. (1992) Mathematical programming and augmented Lagrangian methods
for frictional contact problems. Proc. Contact Mechanics Int. Symp., 409 - 422.
Pang, J.-S. (1990) Newton's method for B-differentiable equations. Math. Operations
Research 15, 311 - 34l.
Stromberg, N., Johansson, L. and Klarbring, A. (1996) Derivation and analysis of a
generalized standard model for contact, friction and wear. Int. J. Solids Structures
33, 1817 - 1836.
Stromberg, N. (1997) An augmented Lagrangian method for fretting problems. Eur.
J. Mech., A/Solids 16, 573 - 593.
Stromberg, N. (1998) Finite element treatment of two-dimensional thermoelastic wear
problems. Compo Meths. in Appl. Mech. Engng. 177, 441 - 455.
CONVERGENCE STUDIES FOR 3D SMOOTH
FRICTIONAL CONTACT ELEMENTS BASED ON
THE QUARTIC BEZIER SURFACES
Lovre Krstulovi6-0para and Peter Wriggers

Institut for Baumechanik und Numerische Mechanik
Universitiit Hannover, AppelstraJ3e 9a, D-30J67 Hannover, Germany
Abstract A 3D smooth triangular frictional node to surface contact element is developed

using an abstract symbolical programming approach. Such an element is used
in combination with tetrahedral continuum elements suitable for the automatic
mesh generation. The C1-continuous smooth contact surface description is
based on the six quartic Bezier surfaces. The approach is based on the elastic-
plastic tangential slip vector decomposition, non-associated frictional law,
penalty method and weak formulation. The presented convergence studies
provide information about the benefit of using the smooth contact approach.
1. Introduction
3D smooth contact elements have already been introduced and described

in Krstulovi6 et al. (2000) and Krstulovi6 (200 l). The present smooth node
to surface contact approach is based on the smoothing of a contact surface by
means of six quartic B6zier surfaces (see Figure In). The surfaces are
spanned between the centroids of the closest triangular contact surface and
three neighboring contact surfaces (Figure 11)' Such a geometry definition
ensures the C l continuity between adjacent quartic surfaces. When combined
with tetrahedral continuum finite elements, this contact element, suitable for
automatic mesh generation, enables smoothing of non-uniform contact
surfaces. For each active contact set, six quartic surfaces are evaluated. The
closest one is then selected and used for the calculation of the residual and
the tangent matrix. Friction is modeled by the non-associated Coulomb law.
The chosen implicit formulation is based on the "elastic-plastic" approach in
371
372
which tangential slip is split into stick ("elastic") and slip ("plastic") part
(Wriggers (1987}), thus enabling quadratic rate of convergence within the
Newton-Raphson iteration loop. The weak fonnulation and the penalty
method are used to describe the contact contribution to the slave nodes.
The main benefit of smoothing the master surface is that even a relatively
coarse mesh can provide results that are satisfactorily close to the physical
reality. To obtain similar results with the simple node-to-surface elements,
the discretized mesh must have a very high density. Hence, if the discretized
mesh density is increased, a point where there is no difference between the
simple and the smooth node-to-segment element should be reached.
According to this, the simple node-to-surface and the smooth elements are
compared in Chapter 3.
Figure 1. The tetrahedral element and the smooth contact surface consisting of the six Bezier
surfaces
2. Definition of the smooth surface geometry
In this sub-chapter the definition of a single Bezier surface, shown in

Figure In, is described. All other five surfaces are fonnulated analogously.
Fifteen Bezier points (see Figure 2n) needed for the definition of a quartic
Bezier surface belong to the two adjacent flat contact segments (Figure 21).
Generally a triangular Bezier surface is defined (see Bohm et al. (1984)
or Farin (1993» in the explicit fonn as:
373
x(n, ~) =L,b;,j,k B;~j,k (n, ~) tI V+ }+k)=4,

!J.
(1)
B;,j,k
4
~n ,~ ) =-'-n'
4'
., "k'
..(
~ J l-n- ~
Y tI V+ }+k)=4.
1.J . .
The symbol!!. denotes the summation over all i,j,k ~ 0 for i + j +k =4.
Due to this particular definition of quartic Bezier surfaces (Figure 1,,), the
C1-continuity between the adjacent surfaces is always satisfied.
1 "
O,
Figure 2. Definition of the B6zier points needed for the definition of one quartic B6zier
surface
2.1. Determination of the closest point on the surface and

calculation of normal and tangential gaps
Once the quartic surface is defmed, the distance vector is defined

between the surface and the slave node n+1xs as:
(2)
The closest point n+1 x{ n+I«, n+I13) is obtained by minimization of equation

(2), using the Newton-Raphson method. Explicit projection of the slave node
on the flat triangle, defmed by b[4,O,O)' b[O,4,O) and b[O,O,4]' is used as the starting
value for the Newton-Raphson method. The procedure described in this sub-
chapter is applied to all other five surfaces. The solution surface, i.e. the
374
closest one, IS the surface that satisfies the following conditions:

o:::;n+la- :::; 1 /\ o:::;n+li3 :::; 1 /\ n+la-+n+li3 :::; 1.
When the closest point is evaluated, the tangents n+1 a", ,,+l ail , and the
normal to the solution point ,,+1 Ii are evaluated from:
Furthermore the contact check, i.e. the detection if the slave node is
inside or outside the master body, is performed. In the case of contact, the
normal gap (penetration) is calculated according to the relation:
n+1 g =II n+1 r(n+1a, n+l~)II.

N (4)
With this relation, the frictionless geometry is completely defmed.

Considering the description of tangential forces (i.e. friction), according to
Agelet de Saracibar (1997), the sliding path definition is referred to the
solution point coordinates and normals. The current and the last converged
solution points are connected using the cubic B6zier curve defined by the
surface normals. The path length s is calculated as the arch length of a B6zier
curve that is defined in the last converged configuration "x=X+"u.
According to this, history variables, i.e. the normal vector and the coordinate
of the solution point, are respectively nn(na-,"j3) and nx(na:~)= x(na:~)
+"u(na;j3). Current values of the surface normal and the solution point are
mapped into the last conver~ed configw:ation, n+lii(n+la;+Ij3) and
"+I~("+la,"+Ii[), ~here n+li(n+l~:+l~)= x(n+la:+l~) +"u("+la;+I~~. The norm~l
n+ln(·+la;+lp) IS calculated m a way that the complete BeZler surface IS
constructed in the last converged configuration. As the solution is already
known, no iterative procedure for fmding the closest point is needed.
Tangents that define the cubic B6zier curve are obtained from the normal
vectors and the vector d (see Figure 3):
d=n+li(n+la, n+I~)-nx(na, n~)

nt 2 =nnxd' I nt =nt 2 xnn=(nnxd)Xnn ' (5)
n+1 t2 =n+1 ii x d, n+1 ~ ="+1 t2 X"+I ii = (n+1 ii x d) x n+ii.
1
The direction of sliding is defined by the vector connecting the previous, "x,
and the current, ,,+l x{'l+la ;+Ij3), solution points, projected into the tangential
375
plane of the current solution, as defmed in equation (61). The position ofthe
second and the third Bezier point is chosen to be third of Ildll multiplied with
the unit tangent vectors. Due to the explicit form and the convex hull
property (for more details see Farin (1993», a cubic Bezier curve is chosen
for the definition of the path. The path length s is evaluated as the arc length
of the Bezier curve from the integral (611).
Figure 3. Geometry of the sliding path
n+le
T ~T
(6)
As the integral in (611) does not have an explicit analytical solution, Gauss
integration is used for the calculation of the arc length s.
The increase in the total tangential slip vector is defmed by equation
(6111). With equations (6) all relations needed for the formulation of the
residual vector and the tangent matrix are defmed. More about the
formulation of residual vector and tangent matrix can be found in Krstulovic
et al. (2000) or Krstulovic (2001).
3. Convergence and error analysis for the 3D frictional

contact elements
A cylindrical contactor (dimensions 2 x 2 x I ) has the curvature radius of

R =3 in the contact surface. It is discretized by tetrahedral elements. These
376
can undergo finite elastic deformation where a neo-Hookean constitutive
equation with Lame constants A = 8.3 .106 and Il = 3.8 .10 6 is used The half-
tube, (dimensions 15 x 08/(6), is discretized using the same elements with
the different Lame constants of A =8.3 .105 and Il =3.8.105 • Frictional
behavior is modeled using the Coulomb law (Il = 0.1). The surface of the
cylindrical contactor is chosen as the slave surface (the half-tube surface
defines the master one). Displacements are applied at the cylindrical
contactor's upper surface. The half-tube is clamped at all edges (cutting
planes). The contactor is pressed into the half-tube (overlapping is J'ZO.063)
and moved from the position in Figure 41 to the one in Figure 4n. Due to the
prescribed displacements, the overlapping differs for various mesh densities.
Therefore, normal and tangential reactions are not equal. The smooth
responses of the contact elements are characterized by the elimination of the
rough responses, typical of simple contact elements. The waviness is caused
by the roughness of the discretized coarse mesh, as well as by the quality of
the description of the continuum when using tetrahedral continuum
elements.
Figure 4. Problem description Figure 5. Increasing the mesh density
To demonstrate the benefit of 3D surface smoothing, simple and smooth

frictional contact elements are compared. It is expected that by increasing
the number of load steps and degrees of freedom, a stage where there is no
difference between the simple and the smooth elements should be reached.
Therefore, the presented example is used for the comparison of the
convergence and error analysis. The analysis is done for the total tangential
reaction at the upper nodes of the contactor where the displacements are
applied. As the contactor is chosen to be the slave surface, the number of
contactor elements is kept constant, while the number of elements for the
master surface (half-tube) is increased as depicted in Figures 4 and 5.
The number of load steps (i.e., prescribed displacements) is also varied.
Loading was performed in: 50, 100,200, 500, and 800 steps. For each step
the total tangential reaction IT is measured. If, for the case of 800 load steps
the mesh density is increased, then Figures 6 show reactions for the meshes
in Figures 4 to 5.
377
---..------.. ·_----------1
0.8 289200F
800 llepl
I O.
I :: Smp_~_1 I
i Smooth frlet_
t 0.3
.. 02
1.2 1.4 1.6 1.8 1.4 1.6 1.8
L. _______ .~...
__fd,IP-I"'--nt)------'--------T-
.... -f-__r--_ _ _ _ ~
...
0.8
17970DOF
1
100llepo
0
•
_ 0.5
1
I
0• - - Smp_fr_nal
.. ..... S..-h frletlOnal
! 03
~ 0.2
"'~'8J
Figure 6. Total tangential reactions for 800 load steps and various mesh densities
For the tangential reaction, the L z norms of the difference between the
reactions of simple (FTRU) and smooth (FSTR) elements are displayed in
Figure 7. The L z norm (Le., the square root error) of the difference in
reactions for simple and smooth elements is calculated from the relation (7,).
From Figure 7 it can be observed that, by increasing the mesh density, the
error of approximately 2.4% corresponds to the mesh in Figure 5". That is a
discretization for which the smooth and the simple approaches give
approximately the same results. The difference in total reaction for various
mesh densities appears to be due to the different overlapping caused by FE
discretization. Therefore, results on the discretization error for systems with
a smaller number of degrees of freedom show the advantage of the smooth
formulation in comparison with the simple one.
From the presented error diagram it can be concluded that the smooth
mesh of 17970 DOF (Figure 5,,) gives the best result. Comparison with other
meshes should show the convergence of the solution when increasing the
number of DOF. For the case of 500 load steps, Figure 8 depicts the Lz norm
of the difference in reactions according to the Lz norm of reaction for the
smooth 17970 DOF mesh. This error is calculated from relation (7,,).
378
Ilfrffilu - frffi7l II L2 Ilfrffilu - f:5771 t ~===_.(7)

Ilfrffilut t('frffilu ')
Ilf)7970DOF t L. (' f)7970DOF ')
~
1=1 1=1
1 03
0.275
0.25
--8----
----<7-~-
--B---
552 OOF
289200F
17970 OOF
l ------
2.5
~~~·---·~~----l,
I
I
I
1";-
r0.2
1=0.175
~0.225
~--------;>,--------"
I
d 0.15
e: 0.125
11f~
I ~ 01
It-0075 ,;r--+------~-----0
1=000:: ~ G-----B------e--~
___
L_~ ~~_~O:_::ioa~~p~OO 600 700 800
Figure 7. L2 nonn of difference in reactions Figure 8. L2 nonn of difference in reac-

tions according to 17970 DOF smooth mesh
References
Agelet de Saracibar, C. (1997) A New Frictional Time Integration Algorithm for Large Slip
Multi-Body Frictional Contact Problems, Computer Methods In Applied Mechanics And
Engineering, 142, 303-334.
Bohm, W., Farin, G., Kahmann, 1. (1984) A survey of curve and surface methods in CAGD,
Computer Aided Geometric Design, 1, 1-60.
Farin, G. (1993) Curves and Surfaces for Computer Aided Geometric Design. A Practical
Guide, third edition, Academic Press, Boston.
Krstulovic-Opara, L., Wriggers, P., and KoreJc, 1. (2000) Symbolically generated 3D smooth
polynomial frictional contact element based on the quartic B6zier surfaces, In Proceedings
of ECCOMAS 2000, Barcelona, Spain.
Krstulovi6-Opara, L. (2001) A C1-continuous formulation for finite deformation contact,
Ph.D. Thesis, FOIlI, IBNM, Hannover.
Wriggers, P. (1987) On Consistent Tangent Matrices for Frictional Problems, In Proceedings
ofNUMETA 87 Conference, Eds. 1. Middleton, G. N. Pande, Nijhoff, Drobrecht.
Wriggers, P., Krstulovi6-0para, L., and KoreJc, 1. (1999) Development of 2D Smooth
Polynomial Frictional Contact Element based on a Symbolic Approach, In Proceedings of
ECCM, Wunderlich (ed.), Miinchen.
MULTICONTACT STRUCTURES AND
PARALLEL COMPUTING
P. Alart
Laboratoire de Mecanique et Genie Civil UMR 5508, CNRS-UniversiU Montpellier 2
Pl. E. Bataillon, 34 095 Montpellier Cedex 5, France
M. Barboteu
Laboratoire de Theorie des Systemes, UniversiU de Perpignan
52, avenue de Villeneuve, 66860 Perpignan Cedex, France
Abstract In this paper we present a numerical strategy to solve large scale fric-
tional contact problems by using balancing domain decomposition meth-
ods which are well adapted to parallel computing (scalability property).
The motivation is given by the study of the mechanical behaviour of
three multicontact structures (rolling shutters, collections of deformable
grains and wood micro cellular media).
1. Introduction
The aim of this paper is to present an efficient numerical scheme for
a multibody system with frictional contact between the bodies or sub-
structures, we call these multicontact structures. We focus our attention
on three applications:
- rolling shutters composed of many slats jointed by a hinge with
clearance and eventually rotative friction;
- collections of deformable grains with contact interfaces between
them;
- wood microstructure composed of a cell wall network with self
contact between the different walls of the cells.
To handle these multicontact problems it is necessary to develop rigorous
modelling and new numerical tools adapted to the strong non-linearity
and non symmetry due to the large proportion of degrees of freedom
related to the contact conditions. To overcome these difficulties, we pro-
pose in Section 3 a domain decomposition method generalized to non
symmetric systems. The balancing preconditioners have to be improved
379
@ 2002 Kluwer Academic Publishers.
380
and some aspects are described in section 3.1. In Section 4, we present

multicontact applications and discuss the numerical performance of the
strategy.
2. Hybrid formulation and Generalized Newton

Method (G.N.M.)
Following the augmented Lagrangian approach presented in Alart and
Curnier (1991), the equilibrium of a discretized contact bodies system is
governed by the system of non-linear equations:
{
Fint - Fext+ F(u, A) = 0 (1 )
-~(A - F(u, A)) = 0
We let u denote the kinematic variables (displacements or rotations) and
A the static variables (contact forces or torques). Fint and Fext denote,
respectively, the internal and the external discretized forces. F(u, A)
defines the discretized contact operator, and r is the penalty coefficient.
We have chosen to treat both variables u and A simultaneously through
Newton's method. The system of equations is split into two parts involv-
ing the pair x = (u, A): a differentiable part G and a non differentiable
frictional contact F,
G(x) + F(x) = O. (2)
To overcome the non differentiability in equation (2), Newton's method
may be extended to the following iterative form (G.N .M.):
xi+! = xi _ (Ki + A~) -1 (G(xi) + F(x i )) , (3)
Ki = 8G(xi), A~ E 8F(x i ).
The matrix Ki is the usual tangent matrix and A~ represents the gener-
alized Jacobian of F at xi. Usually, xi is in a region of linearity, and A~
reduces to a single classical Jacobian matrix. Each region of differentia-
bility of the operator corresponds to a status of each contact node of the
discretization. The G.N .M. leads us to solve at each iteration i a non
symmetric linear system. The non symmetry of the matrix A;:" is due to
the friction terms. The contact interface is discretized by contact finite
elements which yield elementary non symmetric tangent matrices when
the friction appears (slip status). To solve the non symmetric linearized
systems, we choose the GMRes method as an iterative non symmetric
domain decomposition method (Schur complement form).
3. Domain decomposition strategies

The basic idea in nonoverlapping domain decomposition methods (see
Le Tallec (1994) for more details) is to split the domain n into N small
381
non overlapping subdomains nn(n = 1, N) and interfaces rn(n = 1, N).

Substructuring techniques consist in reducing the original global system
to an interface problem by a block Gaussian elimination of the internal
degrees of freedom, and then iteratively solving the resulting variational
interface problem:
:Ju E V / < Su, v >=< f, v> "iv E V = Tr H(n)lr. (4)

N
The matrices S = L RiSi(Ri)t and Si denote, respectively, the global
i=l
Schur complement matrix (defined on r) and the local Schur comple-
ment matrices (defined on ri by Si = Ri - (Bi)t(1{i)-lBi). Above,
(Ri)t is the restriction operator from r to r i,and Ki = (Ki

(Bi)t
Bi)
:Ri
denotes the subdomain stiffness matrix, the first block corresponds to the
internal degrees of freedom Xi, the second corresponds to the interface
degrees Xi . The interface problem (4) can be solved by a preconditioned
conjugate gradient method (symmetric cases) or GMRes method (non
symmetric cases). Hereafter, we use the multilevel Neumann-Neumann
preconditioner M. This iterative technique does not require the explicit
calculation of the matrix S. We just have to form the matrix vector
products Sp and M-1r by solving independent auxiliary Dirichlet and
Neumann problems on the local subdomains and a global coarse problem
defined on a space of singular (rigid body) motions.
3.1. N on symmetric preconditioner

In the framework of Domain Decomposition methods applied to sym-
metric problems, the two level Neumann-Neumann preconditioner solves
the interface problem using an additive Schwarz algorithm. This method
can be generalized to non symmetric problems by defining an ap~ropri
ate coarse space. It consists in developing the interface space V into
coarse and fine components Va and Vf, respectively. These spaces must
be constructed so as to dissociate the global coarse problem from the
local fine problems, and their elements must satisfy the following or-
thogonality conditions:
"iva EVa· (5)
Here, the space Va contains the elements of the kernel of S; (and not
Si as for the symmetric case !). The preconditioning step consists in
multiplying the preconditioning matrix M- 1 by the residual vector r.
This product, ¢, can be decomposed into the sum of ¢f E Vf = L~l ~.l
382
and 4>'G EVa:

N
M-lr = 4> E V = Va + Vf where 4> = 4>'G + 4>f = 4>'G + L 4>t· (6)
i=l
The terms 4>t and 4>'G are given by the solutions of local Neumann prob-
lems and a global coarse problem which take the following variational
forms:
for i=l, .. ,N 4>t E l/i.L

i.L -.L
< S i'+'i,
/ Vi >=<
D ir,vi
t - -.L
>
4>'G E Va / < S4>'G, v'G >=< r, v'G >
By construction, the 2-level non symmetric Neumann-Neumann precon-
ditioner takes the classical form:
N
M-lr = 4>'G + L(I - Pc)DiSilD;(I - Pc)tr. (9)
i=l
where the matrices Si l denote the pseudo inverse of the matrices Si

obtained by elimination of null pivots connected to standard rigid modes.
The numerical results presented below validate this new construction of
a Neumann-Neumann preconditioner. In non symmetric situations such
as multicontact with friction, it produces optimal scalable results in all
cases.
3.2. Substructuring strategies

One feature of this non-linear non symmetric domain decomposition
strategy consists in putting the numerical subdomain interfaces away
from the physical contact interfaces, see Barboteu et al. (2001). Con-
trary to current approaches, we therefore suggest treating the physical
contact interfaces as internal surfaces: the contact interfaces (hinges
for shutters and contact area for deformable grains) must be inside the
subdomains and not constitute decomposition interfaces. Thus, the de-
composition is not forced to respect the geometry of its components;
such a strategy is shown in Figure 3. This choice allows a better balance
of the size of the subdomains and leads to an optimal decomposition for
parallel efficiency.
4. M ulticontact applications
The efficiency of this domain decomposition strategy will be assessed
in three examples of multicontact structures. Some characteristics of
these examples are summarized in the following table.
383
Multicontact Rolling Deformable Cellular

problem shutters grains medium
body(ies) plates 2D grains 2D cells
linear modelling linear elasticity nonlinear elasticity
bi-unilateral contact frictional contact frictional contact
interface(s) rotative friction moderate slips => large slips =>
small rotative slips bi-facet elements multi-facet elements
typical non connected divided connected
features medium medium medium
specific rigid motions many rigid motions nonlinear coupling
4.1. Rolling shutters

Rolling shutters for shops, stores and hangars are formed by a suc-
cession of slats (elastic plates in flexion and torsion) jointed by hinges
(numerous contact zones). The edges of the slats are designed so that
the slats fit into each other. To facilitate the rolling of the shutters on
opening, the profile of the slat requires a clearance in the hinge. Our
specific model takes into account the clearance (-g,g) in the hinge and
the friction according to the relative rotation of the slats, see Barboteu
et al. (2001). The performance analysis is summarized in the Figure
"floating" subdomain
with a hinge (contact interface)
~~e+bodYI
body 2
Figure 1. Rolling shutters, one subdomain and a hinge contact element.
2. The first result described gives the evolution of average number of

GMRes iterations for different values of the friction coefficient varying
from 0 to 2 for a rolling shutter with 16 slats and 30 subdomains (26
floating subdomains). We observe that the new non symmetric pre-
conditioner (curve 0) makes the interface solver insensitive to the non
symmetry. Moreover, we analyse the scalability properties of the differ-
ent Neumann-Neumann preconditioners. Without friction (curve *) the
standard 2-level preconditioner gives a scalable method, i.e., it is inde-
pendent of the number of subdomains. But with friction the standard
384
proced ure (curve 6.) leads to a high increase of the number of iterations,
even higher than without coarse solver (curve with squares). A specific
preconditionner (curve 0) based on mechanical considerations, see Bar-
boteu et al. (2001), improves the behaviour of the algorithm. But only
the non symmetric preconditioner (curve 0) leads to a full recovery of
the numerical scalability properties, see Alart et al. (2001).
StandanI 2- level eumann-Neumann <11 · 0)

Standard 2- level Neumann-Neumann
250 , ., Specific 2-1e.eI Neumann-Neumann
F
~
~
[ Non symmetric 2- le.el Newnann- Neumann
I- level Newnann-Neumann
I ""f)~ •~ 1::
ifau ~ 200
~ ~~. °1 00
50'-:'-:-~-""""''''''''''''--'- o~~~~~~~~
0.0 0.5 1.0 1.5 2.0 10 20 30 40 50 60 70
friclion coefficient Number of subdomains (II = 0.2)
Figure 2. Performance of the "Newton-Schur" domain decomposition algorithm.
4.2. Collection of deformable grains

Our motivation here is to study the behaviour of a collection of de-
formable grains under shear or compression in order to compare with
the simulations of a collection of rigid grains as usually done in granular
media modelling, see Moreau and Jean (1996). From a numerical point
• elastic node
one subdomain
Collection of defonnable grains
Figure 3. Deformable grains, one subdomain and a bi-facet contact element.
of view, each su bdomain contains many grains or parts of grains accord-

ing to our su bstructuring strategy (Section 3.2). Consequently a large
number of rigid motions have to be taken into account in the two-level
385
preconditioner; the convergence behaviour is then similar to the previous

example.
4.3. Wood cellular microstructure

Some materials, such as wood or foam, have a microstructure com-
posed of a cell network. The densification process consists of the buckling
cellular media one subdomain multi-facet contact element
Figure 4. Cellular media, one subdomain and a multi-facet contact element.
of the walls and self contact inside the cells between the walls. The simu-
lation of such a process requires accounting for the large deformations of
the walls and eventual large slips. The connectivity of the medium leads
to identifying a subdomain as a cell, with half walls, or a set of cells.
In the first approach the microstructure is fully periodic. Using only a
Figure 5. Deformed configuration of the cellular media composed by 39 cells.
sample of 39 cells under compression, the first results, cf Figure 5, show

different steps: a first micro-buckling mode of 5 cells without contact, a
second buckling mode of the whole sample with frictional contact in 8
cells and finally a shear band where the contact is concentrated. These
steps are recovered on the force-displacement graph in the figure 6.
5. Conclusions
The numerical tool developed above may be used for cellular media,
such as wood, in order to understand the densification process of wood,
to determine the region with micro-buckling in an indentation test of the
386
0.8
'"....
Q)
<) 0.6
<E
s:: 0.4
.2
t)
..s
0
..... 0.2 linear behaviour
0.0
0 5 10 15 20 25 30 35
compression ratio
Figure 6. Graph of reaction forces - compression ratio.
wood, and to simulate the propagation of shear bands in dynamic impact

of metallic foams. However, the simulation of collections of rigid bodies
is usually treated with the Contact Dynamics method, see Moreau and
Jean (1996). The parallel version of such an algorithm has to be based
on tools others than those used for the Generalized Newton Method as
presented above because the solver of the Contact Dynamics is not a
linear iterative method but may be viewed as a non-linear Gauss Seidel
algorithm , see Jourdan et al. (1998).
Acknowledgment: This work was supported in part by CINES, Cen-

tre Informatique National de l'Enseignement Superieur, Montpellier,
Project mgc2547.
References
Alart P. and Curnier A. (1991) A mixed formulation for frictional contact problems
prone to Newton like solution methods, Comp o Meth . Appl. Mech . Eng, 92, 353-
375.
Alart P. , Barboteu M., Le Tallec P. and Vidrascu M. (2001) Additive Schwarz method
for non symmetric problems : application to frictional multicontact problems, to
appear in the proceeding 's of the 13 t h International Conference on Domain De-
composition Method (plenary lecture).
Barboteu M., Alart P. and Vidrascu M. (2001) A domain decomposition strategy
for non classical frictional multicontact problems, Compo Meth. Appl. Mech. Eng,
190/37-38, 4785- 4803.
Jourdan F., Alart P. and Jean M. (1998) A Gauss-Seidel-like algorithm to solve fric-
tional contact problems, Compo Meth . Appl. Mech. Eng, 155, 31-47.
Le Tallec P. (1994) Domain decomposition methods in computational mechanics,
Compo Mech. Adv., 1, 21-220.
Moreau M. and Jean M. (1996) Numerical treatment of contact and friction: the
contact dynamics method, Eng. Sys. Des. Anal., 76, 201-207.
A NEW APPROACH TO THE
2D TRANSIENT ROLLING
CONTACT PROBLEM
J. A. Gonzalez and R. Abascal

Escuela Superior de Ingenieros
University of Seville, Spain
Abstract This work presents a new approach to the transient rolling contact of
two-dimensional elastic bodies. Solutions will be obtained by mini-
mizing a general B-differentiable function representing the equilibrium
equations and the contact conditions at each time step. Inertial effects
are not taken into account and the Boundary Element Method is used to
compute the elastic influence coefficients of the surface points involved
in contact.
1. Introduction
The mechanical study of rolling contact problems started in 1875 with
Reynold's investigation of the creep phenomena, observed when a rubber
cylinder rolling on a steel plate moves forward more than its undeformed
rolled perimeter. He attributed this effect to the elastic deformations in
the contact area and introduced the idea that the contact region would
be composed of stick and slip zones. It was about fifty years later in 1926,
when Carter published his work presenting the steady-state analytical
solution for the two dimensional case of similar material cylinders, and
the non-linear relation between longitudinal creep force and creepage.
Historically, the first approximation to the transient rolling contact
problem was reported by Kalker in 1970 for the particular no-slip case;
the formulation was extended in 1971 to the complete adhesion-slip case
using an original minimum principle. In both works, a constant Hertz
normal pressure distribution over time was assumed, and uncoupled
normal-tangential problems were solved.
387
388
2. The transient rolling contact problem

Transient rolling contact of two cylinders with parallel axes is the
scope of this work. The study is carried out from a two-dimensional
point of view, assimilating the problem to plane strain.
To start time integration and simulate transient effects, some infor-
mation has to be given: initial conditions and load history. Load history
will be provided in the form of normal and tangential net loads versus
time.
It is assumed that the friction between the cylinders obeys the Coulomb
law. Adopting this approach, each pair of points belonging to each of
the cylinders in rolling A and B (see Figure 1) can experience any of the
following contact states: separation, adhesion or slip.
Driven
Roller 2a SOLIDB
Figure 1. Rolling cylinders. Figure 2. BEM mesh.
2.1 Contact conditions

The mathematical expressions of the contact conditions, using Kalk-
er's notation (Kalker 1990), will be classified in two groups, normal (1)
and tangential (2) :
Pn ~ 0 (1)
Pt E CJ.LP;; ; St(Pt - pt) ~ 0 ; \:fp; E CJ.LP;; (2)

where C9 is a closed interval of radius g.
2.2 Problem description

Normal and tangential tractions will depend on the Eulerian coordi-
nate x (relative to a fixed axis situated in the contact area) which is
389
used to allocate each pair of points relative to a rigid body position of
the cylinders at certain time r, i.e. Pn(x, r) and Pt(x, r).
The cylinders are rolling under the action of external forces P( r) and
Q(r). If these external forces remain constant in time, the system will
tend to a steady rolling state.
Body separation in normal direction, 8n , will be
(3)
where 8no (x, r) = 8g (x)-80(r) is the initial distance between the cylinder
contact points, 8g being the geometric initial separation between two
contact points located at x, and 80 the sum of two approaching rigid
body displacements externally imposed on each cylinder, producing an
overlap between them.
Using a Eulerian description of the particles moving through the con-
tact area, the relative tangential slip velocity between two points of the
cylinders A and B, is defined as:
(4)
where V(.) jVr is the Total time derivative, sometimes called Substantial
or Material derivative (following a particle of fixed identity relative to
fixed axes) and 8t represents the tangential separation, given by
8t (x, r) = {xA(r) - xB(r)} + {uf(x, r) + uf(x, r)} . (5)
We assume quasi-flat contact surfaces, and that the mean rolling ve-
locity V(r) governs the rotation of the two cylinders. Under these hy-
potheses the individual velocity Va(r) of each cylinder can be substi-
tuted in the equations by the mean velocity V (r).
The slip velocity is obtained from equations (5) and (4), based on the
Total time derivative
(6)
and presents the advantage that the formulation for each time step only
involves tangential displacements at the current and the previous time
step and the drawback that we have to interpolate them between nodes
with the shape functions, if tlTJ = V(r)tlr is not equal to the nodal
separation.
3. Problem formulation
The step to step time integration scheme is adopted, forcing the ful-
fillment of contact constraints at all nodes for every discrete time step.
390
The suggested algorithm is based on Mathematical Programming tools
and solves the problem in a combined manner, without separating it into
normal and tangential parts, and with a fully automated iteration con-
trol.
3.1. The elastic equations

A geometrical discretisation process and subsequent application of
the BEM static formulation (see Abascal 2000), provide us with each
cylinder's elastostatic equations at a specified traversed distance TJk after
a discretisation of the traversed distance parameter on k constant steps
(TJk = TJk-l + ~TJ):
10k --
un fkn + SnnPnk + SntPtk ; k = l...nts (7)
where nts is the total number of steps, and p~,t = p~,{ (TJk) are the
surface nodal tractions for discrete k-th step.
Similarly, the tangential displacement difference can be obtained in
condensed form as
u tk = gtk + StnP nk + S ttPtk (8)
After the resolution of the elastic problem using the BEM and later
condensation, we have obtained the linear relation between gap-tangential
displacements vector (o~, u~) and normal-tangential tractions vector in
the contact area (p~, p~), expressed as
( ~~ ) = ( !1 )+ [~:: ~::] ( ~~ ) (9)
The equilibrium equation (9) together with the contact conditions ex-
pressed by equations (1) and (2), for all contact point pairs, completes
the mathematical description of the problem. To calculate the slip ve-
locities at the contact nodes, the Material formulation based on (6) is
adopted.
3.2. Material Formulation

Tangential relative displacements are discretised at the contact area
with continuous quadratic elements in their usual form,
ne
Ut(x, T) c:::: L [Nj(x) Utj(T)] (10)
j=l
where ne is the number of nodes per element, N j the element shape

function associated to node j (located at coordinate Xj), and Utj(T) is
391
the nodal tangential displacement at instant T. These shape functions

N j are the same for each couple of elements in the contact area because
their nodes are located at the same x-coordinate.
When approximation (6) of St is used, we need the tangential relative
displacement of the pair of particles now located at Xi coordinate, when
and where they were tlT time ago. This information can be organised
in a vector named Ut.
If equations are organised in vector form, we can write the slip velocity
for a step 17k as
(11)
Introducing (8) into the slip velocity main expression (11) and con-
sidering (9), we finally obtain the grouped equation for the Material
approach
1 [_<lk-l
k t:k
St = '" + tl17 'U + gtk + S tnPnk + S ttPtk] (12)
with iJk-l = -R (g:-l 1)

+ Stnp~-l + Sttp:- only depending on known
information about the previous time step and the time step size.
3.3. The contact constraints

In this work, we formulate the contact conditions in a quasi-Lagrangian
way, reducing the solution of the problem to finding the zero of a vecto-
rial function.
The sum of normal and tangential loads (P(17), Q(17)) applied on the
discretised contact surface can be calculated by integration over the con-
tact area and expressed in vector form as fT p~ t and easily included in
the general formulation. '
Two new equations having been formulated, it is necessary to consider
as unknowns two other variables which vary with distance: the creep
~0(17), and the overlap 150(17). The initial normal separation and the creep
e
vector can be expressed as OnO = Og - JO(17)e and =~o(17)e~ where e is
the identity vector.
Now the vector of unknowns at the time step k, zk, is defined as
zk = (o~,s~,p~,p~,JO(17k),~o(17k)) and the coupled functions describing
the problem using the Material formulation can be written as
392
= 0,
(13)
which is a B-differentiable function, because of the non F-differentiable
contact conditions, that involve non-linear directional derivatives.
To solve the problem the Generalised Newton's Method with line
search (GNMIs) has been used. GNMls is an extension of Newton's
Method for B-differentiable functions formulated by Pang in 1990.
4. Results
The example problems presented are for cylinders with the same ma-
terial properties. The BE mesh is represented in Figure 2. The first
transient rolling contact problem consists of two rotating cylinders of
identical material. This problem does not present normal-tangential
coupling under the half-space hypothesis, and Hertz normal pressure
solution is exact for normal tractions.
The friction coefficient between surfaces is J1 = 0.1 and two cases
are considered, both starting from a static contact situation but with
different applied forces.
4.1. Constant external loads

Let us suppose first that the two cylinders are initially braked with
applied external loads Q / J1P = 0.75. Initial tractions of this problem
are known analytically: the Hertz distribution for normal tractions, and
the Cattaneo distribution for tangential tractions. When the brake is
released, transient phenomena start, and end with another analytical
solution obtained by Carter for similar cylinders in steady state rolling
motion.
In Figure 3 the BEM tangential tractions are represented and com-
pared with the solution obtained by Kalker for different travelled dis-
tances. The normalisation factors are: J1IPnHI, the friction coefficient
times the absolute value of the maximum normal traction for the equiv-
alent Hertz problem; and bH = 2aH , the length of the contact area.
Agreement between the two results is very good but with some small
differences.
393
-CAnANEQ -KALJ(ER -KAU<ER

• OEM'-- ...... • BEMTMCl '" • BEMTIMt
.....~ HEMSIIp ..•.•~ BEMSilp
i"
~
a.- 0.4
Figure 3. From Cattaneo to Carter, Q/J-tP = 0.75. Similar cylinders .
...... ..... T_
..... -O-Slip
Figure 4.Ramp load Q/J-tP from 0 to 0.75. Similar cylinders.

394
4.2 Varying external loads
The previous example is now revisited but applying different loads;
let us suppose we start with the contacting cylinders not clamped but
stopped with net tangential force equal to zero. The load Q/ J-lP = 0 is
then linearly increased until it reaches 0.75 when the distance travelled
is rJ/aH = 8/3, keeping the normal load P constant. Also a friction
coefficient of 0.1, time step !:lrJ = Le /2 and the same discretisation of
the previous example, are used.
The tangential tractions obtained are represented in Figure 4 together
with the dimensionless relative slip velocity. The first curve corresponds
to the initial time step (rJ/aH = 1/30). Logically, the final steady state
coincides with that obtained in the previous example because the final
applied forces are the same.
5. Conclusions
The transient two-dimensional rolling contact problem between cylin-
ders can be solved using a new numerical technology based on the use
of Boundary Element Method and Mathematical Programming Tech-
niques.
In this new technology BEM has been used to compute the constant
influence coefficients based on true cylinder geometry, not on the half-
space approximation, while MPT was used to model the contact condi-
tions and to solve the highly non-linear problem associated with tran-
sient rolling at each time step. The algorithm and proposed formulations
prove to be very strong and efficient for 2D, similar transient problems.
References
F. W. Carter. On the action of a locomotive driving wheel. Proc. Roy. Soc. Ser. A.,
112:151157, 1926.
J. A. Gonzalez and R. Abascal. An algorithm to solve coupled 2d rolling contact
problems. Int. J. Num. Meth. Engng, 49:1143-1167, 2000.
J. J. Kalker. Transient phenomena in two elastic cylinders rolling over each other with
dry friction. Jour. Appl. Mech., 37:677-688, 1970.
J. J. Kalker. A minimun principle for the law of dry friction, with applications to
elastic cylinders in rolling contact. part 2: Application to nonsteadily rolling elastic
cylinders. Jour. Appl. Mech., 38:881-887, 1971.
J. J. Kalker. Three-dimensional elastic bodies in rolling contact. Kluwer Academic
Press, Dordrecht, 1990.
J. S. Pang. Newton's method for b-differentiable equations. Math. Oper. Research,
15(2):311-341, 1990.
O. Reynolds. On rolling friction. Philosophical Transactions of the Royal Society of
London, 1:155 174,1876.
NUMERICAL APPROXIMATION OF THE
ELASTIC-VISCOPLASTIC CONTACT
PROBLEM USING NONCOINCIDING
FINITE ELEMENT MESHES
Jose R. Fernandez-Garda, Juan M. Viano

Departamento de Matematica Aplicada. Facultade de Matematicas.
University of Santiago de Compostela. 15706 Santiago de Compostela, Spain
Patrick Hild
Laboratoire de Mathimatiques. University of Savoie
73376 Le Bourget du Lac, France
Abstract We are interested in the unilateral contact problem of elastic-viscoplastic bo-

dies in a bidimensional context. The corresponding time dependent variational
inequality is approximated in the space variable with nonconforming finite ele-
ment methods, which allow the handling of nonmatching meshes on the contact
zone. Error estimates are established and the corresponding numerical experi-
ments are performed.
1. Introduction
The object of this paper is to study the convergence of finite element me-
thods for elastic-viscoplastic models submitted to frictionless unilateral con-
tact conditions. Considering the contact problem between two bodies, we are
particularly ~nterested in finite element approximations involving nonmatching
meshes on the contact part in which the nodes inherited from the triangulations
of the bodies may not coincide.
By adapting and extending the techniques developed for the linear elastic
case with nonmatching meshes in Ben Belgacem-Hild-Laborde (1999) and for
the elastic-viscoplastic case with matching meshes in Han and Sofonea (2000),
we are able to study the convergence of different finite element approaches
handling nonmatching meshes: a first method where the discrete contact con-
ditions are of local type and another one where global contact conditions hold.
395
l.A. C. Martins and M.D.P. Monteiro Marques (eds.), Contact Mechanics, 395-402.
396
2. Physical setting and variational formulation

Let us consider two elastic-viscoplastic bodies which occupy two bounded
domains 0 1 and 0 2 of the bidimensional space and let [0, T], T > 0, denote
the time interval in which the evolution of the bodies will be investigated.
For £ = 1, 2, the boundary ri of Oi is assumed to be "smooth" and is the
union of three nonoverlapping portions rb, r},r and rb with meas(rb) > 0.
Assume that both bodies are submitted to prescribed displacements (supposed
equal to zero for the sake of simplicity) on rb x (0, T). The unit outward
normal vector on ri is denoted vi = (lIf, II~). Both bodies are subjected to
densities of volume forces fi(t) = (f[(t), fi(t)) on Oi x (0, T) and densities
of surface forces gi (t) = (gf (t), g~ (t)) are assumed on the boundary part
r},r x (0, T). At the initial time the bodies are in contact on their common
boundary part rh = rb which we denote by r c, and frictionless unilateral
contact conditions without gap are assumed on rc x (0, T). We also suppose
that the contact process is quasistatic.
Let the displacement field u = (u\ u 2) (where the notation u i stands for
ulnl) with u i = (un, 1 ~ i ~ 2, and the stress fields u = (u\ ( 2 ) (where
the notation u i stands for ulnd with u i = (u~j)' 1 ~ i,j ~ 2, induced in the
bodies 0 1 and 0 2 verify the initial conditions ui(O) = u~ and ui(O) = u~.
In order to give the variational formulation associated with this problem, we
introduce the following variational spaces for £ = 1, 2:
V(Oi) = {vi = (v~)

~ °
., v~~ E H1(Oi) , v~z = on riD, 1 -< i <
- 2} ,
Q(Oi) = {Ti = (T~); T~ E L2(Oi), 1 ~ i,j ~ 2},
Q1(Oi) = {Ti E Q(Oi); Div T E [L2(Oi)]2}.
The Hilbert spaces V(Oi), Q(Oi) and Q1 (Oi) are endowed with their canoni-
cal inner products denoted (., ·)V(nl), (., ·)Q(nl) and (., ·)Ql(nl). Setting V =
V(01) x V(02), Q = Q(01) x Q(02) and Q1 = Q1(01) x Q1(02), these
product spaces are equipped with the inner products (., ·)v, (., .)Q and (., ·)Ql
and the associated norms are denoted I· Iv , I·IQ and 1·IQl respectively. Let 82
be the space of second order symmetric tensors on ~2 and let· and I . I denote
the inner product and the norm on this space.
Next, we specify the properties satisfied by the functions t;i and C i (£ =
1, 2) which will be incorporated in the viscoplastic constitutive law.
The function t;i : Oi x 82 -+ 82 is a bounded symmetric positive definite
fourth order tensor
{
t;fjkl E Loo(Oi), 1 ~ i,j, k, 1 ~ 2,
£iui . Ti = ui· £iTi vui, Ti E 8 2 ,
°
(1)
3 a i > s.t. t;iTi . Ti ~ a i lTil2 VT i E 82 .
397
The function G" : O" X 8 2 X 8 2 -+ 8 2 satisfies

°
:3 £l > s.t. IGl(:z:, O"l, E"l - Gl(:z:, u l , i")1
{ ~ £l(IO"l _ u"l + le" - i"l) T/O"l, u l , el , i" E 8 2 , a.e. in Ol,
:z: J--t Gl(:z:,ui,e l ) is measurable for any ui,e l E 8 2 ,
Gl(:z:, 0, 0) E Q(Ol) a.e. :z: E O".
(2)
For the sake of simplicity, we shall adopt the notations
l _ 1 (8vf 8v1)
e(v ) - 2" 8xj + 8Xi '
e(v) = (e(v 1),e(v 2)) T/v = (V 1,V2) E V,
ce = (c 1e 1 , c 2e2) T/E: = (e 1, e2) E Q,
G(O",e) = (G 1(0"1,e 1),G2(0"2,e 2)) T/e, 0" E Q.
For all t E [0, T], let F(t) represent the element of Vi given by
2 2
(F(t),v) = L:(f"(t),v")[L2(nt )]2 + L:(gl(t),vl)[L2(r~)J2 T/v E V.
~1 ~1
Suppose that
I" E W 1,OO(0, Tj [£2(0")]2), y! E W 1,OO(0, Tj [£2(r1v )]2)j (3)
then F E W 1,OO(0, Tj V').

Thus, the convex of admissible displacements is defined as follows
U={v=(vl,v 2)EVj [v.v]:=V1.li+v2.v2~0 on rc}. (4)
Assume that
1.£0 = (uA, u~) E U, (0"0, e(uo))Q = (F(O),uo). (5)
The variational formulation of the elastic-viscoplastic unilateral contact pro-
blem (see Rochdi and Sofonea (1997» consists then of finding the displace-
ment fields 1.£ : [0, T] -+ U and the stress fields 0" : [0, T] -+ Q1 satisfying:
1.£(0) = 1.£0, 0"(0) = 0"0,
u(t)=ce(u(t))+G(O"(t),e(u(t))) a.e. tE(O,T), (6)
(O"(t),e(v) - e(u(t)))Q ~ (F(t),v- u(t)) T/v E U, t E [O,T].
The existence and uniqueness statement for this problem has been proved
by Rochdi and Sofonea (1997). We recall this result in the following theorem.
Theorem 1 Let the assumptions (1), (2), (3) and (5) hold. Then there exists a
unique solution ojproblem (6) such that
(1.£,0") E W1,OO(0,TjU x Q1).
398
3. The finite element approximations

We suppose that each subdomain Oi, l = 1,2, is a polygon and, with each
subdomain ot, we then associate a regular family of discretizations made r:
of triangular elements denoted r;, such that ot = U
K. The discretization
K.Ert
parameter hi on ot is given by ht = max hT where hT denotes the diameter
TErt
of the triangle T. We suppose that the end points Cl and C2 of the contact zone
re are common nodes of the triangulations r",1 and ~ and that the monodi-
mensional traces of triangulations of 1jf and ~ on r e denoted Ok and O~ are
uniformly regular. For any integer q ~ 1, the notation Pq(T) represents the
space of polynomials with global degree ~ q on T.
The following finite element spaces are chosen for the approximation of
Q(ot) and V(ot).
Qh(Oi) = {T~ E [L2(Oi)]~X2, T~IT E [Po(T)]~X2 'tiT Elf},
Vh(Ot) = {v~ E [C(Ot)]2, V~IT E [Pl(T)]2, 'tiT E If vilrlD = O},
Qh = Qh(OI) x Qh(02), Vh = Vh(OI) x Vh(02).
It is straightforward thate(Vh) C Qh. Moreover, we assume thatG(Qh, Qh) C
Qh and ee(Vh) c Qh.
To translate in the finite element context the nonpenetration conditions con-
tained in (4), let us define
wt(re) = {¢h : re
-+ lR; ¢hlT E PdT) 'tiT E O~,
¢h continuous on each straight line segment of r e },
Mt(re) = {.,ph E wt(re); .,phiT E P1 (T), .,phi T E PO(T) if Cl or C2 E T}.
This allows us to introduce the following operators 1r~ (see Ben Belgacem
(1999» andI~ (piecewise interpolation) on wt(re), defined for any function
r r
¢ : e -+ lR, piecewise continuous on e, by:
1ri¢ E wt(re), (~¢)(Ci) = ¢(Ci) for i = 1, 2,
{ (¢ -1r~¢).,ph ar =0 'tI.,ph E Mt(re).
Jre
I~¢ E wt(re), (I~¢)IT(yi) = ¢IT(yi), i = 1,2, 'tiT E~,
where yr, Y2 are the nodes of T E of.
The next step consists of defining the discrete convex cone (approximating
U) involving these operators:
Ui = {Vh = (vk, v~) E Vh; vk .1.1 1 + 1rHv~ . 1.12) ~ 0on re},

Uh = {Vh = (vk, v~) E Vh; vk . 1.11 +IHv~ . 1.1 2) ~ 0 on fe}.
399
Note that both approximations are different and nonconforming so that gene-
rally U'h i= ul, U'h rt rt
U and Ul u. This means that (slight) interpenetra-
tions are allowed in these approximations.
Now, we consider a fully discrete approximation of problem (6). Given a
partition of the time interval [0, T]: 0= to < tl < ... < tN = T, we denote
the step size by k n = tn - tn-I, for n = 1,2, ... , N. Let k = maxn k n be
the time discretization parameter. For a sequence {w n };;=o, we define 6w n =
(w n - wn- 1)jkn. Finally, we denote Fn = F(t n ), un = u(tn),u1,n =
u1(tn), u 2,n = u 2(t n ), un = u(tn ), u~ = uv(tn) for n = 0, 1, ... ,N.
The discretized problem issued from (6) uses the above-mentioned noncon-
forming finite element approaches and a backward Euler scheme. It consists
then of finding the displacement fields Uhk = {Uhk};;=O c Uh and the stress
fields Uhk = {uhd;;=o c Qh such that ug k E Uh, ugk E Qh and for
n = 1,2, ... ,N,
{ 6U hk = E6g(uhk) + G(uhk' g(u hk )),

(u~k' g(Vh - u~k))Q ~ (Fn, Vh - u~k) VVh E Uh,
(7)
Using fixed point arguments, the existence and uniqueness of solution to pro-
blem (7) is proved for a time parameter k small enough. In order to obtain
error estimates for the discrete solutions, we first establish the following result.
Proposition 1 (Fernandez Garda-Hild-Viano (2001)) Let (1.1., u) be the so-

lution of (6). Suppose that 1.1. is such that 1.1. 1 E LOO(O, T; [H2(Ol W) and
1.1.2 E LOO(O, T; [H2(02)]2). Let (Uhk' Uhk) be the solution of(7). Then
max {Iu n -
l::;n::;N
u~klQ + lu n - u~klv} :::; C(luo - ugklQ + 11.1.0 - u~klv
+ max {
l::;n::;N
inf (IVh - unlv
'lJhEUh
+\(
Jre
U~[(Vh - un). V]dI'\t)
+ inf I (
Jre u~[(V - Uhk)' V]dI'/t
t
'lJEU
+1 G(u(s),e(u(s))) ds - ~k;G(ui,e(ui))IQ }),

where Uh = UK or Uh = Ul and constant C is independent of hI, h2 and k.
Clearly, the previous proposition divides the error of the approximation into
four different quantities: the error on the initial conditions, the approximation
error comprising a norm and an integral term, the consistency error coming
from the nonconformity of the finite element methods and an integration er-
ror term on G. We restrict ourselves to the case where rc is a straight line
400
segment for many reasons: because the operator 1rf
could be decomposed on
each straight line segment of f e into different projection operators, to use the
already existing results of Ben Belgacem-Hild-Laborde (1999), established in
the case of a straight line segment, and to avoid a lenghty supplementary con-
vergence study in this paper. Therefore, in the following lemma, we give the
estimates for the approximation and the consistency error terms in the case
Uh = UK and Uh = Uf
Lemma 1 We assume that f e is a straight line segment and let (u, u) be the
solution of(6). Suppose that u is such that u i E V)O(O, T; [H2(Oi)]2), (u 2 .
v 2)(x) = 0, 'Vx E fe nf~ and Uv E LOO(O, T; Ht (fe)). Let (Uhk, Uhk) be
the solution of(7) and n = 0, 1,2, ... , N.
There exists Vh E Uh = UK or Uh = Ul that satisfies the estimates
IVh - unlv + 1( U~[(Vh - un). v] drl t ~ C(ht + h2),

ire
where constant C depend,; linearly on lu 1,nl[H2(f21)]2 and lu2,nl[H2(f22)]2'
Also, there exists v E U that satisfies the estimates
1ire
( u~[(v - u~k) . v] drl t ~ (C hflun - u~klv ) t + C(hl + h2),
where constant C depend,; linearly on lu1,nl[H2(f21))2 and lu 2,nl[H2(f22)]2 and
p, = 1, r = 1 ijUh = UK and p, = ~, r = ~ ijUh = Uf
We finally obtain the convergence result of the fully discrete approximation.
Theorem 2 (Fernandez Garda-Hild-Viano (2001)) Let conditions of Lemma

1 hold and let h = max(h 1 , h2)' We have thefollowing error estimates:
max {Iu n - Uhklv

l'5,.n'5,.N
+ Iu n - UhklQ} ~ C(luo - u~klv
+Iuo - u~klQ + hr + k).
In this section we show some numerical results for constitutive models (6)
when Gi is Perzyna's viscoplastic function (see Ionescu-Sofonea (1993)):
ai(ui , ei ) = - 2~i£i(ui - PBl(Ui )),
where Ai is a viscosity constant and PBl is the projection operator (with respect
to the norm l'T i l2 = ('Ti, £i'Ti)Q(f2l)) on the plasticity convex set Bi C 8 2
401
defined by:
where O"~ represents the uniaxial yield stress. Moreover, both plane stress
elasticity tensors [l (f = 1,2) are given by:
where El and y;,l denote Young's modulus and Poisson's ratio of the material
that occupies the domain nl.
In order to compare results obtained via (local) approximation set Ul and
(global) approximation set U'h, we implement both methods for the simple
problem shown in Figure 1 (see Fernandez Garcia-Hild-Viafio (2000» invol-
ving two viscoplastic bodies (El = 3 X 105 M Pa, y;,l = 0.3, )..l = 100 N .
sec/m2 , O"~ = yTON/m 2, f = 1,2). A density of forces gl(Xl,x2,t) =
(0, -lOt) N/m linearly increasing in time is applied on the upper boundary
of n1 whereas body forces are absent. Also, we consider T = 1 sec, 0'0 =
ON/m 2 anduo = Om.
:r.-'
N
:r::Z
o
Q-------Q------Q------~------Q--------------
Figure 1. Uk: Initial and deformed boundaries at final time T = 1 sec. and contact area.
Figure 1 shows also the deformed boundaries and the deformed meshes
when U'h is assumed. We remark that no interpenetration has been produced
and the obtained stress field is constant (0"11 = 0"12 = 0, 0"22 (Xl, X2, t) = lOt).
In the case where the local approach is chosen (Uh = Ul), the deformed
meshes near the contact area are shown in Figure 2. We now discover a
non negligible and non realistic penetration of n2 into n1 as well as artificial
stresses, particularly near the contact zone.
402
:aD _ 23.05
i. _ 21.15
18 _ 20.51
17 _ 11.80
l' _ 19.03
15 _ 11.25
1& _ 17.&'
13 _ lli.l1
12 _ 15.9'
11 _ 15.16
10 _ 1.... "
,_ 13.11
• _ 12 ....
7 _ 12.0'7
5 _ 11.30
5 _ 10.52
1.750
8.176
•• 203
1 _ 7.&3D
Figure 2. U[: Contact area and stress field at final time T = 1 sec.
References
Ben Belgacem E (1999) The mortar finite element method with Lagrange multipliers, Numer.
Math. 84,173-197.
Ben Belgacem E, Hild P. and Laborde P. (1999) Extension of the mortar finite element method
to a variational inequality modeling unilateral contact, in Math. Mod. Meth. Appl. Sci. 9 (2),
287-303.
Fernandez Garcia, 1.R., Hild, P. and Viano, 1.M. (2000), Resolution numenque d'un probleme
de contact entre corps elasto-viscoplastiques et maillages elements finis incompatibles, C. R.
Acad. Sci. Paris, Serie 1331, 833-838.
Fernandez Garcia, 1.R., Hild, P. and Viano, lM. (2001), Numerical approximation ofthe elastic-
viscoplastic contact problem with non-matching meshes, Prepublication LAMA, Ol-Ola
Submitted.
Han W. and Sofonea M. (2000) Numerical analysis of a frictionless contact problem for elastic-
viscoplastic materials, Comput. Methods Appl. Mech. Engrg. 190, 179-191.
Ionescu, I. R. and Sofonea, M. (1993) Functional and Numerical Methods in Viscoplasticity,
Oxford University Press, Oxford.
Rochdi M. and Sofonea M. (1997) On frictionless contact between two elastic-viscoplastic bod-
ies, Quart. J. Mech. Appl. Math. SO, 481-496.
CONTACT IN THE ARLEQUIN FRAME-
WORK
Hachmi Ben Dhia

Ecole Centrale de Paris
Laboratoire MSS-Mat, CNRS UMR 8579, lirance
Malek Zarroug
Ecole Centrale de Paris
Laboratoire MSS-Mat, CNRS UMR 8579, lirance
Abstract Local refinement of discrete contact problems is needed to analyze ac-

curately contact zones and their neighborhoods. Let us mention in par-
ticular the fine estimation of the contact zone size, stress concentrations
near contact edges, wear and so on. Classically, adaptive approaches are
used to address this issue. The Arlequin method (cf. Ben Dhia (1998,
1999)) is suggested here as (hopefully) a more flexible numerical tool.
1. Introduction
There is an important range of mechanical contact problems for which
the choice of appropriate refined numerical discrete schemes is an im-
portant issue. Among other applications, let us mention the case of
contact edges and more generally problems for which the curvature of
the candidate contact surface is sufficiently large, at least locally. A
soil/foundation interaction problem is a typical example_ Another one
is given by the classical contact test used to measure a material tough-
ness. The multiscale character of contact problems and impact are other
sources of motivation. There is clearly for such kind of problems a need
of a (local at least) numerical refinement in order to capture mechanical
quantities of interest with sufficient accuracy. Classical h-adaptive ap-
proaches (cf. ego Johnson (1991) and Wriggers et al (1998)) can of course
be used to address these contact problems. To ensure the improvement
of the accuracy most of these methods modify the approximation space
by adding independent vectors to enrich the coarse base where needed.
403
404
But let us quote here that this independence requirement constitutes a
lack of "flexibility" of these classical adaptive methods.
Here, the Arlequin method introduced by Ben Dhia (1998), is used as

an alternative. It is based on a superposition of mechanical states and
hence does not suffer the aforementioned lack of flexibility. The aim of
this work is to show how this can be done for contact problems. An
outline of the paper is the following: the Arlequin principles are briefly
recalled in the next section. An academic Laplace problem is used for
this purpose. Our hybrid contact formulation (cf. Ben Dhia et al (2000))
is recalled in section three. Section four is devoted to the formulation
of the same contact problem in the Arlequin framework. The numerical
solution strategy is developed in section five and first numerical results
are shown in the last section.
2. Recall of Arlequin method

Consider the following bidimensional Laplace equation consisting of
finding a scalar displacement field u such that:
{ -ct:.u = f in 0
(1)
u = 0 onaO
where c is a positive real coefficient and where 0 is a regular bounded
domain of 'R2. The weak "mono-model" formulation of this problem is
the following (well-posed one):
Find u in HJ(O) such that for all v in HJ(O),
A1,n(u,v) = l fvdO (2)
with
Aa,w(u,v) = L acVu.Vvdw (3)
Let now S be a sub-domain of O. Set V = HJ(O) X (Hl(S))2. An

equivalent multi-model formulation of Arlequin type can be defined by:
Find (u,u 8,F) in V such that for all (v,v8,F*) in V,

Aa,n(u, v) + A,B,s(uS, VB) - (F, v - VS)s - (F*, u - u 8)s
-/nfvdO=O (4)
405
where (,)8 stands for an equivalent Hl(S) scalar product,:F is a kind of
"glue volumic forces", a and fl are positive functions defined on 0 and
S, respectively, such that :
a + fl = 1 in S, and a = 1 in O\S (5)
It is proved in Ben Dhia and Rateau (2001) that this mixed problem
of the Arlequin family of formulations is well-posed.
The fundamental benefit of Arlequin formulations is that they allow

to approximate the mechanical fields (here u and US) differently. This is
used here for contact problems.
3. Recall of the hybrid contact formulation

This formulation is obtained essentially by writing the contact and
friction laws as (multi-valued) local equations, which, by residual weight-
ing techniques, are weakly formulated. Here, to focus on the application
of the Arlequin method to contact problems and without loss of gener-
ality, we consider the case of the obstacle problem.
Let G denote the function describing the obstacle surface and let A
be the contact pressure on the potential contact surface Oc (figure 1).
The local system taking into account the Signorini unilateral contact
Figure 1. Obstacle problem.
condition is the following:

-c~u - f+A inn
u o onan
d(u) = u - G < 0 in nc (6)
A < 0 in Oc
Ad(u) o in Oc
It is easy to check that Signorini laws are formally equivalent to the
following multi-valued equation:
A- P_(A - p d(u)) = 0, (7)
406
where P_ is the projection on the negative real semi-axis, p is a strictly
positive coefficient and Oc denotes the contact domain, which is assumed
to be strictly contained in O. The hybrid formulation is obtained by a
weak formulation of the equilibrium equation (in (6» and a weighting
of (7). Setting H = HJ(O) x L2(Oc), it reads:
Find (U,A) in H such that for all (v,J') in H,
Al,n(U, v) - { P_(A - pd(u»vd"(

inc
- { !(A - P_(A - pd(u»)J'd"( = ( fvdO (8)
inc p in
The formulation (8) is formal. For the mathematical analysis of mixed
contact formulations, the reader is referred to the works of Brezzi et al
(1978) and to those of Haslinger et al (1991).
4. An Arlequin obstacle problem formulation

Now, an Arlequin hybrid contact formulation for the obstacle problem
can be derived in a rather straightforward manner:
i) set S = Oc and introduce an auxiliary displacement field uC ,
ii) join in Oc the global displacement field u to the local one U C by
following the lines of the Arlequin approach recalled in section 2,
iii) treat the contact conditions only for the local model.
More precisely, by setting W = HJ(O) x (L2(Oc»2, this leads to the
following formulation:
Find (u,.r,U C, A) in W such that for all (v,P,vc,J') in W,
AQ,n (u, v) + A,B ,nc (U C, VC) - { P_ (A - pd(UC»V Cd"(

inc
-(.r,v - vC)nc - (.r*,u - uC)nc - { !(A- P_(A - pd(uC»)J'd"(
inc p
= llvdO (9)
Notice here that if (u,.r, uC, A) is a solution of the Arlequin-obstacle

problem (9), then (u, A) solves the obstacle problem (8). The major
interest of th Arlequin formulation is recovered when the fields u and U C
are approximated differently. For instance, a coarse finite element space
407
can be used to approximate the global solution u while a significantly
finer space is used to approximate the local fields U C and A. Moreover,
let us quote some practical remarks:
1 The contact problem is supported by the local fine model. This

can be exploited for deriving well-suited numerical solvers.
2 The strategy may be viewed as a "volumic" sub-structuring method

and notice that it is easier to deal with the HI-scalar product (the
junction operator) than with the Ht one used in classical dual
substructuring techniques.
5. Solution strategy
The problem (9) is a nonlinear one. This is due to the projection
application. A numerical strategy is needed to solve it. The one we use
consists of introducing the following status function:
X(g) = 1 if 9 ~ 0
{ = 0 otherwise (10)
9 = A - pd(U C ) in !lc
Then, by assigning a given status Xk at each iteration k, we solve, in a

first step, the following linear problem:
Find (uk,:Fk,uck,Ak) in W such that for all (v,:F*,VC,p,) in W,
Aa,n(u k,v) +A,B,ne(uck,vc) - r Xk (Ak - pd(uCk))vcd'Y

Jn e
- (:Fk , V - VC)ne - (:F*, uk - uck)ne

- Jnr !(Ak_Xk(Ak_pd(uCk)))p,d'Y=
e p
r fvd!l
Jn
(11)
The second step consists of updating the status function. If a change

is detected then the first step is executed with the new status function.
The convergence is achieved when no (significant) change is detected
between the given and the updated status functions.
The linear problem, defined by (11), is solved by means of the finite

element/particle method. More precisely, let W h be a finite element
space contained in W. We solve the following finite dimensional linear
system (where the reference to iterations is suppressed for clarity):
408
Find (Uhl UX, Fhl Ah) in W h such that for all (Vhl vhl Fit, Ah) in
W hl
Aa,n(Uhl Vh) + A,B,nc(uhl vh) - IAc (Xkl Ahl Uhl Vh)
-(Fhl Vh - vh)nc = L jVh dO (12)
Ific (Xkl Ahl Uhl P,h) = 0 (13)

(Fhl Uh - u~)nc = 0 (14)
where IAc 0 and TAc 0 refer to an approximation of the contact integrals
by using an appropriate integration scheme (see Ben Dhia and Zarroug
(2001)). Finally, we have to solve the following system:
[ o
o K2 +0 Ak (Bkf
K1 0 (O )T
(02)T
l
1(U~) = ( L1 )
U~ L~
Bk Fk 0 Ak Lk (15)
h 3
0 1 0 2 0 0 Fk 0
where the matrices and the right hand side terms of (15) can be evalu-
ated in a classical manner and where k refers to the iteration.
The methodology can be extended to more general frictional contact

problems involving solids undergoing large transformations. For these
general problems, the integration scheme of contact terms becomes how-
ever an important issue. Let us briefly comment this numerical aspect.
When two deformable bodies are taken into contact, one of them is (clas-
sically) chosen as the "master" body while the other is called the "slave"
body (cf. Hallquist et al (1985)). Since these two bodies can be meshed
and/or evolve differently from one another, the contact surfaces numeri-
cal models are generally incompatible. In Ben Dhia and Zarroug (2001)
we have shown that it is important to integrate accurately the contact
terms which involve a kind of "heterogeneity". An alternative approach
consists of (using basically Arlequin ideas) distributing the computation
of the contact terms between the two contact surfaces. More precisely,
let us consider the following generic contact coupling integral:
(16)
where r~, Ah, v~ and Al stand for the "slave" surface, the contact pres-
sure, the virtual displacement-like field, defined on r~ (the "master"
surface) and the classical target application, respectively. A promising
409
idea consists of distributing this integral as follows:
with 61 and 62 given such that 61 , 62 ~ 0 and 61 + 62 = 1. Notice that

if the trace of the contact fields is piecewise linear with respect to the
appropriate mesh, either the vertices or the mid-points of the edges of
the elements constituting the meshes of r~ and r~ can be respectively
used for the numerical estimation of the two integrals in (17). These
(seemingly new, rather symmetric) integration techniques will be tested
in a forthcoming work.
6. First numerical results

We consider a Boussinesq-like problem. It consists of indenting an
t,.,.,.v ...• u,
, ,,
,
, '
,'
: :
: :
, ,
: :
, ,
, ,
~~,wh~~),;}/.i'##~/.i'.d7ld
Figure 11. Boussinesq problem: arCoarse meshes. b-Arlequin model.
elastic half-space by a square foundation as shown in figure 2-a. A

given displacement Uo is prescribed in the upper boundary of the foun-
dation, while appropriate boundary conditions are applied to the system
as shown in figure 2-a. The particularity of this contact problem is the
singularity (in the elastic range) caused by the edge of contact. To cap-
ture the normal contact pressure singularity (that cannot be seen by
coarse models), fine models are superposed to the coarse ones in the
vicinity of the contact interface (see figure 2-b). We recall that contact
interactions are treated only between fine interfaces (chosen here com-
patible). We note also that we have used constant weight functions Q
and f3 equal to ~ in the glue zones and elastic-like glue forces.
410
Contact pressure
-+- Arlequin approach
- - 6 - Coarse model ,
i
if
JI
Ii
-------.-- -------......
__ -------:-+-->-.+':::<.!!.~(+
__ __ __ x
Figure 3. Contact forces.
Contact pressures obtained by both the coarse model (figure 2-a) and
the Arlequin one (figure 2-b) are plotted in figure 3 where one can notice
the improvement of the accuracy of the result by Arlequin approach.
References
Ben Dhia, H. (1998) Multiscale mechanical problems: The Arlequin method, C. R.
Acad. Sci. Paris, Ser. II, 326, 899 - 904.
Ben Dhia, H. (1999) Numerical modelling of multiscale mechanical problems: the Ar-
lequin method, ECCM'99 CD-ROM, Miinchen, Germany, August 31 - September
3.
Ben Dhia, H., Vautier, L, and Zarroug, M. (2000) Problemes de contact frottant en
grandes transformations: du continu au discret, Revue Europeenne des Elements
Finis, 9, 243 - 26l.
Ben Dhia, H., Rateau, G., (2001) Mathematical analysis of the Arlequin method,
C.R. A cad. Sci. Paris, Ser. I, 332, 649 - 654.
Ben Dhia, H., Zarroug, M., (2001) Mixed frictional contact elements and local "volu-
mic" contact interfaces, ECCM'01 CD-ROM, Cracow, Poland, June 26- 29.
Brezzi, F., Hager, W., and Raviart, P.A., (1978) Error estimates for the finite element
solution of variational inequalities, Numer. Math., 31, 1 -16.
Hallquist, J.O. et al (1985) Sliding interfaces with contact-impact in large-scale La-
grangian computations, Comput. Meth. Appl. Mech. Engng., 51, 107 -137.
Haslinger, J., Hlavacek, 1., and Necas, J., (1991) Numerical methods for unilateral
problems in solid mechanics, In Handbook of Numerical Analysis, IV Part 2, Eds.
P.G. Ciarlet and J.-L. Lions, North Holland, 313 - 485.
Johnson, C., (1991) Adaptative finite element methods for the obstacle problem,
Technical Report, Chalmers University of Technology.
Wriggers, P., Scherif, 0., (1998), Different a posteriori error estimators and indicators
for contact problems, Mathl. Comput. Modelling, 28,4-8, 437 - 447.
Appendix A
SCIENTIFIC PROGRAM OF THE

CMIS 2001
June 17 - 21,2001
Praia da Consolaryoo, Peniche, PORTUGAL
411
413
~onda~June18,2001
Session DYNAMICS AND IMPACT

Chairman: Michel Jean
Jean Jacques Moreau, Numerical dynamics of granular materials (main lecture).
Friedrich Pfeiffer, Measurements of impacts with friction.
Patrick Ballard, Formulation and well-posedness of unilateral multibody dynamics.
Eric Dimnet, Collisions in systems made of rigid bodies.
Session DYNAMICS AND IMPACT (continued)

Chairman: Gianpietro Del Piero
Christoph Glocker, Impacts with global dissipation index at reentrant corners.
Laetitia Paoli, Michelle Schatzman*, Understanding impact through continuous medium
vibrations.
Cuneyt Yilmaz, Yildirim Hurmuzlu*, Solving rocking block problems with multiple
impacts.
Bernard Brogliato*, Mongi Mabrouk, Arturo Zavala Rio, A controllability criterion for
linear juggling mechanical systems.
Mihai Anitescu*, Florian A. Potra, On integrating stiff multibody dynamics with contact
and friction.
A.P. Ivanov, Analysis of systems with multiple frictional contacts.
Session INSTABILITIES, OSCILLATIONS AND WAVES

Chairman: James R. Barber
James R. Rice, Frictional instability in sliding between elastic solids, with geophysical
applications (main lecture).
loan R. lonescu*, Michel Campillo, Cristian Dascalu, Pascal Favreau, Christophe
Voisin, Initiation of friction instability on a plane fault system.
Franck Moirot, Quoc-Son Nguyen*, An example of stick-slip waves.
Werner Schiehlen, Bin Hu*, Peter Eberhard, Longitudinal waves in elastic rods with
discontinuous cross sections.
Session INSTABILITIES, OSCILLATIONS AND WAVES (continued)

Chairman: Michel Raous
Hans True*, Per Grove Thomsen, Rolf Asmund, The dynamics of a railway freight wagon
wheel set with dry friction damping.
M. Rudolph*, K. Popp, The influence of contact properties on friction-induced brake
vibrations.
Riad Hassani, Patrick Hild*, loan lonescu, Analysis of eigenvalue problems modelling
friction: sufficient conditions of non-uniqueness for the elastic equilibrium.
414
Tuesday, June 19, 2001
Session CONTACT, ADHESION AND FRICTION

Chainnan: Michel Fremond
K. L. Johnson*, J. A. Greenwood, Adhesion of viscoelastic spherical solids (main lecture).
Christian Talon, Alain Curnier*, A model of adhesion added to contact with friction.
D.A. Hills*, A. Sackfield*, Analysis of a "walking" punch.
Stanislaw Stupkiewicz*, Zenon Mr6z, Phenomenological model of friction accounting for
subsurface plastic deformation in metal forming.
Session WEAR, THERMOELASTIC AND ELECTRICAL PHENOMENA

Chainnan: K.L. Johnson
Yun-Bo Yi, J.R. Barber*, D.L. Hartsock, Thermoelastic instabilities in automotive disc
brakes - finite element analysis and experimental verification (main lecture).
Nadine Stalin*, Jean Christophe Eytard, Friction coefficient evolution during wear tests
related to infrared emission.
Alfred Zmitrowicz, Variational formulations for contact problems with wear debris at an
interface between sliding surfaces.
Paolo Bisegna, Frederic Lebon*, Franco Maceri, The unilateral frictional contact of a
piezoelectric body with a rigid support (dedicated to the memory of Professor Panagiotis D.
Panagiotopoulos).
G. Zavarise*, D. B05o, B.A. Schrefler, A contact formulation for electrical and mechanical
resistance.
Parallel Poster Session NUMERICAL METHODS AND

MATHEMATICAL ANALYSIS
Chainnan: David E. Stewart
P. Wriggers*, A. Rieger, Adaptive finite element methods for thermomechanical contact
problems.
Laurent Baillet*, Taoufik Sassi, Nonconforming finite element methods for unilateral
contact problems with friction.
Alex Alves Bandeira*, Peter Wriggers, Paulo de Mattos Pimenta, Homogenization
methods leading to interface laws of contact mechanics - a finite element approach for large
3D deformation using augmented lagrangian method.
Gonzalo Alduncin, Nonconforming multidomain finite element approximations of contact
problems.
Jose R. Fernandez-Garcia, Mircea Sofonea, Juan M. Viano*, A convergence result in the
study of frictionless contact problem for elastic-viscoplastic materials with internal state
variables.
Patrice Coorevits, Patrick Hild*, Mohammed Hjiaj, Adaptive finite elements for frictional
contact problems.
L. Slimane, F. Ben Belgacem, Y. Renard*, Mixed finite elements for unilateral contact
problems in incompressible elasticity.
Dominique Chamoret*, Jean-Michel Bergheau, Patrick Saillard, Alain Rassineux, Some
considerations to smooth contact.
Mikiiel Barboteu*, Mircea Sofonea, Variational analysis and numerical simulations in the
study of a contact problem with rotative friction.
Frederic Lebon*, Christian Licht, Fouad Zaittouni, A family of asymptotic contact laws.
415
G. Bayada, M. Chambat*, Multiple scale aspects in micropolar lubrication.

Thierry-Vincent Hoarau-Mantel*, Viorica Venera Motreanu, Quasistatic frictional
contact problems for linear viscoelastic materials.
Bassam Awbi, Variational analysis of a frictional contact problem for viscoelastic bodies.
Oanh Chau, Quasistatic frictional problems with subdifferential boundary conditions.
Parallel Poster Session DYNAMICS AND INSTABILITIES

Chainnan: Christoph Glocker
Michel Fremond*, Raoul Gormas, Jorge San Martin, Collisions of solids and fluids.
J.A. Battle, Assessing the energetical inconsistency of Newton's and Poisson's restitution
rules.
Lars Johansson, Solution of rigid body frictional impact problems with multiple contact
points using a Newton method.
Jens Pfister*, Peter Eberhard, Frictional contact of spatial multibody systems with
kinematic loops.
Asterios A. Liolios, A numerical approach for the dynamic frictional unilateral contact
problem of the seismic interaction between adjacent buildings under second-order
instabilizing effects (dedicated to the memory of Prof Panagiotis D. Panagiatopoulos).
Chuan-yu Wu*, Long-yuan Li, Colin Thornton, The effect of friction and deformation on
the oblique impact behaviour of particles.
T. Zeghloul*, B. Villechaise, Experimental study of stress waves in a dry contact.
Sylvie Wolf*, loan lonescu, Michel Campillo, Instability of a system of two parallel faults
under slip weakening friction.
R. Schroth*, N. Hoffmann, Friction induced oscillations in a pin on disc system.
C.H. Lamarque, O. Janin*, Local stability of singular periodic responses in vibro-impact
oscillators.
Parallel Poster Session CONTACT MODELS, RESULTS AND

APPLICATIONS
Chainnan: Alain Curnier
C.E. Truman*, S.G. Wallace, A. Sackfield, Modelling of multilayered structures subjected
to normal contact loads.
F. Bucher, K. Knothe, A. Liinenschloss*, Threedimensional normal contact problem of
surfaces with measured roughness using multilevel techniques.
H. Hamdi*, M. Dursapt, J.-M. Bergheau, H. Zahouani, Experimental and numerical
investigations on the deformation mechanisms of a rough surface.
H. Zahouani*, M. Dursapt, J.-M. Bergheau, F. Sidoroff, The effect of the length scale of
roughness in contact problems.
M. Borri-Brunetto*, B. Chiaia, Multi-scale analysis of contact between fractal surfaces.
D. Boso*, G. Zavarise, B.A. Schrefler, F. Bellina, Wire-to-wire coupled contact problems in
superconducting coils.
V.J. Pauk, Axisymmetric contact problems involving frictional heating for a layer.
M. Ertz*, K. Knothe, Temperature effects in wheel/rail contact.
A. Bohmer, Shakedown-maps for rail steel considering combined isotropic and kinematic
hardening.
V.L. Popov, S.G. Psakhie, A. Dmitriev, E. Shilko, K. Knothe, M. Ertz*, F. Bucher,
Micro- and nanoscale modeling of friction processes in contact rail-wheel.
416
K. Benzarti*, T. Chaussadent, Adhesive bond in civil engineering structures: contribution

of physico-chemistry to the refinement of mechanical modeling.
S. Launis*, E. Keskinen, M. Cotsaftis, Improvement of wood grinding system by means of
rule based log pressing control.
Session OTHER TOPICS AND APPLICATIONS

Chairman: Friedrich Pfeiffer
Frank Bucher*, Rob S. Dwyer-Joyce, The real area of contact - a combination of
experimental and numerical approaches.
Tomas Jendel*, Mats Berg, Prediction of wheel wear for rail vehicles - methodology and
verification.
Guy Bayada*, Mahdi Boukrouche, Coulomb fluid-solid interface law in lubrication.
Erno Keskinen*, Sirpa Launis, Juha-Matti Kivinen, Multi-body modeling of paper
calendering unit by contact dynamics formulation.
Wednesday, June 20, 2001
Session MATHEMATICAL ANALYSIS

Chairman: Franco Maceri
Lars-Erik Andersson, Anders Kiarbring*, Existence and uniqueness for quasi static contact
problems with friction (main lecture).
W.R. Bielski*, A. Galka, J.J. Telega, Augmented lagrangian methods for a class of
nonconvex contact problems in structural mechanics.
Marius Cocou*, Remi Rocca, Approximation of quasi static Signorini problems with local
friction by a mixed method.
W. Han, K.L. Kuttler, M. Shillor, M. Sofonea*, A beam in adhesive contact.
Session MATHEMATICAL ANALYSIS (continued)

Chairman: Anders Klarbring
Jean-Claude Paumier, Yves Renard*, Elastodynamic friction problem with a "surface
inertia" perturbation.
Christof Eck, Jiri Jarusek*, Solvability of thermo-viscoelastic contact problems with
Coulomb friction and nonlinear heat conductivity.
Jeongho Ahn, David E. Stewart*, A simplified model of impact.
E. Pratt*, J.-M. Ricaud, Convergence for a time discretization of dynamic contact problems
with friction.
Jaroslav Haslinger, Shape optimization in contact problems with Coulomb friction.
Isabel N. Figueiredo*, Joaquim J. JUdice, Silverio S. Rosa, Solution methods for structural
optimization in contact rod problems.
417
Thursday, June 21, 2001
Session NUMERICAL METHODS

Chairman: Serge Cescotto
Michel Raous*, Yann Monerie, Unilateral contact, friction and adhesion: 3D cracks in
composite materials (main lecture).
Przemyslaw Litewka*, Peter Wriggers, Contact between 3D beams with rectangular cross-
sections.
Peter Ireman*, Anders Klarbring, Niclas Stromberg, Algorithms for thennoelastic wear
problems.
Lovre Krstulovic-Opara*, Peter Wriggers, Convergence studies for 3D smooth frictional
contact elements based on the quartic Bezier surfaces.
Session NUMERICAL METHODS (continued)

Chairman: Peter Wriggers
Serge Cescotto, Variational approaches of unilateral contact with friction between
defonnable solids undergoing large strains.
P. Alart*, M. Barboteu, Multicontact structures and parallel computing.
J.A. Gonzalez*, R. Abascal, A new approach to the 2D transient rolling contact problem.
Jose R. Fernandez-Garcia, Juan Viafio*, Patrick Hild, Numerical approximation of the
elastic-viscoplastic contact problem using noncoinciding finite element meshes.
Hachmi Ben Dhia*, Malek Zarroug, Contact in the Arlequin framework.
Appendix B
LIST OF PARTICIPANTS IN THE
CMIS 2001
June 17-21, 2001
Praia da Conso/ac;:iio, Peniche, PORTUGAL
419
421
Alart Pierre Alduncin Gonzalo

L.M.O.C. Instituto de Geofisica
Universite Montpellier II Universidad Nacional Aut6noma Mexico
Cc. 48, Place Eugene Battaillon Ciudad Universitaria Del. Coyoacan
34095 Montpellier Cedex 5 Mexico, D.F. 04510
France Mexico
alart@lmgc.univ-montp2.fr alduncin@igeofcu.unam.mx
Amassad Amina Anitescu Mihai

Laboratoire l-A Dieudonne Dept. of Mathematics
Universite de Nice Sophia-Antipolis University of Pittsburgh
Parc Valrose Thackeray 301
06108 Nice Cedex 2 Pittsburgh, PA 15260
France U.S.A.
amassad@math.unice.fr anitescu@math.pitt.edu
Arantes e Oliveira Eduardo R. AwbiBassam

Departamento de Engenharia Civil Laboratoire de Theorie des Systemes
Instituto Superior Tecnico Universite de Perpignan
Av. Rovisco Pais 52, avenue de Villeneuve
1049-001 LISBOA 66860 Perpignan Cedex
Portugal France
arantes@civil.ist.utl.pt awbi@univ-perp.fr
Baillet Laurent Ballard Patrick

Laboratoire de Mecanique du Contact Laboratoire de Mecanique des Solides
lnstitut National Sciences AppJiquees Lyon Ecole Polytechnique
20 Av. Albert Einstein, Bat. 113
69621 Villeurbanne Cedex 91128 Palaiseau Cedex
France France
laurent. bai Ilet@lmc.insa-lyon.fr ballard@lms.polytechnique.fr
Bandeira Alex Alves Barbarosie Cristian

Institute for Structural Engineering Centro Matematica ApJic~oes Fundamentais
Universidade de Soo Paulo Universidade de Lisboa
Rua Mourato Coelho 738/42, Pinheiros Av. Prof. Gama Pinto 2
Sao Paulo, SP 1649-003 LISBOA
Brasil Portugal
alex_bandeira@yahoo.com.br barbaros@lmc.fc.ul.pt
Barber James R. Barboteu Mikliel

Department of Mechanical Engineering Laboratoire de Theorie des Systemes
University of Michigan Universite de Perpignan
2350 Hayward St 52, avenue de Villeneuve
Ann Arbor MI 48109-2125 66860 Perpignan Cedex
U.S.A. France
jbarber@engin.umich.edu barboteu@univ-perp.fr
422
Batlle 1. A. Bayada Guy

Dept. of Mechanical Engineering INSA Mathematique
Polytechnical University of Catalunya Institut National Sciences Appliquees Lyon
Campus Sud - Edif. PO, Av. Diagonal, 647 Bat. 40 I MATH
08028 Barcelona 69621 VILLEURBANNE Cedex
Spain France
agullo@em.upc.es bayada@insa.insa-Iyon.fr
Ben Dhia Hachmi Benzarti K.

LMSS & Mat Service PCM
Ecole Centrale de Paris Laboratoire Central des Ponts et Chaussees
Grande Voie des Vignes 58 Boulevard Lefebvre
92295 CHATENAY-MALABRY Cedex 75732 Paris Cedex IS
France France
bendhia@mss.ecp.fr Benzarti@\cpc.fr
Berg Mats Bergheau Jean-Michel

Dep. Vehicle Eng. - Div. Railway Technology Lab. Tribologie et Dynamique des Systemes
KTH-Royallnstitut of Technology ENISE
Teknikringen 8 58, rue Jean Parot
SE-I0044 Stockholm 42023 Saint Etienne Cedex 2
Sweden France
mabe@fkt.kth.se bergheau@enise.fr
Bielski Wlodzimierz R. Bohmer Alexander

Institute of Geophysics Aerospace Institute
Polish Academy of Sciences Technische Universitiit Berlin
ul. K. Janusza 64 Sekr.5, Marchstr. 12
01-452 Warszaw 0-10587 Berlin
Poland Germany
wbielski@igf.edu.pl Alex.Boehmer@TU-BerIin.de
Borri-Brunetto Mauro B050 Daniela

Dept. of Structural and Geotechnical Eng. Dep. Construction and Transport. Engineering
Politecnico di Torino University of Padua
Corso Duca degli Abruzzi 24 Via F. Marzolo, 9
10129 Torino 35131 Padova
Italy Italy
borrib@polito.it boso@caronte.dic.unipd.it
Brogliato Bernard Bucher Frank

INRIA Rhone-Alpes Aerospace Institute
ZIRST Montbonnot Technical University of Berlin
655 avenue de I'Europe Sekr.F5, Marchstr.12
38334 Saint Ismier Cedex 0-10587 Berlin
France Germany
Bemard.Brogliato@inrialpes.fr frank.bucher@tu-berlin.de
423
Cescotto Serge Cham bat Michele

Dept. Mecanique des Materiaux et des Solides MAPL Y
Universite de Liege Universite Lyon I
Chemin Chevreuils I, Bat 52!3, Sart TILMAN Bat. 101
B-4000 LIEGE 69622 VILLEURBANNE Cedex
Belgium France
Serge. Cescotto@ulg.ac.be chambat@maply.univ-lyonIJr
Chamoret Dominique Chau Oanh

Laboratoire de Theorie des Systemes
ESI Software Universite de Perpignan
Le Discover, 84 bd Vivier Merle 4, rue du theatre
69485 Lyon Cedex 03 66000 Perpignan
France France
chamoret@enise.fr chau@univ-perp.fr
Cocou Marius Curnier Alain

Equipe Mecanique du Contact Departement de Mecanique
Laboratoire de Mecanique et d'Acoustique Ecole Polytechnique Federale de Lausanne
31, chemin Joseph-Aiguier LMAF-DGM-EPFL
13402 MARSEILLE CEDEX 20 CH - 10 15 Lausanne
France Switzerland
cocou@lma.cnrs-mrs.fr Alain.Cumier@epfl.ch
Del Piero Gianpietro Dimnet Eric

Dipartimento di Ingegneria
Universita di Ferrara Laboratoire Central des Ponts et Chaussees
Via Saragat I 58, boulevard Lefebvre
44100 FERRARA 75732 Paris cedex 15
Italy France
gdpiero@ing.unife.it dimnet@lcpc.fr
Dursapt Michel Ertz Martin

Lab. Tribologie et Dynamique des Systemes Aerospace Institute
ENISE Technische Universitat Berlin
58, rue Jean Parot Sekr. F5, Marchstr. 12
42023 Saint Etienne Cedex 2 0-10587 Berlin
France Germany
dursapt@enise.fr martin@ice.fb12.tu-berlin.de
Figueiredo Isabel N. Fremond Michel

Departamento de Matematica LMSGC-UMR 113
Universidade de Coimbra Laboratoire Central des Ponts et Chaussees
Apartado 3008 Cite Descartes, 2 Allee Kepler
3001-454 Coimbra 77420 CHAMPS SUR MARNE
Portugal France
isabelf@mat.uc.pt fremond@lcpc.fr
424
Glocker Christoph Gonzalez Jose A.

Dept. of Mechanical Engineering Departamento de Estructuras
Institute of Mechanical Systems Escuela Superior de Ingenieros
ETH Zentrum, CLA 121.1 Avda. de los Descrubrimientos sin
CH-8092 ZUrich 41092 Sevilla
Switzerland Spain
christoph.glocker@imes.mavt.ethz.ch japerez@cica.es
Hamdi Hedi Haslinger Jaroslav

Lab. Tribologie et Dynamique des Systemes University Karlovy
ENISE Matematicko-Fyzikalni Fakulta
58, rue Jean Parot Ke Karlovu 5
42023 Saint Etienne Cedex 2 12116 PRAHA 2
France Czech Republic
hamdi@enise.fr haslin@met.mff.cuni.cz
Hassani Riad Hild Patrick

Lab. Geophysique Interne et Tectonophys. Laboratoire de Mathematiques
Universite de Savoie Universite de Savoie
Campus Scientifique Campus Scientifique
73376 Le Bourget-du-Lac 73376 Le Bourget-du-Lac
France France
hassani@univ-savoie.fr hild@univ-savoie.fr
Hills David A. Hoarau-Mantel Thierry-Vincent

Department of Engineering Science Laboratoire de Theorie des Systemes
University of Oxford Universite de Perpignan
Parks Road 52, avenue de Villeneuve
OXFORD OXI 3PJ 66860 Perpignan Cedex
United Kingdom France
david.hills@eng.ox.ac.uk hoarau@univ-perpJr
Hu Bin Hurmuzlu Yildirim

Institut B fur Mechanik Mechanical Engineering Department
Universitat Stuttgart Southern Methodist University
Pfaffenwaldring 9 3160 SMU Blvd.
D-70550 Stuttgart Dallas, TX 75275
Germany U.S.A.
hbi@mechb.uni-stuttgart.de hurmuzlu@seas.smu.edu
[onescu loan R. Ireman Peter

Laboratoire de Mathematiques Department of Mechanical Engineering
Universite de Savoie Linkoping University
Campus Scientifique
73376 Le Bourget-du-Lac SE-581 83 Linkoping
France Sweden
ionescu@univ-savoie.fr petir@ikp.liu.se
425
Ivanov Alexander P. Janin Olivier

Mathematical Modelling Laboratoire Geomateriaux
Moscow State Technical University Ecole Nationale des Travaux Publics de I'Etat
Sev.Chertanovo,5 a-23 I, rue Maurice Audin
113648 Moscow 69518 V AULX-EN-VELIN CEDEX
Russia France
apivanov@orc.ru olivier.janin@entpe.fr
Jarnsek liri Jean Michel

Mathematical Institute Equipe ModeIes Numeriques
Acad. Sci. of Czech Rep. Institut Mediterraneen de Technologie
Zitmi25 Technop61e de Chateau-Gombert
11567 Praha I 13451 Marseille Cedex 20
Czech Republic France
jarusek@math.cas.cz mjean@imtumn.esm2.imt-mrs.fr
Johansson Lars Johnson K. L.

Division of Mechanics Department of Engineering
Link6ping University University of Cambridge
Trumpington Street
SE-581 83 Link6ping CAMBRIDGE CB2 1PZ
Sweden United Kingdom
larjo@ikp.1iu.se klj 1000@eng.cam.ac.uk
Keskinen Emo Klarbring Anders

Laboratory of Machine Dynamics Division of Mechanics
Tampere University of Technology Link6ping University
PO BOX 589
33101 Tampere SE-581 83 Linkoping
Finland Sweden
emo.keskinen@tut.fi andkl@ikp.liu.se
Krstnlovic-Opara Lovre Lannis Sirpa S.

Faculty Electrical, Mechanical and Naval Eng. Laboratory of Machine Dynamics
University of Split Tampere University of Technology
Rudera Boskovica bb PO BOX 589
HR-21000 Split 33101 Tampere
Croatia Finland
lovre.krstulovic-opara@fesb.hr sirpa.launis@tut.fi
Lebon Frederic Leger Alain

L.M.G.C. Mecanique et Modeles Numeriques
Universite de Montpellier II Direction des Etudes et Recherches EDF
CC.048 I, Avenue General de Gaulle, BP. 408
34095 MONTPELLIER Cedex 5 92141 CLAMART Cedex
France France
fred@lmgc.univ-montp2.fr Alain.LEGER@der.edfgdf.fr
426
Liolios Asterios Litewka Przemyslaw
Dept. Civil Engineering Institute of Structural Engineering
Democritus University of Thrace Poznan University of Technology
Inst. Struct. Mechanics and Earthquake Eng. ul. Piotrowo 5
GR- 67100 Xanthi 60-965 Poznan
Greece Poland
liolios@demokritos.cc.duth.gr litewkap@put.poznan.pl
Liinenschloss Alexander Maceri Franco

Dip. di Ing. Civile
Technical University of Berlin Universita di Roma "Tor Vergata"
Dunckerstr. 90a Via di Tor Vergata 110
D-I0437 Berlin 00133 ROMA
Germany Italy
luenenschloss@bigfoot.com maceri@ing.uniroma2.it
Marques Manuel D. P. Monteiro Martins Joao A. C.

Centro Matematica Aplica.;:Oes Fundamentais Dep. Eng. Civil
Universidade de Lisboa Instituto Superior Tecnico
Av. Prof. Gama Pinto 2 Av. Rovisco Pais
1649-003 LlSBOA 1049-001 Lisboa
Portugal Portugal
mmarques@lmc.fc.ul.pt jmartins@civil.ist.utl.pt
Mikkola Martti Moreau J. J.

Lab. of Theoretical and Applied Mechanics L.M.G.C.
Helsinki University of Technology Universite Montpellier II
P.O. Box 1100 Cc. 048, Place Eugene Battaillon
FIN - 02015 HUT 34095 MONTPELLIER Cedex 5
Finland France
martti.mikkola@hut.fi moreau@lmgc.univ-montp2.fr
Nguyen Quoc Son Pauk Volodymyr J.

Laboratoire de Mecanique des Solides Dept. Geotechnics and Structural Eng. (K-66)
Ecole Polytechnique Lodz University of Technology
AI. Politechniki 6
91128 Palaiseau Cedex 90-924 Lodz
France Poland
son@lms.polytechnique.fr pauk@ck-sg.p.lodz.pl
Pfeiffer Friedrich Pfister Jens

Lehrstuhl flir Angewandte Mechanik Institut B fur Mechanik
Technische Universitiit Miinchen Universitiit Stuttgart
Boltzmannstr.15 Pfaffenwaldring 9
D-85748 Garching D-70550 Stuttgart
Germany Germany
pfeiffer@lbm.mw.tu-muenchen.de jp@mechb.uni-stuttgart.de
427
Pinto da Costa Antonio Pires Eduardo Borges

Dept. Eng. Civil Departamento de Engenharia Civil
Instituto Superior Tecnico Instituto Superior Tecnico
Av. Rovisco Pais Av. Rovisco Pais
1049-00 I Lisboa 1096-00 I Lisboa
Portugal Portugal
apcosta@civil.ist.utl.pt bpires@civil.ist.utl.pt
Pratt Elaine Raous Michel

Equipe Mecanique du Contact Equipe Mecanique du Contact
Laboratoire de Mecanique et d'Acoustique Laboratoire de Mecanique et d'Acoustique
31 chemin Joseph-Aiguier 31, Chemin Joseph-Aiguier
13402 MARSEILLE CEDEX 20 13402 MARSEILLE CEDEX 20
France France
pratt@lma.cnrs-mrs.fr raous@lma.cnrs-mrs.fr
Renard Yves Rice James R.

Departement de Mathematiques Dept. of Earth and Planetary Sciences and
MIP - INSA de Toulouse Division of Engineering and Applied Sciences
135 avenue de Rangueil Harvard University, 20 Oxford Street
31077 Toulouse Cedex 4 Cambridge MA 02138
France U.S.A.
Yves. Renard@gmm.insa-t1se.fr rice@esag.harvard.edu
Rietz Andreas Rudolph Martin

Department of Mathematics Institut fUr Mechanik
Linkoping University Universitiit Hannover
Appelstr. II
SE-581 83 Linkoping D-30165 Hannover
Sweden Germany
anrie@mai.liu.se rudolph@ifrn.uni-hannover.de
Sackfield Anthony Sassi Taoufik

Mathematics Department MAPLY
Nottingham Trent University Institut National Sciences Appliquees Lyon
Burton Street 20, avo Albert Einstein, Bat. 401
Nottingham NG I 4BU 69621 Villeurbanne
anthony.sackfield@ntu.ac.uk taoufik.sassi@insa-Iyon.fr
Schatzman Michelle Schroth Ruediger

Lab. Mathematiques Appliq. Lyon MAPLY FV/FLP
Universite Claude Bernard Lyon I Robert Bosch GmbH
21 avo Claude Bernard Postfach 10 60 50
69622 Villeurbanne Cedex D-70049 Stuttgart
France Germany
schatz@maply.univ-Iyonl.fr ruediger.schroth@de.bosch.com
428
Shillor Meir Simoes Fernando M. F.

Department of Mathematics and Statistics Departamento de Engenharia Civil
Oakland University Instituto Superior Tecnico
Av. Rovisco Pais
Rochester, MI 48309-4401 1096-001 Lisboa
U.S.A. Portugal
shillor@oakland.edu fsimoes@civil.ist.utl.pt
Sofonea Mircea Stalin Nadine

Laboratoire de Theorie des Systemes Direction des Constructions Navales
Universite de Perpignan Etablissement S.c., Site du Mourillon
52, avenue de Villeneuve BP 30
66860 Perpignan Cedex 83800 Toulon Naval
France France
sofonea@univ-perp.fr
Stewart David E. Stupkiewicz Stanislaw

Department of Mathematics Institute Fundamental Technological Research
The University of Iowa Polish Academy of Sciences
14 Mc Lean Hall ul. Swietokrzyska 21
Iowa City IA 52242-1419 00-049 WARSZA WA
U.S.A. Poland
dstewart@math.uiowa.edu sstupkie@ippt.gov.pl
Toader Anca-Maria True Hans

Centro Matematica Aplicayoes Fundamentais Department of Mathematical Modelling
Universidade de Lisboa Technical University of Denmark
Av. Prof. Gama Pinto 2 Richard Petersens Plads, Building 321
1649-003 LISBOA DK-2800 Lyngby
Portugal Denmark
amtan@lmc.fc.ul.pt ht@imm.dtu.dk
Truman Christopher E. Viano Juan M.

Department of Mechanical Engineering Departamento de Matematica Aplicada
University of Bristol Universidade de Santiago de Compostela
Queen's Building, University Walk Campus Universitario Sur, sin
Bristol BS8 1TR 15706 SANTIAGO DE COMPOSTELA
United Kingdom Spain
c.e.truman@bris.ac.uk maviano@usc.es
Wallace Steven G. Wolf Sylvie

Department of Mathematics LGIT - Maison des Geosciences
Nottingham Trent University Universite Joseph Fourier
Burton Street 1381 rue de la Piscine, B.P. 53 X
Nottingham NG 1 4BU 38041 GRENOBLE Cedex 9
steven. waI1ace@ntu.ac.uk sylvie.wolf@obs.ujf-grenoble.fr
429
Wriggers Peter Wu Chuan-yu

Institut fUr Baumechanik und Num. Mechanik Department of Civil Engineering
Universitat Hannover Aston University
Appelstr. 9a Aston Triangle
D-30167 Hannover Birmingham B4 7ET
Germany United Kingdom
wriggers@ibnm.uni-hannover.de wucy@aston.ac.uk
Zahouani Hassan Zavarise Giorgio

Lab. Tribologie et Dynamique des Systemes Dept. of Structural and Geotechnical Eng.
ENISE Politecnico di Torino
58, rue Jean Parot Corso Duca degli Abruzzi 24
42023 Saint Etienne Cedex 2 10129 Torino
France Italy
zahouani@enise.fr zavarise@athena.polito.it
Zeghloul Thami Zmitrowicz Alfred

Laboratoire de Mecanique des Solides Institute of Fluid-Flow Machinery
Universite de Poitiers Polish Academy of Sciences
IUT Angouleme, 4 Av. de Varsovie ul. J. Fiszera 14
16 021 Angouleme Cedex PL- 80-952 Gdansk
France Poland
tzeghloul@iutang.univ-poitiers.fr azmit@imp.pg.gda.pl
Mechanics
Series Editor: G.M.L. Gladwell
Aims and Scope of the Series
The fundamental questions arising in mechanics are: Why?, How?, and How much? The aim of
this series is to provide lucid accounts written by authoritative researchers giving vision and insight
in answering these questions on the subject of mechanics as it relates to solids. The scope of the
series covers the entire spectrum of solid mechanics. Thus it includes the foundation of mechanics;
variational formulations; computational mechanics; statics, kinematics and dynamics of rigid and
elastic bodies; vibrations of solids and structures; dynamical systems and chaos; the theories of
elasticity, plasticity and viscoelasticity; composite materials; rods, beams, shells and membranes;
structural control and stability; soils, rocks and geomechanics; fracture; tribology; experimental
mechanics; biomechanics and machine design.
1. R.T. Haftka, Z. Giirdal and M.P. Kamat: Elements ofStructural Optimization. 2nd rev.ed., 1990
ISBN 0-7923-0608-2
2. U. Kalker: Three-Dimensional Elastic Bodies in Rolling Contact. 1990 ISBN 0-7923-0712-7
3. P. Karasudhi: Foundations of Solid Mechanics. 1991 ISBN 0-7923-0772-0
4. Not published
5. Not published.
6. IF. Doyle: Static and Dynamic Analysis of Structures. With an Emphasis on Mechanics and
Computer Matrix Methods. 1991 ISBN 0-7923-1124-8; Pb 0-7923-1208-2
7. 0.0. Ochoa and J.N. Reddy: Finite Element Analysis of Composite Laminates.
ISBN 0-7923-1125-6
8. M.H. Aliabadi and D.P. Rooke: Numerical Fracture Mechanics. ISBN 0-7923-1175-2
9. 1 Angeles and e.S. Lopez-Cajun: Optimization of Cam Mechanisms. 1991
ISBN 0-7923': 1355-0
10. D.E. Grierson, A. Franchi and P. Riva (eds.): Progress in Structural Engineering. 1991
ISBN 0-7923-1396-8
11. R.T. Haftka and Z. Giirdal: Elements of Structural Optimization. 3rd rev. and expo ed. 1992
ISBN 0-7923-1504-9; Pb 0-7923-1505-7
12. J.R. Barber: Elasticity. 1992 ISBN 0-7923-1609-6; Pb 0-7923-161O-X
13. H.S. Tzou and G.L. Anderson (eds.): Intelligent Structural Systems. 1992
ISBN 0-7923-1920-6
14. E.E. Gdoutos: Fracture Mechanics. An Introduction. 1993 ISBN 0-7923-1932-X
15. J.P. Ward: Solid Mechanics. An Introduction. 1992 ISBN 0-7923-1949-4
16. M. Farshad: Design and Analysis of Shell Structures. 1992 ISBN 0-7923-1950-8
17. H.S. Tzou and T. Fukuda (eds.): Precision Sensors, Actuators and Systems. 1992
ISBN 0-7923-2015-8
18. lR. Vinson: The Behavior of Shells Composed of Isotropic and Composite Materials. 1993
ISBN 0-7923-2113-8
19. H.S. Tzou: Piezoelectric Shells. Distributed Sensing and Control of Continua. 1993
ISBN 0-7923-2186-3
20. W. Schiehlen (ed.): Advanced Multibody System Dynamics. Simulation and Software Tools.
1993 ISBN 0-7923-2192-8
21. C.-W. Lee: Vibration Analysis of Rotors. 1993 ISBN 0-7923-2300-9
22. D.R. Smith: An Introduction to Continuum Mechanics. 1993 ISBN 0-7923-2454-4
23. G.M.L. Gladwell: Inverse Problems in Scattering. An Introduction. 1993 ISBN 0-7923-2478-1
Mechanics
24. G. Prathap: The Finite Element Method in Structural Mechanics. 1993 ISBN 0-7923-2492-7
25. 1 Herskovits (ed.): Advances in Structural Optimization. 1995 ISBN 0-7923-2510-9
26. M.A. Gonzalez-Palacios and J. Angeles: Cam Synthesis. 1993 ISBN 0-7923-2536-2
27. W.S. Hall: The Boundary Element Method. 1993 ISBN 0-7923-2580-X
28. 1 Angeles, G. Hommel and P. Kovacs (eds.): Computational Kinematics. 1993
ISBN 0-7923-2585-0
29. A. Curnier: Computational Methods in Solid Mechanics. 1994 ISBN 0-7923-2761-6
30. D.A. Hills and D. Nowell: Mechanics of Fretting Fatigue. 1994 ISBN 0-7923-2866-3
31. B. Tabarrok and EP.I. Rimrott: Variational Methods and Complementary Formulations in
Dynamics. 1994 ISBN 0-7923-2923-6
32. E.H. Dowell (ed.), E.E Crawley, H.C. Curtiss Jr., D.A. Peters, R. H. Scanlan and E Sisto: A
Modem Course in Aeroelasticity. Third Revised and Enlarged Edition. 1995
ISBN 0-7923-2788-8; Pb: 0-7923-2789-6
33. A. Preumont: Random Vibration and Spectral Analysis. 1994 ISBN 0-7923-3036-6
34. J.N. Reddy (ed.): Mechanics of Composite Materials. Selected works of Nicholas 1 Pagano.
1994 ISBN 0-7923-3041-2
35. A.P.S. Selvadurai (ed.): Mechanics of Poroelastic Media. 1996 ISBN 0-7923-3329-2
36. Z. Mr6z, D. Weichert, S. Dorosz (eds.): Inelastic Behaviour of Structures under Variable
Loads. 1995 ISBN 0-7923-3397-7
37. R. Pyrz (ed.): IUTAM Symposium on Microstructure-Property Interactions in Composite
Materials. Proceedings of the IUTAM Symposium held in Aalborg, Denmark. 1995
ISBN 0-7923-3427-2
38. M.I. Friswell and lE. Mottershead: Finite Element Model Updating in Structural Dynamics.
1995 ISBN 0-7923-3431-0
39. D.E Parker and A.H. England (eds.): IUTAM Symposium on Anisotropy, Inhomogeneity and
Nonlinearity in Solid Mechanics. Proceedings of the IUTAM Symposium held in Nottingham,
U.K. 1995 ISBN 0-7923-3594-5
40. J.-P. MerIet and B. Ravani (eds.): Computational Kinematics '95. 1995 ISBN 0-7923-3673-9
41. L.P. Lebedev, l.l. Vorovich and G.M.L. Gladwell: Functional Analysis. Applications in Mech-
anics and Inverse Problems. 1996 ISBN 0-7923-3849-9
42. J. Mencik: Mechanics of Components with Treated or Coated Surfaces. 1996
ISBN 0-7923-3700-X
43. D. Bestle and W. SchieWen (eds.): IUTAM Symposium on Optimization ofMechanical Systems.
Proceedings of the IUTAM Symposium held in Stuttgart, Germany. 1996
ISBN 0-7923-3830-8
44. D.A. Hills, P.A. Kelly, D.N. Dai and A.M. Korsunsky: Solution of Crack Problems. The
Distributed Dislocation Technique. 1996 ISBN 0-7923-3848-0
45. V.A. Squire, R.I. Hosking, A.D. Kerr and P.I. Langhorne: Moving Loads on Ice Plates. 1996
ISBN 0-7923-3953-3
46. A. Pineau and A. Zaoui (eds.): IUTAM Symposium on Micromechanics of Plasticity and
Damage of Multiphase Materials. Proceedings of the IUTAM Symposium held in Sevres,
Paris, France. 1996 ISBN 0-7923-4188-0
47. A. Naess and S. Krenk (eds.): IUTAM Symposium on Advances in Nonlinear Stochastic
Mechanics. Proceedings of the IUTAM Symposium held in Trondheim, Norway. 1996
ISBN 0-7923-4193-7
48. D. Ie~an and A. Scalia: Thermoelastic Deformations. 1996 ISBN 0-7923-4230-5
Mechanics
SOUD MECHANICS AND ITS APPLICATIONS
49. J.R. Willis (ed.): IUTAM Symposium on Nonlinear Analysis of Fracture. Proceedings of the
IUTAM Symposium held in Cambridge, U.K. 1997 ISBN 0-7923-4378-6
50.· A. Preumont: Vibration Control ofActive Structures. An Introduction. 1997
ISBN 0-7923-4392-1
51. O.P. Cherepanov: Methods of Fracture Mechanics: Solid Matter Physics. 1997
ISBN 0-7923-4408-1
52. D.H. van Campen (ed.): IUTAM Symposium on Interaction betwten Dynamics and Control in
Advanced Mechanical Systems. Proceedings of the IUTAM Symposium held in Eindhoven,
The Netherlands. 1997 ISBN 0-7923-4429-4
53. N.A. FleckandA.C.F. Cocks (eds.): IUTAM Symposium on Mechanics ofGranular and Porous
Materials. Proceedings of the IUTAM Symposium held in Cambridge, U.K. 1997
ISBN 0-7923-4553-3
54. J. Roorda and N.K. Srivastava (eds.): Trends in Structural Mechanics. Theory, Practice, Edu-
cation. 1997 ISBN 0-7923-4603-3
55. Yu.A. Mitropolskii and N. Van Dao: Applied Asymptotic Methods in Nonlinear Oscillations.
1997 ISBN 0-7923-4605-X
56. C. Ouedes Soares (ed.): Probabilistic Methods for Structural Design. 1997
ISBN 0-7923-4670-X
57. D. Fran\rois, A. Pineau and A. Zaoui: Mechanical Behaviour ofMaterials. Volume I: Elasticity
and Plasticity. 1998 ISBN 0-7923-4894-X
58. D. Fran\rois, A. Pineau and A. Zaoui: Mechanical Behaviour of Materials. Volume II: Visco-
plasticity, Damage, Fracture and Contact Mechanics. 1998 ISBN 0-7923-4895-8
59. L.T. Tenek and J. Argyris: Finite Element Analysisfor Composite Structures. 1998
ISBN 0-7923-4899-0
60. Y.A. Bahei-EI-Din and OJ. Dvorak (eds.): IUTAM Symposium on Transformation Problems
in Composite and Active Materials. Proceedings of the IUTAM Symposium held in Cairo,
Egypt. 1998 ISBN 0-7923-5122-3
61. 1.0. Ooryacheva: Contact Mechanics in Tribology. 1998 ISBN 0-7923-5257-2
62. O.T. Bruhns and E. Stein (eds.): IUTAM Symposium on Micro- and Macrostructural Aspects
of Thermoplasticity. Proceedings of the IUTAM Symposium held in Bochum, Germany. 1999
ISBN 0-7923-5265-3
63. F.C. Moon: IUTAM Symposium on New Applications of Nonlinear and Chaotic Dynamics in
Mechanics. Proceedings of the IUTAM Symposium held in Ithaca, NY, USA. 1998
ISBN 0-7923-5276-9
64. R. Wang: IUTAM Symposium on Rheology of Bodies with Defects. Proceedings of the IUTAM
Symposium held in Beijing, China. 1999 ISBN 0-7923-5297-1
65. Yu.I. Dimitrienko: Thermomechanics of Composites under High Temperatures. 1999
ISBN 0-7923-4899-0
66. P. Argoul, M. Fremond and Q.S. Nguyen (eds.): IUTAM Symposium on Variations ofDomains
and Free-Boundary Problems in Solid Mechanics. Proceedings of the IUTAM Symposium
held in Paris, France. 1999 ISBN 0-7923-5450-8
67. FJ. Fahy and W.O. Price (eds.): IUTAM Symposium on Statistical Energy Analysis. Proceedings
of the IUTAM Symposium held in Southampton, U.K. 1999 ISBN 0-7923-5457-5
68. H.A. Mang and F.O. Rammerstorfer (eds.): IUTAM Symposium on Discretization Methods in
Structural Mechanics. Proceedings of the IUTAM Symposium held in Vienna, Austria. 1999
ISBN 0-7923-5591-1
Mechanics
69. P. Pedersen and M.P. Bendsj/le (eds.): IUTAM Symposium on Synthesis in Bio Solid Mechanics.
Proceedings of the IUTAM Symposium held in Copenhagen, Denmark. 1999
ISBN 0-7923-5615-2
70. S.K. Agrawal and B.C. Fabien: Optimization of Dynamic Systems. 1999
ISBN 0-7923-5681-0
71. A. Carpinteri: Nonlinear Crack Models for Nonmetallic Materials. 1999
ISBN 0-7923-5750-7
72. F. Pfeifer (ed.): IUTAM Symposium on Unilateral Multibody Contacts. Proceedings of the
IUTAM Symposium held in Munich, Germany. 1999 ISBN 0-7923-6030-3
73. E. Lavendelis and M. Zakrzhevsky (eds.): IUTAMlIFToMM Symposium on Synthesis ofNon-
linear Dynamical Systems. Proceedings of the IUTAMlIFfoMM Symposium held in Riga,
Latvia. 2000 ISBN 0-7923-6106-7
74. J.-P. Merlet: Parallel Robots. 2000 ISBN 0-7923-6308-6
75. J.T. Pindera: Techniques of Tomographic Isodyne Stress Analysis. 2000 ISBN 0-7923-6388-4
76. G.A Maugin, R. Drouot and F. Sidoroff (eds.): Continuum Thermomechanics. The Art and
Science of Modelling Material Behaviour. 2000 ISBN 0-7923-6407-4
77. N. Van Dao and EJ. Kreuzer (eds.): IUTAM Symposium on Recent Developments in Non-linear
Oscillations of Mechanical Systems. 2000 ISBN 0-7923-6470-8
78. S.D. Akbarov and AN. Guz: Mechanics of Curved Composites. 2000 ISBN 0-7923-6477-5
79. M.B. Rubin: Cosserat Theories: Shells, Rods and Points. 2000 ISBN 0-7923-6489-9
80. S. Pellegrino and S.D. Guest (eds.): IUTAM-IASS Symposium on Deployable Structures: Theory
and Applications. Proceedings ofthe IUTAM-IASS Symposium held in Cambridge, U.K., 6-9
September 1998. 2000 ISBN 0-7923-6516-X
81. A.D. Rosato and D.L. Blackmore (eds.): IUTAM Symposium on Segregation in Granular
Flows. Proceedings of the IUTAM Symposium held in Cape May, NJ, U.S.A, June 5-10,
1999.2000 ISBN 0-7923-6547-X
82. A. Lagarde (ed.): IUTAM Symposium on Advanced Optical Methods and Applications in Solid
Mechanics. Proceedings of the IUTAM Symposium held in Futuroscope, Poitiers, France,
August 3 I-September 4, 1998. 2000 ISBN 0-7923-6604-2
83. D. Weichert and G. Maier (eds.): Inelastic Analysis ofStructures under Variable Loads. Theory
and Engineering Applications. 2000 ISBN 0-7923-6645-X
84. T.-J. Chuang and J.w. Rudnicki (eds.): Multiscale Deformation and Fracture in Materials and
Structures. The James R. Rice 60th Anniversary Volume. 2001 ISBN 0-7923-6718-9
85. S. Narayanan and R.N. Iyengar (eds.): IUTAM Symposium on Nonlinearity and Stochastic
Structural Dynamics. Proceedings of the IUTAM Symposium held in Madras, Chennai, India,
4-8 January 1999 ISBN 0-7923-6733-2
86. S. Murakami and N. Ohno (eds.): IUTAM Symposium on Creep in Structures. Proceedings of
the IUTAM Symposium held in Nagoya, Japan, 3-7 April 2000. 2001 ISBN 0-7923-6737-5
87. W. Ehlers (ed.): IUTAM Symposium on Theoretical and Numerical Methods in Continuum
Mechanics of Porous Materials. Proceedings of the IUTAM Symposium held at the University
of Stuttgart, Germany, September 5-10, 1999. 2001 ISBN 0-7923-6766-9
88. D. Durban, D. Givoli and J.G. Simmonds (eds.): Advances in the Mechanis o/Plates and Shells
The Avinoam Libai Anniversary Volume. 2001 ISBN 0-7923-6785-5
89. U. Gabbert and H.-S. Tzou (eds.): IUTAM Symposium on Smart Structures and Structonic Sys-
tems. Proceedings of the IUTAM Symposium held in Magdeburg, Germany, 26-29 September
2000.2001 ISBN 0-7923-6968-8
Mechanics
90. Y. Ivanov, V. Cheshkov and M. Natova: Polymer Composite Materials - Interface Phenomena
& Processes. 2001 ISBN 0-7923-7008-2
91. R.C. McPhedran, L.C. Botten and N.A. Nicorovici (eds.): IUTAM Symposium on Mechanical
and Electromagnetic Waves in Structured Media. Proceedings of the IUTAM Symposium held
in Sydney, NSW, Australia, 18-22 Januari 1999.2001 ISBN 0-7923-7038-4
92. D.A. Sotiropoulos (ed.): IUTAM Symposium on Mechanical Waves for Composite Structures
Characterization. Proceedings of the IUTAM Symposium held in Chania, Crete, Greece, June
14-17,2000.2001 ISBN 0-7923-7164-X
93. V.M. Alexandrov and D.A. Pozharskii: Three-Dimensional Contact Problems. 2001
ISBN 0-7923-7165-8
94. J.P. Dempsey and H.H. Shen (eds.): IUTAM Symposium on Scaling Laws in Ice Mechanics
and Ice Dynamics. Proceedings of the IUTAM Symposium held in Fairbanks, Alaska, U.S.A.,
13-16 June 2000. 2001
ISBN 1-4020-0171-1
95. U. Kirsch: Design-Oriented Analysis of Structures. A Unified Approach. 2002
ISBN 1-4020-0443-5
96. A. Preumont: Vibration Control ofActive Structures. An Introduction (2 nd Edition). 2002
ISBN 1-4020-0496-6
97. B.L. Karihaloo (ed.): IUTAM Symposium on Analytical and Computational Fracture Mechan-
ics of Non-Homogeneous Materials. Proceedings of the IUTAM Symposium held in Cardiff,
U.K., 18-22 June 2001. 2002 ISBN 1-4020-0510-5
98. S.M. Han and H. Benaroya: Nonlinear and Stochastic Dynamics of Compliant Offshore Struc-
tures.2oo2 ISBN 1-4020-0573-3
99. A.M. Linkov: Boundary Integral Equations in Elasticity Theory. 2002
ISBN 1-4020-0574-1
100. L.P. Lebedev, 1.1. Vorovich and G.M.L. Gladwell: Functional Analysis. Applications in Me-
chanics and Inverse Problems (2 nd Edition). 2002 ISBN 1-4020-0667-5; Pb 1-4020-0756-6
Kluwer Academic Publishers - Dordrecht / Boston / London

ICASElLaRC Interdisciplinary Series in Science and Engineering
1. 1 Buckmaster, T.L. Jackson and A. Kumar (eds.): Combustion in High-Speed Flows.

1994 ISBN 0-7923-2086-X
2. M.Y. Hussaini, T.B. Gatski and TL Jackson (eds.): Transition, Turbulence and
Combustion. Volume I: Transition. 1994
ISBN 0-7923-3084-6; set 0-7923-3086-2
3. M.Y. Hussaini, T.B. Gatski and T.L. Jackson (eds.): Transition, Turbulence and
Combustion. Volume II: Turbulence and Combustion. 1994
ISBN 0-7923-3085-4; set 0-7923-3086-2
4. D.E. Keyes, A. Sameh and V. Venkatakrishnan (eds): Parallel Numerical Algorithms.
1997 ISBN 0-7923-4282-8
5. T.G. Campbell, R.A. Nicolaides and M.D. Salas (eds.): Computational Electromag-
netics and Its Applications. 1997 ISBN 0-7923-4733-1
6. V. Venkatakrishnan, M.D. Salas and S.R. Chakravarthy (eds.): Barriers and Chal-
lenges in Computational Fluid Dynamics. 1998 ISBN 0-7923-4855-9
7. M.D. Salas, IN. Hefner and L. Sakell (eds.): Modeling Complex Turbulent Flows.
1999 ISBN 0-7923-5590-3
KLUWER ACADEMIC PUBLISHERS - DORDRECHT I BOSTON I LONDON

Contact Mechanics - Proceedings of The 3rd Contact Mechanics International Symposium

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Contact Mechanics - Proceedings of The 3rd Contact Mechanics International Symposium

Uploaded by

Copyright:

Available Formats

SOLID MECHANICS AND ITS APPLICATIONS

Series Editor: G.M.L. GLADWELL

Aims and Scope of the Series

For a list of related mechanics titles, see final pages.

MANUEL D.P. MONTEIRO MARQUES

SPRINGER-SCIENCE+BUSINESS MEDIA, B.Y.

ISBN 978-90-481-6099-0 ISBN 978-94-017-1154-8 (eBook)

Printed on acid-free paper

All Rights Reserved

Third Contact Mechanics International Symposium..... .... xi

1. Dynamics and impact

Formulation and well-posedness of unilateral multibody dynamics......................... 25

Impacts with global dissipation index at reentrant corners................................. ... 45

Understanding impact through continuous medium vibrations.............................. 53

Solving rocking block problems with multiple impacts.............................. .......... 65

On integrating stiffmultibody dynamics with contact and friction........................... 83

Analysis of systems with multiple frictional contacts....................................... ... 91

2. Instabilities, oscillations and waves

An example of stick-slip waves. . .... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... . . . . . . . . . . . 109

The influence of contact properties on friction-induced brake vibrations................... 125

Analysis of eigenvalue problems modeIling friction: sufficient conditions of non-

3. Contact models, results and applications

A model of adhesion added to contact with friction........................................... 161

Analysis of a "walking" punch.................................................................... 169

Phenomenological model of friction accounting for subsurface plastic deformation in

Thermoelastic instabilities in automotive disc brakes - finite element analysis and

A contact formulation for electrical and mechanical resistance. ............................ 211

Approximation of quasi static Signorini problems with local friction by a mixed

A beam in adhesive contact........................................................................ 277

Coulomb fluid-solid interface law in lubrication............................................... 285

Elastodynamic friction problem with a "surface inertia" perturbation....................... 293

Solvability of thenno-viscoelastic contact problems with Coulomb friction and nonlin-

A simplified model of impact..................................................................... 309

Solution methods for structural optimization in contact rod problems...................... 325

Contact between 3D beams with rectangular cross-sections.. ............................. ... 355

Algorithms for thermoelastic wear problems................................................ .... 363

A new approach to the 2D transient rolling contact problem............................. .... 387

Numerical approximation of the elastic-viscoplastic contact problem using noncoin-

Contact in the Arlequin framework....... . . . . . . . . . . . . . . . . . .. .. . . . . . . . . . . . . . . . . . . . . . . . . . ..... . . . . . 403

Appendix A. Scientific program of the Third Contact

Appendix B. List of participants in the Third Contact

This volume contains 44 papers presented at the Third Contact

Joao Martins and Manuel Monteiro Marques

Lisbon, February 2002

3. Sponsors and Acknowledgments

The Organizing Committee of CMIS 2001 gratefully acknowledges the

• Instituto de Engenharia de Estruturas, Territorio e Constru'Yao (ICIST),

Jean Jacques MOREAU

Abstract Contact Dynamics is a numerical method, suitable for computing the

1.2. Analytical setting

ga(t,q)?O, aE{1, ... ,~}, (1)

The system motion on a time-interval I, with origin to, is described

A(t,q)~~ = F(t,q,u) + ~ro, (3)

to hold for almost every t. Here A denotes in general the n x n inertia

Wa E E 3 , a known function of t and q, accounts for the given motion of

2. The 'Contact Dynamics' approach

go(t, q) ::; 0 :::} nO. Uo ~ 0 (6)

We propose to say that a package of information concerning the possi-

2.2. Nonsmoothness in time and collisions

A{t,q)du = F{t,q,u)dt + Lds a , (8)

motion. The laws of restitution of Newton, Poisson or others amount to

2.3. The CD handling of nonsmooth evolutions